From viridia at gmail.com  Mon Oct 19 00:12:43 2009
From: viridia at gmail.com (Talin)
Date: Sun, 18 Oct 2009 22:12:43 -0700
Subject: [llvm-commits] PATCH: Additional DebugFactory methods
In-Reply-To: <7083196B-95FE-4F1D-89AE-2D38B2D0D72F@apple.com>
References: <d1b71f830910052248naeeba24pe4822e09598119c3@mail.gmail.com> 
	<7083196B-95FE-4F1D-89AE-2D38B2D0D72F@apple.com>
Message-ID: <d1b71f830910182212t4257ab63g5d860b65197b8950@mail.gmail.com>

Here's an updated version of the patch which uses StringRef as requested.
Also, I have changed the method names to avoid the ambiguity problems
encountered in the last patch. The new methods end with "Ex" which can stand
either for "Expression" or "Extension" depending on your preference. :)

On Sat, Oct 17, 2009 at 9:32 PM, Chris Lattner <clattner at apple.com> wrote:

>
> On Oct 5, 2009, at 10:48 PM, Talin wrote:
>
>  The attached patch adds some additional convenience methods to
>> DebugFactory. Specifically, methods that take a uint64 for size, alignment
>> or offset, can now also take a Constant *. This is very useful for
>> target-independent frontends which can use things like
>> ConstantExpr::getSizeof(). I've been using this for about a week in my own
>> code and it seems to work fine,
>>
>
> Hi Talin,
>
> Sorry for the delay, still getting caught up.  Please change this to use
> StringRef like the current DebugInfo.h methods do.
>
> Devang, can you review this in more detail?
>
> -Chris
>
>


-- 
-- Talin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091018/3dbd8077/attachment.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: debug_constants.patch
Type: application/octet-stream
Size: 7759 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091018/3dbd8077/attachment.obj 

From sabre at nondot.org  Mon Oct 19 00:31:11 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 05:31:11 -0000
Subject: [llvm-commits] [llvm] r84463 - in /llvm/trunk/lib/AsmParser:
	LLParser.cpp LLToken.h
Message-ID: <200910190531.n9J5VBQD029063@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 00:31:10 2009
New Revision: 84463

URL: http://llvm.org/viewvc/llvm-project?rev=84463&view=rev
Log:
clean up after metadata changes.

Modified:
    llvm/trunk/lib/AsmParser/LLParser.cpp
    llvm/trunk/lib/AsmParser/LLToken.h

Modified: llvm/trunk/lib/AsmParser/LLParser.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/AsmParser/LLParser.cpp?rev=84463&r1=84462&r2=84463&view=diff

==============================================================================
--- llvm/trunk/lib/AsmParser/LLParser.cpp (original)
+++ llvm/trunk/lib/AsmParser/LLParser.cpp Mon Oct 19 00:31:10 2009
@@ -1051,14 +1051,12 @@
 ///   ::= /* empty */
 ///   ::= !dbg !42
 bool LLParser::ParseOptionalCustomMetadata() {
-
-  std::string Name;
-  if (Lex.getKind() == lltok::NamedOrCustomMD) {
-    Name = Lex.getStrVal();
-    Lex.Lex();
-  } else
+  if (Lex.getKind() != lltok::NamedOrCustomMD)
     return false;
 
+  std::string Name = Lex.getStrVal();
+  Lex.Lex();
+
   if (Lex.getKind() != lltok::Metadata)
     return TokError("Expected '!' here");
   Lex.Lex();
@@ -2880,8 +2878,6 @@
   if (ParseType(Ty, true /*void allowed*/)) return true;
 
   if (Ty->isVoidTy()) {
-    if (EatIfPresent(lltok::comma))
-      if (ParseOptionalCustomMetadata()) return true;
     Inst = ReturnInst::Create(Context);
     return false;
   }
@@ -2917,8 +2913,6 @@
       }
     }
   }
-  if (EatIfPresent(lltok::comma))
-    if (ParseOptionalCustomMetadata()) return true;
 
   Inst = ReturnInst::Create(Context, RV);
   return false;

Modified: llvm/trunk/lib/AsmParser/LLToken.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/AsmParser/LLToken.h?rev=84463&r1=84462&r2=84463&view=diff

==============================================================================
--- llvm/trunk/lib/AsmParser/LLToken.h (original)
+++ llvm/trunk/lib/AsmParser/LLToken.h Mon Oct 19 00:31:10 2009
@@ -64,7 +64,6 @@
     kw_sideeffect,
     kw_msasm,
     kw_gc,
-    kw_dbg,
     kw_c,
 
     kw_cc, kw_ccc, kw_fastcc, kw_coldcc,




From sabre at nondot.org  Mon Oct 19 00:34:14 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 05:34:14 -0000
Subject: [llvm-commits] [llvm] r84464 - in /llvm/trunk/test/Feature:
 md_on_instruction.ll md_on_instruction2.ll
Message-ID: <200910190534.n9J5YE8Q029161@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 00:34:14 2009
New Revision: 84464

URL: http://llvm.org/viewvc/llvm-project?rev=84464&view=rev
Log:
eliminate md_on_instruction.ll, md_on_instruction2.ll is a superset of it.

Added:
    llvm/trunk/test/Feature/md_on_instruction.ll
      - copied unchanged from r84459, llvm/trunk/test/Feature/md_on_instruction2.ll
Removed:
    llvm/trunk/test/Feature/md_on_instruction2.ll

Removed: llvm/trunk/test/Feature/md_on_instruction2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Feature/md_on_instruction2.ll?rev=84463&view=auto

==============================================================================
--- llvm/trunk/test/Feature/md_on_instruction2.ll (original)
+++ llvm/trunk/test/Feature/md_on_instruction2.ll (removed)
@@ -1,22 +0,0 @@
-; RUN: llvm-as < %s | llvm-dis | grep " !dbg " | count 4
-define i32 @foo() nounwind ssp {
-entry:
-  %retval = alloca i32                            ; <i32*> [#uses=2]
-  call void @llvm.dbg.func.start(metadata !0)
-  store i32 42, i32* %retval, !dbg !3
-  br label %0, !dbg !3
-
-; <label>:0                                       ; preds = %entry
-  call void @llvm.dbg.region.end(metadata !0)
-  %1 = load i32* %retval, !dbg !3                  ; <i32> [#uses=1]
-  ret i32 %1, !dbg !3
-}
-
-declare void @llvm.dbg.func.start(metadata) nounwind readnone
-
-declare void @llvm.dbg.region.end(metadata) nounwind readnone
-
-!0 = metadata !{i32 458798, i32 0, metadata !1, metadata !"foo", metadata !"foo", metadata !"foo", metadata !1, i32 1, metadata !2, i1 false, i1 true}
-!1 = metadata !{i32 458769, i32 0, i32 12, metadata !"foo.c", metadata !"/tmp", metadata !"clang 1.0", i1 true, i1 false, metadata !"", i32 0}
-!2 = metadata !{i32 458788, metadata !1, metadata !"int", metadata !1, i32 0, i64 32, i64 32, i64 0, i32 0, i32 5}
-!3 = metadata !{i32 1, i32 13, metadata !1, metadata !1}




From sabre at nondot.org  Mon Oct 19 00:51:03 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 05:51:03 -0000
Subject: [llvm-commits] [llvm] r84465 -
	/llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
Message-ID: <200910190551.n9J5p3m1029698@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 00:51:03 2009
New Revision: 84465

URL: http://llvm.org/viewvc/llvm-project?rev=84465&view=rev
Log:
simplify.

Modified:
    llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp

Modified: llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp?rev=84465&r1=84464&r2=84465&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp (original)
+++ llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp Mon Oct 19 00:51:03 2009
@@ -517,9 +517,7 @@
       }
 
       // Code: [strchar x N]
-      const char *StrBegin = MDS->begin();
-      for (unsigned i = 0, e = MDS->length(); i != e; ++i)
-        Record.push_back(StrBegin[i]);
+      Record.append(MDS->begin(), MDS->end());
 
       // Emit the finished record.
       Stream.EmitRecord(bitc::METADATA_STRING, Record, MDSAbbrev);




From jyasskin at google.com  Mon Oct 19 01:57:07 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Sun, 18 Oct 2009 23:57:07 -0700
Subject: [llvm-commits] [PATCH] Add a ValueMap type,
	whose keys act like 	WeakVHs
In-Reply-To: <bb55930e0910171952p4e5ef885nf35c860c794d626@mail.gmail.com>
References: <0016363b7bde0f990504761ca9c4@google.com>
	<A2FC2BDC-33EB-4611-A76E-273983671CE1@apple.com>
	<bb55930e0910171952p4e5ef885nf35c860c794d626@mail.gmail.com>
Message-ID: <bb55930e0910182357t2e5538f8r2f6e348190b31082@mail.gmail.com>

There's a new and more correct patch at
http://codereview.appspot.com/download/issue132055_1002.diff. I made 3
changes:

1. Support const keys.
2. Support a lock around map changes from the CallbackVH. The
ExecutionEngine needs this since the CallbackVH can be fired from a
place the user doesn't really control.
3. Fix a bug involving callbacks that call back into the ValueMap.

This version passes tests as a replacement for the CallbackVHs in
http://codereview.appspot.com/133043/show.

On Sat, Oct 17, 2009 at 7:52 PM, Jeffrey Yasskin <jyasskin at google.com> wrote:
> On Sat, Oct 17, 2009 at 4:05 PM, Chris Lattner <clattner at apple.com> wrote:
>>
>> On Oct 17, 2009, at 12:42 AM, jyasskin at gmail.com wrote:
>>>
>>> I guess I'll have to figure out how to fix it before submitting.
>>
>> I'll review it when it is fixed.
>
> I've fixed lit.py in r84387. This patch didn't need to change.
>
>>> Description:
>>> ValueMap<Value*, T> is safe to use even when keys get RAUWed and
>>> deleted
>>> during its lifetime. By default the keys act like WeakVHs, but users
>>> can
>>> pass a third template parameter to configure how updates work and
>>> whether to do anything beyond updating the map on each action.
>>>
>>> This type should make http://codereview.appspot.com/133043/diff/1/2
>>> quite a bit simpler, but I haven't yet updated it so I could have
>>> missed
>>> something.
>>
>> Sounds nice, what happens when X->RAUW(Y) and both X and Y are already
>> in the map?
>
> Right now, you wind up with Y mapped to X's old target, and Y's old
> target lost. ValueMapTest.DefaultCollisionBehavior tests for this. It
> should be easy to add the ability to configure this behavior
> (Config::Merge(ValueT& OldTarget, const ValueT& NewTarget), maybe with
> extra parameters), but I didn't need it for my other changes.
>


From clattner at apple.com  Mon Oct 19 02:06:31 2009
From: clattner at apple.com (Chris Lattner)
Date: Mon, 19 Oct 2009 00:06:31 -0700
Subject: [llvm-commits] metadata.h/cpp review stuff
Message-ID: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>

Hi Devang,

I'm just now getting caught up on your work.  Here are some thoughts/ 
questions on Metadata.h/cpp:


The metadata hierarchy is:

User
   -> MetadataBase
      -> MDString
      -> MDNode
      -> NamedMDNode

Why not:

Value
   -> MetadataBase
      -> MDString
      -> MDNode
      -> NamedMDNode

This would shrink them significantly and reduce the games you have to  
play with hiding getNumOperands, overloading operator new, etc.  Also,  
MetadataBase doesn't have much common state, it isn't clear it needs  
to exist at all.

There is some seriously confused code circling around  
MetadataBase::resizeOperands.  It looks like half the code thinks that  
operands can happen, but half doesn't.  Please remove all this stuff.


In Metadata.h:


#include "llvm/User.h"
#include "llvm/Type.h"
#include "llvm/OperandTraits.h"
#include "llvm/ADT/FoldingSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/ADT/ilist_node.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/ValueHandle.h"

This is including way too much.  Please reduce this as aggressively as  
possible.



class MDString : public MetadataBase {
...
   explicit MDString(LLVMContext &C, const char *begin, unsigned l)

This ctor should just take a StringRef.


...
   static MDString *get(LLVMContext &Context, const StringRef &Str);

Please take StringRef's by value, not by const&.


class MDNode : public MetadataBase, public FoldingSetNode {
...
   friend class ElementVH;
   // Use CallbackVH to hold MDNOde elements.
   struct ElementVH : public CallbackVH {
     MDNode *Parent;
     ElementVH(Value *V, MDNode *P) : CallbackVH(V), Parent(P) {}
     ~ElementVH() {}

     virtual void deleted() {
       Parent->replaceElement(this->operator Value*(), 0);
     }

     virtual void allUsesReplacedWith(Value *NV) {
       Parent->replaceElement(this->operator Value*(), NV);
     }
   };
   // Replace each instance of F from the element list of this node  
with T.
   void replaceElement(Value *F, Value *T);

   SmallVector<ElementVH, 4> Node;


Please move all of this stuff out of line to Metadata.cpp.  If needbe,  
Node should be declared as a 'void*' in the header.  Also, there is no  
reason to use a SmallVector here: the size is fixed on node  
construction.  Just use a pointer to array and new[]/delete[] the  
array.  Keeping a SmallVector in NamedMDNode (but not MDNode) is  
completely fine.


Why does MDNode::dropAllReferences need to exist?  It seems *highly*  
unsafe because it clears 'Node' but doesn't update the FoldingSet.


   Value *getElement(unsigned i) const {
     assert(getNumElements() > i && "Invalid element number!");

This is fine, but please use "i < getNumElements()" because that is  
more idiomatic.  This happens multiple times in the file.


/// WeakMetadataVH - a weak value handle for metadata.
class WeakMetadataVH : public WeakVH {
public:

I think this should be removed and the use replaced with TrackingVH??



NamedMDNode::setParent shouldn't be public.  Does  
NamedMDNode::dropAllReferences need to exist?

I don't understand NamedMDNode::Create(const NamedMDNode *NMD, Module  
*M).  Is this a weird copy ctor?  Why does it need to exist?


   void addElement(MetadataBase *M) {
     resizeOperands(0);
     OperandList[NumOperands++] = M;
     Node.push_back(WeakMetadataVH(M));
   }

How can this work?  Isn't this an invalid access into OperandList?


/// MetadataContext -
/// MetadataContext handles uniquing and assignment of IDs for custom  
metadata
/// types. Custom metadata handler names do not contain spaces. And  
the name
/// must start with an alphabet. The regular expression used to check  
name

"start with an alphabet" is missing something.

class MetadataContext {
public:
   typedef std::pair<unsigned, WeakVH> MDPairTy;
   typedef SmallVector<MDPairTy, 2> MDMapTy;
   typedef DenseMap<const Instruction *, MDMapTy> MDStoreTy;
   friend class BitcodeReader;
private:

   /// MetadataStore - Collection of metadata used in this context.
   MDStoreTy MetadataStore;

   /// MDHandlerNames - Map to hold metadata handler names.
   StringMap<unsigned> MDHandlerNames;


These datastructures are all implementation details.  They should be  
hidden behind a void*, allowing you to avoid #include of DenseMap.h,  
StringMap.h etc.  MDPairTy should be replaced with TrackingVH<MDNode>.



   /// RegisterMDKind - Register a new metadata kind and return its ID.
   /// A metadata kind can be registered only once.
   unsigned RegisterMDKind(const char *Name);

   /// getMDKind - Return metadata kind. If the requested metadata kind
   /// is not registered then return 0.
   unsigned getMDKind(const char *Name);

   /// validName - Return true if Name is a valid custom metadata  
handler name.
   bool validName(const char *Name);

These methods should all take StringRef for consistency.  The second  
should be const, the third should be static.  validName ->  
isValidName.  RegisterMDKind -> registerMDKind.


   /// getMD - Get the metadata of given kind attached with an  
Instruction.
   /// If the metadata is not found then return 0.
   MDNode *getMD(unsigned Kind, const Instruction *Inst);

"with an instruction" -> "to an instruction".   This occurs several  
times in both files.

   /// removeMDs - Remove all metadata attached with an instruction.
   void removeMDs(const Instruction *Inst);

removeAllMetadata()?



   /// getHandlerNames - Get handler names. This is used by bitcode
   /// writer.
   const StringMap<unsigned> *getHandlerNames();

This is gross for several reasons: at the least, it should be const.   
It also exposes implementation details.  Finally, the bitcode writer  
is currently writing these out in iterator order which is not at all  
deterministic.  It would be better to have something like:

   void getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef>  
 > &) const;

and have the implementation of getHandlerNames do the sort (with  
array_pod_sort).




   /// ValueIsDeleted - This handler is used to update metadata store
   /// when a value is deleted.
   void ValueIsDeleted(const Value *) {}
   void ValueIsDeleted(const Instruction *Inst) {
     removeMDs(Inst);
   }
   void ValueIsRAUWd(Value *V1, Value *V2);

   /// ValueIsCloned - This handler is used to update metadata store
   /// when In1 is cloned to create In2.
   void ValueIsCloned(const Instruction *In1, Instruction *In2);

These should not be public and/or should not be in the MetadataContext  
class.


Metadata.cpp:

void MDNode::replaceElement(Value *From, Value *To) {
...

You don't know that MDNode::replaceElement will be called before or  
after the VH operands uses are updated.  I don't think this method in  
general is safe because a bunch of stuff depends on pointer equality  
with the "from" value.


   // MDNode only lists metadata elements in operand list, because  
MDNode
   // used by MDNode is considered a valid use. However on the side,  
MDNode
   // using a non-metadata value is not considered a "use" of non- 
metadata
   // value.

Why do MD uses try to show up in the operand list?  This seems really  
complex, what benefit does it provide?

   if (MetadataBase *MDTo = dyn_cast_or_null<MetadataBase>(To)) {
     for (SmallVector<unsigned, 4>::iterator OI = OpIndexes.begin(),
            OE = OpIndexes.end(); OI != OE; ++OI)
       setOperand(*OI, MDTo);
   } else {
     for (SmallVector<unsigned, 4>::iterator OI = OpIndexes.begin(),
            OE = OpIndexes.end(); OI != OE; ++OI)
       setOperand(*OI, 0);
   }

This doesn't make sense to me.  If one MD operand is replaced, it sets  
them all to the new thing?

NamedMDNode::NamedMDNode(LLVMContext &C, const Twine &N,
...
   resizeOperands(NumMDs);

   for (unsigned i = 0; i != NumMDs; ++i) {
     if (MDs[i])
       OperandList[NumOperands++] = MDs[i];
     Node.push_back(WeakMetadataVH(MDs[i]));
   }

This is wrong unless resizeOperands does something really strange  
(likely).  Why does NamedMDNode put anything in the 'User' operands  
list at all?  I think this should just be removed.


unsigned MetadataContext::RegisterMDKind(const char *Name) {
..
   assert(MDHandlerNames.find(Name) == MDHandlerNames.end()
          && "Already registered MDKind!");

Just use: assert(MDHandlerNames.count(Name) == 0 && ...



   MDHandlerNames[Name] = Count + 1;
   return Count + 1;

How about: return MDHandlerNames[Name] = Count + 1;


unsigned MetadataContext::getMDKind(const char *Name) {
   assert(validName(Name) && "Invalid custome metadata name!");
   StringMap<unsigned>::iterator I = MDHandlerNames.find(Name);

Random microoptimization for assert builds: the check for a valid name  
can be sunk into the case where the query fails.  If the query  
succeeds (common case) the name was already validated.


void MetadataContext::addMD(unsigned MDKind, MDNode *Node, Instruction  
*Inst) {
...
   MDStoreTy::iterator I = MetadataStore.find(Inst);
   if (I == MetadataStore.end()) {
     MDMapTy Info;
     Info.push_back(std::make_pair(MDKind, Node));
     MetadataStore.insert(std::make_pair(Inst, Info));
     return;
   }

It is actually better to just rely on default construction in the  
map.  Just use:
   MDMapTy &Info = MetadataStore[Inst];
and delete the code above.


/// removeMD - Remove metadata of given kind attached with an  
instuction.
void MetadataContext::removeMD(unsigned Kind, Instruction *Inst) {
   MDStoreTy::iterator I = MetadataStore.find(Inst);
   if (I == MetadataStore.end())
     return;
   MDMapTy &Info = I->second;

This seems wrong.  The method should only be called if the  
'HasMetadata' bit is set on the instruction.  If it is set and there  
is no entry in MetaDataStore, then it is a logic bug (so it should  
assert).  Just do:

MDMapTy &Info = MetadataStore[Inst];
assert(!Info.empty() && ..

or whatever.



/// removeMDs - Remove all metadata attached with an instruction.
void MetadataContext::removeMDs(const Instruction *Inst) {
   // Find Metadata handles for this instruction.
   MDStoreTy::iterator I = MetadataStore.find(Inst);
   assert(I != MetadataStore.end() && "Invalid custom metadata info!");
   MDMapTy &Info = I->second;

   // FIXME : Give all metadata handlers a chance to adjust.

   // Remove the entries for this instruction.
   Info.clear();
   MetadataStore.erase(I);
}


This doesn't look fully implemented.   In addition to the FIXME (What  
does it mean?) this should clear the  HasMetadata bit on the  
instruction.  Also, don't use find here.  This entire method can  
currently be replaced with this, which is identical in behavior:

void MetadataContext::removeMDs(const Instruction *Inst) {
   MetadataStore.erase(Inst);
}


void MetadataContext::copyMD(Instruction *In1, Instruction *In2) {
   assert(In1 && In2 && "Invalid instruction!");
    MDStoreTy::iterator I = MetadataStore.find(In1);
   if (I == MetadataStore.end())
     return;

   MDMapTy &In1Info = I->second;
   MDMapTy In2Info;
   for (MDMapTy::iterator I = In1Info.begin(), E = In1Info.end(); I !=  
E; ++I)
     if (MDNode *MD = dyn_cast_or_null<MDNode>(I->second))
       addMD(I->first, MD, In2);


Similar comments to above.  Also, In2Info is dead here.   
dyn_cast_or_null should go away when you start using trackingvh  
instead of weakvh (not just here, check the rest of the file).



MDNode *MetadataContext::getMD(unsigned MDKind, const Instruction  
*Inst) {
   MDStoreTy::iterator I = MetadataStore.find(Inst);
   if (I == MetadataStore.end())
     return NULL;

Should be an assert as above.



const MetadataContext::MDMapTy *
MetadataContext::getMDs(const Instruction *Inst) {

I really don't like this method.  Exposing implementation details like  
this is bad.  IF this is for the bc writer, please follow something  
similar to the getHandlerNames approach above.

  void MetadataContext::ValueIsCloned(const Instruction *In1,  
Instruction *In2) {
   // Find Metadata handles for In1.
   MDStoreTy::iterator I = MetadataStore.find(In1);
   assert(I != MetadataStore.end() && "Invalid custom metadata info!");

   // FIXME : Give all metadata handlers a chance to adjust.

   MDMapTy &In1Info = I->second;
   MDMapTy In2Info;
   for (MDMapTy::iterator I = In1Info.begin(), E = In1Info.end(); I !=  
E; ++I)
     if (MDNode *MD = dyn_cast_or_null<MDNode>(I->second))
       addMD(I->first, MD, In2);
}

How is this different than copyMD?

/// ValueIsRAUWd - This handler is used when V1's all uses are  
replaced by
/// V2.
void MetadataContext::ValueIsRAUWd(Value *V1, Value *V2) {
   Instruction *I1 = dyn_cast<Instruction>(V1);
   Instruction *I2 = dyn_cast<Instruction>(V2);
   if (!I1 || !I2)

This seems very very wrong.  If I do:

X->RAUW(Y);

Then the properties on X are not necessarily true in Y's scope.  The  
debug info (for example) should just be discarded.  Value range info  
is another example: X may be defined lower in the dom tree where some  
property holds that doesn't hold on Y.

-Chris

















From sabre at nondot.org  Mon Oct 19 02:11:00 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 07:11:00 -0000
Subject: [llvm-commits] [llvm] r84471 - in /llvm/trunk:
 include/llvm/Metadata.h lib/VMCore/Metadata.cpp
Message-ID: <200910190711.n9J7B0iu032296@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 02:10:59 2009
New Revision: 84471

URL: http://llvm.org/viewvc/llvm-project?rev=84471&view=rev
Log:
various cleanups.

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84471&r1=84470&r2=84471&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Mon Oct 19 02:10:59 2009
@@ -47,12 +47,6 @@
 
   void resizeOperands(unsigned NumOps);
 public:
-  /// isNullValue - Return true if this is the value that would be returned by
-  /// getNullValue.  This always returns false because getNullValue will never
-  /// produce metadata.
-  virtual bool isNullValue() const {
-    return false;
-  }
 
   /// Methods for support type inquiry through isa, cast, and dyn_cast:
   static inline bool classof(const MetadataBase *) { return true; }
@@ -85,15 +79,17 @@
   
   StringRef getString() const { return Str; }
 
-  unsigned length() const { return Str.size(); }
+  unsigned getLength() const { return Str.size(); }
 
+  typedef StringRef::iterator iterator;
+  
   /// begin() - Pointer to the first byte of the string.
   ///
-  const char *begin() const { return Str.begin(); }
+  iterator begin() const { return Str.begin(); }
 
   /// end() - Pointer to one byte past the end of the string.
   ///
-  const char *end() const { return Str.end(); }
+  iterator end() const { return Str.end(); }
 
   /// Methods for support type inquiry through isa, cast, and dyn_cast:
   static inline bool classof(const MDString *) { return true; }
@@ -133,7 +129,7 @@
   SmallVector<ElementVH, 4> Node;
 
 protected:
-  explicit MDNode(LLVMContext &C, Value*const* Vals, unsigned NumVals);
+  explicit MDNode(LLVMContext &C, Value *const *Vals, unsigned NumVals);
 public:
   // Do not allocate any space for operands.
   void *operator new(size_t s) {
@@ -141,7 +137,7 @@
   }
   // Constructors and destructors.
   static MDNode *get(LLVMContext &Context, 
-                     Value* const* Vals, unsigned NumVals);
+                     Value *const *Vals, unsigned NumVals);
 
   /// dropAllReferences - Remove all uses and clear node vector.
   void dropAllReferences();
@@ -151,7 +147,7 @@
   
   /// getElement - Return specified element.
   Value *getElement(unsigned i) const {
-    assert (getNumElements() > i && "Invalid element number!");
+    assert(getNumElements() > i && "Invalid element number!");
     return Node[i];
   }
 
@@ -170,21 +166,10 @@
   elem_iterator elem_begin()             { return Node.begin(); }
   elem_iterator elem_end()               { return Node.end();   }
 
-  /// isNullValue - Return true if this is the value that would be returned by
-  /// getNullValue.  This always returns false because getNullValue will never
-  /// produce metadata.
-  virtual bool isNullValue() const {
-    return false;
-  }
-
   /// Profile - calculate a unique identifier for this MDNode to collapse
   /// duplicates
   void Profile(FoldingSetNodeID &ID) const;
 
-  virtual void replaceUsesOfWithOnConstant(Value *, Value *, Use *) {
-    llvm_unreachable("This should never be called because MDNodes have no ops");
-  }
-
   /// Methods for support type inquiry through isa, cast, and dyn_cast:
   static inline bool classof(const MDNode *) { return true; }
   static bool classof(const Value *V) {
@@ -229,7 +214,7 @@
   typedef SmallVectorImpl<WeakMetadataVH>::iterator elem_iterator;
 
 protected:
-  explicit NamedMDNode(LLVMContext &C, const Twine &N, MetadataBase*const* Vals, 
+  explicit NamedMDNode(LLVMContext &C, const Twine &N, MetadataBase*const *Vals, 
                        unsigned NumVals, Module *M = 0);
 public:
   // Do not allocate any space for operands.
@@ -237,7 +222,7 @@
     return User::operator new(s, 0);
   }
   static NamedMDNode *Create(LLVMContext &C, const Twine &N, 
-                             MetadataBase*const*MDs, 
+                             MetadataBase *const *MDs, 
                              unsigned NumMDs, Module *M = 0) {
     return new NamedMDNode(C, N, MDs, NumMDs, M);
   }
@@ -261,7 +246,7 @@
 
   /// getElement - Return specified element.
   MetadataBase *getElement(unsigned i) const {
-    assert (getNumElements() > i && "Invalid element number!");
+    assert(getNumElements() > i && "Invalid element number!");
     return Node[i];
   }
 
@@ -284,18 +269,6 @@
   elem_iterator elem_begin()             { return Node.begin(); }
   elem_iterator elem_end()               { return Node.end();   }
 
-  /// isNullValue - Return true if this is the value that would be returned by
-  /// getNullValue.  This always returns false because getNullValue will never
-  /// produce metadata.
-  virtual bool isNullValue() const {
-    return false;
-  }
-
-  virtual void replaceUsesOfWithOnConstant(Value *, Value *, Use *) {
-    llvm_unreachable(
-                "This should never be called because NamedMDNodes have no ops");
-  }
-
   /// Methods for support type inquiry through isa, cast, and dyn_cast:
   static inline bool classof(const NamedMDNode *) { return true; }
   static bool classof(const Value *V) {

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84471&r1=84470&r2=84471&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Mon Oct 19 02:10:59 2009
@@ -20,7 +20,7 @@
 using namespace llvm;
 
 //===----------------------------------------------------------------------===//
-//MetadataBase implementation
+// MetadataBase implementation.
 //
 
 /// resizeOperands - Metadata keeps track of other metadata uses using 
@@ -48,7 +48,7 @@
   if (OldOps) Use::zap(OldOps, OldOps + e, true);
 }
 //===----------------------------------------------------------------------===//
-//MDString implementation
+// MDString implementation.
 //
 MDString *MDString::get(LLVMContext &Context, const StringRef &Str) {
   LLVMContextImpl *pImpl = Context.pImpl;
@@ -56,16 +56,15 @@
   StringMapEntry<MDString *> &Entry = 
     pImpl->MDStringCache.GetOrCreateValue(Str);
   MDString *&S = Entry.getValue();
-  if (!S) S = new MDString(Context, Entry.getKeyData(),
-                           Entry.getKeyLength());
-
-  return S;
+  if (S) return S;
+  
+  return S = new MDString(Context, Entry.getKeyData(), Entry.getKeyLength());
 }
 
 //===----------------------------------------------------------------------===//
-//MDNode implementation
+// MDNode implementation.
 //
-MDNode::MDNode(LLVMContext &C, Value*const* Vals, unsigned NumVals)
+MDNode::MDNode(LLVMContext &C, Value *const *Vals, unsigned NumVals)
   : MetadataBase(Type::getMetadataTy(C), Value::MDNodeVal) {
   NumOperands = 0;
   resizeOperands(NumVals);
@@ -91,19 +90,20 @@
   for (unsigned i = 0; i != NumVals; ++i)
     ID.AddPointer(Vals[i]);
 
-  pImpl->ConstantsLock.reader_acquire();
   void *InsertPoint;
-  MDNode *N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
-  pImpl->ConstantsLock.reader_release();
+  MDNode *N;
+  {
+    sys::SmartScopedReader<true> Reader(pImpl->ConstantsLock);
+    N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
+  }  
+  if (N) return N;
   
+  sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
+  N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
   if (!N) {
-    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
-    N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
-    if (!N) {
-      // InsertPoint will have been set by the FindNodeOrInsertPos call.
-      N = new MDNode(Context, Vals, NumVals);
-      pImpl->MDNodeSet.InsertNode(N, InsertPoint);
-    }
+    // InsertPoint will have been set by the FindNodeOrInsertPos call.
+    N = new MDNode(Context, Vals, NumVals);
+    pImpl->MDNodeSet.InsertNode(N, InsertPoint);
   }
 
   return N;
@@ -209,10 +209,10 @@
 }
 
 //===----------------------------------------------------------------------===//
-//NamedMDNode implementation
+// NamedMDNode implementation.
 //
 NamedMDNode::NamedMDNode(LLVMContext &C, const Twine &N,
-                         MetadataBase*const* MDs, 
+                         MetadataBase *const *MDs, 
                          unsigned NumMDs, Module *ParentModule)
   : MetadataBase(Type::getMetadataTy(C), Value::NamedMDNodeVal), Parent(0) {
   setName(N);
@@ -229,7 +229,7 @@
 }
 
 NamedMDNode *NamedMDNode::Create(const NamedMDNode *NMD, Module *M) {
-  assert (NMD && "Invalid source NamedMDNode!");
+  assert(NMD && "Invalid source NamedMDNode!");
   SmallVector<MetadataBase *, 4> Elems;
   for (unsigned i = 0, e = NMD->getNumElements(); i != e; ++i)
     Elems.push_back(NMD->getElement(i));
@@ -254,13 +254,13 @@
 }
 
 //===----------------------------------------------------------------------===//
-//Metadata implementation
+// MetadataContext implementation.
 //
 
 /// RegisterMDKind - Register a new metadata kind and return its ID.
 /// A metadata kind can be registered only once. 
 unsigned MetadataContext::RegisterMDKind(const char *Name) {
-  assert (validName(Name) && "Invalid custome metadata name!");
+  assert(validName(Name) && "Invalid custome metadata name!");
   unsigned Count = MDHandlerNames.size();
   assert(MDHandlerNames.find(Name) == MDHandlerNames.end() 
          && "Already registered MDKind!");
@@ -292,7 +292,7 @@
 /// getMDKind - Return metadata kind. If the requested metadata kind
 /// is not registered then return 0.
 unsigned MetadataContext::getMDKind(const char *Name) {
-  assert (validName(Name) && "Invalid custome metadata name!");
+  assert(validName(Name) && "Invalid custome metadata name!");
   StringMap<unsigned>::iterator I = MDHandlerNames.find(Name);
   if (I == MDHandlerNames.end())
     return 0;
@@ -302,7 +302,7 @@
 
 /// addMD - Attach the metadata of given kind with an Instruction.
 void MetadataContext::addMD(unsigned MDKind, MDNode *Node, Instruction *Inst) {
-  assert (Node && "Unable to add custome metadata");
+  assert(Node && "Invalid null MDNode");
   Inst->HasMetadata = true;
   MDStoreTy::iterator I = MetadataStore.find(Inst);
   if (I == MetadataStore.end()) {
@@ -349,7 +349,7 @@
 void MetadataContext::removeMDs(const Instruction *Inst) {
   // Find Metadata handles for this instruction.
   MDStoreTy::iterator I = MetadataStore.find(Inst);
-  assert (I != MetadataStore.end() && "Invalid custom metadata info!");
+  assert(I != MetadataStore.end() && "Invalid custom metadata info!");
   MDMapTy &Info = I->second;
   
   // FIXME : Give all metadata handlers a chance to adjust.
@@ -362,7 +362,7 @@
 /// copyMD - If metadata is attached with Instruction In1 then attach
 /// the same metadata to In2.
 void MetadataContext::copyMD(Instruction *In1, Instruction *In2) {
-  assert (In1 && In2 && "Invalid instruction!");
+  assert(In1 && In2 && "Invalid instruction!");
    MDStoreTy::iterator I = MetadataStore.find(In1);
   if (I == MetadataStore.end())
     return;
@@ -389,12 +389,13 @@
 }
 
 /// getMDs - Get the metadata attached with an Instruction.
-const MetadataContext::MDMapTy *MetadataContext::getMDs(const Instruction *Inst) {
+const MetadataContext::MDMapTy *
+MetadataContext::getMDs(const Instruction *Inst) {
   MDStoreTy::iterator I = MetadataStore.find(Inst);
   if (I == MetadataStore.end())
     return NULL;
   
-  return &(I->second);
+  return &I->second;
 }
 
 /// getHandlerNames - Get handler names. This is used by bitcode
@@ -408,7 +409,7 @@
 void MetadataContext::ValueIsCloned(const Instruction *In1, Instruction *In2) {
   // Find Metadata handles for In1.
   MDStoreTy::iterator I = MetadataStore.find(In1);
-  assert (I != MetadataStore.end() && "Invalid custom metadata info!");
+  assert(I != MetadataStore.end() && "Invalid custom metadata info!");
 
   // FIXME : Give all metadata handlers a chance to adjust.
 




From edwintorok at gmail.com  Mon Oct 19 02:38:01 2009
From: edwintorok at gmail.com (=?ISO-8859-1?Q?T=F6r=F6k_Edwin?=)
Date: Mon, 19 Oct 2009 10:38:01 +0300
Subject: [llvm-commits] patch: preserve branch to unreachable	until
 codegenprep time
In-Reply-To: <7999D788-8D7E-44D4-A759-8DBDB44F3707@apple.com>
References: <4ADAC0C9.6060508@mxc.ca>	<A586E31E-36A3-4C18-A64E-9394FA98AC20@apple.com>
	<7999D788-8D7E-44D4-A759-8DBDB44F3707@apple.com>
Message-ID: <4ADC1759.6010908@gmail.com>

On 2009-10-19 07:47, Chris Lattner wrote:
> On Oct 18, 2009, at 10:11 AM, Dan Gohman wrote:
>
>   
>> On Oct 18, 2009, at 12:16 AM, Nick Lewycky <nicholas at mxc.ca> wrote:
>>
>>     
>>> This patch moves the code that optimizes "branch to unreachable"
>>> from SimplifyCFG to CodeGenPrep. Most of this patch is just updating
>>> the unit tests that use unreachable everywhere. I've done my best to
>>> preserve the intent of the tests.
>>>       
>> I haven't read the patch, but I'm concerned about this interfering
>> with other optimizations.  Deleting the condition can expose further
>> dce and related optimizations.  Also if the branch is a loop exit,
>> deleting it makes it easier to compute a trip count.
>>     
>
> I agree, but I still think this is a good idea.  :)
>
> LSR and other passes can special case this for trip count handling.   
> We can make simplifycfg delete the branch to unreachable if the  
> condition is "too complicated" to be useful for the optimizer.
>
> One area that does concern me is in the various code size evaluations  
> (inlining, jump threading, unswitch,  ...).  This feature will make  
> those completely wrong for code involved in these sorts of checks.
>
>   

How about splitting the SSA value (like SSI does), and attach some
information to it that say:
conditionX is known to be true when this is reached, or conditionX is
known to be false.
Or perhaps a new intrinsic, that is not a terminator, something like this:

...
call @bar(%i)
%conditionX = icmp ult, i32 %i, i32 %n
%conditionY = icmp ult i32 %j, i32 %n
br i1 %conditionY, %tru1, %fals1
%tru1:
br i1 %conditionX, %true, %false
%true:
    unreachable
%false:
  call @foo(%i)
  br %for.cond
%fals1:
  call @bar(%j)
. ..

--becomes-->
call @bar(%i)
%conditionX = icmp ult, i32 %i, i32 %n
%conditionY = icmp ult, i32 %j, i32 %n
br i1 %conditionY, %tru1, %fals1
%tru1:
llvm.assume(@conditionX, 1)
call @foo(%i)
br %for.cond
%fals1:
call @bar(%j)

You can then do all sorts of optimizations knowing that conditionX is
always 1, and even constant propagate that.
Problem is that intrinsic cannot be moved over any branches and other
intrinsics of the same kind.
I'm not sure how that can be modeled, modeling as writes to global
memory may prohibit optimizations, so maybe we should just teach LICM
that this intrinsic is not hoistable, sinkable.
Is there any other pass that moves instructions across BBs?

Best regards,
--Edwin


From daniel at zuster.org  Mon Oct 19 04:18:27 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 09:18:27 -0000
Subject: [llvm-commits] [llvm] r84476 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910190918.n9J9ISTm018439@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 04:18:24 2009
New Revision: 84476

URL: http://llvm.org/viewvc/llvm-project?rev=84476&view=rev
Log:
Regroup NewNightlyTest.pl options

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84476&r1=84475&r2=84476&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 04:18:24 2009
@@ -15,19 +15,52 @@
 # Syntax:   NightlyTest.pl [OPTIONS] [CVSROOT BUILDDIR WEBDIR]
 #   where
 # OPTIONS may include one or more of the following:
+#
+# MAIN OPTIONS:
+#  -config LLVMPATH If specified, use an existing LLVM build and only run and
+#                   report the test information. The LLVMCONFIG argument should
+#                   be the path to the llvm-config executable in the LLVM build.
+#                   This should be the first argument if given. NOT YET
+#                   IMPLEMENTED.
+#  -nickname NAME   The NAME argument specifieds the nickname this script
+#                   will submit to the nightlytest results repository.
+#  -submit-server   Specifies a server to submit the test results too. If this
+#                   option is not specified it defaults to
+#                   llvm.org. This is basically just the address of the
+#                   webserver
+#  -submit-script   Specifies which script to call on the submit server. If
+#                   this option is not specified it defaults to
+#                   /nightlytest/NightlyTestAccept.php. This is basically
+#                   everything after the www.yourserver.org.
+#  -submit-aux      If specified, an auxiliary script to run in addition to the
+#                   normal submit script. The script will be passed the path to
+#                   the "sentdata.txt" file as its sole argument.
+#  -nosubmit        Do not report the test results back to a submit server.
+#
+#
+# BUILD OPTIONS (not used with -config):
 #  -nocheckout      Do not create, checkout, update, or configure
 #                   the source tree.
 #  -noremove        Do not remove the BUILDDIR after it has been built.
 #  -noremoveresults Do not remove the WEBDIR after it has been built.
 #  -nobuild         Do not build llvm. If tests are enabled perform them
 #                   on the llvm build specified in the build directory
-#  -notest          Do not even attempt to run the test programs.
-#  -nodejagnu       Do not run feature or regression tests
-#  -parallel        Run parallel jobs with GNU Make (see -parallel-jobs).
-#  -parallel-jobs   The number of parallel Make jobs to use (default is two).
-#  -with-clang      Checkout Clang source into tools/clang.
 #  -release         Build an LLVM Release version
 #  -release-asserts Build an LLVM ReleaseAsserts version
+#  -disable-bindings     Disable building LLVM bindings.
+#  -cvstag          Check out a specific CVS tag to build LLVM (useful for
+#                   testing release branches)
+#  -usecvs          Check code out from the (old) CVS Repository instead of from
+#                   the standard Subversion repository.
+#  -with-clang      Checkout Clang source into tools/clang.
+#  -compileflags    Next argument specifies extra options passed to make when
+#                   building LLVM.
+#  -use-gmake       Use gmake instead of the default make command to build
+#                   llvm and run tests.
+#
+# TESTING OPTIONS:
+#  -notest          Do not even attempt to run the test programs.
+#  -nodejagnu       Do not run feature or regression tests
 #  -enable-llcbeta  Enable testing of beta features in llc.
 #  -enable-lli      Enable testing of lli (interpreter) features, default is off
 #  -disable-pic	    Disable building with Position Independent Code.
@@ -35,19 +68,25 @@
 #  -disable-jit     Disable JIT tests in the nightly tester.
 #  -disable-cbe     Disable C backend tests in the nightly tester.
 #  -disable-lto     Disable link time optimization.
-#  -disable-bindings     Disable building LLVM bindings.
+#  -test-cflags     Next argument specifies that C compilation options that
+#                   override the default when running the testsuite.
+#  -test-cxxflags   Next argument specifies that C++ compilation options that
+#                   override the default when running the testsuite.
+#  -extraflags      Next argument specifies extra options that are passed to
+#                   compile the tests.
+#  -noexternals     Do not run the external tests (for cases where povray
+#                   or SPEC are not installed)
+#  -with-externals  Specify a directory where the external tests are located.
+#
+# OTHER OPTIONS:
+#  -parallel        Run parallel jobs with GNU Make (see -parallel-jobs).
+#  -parallel-jobs   The number of parallel Make jobs to use (default is two).
 #  -verbose         Turn on some debug output
 #  -debug           Print information useful only to maintainers of this script.
 #  -nice            Checkout/Configure/Build with "nice" to reduce impact
 #                   on busy servers.
 #  -f2c             Next argument specifies path to F2C utility
-#  -nickname        The next argument specifieds the nickname this script
-#                   will submit to the nightlytest results repository.
 #  -gccpath         Path to gcc/g++ used to build LLVM
-#  -cvstag          Check out a specific CVS tag to build LLVM (useful for
-#                   testing release branches)
-#  -usecvs          Check code out from the (old) CVS Repository instead of from
-#                   the standard Subversion repository.
 #  -target          Specify the target triplet
 #  -cflags          Next argument specifies that C compilation options that
 #                   override the default.
@@ -55,33 +94,6 @@
 #                   override the default.
 #  -ldflags         Next argument specifies that linker options that override
 #                   the default.
-#  -test-cflags     Next argument specifies that C compilation options that
-#                   override the default when running the testsuite.
-#  -test-cxxflags   Next argument specifies that C++ compilation options that
-#                   override the default when running the testsuite.
-#  -compileflags    Next argument specifies extra options passed to make when
-#                   building LLVM.
-#  -use-gmake       Use gmake instead of the default make command to build
-#                   llvm and run tests.
-#
-#  ---------------- Options to configure llvm-test ----------------------------
-#  -extraflags      Next argument specifies extra options that are passed to
-#                   compile the tests.
-#  -noexternals     Do not run the external tests (for cases where povray
-#                   or SPEC are not installed)
-#  -with-externals  Specify a directory where the external tests are located.
-#  -submit-server   Specifies a server to submit the test results too. If this
-#                   option is not specified it defaults to
-#                   llvm.org. This is basically just the address of the
-#                   webserver
-#  -submit-script   Specifies which script to call on the submit server. If
-#                   this option is not specified it defaults to
-#                   /nightlytest/NightlyTestAccept.php. This is basically
-#                   everything after the www.yourserver.org.
-#  -submit-aux      If specified, an auxiliary script to run in addition to the
-#                   normal submit script. The script will be passed the path to
-#                   the "sentdata.txt" file as its sole argument.
-#  -nosubmit        Do not report the test results back to a submit server.
 #
 # CVSROOT is the CVS repository from which the tree will be checked out,
 #  specified either in the full :method:user at host:/dir syntax, or
@@ -133,6 +145,7 @@
 # Parse arguments...
 #
 ##############################################################
+$CONFIG_PATH="";
 $CONFIGUREARGS="";
 $nickname="";
 $NOTEST=0;
@@ -150,6 +163,7 @@
   last if /^--$/;  # Stop processing arguments on --
 
   # List command line options here...
+  if (/^-config$/)         { $CONFIG_PATH = "$ARGV[0]"; shift; next; }
   if (/^-nocheckout$/)     { $NOCHECKOUT = 1; next; }
   if (/^-nocvsstats$/)     { $NOCVSSTATS = 1; next; }
   if (/^-noremove$/)       { $NOREMOVE = 1; next; }
@@ -222,6 +236,10 @@
   print "Unknown option: $_ : ignoring!\n";
 }
 
+if (!($CONFIG_PATH eq "")) {
+  die "error: -config mode is not yet implemented,";
+}
+
 if ($ENV{'LLVMGCCDIR'}) {
   $CONFIGUREARGS .= " --with-llvmgccdir=" . $ENV{'LLVMGCCDIR'};
   $LLVMGCCPATH = $ENV{'LLVMGCCDIR'} . '/bin';
@@ -265,7 +283,7 @@
 
 ##############################################################
 #
-#define the file names we'll use
+# Define the file names we'll use
 #
 ##############################################################
 my $Prefix = "$WebDir/$DATE";
@@ -303,6 +321,7 @@
 # Helper functions
 #
 ##############################################################
+
 sub GetDir {
   my $Suffix = shift;
   opendir DH, $WebDir;
@@ -499,10 +518,10 @@
 # to our central server via the post method
 #
 #~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-sub SendData{
+sub SendData {
     $host = $_[0];
     $file = $_[1];
-    $variables=$_[2];
+    $variables = $_[2];
 
     # Write out the "...-sentdata.txt" file.
 
@@ -517,7 +536,7 @@
         system "$SUBMITAUX \"$Prefix-sentdata.txt\"";
     }
 
-    if (!$SUBMIT) { 
+    if (!$SUBMIT) {
         return "Skipped standard submit.\n";
     }
 
@@ -531,9 +550,9 @@
     }
 
     # Send the data to the server.
-    # 
+    #
     # FIXME: This code should be more robust?
-    
+
     $port=80;
     $socketaddr= sockaddr_in $port, inet_aton $host or die "Bad hostname\n";
     socket SOCK, PF_INET, SOCK_STREAM, getprotobyname('tcp') or




From daniel at zuster.org  Mon Oct 19 04:18:38 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 09:18:38 -0000
Subject: [llvm-commits] [llvm] r84477 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910190918.n9J9Icvk018459@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 04:18:37 2009
New Revision: 84477

URL: http://llvm.org/viewvc/llvm-project?rev=84477&view=rev
Log:
NNT: Remove -debug argument, it is unused.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84477&r1=84476&r2=84477&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 04:18:37 2009
@@ -82,7 +82,6 @@
 #  -parallel        Run parallel jobs with GNU Make (see -parallel-jobs).
 #  -parallel-jobs   The number of parallel Make jobs to use (default is two).
 #  -verbose         Turn on some debug output
-#  -debug           Print information useful only to maintainers of this script.
 #  -nice            Checkout/Configure/Build with "nice" to reduce impact
 #                   on busy servers.
 #  -f2c             Next argument specifies path to F2C utility
@@ -194,7 +193,6 @@
   if (/^-test-opts$/)      { $PROGTESTOPTS .= " $ARGV[0]"; shift; next; }
   if (/^-verbose$/)        { $VERBOSE = 1; next; }
   if (/^-teelogs$/)        { $TEELOGS = 1; next; }
-  if (/^-debug$/)          { $DEBUG = 1; next; }
   if (/^-nice$/)           { $NICE = "nice "; next; }
   if (/^-f2c$/)            { $CONFIGUREARGS .= " --with-f2c=$ARGV[0]";
                              shift; next; }




From daniel at zuster.org  Mon Oct 19 04:18:46 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 09:18:46 -0000
Subject: [llvm-commits] [llvm] r84478 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910190918.n9J9IlF6018476@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 04:18:46 2009
New Revision: 84478

URL: http://llvm.org/viewvc/llvm-project?rev=84478&view=rev
Log:
NNT: Remove -usecvs option, this is very old.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84478&r1=84477&r2=84478&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 04:18:46 2009
@@ -50,8 +50,6 @@
 #  -disable-bindings     Disable building LLVM bindings.
 #  -cvstag          Check out a specific CVS tag to build LLVM (useful for
 #                   testing release branches)
-#  -usecvs          Check code out from the (old) CVS Repository instead of from
-#                   the standard Subversion repository.
 #  -with-clang      Checkout Clang source into tools/clang.
 #  -compileflags    Next argument specifies extra options passed to make when
 #                   building LLVM.
@@ -148,7 +146,6 @@
 $CONFIGUREARGS="";
 $nickname="";
 $NOTEST=0;
-$USESVN=1;
 $MAKECMD="make";
 $SUBMITSERVER = "llvm.org";
 $SUBMITSCRIPT = "/nightlytest/NightlyTestAccept.php";
@@ -211,7 +208,6 @@
   else                     { $GCCPATH=""; }
   if (/^-cvstag/)          { $CVSCOOPT .= " -r $ARGV[0]"; shift; next; }
   else                     { $CVSCOOPT="";}
-  if (/^-usecvs/)          { $USESVN = 0; }
   if (/^-target/)          { $CONFIGUREARGS .= " --target=$ARGV[0]";
                              shift; next; }
   if (/^-cflags/)          { $MAKEOPTS = "$MAKEOPTS C.Flags=\'$ARGV[0]\'";
@@ -302,11 +298,7 @@
 
 if ($VERBOSE) {
   print "INITIALIZED\n";
-  if ($USESVN) {
-    print "SVN URL  = $SVNURL\n";
-  } else {
-    print "CVS Root = $CVSRootDir\n";
-  }
+  print "SVN URL  = $SVNURL\n";
   print "COLog    = $COLog\n";
   print "BuildDir = $BuildDir\n";
   print "WebDir   = $WebDir\n";
@@ -614,25 +606,15 @@
 ##############################################################
 if (!$NOCHECKOUT) {
   ChangeDir( $BuildDir, "checkout directory" );
-  if ($USESVN) {
-      my $SVNCMD = "$NICE svn co --non-interactive $SVNURL";
-      my $SVNCMD2 = "$NICE svn co --non-interactive $TestSVNURL";
-      RunLoggedCommand("( time -p $SVNCMD/llvm/trunk llvm; cd llvm/projects ; " .
-                       "$SVNCMD2/test-suite/trunk llvm-test )", $COLog,
-                       "CHECKOUT LLVM");
-      if ($WITHCLANG) {
-        my $SVNCMD = "$NICE svn co --non-interactive $SVNURL/cfe/trunk";
-        RunLoggedCommand("( time -p cd llvm/tools ; $SVNCMD clang )", $COLog,
-                         "CHECKOUT CLANG");
-      }
-  } else {
-    my $CVSOPT = "";
-    $CVSOPT = "-z3" # Use compression if going over ssh.
-      if $CVSRootDir =~ /^:ext:/;
-    my $CVSCMD = "$NICE cvs $CVSOPT -d $CVSRootDir co -P $CVSCOOPT";
-    RunLoggedCommand("( time -p $CVSCMD llvm; cd llvm/projects ; " .
-                     "$CVSCMD llvm-test )", $COLog,
-                     "CHECKOUT LLVM-TEST");
+  my $SVNCMD = "$NICE svn co --non-interactive $SVNURL";
+  my $SVNCMD2 = "$NICE svn co --non-interactive $TestSVNURL";
+  RunLoggedCommand("( time -p $SVNCMD/llvm/trunk llvm; cd llvm/projects ; " .
+                   "$SVNCMD2/test-suite/trunk llvm-test )", $COLog,
+                   "CHECKOUT LLVM");
+  if ($WITHCLANG) {
+      my $SVNCMD = "$NICE svn co --non-interactive $SVNURL/cfe/trunk";
+      RunLoggedCommand("( time -p cd llvm/tools ; $SVNCMD clang )", $COLog,
+                       "CHECKOUT CLANG");
   }
 }
 ChangeDir( $LLVMSrcDir , "llvm source directory") ;
@@ -651,13 +633,8 @@
 
 my $NumFilesInCVS = 0;
 my $NumDirsInCVS  = 0;
-if ($USESVN) {
-  $NumFilesInCVS = `egrep '^A' $COLog | wc -l` + 0;
-  $NumDirsInCVS  = `sed -e 's#/[^/]*\$##' $COLog | sort | uniq | wc -l` + 0;
-} else {
-  $NumFilesInCVS = `egrep '^U' $COLog | wc -l` + 0;
-  $NumDirsInCVS  = `egrep '^cvs (checkout|server|update):' $COLog | wc -l` + 0;
-}
+$NumFilesInCVS = `egrep '^A' $COLog | wc -l` + 0;
+$NumDirsInCVS  = `sed -e 's#/[^/]*\$##' $COLog | sort | uniq | wc -l` + 0;
 
 ##############################################################
 #
@@ -690,89 +667,49 @@
 if (!$NOCVSSTATS) {
   if ($VERBOSE) { print "CHANGE HISTORY ANALYSIS STAGE\n"; }
 
-  if ($USESVN) {
-    @SVNHistory = split /<logentry/, `svn log --non-interactive --xml --verbose -r{$DATE}:HEAD`;
-    # Skip very first entry because it is the XML header cruft
-    shift @SVNHistory;
-    my $Now = time();
-    foreach $Record (@SVNHistory) {
+  @SVNHistory = split /<logentry/, `svn log --non-interactive --xml --verbose -r{$DATE}:HEAD`;
+  # Skip very first entry because it is the XML header cruft
+  shift @SVNHistory;
+  my $Now = time();
+  foreach $Record (@SVNHistory) {
       my @Lines = split "\n", $Record;
       my ($Author, $Date, $Revision);
       # Get the date and see if its one we want to process.
       my ($Year, $Month, $Day, $Hour, $Min, $Sec);
       if ($Lines[3] =~ /<date>(\d{4})-(\d{2})-(\d{2})T(\d{2}):(\d{2}):(\d{2})/){
-        $Year = $1; $Month = $2; $Day = $3; $Hour = $4; $Min = $5; $Sec = $6;
+          $Year = $1; $Month = $2; $Day = $3; $Hour = $4; $Min = $5; $Sec = $6;
       }
       my $Then = ConvertToSeconds($Sec, $Min, $Hour, $Day, $Month, $Year);
       # Get the current date and compute when "yesterday" is.
       my ($NSec, $NMin, $NHour, $NDay, $NMon, $NYear) = gmtime();
       my $Now = ConvertToSeconds( $NSec, $NMin, $NHour, $NDay, $NMon, $NYear);
       if (($Now - 24*60*60) > $Then) {
-        next;
+          next;
       }
       if ($Lines[1] =~ /   revision="([0-9]*)">/) {
-        $Revision = $1;
+          $Revision = $1;
       }
       if ($Lines[2] =~ /<author>([^<]*)<\/author>/) {
-        $Author = $1;
+          $Author = $1;
       }
       $UsersCommitted{$Author} = 1;
       $Date = $Year . "-" . $Month . "-" . $Day;
       $Time = $Hour . ":" . $Min . ":" . $Sec;
       print "Rev: $Revision, Author: $Author, Date: $Date, Time: $Time\n";
       for ($i = 6; $i < $#Lines; $i += 2 ) {
-        if ($Lines[$i] =~ /^   action="(.)">([^<]*)</) {
-          if ($1 == "A") {
-            $AddedFiles{$2} = 1;
-          } elsif ($1 == 'D') {
-            $RemovedFiles{$2} = 1;
-          } elsif ($1 == 'M' || $1 == 'R' || $1 == 'C') {
-            $ModifiedFiles{$2} = 1;
-          } else {
-            print "UNMATCHABLE: $Lines[$i]\n";
+          if ($Lines[$i] =~ /^   action="(.)">([^<]*)</) {
+              if ($1 == "A") {
+                  $AddedFiles{$2} = 1;
+              } elsif ($1 == 'D') {
+                  $RemovedFiles{$2} = 1;
+              } elsif ($1 == 'M' || $1 == 'R' || $1 == 'C') {
+                  $ModifiedFiles{$2} = 1;
+              } else {
+                  print "UNMATCHABLE: $Lines[$i]\n";
+              }
           }
-        }
       }
-    }
-  } else {
-    @CVSHistory = split "\n", `cvs history -D '1 day ago' -a -xAMROCGUW`;
-#print join "\n", @CVSHistory; print "\n";
-
-    my $DateRE = '[-/:0-9 ]+\+[0-9]+';
-
-# Loop over every record from the CVS history, filling in the hashes.
-    foreach $File (@CVSHistory) {
-        my ($Type, $Date, $UID, $Rev, $Filename);
-        if ($File =~ /([AMRUGC]) ($DateRE) ([^ ]+) +([^ ]+) +([^ ]+) +([^ ]+)/) {
-            ($Type, $Date, $UID, $Rev, $Filename) = ($1, $2, $3, $4, "$6/$5");
-        } elsif ($File =~ /([W]) ($DateRE) ([^ ]+)/) {
-            ($Type, $Date, $UID, $Rev, $Filename) = ($1, $2, $3, "", "");
-        } elsif ($File =~ /([O]) ($DateRE) ([^ ]+) +([^ ]+)/) {
-            ($Type, $Date, $UID, $Rev, $Filename) = ($1, $2, $3, "", "$4/");
-        } else {
-            print "UNMATCHABLE: $File\n";
-            next;
-        }
-        # print "$File\nTy = $Type Date = '$Date' UID=$UID Rev=$Rev File = '$Filename'\n";
-
-        if ($Filename =~ /^llvm/) {
-            if ($Type eq 'M') {        # Modified
-                $ModifiedFiles{$Filename} = 1;
-                $UsersCommitted{$UID} = 1;
-            } elsif ($Type eq 'A') {   # Added
-                $AddedFiles{$Filename} = 1;
-                $UsersCommitted{$UID} = 1;
-            } elsif ($Type eq 'R') {   # Removed
-                $RemovedFiles{$Filename} = 1;
-                $UsersCommitted{$UID} = 1;
-            } else {
-                $UsersUpdated{$UID} = 1;
-            }
-        }
-    }
-
-    my $TestError = 1;
-  } #$USESVN
+  }
 }#!NOCVSSTATS
 
 my $CVSAddedFiles = join "\n", sort keys %AddedFiles;




From daniel at zuster.org  Mon Oct 19 04:18:59 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 09:18:59 -0000
Subject: [llvm-commits] [llvm] r84479 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910190919.n9J9J09x018492@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 04:18:54 2009
New Revision: 84479

URL: http://llvm.org/viewvc/llvm-project?rev=84479&view=rev
Log:
NNT: Remove now-unused -cvstag argument and CVSROOT code.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84479&r1=84478&r2=84479&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 04:18:54 2009
@@ -48,8 +48,6 @@
 #  -release         Build an LLVM Release version
 #  -release-asserts Build an LLVM ReleaseAsserts version
 #  -disable-bindings     Disable building LLVM bindings.
-#  -cvstag          Check out a specific CVS tag to build LLVM (useful for
-#                   testing release branches)
 #  -with-clang      Checkout Clang source into tools/clang.
 #  -compileflags    Next argument specifies extra options passed to make when
 #                   building LLVM.
@@ -92,12 +90,10 @@
 #  -ldflags         Next argument specifies that linker options that override
 #                   the default.
 #
-# CVSROOT is the CVS repository from which the tree will be checked out,
-#  specified either in the full :method:user at host:/dir syntax, or
-#  just /dir if using a local repo.
+# CVSROOT is ignored, it is passed for backwards compatibility.
 # BUILDDIR is the directory where sources for this test run will be checked out
 #  AND objects for this test run will be built. This directory MUST NOT
-#  exist before the script is run; it will be created by the cvs checkout
+#  exist before the script is run; it will be created by the svn checkout
 #  process and erased (unless -noremove is specified; see above.)
 # WEBDIR is the directory into which the test results web page will be written,
 #  AND in which the "index.html" is assumed to be a symlink to the most recent
@@ -115,8 +111,6 @@
 $SVNURL        = 'http://llvm.org/svn/llvm-project' unless $SVNURL;
 my $TestSVNURL = $ENV{"TestSVNURL"};
 $TestSVNURL    = 'http://llvm.org/svn/llvm-project' unless $TestSVNURL;
-my $CVSRootDir = $ENV{'CVSROOT'};
-$CVSRootDir    = "/home/vadve/shared/PublicCVS" unless $CVSRootDir;
 my $BuildDir   = $ENV{'BUILDDIR'};
 $BuildDir      = "$HOME/buildtest" unless $BuildDir;
 my $WebDir     = $ENV{'WEBDIR'};
@@ -206,8 +200,6 @@
                              " CC=$ARGV[0]/gcc CXX=$ARGV[0]/g++";
                              $GCCPATH=$ARGV[0]; shift;  next; }
   else                     { $GCCPATH=""; }
-  if (/^-cvstag/)          { $CVSCOOPT .= " -r $ARGV[0]"; shift; next; }
-  else                     { $CVSCOOPT="";}
   if (/^-target/)          { $CONFIGUREARGS .= " --target=$ARGV[0]";
                              shift; next; }
   if (/^-cflags/)          { $MAKEOPTS = "$MAKEOPTS C.Flags=\'$ARGV[0]\'";
@@ -254,16 +246,14 @@
 }
 
 if (@ARGV == 3) {
-  $CVSRootDir = $ARGV[0];
+  # ARGV[0] used to be the CVS root, ignored for backward compatibility.
   $BuildDir   = $ARGV[1];
   $WebDir     = $ARGV[2];
 }
 
-if ($CVSRootDir eq "" or
-    $BuildDir   eq "" or
+if ($BuildDir   eq "" or
     $WebDir     eq "") {
-  die("please specify a cvs root directory, a build directory, and a ".
-       "web directory");
+  die("please specify a build directory, and a web directory");
  }
 
 if ($nickname eq "") {
@@ -577,7 +567,7 @@
 
 ##############################################################
 #
-# Create the CVS repository directory
+# Create the source repository directory
 #
 ##############################################################
 if (!$NOCHECKOUT) {
@@ -601,7 +591,7 @@
 
 ##############################################################
 #
-# Check out the llvm tree, using either SVN or CVS
+# Check out the llvm tree with SVN
 #
 ##############################################################
 if (!$NOCHECKOUT) {
@@ -621,7 +611,7 @@
 
 ##############################################################
 #
-# Get some static statistics about the current state of CVS
+# Get some static statistics about the current source code
 #
 # This can probably be put on the server side
 #
@@ -1110,7 +1100,7 @@
 
 ##############################################################
 #
-# Remove the cvs tree...
+# Remove the source tree...
 #
 ##############################################################
 system ( "$NICE rm -rf $BuildDir")




From daniel at zuster.org  Mon Oct 19 04:19:11 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 09:19:11 -0000
Subject: [llvm-commits] [llvm] r84480 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910190919.n9J9JCDj018518@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 04:19:09 2009
New Revision: 84480

URL: http://llvm.org/viewvc/llvm-project?rev=84480&view=rev
Log:
NNT: Remove "CVS Stats", this isn't particularly useful and can be better done by the server or user.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84480&r1=84479&r2=84480&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 04:19:09 2009
@@ -155,7 +155,6 @@
   # List command line options here...
   if (/^-config$/)         { $CONFIG_PATH = "$ARGV[0]"; shift; next; }
   if (/^-nocheckout$/)     { $NOCHECKOUT = 1; next; }
-  if (/^-nocvsstats$/)     { $NOCVSSTATS = 1; next; }
   if (/^-noremove$/)       { $NOREMOVE = 1; next; }
   if (/^-noremoveatend$/)  { $NOREMOVEATEND = 1; next; }
   if (/^-noremoveresults$/){ $NOREMOVERESULTS = 1; next; }
@@ -621,93 +620,6 @@
 my $CheckoutTime_Sys = GetRegex "([0-9.]+)", `grep '^sys' $COLog`;
 my $CheckoutTime_CPU = $CVSCheckoutTime_User + $CVSCheckoutTime_Sys;
 
-my $NumFilesInCVS = 0;
-my $NumDirsInCVS  = 0;
-$NumFilesInCVS = `egrep '^A' $COLog | wc -l` + 0;
-$NumDirsInCVS  = `sed -e 's#/[^/]*\$##' $COLog | sort | uniq | wc -l` + 0;
-
-##############################################################
-#
-# Extract some information from the CVS history... use a hash so no duplicate
-# stuff is stored. This gets the history from the previous days worth
-# of cvs activity and parses it.
-#
-##############################################################
-
-# This just computes a reasonably accurate #of seconds since 2000. It doesn't
-# have to be perfect as its only used for comparing date ranges within a couple
-# of days.
-sub ConvertToSeconds {
-  my ($sec, $min, $hour, $day, $mon, $yr) = @_;
-  my $Result = ($yr - 2000) * 12;
-  $Result += $mon;
-  $Result *= 31;
-  $Result += $day;
-  $Result *= 24;
-  $Result += $hour;
-  $Result *= 60;
-  $Result += $min;
-  $Result *= 60;
-  $Result += $sec;
-  return $Result;
-}
-
-my (%AddedFiles, %ModifiedFiles, %RemovedFiles, %UsersCommitted, %UsersUpdated);
-
-if (!$NOCVSSTATS) {
-  if ($VERBOSE) { print "CHANGE HISTORY ANALYSIS STAGE\n"; }
-
-  @SVNHistory = split /<logentry/, `svn log --non-interactive --xml --verbose -r{$DATE}:HEAD`;
-  # Skip very first entry because it is the XML header cruft
-  shift @SVNHistory;
-  my $Now = time();
-  foreach $Record (@SVNHistory) {
-      my @Lines = split "\n", $Record;
-      my ($Author, $Date, $Revision);
-      # Get the date and see if its one we want to process.
-      my ($Year, $Month, $Day, $Hour, $Min, $Sec);
-      if ($Lines[3] =~ /<date>(\d{4})-(\d{2})-(\d{2})T(\d{2}):(\d{2}):(\d{2})/){
-          $Year = $1; $Month = $2; $Day = $3; $Hour = $4; $Min = $5; $Sec = $6;
-      }
-      my $Then = ConvertToSeconds($Sec, $Min, $Hour, $Day, $Month, $Year);
-      # Get the current date and compute when "yesterday" is.
-      my ($NSec, $NMin, $NHour, $NDay, $NMon, $NYear) = gmtime();
-      my $Now = ConvertToSeconds( $NSec, $NMin, $NHour, $NDay, $NMon, $NYear);
-      if (($Now - 24*60*60) > $Then) {
-          next;
-      }
-      if ($Lines[1] =~ /   revision="([0-9]*)">/) {
-          $Revision = $1;
-      }
-      if ($Lines[2] =~ /<author>([^<]*)<\/author>/) {
-          $Author = $1;
-      }
-      $UsersCommitted{$Author} = 1;
-      $Date = $Year . "-" . $Month . "-" . $Day;
-      $Time = $Hour . ":" . $Min . ":" . $Sec;
-      print "Rev: $Revision, Author: $Author, Date: $Date, Time: $Time\n";
-      for ($i = 6; $i < $#Lines; $i += 2 ) {
-          if ($Lines[$i] =~ /^   action="(.)">([^<]*)</) {
-              if ($1 == "A") {
-                  $AddedFiles{$2} = 1;
-              } elsif ($1 == 'D') {
-                  $RemovedFiles{$2} = 1;
-              } elsif ($1 == 'M' || $1 == 'R' || $1 == 'C') {
-                  $ModifiedFiles{$2} = 1;
-              } else {
-                  print "UNMATCHABLE: $Lines[$i]\n";
-              }
-          }
-      }
-  }
-}#!NOCVSSTATS
-
-my $CVSAddedFiles = join "\n", sort keys %AddedFiles;
-my $CVSModifiedFiles = join "\n", sort keys %ModifiedFiles;
-my $CVSRemovedFiles = join "\n", sort keys %RemovedFiles;
-my $UserCommitList = join "\n", sort keys %UsersCommitted;
-my $UserUpdateList = join "\n", sort keys %UsersUpdated;
-
 ##############################################################
 #
 # Build the entire tree, saving build messages to the build log
@@ -727,15 +639,6 @@
 # Get some statistics about the build...
 #
 ##############################################################
-#this can de done on server
-#my @Linked = split '\n', `grep Linking $BuildLog`;
-#my $NumExecutables = scalar(grep(/executable/, @Linked));
-#my $NumLibraries   = scalar(grep(!/executable/, @Linked));
-#my $NumObjects     = `grep ']\: Compiling ' $BuildLog | wc -l` + 0;
-
-# Get the number of lines of source code. Must be here after the build is done
-# because countloc.sh uses the llvm-config script which must be built.
-my $LOC = `utils/countloc.sh -topdir $LLVMSrcDir`;
 
 # Get the time taken by the configure script
 my $ConfigTimeU = GetRegexNum "^user", 0, "([0-9.]+)", "$BuildLog";
@@ -1058,14 +961,14 @@
   'buildtime_wall' => $BuildWallTime,
   'buildtime_cpu' => $BuildTime,
   'warnings' => $WarningsFile,
-  'cvsusercommitlist' => $UserCommitList,
-  'cvsuserupdatelist' => $UserUpdateList,
-  'cvsaddedfiles' => $CVSAddedFiles,
-  'cvsmodifiedfiles' => $CVSModifiedFiles,
-  'cvsremovedfiles' => $CVSRemovedFiles,
-  'lines_of_code' => $LOC,
-  'cvs_file_count' => $NumFilesInCVS,
-  'cvs_dir_count' => $NumDirsInCVS,
+  'cvsusercommitlist' => "",
+  'cvsuserupdatelist' => "",
+  'cvsaddedfiles' => "",
+  'cvsmodifiedfiles' => "",
+  'cvsremovedfiles' => "",
+  'lines_of_code' => "",
+  'cvs_file_count' => 0,
+  'cvs_dir_count' => 0,
   'buildstatus' => $BuildStatus,
   'singlesource_programstable' => $SingleSourceProgramsTable,
   'multisource_programstable' => $MultiSourceProgramsTable,




From daniel at zuster.org  Mon Oct 19 04:19:19 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 09:19:19 -0000
Subject: [llvm-commits] [llvm] r84481 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910190919.n9J9JJ10018533@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 04:19:19 2009
New Revision: 84481

URL: http://llvm.org/viewvc/llvm-project?rev=84481&view=rev
Log:
NNT: Remove hard coded BuildDir and WebDir, users should have to specify these.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84481&r1=84480&r2=84481&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 04:19:19 2009
@@ -112,9 +112,7 @@
 my $TestSVNURL = $ENV{"TestSVNURL"};
 $TestSVNURL    = 'http://llvm.org/svn/llvm-project' unless $TestSVNURL;
 my $BuildDir   = $ENV{'BUILDDIR'};
-$BuildDir      = "$HOME/buildtest" unless $BuildDir;
 my $WebDir     = $ENV{'WEBDIR'};
-$WebDir        = "$HOME/cvs/testresults-X86" unless $WebDir;
 
 my $LLVMSrcDir   = $ENV{'LLVMSRCDIR'};
 $LLVMSrcDir    = "$BuildDir/llvm" unless $LLVMSrcDir;
@@ -221,10 +219,6 @@
   print "Unknown option: $_ : ignoring!\n";
 }
 
-if (!($CONFIG_PATH eq "")) {
-  die "error: -config mode is not yet implemented,";
-}
-
 if ($ENV{'LLVMGCCDIR'}) {
   $CONFIGUREARGS .= " --with-llvmgccdir=" . $ENV{'LLVMGCCDIR'};
   $LLVMGCCPATH = $ENV{'LLVMGCCDIR'} . '/bin';
@@ -237,14 +231,15 @@
   $CONFIGUREARGS .= " --enable-jit";
 }
 
-if (@ARGV != 0 and @ARGV != 3 and $VERBOSE) {
-  foreach $x (@ARGV) {
-    print "$x\n";
-  }
-  print "Must specify 0 or 3 options!";
+if (@ARGV != 0 and @ARGV != 3) {
+  die "error: must specify 0 or 3 options!";
 }
 
 if (@ARGV == 3) {
+  if ($CONFIG_PATH ne "") {
+      die "error: arguments are unsupported in -config mode,";
+  }
+
   # ARGV[0] used to be the CVS root, ignored for backward compatibility.
   $BuildDir   = $ARGV[1];
   $WebDir     = $ARGV[2];
@@ -264,6 +259,10 @@
   $BUILDTYPE = "debug";
 }
 
+if ($CONFIG_PATH ne "") {
+  die "error: -config mode is not yet implemented,";
+}
+
 ##############################################################
 #
 # Define the file names we'll use




From daniel at zuster.org  Mon Oct 19 04:19:33 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 09:19:33 -0000
Subject: [llvm-commits] [llvm] r84482 -
	/llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp
Message-ID: <200910190919.n9J9JXAV018581@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 04:19:32 2009
New Revision: 84482

URL: http://llvm.org/viewvc/llvm-project?rev=84482&view=rev
Log:
Also check for __POWERPC__ when skipping these tests.

Modified:
    llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp

Modified: llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp?rev=84482&r1=84481&r2=84482&view=diff

==============================================================================
--- llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp (original)
+++ llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp Mon Oct 19 04:19:32 2009
@@ -166,7 +166,7 @@
   EXPECT_EQ(8, TestFunctionPtr());
 }
 
-#if !defined(__arm__) && !defined(__ppc__)
+#if !defined(__arm__) && !defined(__POWERPC__) && !defined(__ppc__)
 // Test a function C which calls A and B which call each other.
 TEST_F(JITTest, NonLazyCompilationStillNeedsStubs) {
   TheJIT->DisableLazyCompilation();




From edwintorok at gmail.com  Mon Oct 19 06:01:00 2009
From: edwintorok at gmail.com (Torok Edwin)
Date: Mon, 19 Oct 2009 11:01:00 -0000
Subject: [llvm-commits] [llvm] r84485 - in /llvm/trunk:
 lib/Target/X86/X86Instr64bit.td
 test/CodeGen/X86/2009-10-19-atomic-cmp-eflags.ll
Message-ID: <200910191101.n9JB11v4022109@zion.cs.uiuc.edu>

Author: edwin
Date: Mon Oct 19 06:00:58 2009
New Revision: 84485

URL: http://llvm.org/viewvc/llvm-project?rev=84485&view=rev
Log:
Fix PR5247, "lock addq" pattern (and other atomics), it DOES modify EFLAGS.
LLC was scheduling compares before the adds causing wrong branches to be taken
in programs, resulting in misoptimized code wherever atomic adds where used.

Added:
    llvm/trunk/test/CodeGen/X86/2009-10-19-atomic-cmp-eflags.ll
Modified:
    llvm/trunk/lib/Target/X86/X86Instr64bit.td

Modified: llvm/trunk/lib/Target/X86/X86Instr64bit.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Instr64bit.td?rev=84485&r1=84484&r2=84485&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86Instr64bit.td (original)
+++ llvm/trunk/lib/Target/X86/X86Instr64bit.td Mon Oct 19 06:00:58 2009
@@ -1515,6 +1515,7 @@
 }
 
 // Optimized codegen when the non-memory output is not used.
+let Defs = [EFLAGS] in {
 // FIXME: Use normal add / sub instructions and add lock prefix dynamically.
 def LOCK_ADD64mr : RI<0x03, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src2),
                       "lock\n\t"
@@ -1544,7 +1545,7 @@
 def LOCK_DEC64m : RI<0xFF, MRM1m, (outs), (ins i64mem:$dst),
                       "lock\n\t"
                       "dec{q}\t$dst", []>, LOCK;
-
+}
 // Atomic exchange, and, or, xor
 let Constraints = "$val = $dst", Defs = [EFLAGS],
                   usesCustomDAGSchedInserter = 1 in {

Added: llvm/trunk/test/CodeGen/X86/2009-10-19-atomic-cmp-eflags.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-10-19-atomic-cmp-eflags.ll?rev=84485&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/X86/2009-10-19-atomic-cmp-eflags.ll (added)
+++ llvm/trunk/test/CodeGen/X86/2009-10-19-atomic-cmp-eflags.ll Mon Oct 19 06:00:58 2009
@@ -0,0 +1,69 @@
+; RUN: llvm-as <%s | llc | FileCheck %s
+; PR 5247
+; check that cmp is not scheduled before the add
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
+target triple = "x86_64-unknown-linux-gnu"
+
+ at .str76843 = external constant [45 x i8]          ; <[45 x i8]*> [#uses=1]
+ at __profiling_callsite_timestamps_live = external global [1216 x i64] ; <[1216 x i64]*> [#uses=2]
+
+define i32 @cl_init(i32 %initoptions) nounwind {
+entry:
+  %retval.i = alloca i32                          ; <i32*> [#uses=3]
+  %retval = alloca i32                            ; <i32*> [#uses=2]
+  %initoptions.addr = alloca i32                  ; <i32*> [#uses=2]
+  tail call void asm sideeffect "cpuid", "~{ax},~{bx},~{cx},~{dx},~{memory},~{dirflag},~{fpsr},~{flags}"() nounwind
+  %0 = tail call i64 @llvm.readcyclecounter() nounwind ; <i64> [#uses=1]
+  store i32 %initoptions, i32* %initoptions.addr
+  %1 = bitcast i32* %initoptions.addr to { }*     ; <{ }*> [#uses=0]
+  call void asm sideeffect "cpuid", "~{ax},~{bx},~{cx},~{dx},~{memory},~{dirflag},~{fpsr},~{flags}"() nounwind
+  %2 = call i64 @llvm.readcyclecounter() nounwind ; <i64> [#uses=1]
+  %call.i = call i32 @lt_dlinit() nounwind        ; <i32> [#uses=1]
+  %tobool.i = icmp ne i32 %call.i, 0              ; <i1> [#uses=1]
+  br i1 %tobool.i, label %if.then.i, label %if.end.i
+
+if.then.i:                                        ; preds = %entry
+  %call1.i = call i32 @warn_dlerror(i8* getelementptr inbounds ([45 x i8]* @.str76843, i32 0, i32 0)) nounwind ; <i32> [#uses=0]
+  store i32 -1, i32* %retval.i
+  br label %lt_init.exit
+
+if.end.i:                                         ; preds = %entry
+  store i32 0, i32* %retval.i
+  br label %lt_init.exit
+
+lt_init.exit:                                     ; preds = %if.end.i, %if.then.i
+  %3 = load i32* %retval.i                        ; <i32> [#uses=1]
+  call void asm sideeffect "cpuid", "~{ax},~{bx},~{cx},~{dx},~{memory},~{dirflag},~{fpsr},~{flags}"() nounwind
+  %4 = call i64 @llvm.readcyclecounter() nounwind ; <i64> [#uses=1]
+  %5 = sub i64 %4, %2                             ; <i64> [#uses=1]
+  %6 = call i64 @llvm.atomic.load.add.i64.p0i64(i64* getelementptr inbounds ([1216 x i64]* @__profiling_callsite_timestamps_live, i32 0, i32 51), i64 %5) nounwind ; <i64> [#uses=0]
+;CHECK: lock
+;CHECK-NEXT: {{xadd|addq}} %rdx, __profiling_callsite_timestamps_live
+;CHECK-NEXT: cmpl $0,
+;CHECK-NEXT: jne
+  %cmp = icmp eq i32 %3, 0                        ; <i1> [#uses=1]
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %lt_init.exit
+  call void @cli_rarload()
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %lt_init.exit
+  store i32 0, i32* %retval
+  %7 = load i32* %retval                          ; <i32> [#uses=1]
+  tail call void asm sideeffect "cpuid", "~{ax},~{bx},~{cx},~{dx},~{memory},~{dirflag},~{fpsr},~{flags}"() nounwind
+  %8 = tail call i64 @llvm.readcyclecounter() nounwind ; <i64> [#uses=1]
+  %9 = sub i64 %8, %0                             ; <i64> [#uses=1]
+  %10 = call i64 @llvm.atomic.load.add.i64.p0i64(i64* getelementptr inbounds ([1216 x i64]* @__profiling_callsite_timestamps_live, i32 0, i32 50), i64 %9) ; <i64> [#uses=0]
+  ret i32 %7
+}
+
+declare void @cli_rarload() nounwind
+
+declare i32 @lt_dlinit()
+
+declare i32 @warn_dlerror(i8*) nounwind
+
+declare i64 @llvm.atomic.load.add.i64.p0i64(i64* nocapture, i64) nounwind
+
+declare i64 @llvm.readcyclecounter() nounwind




From daniel at zuster.org  Mon Oct 19 07:22:37 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 05:22:37 -0700
Subject: [llvm-commits] [llvm] r84327 -
	/llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp
In-Reply-To: <5671CCF8-74F5-4544-893B-711524124B41@apple.com>
References: <200910170933.n9H9XeSH030932@zion.cs.uiuc.edu>
	<B4FE2DBB-405E-419B-8C8C-255D4203BEA4@apple.com>
	<5671CCF8-74F5-4544-893B-711524124B41@apple.com>
Message-ID: <6a8523d60910190522w2b8bc18eu8af2b5c2b0bee681@mail.gmail.com>

On Sun, Oct 18, 2009 at 8:36 PM, Evan Cheng <evan.cheng at apple.com> wrote:
>
> On Oct 17, 2009, at 10:44 AM, Chris Lattner wrote:
>
>>
>> On Oct 17, 2009, at 2:33 AM, Daniel Dunbar wrote:
>>
>>> Author: ddunbar
>>> Date: Sat Oct 17 04:33:00 2009
>>> New Revision: 84327
>>>
>>> URL: http://llvm.org/viewvc/llvm-project?rev=84327&view=rev
>>> Log:
>>> Suppress -Asserts warning.
>>
>> Despite the name, I think that AnalyzeBranch does mutate the code in
>> some cases on some targets. ?Is this true?
>>
>
> Not in this case since AllowModify is false.
>
> virtual bool AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? MachineBasicBlock *&FBB,
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? SmallVectorImpl<MachineOperand> &Cond,
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? bool AllowModify = false) const {

Ok. Please put it back inside the assert if it doesn't do anything
else and you don't want it called in a -Asserts build, I just wanted
the build warning to go away.

 - Daniel

> Evan
>
>> -Chris
>>
>>>
>>> Modified:
>>> ?llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp
>>>
>>> Modified: llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp
>>> URL:
>>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp?rev=84327&r1=84326&r2=84327&view=diff
>>>
>>> =
>>> =
>>> =
>>> =
>>> =
>>> =
>>> =
>>> =
>>> ======================================================================
>>> --- llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp (original)
>>> +++ llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp Sat Oct 17 04:33:00
>>> 2009
>>> @@ -153,8 +153,8 @@
>>>
>>> ?MachineBasicBlock *TBB = 0, *FBB = 0;
>>> ?SmallVector<MachineOperand, 4> Cond;
>>> - ?bool B = TII->AnalyzeBranch(*MBB, TBB, FBB, Cond);
>>> - ?assert(!B && "UpdateTerminators requires analyzable
>>> predecessors!");
>>> + ?assert(!TII->AnalyzeBranch(*MBB, TBB, FBB, Cond) &&
>>> + ? ? ? ? "UpdateTerminators requires analyzable predecessors!");
>>> ?if (Cond.empty()) {
>>> ? if (TBB) {
>>> ? ? // The block has an unconditional branch. If its successor is
>>> now
>>>
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>



From daniel at zuster.org  Mon Oct 19 07:47:06 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 12:47:06 -0000
Subject: [llvm-commits] [test-suite] r84486 -
	/test-suite/trunk/Makefile.programs
Message-ID: <200910191247.n9JCl7PO026089@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 07:47:03 2009
New Revision: 84486

URL: http://llvm.org/viewvc/llvm-project?rev=84486&view=rev
Log:
Add ENABLE_PARALLEL_REPORT which prevents the makefiles from forcing a serial build.

Modified:
    test-suite/trunk/Makefile.programs

Modified: test-suite/trunk/Makefile.programs
URL: http://llvm.org/viewvc/llvm-project/test-suite/trunk/Makefile.programs?rev=84486&r1=84485&r2=84486&view=diff

==============================================================================
--- test-suite/trunk/Makefile.programs (original)
+++ test-suite/trunk/Makefile.programs Mon Oct 19 07:47:03 2009
@@ -788,8 +788,12 @@
 #
 GENERATEREPORT := $(PROJ_SRC_ROOT)/GenerateReport.pl
 
+ifndef ENABLE_PARALLEL_REPORT
+FORCE_SERIAL_ARG := -j1
+endif
+
 report.$(TEST).raw.out: $(REPORT_DEPENDENCIES) $(TestMakefile)
-	$(MAKE) -j1 TEST=$(TEST) 2>&1 | tee $@
+	$(MAKE) $(FORCE_SERIAL_ARG) TEST=$(TEST) 2>&1 | tee $@
 
 ifneq ($(TestReport),)
 report.$(TEST).txt: report.$(TEST).raw.out $(TestReport) $(GENERATEREPORT)




From daniel at zuster.org  Mon Oct 19 08:19:54 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:19:54 -0000
Subject: [llvm-commits] [llvm] r84487 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191319.n9JDJsWF027189@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:19:53 2009
New Revision: 84487

URL: http://llvm.org/viewvc/llvm-project?rev=84487&view=rev
Log:
NNT: Remove code to track build warnings, the buildbots cover this.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84487&r1=84486&r2=84487&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:19:53 2009
@@ -278,7 +278,7 @@
 my $DejagnuSum = "$Prefix-Dejagnu-testrun.sum";
 my $DejagnuTestsLog = "$Prefix-DejagnuTests-Log.txt";
 if (! -d $WebDir) {
-  mkdir $WebDir, 0777;
+  mkdir $WebDir, 0777 or die "Unable to create web directory: '$WebDir'.";
   if($VERBOSE){
     warn "$WebDir did not exist; creating it.\n";
   }
@@ -347,28 +347,6 @@
 }
 
 #~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-#
-# DiffFiles - Diff the current version of the file against the last version of
-# the file, reporting things added and removed.  This is used to report, for
-# example, added and removed warnings.  This returns a pair (added, removed)
-#
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-sub DiffFiles {
-  my $Suffix = shift;
-  my @Others = GetDir $Suffix;
-  if (@Others == 0) {  # No other files?  We added all entries...
-    return (`cat $WebDir/$DATE$Suffix`, "");
-  }
-# Diff the files now...
-  my @Diffs = split "\n", `diff $WebDir/$DATE$Suffix $WebDir/$Others[0]`;
-  my $Added   = join "\n", grep /^</, @Diffs;
-  my $Removed = join "\n", grep /^>/, @Diffs;
-  $Added =~ s/^< //gm;
-  $Removed =~ s/^> //gm;
-  return ($Added, $Removed);
-}
-
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 #~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 sub GetRegex {   # (Regex with ()'s, value)
   $_[1] =~ /$_[0]/m;
@@ -723,40 +701,8 @@
 $DejagnuTime     = "0.0" unless $DejagnuTime;
 $DejagnuWallTime = "0.0" unless $DejagnuWallTime;
 
-##############################################################
-#
-# Get warnings from the build
-#
-##############################################################
 if (!$NODEJAGNU) {
   if ( $VERBOSE ) { print "BUILD INFORMATION COLLECTION STAGE\n"; }
-  my @Warn = split "\n", `egrep 'warning:|Entering dir' $BuildLog`;
-  my @Warnings;
-  my $CurDir = "";
-
-  foreach $Warning (@Warn) {
-    if ($Warning =~ m/Entering directory \`([^\`]+)\'/) {
-      $CurDir = $1;                 # Keep track of directory warning is in...
-      # Remove buildir prefix if included
-      if ($CurDir =~ m#$LLVMSrcDir/(.*)#) { $CurDir = $1; }
-    } else {
-      push @Warnings, "$CurDir/$Warning";     # Add directory to warning...
-    }
-  }
-  my $WarningsFile =  join "\n", @Warnings;
-  $WarningsFile =~ s/:[0-9]+:/::/g;
-
-  # Emit the warnings file, so we can diff...
-  WriteFile "$WebDir/$DATE-Warnings.txt", $WarningsFile . "\n";
-  my ($WarningsAdded, $WarningsRemoved) = DiffFiles "-Warnings.txt";
-
-  # Output something to stdout if something has changed
-  #print "ADDED   WARNINGS:\n$WarningsAdded\n\n" if (length $WarningsAdded);
-  #print "REMOVED WARNINGS:\n$WarningsRemoved\n\n" if (length $WarningsRemoved);
-
-  #my @TmpWarningsAdded = split "\n", $WarningsAdded; ~PJ on upgrade
-  #my @TmpWarningsRemoved = split "\n", $WarningsRemoved; ~PJ on upgrade
-
 } #endif !NODEGAGNU
 
 ##############################################################
@@ -959,7 +905,7 @@
   'configtime_cpu'=> $ConfigTime,
   'buildtime_wall' => $BuildWallTime,
   'buildtime_cpu' => $BuildTime,
-  'warnings' => $WarningsFile,
+  'warnings' => "",
   'cvsusercommitlist' => "",
   'cvsuserupdatelist' => "",
   'cvsaddedfiles' => "",
@@ -973,8 +919,8 @@
   'multisource_programstable' => $MultiSourceProgramsTable,
   'externalsource_programstable' => $ExternalProgramsTable,
   'llcbeta_options' => $multisource_llcbeta_options,
-  'warnings_removed' => $WarningsRemoved,
-  'warnings_added' => $WarningsAdded,
+  'warnings_removed' => "",
+  'warnings_added' => "",
   'passing_tests' => $passes,
   'expfail_tests' => $xfails,
   'unexpfail_tests' => $fails,




From daniel at zuster.org  Mon Oct 19 08:20:00 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:00 -0000
Subject: [llvm-commits] [llvm] r84488 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDK0hL027203@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:00 2009
New Revision: 84488

URL: http://llvm.org/viewvc/llvm-project?rev=84488&view=rev
Log:
NNT: Remove .{o,a} size info, this is better tracked elsewhere.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84488&r1=84487&r2=84488&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:00 2009
@@ -646,31 +646,6 @@
 }
 if ($BuildError) { $NODEJAGNU=1; }
 
-my $a_file_sizes="";
-my $o_file_sizes="";
-if (!$BuildError) {
-  print "Organizing size of .o and .a files\n"
-    if ( $VERBOSE );
-  ChangeDir( "$LLVMObjDir", "Build Directory" );
-
-  my @dirs = ('utils', 'lib', 'tools');
-  if($BUILDTYPE eq "release"){
-    push @dirs, 'Release';
-  } elsif($BUILDTYPE eq "release-asserts") {
-    push @dirs, 'Release-Asserts';
-  } else {
-    push @dirs, 'Debug';
-  }
-
-  find(sub {
-      $a_file_sizes .= (-s $_)." $File::Find::name $BUILDTYPE\n" if /\.a$/i;
-      $o_file_sizes .= (-s $_)." $File::Find::name $BUILDTYPE\n" if /\.o$/i;
-    }, @dirs);
-} else {
-  $a_file_sizes="No data due to a bad build.";
-  $o_file_sizes="No data due to a bad build.";
-}
-
 ##############################################################
 #
 # Running dejagnu tests
@@ -931,8 +906,8 @@
   'dejagnutests_log' => $dejagnulog_full,
   'starttime' => $starttime,
   'endtime' => $endtime,
-  'o_file_sizes' => $o_file_sizes,
-  'a_file_sizes' => $a_file_sizes,
+  'o_file_sizes' => "",
+  'a_file_sizes' => "",
   'target_triple' => $targetTriple
 );
 




From daniel at zuster.org  Mon Oct 19 08:20:07 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:07 -0000
Subject: [llvm-commits] [llvm] r84489 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDK76m027228@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:06 2009
New Revision: 84489

URL: http://llvm.org/viewvc/llvm-project?rev=84489&view=rev
Log:
NNT: Move source checkout code into subroutine.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84489&r1=84488&r2=84489&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:06 2009
@@ -546,59 +546,41 @@
 # Create the source repository directory
 #
 ##############################################################
-if (!$NOCHECKOUT) {
-  if (-d $BuildDir) {
-    if (!$NOREMOVE) {
-      if ( $VERBOSE ) {
-        print "Build directory exists! Removing it\n";
+sub CheckoutSource {
+  if (!$NOCHECKOUT) {
+    if (-d $BuildDir) {
+      if (!$NOREMOVE) {
+        if ( $VERBOSE ) {
+          print "Build directory exists! Removing it\n";
+        }
+        system "rm -rf $BuildDir";
+        mkdir $BuildDir or die "Could not create checkout directory $BuildDir!";
+      } else {
+        if ( $VERBOSE ) {
+          print "Build directory exists!\n";
+        }
       }
-      system "rm -rf $BuildDir";
-      mkdir $BuildDir or die "Could not create checkout directory $BuildDir!";
     } else {
-      if ( $VERBOSE ) {
-        print "Build directory exists!\n";
-      }
+      mkdir $BuildDir or die "Could not create checkout directory $BuildDir!";
     }
-  } else {
-    mkdir $BuildDir or die "Could not create checkout directory $BuildDir!";
-  }
-}
-
 
-##############################################################
-#
-# Check out the llvm tree with SVN
-#
-##############################################################
-if (!$NOCHECKOUT) {
-  ChangeDir( $BuildDir, "checkout directory" );
-  my $SVNCMD = "$NICE svn co --non-interactive $SVNURL";
-  my $SVNCMD2 = "$NICE svn co --non-interactive $TestSVNURL";
-  RunLoggedCommand("( time -p $SVNCMD/llvm/trunk llvm; cd llvm/projects ; " .
-                   "$SVNCMD2/test-suite/trunk llvm-test )", $COLog,
-                   "CHECKOUT LLVM");
-  if ($WITHCLANG) {
-      my $SVNCMD = "$NICE svn co --non-interactive $SVNURL/cfe/trunk";
-      RunLoggedCommand("( time -p cd llvm/tools ; $SVNCMD clang )", $COLog,
-                       "CHECKOUT CLANG");
+    ChangeDir( $BuildDir, "checkout directory" );
+    my $SVNCMD = "$NICE svn co --non-interactive";
+    RunLoggedCommand("( time -p $SVNCMD $SVNURL/llvm/trunk llvm; cd llvm/projects ; " .
+                     "  $SVNCMD $TestSVNURL/test-suite/trunk llvm-test )", $COLog,
+                     "CHECKOUT LLVM");
+    if ($WITHCLANG) {
+        RunLoggedCommand("( cd llvm/tools ; " .
+                         "  $SVNCMD $SVNURL/cfe/trunk clang )", $COLog,
+                         "CHECKOUT CLANG");
+    }
   }
 }
+CheckoutSource();
 ChangeDir( $LLVMSrcDir , "llvm source directory") ;
 
 ##############################################################
 #
-# Get some static statistics about the current source code
-#
-# This can probably be put on the server side
-#
-##############################################################
-my $CheckoutTime_Wall = GetRegex "([0-9.]+)", `grep '^real' $COLog`;
-my $CheckoutTime_User = GetRegex "([0-9.]+)", `grep '^user' $COLog`;
-my $CheckoutTime_Sys = GetRegex "([0-9.]+)", `grep '^sys' $COLog`;
-my $CheckoutTime_CPU = $CVSCheckoutTime_User + $CVSCheckoutTime_Sys;
-
-##############################################################
-#
 # Build the entire tree, saving build messages to the build log
 #
 ##############################################################
@@ -651,7 +633,7 @@
 # Running dejagnu tests
 #
 ##############################################################
-my $DejangnuTestResults=""; # String containing the results of the dejagnu
+my $DejagnuTestResults=""; # String containing the results of the dejagnu
 my $dejagnu_output = "$DejagnuTestsLog";
 if (!$NODEJAGNU) {
   #Run the feature and regression tests, results are put into testrun.sum
@@ -813,11 +795,6 @@
                 "date: ".`date \"+20%y-%m-%d\"`.
                 "time: ".`date +\"%H:%M:%S\"`;
 
-my @CVS_DATA;
-my $cvs_data;
- at CVS_DATA = ReadFile "$COLog";
-$cvs_data = join("\n", @CVS_DATA);
-
 my @BUILD_DATA;
 my $build_data;
 @BUILD_DATA = ReadFile "$BuildLog";
@@ -867,6 +844,11 @@
 
 if ( $VERBOSE ) { print "SEND THE DATA VIA THE POST REQUEST\n"; }
 
+my $CheckoutTime_Wall = GetRegex "([0-9.]+)", `grep '^real' $COLog`;
+my $CheckoutTime_User = GetRegex "([0-9.]+)", `grep '^user' $COLog`;
+my $CheckoutTime_Sys = GetRegex "([0-9.]+)", `grep '^sys' $COLog`;
+my $CheckoutTime_CPU = $CVSCheckoutTime_User + $CVSCheckoutTime_Sys;
+
 my %hash_of_data = (
   'machine_data' => $machine_data,
   'build_data' => $build_data,




From daniel at zuster.org  Mon Oct 19 08:20:13 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:13 -0000
Subject: [llvm-commits] [llvm] r84490 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDKDvh027241@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:13 2009
New Revision: 84490

URL: http://llvm.org/viewvc/llvm-project?rev=84490&view=rev
Log:
NNT: Move build code into subroutine.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84490&r1=84489&r2=84490&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:13 2009
@@ -584,49 +584,35 @@
 # Build the entire tree, saving build messages to the build log
 #
 ##############################################################
-if (!$NOCHECKOUT && !$NOBUILD) {
-  my $EXTRAFLAGS = "--enable-spec --with-objroot=.";
-  RunLoggedCommand("(time -p $NICE ./configure $CONFIGUREARGS $EXTRAFLAGS) ",
-                   $BuildLog, "CONFIGURE");
-  # Build the entire tree, capturing the output into $BuildLog
-  RunAppendingLoggedCommand("(time -p $NICE $MAKECMD clean)", $BuildLog, "BUILD CLEAN");
-  RunAppendingLoggedCommand("(time -p $NICE $MAKECMD $MAKEOPTS)", $BuildLog, "BUILD");
-}
-
-##############################################################
-#
-# Get some statistics about the build...
-#
-##############################################################
-
-# Get the time taken by the configure script
-my $ConfigTimeU = GetRegexNum "^user", 0, "([0-9.]+)", "$BuildLog";
-my $ConfigTimeS = GetRegexNum "^sys", 0, "([0-9.]+)", "$BuildLog";
-my $ConfigTime  = $ConfigTimeU+$ConfigTimeS;  # ConfigTime = User+System
-my $ConfigWallTime = GetRegexNum "^real", 0,"([0-9.]+)","$BuildLog";
-
-$ConfigTime=-1 unless $ConfigTime;
-$ConfigWallTime=-1 unless $ConfigWallTime;
-
-my $BuildTimeU = GetRegexNum "^user", 1, "([0-9.]+)", "$BuildLog";
-my $BuildTimeS = GetRegexNum "^sys", 1, "([0-9.]+)", "$BuildLog";
-my $BuildTime  = $BuildTimeU+$BuildTimeS;  # BuildTime = User+System
-my $BuildWallTime = GetRegexNum "^real", 1, "([0-9.]+)","$BuildLog";
+sub BuildLLVM {
+  if (!$NOCHECKOUT && !$NOBUILD) {
+    my $EXTRAFLAGS = "--enable-spec --with-objroot=.";
+    RunLoggedCommand("(time -p $NICE ./configure $CONFIGUREARGS $EXTRAFLAGS) ",
+                     $BuildLog, "CONFIGURE");
+    # Build the entire tree, capturing the output into $BuildLog
+    RunAppendingLoggedCommand("(time -p $NICE $MAKECMD clean)", $BuildLog, "BUILD CLEAN");
+    RunAppendingLoggedCommand("(time -p $NICE $MAKECMD $MAKEOPTS)", $BuildLog, "BUILD");
+  }
 
-$BuildTime=-1 unless $BuildTime;
-$BuildWallTime=-1 unless $BuildWallTime;
+  # Check for build error.
+  my $HadError = 0, $Status = "OK";
+  if ($NOBUILD) {
+    $Status = "Skipped by user";
+  }
+  elsif (`grep '^$MAKECMD\[^:]*: .*Error' $BuildLog | wc -l` + 0 ||
+    `grep '^$MAKECMD: \*\*\*.*Stop.' $BuildLog | wc -l`+0) {
+    $Status = "Error: compilation aborted";
+    $HadError = 1;
+  }
 
-my $BuildError = 0, $BuildStatus = "OK";
-if ($NOBUILD) {
-  $BuildStatus = "Skipped by user";
+  return ($HadError, $Status);
 }
-elsif (`grep '^$MAKECMD\[^:]*: .*Error' $BuildLog | wc -l` + 0 ||
-  `grep '^$MAKECMD: \*\*\*.*Stop.' $BuildLog | wc -l`+0) {
-  $BuildStatus = "Error: compilation aborted";
-  $BuildError = 1;
-  if( $VERBOSE) { print  "\n***ERROR BUILDING TREE\n\n"; }
+($BuildError, $BuildStatus) = BuildLLVM();
+
+if ($BuildError) {
+    if( $VERBOSE) { print  "\n***ERROR BUILDING TREE\n\n"; }
+    $NODEJAGNU=1;
 }
-if ($BuildError) { $NODEJAGNU=1; }
 
 ##############################################################
 #
@@ -844,11 +830,29 @@
 
 if ( $VERBOSE ) { print "SEND THE DATA VIA THE POST REQUEST\n"; }
 
+# Checkout info.
 my $CheckoutTime_Wall = GetRegex "([0-9.]+)", `grep '^real' $COLog`;
 my $CheckoutTime_User = GetRegex "([0-9.]+)", `grep '^user' $COLog`;
 my $CheckoutTime_Sys = GetRegex "([0-9.]+)", `grep '^sys' $COLog`;
 my $CheckoutTime_CPU = $CVSCheckoutTime_User + $CVSCheckoutTime_Sys;
 
+# Configure info.
+my $ConfigTimeU = GetRegexNum "^user", 0, "([0-9.]+)", "$BuildLog";
+my $ConfigTimeS = GetRegexNum "^sys", 0, "([0-9.]+)", "$BuildLog";
+my $ConfigTime  = $ConfigTimeU+$ConfigTimeS;  # ConfigTime = User+System
+my $ConfigWallTime = GetRegexNum "^real", 0,"([0-9.]+)","$BuildLog";
+$ConfigTime=-1 unless $ConfigTime;
+$ConfigWallTime=-1 unless $ConfigWallTime;
+
+# Build info.
+my $BuildTimeU = GetRegexNum "^user", 1, "([0-9.]+)", "$BuildLog";
+my $BuildTimeS = GetRegexNum "^sys", 1, "([0-9.]+)", "$BuildLog";
+my $BuildTime  = $BuildTimeU+$BuildTimeS;  # BuildTime = User+System
+my $BuildWallTime = GetRegexNum "^real", 1, "([0-9.]+)","$BuildLog";
+$BuildTime=-1 unless $BuildTime;
+$BuildWallTime=-1 unless $BuildWallTime;
+
+
 my %hash_of_data = (
   'machine_data' => $machine_data,
   'build_data' => $build_data,




From daniel at zuster.org  Mon Oct 19 08:20:19 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:19 -0000
Subject: [llvm-commits] [llvm] r84491 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDKK2Z027262@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:19 2009
New Revision: 84491

URL: http://llvm.org/viewvc/llvm-project?rev=84491&view=rev
Log:
NNT: Tweaks and simplifications.
 - Split out configure log.
 - Kill off GetRegexNum.
 - Fix GetRegex to not return previous match on failure.
 - Remove dead code.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84491&r1=84490&r2=84491&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:19 2009
@@ -269,6 +269,7 @@
 #
 ##############################################################
 my $Prefix = "$WebDir/$DATE";
+my $ConfigureLog = "$Prefix-Configure-Log.txt";
 my $BuildLog = "$Prefix-Build-Log.txt";
 my $COLog = "$Prefix-CVS-Log.txt";
 my $SingleSourceLog = "$Prefix-SingleSource-ProgramTest.txt.gz";
@@ -276,7 +277,7 @@
 my $ExternalLog = "$Prefix-External-ProgramTest.txt.gz";
 my $DejagnuLog = "$Prefix-Dejagnu-testrun.log";
 my $DejagnuSum = "$Prefix-Dejagnu-testrun.sum";
-my $DejagnuTestsLog = "$Prefix-DejagnuTests-Log.txt";
+my $DejagnuLog = "$Prefix-DejagnuTests-Log.txt";
 if (! -d $WebDir) {
   mkdir $WebDir, 0777 or die "Unable to create web directory: '$WebDir'.";
   if($VERBOSE){
@@ -346,25 +347,13 @@
   }
 }
 
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 sub GetRegex {   # (Regex with ()'s, value)
-  $_[1] =~ /$_[0]/m;
-  return $1
-    if (defined($1));
+  if ($_[1] =~ /$_[0]/m) {
+    return $1;
+  }
   return "0";
 }
 
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-sub GetRegexNum {
-  my ($Regex, $Num, $Regex2, $File) = @_;
-  my @Items = split "\n", `grep '$Regex' $File`;
-  return GetRegex $Regex2, $Items[$Num];
-}
-
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 sub ChangeDir { # directory, logical name
   my ($dir,$name) = @_;
   chomp($dir);
@@ -377,8 +366,6 @@
   return true;
 }
 
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 sub ReadFile {
   if (open (FILE, $_[0])) {
     undef $/;
@@ -392,16 +379,12 @@
   }
 }
 
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 sub WriteFile {  # (filename, contents)
   open (FILE, ">$_[0]") or die "Could not open file '$_[0]' for writing!\n";
   print FILE $_[1];
   close FILE;
 }
 
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 sub CopyFile { #filename, newfile
   my ($file, $newfile) = @_;
   chomp($file);
@@ -410,32 +393,6 @@
 }
 
 #~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-sub AddRecord {
-  my ($Val, $Filename,$WebDir) = @_;
-  my @Records;
-  if (open FILE, "$WebDir/$Filename") {
-    @Records = grep !/$DATE/, split "\n", <FILE>;
-    close FILE;
-  }
-  push @Records, "$DATE: $Val";
-  WriteFile "$WebDir/$Filename", (join "\n", @Records) . "\n";
-}
-
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-#
-# FormatTime - Convert a time from 1m23.45 into 83.45
-#
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-sub FormatTime {
-  my $Time = shift;
-  if ($Time =~ m/([0-9]+)m([0-9.]+)/) {
-    $Time = sprintf("%7.4f", $1*60.0+$2);
-  }
-  return $Time;
-}
-
-#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 #
 # This function is meant to read in the dejagnu sum file and
 # return a string with only the results (i.e. PASS/FAIL/XPASS/
@@ -588,9 +545,9 @@
   if (!$NOCHECKOUT && !$NOBUILD) {
     my $EXTRAFLAGS = "--enable-spec --with-objroot=.";
     RunLoggedCommand("(time -p $NICE ./configure $CONFIGUREARGS $EXTRAFLAGS) ",
-                     $BuildLog, "CONFIGURE");
+                     $ConfigureLog, "CONFIGURE");
     # Build the entire tree, capturing the output into $BuildLog
-    RunAppendingLoggedCommand("(time -p $NICE $MAKECMD clean)", $BuildLog, "BUILD CLEAN");
+    RunAppendingLoggedCommand("($NICE $MAKECMD clean)", $BuildLog, "BUILD CLEAN");
     RunAppendingLoggedCommand("(time -p $NICE $MAKECMD $MAKEOPTS)", $BuildLog, "BUILD");
   }
 
@@ -619,35 +576,22 @@
 # Running dejagnu tests
 #
 ##############################################################
-my $DejagnuTestResults=""; # String containing the results of the dejagnu
-my $dejagnu_output = "$DejagnuTestsLog";
+
+# String containing the results of the dejagnu
+my $DejagnuTestResults="Dejagnu skipped by user choice.";
+
 if (!$NODEJAGNU) {
   #Run the feature and regression tests, results are put into testrun.sum
   #Full log in testrun.log
-  RunLoggedCommand("(time -p $MAKECMD $MAKEOPTS check)", $dejagnu_output, "DEJAGNU");
+  RunLoggedCommand("(time -p $MAKECMD $MAKEOPTS check)", $DejagnuLog, "DEJAGNU");
 
   #Copy the testrun.log and testrun.sum to our webdir
   CopyFile("test/testrun.log", $DejagnuLog);
   CopyFile("test/testrun.sum", $DejagnuSum);
   #can be done on server
   $DejagnuTestResults = GetDejagnuTestResults($DejagnuSum, $DejagnuLog);
-  $unexpfail_tests = $DejagnuTestResults;
 }
 
-#Extract time of dejagnu tests
-my $DejagnuTimeU = GetRegexNum "^user", 0, "([0-9.]+)", "$dejagnu_output";
-my $DejagnuTimeS = GetRegexNum "^sys", 0, "([0-9.]+)", "$dejagnu_output";
-$DejagnuTime  = $DejagnuTimeU+$DejagnuTimeS;  # DejagnuTime = User+System
-$DejagnuWallTime = GetRegexNum "^real", 0,"([0-9.]+)","$dejagnu_output";
-$DejagnuTestResults =
-  "Dejagnu skipped by user choice." unless $DejagnuTestResults;
-$DejagnuTime     = "0.0" unless $DejagnuTime;
-$DejagnuWallTime = "0.0" unless $DejagnuWallTime;
-
-if (!$NODEJAGNU) {
-  if ( $VERBOSE ) { print "BUILD INFORMATION COLLECTION STAGE\n"; }
-} #endif !NODEGAGNU
-
 ##############################################################
 #
 # If we built the tree successfully, run the nightly programs tests...
@@ -665,6 +609,7 @@
   my $ProgramTestLog = "$Prefix-$SubDir-ProgramTest.txt";
 
   # Run the programs tests... creating a report.nightly.csv file
+  my $LLCBetaOpts = "";
   if (!$NOTEST) {
     if( $VERBOSE) {
       print "$MAKECMD -k $MAKEOPTS $PROGTESTOPTS report.nightly.csv ".
@@ -673,7 +618,7 @@
     RunLoggedCommand("$MAKECMD -k $MAKEOPTS $PROGTESTOPTS report.nightly.csv ".
                      "$TESTFLAGS TEST=nightly",
                      $ProgramTestLog, "TEST DIRECTORY $SubDir");
-    $llcbeta_options=`$MAKECMD print-llcbeta-option`;
+    $LLCBetaOpts = `$MAKECMD print-llcbeta-option`;
   }
 
   my $ProgramsTable;
@@ -696,8 +641,8 @@
   $ProgramsTable = ReadFile "report.nightly.csv";
 
   ChangeDir( "../../..", "Programs Test Parent Directory" );
-  return ($ProgramsTable, $llcbeta_options);
-} #end sub TestDirectory
+  return ($ProgramsTable, $LLCBetaOpts);
+}
 
 ##############################################################
 #
@@ -705,8 +650,7 @@
 #
 ##############################################################
 if (!$BuildError) {
-  ($SingleSourceProgramsTable, $llcbeta_options) =
-    TestDirectory("SingleSource");
+  ($SingleSourceProgramsTable, $llcbeta_options) = TestDirectory("SingleSource");
   WriteFile "$Prefix-SingleSource-Performance.txt", $SingleSourceProgramsTable;
   ($MultiSourceProgramsTable, $llcbeta_options) = TestDirectory("MultiSource");
   WriteFile "$Prefix-MultiSource-Performance.txt", $MultiSourceProgramsTable;
@@ -732,32 +676,22 @@
                " | sort > $Prefix-Performance.txt";
   }
 
-  ##############################################################
-  #
-  #
-  # gathering tests added removed broken information here
-  #
-  #
-  ##############################################################
-  my $dejagnu_test_list = ReadFile "$Prefix-Tests.txt";
-  my @DEJAGNU = split "\n", $dejagnu_test_list;
+  # Compile passes, fails, xfails.
+  my @TestSuiteResultLines = split "\n", (ReadFile "$Prefix-Tests.txt");
   my ($passes, $fails, $xfails) = "";
 
-  if(!$NODEJAGNU) {
-    for ($x=0; $x<@DEJAGNU; $x++) {
-      if ($DEJAGNU[$x] =~ m/^PASS:/) {
-        $passes.="$DEJAGNU[$x]\n";
-      }
-      elsif ($DEJAGNU[$x] =~ m/^FAIL:/) {
-        $fails.="$DEJAGNU[$x]\n";
-      }
-      elsif ($DEJAGNU[$x] =~ m/^XFAIL:/) {
-        $xfails.="$DEJAGNU[$x]\n";
-      }
+  for ($x=0; $x < @TestSuiteResultLines; $x++) {
+    if (@TestSuiteResultLines[$x] =~ m/^PASS:/) {
+      $passes .= "$TestSuiteResultLines[$x]\n";
+    }
+    elsif (@TestSuiteResultLines[$x] =~ m/^FAIL:/) {
+      $fails .= "$TestSuiteResultLines[$x]\n";
+    }
+    elsif (@TestSuiteResultLines[$x] =~ m/^XFAIL:/) {
+      $xfails .= "$TestSuiteResultLines[$x]\n";
     }
   }
-
-} #end if !$BuildError
+}
 
 ##############################################################
 #
@@ -781,46 +715,24 @@
                 "date: ".`date \"+20%y-%m-%d\"`.
                 "time: ".`date +\"%H:%M:%S\"`;
 
-my @BUILD_DATA;
-my $build_data;
- at BUILD_DATA = ReadFile "$BuildLog";
-$build_data = join("\n", @BUILD_DATA);
-
-my (@DEJAGNU_LOG, @DEJAGNU_SUM, @DEJAGNULOG_FULL, @GCC_VERSION);
-my ($dejagnutests_log ,$dejagnutests_sum, $dejagnulog_full) = "";
-my ($gcc_version, $gcc_version_long) = "";
-
-$gcc_version_long="";
+my $gcc_version_long = "";
 if ($GCCPATH ne "") {
-	$gcc_version_long = `$GCCPATH/gcc --version`;
+  $gcc_version_long = `$GCCPATH/gcc --version`;
 } elsif ($ENV{"CC"}) {
-	$gcc_version_long = `$ENV{"CC"} --version`;
+  $gcc_version_long = `$ENV{"CC"} --version`;
 } else {
-	$gcc_version_long = `gcc --version`;
+  $gcc_version_long = `gcc --version`;
 }
- at GCC_VERSION = split '\n', $gcc_version_long;
-$gcc_version = $GCC_VERSION[0];
+my $gcc_version = (split '\n', $gcc_version_long)[0];
 
-$llvmgcc_version_long="";
+my $llvmgcc_version_long = "";
 if ($LLVMGCCPATH ne "") {
   $llvmgcc_version_long = `$LLVMGCCPATH/llvm-gcc -v 2>&1`;
 } else {
   $llvmgcc_version_long = `llvm-gcc -v 2>&1`;
 }
- at LLVMGCC_VERSION = split '\n', $llvmgcc_version_long;
-$llvmgcc_versionTarget = $LLVMGCC_VERSION[1];
-$llvmgcc_versionTarget =~ /Target: (.+)/;
-$targetTriple = $1;
-
-if(!$BuildError){
-  @DEJAGNU_LOG = ReadFile "$DejagnuLog";
-  @DEJAGNU_SUM = ReadFile "$DejagnuSum";
-  $dejagnutests_log = join("\n", @DEJAGNU_LOG);
-  $dejagnutests_sum = join("\n", @DEJAGNU_SUM);
-
-  @DEJAGNULOG_FULL = ReadFile "$DejagnuTestsLog";
-  $dejagnulog_full = join("\n", @DEJAGNULOG_FULL);
-}
+(split '\n', $llvmgcc_version_long)[1] =~ /Target: (.+)/;
+my $targetTriple = $1;
 
 ##############################################################
 #
@@ -830,32 +742,45 @@
 
 if ( $VERBOSE ) { print "SEND THE DATA VIA THE POST REQUEST\n"; }
 
+# Logs.
+my $ConfigureLogData = ReadFile $ConfigureLog;
+my $BuildLogData = ReadFile $BuildLog;
+my $DejagnuLogData = ReadFile $DejagnuLog;
+my $CheckoutLogData = ReadFile $COLog;
+
 # Checkout info.
-my $CheckoutTime_Wall = GetRegex "([0-9.]+)", `grep '^real' $COLog`;
-my $CheckoutTime_User = GetRegex "([0-9.]+)", `grep '^user' $COLog`;
-my $CheckoutTime_Sys = GetRegex "([0-9.]+)", `grep '^sys' $COLog`;
+my $CheckoutTime_Wall = GetRegex "^real ([0-9.]+)", $CheckoutLogData;
+my $CheckoutTime_User = GetRegex "^user ([0-9.]+)", $CheckoutLogData;
+my $CheckoutTime_Sys = GetRegex "^sys ([0-9.]+)", $CheckoutLogData;
 my $CheckoutTime_CPU = $CVSCheckoutTime_User + $CVSCheckoutTime_Sys;
 
 # Configure info.
-my $ConfigTimeU = GetRegexNum "^user", 0, "([0-9.]+)", "$BuildLog";
-my $ConfigTimeS = GetRegexNum "^sys", 0, "([0-9.]+)", "$BuildLog";
+my $ConfigTimeU = GetRegex "^user ([0-9.]+)", $ConfigureLogData;
+my $ConfigTimeS = GetRegex "^sys ([0-9.]+)", $ConfigureLogData;
 my $ConfigTime  = $ConfigTimeU+$ConfigTimeS;  # ConfigTime = User+System
-my $ConfigWallTime = GetRegexNum "^real", 0,"([0-9.]+)","$BuildLog";
+my $ConfigWallTime = GetRegex "^real ([0-9.]+)",$ConfigureLogData;
 $ConfigTime=-1 unless $ConfigTime;
 $ConfigWallTime=-1 unless $ConfigWallTime;
 
 # Build info.
-my $BuildTimeU = GetRegexNum "^user", 1, "([0-9.]+)", "$BuildLog";
-my $BuildTimeS = GetRegexNum "^sys", 1, "([0-9.]+)", "$BuildLog";
+my $BuildTimeU = GetRegex "^user ([0-9.]+)", $BuildLogData;
+my $BuildTimeS = GetRegex "^sys ([0-9.]+)", $BuildLogData;
 my $BuildTime  = $BuildTimeU+$BuildTimeS;  # BuildTime = User+System
-my $BuildWallTime = GetRegexNum "^real", 1, "([0-9.]+)","$BuildLog";
+my $BuildWallTime = GetRegex "^real ([0-9.]+)", $BuildLogData;
 $BuildTime=-1 unless $BuildTime;
 $BuildWallTime=-1 unless $BuildWallTime;
 
+# DejaGNU info.
+my $DejagnuTimeU = GetRegex "^user ([0-9.]+)", $DejagnuLogData;
+my $DejagnuTimeS = GetRegex "^sys ([0-9.]+)", $DejagnuLogData;
+$DejagnuTime  = $DejagnuTimeU+$DejagnuTimeS;  # DejagnuTime = User+System
+$DejagnuWallTime = GetRegex "^real ([0-9.]+)", $DejagnuLogData;
+$DejagnuTime     = "0.0" unless $DejagnuTime;
+$DejagnuWallTime = "0.0" unless $DejagnuWallTime;
 
 my %hash_of_data = (
   'machine_data' => $machine_data,
-  'build_data' => $build_data,
+  'build_data' => $ConfigureLogData . $BuildLogData,
   'gcc_version' => $gcc_version,
   'nickname' => $nickname,
   'dejagnutime_wall' => $DejagnuWallTime,
@@ -879,7 +804,7 @@
   'singlesource_programstable' => $SingleSourceProgramsTable,
   'multisource_programstable' => $MultiSourceProgramsTable,
   'externalsource_programstable' => $ExternalProgramsTable,
-  'llcbeta_options' => $multisource_llcbeta_options,
+  'llcbeta_options' => $llcbeta_options,
   'warnings_removed' => "",
   'warnings_added' => "",
   'passing_tests' => $passes,
@@ -889,7 +814,7 @@
   'new_tests' => "",
   'removed_tests' => "",
   'dejagnutests_results' => $DejagnuTestResults,
-  'dejagnutests_log' => $dejagnulog_full,
+  'dejagnutests_log' => $DejagnuLogData,
   'starttime' => $starttime,
   'endtime' => $endtime,
   'o_file_sizes' => "",




From daniel at zuster.org  Mon Oct 19 08:20:26 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:26 -0000
Subject: [llvm-commits] [llvm] r84492 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDKQfI027274@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:25 2009
New Revision: 84492

URL: http://llvm.org/viewvc/llvm-project?rev=84492&view=rev
Log:
NNT: Sink code for running nightly test into subroutine.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84492&r1=84491&r2=84492&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:25 2009
@@ -649,50 +649,57 @@
 # Calling sub TestDirectory
 #
 ##############################################################
-if (!$BuildError) {
-  ($SingleSourceProgramsTable, $llcbeta_options) = TestDirectory("SingleSource");
-  WriteFile "$Prefix-SingleSource-Performance.txt", $SingleSourceProgramsTable;
-  ($MultiSourceProgramsTable, $llcbeta_options) = TestDirectory("MultiSource");
-  WriteFile "$Prefix-MultiSource-Performance.txt", $MultiSourceProgramsTable;
-  if ( ! $NOEXTERNALS ) {
-    ($ExternalProgramsTable, $llcbeta_options) = TestDirectory("External");
-    WriteFile "$Prefix-External-Performance.txt", $ExternalProgramsTable;
-    system "cat $Prefix-SingleSource-Tests.txt " .
-               "$Prefix-MultiSource-Tests.txt ".
-               "$Prefix-External-Tests.txt | sort > $Prefix-Tests.txt";
-    system "cat $Prefix-SingleSource-Performance.txt " .
-               "$Prefix-MultiSource-Performance.txt ".
-               "$Prefix-External-Performance.txt | sort > $Prefix-Performance.txt";
-  } else {
-    $ExternalProgramsTable = "External TEST STAGE SKIPPED\n";
-    if ( $VERBOSE ) {
-      print "External TEST STAGE SKIPPED\n";
+sub RunNightlyTest() {
+  if (!$BuildError) {
+    ($SSProgs, $llcbeta_options) = TestDirectory("SingleSource");
+    WriteFile "$Prefix-SingleSource-Performance.txt", $SSProgs;
+    ($MSProgs, $llcbeta_options) = TestDirectory("MultiSource");
+    WriteFile "$Prefix-MultiSource-Performance.txt", $MSProgs;
+    if ( ! $NOEXTERNALS ) {
+      ($ExtProgs, $llcbeta_options) = TestDirectory("External");
+      WriteFile "$Prefix-External-Performance.txt", $ExtProgs;
+      system "cat $Prefix-SingleSource-Tests.txt " .
+                 "$Prefix-MultiSource-Tests.txt ".
+                 "$Prefix-External-Tests.txt | sort > $Prefix-Tests.txt";
+      system "cat $Prefix-SingleSource-Performance.txt " .
+                 "$Prefix-MultiSource-Performance.txt ".
+                 "$Prefix-External-Performance.txt | sort > $Prefix-Performance.txt";
+    } else {
+      $ExtProgs = "External TEST STAGE SKIPPED\n";
+      if ( $VERBOSE ) {
+        print "External TEST STAGE SKIPPED\n";
+      }
+      system "cat $Prefix-SingleSource-Tests.txt " .
+                 "$Prefix-MultiSource-Tests.txt ".
+                 " | sort > $Prefix-Tests.txt";
+      system "cat $Prefix-SingleSource-Performance.txt " .
+                 "$Prefix-MultiSource-Performance.txt ".
+                 " | sort > $Prefix-Performance.txt";
     }
-    system "cat $Prefix-SingleSource-Tests.txt " .
-               "$Prefix-MultiSource-Tests.txt ".
-               " | sort > $Prefix-Tests.txt";
-    system "cat $Prefix-SingleSource-Performance.txt " .
-               "$Prefix-MultiSource-Performance.txt ".
-               " | sort > $Prefix-Performance.txt";
-  }
 
-  # Compile passes, fails, xfails.
-  my @TestSuiteResultLines = split "\n", (ReadFile "$Prefix-Tests.txt");
-  my ($passes, $fails, $xfails) = "";
-
-  for ($x=0; $x < @TestSuiteResultLines; $x++) {
-    if (@TestSuiteResultLines[$x] =~ m/^PASS:/) {
-      $passes .= "$TestSuiteResultLines[$x]\n";
-    }
-    elsif (@TestSuiteResultLines[$x] =~ m/^FAIL:/) {
-      $fails .= "$TestSuiteResultLines[$x]\n";
-    }
-    elsif (@TestSuiteResultLines[$x] =~ m/^XFAIL:/) {
-      $xfails .= "$TestSuiteResultLines[$x]\n";
+    # Compile passes, fails, xfails.
+    my @TestSuiteResultLines = split "\n", (ReadFile "$Prefix-Tests.txt");
+    my ($Passes, $Fails, $XFails) = "";
+
+    for ($x=0; $x < @TestSuiteResultLines; $x++) {
+      if (@TestSuiteResultLines[$x] =~ m/^PASS:/) {
+        $Passes .= "$TestSuiteResultLines[$x]\n";
+      }
+      elsif (@TestSuiteResultLines[$x] =~ m/^FAIL:/) {
+        $Fails .= "$TestSuiteResultLines[$x]\n";
+      }
+      elsif (@TestSuiteResultLines[$x] =~ m/^XFAIL:/) {
+        $XFails .= "$TestSuiteResultLines[$x]\n";
+      }
     }
   }
+
+  return ($SSProgs, $MSProgs, $ExtProgs, $Passes, $Fails, $XFails)
 }
 
+my ($SingleSourceProgramsTable, $MultiSourceProgramsTable, $ExternalProgramsTable,
+    $passes, $fails, $xfails) = RunNightlyTest();
+
 ##############################################################
 #
 # Getting end timestamp




From daniel at zuster.org  Mon Oct 19 08:20:32 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:32 -0000
Subject: [llvm-commits] [llvm] r84493 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDKWdX027289@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:31 2009
New Revision: 84493

URL: http://llvm.org/viewvc/llvm-project?rev=84493&view=rev
Log:
NNT: Now that build & test steps are factored out, coalesce all the logic together.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84493&r1=84492&r2=84493&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:31 2009
@@ -493,16 +493,11 @@
 
 ##############################################################
 #
-# Getting Start timestamp
+# Individual Build & Test Functions
 #
 ##############################################################
-$starttime = `date "+20%y-%m-%d %H:%M:%S"`;
 
-##############################################################
-#
-# Create the source repository directory
-#
-##############################################################
+# Create the source repository directory.
 sub CheckoutSource {
   if (!$NOCHECKOUT) {
     if (-d $BuildDir) {
@@ -533,14 +528,8 @@
     }
   }
 }
-CheckoutSource();
-ChangeDir( $LLVMSrcDir , "llvm source directory") ;
 
-##############################################################
-#
-# Build the entire tree, saving build messages to the build log
-#
-##############################################################
+# Build the entire tree, saving build messages to the build log.
 sub BuildLLVM {
   if (!$NOCHECKOUT && !$NOBUILD) {
     my $EXTRAFLAGS = "--enable-spec --with-objroot=.";
@@ -564,43 +553,27 @@
 
   return ($HadError, $Status);
 }
-($BuildError, $BuildStatus) = BuildLLVM();
-
-if ($BuildError) {
-    if( $VERBOSE) { print  "\n***ERROR BUILDING TREE\n\n"; }
-    $NODEJAGNU=1;
-}
 
-##############################################################
-#
-# Running dejagnu tests
-#
-##############################################################
+# Running dejagnu tests and save results to log.
+sub RunDejaGNUTests {
+  my $Res = "Dejagnu skipped by user choice.";
+
+  if (!$NODEJAGNU) {
+    #Run the feature and regression tests, results are put into testrun.sum
+    #Full log in testrun.log
+    RunLoggedCommand("(time -p $MAKECMD $MAKEOPTS check)", $DejagnuLog, "DEJAGNU");
+
+    #Copy the testrun.log and testrun.sum to our webdir
+    CopyFile("test/testrun.log", $DejagnuLog);
+    CopyFile("test/testrun.sum", $DejagnuSum);
 
-# String containing the results of the dejagnu
-my $DejagnuTestResults="Dejagnu skipped by user choice.";
+    $Res = GetDejagnuTestResults($DejagnuSum, $DejagnuLog);
+  }
 
-if (!$NODEJAGNU) {
-  #Run the feature and regression tests, results are put into testrun.sum
-  #Full log in testrun.log
-  RunLoggedCommand("(time -p $MAKECMD $MAKEOPTS check)", $DejagnuLog, "DEJAGNU");
-
-  #Copy the testrun.log and testrun.sum to our webdir
-  CopyFile("test/testrun.log", $DejagnuLog);
-  CopyFile("test/testrun.sum", $DejagnuSum);
-  #can be done on server
-  $DejagnuTestResults = GetDejagnuTestResults($DejagnuSum, $DejagnuLog);
+  return $Res;
 }
 
-##############################################################
-#
-# If we built the tree successfully, run the nightly programs tests...
-#
-# A set of tests to run is passed in (i.e. "SingleSource" "MultiSource"
-# "External")
-#
-##############################################################
-
+# Run the named tests (i.e. "SingleSource" "MultiSource" "External")
 sub TestDirectory {
   my $SubDir = shift;
   ChangeDir( "$LLVMTestDir/$SubDir",
@@ -644,11 +617,8 @@
   return ($ProgramsTable, $LLCBetaOpts);
 }
 
-##############################################################
-#
-# Calling sub TestDirectory
-#
-##############################################################
+# Run all the nightly tests and return the program tables and the number of
+# passes, fails, and xfails.
 sub RunNightlyTest() {
   if (!$BuildError) {
     ($SSProgs, $llcbeta_options) = TestDirectory("SingleSource");
@@ -694,25 +664,36 @@
     }
   }
 
-  return ($SSProgs, $MSProgs, $ExtProgs, $Passes, $Fails, $XFails)
+  return ($SSProgs, $MSProgs, $ExtProgs, $Passes, $Fails, $XFails);
 }
 
-my ($SingleSourceProgramsTable, $MultiSourceProgramsTable, $ExternalProgramsTable,
-    $passes, $fails, $xfails) = RunNightlyTest();
-
 ##############################################################
 #
-# Getting end timestamp
+# The actual NewNightlyTest logic.
 #
 ##############################################################
-$endtime = `date "+20%y-%m-%d %H:%M:%S"`;
 
+$starttime = `date "+20%y-%m-%d %H:%M:%S"`;
+
+CheckoutSource();
+ChangeDir( $LLVMSrcDir , "llvm source directory") ;
+($BuildError, $BuildStatus) = BuildLLVM();
+if ($BuildError) {
+    if( $VERBOSE) { print  "\n***ERROR BUILDING TREE\n\n"; }
+    $NODEJAGNU=1;
+}
+my $DejagnuTestResults = RunDejaGNUTests();
+my ($SingleSourceProgramsTable, $MultiSourceProgramsTable, $ExternalProgramsTable,
+    $passes, $fails, $xfails) = RunNightlyTest();
+
+$endtime = `date "+20%y-%m-%d %H:%M:%S"`;
 
 ##############################################################
 #
-# Place all the logs neatly into one humungous file
+# Accumulate the information to send to the server.
 #
 ##############################################################
+
 if ( $VERBOSE ) { print "PREPARING LOGS TO BE SENT TO SERVER\n"; }
 
 $machine_data = "uname: ".`uname -a`.
@@ -722,6 +703,7 @@
                 "date: ".`date \"+20%y-%m-%d\"`.
                 "time: ".`date +\"%H:%M:%S\"`;
 
+# Get gcc version.
 my $gcc_version_long = "";
 if ($GCCPATH ne "") {
   $gcc_version_long = `$GCCPATH/gcc --version`;
@@ -732,6 +714,7 @@
 }
 my $gcc_version = (split '\n', $gcc_version_long)[0];
 
+# Get llvm-gcc target triple.
 my $llvmgcc_version_long = "";
 if ($LLVMGCCPATH ne "") {
   $llvmgcc_version_long = `$LLVMGCCPATH/llvm-gcc -v 2>&1`;
@@ -741,12 +724,6 @@
 (split '\n', $llvmgcc_version_long)[1] =~ /Target: (.+)/;
 my $targetTriple = $1;
 
-##############################################################
-#
-# Send data via a post request
-#
-##############################################################
-
 if ( $VERBOSE ) { print "SEND THE DATA VIA THE POST REQUEST\n"; }
 
 # Logs.




From daniel at zuster.org  Mon Oct 19 08:20:38 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:38 -0000
Subject: [llvm-commits] [llvm] r84494 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDKc2A027304@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:38 2009
New Revision: 84494

URL: http://llvm.org/viewvc/llvm-project?rev=84494&view=rev
Log:
NNT: Lift conditional logic out of test steps.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84494&r1=84493&r2=84494&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:38 2009
@@ -499,78 +499,63 @@
 
 # Create the source repository directory.
 sub CheckoutSource {
-  if (!$NOCHECKOUT) {
-    if (-d $BuildDir) {
-      if (!$NOREMOVE) {
-        if ( $VERBOSE ) {
-          print "Build directory exists! Removing it\n";
-        }
-        system "rm -rf $BuildDir";
-        mkdir $BuildDir or die "Could not create checkout directory $BuildDir!";
-      } else {
-        if ( $VERBOSE ) {
-          print "Build directory exists!\n";
-        }
+  if (-d $BuildDir) {
+    if (!$NOREMOVE) {
+      if ( $VERBOSE ) {
+        print "Build directory exists! Removing it\n";
       }
-    } else {
+      system "rm -rf $BuildDir";
       mkdir $BuildDir or die "Could not create checkout directory $BuildDir!";
+    } else {
+      if ( $VERBOSE ) {
+        print "Build directory exists!\n";
+      }
     }
+  } else {
+    mkdir $BuildDir or die "Could not create checkout directory $BuildDir!";
+  }
 
-    ChangeDir( $BuildDir, "checkout directory" );
-    my $SVNCMD = "$NICE svn co --non-interactive";
-    RunLoggedCommand("( time -p $SVNCMD $SVNURL/llvm/trunk llvm; cd llvm/projects ; " .
-                     "  $SVNCMD $TestSVNURL/test-suite/trunk llvm-test )", $COLog,
-                     "CHECKOUT LLVM");
-    if ($WITHCLANG) {
-        RunLoggedCommand("( cd llvm/tools ; " .
-                         "  $SVNCMD $SVNURL/cfe/trunk clang )", $COLog,
-                         "CHECKOUT CLANG");
-    }
+  ChangeDir( $BuildDir, "checkout directory" );
+  my $SVNCMD = "$NICE svn co --non-interactive";
+  RunLoggedCommand("( time -p $SVNCMD $SVNURL/llvm/trunk llvm; cd llvm/projects ; " .
+                   "  $SVNCMD $TestSVNURL/test-suite/trunk llvm-test )", $COLog,
+                   "CHECKOUT LLVM");
+  if ($WITHCLANG) {
+      RunLoggedCommand("( cd llvm/tools ; " .
+                       "  $SVNCMD $SVNURL/cfe/trunk clang )", $COLog,
+                       "CHECKOUT CLANG");
   }
 }
 
-# Build the entire tree, saving build messages to the build log.
+# Build the entire tree, saving build messages to the build log. Returns false
+# on build failure.
 sub BuildLLVM {
-  if (!$NOCHECKOUT && !$NOBUILD) {
-    my $EXTRAFLAGS = "--enable-spec --with-objroot=.";
-    RunLoggedCommand("(time -p $NICE ./configure $CONFIGUREARGS $EXTRAFLAGS) ",
-                     $ConfigureLog, "CONFIGURE");
-    # Build the entire tree, capturing the output into $BuildLog
-    RunAppendingLoggedCommand("($NICE $MAKECMD clean)", $BuildLog, "BUILD CLEAN");
-    RunAppendingLoggedCommand("(time -p $NICE $MAKECMD $MAKEOPTS)", $BuildLog, "BUILD");
+  my $EXTRAFLAGS = "--enable-spec --with-objroot=.";
+  RunLoggedCommand("(time -p $NICE ./configure $CONFIGUREARGS $EXTRAFLAGS) ",
+                   $ConfigureLog, "CONFIGURE");
+  # Build the entire tree, capturing the output into $BuildLog
+  RunAppendingLoggedCommand("($NICE $MAKECMD clean)", $BuildLog, "BUILD CLEAN");
+  RunAppendingLoggedCommand("(time -p $NICE $MAKECMD $MAKEOPTS)", $BuildLog, "BUILD");
+
+  if (`grep '^$MAKECMD\[^:]*: .*Error' $BuildLog | wc -l` + 0 ||
+      `grep '^$MAKECMD: \*\*\*.*Stop.' $BuildLog | wc -l` + 0) {
+    return 0;
   }
 
-  # Check for build error.
-  my $HadError = 0, $Status = "OK";
-  if ($NOBUILD) {
-    $Status = "Skipped by user";
-  }
-  elsif (`grep '^$MAKECMD\[^:]*: .*Error' $BuildLog | wc -l` + 0 ||
-    `grep '^$MAKECMD: \*\*\*.*Stop.' $BuildLog | wc -l`+0) {
-    $Status = "Error: compilation aborted";
-    $HadError = 1;
-  }
-
-  return ($HadError, $Status);
+  return 1;
 }
 
 # Running dejagnu tests and save results to log.
 sub RunDejaGNUTests {
-  my $Res = "Dejagnu skipped by user choice.";
-
-  if (!$NODEJAGNU) {
-    #Run the feature and regression tests, results are put into testrun.sum
-    #Full log in testrun.log
-    RunLoggedCommand("(time -p $MAKECMD $MAKEOPTS check)", $DejagnuLog, "DEJAGNU");
-
-    #Copy the testrun.log and testrun.sum to our webdir
-    CopyFile("test/testrun.log", $DejagnuLog);
-    CopyFile("test/testrun.sum", $DejagnuSum);
-
-    $Res = GetDejagnuTestResults($DejagnuSum, $DejagnuLog);
-  }
+  # Run the feature and regression tests, results are put into testrun.sum and
+  # the full log in testrun.log.
+  RunLoggedCommand("(time -p $MAKECMD $MAKEOPTS check)", $DejagnuLog, "DEJAGNU");
+
+  # Copy the testrun.log and testrun.sum to our webdir.
+  CopyFile("test/testrun.log", $DejagnuLog);
+  CopyFile("test/testrun.sum", $DejagnuSum);
 
-  return $Res;
+  return GetDejagnuTestResults($DejagnuSum, $DejagnuLog);
 }
 
 # Run the named tests (i.e. "SingleSource" "MultiSource" "External")
@@ -581,35 +566,28 @@
 
   my $ProgramTestLog = "$Prefix-$SubDir-ProgramTest.txt";
 
-  # Run the programs tests... creating a report.nightly.csv file
+  # Run the programs tests... creating a report.nightly.csv file.
   my $LLCBetaOpts = "";
-  if (!$NOTEST) {
-    if( $VERBOSE) {
-      print "$MAKECMD -k $MAKEOPTS $PROGTESTOPTS report.nightly.csv ".
-            "$TESTFLAGS TEST=nightly > $ProgramTestLog 2>&1\n";
-    }
-    RunLoggedCommand("$MAKECMD -k $MAKEOPTS $PROGTESTOPTS report.nightly.csv ".
-                     "$TESTFLAGS TEST=nightly",
-                     $ProgramTestLog, "TEST DIRECTORY $SubDir");
-    $LLCBetaOpts = `$MAKECMD print-llcbeta-option`;
+  if( $VERBOSE) {
+    print "$MAKECMD -k $MAKEOPTS $PROGTESTOPTS report.nightly.csv ".
+          "$TESTFLAGS TEST=nightly > $ProgramTestLog 2>&1\n";
   }
+  RunLoggedCommand("$MAKECMD -k $MAKEOPTS $PROGTESTOPTS report.nightly.csv ".
+                   "$TESTFLAGS TEST=nightly",
+                   $ProgramTestLog, "TEST DIRECTORY $SubDir");
+  $LLCBetaOpts = `$MAKECMD print-llcbeta-option`;
 
   my $ProgramsTable;
   if (`grep '^$MAKECMD\[^:]: .*Error' $ProgramTestLog | wc -l` + 0) {
-    $TestError = 1;
     $ProgramsTable="Error running test $SubDir\n";
     print "ERROR TESTING\n";
   } elsif (`grep '^$MAKECMD\[^:]: .*No rule to make target' $ProgramTestLog | wc -l` + 0) {
-    $TestError = 1;
     $ProgramsTable="Makefile error running tests $SubDir!\n";
     print "ERROR TESTING\n";
   } else {
-    $TestError = 0;
-  #
-  # Create a list of the tests which were run...
-  #
-  system "egrep 'TEST-(PASS|FAIL)' < $ProgramTestLog ".
-         "| sort > $Prefix-$SubDir-Tests.txt";
+    # Create a list of the tests which were run...
+    system "egrep 'TEST-(PASS|FAIL)' < $ProgramTestLog ".
+           "| sort > $Prefix-$SubDir-Tests.txt";
   }
   $ProgramsTable = ReadFile "report.nightly.csv";
 
@@ -620,47 +598,45 @@
 # Run all the nightly tests and return the program tables and the number of
 # passes, fails, and xfails.
 sub RunNightlyTest() {
-  if (!$BuildError) {
-    ($SSProgs, $llcbeta_options) = TestDirectory("SingleSource");
-    WriteFile "$Prefix-SingleSource-Performance.txt", $SSProgs;
-    ($MSProgs, $llcbeta_options) = TestDirectory("MultiSource");
-    WriteFile "$Prefix-MultiSource-Performance.txt", $MSProgs;
-    if ( ! $NOEXTERNALS ) {
-      ($ExtProgs, $llcbeta_options) = TestDirectory("External");
-      WriteFile "$Prefix-External-Performance.txt", $ExtProgs;
-      system "cat $Prefix-SingleSource-Tests.txt " .
-                 "$Prefix-MultiSource-Tests.txt ".
-                 "$Prefix-External-Tests.txt | sort > $Prefix-Tests.txt";
-      system "cat $Prefix-SingleSource-Performance.txt " .
-                 "$Prefix-MultiSource-Performance.txt ".
-                 "$Prefix-External-Performance.txt | sort > $Prefix-Performance.txt";
-    } else {
-      $ExtProgs = "External TEST STAGE SKIPPED\n";
-      if ( $VERBOSE ) {
-        print "External TEST STAGE SKIPPED\n";
-      }
-      system "cat $Prefix-SingleSource-Tests.txt " .
-                 "$Prefix-MultiSource-Tests.txt ".
-                 " | sort > $Prefix-Tests.txt";
-      system "cat $Prefix-SingleSource-Performance.txt " .
-                 "$Prefix-MultiSource-Performance.txt ".
-                 " | sort > $Prefix-Performance.txt";
+  ($SSProgs, $llcbeta_options) = TestDirectory("SingleSource");
+  WriteFile "$Prefix-SingleSource-Performance.txt", $SSProgs;
+  ($MSProgs, $llcbeta_options) = TestDirectory("MultiSource");
+  WriteFile "$Prefix-MultiSource-Performance.txt", $MSProgs;
+  if ( ! $NOEXTERNALS ) {
+    ($ExtProgs, $llcbeta_options) = TestDirectory("External");
+    WriteFile "$Prefix-External-Performance.txt", $ExtProgs;
+    system "cat $Prefix-SingleSource-Tests.txt " .
+               "$Prefix-MultiSource-Tests.txt ".
+               "$Prefix-External-Tests.txt | sort > $Prefix-Tests.txt";
+    system "cat $Prefix-SingleSource-Performance.txt " .
+               "$Prefix-MultiSource-Performance.txt ".
+               "$Prefix-External-Performance.txt | sort > $Prefix-Performance.txt";
+  } else {
+    $ExtProgs = "External TEST STAGE SKIPPED\n";
+    if ( $VERBOSE ) {
+      print "External TEST STAGE SKIPPED\n";
     }
+    system "cat $Prefix-SingleSource-Tests.txt " .
+               "$Prefix-MultiSource-Tests.txt ".
+               " | sort > $Prefix-Tests.txt";
+    system "cat $Prefix-SingleSource-Performance.txt " .
+               "$Prefix-MultiSource-Performance.txt ".
+               " | sort > $Prefix-Performance.txt";
+  }
 
-    # Compile passes, fails, xfails.
-    my @TestSuiteResultLines = split "\n", (ReadFile "$Prefix-Tests.txt");
-    my ($Passes, $Fails, $XFails) = "";
-
-    for ($x=0; $x < @TestSuiteResultLines; $x++) {
-      if (@TestSuiteResultLines[$x] =~ m/^PASS:/) {
-        $Passes .= "$TestSuiteResultLines[$x]\n";
-      }
-      elsif (@TestSuiteResultLines[$x] =~ m/^FAIL:/) {
-        $Fails .= "$TestSuiteResultLines[$x]\n";
-      }
-      elsif (@TestSuiteResultLines[$x] =~ m/^XFAIL:/) {
-        $XFails .= "$TestSuiteResultLines[$x]\n";
-      }
+  # Compile passes, fails, xfails.
+  my @TestSuiteResultLines = split "\n", (ReadFile "$Prefix-Tests.txt");
+  my ($Passes, $Fails, $XFails) = "";
+
+  for ($x=0; $x < @TestSuiteResultLines; $x++) {
+    if (@TestSuiteResultLines[$x] =~ m/^PASS:/) {
+      $Passes .= "$TestSuiteResultLines[$x]\n";
+    }
+    elsif (@TestSuiteResultLines[$x] =~ m/^FAIL:/) {
+      $Fails .= "$TestSuiteResultLines[$x]\n";
+    }
+    elsif (@TestSuiteResultLines[$x] =~ m/^XFAIL:/) {
+      $XFails .= "$TestSuiteResultLines[$x]\n";
     }
   }
 
@@ -675,16 +651,37 @@
 
 $starttime = `date "+20%y-%m-%d %H:%M:%S"`;
 
-CheckoutSource();
+if (!$NOCHECKOUT) {
+  CheckoutSource();
+}
+
+# Build LLVM.
+my $BuildError = 0, $BuildStatus = "OK";
 ChangeDir( $LLVMSrcDir , "llvm source directory") ;
-($BuildError, $BuildStatus) = BuildLLVM();
-if ($BuildError) {
+if ($NOCHECKOUT || $NOBUILD) {
+  $BuildStatus = "Skipped by user";
+} else {
+  if (!BuildLLVM()) {
     if( $VERBOSE) { print  "\n***ERROR BUILDING TREE\n\n"; }
+    $BuildError = 1;
+    $BuildStatus = "Error: compilation aborted";
     $NODEJAGNU=1;
+  }
 }
-my $DejagnuTestResults = RunDejaGNUTests();
+
+# Run DejaGNU.
+my $DejagnuTestResults = "Dejagnu skipped by user choice.";
+if (!$NODEJAGNU && !$BuildError) {
+  $DejagnuTestResults = RunDejaGNUTests();
+}
+
+# Run the llvm-test tests.
 my ($SingleSourceProgramsTable, $MultiSourceProgramsTable, $ExternalProgramsTable,
-    $passes, $fails, $xfails) = RunNightlyTest();
+    $passes, $fails, $xfails) = "";
+if (!$NOTEST && !$BuildError) {
+  ($SingleSourceProgramsTable, $MultiSourceProgramsTable, $ExternalProgramsTable,
+   $passes, $fails, $xfails) = RunNightlyTest();
+}
 
 $endtime = `date "+20%y-%m-%d %H:%M:%S"`;
 




From daniel at zuster.org  Mon Oct 19 08:20:45 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:45 -0000
Subject: [llvm-commits] [llvm] r84495 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDKjNR027331@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:44 2009
New Revision: 84495

URL: http://llvm.org/viewvc/llvm-project?rev=84495&view=rev
Log:
NNT: Fix refactoro, I dropped the list of all (llvm-test) tests. I'm sure it was named dejagnu_test_list for a good reason.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84495&r1=84494&r2=84495&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:44 2009
@@ -595,7 +595,7 @@
   return ($ProgramsTable, $LLCBetaOpts);
 }
 
-# Run all the nightly tests and return the program tables and the number of
+# Run all the nightly tests and return the program tables and the list of tests,
 # passes, fails, and xfails.
 sub RunNightlyTest() {
   ($SSProgs, $llcbeta_options) = TestDirectory("SingleSource");
@@ -625,7 +625,8 @@
   }
 
   # Compile passes, fails, xfails.
-  my @TestSuiteResultLines = split "\n", (ReadFile "$Prefix-Tests.txt");
+  my $All = (ReadFile "$Prefix-Tests.txt");
+  my @TestSuiteResultLines = split "\n", $All;
   my ($Passes, $Fails, $XFails) = "";
 
   for ($x=0; $x < @TestSuiteResultLines; $x++) {
@@ -640,7 +641,7 @@
     }
   }
 
-  return ($SSProgs, $MSProgs, $ExtProgs, $Passes, $Fails, $XFails);
+  return ($SSProgs, $MSProgs, $ExtProgs, $All, $Passes, $Fails, $XFails);
 }
 
 ##############################################################
@@ -677,10 +678,10 @@
 
 # Run the llvm-test tests.
 my ($SingleSourceProgramsTable, $MultiSourceProgramsTable, $ExternalProgramsTable,
-    $passes, $fails, $xfails) = "";
+    $all_tests, $passes, $fails, $xfails) = "";
 if (!$NOTEST && !$BuildError) {
   ($SingleSourceProgramsTable, $MultiSourceProgramsTable, $ExternalProgramsTable,
-   $passes, $fails, $xfails) = RunNightlyTest();
+   $all_tests, $passes, $fails, $xfails) = RunNightlyTest();
 }
 
 $endtime = `date "+20%y-%m-%d %H:%M:%S"`;
@@ -791,7 +792,7 @@
   'passing_tests' => $passes,
   'expfail_tests' => $xfails,
   'unexpfail_tests' => $fails,
-  'all_tests' => $dejagnu_test_list,
+  'all_tests' => $all_tests,
   'new_tests' => "",
   'removed_tests' => "",
   'dejagnutests_results' => $DejagnuTestResults,




From daniel at zuster.org  Mon Oct 19 08:20:51 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:51 -0000
Subject: [llvm-commits] [llvm] r84496 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDKpel027346@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:50 2009
New Revision: 84496

URL: http://llvm.org/viewvc/llvm-project?rev=84496&view=rev
Log:
NNT: Don't hard code -l3.0 argument to make, this is very server dependent. Users who care can use -compileflags for this.

Also, fix make clean call and a few other tweaks.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84496&r1=84495&r2=84496&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:50 2009
@@ -159,7 +159,7 @@
   if (/^-notest$/)         { $NOTEST = 1; next; }
   if (/^-norunningtests$/) { next; } # Backward compatibility, ignored.
   if (/^-parallel-jobs$/)  { $PARALLELJOBS = "$ARGV[0]"; shift; next;}
-  if (/^-parallel$/)       { $MAKEOPTS = "$MAKEOPTS -j$PARALLELJOBS -l3.0"; next; }
+  if (/^-parallel$/)       { $MAKEOPTS = "$MAKEOPTS -j$PARALLELJOBS"; next; }
   if (/^-with-clang$/)     { $WITHCLANG = 1; next; }
   if (/^-release$/)        { $MAKEOPTS = "$MAKEOPTS ENABLE_OPTIMIZED=1 ".
                              "OPTIMIZE_OPTION=-O2"; $BUILDTYPE="release"; next;}
@@ -360,7 +360,7 @@
   if ( $VERBOSE ) { print "Changing To: $name ($dir)\n"; }
   $result = chdir($dir);
   if (!$result) {
-    print "ERROR!!! Cannot change directory to: $name ($dir) because $!";
+    print "ERROR!!! Cannot change directory to: $name ($dir) because $!\n";
     return false;
   }
   return true;
@@ -534,7 +534,7 @@
   RunLoggedCommand("(time -p $NICE ./configure $CONFIGUREARGS $EXTRAFLAGS) ",
                    $ConfigureLog, "CONFIGURE");
   # Build the entire tree, capturing the output into $BuildLog
-  RunAppendingLoggedCommand("($NICE $MAKECMD clean)", $BuildLog, "BUILD CLEAN");
+  RunAppendingLoggedCommand("($NICE $MAKECMD $MAKEOPTS clean)", $BuildLog, "BUILD CLEAN");
   RunAppendingLoggedCommand("(time -p $NICE $MAKECMD $MAKEOPTS)", $BuildLog, "BUILD");
 
   if (`grep '^$MAKECMD\[^:]*: .*Error' $BuildLog | wc -l` + 0 ||
@@ -549,6 +549,7 @@
 sub RunDejaGNUTests {
   # Run the feature and regression tests, results are put into testrun.sum and
   # the full log in testrun.log.
+  system "rm -f test/testrun.log test/testrun.sum";
   RunLoggedCommand("(time -p $MAKECMD $MAKEOPTS check)", $DejagnuLog, "DEJAGNU");
 
   # Copy the testrun.log and testrun.sum to our webdir.
@@ -686,6 +687,9 @@
 
 $endtime = `date "+20%y-%m-%d %H:%M:%S"`;
 
+# The last bit of logic is to remove the build and web dirs, after sending data
+# to the server.
+
 ##############################################################
 #
 # Accumulate the information to send to the server.
@@ -722,8 +726,6 @@
 (split '\n', $llvmgcc_version_long)[1] =~ /Target: (.+)/;
 my $targetTriple = $1;
 
-if ( $VERBOSE ) { print "SEND THE DATA VIA THE POST REQUEST\n"; }
-
 # Logs.
 my $ConfigureLogData = ReadFile $ConfigureLog;
 my $BuildLogData = ReadFile $BuildLog;
@@ -760,6 +762,8 @@
 $DejagnuTime     = "0.0" unless $DejagnuTime;
 $DejagnuWallTime = "0.0" unless $DejagnuWallTime;
 
+if ( $VERBOSE ) { print "SEND THE DATA VIA THE POST REQUEST\n"; }
+
 my %hash_of_data = (
   'machine_data' => $machine_data,
   'build_data' => $ConfigureLogData . $BuildLogData,
@@ -773,35 +777,37 @@
   'configtime_cpu'=> $ConfigTime,
   'buildtime_wall' => $BuildWallTime,
   'buildtime_cpu' => $BuildTime,
-  'warnings' => "",
-  'cvsusercommitlist' => "",
-  'cvsuserupdatelist' => "",
-  'cvsaddedfiles' => "",
-  'cvsmodifiedfiles' => "",
-  'cvsremovedfiles' => "",
-  'lines_of_code' => "",
-  'cvs_file_count' => 0,
-  'cvs_dir_count' => 0,
   'buildstatus' => $BuildStatus,
   'singlesource_programstable' => $SingleSourceProgramsTable,
   'multisource_programstable' => $MultiSourceProgramsTable,
   'externalsource_programstable' => $ExternalProgramsTable,
   'llcbeta_options' => $llcbeta_options,
-  'warnings_removed' => "",
-  'warnings_added' => "",
   'passing_tests' => $passes,
   'expfail_tests' => $xfails,
   'unexpfail_tests' => $fails,
   'all_tests' => $all_tests,
-  'new_tests' => "",
-  'removed_tests' => "",
   'dejagnutests_results' => $DejagnuTestResults,
   'dejagnutests_log' => $DejagnuLogData,
   'starttime' => $starttime,
   'endtime' => $endtime,
+  'target_triple' => $targetTriple,
+
+  # Unused, but left around for backwards compatability.
+  'warnings' => "",
+  'cvsusercommitlist' => "",
+  'cvsuserupdatelist' => "",
+  'cvsaddedfiles' => "",
+  'cvsmodifiedfiles' => "",
+  'cvsremovedfiles' => "",
+  'lines_of_code' => "",
+  'cvs_file_count' => 0,
+  'cvs_dir_count' => 0,
+  'warnings_removed' => "",
+  'warnings_added' => "",
+  'new_tests' => "",
+  'removed_tests' => "",
   'o_file_sizes' => "",
-  'a_file_sizes' => "",
-  'target_triple' => $targetTriple
+  'a_file_sizes' => ""
 );
 
 if ($SUBMIT || !($SUBMITAUX eq "")) {




From daniel at zuster.org  Mon Oct 19 08:20:57 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 13:20:57 -0000
Subject: [llvm-commits] [llvm] r84497 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910191320.n9JDKvRo027358@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 08:20:56 2009
New Revision: 84497

URL: http://llvm.org/viewvc/llvm-project?rev=84497&view=rev
Log:
NNT: Add -parallel-test option, which runs llvm-test with
ENABLE_PARALLEL_REPORT.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84497&r1=84496&r2=84497&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 08:20:56 2009
@@ -77,6 +77,7 @@
 # OTHER OPTIONS:
 #  -parallel        Run parallel jobs with GNU Make (see -parallel-jobs).
 #  -parallel-jobs   The number of parallel Make jobs to use (default is two).
+#  -parallel-test   Allow parallel execution of llvm-test
 #  -verbose         Turn on some debug output
 #  -nice            Checkout/Configure/Build with "nice" to reduce impact
 #                   on busy servers.
@@ -160,6 +161,7 @@
   if (/^-norunningtests$/) { next; } # Backward compatibility, ignored.
   if (/^-parallel-jobs$/)  { $PARALLELJOBS = "$ARGV[0]"; shift; next;}
   if (/^-parallel$/)       { $MAKEOPTS = "$MAKEOPTS -j$PARALLELJOBS"; next; }
+  if (/^-parallel-test$/)  { $PROGTESTOPTS .= " ENABLE_PARALLEL_REPORT=1"; next; }
   if (/^-with-clang$/)     { $WITHCLANG = 1; next; }
   if (/^-release$/)        { $MAKEOPTS = "$MAKEOPTS ENABLE_OPTIMIZED=1 ".
                              "OPTIMIZE_OPTION=-O2"; $BUILDTYPE="release"; next;}




From gohman at apple.com  Mon Oct 19 09:47:32 2009
From: gohman at apple.com (Dan Gohman)
Date: Mon, 19 Oct 2009 14:47:32 -0000
Subject: [llvm-commits] [llvm] r84502 -
	/llvm/trunk/lib/Transforms/Utils/InstructionNamer.cpp
Message-ID: <200910191447.n9JElWOt031285@zion.cs.uiuc.edu>

Author: djg
Date: Mon Oct 19 09:47:32 2009
New Revision: 84502

URL: http://llvm.org/viewvc/llvm-project?rev=84502&view=rev
Log:
Change instnamer to name arguments "arg" instead of "tmp" for clarity, and
to name basic blocks "bb" instead of "BB", for consistency.

Modified:
    llvm/trunk/lib/Transforms/Utils/InstructionNamer.cpp

Modified: llvm/trunk/lib/Transforms/Utils/InstructionNamer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/InstructionNamer.cpp?rev=84502&r1=84501&r2=84502&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Utils/InstructionNamer.cpp (original)
+++ llvm/trunk/lib/Transforms/Utils/InstructionNamer.cpp Mon Oct 19 09:47:32 2009
@@ -33,11 +33,11 @@
       for (Function::arg_iterator AI = F.arg_begin(), AE = F.arg_end();
            AI != AE; ++AI)
         if (!AI->hasName() && AI->getType() != Type::getVoidTy(F.getContext()))
-          AI->setName("tmp");
+          AI->setName("arg");
 
       for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
         if (!BB->hasName())
-          BB->setName("BB");
+          BB->setName("bb");
         
         for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
           if (!I->hasName() && I->getType() != Type::getVoidTy(F.getContext()))




From gohman at apple.com  Mon Oct 19 09:52:05 2009
From: gohman at apple.com (Dan Gohman)
Date: Mon, 19 Oct 2009 14:52:05 -0000
Subject: [llvm-commits] [llvm] r84503 -
	/llvm/trunk/lib/CodeGen/MachineSink.cpp
Message-ID: <200910191452.n9JEq5f3031462@zion.cs.uiuc.edu>

Author: djg
Date: Mon Oct 19 09:52:05 2009
New Revision: 84503

URL: http://llvm.org/viewvc/llvm-project?rev=84503&view=rev
Log:
Change a few instance variables to be local variables.

Modified:
    llvm/trunk/lib/CodeGen/MachineSink.cpp

Modified: llvm/trunk/lib/CodeGen/MachineSink.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/MachineSink.cpp?rev=84503&r1=84502&r2=84503&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/MachineSink.cpp (original)
+++ llvm/trunk/lib/CodeGen/MachineSink.cpp Mon Oct 19 09:52:05 2009
@@ -34,10 +34,8 @@
 
 namespace {
   class VISIBILITY_HIDDEN MachineSinking : public MachineFunctionPass {
-    const TargetMachine   *TM;
     const TargetInstrInfo *TII;
     const TargetRegisterInfo *TRI;
-    MachineFunction       *CurMF; // Current MachineFunction
     MachineRegisterInfo  *RegInfo; // Machine register information
     MachineDominatorTree *DT;   // Machine dominator tree
     AliasAnalysis *AA;
@@ -92,19 +90,16 @@
   return true;
 }
 
-
-
 bool MachineSinking::runOnMachineFunction(MachineFunction &MF) {
   DEBUG(errs() << "******** Machine Sinking ********\n");
   
-  CurMF = &MF;
-  TM = &CurMF->getTarget();
-  TII = TM->getInstrInfo();
-  TRI = TM->getRegisterInfo();
-  RegInfo = &CurMF->getRegInfo();
+  const TargetMachine &TM = MF.getTarget();
+  TII = TM.getInstrInfo();
+  TRI = TM.getRegisterInfo();
+  RegInfo = &MF.getRegInfo();
   DT = &getAnalysis<MachineDominatorTree>();
   AA = &getAnalysis<AliasAnalysis>();
-  AllocatableSet = TRI->getAllocatableSet(*CurMF);
+  AllocatableSet = TRI->getAllocatableSet(MF);
 
   bool EverMadeChange = false;
   
@@ -112,7 +107,7 @@
     bool MadeChange = false;
 
     // Process all basic blocks.
-    for (MachineFunction::iterator I = CurMF->begin(), E = CurMF->end(); 
+    for (MachineFunction::iterator I = MF.begin(), E = MF.end(); 
          I != E; ++I)
       MadeChange |= ProcessBlock(*I);
     




From gohman at apple.com  Mon Oct 19 09:56:05 2009
From: gohman at apple.com (Dan Gohman)
Date: Mon, 19 Oct 2009 14:56:05 -0000
Subject: [llvm-commits] [llvm] r84504 -
	/llvm/trunk/lib/CodeGen/MachineSink.cpp
Message-ID: <200910191456.n9JEu5b4031648@zion.cs.uiuc.edu>

Author: djg
Date: Mon Oct 19 09:56:05 2009
New Revision: 84504

URL: http://llvm.org/viewvc/llvm-project?rev=84504&view=rev
Log:
Fix a typo in a comment.

Modified:
    llvm/trunk/lib/CodeGen/MachineSink.cpp

Modified: llvm/trunk/lib/CodeGen/MachineSink.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/MachineSink.cpp?rev=84504&r1=84503&r2=84504&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/MachineSink.cpp (original)
+++ llvm/trunk/lib/CodeGen/MachineSink.cpp Mon Oct 19 09:56:05 2009
@@ -251,7 +251,7 @@
   if (SuccToSinkTo->isLandingPad())
     return false;
   
-  // If is not possible to sink an instruction into its own block.  This can
+  // It is not possible to sink an instruction into its own block.  This can
   // happen with loops.
   if (MI->getParent() == SuccToSinkTo)
     return false;




From gohman at apple.com  Mon Oct 19 11:04:50 2009
From: gohman at apple.com (Dan Gohman)
Date: Mon, 19 Oct 2009 16:04:50 -0000
Subject: [llvm-commits] [llvm] r84505 - in /llvm/trunk:
 lib/Transforms/Utils/BasicBlockUtils.cpp
 test/Transforms/LoopSimplify/unreachable-loop-pred.ll
Message-ID: <200910191604.n9JG4pTB002152@zion.cs.uiuc.edu>

Author: djg
Date: Mon Oct 19 11:04:50 2009
New Revision: 84505

URL: http://llvm.org/viewvc/llvm-project?rev=84505&view=rev
Log:
Fix SplitBlockPredecessors' LoopInfo updating code to handle the case
where a loop's header is being split and it has predecessors which are not
contained by the most-nested loop which contains the loop.
This fixes PR5235.

Added:
    llvm/trunk/test/Transforms/LoopSimplify/unreachable-loop-pred.ll
Modified:
    llvm/trunk/lib/Transforms/Utils/BasicBlockUtils.cpp

Modified: llvm/trunk/lib/Transforms/Utils/BasicBlockUtils.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/BasicBlockUtils.cpp?rev=84505&r1=84504&r2=84505&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Utils/BasicBlockUtils.cpp (original)
+++ llvm/trunk/lib/Transforms/Utils/BasicBlockUtils.cpp Mon Oct 19 11:04:50 2009
@@ -425,14 +425,26 @@
 
   if (L) {
     if (IsLoopEntry) {
-      if (Loop *PredLoop = LI->getLoopFor(Preds[0])) {
-        // Add the new block to the nearest enclosing loop (and not an
-        // adjacent loop).
-        while (PredLoop && !PredLoop->contains(BB))
-          PredLoop = PredLoop->getParentLoop();
-        if (PredLoop)
-          PredLoop->addBasicBlockToLoop(NewBB, LI->getBase());
-      }
+      // Add the new block to the nearest enclosing loop (and not an
+      // adjacent loop). To find this, examine each of the predecessors and
+      // determine which loops enclose them, and select the most-nested loop
+      // which contains the loop containing the block being split.
+      Loop *InnermostPredLoop = 0;
+      for (unsigned i = 0; i != NumPreds; ++i)
+        if (Loop *PredLoop = LI->getLoopFor(Preds[i])) {
+          // Seek a loop which actually contains the block being split (to
+          // avoid adjacent loops).
+          while (PredLoop && !PredLoop->contains(BB))
+            PredLoop = PredLoop->getParentLoop();
+          // Select the most-nested of these loops which contains the block.
+          if (PredLoop &&
+              PredLoop->contains(BB) &&
+              (!InnermostPredLoop ||
+               InnermostPredLoop->getLoopDepth() < PredLoop->getLoopDepth()))
+            InnermostPredLoop = PredLoop;
+        }
+      if (InnermostPredLoop)
+        InnermostPredLoop->addBasicBlockToLoop(NewBB, LI->getBase());
     } else {
       L->addBasicBlockToLoop(NewBB, LI->getBase());
       if (SplitMakesNewLoopHeader)

Added: llvm/trunk/test/Transforms/LoopSimplify/unreachable-loop-pred.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopSimplify/unreachable-loop-pred.ll?rev=84505&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/LoopSimplify/unreachable-loop-pred.ll (added)
+++ llvm/trunk/test/Transforms/LoopSimplify/unreachable-loop-pred.ll Mon Oct 19 11:04:50 2009
@@ -0,0 +1,20 @@
+; RUN: opt -S -loopsimplify -disable-output -verify-loop-info -verify-dom-info < %s
+; PR5235
+
+; When loopsimplify inserts a preheader for this loop, it should add the new
+; block to the enclosing loop and not get confused by the unreachable
+; bogus loop entry.
+
+define void @is_extract_cab() nounwind {
+entry:
+  br label %header
+
+header:                                       ; preds = %if.end206, %cond.end66, %if.end23
+  br label %while.body115
+
+while.body115:                                    ; preds = %9, %if.end192, %if.end101
+  br i1 undef, label %header, label %while.body115
+
+foo:
+  br label %while.body115
+}




From gohman at apple.com  Mon Oct 19 11:21:59 2009
From: gohman at apple.com (Dan Gohman)
Date: Mon, 19 Oct 2009 09:21:59 -0700
Subject: [llvm-commits] patch: preserve branch to unreachable	until
 codegenprep time
In-Reply-To: <7999D788-8D7E-44D4-A759-8DBDB44F3707@apple.com>
References: <4ADAC0C9.6060508@mxc.ca>
	<A586E31E-36A3-4C18-A64E-9394FA98AC20@apple.com>
	<7999D788-8D7E-44D4-A759-8DBDB44F3707@apple.com>
Message-ID: <AEC52D0E-CBA3-4F62-A7AD-CA5CA1EC1319@apple.com>



On Oct 18, 2009, at 9:47 PM, Chris Lattner <clattner at apple.com> wrote:

>
> On Oct 18, 2009, at 10:11 AM, Dan Gohman wrote:
>
>>
>>
>> On Oct 18, 2009, at 12:16 AM, Nick Lewycky <nicholas at mxc.ca> wrote:
>>
>>> This patch moves the code that optimizes "branch to unreachable"
>>> from SimplifyCFG to CodeGenPrep. Most of this patch is just updating
>>> the unit tests that use unreachable everywhere. I've done my best to
>>> preserve the intent of the tests.
>>
>> I haven't read the patch, but I'm concerned about this interfering
>> with other optimizations.  Deleting the condition can expose further
>> dce and related optimizations.  Also if the branch is a loop exit,
>> deleting it makes it easier to compute a trip count.
>
> I agree, but I still think this is a good idea.  :)
>
> LSR and other passes can special case this for trip count handling.   
> We can make simplifycfg delete the branch to unreachable if the  
> condition is "too complicated" to be useful for the optimizer.

Speaking of LSR, if the condition has an IV expr use, this could cause  
LSR to insert an additional IV, or miss an opportunity to eliminate one.

And to resond to Nick's comment about CGP doing trivial DCE, what  
about non-trivial DSE oppportunities only exposed when dead loads are  
deleted?

CGP is very late. Have you considered removing the unreachable blocks  
sometime earlier?

Dan


From gohman at apple.com  Mon Oct 19 11:30:51 2009
From: gohman at apple.com (Dan Gohman)
Date: Mon, 19 Oct 2009 09:30:51 -0700
Subject: [llvm-commits] patch: preserve branch to unreachable until
 codegenprep time
In-Reply-To: <4ADB4FF7.8040904@mxc.ca>
References: <4ADAC0C9.6060508@mxc.ca>
	<A586E31E-36A3-4C18-A64E-9394FA98AC20@apple.com>
	<4ADB4FF7.8040904@mxc.ca>
Message-ID: <5C3B3816-852C-4C3A-A704-A51367992123@apple.com>



On Oct 18, 2009, at 10:27 AM, Nick Lewycky <nicholas at mxc.ca> wrote:

> Dan Gohman wrote:
>> On Oct 18, 2009, at 12:16 AM, Nick Lewycky <nicholas at mxc.ca> wrote:
>>> This patch moves the code that optimizes "branch to unreachable"  
>>> from SimplifyCFG to CodeGenPrep. Most of this patch is just  
>>> updating the unit tests that use unreachable everywhere. I've done  
>>> my best to preserve the intent of the tests.
>> I haven't read the patch, but I'm concerned about this interfering  
>> with other optimizations.  Deleting the condition can expose  
>> further dce and related optimizations.
>
> Yes, having extra instructions could interfere with optimizations.  
> Simple DCE isn't the problem though; when CodeGenPrep eliminates the  
> unreachables, it also DCEs the chain of computations.
>
> One case that could cause trouble is if an instruction had two uses,  
> one which fed into the "dead" branch and one which didn't, and  
> instcombine couldn't fold the instruction away only because it had  
> more than one use.

Indeed. Instcombine relies heavily on hasOneUse.

>
>  Also if the branch is a loop exit, deleting it
>> makes it easier to compute a trip count.
>
> They're easy to detect, just look for a BB whose first instruction  
> is "unreachable". I'll try making SCEV filter such blocks out of its  
> ExitingBlocks list.

LoopInfo would also need special-case code.

Dan


From johnny.chen at apple.com  Mon Oct 19 11:33:01 2009
From: johnny.chen at apple.com (Johnny Chen)
Date: Mon, 19 Oct 2009 09:33:01 -0700
Subject: [llvm-commits] Patch for encodings of VEXT (Vector Extract)
In-Reply-To: <AD31FE2C-054B-424B-A0DC-27C3F3A6BC4A@apple.com>
References: <AD31FE2C-054B-424B-A0DC-27C3F3A6BC4A@apple.com>
Message-ID: <D6C13168-01E3-419D-95D6-7425D9706AE3@apple.com>

This is bad english.  Correction follows:

Refs: A8-598.
For VEXT, Inst{11-8} represents the starting byte index of the extracted result in the
concatenation of the operands and should be left unspecified.

Thanks.

On Oct 16, 2009, at 11:17 AM, Johnny Chen wrote:

> Hi,
> 
> Refs: A8-598.
> Leave Inst{11-8}, which represents the starting byte index of the extracted result in the
> concatenation of the operands and is left unspecified.
> 
> <ARMInstrNEON.patch5>




From grosbach at apple.com  Mon Oct 19 12:27:06 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Mon, 19 Oct 2009 10:27:06 -0700
Subject: [llvm-commits] [llvm] r84355 - in /llvm/trunk:
	include/llvm/Support/raw_ostream.h lib/Support/raw_ostream.cpp
	unittests/Support/raw_ostream_test.cpp
In-Reply-To: <200910172043.n9HKh9O1022980@zion.cs.uiuc.edu>
References: <200910172043.n9HKh9O1022980@zion.cs.uiuc.edu>
Message-ID: <C7847C3C-6DFD-46B5-941C-B0995D20B123@apple.com>

Hi Daniel,

This looks very nice. Definitely an improvement over EscapeString.

I assume the choice to restrict the list handled explicitly and not  
include things like '\v' and '\b' is due to those being uncommon  
enough to just leave to the generic octal expansion?

-Jim

On Oct 17, 2009, at 1:43 PM, Daniel Dunbar wrote:

> Author: ddunbar
> Date: Sat Oct 17 15:43:08 2009
> New Revision: 84355
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84355&view=rev
> Log:
> Add raw_ostream::write_escaped, for writing escaped strings.
>
> Modified:
>    llvm/trunk/include/llvm/Support/raw_ostream.h
>    llvm/trunk/lib/Support/raw_ostream.cpp
>    llvm/trunk/unittests/Support/raw_ostream_test.cpp
>
> Modified: llvm/trunk/include/llvm/Support/raw_ostream.h
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/raw_ostream.h?rev=84355&r1=84354&r2=84355&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/include/llvm/Support/raw_ostream.h (original)
> +++ llvm/trunk/include/llvm/Support/raw_ostream.h Sat Oct 17  
> 15:43:08 2009
> @@ -216,6 +216,10 @@
>   /// write_hex - Output \arg N in hexadecimal, without any prefix  
> or padding.
>   raw_ostream &write_hex(unsigned long long N);
>
> +  /// write_escaped - Output \arg Str, turning '\\', '\t', '\n',  
> '"', and
> +  /// anything that doesn't satisfy std::isprint into an escape  
> sequence.
> +  raw_ostream &write_escaped(StringRef Str);
> +
>   raw_ostream &write(unsigned char C);
>   raw_ostream &write(const char *Ptr, size_t Size);
>
>
> Modified: llvm/trunk/lib/Support/raw_ostream.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Support/raw_ostream.cpp?rev=84355&r1=84354&r2=84355&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/lib/Support/raw_ostream.cpp (original)
> +++ llvm/trunk/lib/Support/raw_ostream.cpp Sat Oct 17 15:43:08 2009
> @@ -168,6 +168,40 @@
>   return write(CurPtr, EndPtr-CurPtr);
> }
>
> +raw_ostream &raw_ostream::write_escaped(StringRef Str) {
> +  for (unsigned i = 0, e = Str.size(); i != e; ++i) {
> +    unsigned char c = Str[i];
> +
> +    switch (c) {
> +    case '\\':
> +      *this << '\\' << '\\';
> +      break;
> +    case '\t':
> +      *this << '\\' << 't';
> +      break;
> +    case '\n':
> +      *this << '\\' << 'n';
> +      break;
> +    case '"':
> +      *this << '\\' << '"';
> +      break;
> +    default:
> +      if (std::isprint(c)) {
> +        *this << c;
> +        break;
> +      }
> +
> +      // Always expand to a 3-character octal escape.
> +      *this << '\\';
> +      *this << char('0' + ((c >> 6) & 7));
> +      *this << char('0' + ((c >> 3) & 7));
> +      *this << char('0' + ((c >> 0) & 7));
> +    }
> +  }
> +
> +  return *this;
> +}
> +
> raw_ostream &raw_ostream::operator<<(const void *P) {
>   *this << '0' << 'x';
>
>
> Modified: llvm/trunk/unittests/Support/raw_ostream_test.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/Support/raw_ostream_test.cpp?rev=84355&r1=84354&r2=84355&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/unittests/Support/raw_ostream_test.cpp (original)
> +++ llvm/trunk/unittests/Support/raw_ostream_test.cpp Sat Oct 17  
> 15:43:08 2009
> @@ -127,4 +127,20 @@
>   EXPECT_EQ("hello1world", OS.str());
> }
>
> +TEST(raw_ostreamTest, WriteEscaped) {
> +  std::string Str;
> +
> +  Str = "";
> +  raw_string_ostream(Str).write_escaped("hi");
> +  EXPECT_EQ("hi", Str);
> +
> +  Str = "";
> +  raw_string_ostream(Str).write_escaped("\\\t\n\"");
> +  EXPECT_EQ("\\\\\\t\\n\\\"", Str);
> +
> +  Str = "";
> +  raw_string_ostream(Str).write_escaped("\1\10\200");
> +  EXPECT_EQ("\\001\\010\\200", Str);
> +}
> +
> }
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From natebegeman at mac.com  Mon Oct 19 12:31:16 2009
From: natebegeman at mac.com (Nate Begeman)
Date: Mon, 19 Oct 2009 17:31:16 -0000
Subject: [llvm-commits] [llvm] r84506 -
	/llvm/trunk/lib/Target/X86/X86InstrFormats.td
Message-ID: <200910191731.n9JHVHKI005600@zion.cs.uiuc.edu>

Author: sampo
Date: Mon Oct 19 12:31:16 2009
New Revision: 84506

URL: http://llvm.org/viewvc/llvm-project?rev=84506&view=rev
Log:
PR 5245 - The imediate size target flag was not set on 3A-prefixed SSSE3 instructions.

Modified:
    llvm/trunk/lib/Target/X86/X86InstrFormats.td

Modified: llvm/trunk/lib/Target/X86/X86InstrFormats.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrFormats.td?rev=84506&r1=84505&r2=84506&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrFormats.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrFormats.td Mon Oct 19 12:31:16 2009
@@ -224,10 +224,10 @@
 
 class SS38I<bits<8> o, Format F, dag outs, dag ins, string asm,
             list<dag> pattern>
-      : I<o, F, outs, ins, asm, pattern>, T8, Requires<[HasSSSE3]>;
+      : Ii8<o, F, outs, ins, asm, pattern>, T8, Requires<[HasSSSE3]>;
 class SS3AI<bits<8> o, Format F, dag outs, dag ins, string asm,
             list<dag> pattern>
-      : I<o, F, outs, ins, asm, pattern>, TA, Requires<[HasSSSE3]>;
+      : Ii8<o, F, outs, ins, asm, pattern>, TA, Requires<[HasSSSE3]>;
 
 // SSE4.1 Instruction Templates:
 // 




From jyasskin at google.com  Mon Oct 19 12:45:11 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Mon, 19 Oct 2009 10:45:11 -0700
Subject: [llvm-commits] [PATCH] Sanitize the JITResolver
	stub/callsite<->function maps
In-Reply-To: <0016367b65c6b96c230475ed8580@google.com>
References: <0016367b65c6b96c230475ed8580@google.com>
Message-ID: <bb55930e0910191045w791e58ddx2aac279dadabdbd4@mail.gmail.com>

Evan, Chris says you're one of the right people to take a look at
this. Could you?

Thanks,
Jeffrey

On Wed, Oct 14, 2009 at 4:28 PM,  <jyasskin at gmail.com> wrote:
> Reviewers: ,
>
> Message:
> Patch at http://codereview.appspot.com/download/issue130080_1.diff
>
> This passes check-lit.
>
> Description:
> The JITResolver maps Functions to their canonical stubs and all
> callsites for lazily-compiled functions to their target Functions. To
> make Function destruction work, I'm going to need to remove all
> callsites on destruction, so this patch also adds the reverse mapping
> for that.
>
> There was an incorrect assumption in here that the only stub for a
> function would be the one caused by needing to lazily compile it, while
> x86-64 far calls and dlsym-stubs could also cause such stubs, but I
> didn't look for a test case that the assumption broke.
>
> This also adds DenseMapInfo<AssertingVH> so I can use DenseMaps instead
> of std::maps.
>
> Please review this at http://codereview.appspot.com/130080
>
> Affected files:
> ? M include/llvm/Support/ValueHandle.h
> ? M lib/ExecutionEngine/JIT/JITEmitter.cpp
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>



From stuart at apple.com  Mon Oct 19 12:53:58 2009
From: stuart at apple.com (Stuart Hastings)
Date: Mon, 19 Oct 2009 17:53:58 -0000
Subject: [llvm-commits] [llvm] r84507 - in /llvm/trunk: Makefile
 utils/buildit/GNUmakefile utils/buildit/build_llvm
Message-ID: <200910191753.n9JHrwL2006834@zion.cs.uiuc.edu>

Author: stuart
Date: Mon Oct 19 12:53:54 2009
New Revision: 84507

URL: http://llvm.org/viewvc/llvm-project?rev=84507&view=rev
Log:
Tweak top-level Makefile to facilitate Apple-style build.

Modified:
    llvm/trunk/Makefile
    llvm/trunk/utils/buildit/GNUmakefile
    llvm/trunk/utils/buildit/build_llvm

Modified: llvm/trunk/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/Makefile?rev=84507&r1=84506&r2=84507&view=diff

==============================================================================
--- llvm/trunk/Makefile (original)
+++ llvm/trunk/Makefile Mon Oct 19 12:53:54 2009
@@ -19,6 +19,7 @@
 #
 # When cross-compiling, there are some things (tablegen) that need to
 # be build for the build system first.
+ifndef RC_ARCHS  # Normal build (not "Apple-style").
 ifeq ($(BUILD_DIRS_ONLY),1)
   DIRS := lib/System lib/Support utils
   OPTIONAL_DIRS :=
@@ -211,3 +212,8 @@
 
 .NOTPARALLEL:
 
+else # Building "Apple-style."
+# N.B. In an Apple-style build, once configuration is done, lines
+# marked "Apple-style" are removed with sed!  See utils/buildit/build_llvm.
+include utils/buildit/GNUmakefile # Building "Apple-style."
+endif # Building "Apple-style."

Modified: llvm/trunk/utils/buildit/GNUmakefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/GNUmakefile?rev=84507&r1=84506&r2=84507&view=diff

==============================================================================
--- llvm/trunk/utils/buildit/GNUmakefile (original)
+++ llvm/trunk/utils/buildit/GNUmakefile Mon Oct 19 12:53:54 2009
@@ -59,7 +59,7 @@
 # NOTE : Always put version numbers at the end because they are optional.
 install: $(OBJROOT) $(SYMROOT) $(DSTROOT)
 	cd $(OBJROOT) && \
-	  $(SRC)/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
+	  $(SRC)/utils/buildit/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
 	    $(SRC) $(PREFIX) $(DSTROOT) $(SYMROOT) \
 	    $(LLVM_ASSERTIONS) $(LLVM_OPTIMIZED) \
 	    $(RC_ProjectSourceVersion) $(RC_ProjectSourceSubversion) 

Modified: llvm/trunk/utils/buildit/build_llvm
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/build_llvm?rev=84507&r1=84506&r2=84507&view=diff

==============================================================================
--- llvm/trunk/utils/buildit/build_llvm (original)
+++ llvm/trunk/utils/buildit/build_llvm Mon Oct 19 12:53:54 2009
@@ -82,6 +82,10 @@
 rm -rf $SRC_DIR || exit 1
 mkdir $SRC_DIR || exit 1
 ln -s $ORIG_SRC_DIR/* $SRC_DIR/ || exit 1
+# We can't use the top-level Makefile as-is.  Remove the soft link:
+rm $SRC_DIR/Makefile || exit 1
+# Now create our own by editing the top-level Makefile, deleting every line marked "Apple-style":
+sed -e '/[Aa]pple-style/d' -e '/include.*GNUmakefile/d' $ORIG_SRC_DIR/Makefile > $SRC_DIR/Makefile || exit 1
 
 # Build the LLVM tree universal.
 mkdir -p $DIR/obj-llvm || exit 1




From sabre at nondot.org  Mon Oct 19 12:59:20 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 17:59:20 -0000
Subject: [llvm-commits] [llvm] r84508 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910191759.n9JHxKKu007043@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 12:59:19 2009
New Revision: 84508

URL: http://llvm.org/viewvc/llvm-project?rev=84508&view=rev
Log:
cleanup doFinalization -> EmitEndOfAsmFile.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84508&r1=84507&r2=84508&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 12:59:19 2009
@@ -38,7 +38,6 @@
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringSet.h"
-#include "llvm/Support/Compiler.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/Mangler.h"
 #include "llvm/Support/MathExtras.h"
@@ -49,7 +48,7 @@
 STATISTIC(EmittedInsts, "Number of machine instrs printed");
 
 namespace {
-  class VISIBILITY_HIDDEN ARMAsmPrinter : public AsmPrinter {
+  class ARMAsmPrinter : public AsmPrinter {
 
     /// Subtarget - Keep a pointer to the ARMSubtarget around so that we can
     /// make the right decision when printing asm code for different targets.
@@ -149,8 +148,8 @@
 
     void printMachineInstruction(const MachineInstr *MI);
     bool runOnMachineFunction(MachineFunction &F);
-    bool doFinalization(Module &M);
     void EmitStartOfAsmFile(Module &M);
+    void EmitEndOfAsmFile(Module &M);
 
     /// EmitMachineConstantPoolValue - Print a machine constantpool value to
     /// the .s file.
@@ -1256,7 +1255,7 @@
 }
 
 
-bool ARMAsmPrinter::doFinalization(Module &M) {
+void ARMAsmPrinter::EmitEndOfAsmFile(Module &M) {
   if (Subtarget->isTargetDarwin()) {
     // All darwin targets use mach-o.
     TargetLoweringObjectFileMachO &TLOFMacho =
@@ -1294,8 +1293,6 @@
     // generates code that does this, it is always safe to set.
     O << "\t.subsections_via_symbols\n";
   }
-
-  return AsmPrinter::doFinalization(M);
 }
 
 // Force static initialization.




From sabre at nondot.org  Mon Oct 19 13:03:08 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 18:03:08 -0000
Subject: [llvm-commits] [llvm] r84509 - in /llvm/trunk/lib/Target:
 ARM/AsmPrinter/ARMAsmPrinter.cpp PowerPC/AsmPrinter/PPCAsmPrinter.cpp
 X86/AsmPrinter/X86AsmPrinter.cpp
Message-ID: <200910191803.n9JI38HF007227@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 13:03:08 2009
New Revision: 84509

URL: http://llvm.org/viewvc/llvm-project?rev=84509&view=rev
Log:
emit .subsections_via_symbols through MCStreamer instead of textually.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/PowerPC/AsmPrinter/PPCAsmPrinter.cpp
    llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84509&r1=84508&r2=84509&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 13:03:08 2009
@@ -1291,7 +1291,7 @@
     // implementation of multiple entry points).  If this doesn't occur, the
     // linker can safely perform dead code stripping.  Since LLVM never
     // generates code that does this, it is always safe to set.
-    O << "\t.subsections_via_symbols\n";
+    OutStreamer.EmitAssemblerFlag(MCStreamer::SubsectionsViaSymbols);
   }
 }
 

Modified: llvm/trunk/lib/Target/PowerPC/AsmPrinter/PPCAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/AsmPrinter/PPCAsmPrinter.cpp?rev=84509&r1=84508&r2=84509&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PowerPC/AsmPrinter/PPCAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/PowerPC/AsmPrinter/PPCAsmPrinter.cpp Mon Oct 19 13:03:08 2009
@@ -1129,7 +1129,7 @@
   // implementation of multiple entry points).  If this doesn't occur, the
   // linker can safely perform dead code stripping.  Since LLVM never generates
   // code that does this, it is always safe to set.
-  O << "\t.subsections_via_symbols\n";
+  OutStreamer.EmitAssemblerFlag(MCStreamer::SubsectionsViaSymbols);
 
   return AsmPrinter::doFinalization(M);
 }

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp?rev=84509&r1=84508&r2=84509&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp Mon Oct 19 13:03:08 2009
@@ -870,7 +870,9 @@
     // implementation of multiple entry points).  If this doesn't occur, the
     // linker can safely perform dead code stripping.  Since LLVM never
     // generates code that does this, it is always safe to set.
-    O << "\t.subsections_via_symbols\n";
+    //O << "\t.subsections_via_symbols\n";
+    
+    OutStreamer.EmitAssemblerFlag(MCStreamer::SubsectionsViaSymbols);
   }
 
   if (Subtarget->isTargetCOFF()) {




From sabre at nondot.org  Mon Oct 19 13:03:41 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 18:03:41 -0000
Subject: [llvm-commits] [llvm] r84510 -
	/llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp
Message-ID: <200910191803.n9JI3f3V007270@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 13:03:41 2009
New Revision: 84510

URL: http://llvm.org/viewvc/llvm-project?rev=84510&view=rev
Log:
remove accidental comment.

Modified:
    llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp?rev=84510&r1=84509&r2=84510&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp Mon Oct 19 13:03:41 2009
@@ -870,8 +870,6 @@
     // implementation of multiple entry points).  If this doesn't occur, the
     // linker can safely perform dead code stripping.  Since LLVM never
     // generates code that does this, it is always safe to set.
-    //O << "\t.subsections_via_symbols\n";
-    
     OutStreamer.EmitAssemblerFlag(MCStreamer::SubsectionsViaSymbols);
   }
 




From evan.cheng at apple.com  Mon Oct 19 13:04:16 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Mon, 19 Oct 2009 11:04:16 -0700
Subject: [llvm-commits] [PATCH] Sanitize the JITResolver
	stub/callsite<->function maps
In-Reply-To: <bb55930e0910191045w791e58ddx2aac279dadabdbd4@mail.gmail.com>
References: <0016367b65c6b96c230475ed8580@google.com>
	<bb55930e0910191045w791e58ddx2aac279dadabdbd4@mail.gmail.com>
Message-ID: <41EFED7B-8CA5-46A6-8E8D-AD7F023C3942@apple.com>

It looks very reasonable to me.

Evan

On Oct 19, 2009, at 10:45 AM, Jeffrey Yasskin wrote:

> Evan, Chris says you're one of the right people to take a look at
> this. Could you?
>
> Thanks,
> Jeffrey
>
> On Wed, Oct 14, 2009 at 4:28 PM,  <jyasskin at gmail.com> wrote:
>> Reviewers: ,
>>
>> Message:
>> Patch at http://codereview.appspot.com/download/issue130080_1.diff
>>
>> This passes check-lit.
>>
>> Description:
>> The JITResolver maps Functions to their canonical stubs and all
>> callsites for lazily-compiled functions to their target Functions. To
>> make Function destruction work, I'm going to need to remove all
>> callsites on destruction, so this patch also adds the reverse mapping
>> for that.
>>
>> There was an incorrect assumption in here that the only stub for a
>> function would be the one caused by needing to lazily compile it,  
>> while
>> x86-64 far calls and dlsym-stubs could also cause such stubs, but I
>> didn't look for a test case that the assumption broke.
>>
>> This also adds DenseMapInfo<AssertingVH> so I can use DenseMaps  
>> instead
>> of std::maps.
>>
>> Please review this at http://codereview.appspot.com/130080
>>
>> Affected files:
>>   M include/llvm/Support/ValueHandle.h
>>   M lib/ExecutionEngine/JIT/JITEmitter.cpp
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>



From sabre at nondot.org  Mon Oct 19 13:08:03 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 18:08:03 -0000
Subject: [llvm-commits] [llvm] r84512 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910191808.n9JI83L6007549@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 13:08:02 2009
New Revision: 84512

URL: http://llvm.org/viewvc/llvm-project?rev=84512&view=rev
Log:
don't bother trying to avoid emitting redundant constant pool alignment directives.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84512&r1=84511&r2=84512&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 13:08:02 2009
@@ -77,13 +77,10 @@
     /// visibility that require non-lazy-pointers for indirect access.
     StringMap<std::string> HiddenGVNonLazyPtrs;
 
-    /// True if asm printer is printing a series of CONSTPOOL_ENTRY.
-    bool InCPMode;
   public:
     explicit ARMAsmPrinter(formatted_raw_ostream &O, TargetMachine &TM,
                            const MCAsmInfo *T, bool V)
-      : AsmPrinter(O, TM, T, V), AFI(NULL), MCP(NULL),
-        InCPMode(false) {
+      : AsmPrinter(O, TM, T, V), AFI(NULL), MCP(NULL) {
       Subtarget = &TM.getSubtarget<ARMSubtarget>();
     }
 
@@ -259,7 +256,6 @@
     if (Subtarget->isTargetDarwin())
       O << "\t" << CurrentFnName;
     O << "\n";
-    InCPMode = false;
   } else {
     EmitAlignment(FnAlign, F);
   }
@@ -1027,17 +1023,8 @@
   ++EmittedInsts;
 
   int Opc = MI->getOpcode();
-  switch (Opc) {
-  case ARM::CONSTPOOL_ENTRY:
-    if (!InCPMode && AFI->isThumbFunction()) {
-      EmitAlignment(2);
-      InCPMode = true;
-    }
-    break;
-  default: {
-    if (InCPMode && AFI->isThumbFunction())
-      InCPMode = false;
-  }}
+  if (Opc == ARM::CONSTPOOL_ENTRY)
+    EmitAlignment(2);
 
   // Call the autogenerated instruction printer routines.
   processDebugLoc(MI, true);




From sabre at nondot.org  Mon Oct 19 13:11:25 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 18:11:25 -0000
Subject: [llvm-commits] [llvm] r84513 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910191811.n9JIBP2N007764@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 13:11:25 2009
New Revision: 84513

URL: http://llvm.org/viewvc/llvm-project?rev=84513&view=rev
Log:
remove dead map

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84513&r1=84512&r2=84513&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 13:11:25 2009
@@ -62,13 +62,6 @@
     /// MachineFunction.
     const MachineConstantPool *MCP;
 
-    /// We name each basic block in a Function with a unique number, so
-    /// that we can consistently refer to them later. This is cleared
-    /// at the beginning of each call to runOnMachineFunction().
-    ///
-    typedef std::map<const Value *, unsigned> ValueMapTy;
-    ValueMapTy NumberForBB;
-
     /// GVNonLazyPtrs - Keeps the set of GlobalValues that require
     /// non-lazy-pointers for indirect access.
     StringMap<std::string> GVNonLazyPtrs;




From sabre at nondot.org  Mon Oct 19 13:18:07 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 18:18:07 -0000
Subject: [llvm-commits] [llvm] r84515 - in /llvm/trunk/test/CodeGen/ARM:
	globals.ll load-global.ll
Message-ID: <200910191818.n9JII7Qw008067@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 13:18:07 2009
New Revision: 84515

URL: http://llvm.org/viewvc/llvm-project?rev=84515&view=rev
Log:
rename test

Added:
    llvm/trunk/test/CodeGen/ARM/globals.ll
      - copied unchanged from r84505, llvm/trunk/test/CodeGen/ARM/load-global.ll
Removed:
    llvm/trunk/test/CodeGen/ARM/load-global.ll

Removed: llvm/trunk/test/CodeGen/ARM/load-global.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/load-global.ll?rev=84514&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/ARM/load-global.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/load-global.ll (removed)
@@ -1,15 +0,0 @@
-; RUN: llc < %s -mtriple=arm-apple-darwin -relocation-model=static | \
-; RUN:   not grep {L_G\$non_lazy_ptr}
-; RUN: llc < %s -mtriple=arm-apple-darwin -relocation-model=dynamic-no-pic | \
-; RUN:   grep {L_G\$non_lazy_ptr} | count 2
-; RUN: llc < %s -mtriple=arm-apple-darwin -relocation-model=pic | \
-; RUN:   grep {ldr.*pc} | count 1
-; RUN: llc < %s -mtriple=arm-linux-gnueabi -relocation-model=pic | \
-; RUN:   grep {GOT} | count 1
-
- at G = external global i32
-
-define i32 @test1() {
-	%tmp = load i32* @G
-	ret i32 %tmp
-}




From asl at math.spbu.ru  Mon Oct 19 13:21:10 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Mon, 19 Oct 2009 18:21:10 -0000
Subject: [llvm-commits] [llvm] r84516 - in /llvm/trunk:
 lib/CodeGen/CodePlacementOpt.cpp
 test/CodeGen/X86/2009-04-20-LinearScanOpt.ll
 test/CodeGen/X86/loop_blocks.ll
Message-ID: <200910191821.n9JILArY008185@zion.cs.uiuc.edu>

Author: asl
Date: Mon Oct 19 13:21:09 2009
New Revision: 84516

URL: http://llvm.org/viewvc/llvm-project?rev=84516&view=rev
Log:
Revert r84295, this unbreaks llvm-gcc bootstrap on x86-64/linux

Removed:
    llvm/trunk/test/CodeGen/X86/loop_blocks.ll
Modified:
    llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp
    llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll

Modified: llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp?rev=84516&r1=84515&r2=84516&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp (original)
+++ llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp Mon Oct 19 13:21:09 2009
@@ -34,6 +34,14 @@
     const TargetInstrInfo *TII;
     const TargetLowering  *TLI;
 
+    /// ChangedMBBs - BBs which are modified by OptimizeIntraLoopEdges.
+    SmallPtrSet<MachineBasicBlock*, 8> ChangedMBBs;
+
+    /// UncondJmpMBBs - A list of BBs which are in loops and end with
+    /// unconditional branches.
+    SmallVector<std::pair<MachineBasicBlock*,MachineBasicBlock*>, 4>
+    UncondJmpMBBs;
+
   public:
     static char ID;
     CodePlacementOpt() : MachineFunctionPass(&ID) {}
@@ -50,15 +58,7 @@
     }
 
   private:
-    bool HasFallthrough(MachineBasicBlock *MBB);
-    bool HasAnalyzableTerminator(MachineBasicBlock *MBB);
-    void Splice(MachineFunction &MF,
-                MachineFunction::iterator InsertPt,
-                MachineFunction::iterator Begin,
-                MachineFunction::iterator End);
-    void UpdateTerminator(MachineBasicBlock *MBB);
-    bool OptimizeIntraLoopEdges(MachineFunction &MF);
-    bool OptimizeIntraLoopEdgesInLoop(MachineFunction &MF, MachineLoop *L);
+    bool OptimizeIntraLoopEdges();
     bool AlignLoops(MachineFunction &MF);
     bool AlignLoop(MachineFunction &MF, MachineLoop *L, unsigned Align);
   };
@@ -70,365 +70,168 @@
   return new CodePlacementOpt();
 }
 
-/// HasFallthrough - Test whether the given branch has a fallthrough, either as
-/// a plain fallthrough or as a fallthrough case of a conditional branch.
+/// OptimizeBackEdges - Place loop back edges to move unconditional branches
+/// out of the loop.
 ///
-bool CodePlacementOpt::HasFallthrough(MachineBasicBlock *MBB) {
-  MachineBasicBlock *TBB = 0, *FBB = 0;
-  SmallVector<MachineOperand, 4> Cond;
-  if (TII->AnalyzeBranch(*MBB, TBB, FBB, Cond))
-    return false;
-  // This conditional branch has no fallthrough.
-  if (FBB)
-    return false;
-  // An unconditional branch has no fallthrough.
-  if (Cond.empty() && TBB)
-    return false;
-  // It has a fallthrough.
-  return true;
-}
-
-/// HasAnalyzableTerminator - Test whether AnalyzeBranch will succeed on MBB.
-/// This is called before major changes are begun to test whether it will be
-/// possible to complete the changes.
+///       A:
+///       ...
+///       <fallthrough to B>
 ///
-/// Target-specific code is hereby encouraged to make AnalyzeBranch succeed
-/// whenever possible.
+///       B:  --> loop header
+///       ...
+///       jcc <cond> C, [exit]
 ///
-bool CodePlacementOpt::HasAnalyzableTerminator(MachineBasicBlock *MBB) {
-  // Conservatively ignore EH landing pads.
-  if (MBB->isLandingPad()) return false;
-
-  // Ignore blocks which look like they might have EH-related control flow.
-  // At the time of this writing, there are blocks which AnalyzeBranch
-  // thinks end in single uncoditional branches, yet which have two CFG
-  // successors. Code in this file is not prepared to reason about such things.
-  if (!MBB->empty() && MBB->back().getOpcode() == TargetInstrInfo::EH_LABEL)
-    return false;
-
-  // Aggressively handle return blocks and similar constructs.
-  if (MBB->succ_empty()) return true;
-
-  // Ask the target's AnalyzeBranch if it can handle this block.
-  MachineBasicBlock *TBB = 0, *FBB = 0;
-  SmallVector<MachineOperand, 4> Cond;
-  // Make the the terminator is understood.
-  if (TII->AnalyzeBranch(*MBB, TBB, FBB, Cond))
-    return false;
-  // Make sure we have the option of reversing the condition.
-  if (!Cond.empty() && TII->ReverseBranchCondition(Cond))
-    return false;
-  return true;
-}
-
-/// Splice - Move the sequence of instructions [Begin,End) to just before
-/// InsertPt. Update branch instructions as needed to account for broken
-/// fallthrough edges and to take advantage of newly exposed fallthrough
-/// opportunities.
+///       C:
+///       ...
+///       jmp B
 ///
-void CodePlacementOpt::Splice(MachineFunction &MF,
-                              MachineFunction::iterator InsertPt,
-                              MachineFunction::iterator Begin,
-                              MachineFunction::iterator End) {
-  assert(Begin != MF.begin() && End != MF.begin() && InsertPt != MF.begin() &&
-         "Splice can't change the entry block!");
-  MachineFunction::iterator OldBeginPrior = prior(Begin);
-  MachineFunction::iterator OldEndPrior = prior(End);
-
-  MF.splice(InsertPt, Begin, End);
-
-  UpdateTerminator(prior(Begin));
-  UpdateTerminator(OldBeginPrior);
-  UpdateTerminator(OldEndPrior);
-}
-
-/// UpdateTerminator - Update the terminator instructions in MBB to account
-/// for changes to the layout. If the block previously used a fallthrough,
-/// it may now need a branch, and if it previously used branching it may now
-/// be able to use a fallthrough.
+/// ==>
 ///
-void CodePlacementOpt::UpdateTerminator(MachineBasicBlock *MBB) {
-  // A block with no successors has no concerns with fall-through edges.
-  if (MBB->succ_empty()) return;
-
-  MachineBasicBlock *TBB = 0, *FBB = 0;
-  SmallVector<MachineOperand, 4> Cond;
-  bool B = TII->AnalyzeBranch(*MBB, TBB, FBB, Cond);
-  (void) B;
-  assert(!B && "UpdateTerminators requires analyzable predecessors!");
-  if (Cond.empty()) {
-    if (TBB) {
-      // The block has an unconditional branch. If its successor is now
-      // its layout successor, delete the branch.
-      if (MBB->isLayoutSuccessor(TBB))
-        TII->RemoveBranch(*MBB);
-    } else {
-      // The block has an unconditional fallthrough. If its successor is not
-      // its layout successor, insert a branch.
-      TBB = *MBB->succ_begin();
-      if (!MBB->isLayoutSuccessor(TBB))
-        TII->InsertBranch(*MBB, TBB, 0, Cond);
-    }
-  } else {
-    if (FBB) {
-      // The block has a non-fallthrough conditional branch. If one of its
-      // successors is its layout successor, rewrite it to a fallthrough
-      // conditional branch.
-      if (MBB->isLayoutSuccessor(TBB)) {
-        TII->RemoveBranch(*MBB);
-        TII->ReverseBranchCondition(Cond);
-        TII->InsertBranch(*MBB, FBB, 0, Cond);
-      } else if (MBB->isLayoutSuccessor(FBB)) {
-        TII->RemoveBranch(*MBB);
-        TII->InsertBranch(*MBB, TBB, 0, Cond);
-      }
-    } else {
-      // The block has a fallthrough conditional branch.
-      MachineBasicBlock *MBBA = *MBB->succ_begin();
-      MachineBasicBlock *MBBB = *next(MBB->succ_begin());
-      if (MBBA == TBB) std::swap(MBBB, MBBA);
-      if (MBB->isLayoutSuccessor(TBB)) {
-        TII->RemoveBranch(*MBB);
-        TII->ReverseBranchCondition(Cond);
-        TII->InsertBranch(*MBB, MBBA, 0, Cond);
-      } else if (!MBB->isLayoutSuccessor(MBBA)) {
-        TII->RemoveBranch(*MBB);
-        TII->InsertBranch(*MBB, TBB, MBBA, Cond);
-      }
-    }
-  }
-}
-
-/// OptimizeIntraLoopEdges - Reposition loop blocks to minimize
-/// intra-loop branching and to form contiguous loops.
+///       A:
+///       ...
+///       jmp B
 ///
-bool CodePlacementOpt::OptimizeIntraLoopEdges(MachineFunction &MF) {
-  bool Changed = false;
-
+///       C:
+///       ...
+///       <fallthough to B>
+///       
+///       B:  --> loop header
+///       ...
+///       jcc <cond> C, [exit]
+///
+bool CodePlacementOpt::OptimizeIntraLoopEdges() {
   if (!TLI->shouldOptimizeCodePlacement())
-    return Changed;
-
-  for (MachineLoopInfo::iterator I = MLI->begin(), E = MLI->end();
-       I != E; ++I)
-    Changed |= OptimizeIntraLoopEdgesInLoop(MF, *I);
-
-  return Changed;
-}
+    return false;
 
-/// OptimizeIntraLoopEdgesInLoop - Reposition loop blocks to minimize
-/// intra-loop branching and to form contiguous loops.
-///
-/// This code takes the approach of making minor changes to the existing
-/// layout to fix specific loop-oriented problems. Also, it depends on
-/// AnalyzeBranch, which can't understand complex control instructions.
-///
-bool CodePlacementOpt::OptimizeIntraLoopEdgesInLoop(MachineFunction &MF,
-                                                    MachineLoop *L) {
   bool Changed = false;
-
-  // Do optimization for nested loops.
-  for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I)
-    Changed |= OptimizeIntraLoopEdgesInLoop(MF, *I);
-
-  // Keep a record of which blocks are in the portion of the loop contiguous
-  // with the loop header.
-  SmallPtrSet<MachineBasicBlock *, 8> ContiguousBlocks;
-  ContiguousBlocks.insert(L->getHeader());
-
-  // Find the loop "top", ignoring any discontiguous parts.
-  MachineBasicBlock *TopMBB = L->getHeader();
-  if (TopMBB != MF.begin()) {
-    MachineBasicBlock *PriorMBB = prior(MachineFunction::iterator(TopMBB));
-    while (L->contains(PriorMBB)) {
-      ContiguousBlocks.insert(PriorMBB);
-      TopMBB = PriorMBB;
-      if (TopMBB == MF.begin()) break;
-      PriorMBB = prior(MachineFunction::iterator(TopMBB));
-    }
-  }
-
-  // Find the loop "bottom", ignoring any discontiguous parts.
-  MachineBasicBlock *BotMBB = L->getHeader();
-  if (BotMBB != prior(MF.end())) {
-    MachineBasicBlock *NextMBB = next(MachineFunction::iterator(BotMBB));
-    while (L->contains(NextMBB)) {
-      ContiguousBlocks.insert(NextMBB);
-      BotMBB = NextMBB;
-      if (BotMBB == next(MachineFunction::iterator(BotMBB))) break;
-      NextMBB = next(MachineFunction::iterator(BotMBB));
-    }
-  }
-
-  // First, move blocks which unconditionally jump to the loop top to the
-  // top of the loop so that they have a fall through. This can introduce a
-  // branch on entry to the loop, but it can eliminate a branch within the
-  // loop. See the @simple case in test/CodeGen/X86/loop_blocks.ll for an
-  // example of this.
-
-  bool BotHasFallthrough = HasFallthrough(BotMBB);
-
-  if (TopMBB == MF.begin() ||
-      HasAnalyzableTerminator(prior(MachineFunction::iterator(TopMBB)))) {
-  new_top:
-    for (MachineBasicBlock::pred_iterator PI = TopMBB->pred_begin(),
-         PE = TopMBB->pred_end(); PI != PE; ++PI) {
-      MachineBasicBlock *Pred = *PI;
-      if (Pred == TopMBB) continue;
-      if (HasFallthrough(Pred)) continue;
-      if (!L->contains(Pred)) continue;
-
-      // Verify that we can analyze all the loop entry edges before beginning
-      // any changes which will require us to be able to analyze them.
-      if (Pred == MF.begin())
-        continue;
-      if (!HasAnalyzableTerminator(Pred))
+  for (unsigned i = 0, e = UncondJmpMBBs.size(); i != e; ++i) {
+    MachineBasicBlock *MBB = UncondJmpMBBs[i].first;
+    MachineBasicBlock *SuccMBB = UncondJmpMBBs[i].second;
+    MachineLoop *L = MLI->getLoopFor(MBB);
+    assert(L && "BB is expected to be in a loop!");
+
+    if (ChangedMBBs.count(MBB)) {
+      // BB has been modified, re-analyze.
+      MachineBasicBlock *TBB = 0, *FBB = 0;
+      SmallVector<MachineOperand, 4> Cond;
+      if (TII->AnalyzeBranch(*MBB, TBB, FBB, Cond) || !Cond.empty())
         continue;
-      if (!HasAnalyzableTerminator(prior(MachineFunction::iterator(Pred))))
+      if (MLI->getLoopFor(TBB) != L || TBB->isLandingPad())
         continue;
+      SuccMBB = TBB;
+    } else {
+      assert(MLI->getLoopFor(SuccMBB) == L &&
+             "Successor is not in the same loop!");
+    }
 
-      // Move the block.
+    if (MBB->isLayoutSuccessor(SuccMBB)) {
+      // Successor is right after MBB, just eliminate the unconditional jmp.
+      // Can this happen?
+      TII->RemoveBranch(*MBB);
+      ChangedMBBs.insert(MBB);
+      ++NumIntraElim;
       Changed = true;
-      ContiguousBlocks.insert(Pred);
+      continue;
+    }
 
-      // Move it and all the blocks that can reach it via fallthrough edges
-      // exclusively, to keep existing fallthrough-edges intact.
-      MachineFunction::iterator Begin = Pred;
-      MachineFunction::iterator End = next(Begin);
-      while (Begin != MF.begin()) {
-        MachineFunction::iterator Prior = prior(Begin);
-        if (Prior == MF.begin())
-          break;
-        // Stop when a non-fallthrough edge is found.
-        if (!HasFallthrough(Prior))
-          break;
-        // Stop if a block which could fall-through out of the loop is found.
-        if (Prior->isSuccessor(End))
-          break;
-        // If we've reached the top, stop scanning.
-        if (Prior == MachineFunction::iterator(TopMBB)) {
-          // We know top currently has a fall through (because we just checked
-          // it) which would be lost if we do the transformation, so it isn't
-          // worthwhile to do the transformation unless it would expose a new
-          // fallthrough edge.
-          if (!Prior->isSuccessor(End))
-            goto next_pred;
-          // Otherwise we can stop scanning and procede to move the blocks.
+    // Now check if the predecessor is fallthrough from any BB. If there is,
+    // that BB should be from outside the loop since edge will become a jmp.
+    bool OkToMove = true;
+    MachineBasicBlock *FtMBB = 0, *FtTBB = 0, *FtFBB = 0;
+    SmallVector<MachineOperand, 4> FtCond;    
+    for (MachineBasicBlock::pred_iterator PI = SuccMBB->pred_begin(),
+           PE = SuccMBB->pred_end(); PI != PE; ++PI) {
+      MachineBasicBlock *PredMBB = *PI;
+      if (PredMBB->isLayoutSuccessor(SuccMBB)) {
+        if (TII->AnalyzeBranch(*PredMBB, FtTBB, FtFBB, FtCond)) {
+          OkToMove = false;
           break;
         }
-        // If we hit a switch or something complicated, don't move anything
-        // for this predecessor.
-        if (!HasAnalyzableTerminator(prior(MachineFunction::iterator(Prior))))
-          break;
-        Begin = Prior;
-        ContiguousBlocks.insert(Begin);
-        ++NumIntraMoved;
-      }
+        if (!FtTBB)
+          FtTBB = SuccMBB;
+        else if (!FtFBB) {
+          assert(FtFBB != SuccMBB && "Unexpected control flow!");
+          FtFBB = SuccMBB;
+        }
+        
+        // A fallthrough.
+        FtMBB = PredMBB;
+        MachineLoop *PL = MLI->getLoopFor(PredMBB);
+        if (PL && (PL == L || PL->getLoopDepth() >= L->getLoopDepth()))
+          OkToMove = false;
 
-      // Update BotMBB, before moving Begin/End around and forgetting where
-      // the new bottom is.
-      if (BotMBB == prior(End))
-        BotMBB = prior(Begin);
-
-      // Move the blocks.
-      Splice(MF, TopMBB, Begin, End);
-
-      // Update TopMBB, now that all the updates requiring the old top are
-      // complete.
-      TopMBB = Begin;
-
-      // We have a new loop top. Iterate on it. We shouldn't have to do this
-      // too many times if BranchFolding has done a reasonable job.
-      goto new_top;
-    next_pred:;
+        break;
+      }
     }
-  }
 
-  // If the loop previously didn't exit with a fall-through and it now does,
-  // we eliminated a branch.
-  if (!BotHasFallthrough && HasFallthrough(BotMBB)) {
-    ++NumIntraElim;
-    BotHasFallthrough = true;
-  }
+    if (!OkToMove)
+      continue;
 
-  // Next, move any loop blocks that are not in the portion of the loop
-  // contiguous with the header. This makes the loop contiguous, provided that
-  // AnalyzeBranch can handle all the relevant branching. See the @cfg_islands
-  // case in test/CodeGen/X86/loop_blocks.ll for an example of this.
-
-  // Determine a position to move orphaned loop blocks to. If TopMBB is not
-  // entered via fallthrough and BotMBB is exited via fallthrough, prepend them
-  // to the top of the loop to avoid loosing that fallthrough. Otherwise append
-  // them to the bottom, even if it previously had a fallthrough, on the theory
-  // that it's worth an extra branch to keep the loop contiguous.
-  MachineFunction::iterator InsertPt = next(MachineFunction::iterator(BotMBB));
-  bool InsertAtTop = false;
-  if (TopMBB != MF.begin() &&
-      !HasFallthrough(prior(MachineFunction::iterator(TopMBB))) &&
-      HasFallthrough(BotMBB)) {
-    InsertPt = TopMBB;
-    InsertAtTop = true;
-  }
-
-  // Find non-contigous blocks and fix them.
-  if (InsertPt != MF.begin() && HasAnalyzableTerminator(prior(InsertPt)))
-    for (MachineLoop::block_iterator BI = L->block_begin(), BE = L->block_end();
-         BI != BE; ++BI) {
-      MachineBasicBlock *BB = *BI;
-
-      // Verify that we can analyze all the loop entry edges before beginning
-      // any changes which will require us to be able to analyze them.
-      if (!HasAnalyzableTerminator(BB))
-        continue;
-      if (!HasAnalyzableTerminator(prior(MachineFunction::iterator(BB))))
-        continue;
-
-      // If the layout predecessor is part of the loop, this block will be
-      // processed along with it. This keeps them in their relative order.
-      if (BB != MF.begin() &&
-          L->contains(prior(MachineFunction::iterator(BB))))
-        continue;
-
-      // Check to see if this block is already contiguous with the main
-      // portion of the loop.
-      if (!ContiguousBlocks.insert(BB))
-        continue;
-
-      // Move the block.
-      Changed = true;
-
-      // Process this block and all loop blocks contiguous with it, to keep
-      // them in their relative order.
-      MachineFunction::iterator Begin = BB;
-      MachineFunction::iterator End = next(MachineFunction::iterator(BB));
-      for (; End != MF.end(); ++End) {
-        if (!L->contains(End)) break;
-        if (!HasAnalyzableTerminator(End)) break;
-        ContiguousBlocks.insert(End);
-        ++NumIntraMoved;
+    // Is it profitable? If SuccMBB can fallthrough itself, that can be changed
+    // into a jmp.
+    MachineBasicBlock *TBB = 0, *FBB = 0;
+    SmallVector<MachineOperand, 4> Cond;
+    if (TII->AnalyzeBranch(*SuccMBB, TBB, FBB, Cond))
+      continue;
+    if (!TBB && Cond.empty())
+      TBB = next(MachineFunction::iterator(SuccMBB));
+    else if (!FBB && !Cond.empty())
+      FBB = next(MachineFunction::iterator(SuccMBB));
+
+    // This calculate the cost of the transformation. Also, it finds the *only*
+    // intra-loop edge if there is one.
+    int Cost = 0;
+    bool HasOneIntraSucc = true;
+    MachineBasicBlock *IntraSucc = 0;
+    for (MachineBasicBlock::succ_iterator SI = SuccMBB->succ_begin(),
+           SE = SuccMBB->succ_end(); SI != SE; ++SI) {
+      MachineBasicBlock *SSMBB = *SI;
+      if (MLI->getLoopFor(SSMBB) == L) {
+        if (!IntraSucc)
+          IntraSucc = SSMBB;
+        else
+          HasOneIntraSucc = false;
       }
 
-      // Update BotMBB.
-      if (!InsertAtTop)
-        BotMBB = prior(End);
-
-      // If we're inserting at the bottom of the loop, and the code we're
-      // moving originally had fall-through successors, bring the sucessors
-      // up with the loop blocks to preserve the fall-through edges.
-      if (!InsertAtTop)
-        for (; End != MF.end(); ++End) {
-          if (L->contains(End)) break;
-          if (!HasAnalyzableTerminator(End)) break;
-          if (!HasFallthrough(prior(End))) break;
+      if (SuccMBB->isLayoutSuccessor(SSMBB))
+        // This will become a jmp.
+        ++Cost;
+      else if (MBB->isLayoutSuccessor(SSMBB)) {
+        // One of the successor will become the new fallthrough.
+        if (SSMBB == FBB) {
+          FBB = 0;
+          --Cost;
+        } else if (!FBB && SSMBB == TBB && Cond.empty()) {
+          TBB = 0;
+          --Cost;
+        } else if (!Cond.empty() && !TII->ReverseBranchCondition(Cond)) {
+          assert(SSMBB == TBB);
+          TBB = FBB;
+          FBB = 0;
+          --Cost;
         }
+      }
+    }
+    if (Cost)
+      continue;
 
-      // Move the blocks.
-      Splice(MF, InsertPt, Begin, End);
-
-      // Update TopMBB.
-      if (InsertAtTop)
-        TopMBB = Begin;
+    // Now, let's move the successor to below the BB to eliminate the jmp.
+    SuccMBB->moveAfter(MBB);
+    TII->RemoveBranch(*MBB);
+    TII->RemoveBranch(*SuccMBB);
+    if (TBB)
+      TII->InsertBranch(*SuccMBB, TBB, FBB, Cond);
+    ChangedMBBs.insert(MBB);
+    ChangedMBBs.insert(SuccMBB);
+    if (FtMBB) {
+      TII->RemoveBranch(*FtMBB);
+      TII->InsertBranch(*FtMBB, FtTBB, FtFBB, FtCond);
+      ChangedMBBs.insert(FtMBB);
     }
+    Changed = true;
+  }
 
+  ++NumIntraMoved;
   return Changed;
 }
 
@@ -452,8 +255,6 @@
   return Changed;
 }
 
-/// AlignLoop - Align loop headers to target preferred alignments.
-///
 bool CodePlacementOpt::AlignLoop(MachineFunction &MF, MachineLoop *L,
                                  unsigned Align) {
   bool Changed = false;
@@ -463,13 +264,13 @@
     Changed |= AlignLoop(MF, *I, Align);
 
   MachineBasicBlock *TopMBB = L->getHeader();
-  if (TopMBB != MF.begin()) {
-    MachineBasicBlock *PredMBB = prior(MachineFunction::iterator(TopMBB));
-    while (L->contains(PredMBB)) {
-      TopMBB = PredMBB;
-      if (TopMBB == MF.begin()) break;
-      PredMBB = prior(MachineFunction::iterator(TopMBB));
-    }
+  if (TopMBB == MF.begin()) return Changed;
+
+  MachineBasicBlock *PredMBB = prior(MachineFunction::iterator(TopMBB));
+  while (MLI->getLoopFor(PredMBB) == L) {
+    TopMBB = PredMBB;
+    if (TopMBB == MF.begin()) return Changed;
+    PredMBB = prior(MachineFunction::iterator(TopMBB));
   }
 
   TopMBB->setAlignment(Align);
@@ -487,9 +288,30 @@
   TLI = MF.getTarget().getTargetLowering();
   TII = MF.getTarget().getInstrInfo();
 
-  bool Changed = OptimizeIntraLoopEdges(MF);
+  // Analyze the BBs first and keep track of BBs that
+  // end with an unconditional jmp to another block in the same loop.
+  for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
+    MachineBasicBlock *MBB = I;
+    if (MBB->isLandingPad())
+      continue;
+    MachineLoop *L = MLI->getLoopFor(MBB);
+    if (!L)
+      continue;
+
+    MachineBasicBlock *TBB = 0, *FBB = 0;
+    SmallVector<MachineOperand, 4> Cond;
+    if (TII->AnalyzeBranch(*MBB, TBB, FBB, Cond) || !Cond.empty())
+      continue;
+    if (MLI->getLoopFor(TBB) == L && !TBB->isLandingPad())
+      UncondJmpMBBs.push_back(std::make_pair(MBB, TBB));
+  }
+
+  bool Changed = OptimizeIntraLoopEdges();
 
   Changed |= AlignLoops(MF);
 
+  ChangedMBBs.clear();
+  UncondJmpMBBs.clear();
+
   return Changed;
 }

Modified: llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll?rev=84516&r1=84515&r2=84516&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll (original)
+++ llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll Mon Oct 19 13:21:09 2009
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=x86_64-apple-darwin10.0 -relocation-model=pic -disable-fp-elim -stats |& grep asm-printer | grep 83
+; RUN: llc < %s -mtriple=x86_64-apple-darwin10.0 -relocation-model=pic -disable-fp-elim -stats |& grep asm-printer | grep 84
 ; rdar://6802189
 
 ; Test if linearscan is unfavoring registers for allocation to allow more reuse

Removed: llvm/trunk/test/CodeGen/X86/loop_blocks.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/loop_blocks.ll?rev=84515&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/X86/loop_blocks.ll (original)
+++ llvm/trunk/test/CodeGen/X86/loop_blocks.ll (removed)
@@ -1,207 +0,0 @@
-; RUN: llc < %s -march=x86-64 -mtriple=x86_64-unknown-linux-gnu -asm-verbose=false | FileCheck %s
-
-; These tests check for loop branching structure, and that the loop align
-; directive is placed in the expected place.
-
-; CodeGen should insert a branch into the middle of the loop in
-; order to avoid a branch within the loop.
-
-; CHECK: simple:
-;      CHECK:   jmp   .LBB1_1
-; CHECK-NEXT:   align
-; CHECK-NEXT: .LBB1_2:
-; CHECK-NEXT:   call loop_latch
-; CHECK-NEXT: .LBB1_1:
-; CHECK-NEXT:   call loop_header
-
-define void @simple() nounwind {
-entry:
-  br label %loop
-
-loop:
-  call void @loop_header()
-  %t0 = tail call i32 @get()
-  %t1 = icmp slt i32 %t0, 0
-  br i1 %t1, label %done, label %bb
-
-bb:
-  call void @loop_latch()
-  br label %loop
-
-done:
-  call void @exit()
-  ret void
-}
-
-; CodeGen should move block_a to the top of the loop so that it
-; falls through into the loop, avoiding a branch within the loop.
-
-; CHECK: slightly_more_involved:
-;      CHECK:   jmp .LBB2_1
-; CHECK-NEXT:   align
-; CHECK-NEXT: .LBB2_4:
-; CHECK-NEXT:   call bar99
-; CHECK-NEXT: .LBB2_1:
-; CHECK-NEXT:   call body
-
-define void @slightly_more_involved() nounwind {
-entry:
-  br label %loop
-
-loop:
-  call void @body()
-  %t0 = call i32 @get()
-  %t1 = icmp slt i32 %t0, 2
-  br i1 %t1, label %block_a, label %bb
-
-bb:
-  %t2 = call i32 @get()
-  %t3 = icmp slt i32 %t2, 99
-  br i1 %t3, label %exit, label %loop
-
-block_a:
-  call void @bar99()
-  br label %loop
-
-exit:
-  call void @exit()
-  ret void
-}
-
-; Same as slightly_more_involved, but block_a is now a CFG diamond with
-; fallthrough edges which should be preserved.
-
-; CHECK: yet_more_involved:
-;      CHECK:   jmp .LBB3_1
-; CHECK-NEXT:   align
-; CHECK-NEXT: .LBB3_7:
-; CHECK-NEXT:   call block_a_true_func
-; CHECK-NEXT:   jmp .LBB3_4
-; CHECK-NEXT: .LBB3_2:
-; CHECK-NEXT:   call bar99
-; CHECK-NEXT:   call get
-; CHECK-NEXT:   cmpl $2999, %eax
-; CHECK-NEXT:   jle .LBB3_7
-; CHECK-NEXT:   call block_a_false_func
-; CHECK-NEXT: .LBB3_4:
-; CHECK-NEXT:   call block_a_merge_func
-; CHECK-NEXT: .LBB3_1:
-; CHECK-NEXT:   call body
-
-define void @yet_more_involved() nounwind {
-entry:
-  br label %loop
-
-loop:
-  call void @body()
-  %t0 = call i32 @get()
-  %t1 = icmp slt i32 %t0, 2
-  br i1 %t1, label %block_a, label %bb
-
-bb:
-  %t2 = call i32 @get()
-  %t3 = icmp slt i32 %t2, 99
-  br i1 %t3, label %exit, label %loop
-
-block_a:
-  call void @bar99()
-  %z0 = call i32 @get()
-  %z1 = icmp slt i32 %z0, 3000
-  br i1 %z1, label %block_a_true, label %block_a_false
-
-block_a_true:
-  call void @block_a_true_func()
-  br label %block_a_merge
-
-block_a_false:
-  call void @block_a_false_func()
-  br label %block_a_merge
-
-block_a_merge:
-  call void @block_a_merge_func()
-  br label %loop
-
-exit:
-  call void @exit()
-  ret void
-}
-
-; CodeGen should move the CFG islands that are part of the loop but don't
-; conveniently fit anywhere so that they are at least contiguous with the
-; loop.
-
-; CHECK: cfg_islands:
-;      CHECK:   jmp     .LBB4_1
-; CHECK-NEXT:   align
-; CHECK-NEXT: .LBB4_7:
-; CHECK-NEXT:   call    bar100
-; CHECK-NEXT:   jmp     .LBB4_1
-; CHECK-NEXT: .LBB4_8:
-; CHECK-NEXT:   call    bar101
-; CHECK-NEXT:   jmp     .LBB4_1
-; CHECK-NEXT: .LBB4_9:
-; CHECK-NEXT:   call    bar102
-; CHECK-NEXT:   jmp     .LBB4_1
-; CHECK-NEXT: .LBB4_5:
-; CHECK-NEXT:   call    loop_latch
-; CHECK-NEXT: .LBB4_1:
-; CHECK-NEXT:   call    loop_header
-
-define void @cfg_islands() nounwind {
-entry:
-  br label %loop
-
-loop:
-  call void @loop_header()
-  %t0 = call i32 @get()
-  %t1 = icmp slt i32 %t0, 100
-  br i1 %t1, label %block100, label %bb
-
-bb:
-  %t2 = call i32 @get()
-  %t3 = icmp slt i32 %t2, 101
-  br i1 %t3, label %block101, label %bb1
-
-bb1:
-  %t4 = call i32 @get()
-  %t5 = icmp slt i32 %t4, 102
-  br i1 %t5, label %block102, label %bb2
-
-bb2:
-  %t6 = call i32 @get()
-  %t7 = icmp slt i32 %t6, 103
-  br i1 %t7, label %exit, label %bb3
-
-bb3:
-  call void @loop_latch()
-  br label %loop
-
-exit:
-  call void @exit()
-  ret void
-
-block100:
-  call void @bar100()
-  br label %loop
-
-block101:
-  call void @bar101()
-  br label %loop
-
-block102:
-  call void @bar102()
-  br label %loop
-}
-
-declare void @bar99() nounwind
-declare void @bar100() nounwind
-declare void @bar101() nounwind
-declare void @bar102() nounwind
-declare void @body() nounwind
-declare void @exit() nounwind
-declare void @loop_header() nounwind
-declare void @loop_latch() nounwind
-declare i32 @get() nounwind
-declare void @block_a_true_func() nounwind
-declare void @block_a_false_func() nounwind
-declare void @block_a_merge_func() nounwind




From sabre at nondot.org  Mon Oct 19 13:27:56 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 18:27:56 -0000
Subject: [llvm-commits] [llvm] r84517 -
	/llvm/trunk/test/CodeGen/ARM/globals.ll
Message-ID: <200910191827.n9JIRuok008468@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 13:27:56 2009
New Revision: 84517

URL: http://llvm.org/viewvc/llvm-project?rev=84517&view=rev
Log:
convert to filecheck syntax and make a lot more aggressive.

Modified:
    llvm/trunk/test/CodeGen/ARM/globals.ll

Modified: llvm/trunk/test/CodeGen/ARM/globals.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/globals.ll?rev=84517&r1=84516&r2=84517&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/ARM/globals.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/globals.ll Mon Oct 19 13:27:56 2009
@@ -1,11 +1,7 @@
-; RUN: llc < %s -mtriple=arm-apple-darwin -relocation-model=static | \
-; RUN:   not grep {L_G\$non_lazy_ptr}
-; RUN: llc < %s -mtriple=arm-apple-darwin -relocation-model=dynamic-no-pic | \
-; RUN:   grep {L_G\$non_lazy_ptr} | count 2
-; RUN: llc < %s -mtriple=arm-apple-darwin -relocation-model=pic | \
-; RUN:   grep {ldr.*pc} | count 1
-; RUN: llc < %s -mtriple=arm-linux-gnueabi -relocation-model=pic | \
-; RUN:   grep {GOT} | count 1
+; RUN: llc < %s -mtriple=arm-apple-darwin -relocation-model=static | FileCheck %s -check-prefix=DarwinStatic
+; RUN: llc < %s -mtriple=arm-apple-darwin -relocation-model=dynamic-no-pic | FileCheck %s -check-prefix=DarwinDynamic
+; RUN: llc < %s -mtriple=arm-apple-darwin -relocation-model=pic | FileCheck %s -check-prefix=DarwinPIC
+; RUN: llc < %s -mtriple=arm-linux-gnueabi -relocation-model=pic | FileCheck %s -check-prefix=LinuxPIC
 
 @G = external global i32
 
@@ -13,3 +9,67 @@
 	%tmp = load i32* @G
 	ret i32 %tmp
 }
+
+; DarwinStatic: _test1:
+; DarwinStatic: 	ldr r0, LCPI1_0
+; DarwinStatic:	        ldr r0, [r0]
+; DarwinStatic:	        bx lr
+
+; DarwinStatic: 	.align	2
+; DarwinStatic:	LCPI1_0:
+; DarwinStatic: 	.long	{{_G$}}
+
+
+; DarwinDynamic: _test1:
+; DarwinDynamic: 	ldr r0, LCPI1_0
+; DarwinDynamic:        ldr r0, [r0]
+; DarwinDynamic:        ldr r0, [r0]
+; DarwinDynamic:        bx lr
+
+; DarwinDynamic: 	.align	2
+; DarwinDynamic:	LCPI1_0:
+; DarwinDynamic: 	.long	L_G$non_lazy_ptr
+
+; DarwinDynamic: 	.section __DATA,__nl_symbol_ptr,non_lazy_symbol_pointers
+; DarwinDynamic:	.align	2
+; DarwinDynamic: L_G$non_lazy_ptr:
+; DarwinDynamic:	.indirect_symbol _G
+; DarwinDynamic:	.long	0
+
+
+
+; DarwinPIC: _test1:
+; DarwinPIC: 	ldr r0, LCPI1_0
+; DarwinPIC: LPC0:
+; DarwinPIC:    ldr r0, [pc, +r0]
+; DarwinPIC:    ldr r0, [r0]
+; DarwinPIC:    bx lr
+
+; DarwinPIC: 	.align	2
+; DarwinPIC: LCPI1_0:
+; DarwinPIC: 	.long	L_G$non_lazy_ptr-(LPC0+8)
+
+; DarwinPIC: 	.section __DATA,__nl_symbol_ptr,non_lazy_symbol_pointers
+; DarwinPIC:	.align	2
+; DarwinPIC: L_G$non_lazy_ptr:
+; DarwinPIC:	.indirect_symbol _G
+; DarwinPIC:	.long	0
+
+
+
+; LinuxPIC: test1:
+; LinuxPIC: 	ldr r0, .LCPI1_0
+; LinuxPIC: 	ldr r1, .LCPI1_1
+	
+; LinuxPIC: .LPC0:
+; LinuxPIC: 	add r0, pc, r0
+; LinuxPIC: 	ldr r0, [r1, +r0]
+; LinuxPIC: 	ldr r0, [r0]
+; LinuxPIC: 	bx lr
+
+; LinuxPIC: .align 2
+; LinuxPIC: .LCPI1_0:
+; LinuxPIC:     .long _GLOBAL_OFFSET_TABLE_-(.LPC0+8)
+; LinuxPIC: .align 2
+; LinuxPIC: .LCPI1_1:
+; LinuxPIC:     .long	G(GOT)




From sabre at nondot.org  Mon Oct 19 13:38:34 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 18:38:34 -0000
Subject: [llvm-commits] [llvm] r84519 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910191838.n9JIcYiE008923@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 13:38:33 2009
New Revision: 84519

URL: http://llvm.org/viewvc/llvm-project?rev=84519&view=rev
Log:
use MachineModuleInfoMachO for non-lazy gv stubs instead of a private map.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84519&r1=84518&r2=84519&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 13:38:33 2009
@@ -22,12 +22,13 @@
 #include "llvm/Assembly/Writer.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/DwarfWriter.h"
-#include "llvm/CodeGen/MachineModuleInfo.h"
+#include "llvm/CodeGen/MachineModuleInfoImpls.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCSectionMachO.h"
 #include "llvm/MC/MCStreamer.h"
-#include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCSymbol.h"
 #include "llvm/Target/TargetData.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
@@ -62,10 +63,6 @@
     /// MachineFunction.
     const MachineConstantPool *MCP;
 
-    /// GVNonLazyPtrs - Keeps the set of GlobalValues that require
-    /// non-lazy-pointers for indirect access.
-    StringMap<std::string> GVNonLazyPtrs;
-
     /// HiddenGVNonLazyPtrs - Keeps the set of GlobalValues with hidden
     /// visibility that require non-lazy-pointers for indirect access.
     StringMap<std::string> HiddenGVNonLazyPtrs;
@@ -166,8 +163,16 @@
           Name = Mang->getMangledName(GV, "$non_lazy_ptr", true);
           if (GV->hasHiddenVisibility())
             HiddenGVNonLazyPtrs[SymName] = Name;
-          else
-            GVNonLazyPtrs[SymName] = Name;
+          else {
+            MCSymbol *Sym = OutContext.GetOrCreateSymbol(Name.c_str());
+            const MCSymbol *&StubSym =
+              MMI->getObjFileInfo<MachineModuleInfoMachO>().getGVStubEntry(Sym);
+            if (StubSym == 0) {
+              //NameStr.clear();
+              //Mang->getNameWithPrefix(NameStr, GV, false);
+              StubSym = OutContext.GetOrCreateSymbol(SymName.c_str());
+            }
+          }
         }
       } else
         Name = Mang->makeNameProper(ACPV->getSymbol());
@@ -1240,19 +1245,23 @@
     // All darwin targets use mach-o.
     TargetLoweringObjectFileMachO &TLOFMacho =
       static_cast<TargetLoweringObjectFileMachO &>(getObjFileLowering());
+    MachineModuleInfoMachO &MMIMacho =
+      MMI->getObjFileInfo<MachineModuleInfoMachO>();
 
     O << '\n';
 
     // Output non-lazy-pointers for external and common global variables.
-    if (!GVNonLazyPtrs.empty()) {
+    MachineModuleInfoMachO::SymbolListTy Stubs = MMIMacho.GetGVStubList();
+    
+    if (!Stubs.empty()) {
       // Switch with ".non_lazy_symbol_pointer" directive.
       OutStreamer.SwitchSection(TLOFMacho.getNonLazySymbolPointerSection());
       EmitAlignment(2);
-      for (StringMap<std::string>::iterator I = GVNonLazyPtrs.begin(),
-           E = GVNonLazyPtrs.end(); I != E; ++I) {
-        O << I->second << ":\n";
-        O << "\t.indirect_symbol " << I->getKeyData() << "\n";
-        O << "\t.long\t0\n";
+      for (unsigned i = 0, e = Stubs.size(); i != e; ++i) {
+        Stubs[i].first->print(O, MAI);
+        O << ":\n\t.indirect_symbol ";
+        Stubs[i].second->print(O, MAI);
+        O << "\n\t.long\t0\n";
       }
     }
 




From sabre at nondot.org  Mon Oct 19 13:44:38 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 18:44:38 -0000
Subject: [llvm-commits] [llvm] r84520 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910191844.n9JIicHe009132@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 13:44:38 2009
New Revision: 84520

URL: http://llvm.org/viewvc/llvm-project?rev=84520&view=rev
Log:
switch hidden gv stubs to use MachineModuleInfoMachO instead of a custom map.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84520&r1=84519&r2=84520&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 13:44:38 2009
@@ -63,10 +63,6 @@
     /// MachineFunction.
     const MachineConstantPool *MCP;
 
-    /// HiddenGVNonLazyPtrs - Keeps the set of GlobalValues with hidden
-    /// visibility that require non-lazy-pointers for indirect access.
-    StringMap<std::string> HiddenGVNonLazyPtrs;
-
   public:
     explicit ARMAsmPrinter(formatted_raw_ostream &O, TargetMachine &TM,
                            const MCAsmInfo *T, bool V)
@@ -159,17 +155,26 @@
           Name = Mang->getMangledName(GV);
         else {
           // FIXME: Remove this when Darwin transition to @GOT like syntax.
-          std::string SymName = Mang->getMangledName(GV);
           Name = Mang->getMangledName(GV, "$non_lazy_ptr", true);
-          if (GV->hasHiddenVisibility())
-            HiddenGVNonLazyPtrs[SymName] = Name;
-          else {
-            MCSymbol *Sym = OutContext.GetOrCreateSymbol(Name.c_str());
+          MCSymbol *Sym = OutContext.GetOrCreateSymbol(Name.c_str());
+
+          if (GV->hasHiddenVisibility()) {
+            const MCSymbol *&StubSym =
+              MMI->getObjFileInfo<MachineModuleInfoMachO>()
+                       .getHiddenGVStubEntry(Sym);
+            if (StubSym == 0) {
+              //NameStr.clear();
+              //Mang->getNameWithPrefix(NameStr, GV, false);
+              std::string SymName = Mang->getMangledName(GV);
+              StubSym = OutContext.GetOrCreateSymbol(SymName.c_str());
+            }
+          } else {
             const MCSymbol *&StubSym =
               MMI->getObjFileInfo<MachineModuleInfoMachO>().getGVStubEntry(Sym);
             if (StubSym == 0) {
               //NameStr.clear();
               //Mang->getNameWithPrefix(NameStr, GV, false);
+              std::string SymName = Mang->getMangledName(GV);
               StubSym = OutContext.GetOrCreateSymbol(SymName.c_str());
             }
           }
@@ -1265,13 +1270,15 @@
       }
     }
 
-    if (!HiddenGVNonLazyPtrs.empty()) {
+    Stubs = MMIMacho.GetHiddenGVStubList();
+    if (!Stubs.empty()) {
       OutStreamer.SwitchSection(getObjFileLowering().getDataSection());
       EmitAlignment(2);
-      for (StringMap<std::string>::iterator I = HiddenGVNonLazyPtrs.begin(),
-             E = HiddenGVNonLazyPtrs.end(); I != E; ++I) {
-        O << I->second << ":\n";
-        O << "\t.long " << I->getKeyData() << "\n";
+      for (unsigned i = 0, e = Stubs.size(); i != e; ++i) {
+        Stubs[i].first->print(O, MAI);
+        O << ":\n\t.long ";
+        Stubs[i].second->print(O, MAI);
+        O << "\n";
       }
     }
 




From sabre at nondot.org  Mon Oct 19 13:49:15 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 18:49:15 -0000
Subject: [llvm-commits] [llvm] r84521 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910191849.n9JInFNH009272@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 13:49:14 2009
New Revision: 84521

URL: http://llvm.org/viewvc/llvm-project?rev=84521&view=rev
Log:
simplify code, reducing string thrashing.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84521&r1=84520&r2=84521&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 13:49:14 2009
@@ -157,26 +157,16 @@
           // FIXME: Remove this when Darwin transition to @GOT like syntax.
           Name = Mang->getMangledName(GV, "$non_lazy_ptr", true);
           MCSymbol *Sym = OutContext.GetOrCreateSymbol(Name.c_str());
-
-          if (GV->hasHiddenVisibility()) {
-            const MCSymbol *&StubSym =
-              MMI->getObjFileInfo<MachineModuleInfoMachO>()
-                       .getHiddenGVStubEntry(Sym);
-            if (StubSym == 0) {
-              //NameStr.clear();
-              //Mang->getNameWithPrefix(NameStr, GV, false);
-              std::string SymName = Mang->getMangledName(GV);
-              StubSym = OutContext.GetOrCreateSymbol(SymName.c_str());
-            }
-          } else {
-            const MCSymbol *&StubSym =
-              MMI->getObjFileInfo<MachineModuleInfoMachO>().getGVStubEntry(Sym);
-            if (StubSym == 0) {
-              //NameStr.clear();
-              //Mang->getNameWithPrefix(NameStr, GV, false);
-              std::string SymName = Mang->getMangledName(GV);
-              StubSym = OutContext.GetOrCreateSymbol(SymName.c_str());
-            }
+          
+          MachineModuleInfoMachO &MMIMachO =
+            MMI->getObjFileInfo<MachineModuleInfoMachO>();
+          const MCSymbol *&StubSym =
+            GV->hasHiddenVisibility() ? MMIMachO.getHiddenGVStubEntry(Sym) :
+                                        MMIMachO.getGVStubEntry(Sym);
+          if (StubSym == 0) {
+            SmallString<128> NameStr;
+            Mang->getNameWithPrefix(NameStr, GV, false);
+            StubSym = OutContext.GetOrCreateSymbol(NameStr.str());
           }
         }
       } else




From jyasskin at google.com  Mon Oct 19 13:49:59 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Mon, 19 Oct 2009 18:49:59 -0000
Subject: [llvm-commits] [llvm] r84522 - in /llvm/trunk:
 include/llvm/Support/ValueHandle.h lib/ExecutionEngine/JIT/JITEmitter.cpp
Message-ID: <200910191850.n9JIo0x3009322@zion.cs.uiuc.edu>

Author: jyasskin
Date: Mon Oct 19 13:49:59 2009
New Revision: 84522

URL: http://llvm.org/viewvc/llvm-project?rev=84522&view=rev
Log:
Clean up the JITResolver stub/callsite<->function maps.

The JITResolver maps Functions to their canonical stubs and all callsites for
lazily-compiled functions to their target Functions. To make Function
destruction work, I'm going to need to remove all callsites on destruction, so
this patch also adds the reverse mapping for that.

There was an incorrect assumption in here that the only stub for a function
would be the one caused by needing to lazily compile it, while x86-64 far calls
and dlsym-stubs could also cause such stubs, but I didn't look for a test case
that the assumption broke.

This also adds DenseMapInfo<AssertingVH> so I can use DenseMaps instead of
std::maps.

Modified:
    llvm/trunk/include/llvm/Support/ValueHandle.h
    llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp

Modified: llvm/trunk/include/llvm/Support/ValueHandle.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/ValueHandle.h?rev=84522&r1=84521&r2=84522&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Support/ValueHandle.h (original)
+++ llvm/trunk/include/llvm/Support/ValueHandle.h Mon Oct 19 13:49:59 2009
@@ -238,6 +238,31 @@
 template<> struct simplify_type<AssertingVH<Value> >
   : public simplify_type<const AssertingVH<Value> > {};
 
+// Specialize DenseMapInfo to allow AssertingVH to participate in DenseMap.
+template<typename T>
+struct DenseMapInfo<AssertingVH<T> > {
+  typedef DenseMapInfo<T*> PointerInfo;
+  static inline AssertingVH<T> getEmptyKey() {
+    return AssertingVH<T>(PointerInfo::getEmptyKey());
+  }
+  static inline T* getTombstoneKey() {
+    return AssertingVH<T>(PointerInfo::getTombstoneKey());
+  }
+  static unsigned getHashValue(const AssertingVH<T> &Val) {
+    return PointerInfo::getHashValue(Val);
+  }
+  static bool isEqual(const AssertingVH<T> &LHS, const AssertingVH<T> &RHS) {
+    return LHS == RHS;
+  }
+  static bool isPod() {
+#ifdef NDEBUG
+    return true;
+#else
+    return false;
+#endif
+  }
+};
+
 /// TrackingVH - This is a value handle that tracks a Value (or Value subclass),
 /// even across RAUW operations.
 ///

Modified: llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp?rev=84522&r1=84521&r2=84522&view=diff

==============================================================================
--- llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp (original)
+++ llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp Mon Oct 19 13:49:59 2009
@@ -63,17 +63,20 @@
 namespace {
   class JITResolverState {
   public:
-    typedef std::map<AssertingVH<Function>, void*> FunctionToStubMapTy;
-    typedef std::map<void*, AssertingVH<Function> > StubToFunctionMapTy;
+    typedef DenseMap<AssertingVH<Function>, void*> FunctionToStubMapTy;
+    typedef std::map<void*, AssertingVH<Function> > CallSiteToFunctionMapTy;
+    typedef DenseMap<AssertingVH<Function>, SmallPtrSet<void*, 1> >
+            FunctionToCallSitesMapTy;
     typedef std::map<AssertingVH<GlobalValue>, void*> GlobalToIndirectSymMapTy;
   private:
     /// FunctionToStubMap - Keep track of the stub created for a particular
     /// function so that we can reuse them if necessary.
     FunctionToStubMapTy FunctionToStubMap;
 
-    /// StubToFunctionMap - Keep track of the function that each stub
-    /// corresponds to.
-    StubToFunctionMapTy StubToFunctionMap;
+    /// CallSiteToFunctionMap - Keep track of the function that each lazy call
+    /// site corresponds to, and vice versa.
+    CallSiteToFunctionMapTy CallSiteToFunctionMap;
+    FunctionToCallSitesMapTy FunctionToCallSitesMap;
 
     /// GlobalToIndirectSymMap - Keep track of the indirect symbol created for a
     /// particular GlobalVariable so that we can reuse them if necessary.
@@ -85,14 +88,78 @@
       return FunctionToStubMap;
     }
 
-    StubToFunctionMapTy& getStubToFunctionMap(const MutexGuard& locked) {
+    GlobalToIndirectSymMapTy& getGlobalToIndirectSymMap(const MutexGuard& locked) {
       assert(locked.holds(TheJIT->lock));
-      return StubToFunctionMap;
+      return GlobalToIndirectSymMap;
     }
 
-    GlobalToIndirectSymMapTy& getGlobalToIndirectSymMap(const MutexGuard& locked) {
+    pair<void *, Function *> LookupFunctionFromCallSite(
+        const MutexGuard &locked, void *CallSite) const {
       assert(locked.holds(TheJIT->lock));
-      return GlobalToIndirectSymMap;
+
+      // The address given to us for the stub may not be exactly right, it might be
+      // a little bit after the stub.  As such, use upper_bound to find it.
+      CallSiteToFunctionMapTy::const_iterator I =
+        CallSiteToFunctionMap.upper_bound(CallSite);
+      assert(I != CallSiteToFunctionMap.begin() &&
+             "This is not a known call site!");
+      --I;
+      return *I;
+    }
+
+    void AddCallSite(const MutexGuard &locked, void *CallSite, Function *F) {
+      assert(locked.holds(TheJIT->lock));
+
+      assert(CallSiteToFunctionMap.insert(std::make_pair(CallSite, F)).second &&
+             "Pair was already in CallSiteToFunctionMap");
+      FunctionToCallSitesMap[F].insert(CallSite);
+    }
+
+    // Returns the Function of the stub if a stub was erased, or NULL if there
+    // was no stub.  This function uses the call-site->function map to find a
+    // relevant function, but asserts that only stubs and not other call sites
+    // will be passed in.
+    Function *EraseStub(const MutexGuard &locked, void *Stub) {
+      CallSiteToFunctionMapTy::iterator C2F_I =
+        CallSiteToFunctionMap.find(Stub);
+      if (C2F_I == CallSiteToFunctionMap.end()) {
+        // Not a stub.
+        return NULL;
+      }
+
+      Function *const F = C2F_I->second;
+#ifndef NDEBUG
+      void *RealStub = FunctionToStubMap.lookup(F);
+      assert(RealStub == Stub &&
+             "Call-site that wasn't a stub pass in to EraseStub");
+#endif
+      FunctionToStubMap.erase(F);
+      CallSiteToFunctionMap.erase(C2F_I);
+
+      // Remove the stub from the function->call-sites map, and remove the whole
+      // entry from the map if that was the last call site.
+      FunctionToCallSitesMapTy::iterator F2C_I = FunctionToCallSitesMap.find(F);
+      assert(F2C_I != FunctionToCallSitesMap.end() &&
+             "FunctionToCallSitesMap broken");
+      assert(F2C_I->second.erase(Stub) &&
+             "FunctionToCallSitesMap broken");
+      if (F2C_I->second.empty())
+        FunctionToCallSitesMap.erase(F2C_I);
+
+      return F;
+    }
+
+    void EraseAllCallSites(const MutexGuard &locked, Function *F) {
+      assert(locked.holds(TheJIT->lock));
+      FunctionToCallSitesMapTy::iterator F2C = FunctionToCallSitesMap.find(F);
+      if (F2C == FunctionToCallSitesMap.end())
+        return;
+      for (SmallPtrSet<void*, 1>::const_iterator I = F2C->second.begin(),
+             E = F2C->second.end(); I != E; ++I) {
+        assert(CallSiteToFunctionMap.erase(*I) == 1 &&
+               "Missing call site->function mapping");
+      }
+      FunctionToCallSitesMap.erase(F2C);
     }
   };
 
@@ -100,7 +167,7 @@
   /// have not yet been compiled.
   class JITResolver {
     typedef JITResolverState::FunctionToStubMapTy FunctionToStubMapTy;
-    typedef JITResolverState::StubToFunctionMapTy StubToFunctionMapTy;
+    typedef JITResolverState::CallSiteToFunctionMapTy CallSiteToFunctionMapTy;
     typedef JITResolverState::GlobalToIndirectSymMapTy GlobalToIndirectSymMapTy;
 
     /// LazyResolverFn - The target lazy resolver function that we actually
@@ -154,7 +221,7 @@
     void *AddCallbackAtLocation(Function *F, void *Location) {
       MutexGuard locked(TheJIT->lock);
       /// Get the target-specific JIT resolver function.
-      state.getStubToFunctionMap(locked)[Location] = F;
+      state.AddCallSite(locked, Location, F);
       return (void*)(intptr_t)LazyResolverFn;
     }
     
@@ -183,8 +250,7 @@
   MutexGuard locked(TheJIT->lock);
 
   // If we already have a stub for this function, recycle it.
-  void *&Stub = state.getFunctionToStubMap(locked)[F];
-  return Stub;
+  return state.getFunctionToStubMap(locked).lookup(F);
 }
 
 /// getFunctionStub - This returns a pointer to a function stub, creating
@@ -230,7 +296,7 @@
 
   // Finally, keep track of the stub-to-Function mapping so that the
   // JITCompilerFn knows which function to compile!
-  state.getStubToFunctionMap(locked)[Stub] = F;
+  state.AddCallSite(locked, Stub, F);
 
   // If we are JIT'ing non-lazily but need to call a function that does not
   // exist yet, add it to the JIT's work list so that we can fill in the stub
@@ -291,10 +357,11 @@
                                     SmallVectorImpl<void*> &Ptrs) {
   MutexGuard locked(TheJIT->lock);
   
-  FunctionToStubMapTy &FM = state.getFunctionToStubMap(locked);
+  const FunctionToStubMapTy &FM = state.getFunctionToStubMap(locked);
   GlobalToIndirectSymMapTy &GM = state.getGlobalToIndirectSymMap(locked);
   
-  for (FunctionToStubMapTy::iterator i = FM.begin(), e = FM.end(); i != e; ++i){
+  for (FunctionToStubMapTy::const_iterator i = FM.begin(), e = FM.end();
+       i != e; ++i){
     Function *F = i->first;
     if (F->isDeclaration() && F->hasExternalLinkage()) {
       GVs.push_back(i->first);
@@ -310,20 +377,15 @@
 
 GlobalValue *JITResolver::invalidateStub(void *Stub) {
   MutexGuard locked(TheJIT->lock);
-  
-  FunctionToStubMapTy &FM = state.getFunctionToStubMap(locked);
-  StubToFunctionMapTy &SM = state.getStubToFunctionMap(locked);
+
   GlobalToIndirectSymMapTy &GM = state.getGlobalToIndirectSymMap(locked);
-  
+
   // Look up the cheap way first, to see if it's a function stub we are
   // invalidating.  If so, remove it from both the forward and reverse maps.
-  if (SM.find(Stub) != SM.end()) {
-    Function *F = SM[Stub];
-    SM.erase(Stub);
-    FM.erase(F);
+  if (Function *F = state.EraseStub(locked, Stub)) {
     return F;
   }
-  
+
   // Otherwise, it might be an indirect symbol stub.  Find it and remove it.
   for (GlobalToIndirectSymMapTy::iterator i = GM.begin(), e = GM.end();
        i != e; ++i) {
@@ -361,14 +423,12 @@
     // JIT lock to be unlocked.
     MutexGuard locked(TheJIT->lock);
 
-    // The address given to us for the stub may not be exactly right, it might be
-    // a little bit after the stub.  As such, use upper_bound to find it.
-    StubToFunctionMapTy::iterator I =
-      JR.state.getStubToFunctionMap(locked).upper_bound(Stub);
-    assert(I != JR.state.getStubToFunctionMap(locked).begin() &&
-           "This is not a known stub!");
-    F = (--I)->second;
-    ActualPtr = I->first;
+    // The address given to us for the stub may not be exactly right, it might
+    // be a little bit after the stub.  As such, use upper_bound to find it.
+    pair<void*, Function*> I =
+      JR.state.LookupFunctionFromCallSite(locked, Stub);
+    F = I.second;
+    ActualPtr = I.first;
   }
 
   // If we have already code generated the function, just return the address.
@@ -383,25 +443,21 @@
                         + F->getName() + "' when lazy compiles are disabled!");
     }
   
-    // We might like to remove the stub from the StubToFunction map.
-    // We can't do that! Multiple threads could be stuck, waiting to acquire the
-    // lock above. As soon as the 1st function finishes compiling the function,
-    // the next one will be released, and needs to be able to find the function
-    // it needs to call.
-    //JR.state.getStubToFunctionMap(locked).erase(I);
-
     DEBUG(errs() << "JIT: Lazily resolving function '" << F->getName()
           << "' In stub ptr = " << Stub << " actual ptr = "
           << ActualPtr << "\n");
 
     Result = TheJIT->getPointerToFunction(F);
   }
-  
-  // Reacquire the lock to erase the stub in the map.
+
+  // Reacquire the lock to update the GOT map.
   MutexGuard locked(TheJIT->lock);
 
-  // We don't need to reuse this stub in the future, as F is now compiled.
-  JR.state.getFunctionToStubMap(locked).erase(F);
+  // We might like to remove the call site from the CallSiteToFunction map, but
+  // we can't do that! Multiple threads could be stuck, waiting to acquire the
+  // lock above. As soon as the 1st function finishes compiling the function,
+  // the next one will be released, and needs to be able to find the function it
+  // needs to call.
 
   // FIXME: We could rewrite all references to this stub if we knew them.
 




From sabre at nondot.org  Mon Oct 19 14:51:45 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 19:51:45 -0000
Subject: [llvm-commits] [llvm] r84526 - in /llvm/trunk/lib/Target/X86:
 AsmPrinter/X86ATTInstPrinter.cpp AsmPrinter/X86ATTInstPrinter.h
 X86Instr64bit.td X86InstrInfo.td
Message-ID: <200910191951.n9JJpkGT011632@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 14:51:42 2009
New Revision: 84526

URL: http://llvm.org/viewvc/llvm-project?rev=84526&view=rev
Log:
remove strings from instructions who are never asmprinted.
All of these "subreg32" modifier instructions are handled
explicitly by the MCInst lowering phase.  If they got to
the asmprinter, they would explode.  They should eventually
be replace with correct use of subregs.

Modified:
    llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.cpp
    llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.h
    llvm/trunk/lib/Target/X86/X86Instr64bit.td
    llvm/trunk/lib/Target/X86/X86InstrInfo.td

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.cpp?rev=84526&r1=84525&r2=84526&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.cpp Mon Oct 19 14:51:42 2009
@@ -57,9 +57,7 @@
   }
 }
 
-void X86ATTInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
-                                     const char *Modifier) {
-  assert(Modifier == 0 && "Modifiers should not be used");
+void X86ATTInstPrinter::printOperand(const MCInst *MI, unsigned OpNo) {
   
   const MCOperand &Op = MI->getOperand(OpNo);
   if (Op.isReg()) {

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.h?rev=84526&r1=84525&r2=84526&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.h (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.h Mon Oct 19 14:51:42 2009
@@ -32,8 +32,7 @@
   static const char *getRegisterName(unsigned RegNo);
 
 
-  void printOperand(const MCInst *MI, unsigned OpNo,
-                    const char *Modifier = 0);
+  void printOperand(const MCInst *MI, unsigned OpNo);
   void printMemReference(const MCInst *MI, unsigned Op);
   void printLeaMemReference(const MCInst *MI, unsigned Op);
   void printSSECC(const MCInst *MI, unsigned Op);

Modified: llvm/trunk/lib/Target/X86/X86Instr64bit.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Instr64bit.td?rev=84526&r1=84525&r2=84526&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86Instr64bit.td (original)
+++ llvm/trunk/lib/Target/X86/X86Instr64bit.td Mon Oct 19 14:51:42 2009
@@ -368,19 +368,15 @@
 // Use movzbl instead of movzbq when the destination is a register; it's
 // equivalent due to implicit zero-extending, and it has a smaller encoding.
 def MOVZX64rr8 : I<0xB6, MRMSrcReg, (outs GR64:$dst), (ins GR8 :$src),
-                   "movz{bl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                   [(set GR64:$dst, (zext GR8:$src))]>, TB;
+                   "", [(set GR64:$dst, (zext GR8:$src))]>, TB;
 def MOVZX64rm8 : I<0xB6, MRMSrcMem, (outs GR64:$dst), (ins i8mem :$src),
-                   "movz{bl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                   [(set GR64:$dst, (zextloadi64i8 addr:$src))]>, TB;
+                   "", [(set GR64:$dst, (zextloadi64i8 addr:$src))]>, TB;
 // Use movzwl instead of movzwq when the destination is a register; it's
 // equivalent due to implicit zero-extending, and it has a smaller encoding.
 def MOVZX64rr16: I<0xB7, MRMSrcReg, (outs GR64:$dst), (ins GR16:$src),
-                   "movz{wl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                   [(set GR64:$dst, (zext GR16:$src))]>, TB;
+                   "", [(set GR64:$dst, (zext GR16:$src))]>, TB;
 def MOVZX64rm16: I<0xB7, MRMSrcMem, (outs GR64:$dst), (ins i16mem:$src),
-                   "movz{wl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                   [(set GR64:$dst, (zextloadi64i16 addr:$src))]>, TB;
+                   "", [(set GR64:$dst, (zextloadi64i16 addr:$src))]>, TB;
 
 // There's no movzlq instruction, but movl can be used for this purpose, using
 // implicit zero-extension. The preferred way to do 32-bit-to-64-bit zero
@@ -390,11 +386,9 @@
 // necessarily all zero. In such cases, we fall back to these explicit zext
 // instructions.
 def MOVZX64rr32 : I<0x89, MRMDestReg, (outs GR64:$dst), (ins GR32:$src),
-                    "mov{l}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                    [(set GR64:$dst, (zext GR32:$src))]>;
+                    "", [(set GR64:$dst, (zext GR32:$src))]>;
 def MOVZX64rm32 : I<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
-                    "mov{l}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                    [(set GR64:$dst, (zextloadi64i32 addr:$src))]>;
+                    "", [(set GR64:$dst, (zextloadi64i32 addr:$src))]>;
 
 // Any instruction that defines a 32-bit result leaves the high half of the
 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
@@ -1455,8 +1449,7 @@
 // Materialize i64 constant where top 32-bits are zero.
 let AddedComplexity = 1, isReMaterializable = 1, isAsCheapAsAMove = 1 in
 def MOV64ri64i32 : Ii32<0xB8, AddRegFrm, (outs GR64:$dst), (ins i64i32imm:$src),
-                        "mov{l}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                        [(set GR64:$dst, i64immZExt32:$src)]>;
+                        "", [(set GR64:$dst, i64immZExt32:$src)]>;
 
 //===----------------------------------------------------------------------===//
 // Thread Local Storage Instructions

Modified: llvm/trunk/lib/Target/X86/X86InstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrInfo.td?rev=84526&r1=84525&r2=84526&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrInfo.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrInfo.td Mon Oct 19 14:51:42 2009
@@ -3392,11 +3392,9 @@
 // of the register here. This has a smaller encoding and avoids a
 // partial-register update.
 def MOVSX16rr8 : I<0xBE, MRMSrcReg, (outs GR16:$dst), (ins GR8 :$src),
-                   "movs{bl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                   [(set GR16:$dst, (sext GR8:$src))]>, TB;
+                   "", [(set GR16:$dst, (sext GR8:$src))]>, TB;
 def MOVSX16rm8 : I<0xBE, MRMSrcMem, (outs GR16:$dst), (ins i8mem :$src),
-                   "movs{bl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                   [(set GR16:$dst, (sextloadi16i8 addr:$src))]>, TB;
+                   "", [(set GR16:$dst, (sextloadi16i8 addr:$src))]>, TB;
 def MOVSX32rr8 : I<0xBE, MRMSrcReg, (outs GR32:$dst), (ins GR8 :$src),
                    "movs{bl|x}\t{$src, $dst|$dst, $src}",
                    [(set GR32:$dst, (sext GR8:$src))]>, TB;
@@ -3414,11 +3412,9 @@
 // of the register here. This has a smaller encoding and avoids a
 // partial-register update.
 def MOVZX16rr8 : I<0xB6, MRMSrcReg, (outs GR16:$dst), (ins GR8 :$src),
-                   "movz{bl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                   [(set GR16:$dst, (zext GR8:$src))]>, TB;
+                   "", [(set GR16:$dst, (zext GR8:$src))]>, TB;
 def MOVZX16rm8 : I<0xB6, MRMSrcMem, (outs GR16:$dst), (ins i8mem :$src),
-                   "movz{bl|x}\t{$src, ${dst:subreg32}|${dst:subreg32}, $src}",
-                   [(set GR16:$dst, (zextloadi16i8 addr:$src))]>, TB;
+                   "", [(set GR16:$dst, (zextloadi16i8 addr:$src))]>, TB;
 def MOVZX32rr8 : I<0xB6, MRMSrcReg, (outs GR32:$dst), (ins GR8 :$src),
                    "movz{bl|x}\t{$src, $dst|$dst, $src}",
                    [(set GR32:$dst, (zext GR8:$src))]>, TB;
@@ -3474,9 +3470,8 @@
                  [(set GR8:$dst, 0)]>;
 // Use xorl instead of xorw since we don't care about the high 16 bits,
 // it's smaller, and it avoids a partial-register update.
-def MOV16r0  : I<0x31, MRMInitReg,  (outs GR16:$dst), (ins),
-                 "xor{l}\t${dst:subreg32}, ${dst:subreg32}",
-                 [(set GR16:$dst, 0)]>;
+def MOV16r0  : I<0x31, MRMInitReg, (outs GR16:$dst), (ins),
+                 "", [(set GR16:$dst, 0)]>;
 def MOV32r0  : I<0x31, MRMInitReg,  (outs GR32:$dst), (ins),
                  "xor{l}\t$dst, $dst",
                  [(set GR32:$dst, 0)]>;




From sabre at nondot.org  Mon Oct 19 14:56:37 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 19:56:37 -0000
Subject: [llvm-commits] [llvm] r84527 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMInstPrinter.cpp ARMInstPrinter.h
 CMakeLists.txt
Message-ID: <200910191956.n9JJuel6011780@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 14:56:26 2009
New Revision: 84527

URL: http://llvm.org/viewvc/llvm-project?rev=84527&view=rev
Log:
stub out a minimal ARMInstPrinter.

Added:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt

Added: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84527&view=auto

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (added)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 14:56:26 2009
@@ -0,0 +1,33 @@
+//===-- ARMInstPrinter.cpp - Convert ARM MCInst to assembly syntax --------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class prints an ARM MCInst to a .s file.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "asm-printer"
+#include "ARMInstPrinter.h"
+#include "llvm/MC/MCInst.h"
+//#include "llvm/MC/MCAsmInfo.h"
+//#include "llvm/MC/MCExpr.h"
+//#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/FormattedStream.h"
+#include "ARMGenInstrNames.inc"
+using namespace llvm;
+
+// Include the auto-generated portion of the assembly writer.
+#define MachineInstr MCInst
+#define ARMAsmPrinter ARMInstPrinter  // FIXME: REMOVE.
+#define NO_ASM_WRITER_BOILERPLATE
+#include "ARMGenAsmWriter.inc"
+#undef MachineInstr
+#undef ARMAsmPrinter
+
+void ARMInstPrinter::printInst(const MCInst *MI) { printInstruction(MI); }
+

Added: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84527&view=auto

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (added)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 14:56:26 2009
@@ -0,0 +1,86 @@
+//===-- ARMInstPrinter.h - Convert ARM MCInst to assembly syntax ----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class prints an ARM MCInst to a .s file.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef ARMINSTPRINTER_H
+#define ARMINSTPRINTER_H
+
+#include "llvm/MC/MCInstPrinter.h"
+
+namespace llvm {
+  class MCOperand;
+  
+class ARMInstPrinter : public MCInstPrinter {
+public:
+  ARMInstPrinter(raw_ostream &O, const MCAsmInfo &MAI) : MCInstPrinter(O, MAI){}
+
+  virtual void printInst(const MCInst *MI);
+  
+  // Autogenerated by tblgen.
+  void printInstruction(const MCInst *MI);
+  static const char *getRegisterName(unsigned RegNo);
+
+
+  void printOperand(const MCInst *MI, unsigned OpNo,
+                    const char *Modifier = 0) {}
+    
+  void printSOImmOperand(const MCInst *MI, unsigned OpNum) {}
+  void printSOImm2PartOperand(const MCInst *MI, unsigned OpNum) {}
+  void printSORegOperand(const MCInst *MI, unsigned OpNum) {}
+  void printAddrMode2Operand(const MCInst *MI, unsigned OpNum) {}
+  void printAddrMode2OffsetOperand(const MCInst *MI, unsigned OpNum) {}
+  void printAddrMode3Operand(const MCInst *MI, unsigned OpNum) {}
+  void printAddrMode3OffsetOperand(const MCInst *MI, unsigned OpNum) {}
+  void printAddrMode4Operand(const MCInst *MI, unsigned OpNum,
+                             const char *Modifier = 0) {}
+  void printAddrMode5Operand(const MCInst *MI, unsigned OpNum,
+                             const char *Modifier = 0) {}
+  void printAddrMode6Operand(const MCInst *MI, unsigned OpNum) {}
+  void printAddrModePCOperand(const MCInst *MI, unsigned OpNum,
+                              const char *Modifier = 0) {}
+  void printBitfieldInvMaskImmOperand (const MCInst *MI, unsigned OpNum) {}
+  
+  void printThumbITMask(const MCInst *MI, unsigned OpNum) {}
+  void printThumbAddrModeRROperand(const MCInst *MI, unsigned OpNum) {}
+  void printThumbAddrModeRI5Operand(const MCInst *MI, unsigned OpNum,
+                                    unsigned Scale) {}
+  void printThumbAddrModeS1Operand(const MCInst *MI, unsigned OpNum) {}
+  void printThumbAddrModeS2Operand(const MCInst *MI, unsigned OpNum) {}
+  void printThumbAddrModeS4Operand(const MCInst *MI, unsigned OpNum) {}
+  void printThumbAddrModeSPOperand(const MCInst *MI, unsigned OpNum) {}
+  
+  void printT2SOOperand(const MCInst *MI, unsigned OpNum) {}
+  void printT2AddrModeImm12Operand(const MCInst *MI, unsigned OpNum) {}
+  void printT2AddrModeImm8Operand(const MCInst *MI, unsigned OpNum) {}
+  void printT2AddrModeImm8s4Operand(const MCInst *MI, unsigned OpNum) {}
+  void printT2AddrModeImm8OffsetOperand(const MCInst *MI, unsigned OpNum) {}
+  void printT2AddrModeSoRegOperand(const MCInst *MI, unsigned OpNum) {}
+  
+  void printPredicateOperand(const MCInst *MI, unsigned OpNum) {}
+  void printSBitModifierOperand(const MCInst *MI, unsigned OpNum) {}
+  void printPCLabel(const MCInst *MI, unsigned OpNum) {}
+  void printRegisterList(const MCInst *MI, unsigned OpNum) {}
+  void printCPInstOperand(const MCInst *MI, unsigned OpNum,
+                          const char *Modifier) {}
+  void printJTBlockOperand(const MCInst *MI, unsigned OpNum) {}
+  void printJT2BlockOperand(const MCInst *MI, unsigned OpNum) {}
+  void printTBAddrMode(const MCInst *MI, unsigned OpNum) {}
+  void printNoHashImmediate(const MCInst *MI, unsigned OpNum) {}
+  
+  
+  // FIXME:
+  void PrintSpecial(const MCInst *MI, const char *Kind) {}
+};
+  
+}
+
+#endif

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt?rev=84527&r1=84526&r2=84527&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt Mon Oct 19 14:56:26 2009
@@ -1,6 +1,7 @@
 include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
 
 add_llvm_library(LLVMARMAsmPrinter
+  ARMInstPrinter.cpp
   ARMAsmPrinter.cpp
   )
-add_dependencies(LLVMARMAsmPrinter ARMCodeGenTable_gen)
\ No newline at end of file
+add_dependencies(LLVMARMAsmPrinter ARMCodeGenTable_gen)




From sabre at nondot.org  Mon Oct 19 14:59:05 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 19:59:05 -0000
Subject: [llvm-commits] [llvm] r84528 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910191959.n9JJx5jj011853@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 14:59:05 2009
New Revision: 84528

URL: http://llvm.org/viewvc/llvm-project?rev=84528&view=rev
Log:
wire up ARM's printMCInst method.  Now llvm-mc should be able to produce 
"something" when printing MCInsts, it will just be missing all the 
operand info.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84528&r1=84527&r2=84528&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 14:59:05 2009
@@ -17,6 +17,7 @@
 #include "ARMAddressingModes.h"
 #include "ARMConstantPoolValue.h"
 #include "ARMMachineFunctionInfo.h"
+#include "ARMInstPrinter.h"
 #include "llvm/Constants.h"
 #include "llvm/Module.h"
 #include "llvm/Assembly/Writer.h"
@@ -73,6 +74,10 @@
     virtual const char *getPassName() const {
       return "ARM Assembly Printer";
     }
+    
+    void printMCInst(const MCInst *MI) {
+      ARMInstPrinter(O, *MAI).printInstruction(MI);
+    }      
 
     void printOperand(const MachineInstr *MI, int OpNum,
                       const char *Modifier = 0);




From resistor at mac.com  Mon Oct 19 15:11:53 2009
From: resistor at mac.com (Owen Anderson)
Date: Mon, 19 Oct 2009 20:11:53 -0000
Subject: [llvm-commits] [llvm] r84529 - in /llvm/trunk/lib/VMCore:
 Constants.cpp LLVMContextImpl.h Metadata.cpp
Message-ID: <200910192011.n9JKBrmn012258@zion.cs.uiuc.edu>

Author: resistor
Date: Mon Oct 19 15:11:52 2009
New Revision: 84529

URL: http://llvm.org/viewvc/llvm-project?rev=84529&view=rev
Log:
Banish ConstantsLock.  It's serving no purpose other than slowing things down
at the moment.

Modified:
    llvm/trunk/lib/VMCore/Constants.cpp
    llvm/trunk/lib/VMCore/LLVMContextImpl.h
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/lib/VMCore/Constants.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Constants.cpp?rev=84529&r1=84528&r2=84529&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Constants.cpp (original)
+++ llvm/trunk/lib/VMCore/Constants.cpp Mon Oct 19 15:11:52 2009
@@ -232,7 +232,6 @@
 
 ConstantInt* ConstantInt::getTrue(LLVMContext &Context) {
   LLVMContextImpl *pImpl = Context.pImpl;
-  sys::SmartScopedWriter<true>(pImpl->ConstantsLock);
   if (pImpl->TheTrueVal)
     return pImpl->TheTrueVal;
   else
@@ -242,7 +241,6 @@
 
 ConstantInt* ConstantInt::getFalse(LLVMContext &Context) {
   LLVMContextImpl *pImpl = Context.pImpl;
-  sys::SmartScopedWriter<true>(pImpl->ConstantsLock);
   if (pImpl->TheFalseVal)
     return pImpl->TheFalseVal;
   else
@@ -261,22 +259,9 @@
   const IntegerType *ITy = IntegerType::get(Context, V.getBitWidth());
   // get an existing value or the insertion position
   DenseMapAPIntKeyInfo::KeyTy Key(V, ITy);
-  
-  Context.pImpl->ConstantsLock.reader_acquire();
   ConstantInt *&Slot = Context.pImpl->IntConstants[Key]; 
-  Context.pImpl->ConstantsLock.reader_release();
-    
-  if (!Slot) {
-    sys::SmartScopedWriter<true> Writer(Context.pImpl->ConstantsLock);
-    ConstantInt *&NewSlot = Context.pImpl->IntConstants[Key]; 
-    if (!Slot) {
-      NewSlot = new ConstantInt(ITy, V);
-    }
-    
-    return NewSlot;
-  } else {
-    return Slot;
-  }
+  if (!Slot) Slot = new ConstantInt(ITy, V);
+  return Slot;
 }
 
 Constant* ConstantInt::get(const Type* Ty, uint64_t V, bool isSigned) {
@@ -405,32 +390,24 @@
   
   LLVMContextImpl* pImpl = Context.pImpl;
   
-  pImpl->ConstantsLock.reader_acquire();
   ConstantFP *&Slot = pImpl->FPConstants[Key];
-  pImpl->ConstantsLock.reader_release();
     
   if (!Slot) {
-    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
-    ConstantFP *&NewSlot = pImpl->FPConstants[Key];
-    if (!NewSlot) {
-      const Type *Ty;
-      if (&V.getSemantics() == &APFloat::IEEEsingle)
-        Ty = Type::getFloatTy(Context);
-      else if (&V.getSemantics() == &APFloat::IEEEdouble)
-        Ty = Type::getDoubleTy(Context);
-      else if (&V.getSemantics() == &APFloat::x87DoubleExtended)
-        Ty = Type::getX86_FP80Ty(Context);
-      else if (&V.getSemantics() == &APFloat::IEEEquad)
-        Ty = Type::getFP128Ty(Context);
-      else {
-        assert(&V.getSemantics() == &APFloat::PPCDoubleDouble && 
-               "Unknown FP format");
-        Ty = Type::getPPC_FP128Ty(Context);
-      }
-      NewSlot = new ConstantFP(Ty, V);
+    const Type *Ty;
+    if (&V.getSemantics() == &APFloat::IEEEsingle)
+      Ty = Type::getFloatTy(Context);
+    else if (&V.getSemantics() == &APFloat::IEEEdouble)
+      Ty = Type::getDoubleTy(Context);
+    else if (&V.getSemantics() == &APFloat::x87DoubleExtended)
+      Ty = Type::getX86_FP80Ty(Context);
+    else if (&V.getSemantics() == &APFloat::IEEEquad)
+      Ty = Type::getFP128Ty(Context);
+    else {
+      assert(&V.getSemantics() == &APFloat::PPCDoubleDouble && 
+             "Unknown FP format");
+      Ty = Type::getPPC_FP128Ty(Context);
     }
-    
-    return NewSlot;
+    Slot = new ConstantFP(Ty, V);
   }
   
   return Slot;
@@ -1908,7 +1885,6 @@
     Replacement = ConstantAggregateZero::get(getType());
   } else {
     // Check to see if we have this array type already.
-    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
     bool Exists;
     LLVMContextImpl::ArrayConstantsTy::MapTy::iterator I =
       pImpl->ArrayConstants.InsertOrGetItem(Lookup, Exists);
@@ -1987,7 +1963,6 @@
     Replacement = ConstantAggregateZero::get(getType());
   } else {
     // Check to see if we have this array type already.
-    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
     bool Exists;
     LLVMContextImpl::StructConstantsTy::MapTy::iterator I =
       pImpl->StructConstants.InsertOrGetItem(Lookup, Exists);

Modified: llvm/trunk/lib/VMCore/LLVMContextImpl.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/LLVMContextImpl.h?rev=84529&r1=84528&r2=84529&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/LLVMContextImpl.h (original)
+++ llvm/trunk/lib/VMCore/LLVMContextImpl.h Mon Oct 19 15:11:52 2009
@@ -96,7 +96,6 @@
 
 class LLVMContextImpl {
 public:
-  sys::SmartRWMutex<true> ConstantsLock;
   typedef DenseMap<DenseMapAPIntKeyInfo::KeyTy, ConstantInt*, 
                          DenseMapAPIntKeyInfo> IntMapTy;
   IntMapTy IntConstants;

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84529&r1=84528&r2=84529&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Mon Oct 19 15:11:52 2009
@@ -52,7 +52,6 @@
 //
 MDString *MDString::get(LLVMContext &Context, const StringRef &Str) {
   LLVMContextImpl *pImpl = Context.pImpl;
-  sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
   StringMapEntry<MDString *> &Entry = 
     pImpl->MDStringCache.GetOrCreateValue(Str);
   MDString *&S = Entry.getValue();
@@ -93,12 +92,10 @@
   void *InsertPoint;
   MDNode *N;
   {
-    sys::SmartScopedReader<true> Reader(pImpl->ConstantsLock);
     N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
   }  
   if (N) return N;
   
-  sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
   N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
   if (!N) {
     // InsertPoint will have been set by the FindNodeOrInsertPos call.
@@ -118,7 +115,6 @@
 MDNode::~MDNode() {
   {
     LLVMContextImpl *pImpl = getType()->getContext().pImpl;
-    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
     pImpl->MDNodeSet.RemoveNode(this);
   }
   dropAllReferences();
@@ -147,10 +143,7 @@
     return;
 
   // Remove "this" from the context map. 
-  {
-    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
-    pImpl->MDNodeSet.RemoveNode(this);
-  }
+  pImpl->MDNodeSet.RemoveNode(this);
 
   // MDNode only lists metadata elements in operand list, because MDNode
   // used by MDNode is considered a valid use. However on the side, MDNode
@@ -186,10 +179,8 @@
   // node with updated "this" node.
   FoldingSetNodeID ID;
   Profile(ID);
-  pImpl->ConstantsLock.reader_acquire();
   void *InsertPoint;
   MDNode *N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
-  pImpl->ConstantsLock.reader_release();
 
   if (N) {
     N->replaceAllUsesWith(this);
@@ -197,14 +188,11 @@
     N = 0;
   }
 
-  {
-    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
-    N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
-    if (!N) {
-      // InsertPoint will have been set by the FindNodeOrInsertPos call.
-      N = this;
-      pImpl->MDNodeSet.InsertNode(N, InsertPoint);
-    }
+  N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
+  if (!N) {
+    // InsertPoint will have been set by the FindNodeOrInsertPos call.
+    N = this;
+    pImpl->MDNodeSet.InsertNode(N, InsertPoint);
   }
 }
 




From sabre at nondot.org  Mon Oct 19 15:20:47 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 20:20:47 -0000
Subject: [llvm-commits] [llvm] r84530 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMAsmPrinter.cpp CMakeLists.txt
Message-ID: <200910192020.n9JKKlep012646@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 15:20:46 2009
New Revision: 84530

URL: http://llvm.org/viewvc/llvm-project?rev=84530&view=rev
Log:
wire up skeletal support for having llc print instructions
through mcinst lowering -> mcinstprinter, when llc is passed
the -enable-arm-mcinst-printer flag.  Currently this
is very "aborty".

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84530&r1=84529&r2=84530&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 15:20:46 2009
@@ -1,3 +1,5 @@
+//===-- ARMAsmPrinter.cpp - Print machine code to an ARM .s file ----------===//
+//
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
@@ -13,11 +15,12 @@
 #define DEBUG_TYPE "asm-printer"
 #include "ARM.h"
 #include "ARMBuildAttrs.h"
-#include "ARMTargetMachine.h"
 #include "ARMAddressingModes.h"
 #include "ARMConstantPoolValue.h"
-#include "ARMMachineFunctionInfo.h"
 #include "ARMInstPrinter.h"
+#include "ARMMachineFunctionInfo.h"
+#include "ARMMCInstLower.h"
+#include "ARMTargetMachine.h"
 #include "llvm/Constants.h"
 #include "llvm/Module.h"
 #include "llvm/Assembly/Writer.h"
@@ -28,6 +31,7 @@
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCSectionMachO.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
@@ -40,15 +44,20 @@
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringSet.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/FormattedStream.h"
 #include "llvm/Support/Mangler.h"
 #include "llvm/Support/MathExtras.h"
-#include "llvm/Support/FormattedStream.h"
 #include <cctype>
 using namespace llvm;
 
 STATISTIC(EmittedInsts, "Number of machine instrs printed");
 
+static cl::opt<bool>
+EnableMCInst("enable-arm-mcinst-printer", cl::Hidden,
+            cl::desc("enable experimental asmprinter gunk in the arm backend"));
+
 namespace {
   class ARMAsmPrinter : public AsmPrinter {
 
@@ -77,7 +86,10 @@
     
     void printMCInst(const MCInst *MI) {
       ARMInstPrinter(O, *MAI).printInstruction(MI);
-    }      
+    }
+    
+    void printInstructionThroughMCStreamer(const MachineInstr *MI);
+    
 
     void printOperand(const MachineInstr *MI, int OpNum,
                       const char *Modifier = 0);
@@ -276,14 +288,13 @@
   for (MachineFunction::const_iterator I = MF.begin(), E = MF.end();
        I != E; ++I) {
     // Print a label for the basic block.
-    if (I != MF.begin()) {
+    if (I != MF.begin())
       EmitBasicBlockStart(I);
-    }
+
+    // Print the assembly for the instruction.
     for (MachineBasicBlock::const_iterator II = I->begin(), E = I->end();
-         II != E; ++II) {
-      // Print the assembly for the instruction.
+         II != E; ++II)
       printMachineInstruction(II);
-    }
   }
 
   if (MAI->hasDotTypeDotSizeDirective())
@@ -1026,7 +1037,13 @@
 
   // Call the autogenerated instruction printer routines.
   processDebugLoc(MI, true);
-  printInstruction(MI);
+  
+  if (EnableMCInst) {
+    printInstructionThroughMCStreamer(MI);
+  } else {
+    printInstruction(MI);
+  }
+  
   if (VerboseAsm && !MI->getDebugLoc().isUnknown())
     EmitComments(*MI);
   O << '\n';
@@ -1291,3 +1308,32 @@
   RegisterAsmPrinter<ARMAsmPrinter> X(TheARMTarget);
   RegisterAsmPrinter<ARMAsmPrinter> Y(TheThumbTarget);
 }
+
+
+//===----------------------------------------------------------------------===//
+
+void ARMAsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI) {
+  ARMMCInstLower MCInstLowering(OutContext, Mang);
+  switch (MI->getOpcode()) {
+  case TargetInstrInfo::DBG_LABEL:
+  case TargetInstrInfo::EH_LABEL:
+  case TargetInstrInfo::GC_LABEL:
+    printLabel(MI);
+    return;
+  case TargetInstrInfo::KILL:
+    return;
+  case TargetInstrInfo::INLINEASM:
+    O << '\t';
+    printInlineAsm(MI);
+    return;
+  case TargetInstrInfo::IMPLICIT_DEF:
+    printImplicitDef(MI);
+    return;
+  default: break;
+  }
+      
+  MCInst TmpInst;
+  MCInstLowering.Lower(MI, TmpInst);
+  
+  printMCInst(&TmpInst);
+}

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt?rev=84530&r1=84529&r2=84530&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/CMakeLists.txt Mon Oct 19 15:20:46 2009
@@ -1,7 +1,8 @@
 include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
 
 add_llvm_library(LLVMARMAsmPrinter
-  ARMInstPrinter.cpp
   ARMAsmPrinter.cpp
+  ARMInstPrinter.cpp
+  ARMMCInstLower.cpp
   )
 add_dependencies(LLVMARMAsmPrinter ARMCodeGenTable_gen)




From sabre at nondot.org  Mon Oct 19 15:21:06 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 20:21:06 -0000
Subject: [llvm-commits] [llvm] r84531 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMMCInstLower.cpp ARMMCInstLower.h
Message-ID: <200910192021.n9JKL6ur012667@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 15:21:05 2009
New Revision: 84531

URL: http://llvm.org/viewvc/llvm-project?rev=84531&view=rev
Log:
add the files that go with the previous rev

Added:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h

Added: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp?rev=84531&view=auto

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp (added)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp Mon Oct 19 15:21:05 2009
@@ -0,0 +1,81 @@
+//===-- ARMMCInstLower.cpp - Convert ARM MachineInstr to an MCInst --------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains code to lower ARM MachineInstrs to their corresponding
+// MCInst records.
+//
+//===----------------------------------------------------------------------===//
+
+#include "ARMMCInstLower.h"
+//#include "ARMMCAsmInfo.h"
+//#include "llvm/CodeGen/MachineModuleInfoImpls.h"
+#include "llvm/CodeGen/MachineInstr.h"
+//#include "llvm/MC/MCContext.h"
+//#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
+//#include "llvm/MC/MCStreamer.h"
+//#include "llvm/Support/FormattedStream.h"
+//#include "llvm/Support/Mangler.h"
+//#include "llvm/ADT/SmallString.h"
+using namespace llvm;
+
+
+#if 0
+const ARMSubtarget &ARMMCInstLower::getSubtarget() const {
+  return AsmPrinter.getSubtarget();
+}
+
+MachineModuleInfoMachO &ARMMCInstLower::getMachOMMI() const {
+  assert(getSubtarget().isTargetDarwin() &&"Can only get MachO info on darwin");
+  return AsmPrinter.MMI->getObjFileInfo<MachineModuleInfoMachO>(); 
+}
+#endif
+
+
+void ARMMCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
+  OutMI.setOpcode(MI->getOpcode());
+  
+  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+    const MachineOperand &MO = MI->getOperand(i);
+    
+    MCOperand MCOp;
+    switch (MO.getType()) {
+    default:
+      MI->dump();
+      assert(0 && "unknown operand type");
+    case MachineOperand::MO_Register:
+      MCOp = MCOperand::CreateReg(MO.getReg());
+      break;
+    case MachineOperand::MO_Immediate:
+      MCOp = MCOperand::CreateImm(MO.getImm());
+      break;
+#if 0
+    case MachineOperand::MO_MachineBasicBlock:
+      MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
+                       AsmPrinter.GetMBBSymbol(MO.getMBB()->getNumber()), Ctx));
+      break;
+    case MachineOperand::MO_GlobalAddress:
+      MCOp = LowerSymbolOperand(MO, GetGlobalAddressSymbol(MO));
+      break;
+    case MachineOperand::MO_ExternalSymbol:
+      MCOp = LowerSymbolOperand(MO, GetExternalSymbolSymbol(MO));
+      break;
+    case MachineOperand::MO_JumpTableIndex:
+      MCOp = LowerSymbolOperand(MO, GetJumpTableSymbol(MO));
+      break;
+    case MachineOperand::MO_ConstantPoolIndex:
+      MCOp = LowerSymbolOperand(MO, GetConstantPoolIndexSymbol(MO));
+      break;
+#endif
+    }
+    
+    OutMI.addOperand(MCOp);
+  }
+  
+}

Added: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h?rev=84531&view=auto

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h (added)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h Mon Oct 19 15:21:05 2009
@@ -0,0 +1,54 @@
+//===-- ARMMCInstLower.h - Lower MachineInstr to MCInst -------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef ARM_MCINSTLOWER_H
+#define ARM_MCINSTLOWER_H
+
+#include "llvm/Support/Compiler.h"
+
+namespace llvm {
+  class MCContext;
+  class MCInst;
+  class MCOperand;
+  class MCSymbol;
+  class MachineInstr;
+  class MachineModuleInfoMachO;
+  class MachineOperand;
+  class Mangler;
+  //class ARMSubtarget;
+  
+/// ARMMCInstLower - This class is used to lower an MachineInstr into an MCInst.
+class VISIBILITY_HIDDEN ARMMCInstLower {
+  MCContext &Ctx;
+  Mangler *Mang;
+
+  //const ARMSubtarget &getSubtarget() const;
+public:
+  ARMMCInstLower(MCContext &ctx, Mangler *mang)
+    : Ctx(ctx), Mang(mang) {}
+  
+  void Lower(const MachineInstr *MI, MCInst &OutMI) const;
+
+/*
+ MCSymbol *GetPICBaseSymbol() const;
+  
+  MCSymbol *GetGlobalAddressSymbol(const MachineOperand &MO) const;
+  MCSymbol *GetExternalSymbolSymbol(const MachineOperand &MO) const;
+  MCSymbol *GetJumpTableSymbol(const MachineOperand &MO) const;
+  MCSymbol *GetConstantPoolIndexSymbol(const MachineOperand &MO) const;
+  MCOperand LowerSymbolOperand(const MachineOperand &MO, MCSymbol *Sym) const;
+  
+private:
+  MachineModuleInfoMachO &getMachOMMI() const;
+ */
+};
+
+}
+
+#endif




From clattner at apple.com  Mon Oct 19 15:25:39 2009
From: clattner at apple.com (Chris Lattner)
Date: Mon, 19 Oct 2009 13:25:39 -0700
Subject: [llvm-commits] [llvm] r84529 - in /llvm/trunk/lib/VMCore:
	Constants.cpp LLVMContextImpl.h Metadata.cpp
In-Reply-To: <200910192011.n9JKBrmn012258@zion.cs.uiuc.edu>
References: <200910192011.n9JKBrmn012258@zion.cs.uiuc.edu>
Message-ID: <43E3E5D7-C7FF-499C-9C02-90ED1C2C3A7E@apple.com>


On Oct 19, 2009, at 1:11 PM, Owen Anderson wrote:

> Author: resistor
> Date: Mon Oct 19 15:11:52 2009
> New Revision: 84529
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84529&view=rev
> Log:
> Banish ConstantsLock.  It's serving no purpose other than slowing  
> things down
> at the moment.

Thanks Owen!

-Chris

>
> Modified:
>    llvm/trunk/lib/VMCore/Constants.cpp
>    llvm/trunk/lib/VMCore/LLVMContextImpl.h
>    llvm/trunk/lib/VMCore/Metadata.cpp
>
> Modified: llvm/trunk/lib/VMCore/Constants.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Constants.cpp?rev=84529&r1=84528&r2=84529&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/lib/VMCore/Constants.cpp (original)
> +++ llvm/trunk/lib/VMCore/Constants.cpp Mon Oct 19 15:11:52 2009
> @@ -232,7 +232,6 @@
>
> ConstantInt* ConstantInt::getTrue(LLVMContext &Context) {
>   LLVMContextImpl *pImpl = Context.pImpl;
> -  sys::SmartScopedWriter<true>(pImpl->ConstantsLock);
>   if (pImpl->TheTrueVal)
>     return pImpl->TheTrueVal;
>   else
> @@ -242,7 +241,6 @@
>
> ConstantInt* ConstantInt::getFalse(LLVMContext &Context) {
>   LLVMContextImpl *pImpl = Context.pImpl;
> -  sys::SmartScopedWriter<true>(pImpl->ConstantsLock);
>   if (pImpl->TheFalseVal)
>     return pImpl->TheFalseVal;
>   else
> @@ -261,22 +259,9 @@
>   const IntegerType *ITy = IntegerType::get(Context, V.getBitWidth());
>   // get an existing value or the insertion position
>   DenseMapAPIntKeyInfo::KeyTy Key(V, ITy);
> -
> -  Context.pImpl->ConstantsLock.reader_acquire();
>   ConstantInt *&Slot = Context.pImpl->IntConstants[Key];
> -  Context.pImpl->ConstantsLock.reader_release();
> -
> -  if (!Slot) {
> -    sys::SmartScopedWriter<true> Writer(Context.pImpl- 
> >ConstantsLock);
> -    ConstantInt *&NewSlot = Context.pImpl->IntConstants[Key];
> -    if (!Slot) {
> -      NewSlot = new ConstantInt(ITy, V);
> -    }
> -
> -    return NewSlot;
> -  } else {
> -    return Slot;
> -  }
> +  if (!Slot) Slot = new ConstantInt(ITy, V);
> +  return Slot;
> }
>
> Constant* ConstantInt::get(const Type* Ty, uint64_t V, bool  
> isSigned) {
> @@ -405,32 +390,24 @@
>
>   LLVMContextImpl* pImpl = Context.pImpl;
>
> -  pImpl->ConstantsLock.reader_acquire();
>   ConstantFP *&Slot = pImpl->FPConstants[Key];
> -  pImpl->ConstantsLock.reader_release();
>
>   if (!Slot) {
> -    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
> -    ConstantFP *&NewSlot = pImpl->FPConstants[Key];
> -    if (!NewSlot) {
> -      const Type *Ty;
> -      if (&V.getSemantics() == &APFloat::IEEEsingle)
> -        Ty = Type::getFloatTy(Context);
> -      else if (&V.getSemantics() == &APFloat::IEEEdouble)
> -        Ty = Type::getDoubleTy(Context);
> -      else if (&V.getSemantics() == &APFloat::x87DoubleExtended)
> -        Ty = Type::getX86_FP80Ty(Context);
> -      else if (&V.getSemantics() == &APFloat::IEEEquad)
> -        Ty = Type::getFP128Ty(Context);
> -      else {
> -        assert(&V.getSemantics() == &APFloat::PPCDoubleDouble &&
> -               "Unknown FP format");
> -        Ty = Type::getPPC_FP128Ty(Context);
> -      }
> -      NewSlot = new ConstantFP(Ty, V);
> +    const Type *Ty;
> +    if (&V.getSemantics() == &APFloat::IEEEsingle)
> +      Ty = Type::getFloatTy(Context);
> +    else if (&V.getSemantics() == &APFloat::IEEEdouble)
> +      Ty = Type::getDoubleTy(Context);
> +    else if (&V.getSemantics() == &APFloat::x87DoubleExtended)
> +      Ty = Type::getX86_FP80Ty(Context);
> +    else if (&V.getSemantics() == &APFloat::IEEEquad)
> +      Ty = Type::getFP128Ty(Context);
> +    else {
> +      assert(&V.getSemantics() == &APFloat::PPCDoubleDouble &&
> +             "Unknown FP format");
> +      Ty = Type::getPPC_FP128Ty(Context);
>     }
> -
> -    return NewSlot;
> +    Slot = new ConstantFP(Ty, V);
>   }
>
>   return Slot;
> @@ -1908,7 +1885,6 @@
>     Replacement = ConstantAggregateZero::get(getType());
>   } else {
>     // Check to see if we have this array type already.
> -    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
>     bool Exists;
>     LLVMContextImpl::ArrayConstantsTy::MapTy::iterator I =
>       pImpl->ArrayConstants.InsertOrGetItem(Lookup, Exists);
> @@ -1987,7 +1963,6 @@
>     Replacement = ConstantAggregateZero::get(getType());
>   } else {
>     // Check to see if we have this array type already.
> -    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
>     bool Exists;
>     LLVMContextImpl::StructConstantsTy::MapTy::iterator I =
>       pImpl->StructConstants.InsertOrGetItem(Lookup, Exists);
>
> Modified: llvm/trunk/lib/VMCore/LLVMContextImpl.h
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/LLVMContextImpl.h?rev=84529&r1=84528&r2=84529&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/lib/VMCore/LLVMContextImpl.h (original)
> +++ llvm/trunk/lib/VMCore/LLVMContextImpl.h Mon Oct 19 15:11:52 2009
> @@ -96,7 +96,6 @@
>
> class LLVMContextImpl {
> public:
> -  sys::SmartRWMutex<true> ConstantsLock;
>   typedef DenseMap<DenseMapAPIntKeyInfo::KeyTy, ConstantInt*,
>                          DenseMapAPIntKeyInfo> IntMapTy;
>   IntMapTy IntConstants;
>
> Modified: llvm/trunk/lib/VMCore/Metadata.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84529&r1=84528&r2=84529&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/lib/VMCore/Metadata.cpp (original)
> +++ llvm/trunk/lib/VMCore/Metadata.cpp Mon Oct 19 15:11:52 2009
> @@ -52,7 +52,6 @@
> //
> MDString *MDString::get(LLVMContext &Context, const StringRef &Str) {
>   LLVMContextImpl *pImpl = Context.pImpl;
> -  sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
>   StringMapEntry<MDString *> &Entry =
>     pImpl->MDStringCache.GetOrCreateValue(Str);
>   MDString *&S = Entry.getValue();
> @@ -93,12 +92,10 @@
>   void *InsertPoint;
>   MDNode *N;
>   {
> -    sys::SmartScopedReader<true> Reader(pImpl->ConstantsLock);
>     N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
>   }
>   if (N) return N;
>
> -  sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
>   N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
>   if (!N) {
>     // InsertPoint will have been set by the FindNodeOrInsertPos call.
> @@ -118,7 +115,6 @@
> MDNode::~MDNode() {
>   {
>     LLVMContextImpl *pImpl = getType()->getContext().pImpl;
> -    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
>     pImpl->MDNodeSet.RemoveNode(this);
>   }
>   dropAllReferences();
> @@ -147,10 +143,7 @@
>     return;
>
>   // Remove "this" from the context map.
> -  {
> -    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
> -    pImpl->MDNodeSet.RemoveNode(this);
> -  }
> +  pImpl->MDNodeSet.RemoveNode(this);
>
>   // MDNode only lists metadata elements in operand list, because  
> MDNode
>   // used by MDNode is considered a valid use. However on the side,  
> MDNode
> @@ -186,10 +179,8 @@
>   // node with updated "this" node.
>   FoldingSetNodeID ID;
>   Profile(ID);
> -  pImpl->ConstantsLock.reader_acquire();
>   void *InsertPoint;
>   MDNode *N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
> -  pImpl->ConstantsLock.reader_release();
>
>   if (N) {
>     N->replaceAllUsesWith(this);
> @@ -197,14 +188,11 @@
>     N = 0;
>   }
>
> -  {
> -    sys::SmartScopedWriter<true> Writer(pImpl->ConstantsLock);
> -    N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
> -    if (!N) {
> -      // InsertPoint will have been set by the FindNodeOrInsertPos  
> call.
> -      N = this;
> -      pImpl->MDNodeSet.InsertNode(N, InsertPoint);
> -    }
> +  N = pImpl->MDNodeSet.FindNodeOrInsertPos(ID, InsertPoint);
> +  if (!N) {
> +    // InsertPoint will have been set by the FindNodeOrInsertPos  
> call.
> +    N = this;
> +    pImpl->MDNodeSet.InsertNode(N, InsertPoint);
>   }
> }
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From sabre at nondot.org  Mon Oct 19 15:59:56 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 20:59:56 -0000
Subject: [llvm-commits] [llvm] r84532 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMAsmPrinter.cpp ARMInstPrinter.cpp
 ARMInstPrinter.h
Message-ID: <200910192059.n9JKxuJp014309@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 15:59:55 2009
New Revision: 84532

URL: http://llvm.org/viewvc/llvm-project?rev=84532&view=rev
Log:
wire up some basic printOperand goodness, giving us stuff like this before
we abort:

_main:
	stm , 
	mov r7, sp
	sub sp, sp, 
	mov r0, 
	str r0, 


Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84532&r1=84531&r2=84532&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 15:59:55 2009
@@ -310,25 +310,25 @@
                                  const char *Modifier) {
   const MachineOperand &MO = MI->getOperand(OpNum);
   switch (MO.getType()) {
+  default:
+    assert(0 && "<unknown operand type>");
   case MachineOperand::MO_Register: {
     unsigned Reg = MO.getReg();
-    if (TargetRegisterInfo::isPhysicalRegister(Reg)) {
-      if (Modifier && strcmp(Modifier, "dregpair") == 0) {
-        unsigned DRegLo = TRI->getSubReg(Reg, 5); // arm_dsubreg_0
-        unsigned DRegHi = TRI->getSubReg(Reg, 6); // arm_dsubreg_1
-        O << '{'
-          << getRegisterName(DRegLo) << ',' << getRegisterName(DRegHi)
-          << '}';
-      } else if (Modifier && strcmp(Modifier, "lane") == 0) {
-        unsigned RegNum = ARMRegisterInfo::getRegisterNumbering(Reg);
-        unsigned DReg = TRI->getMatchingSuperReg(Reg, RegNum & 1 ? 2 : 1,
-                                                 &ARM::DPR_VFP2RegClass);
-        O << getRegisterName(DReg) << '[' << (RegNum & 1) << ']';
-      } else {
-        O << getRegisterName(Reg);
-      }
-    } else
-      llvm_unreachable("not implemented");
+    assert(TargetRegisterInfo::isPhysicalRegister(Reg));
+    if (Modifier && strcmp(Modifier, "dregpair") == 0) {
+      unsigned DRegLo = TRI->getSubReg(Reg, 5); // arm_dsubreg_0
+      unsigned DRegHi = TRI->getSubReg(Reg, 6); // arm_dsubreg_1
+      O << '{'
+        << getRegisterName(DRegLo) << ',' << getRegisterName(DRegHi)
+        << '}';
+    } else if (Modifier && strcmp(Modifier, "lane") == 0) {
+      unsigned RegNum = ARMRegisterInfo::getRegisterNumbering(Reg);
+      unsigned DReg = TRI->getMatchingSuperReg(Reg, RegNum & 1 ? 2 : 1,
+                                               &ARM::DPR_VFP2RegClass);
+      O << getRegisterName(DReg) << '[' << (RegNum & 1) << ']';
+    } else {
+      O << getRegisterName(Reg);
+    }
     break;
   }
   case MachineOperand::MO_Immediate: {
@@ -376,8 +376,6 @@
     O << MAI->getPrivateGlobalPrefix() << "JTI" << getFunctionNumber()
       << '_' << MO.getIndex();
     break;
-  default:
-    O << "<unknown operand type>"; abort (); break;
   }
 }
 

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84532&r1=84531&r2=84532&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 15:59:55 2009
@@ -31,3 +31,19 @@
 
 void ARMInstPrinter::printInst(const MCInst *MI) { printInstruction(MI); }
 
+void ARMInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
+                                  const char *Modifier) {
+  assert((Modifier == 0 || Modifier[0] == 0) && "Cannot print modifiers");
+  
+  const MCOperand &Op = MI->getOperand(OpNo);
+  if (Op.isReg()) {
+    O << getRegisterName(Op.getReg());
+  } else if (Op.isImm()) {
+    O << '#' << Op.getImm();
+  } else {
+    assert(Op.isExpr() && "unknown operand kind in printOperand");
+    assert(0 && "UNIMP");
+    //O << '$';
+    //Op.getExpr()->print(O, &MAI);
+  }
+}

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84532&r1=84531&r2=84532&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 15:59:55 2009
@@ -31,7 +31,7 @@
 
 
   void printOperand(const MCInst *MI, unsigned OpNo,
-                    const char *Modifier = 0) {}
+                    const char *Modifier = 0);
     
   void printSOImmOperand(const MCInst *MI, unsigned OpNum) {}
   void printSOImm2PartOperand(const MCInst *MI, unsigned OpNum) {}




From daniel at zuster.org  Mon Oct 19 16:08:02 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 14:08:02 -0700
Subject: [llvm-commits] [llvm] r84355 - in /llvm/trunk:
	include/llvm/Support/raw_ostream.h lib/Support/raw_ostream.cpp
	unittests/Support/raw_ostream_test.cpp
In-Reply-To: <C7847C3C-6DFD-46B5-941C-B0995D20B123@apple.com>
References: <200910172043.n9HKh9O1022980@zion.cs.uiuc.edu>
	<C7847C3C-6DFD-46B5-941C-B0995D20B123@apple.com>
Message-ID: <6a8523d60910191408k31514ce0qe91366c4ae89c19d@mail.gmail.com>

On Mon, Oct 19, 2009 at 10:27 AM, Jim Grosbach <grosbach at apple.com> wrote:
> Hi Daniel,
>
> This looks very nice. Definitely an improvement over EscapeString.
>
> I assume the choice to restrict the list handled explicitly and not include
> things like '\v' and '\b' is due to those being uncommon enough to just
> leave to the generic octal expansion?

I was just stealing the logic from EscapeString. But I do tend to
think its better to just go to octal, it is more likely to be helpful
to users of this function I expect (unless they have the \v and \b
values memorized), and the clients of this function don't need to
worry as much about whether whatever is parsing the output understands
\v and \b.

 - Daniel

> -Jim
>
> On Oct 17, 2009, at 1:43 PM, Daniel Dunbar wrote:
>
>> Author: ddunbar
>> Date: Sat Oct 17 15:43:08 2009
>> New Revision: 84355
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=84355&view=rev
>> Log:
>> Add raw_ostream::write_escaped, for writing escaped strings.
>>
>> Modified:
>> ? llvm/trunk/include/llvm/Support/raw_ostream.h
>> ? llvm/trunk/lib/Support/raw_ostream.cpp
>> ? llvm/trunk/unittests/Support/raw_ostream_test.cpp
>>
>> Modified: llvm/trunk/include/llvm/Support/raw_ostream.h
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/raw_ostream.h?rev=84355&r1=84354&r2=84355&view=diff
>>
>>
>> ==============================================================================
>> --- llvm/trunk/include/llvm/Support/raw_ostream.h (original)
>> +++ llvm/trunk/include/llvm/Support/raw_ostream.h Sat Oct 17 15:43:08 2009
>> @@ -216,6 +216,10 @@
>> ?/// write_hex - Output \arg N in hexadecimal, without any prefix or
>> padding.
>> ?raw_ostream &write_hex(unsigned long long N);
>>
>> + ?/// write_escaped - Output \arg Str, turning '\\', '\t', '\n', '"', and
>> + ?/// anything that doesn't satisfy std::isprint into an escape sequence.
>> + ?raw_ostream &write_escaped(StringRef Str);
>> +
>> ?raw_ostream &write(unsigned char C);
>> ?raw_ostream &write(const char *Ptr, size_t Size);
>>
>>
>> Modified: llvm/trunk/lib/Support/raw_ostream.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Support/raw_ostream.cpp?rev=84355&r1=84354&r2=84355&view=diff
>>
>>
>> ==============================================================================
>> --- llvm/trunk/lib/Support/raw_ostream.cpp (original)
>> +++ llvm/trunk/lib/Support/raw_ostream.cpp Sat Oct 17 15:43:08 2009
>> @@ -168,6 +168,40 @@
>> ?return write(CurPtr, EndPtr-CurPtr);
>> }
>>
>> +raw_ostream &raw_ostream::write_escaped(StringRef Str) {
>> + ?for (unsigned i = 0, e = Str.size(); i != e; ++i) {
>> + ? ?unsigned char c = Str[i];
>> +
>> + ? ?switch (c) {
>> + ? ?case '\\':
>> + ? ? ?*this << '\\' << '\\';
>> + ? ? ?break;
>> + ? ?case '\t':
>> + ? ? ?*this << '\\' << 't';
>> + ? ? ?break;
>> + ? ?case '\n':
>> + ? ? ?*this << '\\' << 'n';
>> + ? ? ?break;
>> + ? ?case '"':
>> + ? ? ?*this << '\\' << '"';
>> + ? ? ?break;
>> + ? ?default:
>> + ? ? ?if (std::isprint(c)) {
>> + ? ? ? ?*this << c;
>> + ? ? ? ?break;
>> + ? ? ?}
>> +
>> + ? ? ?// Always expand to a 3-character octal escape.
>> + ? ? ?*this << '\\';
>> + ? ? ?*this << char('0' + ((c >> 6) & 7));
>> + ? ? ?*this << char('0' + ((c >> 3) & 7));
>> + ? ? ?*this << char('0' + ((c >> 0) & 7));
>> + ? ?}
>> + ?}
>> +
>> + ?return *this;
>> +}
>> +
>> raw_ostream &raw_ostream::operator<<(const void *P) {
>> ?*this << '0' << 'x';
>>
>>
>> Modified: llvm/trunk/unittests/Support/raw_ostream_test.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/Support/raw_ostream_test.cpp?rev=84355&r1=84354&r2=84355&view=diff
>>
>>
>> ==============================================================================
>> --- llvm/trunk/unittests/Support/raw_ostream_test.cpp (original)
>> +++ llvm/trunk/unittests/Support/raw_ostream_test.cpp Sat Oct 17 15:43:08
>> 2009
>> @@ -127,4 +127,20 @@
>> ?EXPECT_EQ("hello1world", OS.str());
>> }
>>
>> +TEST(raw_ostreamTest, WriteEscaped) {
>> + ?std::string Str;
>> +
>> + ?Str = "";
>> + ?raw_string_ostream(Str).write_escaped("hi");
>> + ?EXPECT_EQ("hi", Str);
>> +
>> + ?Str = "";
>> + ?raw_string_ostream(Str).write_escaped("\\\t\n\"");
>> + ?EXPECT_EQ("\\\\\\t\\n\\\"", Str);
>> +
>> + ?Str = "";
>> + ?raw_string_ostream(Str).write_escaped("\1\10\200");
>> + ?EXPECT_EQ("\\001\\010\\200", Str);
>> +}
>> +
>> }
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>



From daniel at zuster.org  Mon Oct 19 16:12:16 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 14:12:16 -0700
Subject: [llvm-commits] [llvm] r84507 - in /llvm/trunk: Makefile
	utils/buildit/GNUmakefile utils/buildit/build_llvm
In-Reply-To: <200910191753.n9JHrwL2006834@zion.cs.uiuc.edu>
References: <200910191753.n9JHrwL2006834@zion.cs.uiuc.edu>
Message-ID: <6a8523d60910191412t4e63186cre37e8db21587e349@mail.gmail.com>

Hi Stuart,

This is breaking Clang's Apple style build, see our projects builder.

Can you find another solution, or perhaps limit this to only affect
the LLVM build?

 - Daniel

On Mon, Oct 19, 2009 at 10:53 AM, Stuart Hastings <stuart at apple.com> wrote:
> Author: stuart
> Date: Mon Oct 19 12:53:54 2009
> New Revision: 84507
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84507&view=rev
> Log:
> Tweak top-level Makefile to facilitate Apple-style build.
>
> Modified:
> ? ?llvm/trunk/Makefile
> ? ?llvm/trunk/utils/buildit/GNUmakefile
> ? ?llvm/trunk/utils/buildit/build_llvm
>
> Modified: llvm/trunk/Makefile
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/Makefile?rev=84507&r1=84506&r2=84507&view=diff
>
> ==============================================================================
> --- llvm/trunk/Makefile (original)
> +++ llvm/trunk/Makefile Mon Oct 19 12:53:54 2009
> @@ -19,6 +19,7 @@
> ?#
> ?# When cross-compiling, there are some things (tablegen) that need to
> ?# be build for the build system first.
> +ifndef RC_ARCHS ?# Normal build (not "Apple-style").
> ?ifeq ($(BUILD_DIRS_ONLY),1)
> ? DIRS := lib/System lib/Support utils
> ? OPTIONAL_DIRS :=
> @@ -211,3 +212,8 @@
>
> ?.NOTPARALLEL:
>
> +else # Building "Apple-style."
> +# N.B. In an Apple-style build, once configuration is done, lines
> +# marked "Apple-style" are removed with sed! ?See utils/buildit/build_llvm.
> +include utils/buildit/GNUmakefile # Building "Apple-style."
> +endif # Building "Apple-style."
>
> Modified: llvm/trunk/utils/buildit/GNUmakefile
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/GNUmakefile?rev=84507&r1=84506&r2=84507&view=diff
>
> ==============================================================================
> --- llvm/trunk/utils/buildit/GNUmakefile (original)
> +++ llvm/trunk/utils/buildit/GNUmakefile Mon Oct 19 12:53:54 2009
> @@ -59,7 +59,7 @@
> ?# NOTE : Always put version numbers at the end because they are optional.
> ?install: $(OBJROOT) $(SYMROOT) $(DSTROOT)
> ? ? ? ?cd $(OBJROOT) && \
> - ? ? ? ? $(SRC)/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
> + ? ? ? ? $(SRC)/utils/buildit/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
> ? ? ? ? ? ?$(SRC) $(PREFIX) $(DSTROOT) $(SYMROOT) \
> ? ? ? ? ? ?$(LLVM_ASSERTIONS) $(LLVM_OPTIMIZED) \
> ? ? ? ? ? ?$(RC_ProjectSourceVersion) $(RC_ProjectSourceSubversion)
>
> Modified: llvm/trunk/utils/buildit/build_llvm
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/build_llvm?rev=84507&r1=84506&r2=84507&view=diff
>
> ==============================================================================
> --- llvm/trunk/utils/buildit/build_llvm (original)
> +++ llvm/trunk/utils/buildit/build_llvm Mon Oct 19 12:53:54 2009
> @@ -82,6 +82,10 @@
> ?rm -rf $SRC_DIR || exit 1
> ?mkdir $SRC_DIR || exit 1
> ?ln -s $ORIG_SRC_DIR/* $SRC_DIR/ || exit 1
> +# We can't use the top-level Makefile as-is. ?Remove the soft link:
> +rm $SRC_DIR/Makefile || exit 1
> +# Now create our own by editing the top-level Makefile, deleting every line marked "Apple-style":
> +sed -e '/[Aa]pple-style/d' -e '/include.*GNUmakefile/d' $ORIG_SRC_DIR/Makefile > $SRC_DIR/Makefile || exit 1
>
> ?# Build the LLVM tree universal.
> ?mkdir -p $DIR/obj-llvm || exit 1
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>



From daniel at zuster.org  Mon Oct 19 16:13:52 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 14:13:52 -0700
Subject: [llvm-commits] [llvm] r84509 - in /llvm/trunk/lib/Target:
	ARM/AsmPrinter/ARMAsmPrinter.cpp
	PowerPC/AsmPrinter/PPCAsmPrinter.cpp
	X86/AsmPrinter/X86AsmPrinter.cpp
In-Reply-To: <200910191803.n9JI38HF007227@zion.cs.uiuc.edu>
References: <200910191803.n9JI38HF007227@zion.cs.uiuc.edu>
Message-ID: <6a8523d60910191413hc352570i845f29b1c473883f@mail.gmail.com>

On Mon, Oct 19, 2009 at 11:03 AM, Chris Lattner <sabre at nondot.org> wrote:
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp (original)
> +++ llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp Mon Oct 19 13:03:08 2009
> @@ -870,7 +870,9 @@
> ? ? // implementation of multiple entry points). ?If this doesn't occur, the
> ? ? // linker can safely perform dead code stripping. ?Since LLVM never
> ? ? // generates code that does this, it is always safe to set.
> - ? ?O << "\t.subsections_via_symbols\n";
> + ? ?//O << "\t.subsections_via_symbols\n";
> +

Was this left intentionally?

 - Daniel



From daniel at zuster.org  Mon Oct 19 16:14:19 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 14:14:19 -0700
Subject: [llvm-commits] [llvm] r84509 - in /llvm/trunk/lib/Target:
	ARM/AsmPrinter/ARMAsmPrinter.cpp
	PowerPC/AsmPrinter/PPCAsmPrinter.cpp
	X86/AsmPrinter/X86AsmPrinter.cpp
In-Reply-To: <6a8523d60910191413hc352570i845f29b1c473883f@mail.gmail.com>
References: <200910191803.n9JI38HF007227@zion.cs.uiuc.edu>
	<6a8523d60910191413hc352570i845f29b1c473883f@mail.gmail.com>
Message-ID: <6a8523d60910191414u6878b22dye5bcd41ec67399a9@mail.gmail.com>

Nevermind, I see you fixed it.

On Mon, Oct 19, 2009 at 2:13 PM, Daniel Dunbar <daniel at zuster.org> wrote:
> On Mon, Oct 19, 2009 at 11:03 AM, Chris Lattner <sabre at nondot.org> wrote:
>> ==============================================================================
>> --- llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp (original)
>> +++ llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp Mon Oct 19 13:03:08 2009
>> @@ -870,7 +870,9 @@
>> ? ? // implementation of multiple entry points). ?If this doesn't occur, the
>> ? ? // linker can safely perform dead code stripping. ?Since LLVM never
>> ? ? // generates code that does this, it is always safe to set.
>> - ? ?O << "\t.subsections_via_symbols\n";
>> + ? ?//O << "\t.subsections_via_symbols\n";
>> +
>
> Was this left intentionally?
>
> ?- Daniel
>



From resistor at mac.com  Mon Oct 19 16:14:57 2009
From: resistor at mac.com (Owen Anderson)
Date: Mon, 19 Oct 2009 21:14:57 -0000
Subject: [llvm-commits] [llvm] r84533 -
	/llvm/trunk/lib/Transforms/Scalar/GVN.cpp
Message-ID: <200910192114.n9JLEvw6014926@zion.cs.uiuc.edu>

Author: resistor
Date: Mon Oct 19 16:14:57 2009
New Revision: 84533

URL: http://llvm.org/viewvc/llvm-project?rev=84533&view=rev
Log:
Simplify some code.

Modified:
    llvm/trunk/lib/Transforms/Scalar/GVN.cpp

Modified: llvm/trunk/lib/Transforms/Scalar/GVN.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/GVN.cpp?rev=84533&r1=84532&r2=84533&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/GVN.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/GVN.cpp Mon Oct 19 16:14:57 2009
@@ -429,31 +429,24 @@
 
   if (CallInst* C = dyn_cast<CallInst>(V)) {
     if (AA->doesNotAccessMemory(C)) {
-      Expression e = create_expression(C);
-
-      DenseMap<Expression, uint32_t>::iterator EI = expressionNumbering.find(e);
-      if (EI != expressionNumbering.end()) {
-        valueNumbering.insert(std::make_pair(V, EI->second));
-        return EI->second;
-      } else {
-        expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-        valueNumbering.insert(std::make_pair(V, nextValueNumber));
-
-        return nextValueNumber++;
-      }
+      Expression exp = create_expression(C);
+      uint32_t& e = expressionNumbering[exp];
+      if (!e) e = nextValueNumber++;
+      valueNumbering[V] = e;
+      return e;
     } else if (AA->onlyReadsMemory(C)) {
-      Expression e = create_expression(C);
-
-      if (expressionNumbering.find(e) == expressionNumbering.end()) {
-        expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-        valueNumbering.insert(std::make_pair(V, nextValueNumber));
-        return nextValueNumber++;
+      Expression exp = create_expression(C);
+      uint32_t& e = expressionNumbering[exp];
+      if (!e) {
+        e = nextValueNumber++;
+        valueNumbering[V] = e;
+        return e;
       }
 
       MemDepResult local_dep = MD->getDependency(C);
 
       if (!local_dep.isDef() && !local_dep.isNonLocal()) {
-        valueNumbering.insert(std::make_pair(V, nextValueNumber));
+        valueNumbering[V] =  nextValueNumber;
         return nextValueNumber++;
       }
 
@@ -461,7 +454,7 @@
         CallInst* local_cdep = cast<CallInst>(local_dep.getInst());
 
         if (local_cdep->getNumOperands() != C->getNumOperands()) {
-          valueNumbering.insert(std::make_pair(V, nextValueNumber));
+          valueNumbering[V] = nextValueNumber;
           return nextValueNumber++;
         }
 
@@ -469,13 +462,13 @@
           uint32_t c_vn = lookup_or_add(C->getOperand(i));
           uint32_t cd_vn = lookup_or_add(local_cdep->getOperand(i));
           if (c_vn != cd_vn) {
-            valueNumbering.insert(std::make_pair(V, nextValueNumber));
+            valueNumbering[V] = nextValueNumber;
             return nextValueNumber++;
           }
         }
 
         uint32_t v = lookup_or_add(local_cdep);
-        valueNumbering.insert(std::make_pair(V, v));
+        valueNumbering[V] = v;
         return v;
       }
 
@@ -513,137 +506,81 @@
       }
 
       if (!cdep) {
-        valueNumbering.insert(std::make_pair(V, nextValueNumber));
+        valueNumbering[V] = nextValueNumber;
         return nextValueNumber++;
       }
 
       if (cdep->getNumOperands() != C->getNumOperands()) {
-        valueNumbering.insert(std::make_pair(V, nextValueNumber));
+        valueNumbering[V] = nextValueNumber;
         return nextValueNumber++;
       }
       for (unsigned i = 1; i < C->getNumOperands(); ++i) {
         uint32_t c_vn = lookup_or_add(C->getOperand(i));
         uint32_t cd_vn = lookup_or_add(cdep->getOperand(i));
         if (c_vn != cd_vn) {
-          valueNumbering.insert(std::make_pair(V, nextValueNumber));
+          valueNumbering[V] = nextValueNumber;
           return nextValueNumber++;
         }
       }
 
       uint32_t v = lookup_or_add(cdep);
-      valueNumbering.insert(std::make_pair(V, v));
+      valueNumbering[V] = v;
       return v;
 
     } else {
-      valueNumbering.insert(std::make_pair(V, nextValueNumber));
+      valueNumbering[V] = nextValueNumber;
       return nextValueNumber++;
     }
   } else if (BinaryOperator* BO = dyn_cast<BinaryOperator>(V)) {
-    Expression e = create_expression(BO);
-
-    DenseMap<Expression, uint32_t>::iterator EI = expressionNumbering.find(e);
-    if (EI != expressionNumbering.end()) {
-      valueNumbering.insert(std::make_pair(V, EI->second));
-      return EI->second;
-    } else {
-      expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-      valueNumbering.insert(std::make_pair(V, nextValueNumber));
-
-      return nextValueNumber++;
-    }
+    Expression exp = create_expression(BO);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) e = nextValueNumber++;
+    valueNumbering[V] = e;
+    return e;
   } else if (CmpInst* C = dyn_cast<CmpInst>(V)) {
-    Expression e = create_expression(C);
-
-    DenseMap<Expression, uint32_t>::iterator EI = expressionNumbering.find(e);
-    if (EI != expressionNumbering.end()) {
-      valueNumbering.insert(std::make_pair(V, EI->second));
-      return EI->second;
-    } else {
-      expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-      valueNumbering.insert(std::make_pair(V, nextValueNumber));
-
-      return nextValueNumber++;
-    }
+    Expression exp = create_expression(C);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) e = nextValueNumber++;
+    valueNumbering[V] = e;
+    return e;
   } else if (ShuffleVectorInst* U = dyn_cast<ShuffleVectorInst>(V)) {
-    Expression e = create_expression(U);
-
-    DenseMap<Expression, uint32_t>::iterator EI = expressionNumbering.find(e);
-    if (EI != expressionNumbering.end()) {
-      valueNumbering.insert(std::make_pair(V, EI->second));
-      return EI->second;
-    } else {
-      expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-      valueNumbering.insert(std::make_pair(V, nextValueNumber));
-
-      return nextValueNumber++;
-    }
+    Expression exp = create_expression(U);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) e = nextValueNumber++;
+    valueNumbering[V] = e;
+    return e;
   } else if (ExtractElementInst* U = dyn_cast<ExtractElementInst>(V)) {
-    Expression e = create_expression(U);
-
-    DenseMap<Expression, uint32_t>::iterator EI = expressionNumbering.find(e);
-    if (EI != expressionNumbering.end()) {
-      valueNumbering.insert(std::make_pair(V, EI->second));
-      return EI->second;
-    } else {
-      expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-      valueNumbering.insert(std::make_pair(V, nextValueNumber));
-
-      return nextValueNumber++;
-    }
+    Expression exp = create_expression(U);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) e = nextValueNumber++;
+    valueNumbering[V] = e;
+    return e;
   } else if (InsertElementInst* U = dyn_cast<InsertElementInst>(V)) {
-    Expression e = create_expression(U);
-
-    DenseMap<Expression, uint32_t>::iterator EI = expressionNumbering.find(e);
-    if (EI != expressionNumbering.end()) {
-      valueNumbering.insert(std::make_pair(V, EI->second));
-      return EI->second;
-    } else {
-      expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-      valueNumbering.insert(std::make_pair(V, nextValueNumber));
-
-      return nextValueNumber++;
-    }
+    Expression exp = create_expression(U);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) e = nextValueNumber++;
+    valueNumbering[V] = e;
+    return e;
   } else if (SelectInst* U = dyn_cast<SelectInst>(V)) {
-    Expression e = create_expression(U);
-
-    DenseMap<Expression, uint32_t>::iterator EI = expressionNumbering.find(e);
-    if (EI != expressionNumbering.end()) {
-      valueNumbering.insert(std::make_pair(V, EI->second));
-      return EI->second;
-    } else {
-      expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-      valueNumbering.insert(std::make_pair(V, nextValueNumber));
-
-      return nextValueNumber++;
-    }
+    Expression exp = create_expression(U);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) e = nextValueNumber++;
+    valueNumbering[V] = e;
+    return e;
   } else if (CastInst* U = dyn_cast<CastInst>(V)) {
-    Expression e = create_expression(U);
-
-    DenseMap<Expression, uint32_t>::iterator EI = expressionNumbering.find(e);
-    if (EI != expressionNumbering.end()) {
-      valueNumbering.insert(std::make_pair(V, EI->second));
-      return EI->second;
-    } else {
-      expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-      valueNumbering.insert(std::make_pair(V, nextValueNumber));
-
-      return nextValueNumber++;
-    }
+    Expression exp = create_expression(U);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) e = nextValueNumber++;
+    valueNumbering[V] = e;
+    return e;
   } else if (GetElementPtrInst* U = dyn_cast<GetElementPtrInst>(V)) {
-    Expression e = create_expression(U);
-
-    DenseMap<Expression, uint32_t>::iterator EI = expressionNumbering.find(e);
-    if (EI != expressionNumbering.end()) {
-      valueNumbering.insert(std::make_pair(V, EI->second));
-      return EI->second;
-    } else {
-      expressionNumbering.insert(std::make_pair(e, nextValueNumber));
-      valueNumbering.insert(std::make_pair(V, nextValueNumber));
-
-      return nextValueNumber++;
-    }
+    Expression exp = create_expression(U);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) e = nextValueNumber++;
+    valueNumbering[V] = e;
+    return e;
   } else {
-    valueNumbering.insert(std::make_pair(V, nextValueNumber));
+    valueNumbering[V] = nextValueNumber;
     return nextValueNumber++;
   }
 }




From sabre at nondot.org  Mon Oct 19 16:21:39 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 21:21:39 -0000
Subject: [llvm-commits] [llvm] r84535 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMAsmPrinter.cpp ARMInstPrinter.cpp
 ARMInstPrinter.h
Message-ID: <200910192121.n9JLLdbS015191@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 16:21:39 2009
New Revision: 84535

URL: http://llvm.org/viewvc/llvm-project?rev=84535&view=rev
Log:
add printing support for SOImm operands, getting us to:

_main:
	stm , 
	mov r7, sp
	sub sp, sp, #4
	mov r0, #0
	str r0, 


Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84535&r1=84534&r2=84535&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 16:21:39 2009
@@ -85,7 +85,7 @@
     }
     
     void printMCInst(const MCInst *MI) {
-      ARMInstPrinter(O, *MAI).printInstruction(MI);
+      ARMInstPrinter(O, *MAI, VerboseAsm).printInstruction(MI);
     }
     
     void printInstructionThroughMCStreamer(const MachineInstr *MI);

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84535&r1=84534&r2=84535&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 16:21:39 2009
@@ -13,8 +13,9 @@
 
 #define DEBUG_TYPE "asm-printer"
 #include "ARMInstPrinter.h"
+#include "ARMAddressingModes.h"
 #include "llvm/MC/MCInst.h"
-//#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCAsmInfo.h"
 //#include "llvm/MC/MCExpr.h"
 //#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
@@ -47,3 +48,34 @@
     //Op.getExpr()->print(O, &MAI);
   }
 }
+
+static void printSOImm(raw_ostream &O, int64_t V, bool VerboseAsm,
+                       const MCAsmInfo *MAI) {
+  // Break it up into two parts that make up a shifter immediate.
+  V = ARM_AM::getSOImmVal(V);
+  assert(V != -1 && "Not a valid so_imm value!");
+  
+  unsigned Imm = ARM_AM::getSOImmValImm(V);
+  unsigned Rot = ARM_AM::getSOImmValRot(V);
+  
+  // Print low-level immediate formation info, per
+  // A5.1.3: "Data-processing operands - Immediate".
+  if (Rot) {
+    O << "#" << Imm << ", " << Rot;
+    // Pretty printed version.
+    if (VerboseAsm)
+      O << ' ' << MAI->getCommentString()
+      << ' ' << (int)ARM_AM::rotr32(Imm, Rot);
+  } else {
+    O << "#" << Imm;
+  }
+}
+
+
+/// printSOImmOperand - SOImm is 4-bit rotate amount in bits 8-11 with 8-bit
+/// immediate in bits 0-7.
+void ARMInstPrinter::printSOImmOperand(const MCInst *MI, unsigned OpNum) {
+  const MCOperand &MO = MI->getOperand(OpNum);
+  assert(MO.isImm() && "Not a valid so_imm value!");
+  printSOImm(O, MO.getImm(), VerboseAsm, &MAI);
+}

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84535&r1=84534&r2=84535&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 16:21:39 2009
@@ -20,8 +20,10 @@
   class MCOperand;
   
 class ARMInstPrinter : public MCInstPrinter {
+  bool VerboseAsm;
 public:
-  ARMInstPrinter(raw_ostream &O, const MCAsmInfo &MAI) : MCInstPrinter(O, MAI){}
+  ARMInstPrinter(raw_ostream &O, const MCAsmInfo &MAI, bool verboseAsm)
+    : MCInstPrinter(O, MAI), VerboseAsm(verboseAsm) {}
 
   virtual void printInst(const MCInst *MI);
   
@@ -33,7 +35,9 @@
   void printOperand(const MCInst *MI, unsigned OpNo,
                     const char *Modifier = 0);
     
-  void printSOImmOperand(const MCInst *MI, unsigned OpNum) {}
+  void printSOImmOperand(const MCInst *MI, unsigned OpNum);
+  
+  
   void printSOImm2PartOperand(const MCInst *MI, unsigned OpNum) {}
   void printSORegOperand(const MCInst *MI, unsigned OpNum) {}
   void printAddrMode2Operand(const MCInst *MI, unsigned OpNum) {}




From sabre at nondot.org  Mon Oct 19 16:23:15 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 21:23:15 -0000
Subject: [llvm-commits] [llvm] r84536 -
	/llvm/trunk/lib/Target/ARM/ARMAddressingModes.h
Message-ID: <200910192123.n9JLNF2o015248@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 16:23:15 2009
New Revision: 84536

URL: http://llvm.org/viewvc/llvm-project?rev=84536&view=rev
Log:
reduce #includes

Modified:
    llvm/trunk/lib/Target/ARM/ARMAddressingModes.h

Modified: llvm/trunk/lib/Target/ARM/ARMAddressingModes.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMAddressingModes.h?rev=84536&r1=84535&r2=84536&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMAddressingModes.h (original)
+++ llvm/trunk/lib/Target/ARM/ARMAddressingModes.h Mon Oct 19 16:23:15 2009
@@ -15,7 +15,6 @@
 #define LLVM_TARGET_ARM_ARMADDRESSINGMODES_H
 
 #include "llvm/CodeGen/SelectionDAGNodes.h"
-#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/MathExtras.h"
 #include <cassert>
 
@@ -38,7 +37,7 @@
 
   static inline const char *getShiftOpcStr(ShiftOpc Op) {
     switch (Op) {
-    default: llvm_unreachable("Unknown shift opc!");
+    default: assert(0 && "Unknown shift opc!");
     case ARM_AM::asr: return "asr";
     case ARM_AM::lsl: return "lsl";
     case ARM_AM::lsr: return "lsr";
@@ -71,7 +70,7 @@
 
   static inline const char *getAMSubModeStr(AMSubMode Mode) {
     switch (Mode) {
-    default: llvm_unreachable("Unknown addressing sub-mode!");
+    default: assert(0 && "Unknown addressing sub-mode!");
     case ARM_AM::ia: return "ia";
     case ARM_AM::ib: return "ib";
     case ARM_AM::da: return "da";
@@ -81,7 +80,7 @@
 
   static inline const char *getAMSubModeAltStr(AMSubMode Mode, bool isLD) {
     switch (Mode) {
-    default: llvm_unreachable("Unknown addressing sub-mode!");
+    default: assert(0 && "Unknown addressing sub-mode!");
     case ARM_AM::ia: return isLD ? "fd" : "ea";
     case ARM_AM::ib: return isLD ? "ed" : "fa";
     case ARM_AM::da: return isLD ? "fa" : "ed";




From foldr at codedgers.com  Mon Oct 19 16:24:28 2009
From: foldr at codedgers.com (Mikhail Glushenkov)
Date: Mon, 19 Oct 2009 21:24:28 -0000
Subject: [llvm-commits] [llvm] r84537 -
	/llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp
Message-ID: <200910192124.n9JLOSaJ015293@zion.cs.uiuc.edu>

Author: foldr
Date: Mon Oct 19 16:24:28 2009
New Revision: 84537

URL: http://llvm.org/viewvc/llvm-project?rev=84537&view=rev
Log:
More refactoring...

Modified:
    llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp

Modified: llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp?rev=84537&r1=84536&r2=84537&view=diff

==============================================================================
--- llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp (original)
+++ llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp Mon Oct 19 16:24:28 2009
@@ -1279,12 +1279,21 @@
     : Callback_(Callback), O_(O)
   {}
 
-  // TODO: Handle lists here.
   void operator() (const Init* Statement, unsigned IndentLevel) {
+
     // Ignore nested 'case' DAG.
     if (!(dynamic_cast<const DagInit*>(Statement) &&
-          GetOperatorName(static_cast<const DagInit*>(Statement)) == "case"))
-      Callback_(Statement, (IndentLevel + Indent1), O_);
+          GetOperatorName(static_cast<const DagInit*>(Statement)) == "case")) {
+      if (typeid(*Statement) == typeid(ListInit)) {
+        const ListInit& DagList = *static_cast<const ListInit*>(Statement);
+        for (ListInit::const_iterator B = DagList.begin(), E = DagList.end();
+             B != E; ++B)
+          Callback_(*B, (IndentLevel + Indent1), O_);
+      }
+      else {
+        Callback_(Statement, (IndentLevel + Indent1), O_);
+      }
+    }
     O_.indent(IndentLevel) << "}\n";
   }
 
@@ -1602,10 +1611,33 @@
   }
 }
 
-/// EmitActionHandler - Emit code that handles actions. Used by
-/// EmitGenerateActionMethod() as an argument to
-/// EmitCaseConstructHandler().
-class EmitActionHandler {
+/// ActionHandlingCallbackBase - Base class of EmitActionHandlersCallback and
+/// EmitPreprocessOptionsCallback.
+struct ActionHandlingCallbackBase {
+
+  void onErrorDag(const DagInit& d,
+                  unsigned IndentLevel, raw_ostream& O) const
+  {
+    O.indent(IndentLevel)
+      << "throw std::runtime_error(\"" <<
+      (d.getNumArgs() >= 1 ? InitPtrToString(d.getArg(0))
+       : "Unknown error!")
+      << "\");\n";
+  }
+
+  void onWarningDag(const DagInit& d,
+                    unsigned IndentLevel, raw_ostream& O) const
+  {
+    checkNumberOfArguments(&d, 1);
+    O.indent(IndentLevel) << "llvm::errs() << \""
+                          << InitPtrToString(d.getArg(0)) << "\";\n";
+  }
+
+};
+
+/// EmitActionHandlersCallback - Emit code that handles actions. Used by
+/// EmitGenerateActionMethod() as an argument to EmitCaseConstructHandler().
+class EmitActionHandlersCallback : ActionHandlingCallbackBase {
   const OptionDescriptions& OptDescs;
 
   void processActionDag(const Init* Statement, unsigned IndentLevel,
@@ -1625,10 +1657,10 @@
         O.indent(IndentLevel) << "vec.push_back(\"" << *B << "\");\n";
     }
     else if (ActionName == "error") {
-      O.indent(IndentLevel) << "throw std::runtime_error(\"" <<
-        (Dag.getNumArgs() >= 1 ? InitPtrToString(Dag.getArg(0))
-         : "Unknown error!")
-        << "\");\n";
+      this->onErrorDag(Dag, IndentLevel, O);
+    }
+    else if (ActionName == "warning") {
+      this->onWarningDag(Dag, IndentLevel, O);
     }
     else if (ActionName == "forward") {
       checkNumberOfArguments(&Dag, 1);
@@ -1682,21 +1714,13 @@
     }
   }
  public:
-  EmitActionHandler(const OptionDescriptions& OD)
+  EmitActionHandlersCallback(const OptionDescriptions& OD)
     : OptDescs(OD) {}
 
-  void operator()(const Init* Statement, unsigned IndentLevel,
-                  raw_ostream& O) const
+  void operator()(const Init* Statement,
+                  unsigned IndentLevel, raw_ostream& O) const
   {
-    if (typeid(*Statement) == typeid(ListInit)) {
-      const ListInit& DagList = *static_cast<const ListInit*>(Statement);
-      for (ListInit::const_iterator B = DagList.begin(), E = DagList.end();
-           B != E; ++B)
-        this->processActionDag(*B, IndentLevel, O);
-    }
-    else {
-      this->processActionDag(Statement, IndentLevel, O);
-    }
+    this->processActionDag(Statement, IndentLevel, O);
   }
 };
 
@@ -1791,7 +1815,7 @@
 
   // For every understood option, emit handling code.
   if (D.Actions)
-    EmitCaseConstructHandler(D.Actions, Indent2, EmitActionHandler(OptDescs),
+    EmitCaseConstructHandler(D.Actions, Indent2, EmitActionHandlersCallback(OptDescs),
                              false, OptDescs, O);
 
   O << '\n';
@@ -2009,9 +2033,9 @@
   O << '\n';
 }
 
-/// PreprocessOptionsCallback - Helper function passed to
+/// EmitPreprocessOptionsCallback - Helper function passed to
 /// EmitCaseConstructHandler() by EmitPreprocessOptions().
-class PreprocessOptionsCallback {
+class EmitPreprocessOptionsCallback : ActionHandlingCallbackBase {
   const OptionDescriptions& OptDescs_;
 
   void onUnsetOption(Init* i, unsigned IndentLevel, raw_ostream& O) {
@@ -2037,18 +2061,11 @@
     const DagInit& d = InitPtrToDag(I);
     const std::string& OpName = GetOperatorName(d);
 
-    // TOFIX: there is some duplication between this function and
-    // EmitActionHandler.
     if (OpName == "warning") {
-      checkNumberOfArguments(&d, 1);
-      O.indent(IndentLevel) << "llvm::errs() << \""
-                            << InitPtrToString(d.getArg(0)) << "\";\n";
+      this->onWarningDag(d, IndentLevel, O);
     }
     else if (OpName == "error") {
-      checkNumberOfArguments(&d, 1);
-      O.indent(IndentLevel) << "throw std::runtime_error(\""
-                            << InitPtrToString(d.getArg(0))
-                            << "\");\n";
+      this->onWarningDag(d, IndentLevel, O);
     }
     else if (OpName == "unset_option") {
       checkNumberOfArguments(&d, 1);
@@ -2071,20 +2088,11 @@
 
 public:
 
-  // TODO: Remove duplication.
   void operator()(const Init* I, unsigned IndentLevel, raw_ostream& O) {
-    if (typeid(*I) == typeid(ListInit)) {
-      const ListInit& DagList = *static_cast<const ListInit*>(I);
-      for (ListInit::const_iterator B = DagList.begin(), E = DagList.end();
-           B != E; ++B)
-        this->processDag(*B, IndentLevel, O);
-    }
-    else {
       this->processDag(I, IndentLevel, O);
-    }
   }
 
-  PreprocessOptionsCallback(const OptionDescriptions& OptDescs)
+  EmitPreprocessOptionsCallback(const OptionDescriptions& OptDescs)
   : OptDescs_(OptDescs)
   {}
 };
@@ -2101,7 +2109,8 @@
   for (RecordVector::const_iterator B = OptionPreprocessors.begin(),
          E = OptionPreprocessors.end(); B!=E; ++B) {
     DagInit* Case = (*B)->getValueAsDag("preprocessor");
-    EmitCaseConstructHandler(Case, Indent1, PreprocessOptionsCallback(OptDecs),
+    EmitCaseConstructHandler(Case, Indent1,
+                             EmitPreprocessOptionsCallback(OptDecs),
                              false, OptDecs, O);
   }
 




From clattner at apple.com  Mon Oct 19 16:24:36 2009
From: clattner at apple.com (Chris Lattner)
Date: Mon, 19 Oct 2009 14:24:36 -0700
Subject: [llvm-commits] [llvm] r84533 -
	/llvm/trunk/lib/Transforms/Scalar/GVN.cpp
In-Reply-To: <200910192114.n9JLEvw6014926@zion.cs.uiuc.edu>
References: <200910192114.n9JLEvw6014926@zion.cs.uiuc.edu>
Message-ID: <BACA3D0B-F397-416D-9C40-A57ECB7967AF@apple.com>


On Oct 19, 2009, at 2:14 PM, Owen Anderson wrote:

>       if (!local_dep.isDef() && !local_dep.isNonLocal()) {
> -        valueNumbering.insert(std::make_pair(V, nextValueNumber));
> +        valueNumbering[V] =  nextValueNumber;

Note that these have different behavior.  In the former case, if V was  
already in the map, insert is a noop.  In the later case, V gets  
updated.  Please remove the double spaces after the =.

-Chris


From sabre at nondot.org  Mon Oct 19 16:45:32 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 21:45:32 -0000
Subject: [llvm-commits] [llvm] r84540 - in
 /llvm/trunk/lib/Target/X86/AsmPrinter: X86MCInstLower.cpp X86MCInstLower.h
Message-ID: <200910192145.n9JLjWnD016059@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 16:45:31 2009
New Revision: 84540

URL: http://llvm.org/viewvc/llvm-project?rev=84540&view=rev
Log:
pass mangler in as a reference instead of a pointer.

Modified:
    llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp
    llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp?rev=84540&r1=84539&r2=84540&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp Mon Oct 19 16:45:31 2009
@@ -59,7 +59,7 @@
     isImplicitlyPrivate = true;
   
   SmallString<128> Name;
-  Mang->getNameWithPrefix(Name, GV, isImplicitlyPrivate);
+  Mang.getNameWithPrefix(Name, GV, isImplicitlyPrivate);
   
   if (getSubtarget().isTargetCygMing()) {
     X86COFFMachineModuleInfo &COFFMMI = 
@@ -86,7 +86,7 @@
     const MCSymbol *&StubSym = getMachOMMI().getGVStubEntry(Sym);
     if (StubSym == 0) {
       Name.clear();
-      Mang->getNameWithPrefix(Name, GV, false);
+      Mang.getNameWithPrefix(Name, GV, false);
       StubSym = Ctx.GetOrCreateSymbol(Name.str());
     }
     return Sym;
@@ -97,7 +97,7 @@
     const MCSymbol *&StubSym = getMachOMMI().getHiddenGVStubEntry(Sym);
     if (StubSym == 0) {
       Name.clear();
-      Mang->getNameWithPrefix(Name, GV, false);
+      Mang.getNameWithPrefix(Name, GV, false);
       StubSym = Ctx.GetOrCreateSymbol(Name.str());
     }
     return Sym;
@@ -108,7 +108,7 @@
     const MCSymbol *&StubSym = getMachOMMI().getFnStubEntry(Sym);
     if (StubSym == 0) {
       Name.clear();
-      Mang->getNameWithPrefix(Name, GV, false);
+      Mang.getNameWithPrefix(Name, GV, false);
       StubSym = Ctx.GetOrCreateSymbol(Name.str());
     }
     return Sym;
@@ -393,7 +393,7 @@
 
 
 void X86AsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI) {
-  X86MCInstLower MCInstLowering(OutContext, Mang, *this);
+  X86MCInstLower MCInstLowering(OutContext, *Mang, *this);
   switch (MI->getOpcode()) {
   case TargetInstrInfo::DBG_LABEL:
   case TargetInstrInfo::EH_LABEL:

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h?rev=84540&r1=84539&r2=84540&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h Mon Oct 19 16:45:31 2009
@@ -27,12 +27,12 @@
 /// X86MCInstLower - This class is used to lower an MachineInstr into an MCInst.
 class VISIBILITY_HIDDEN X86MCInstLower {
   MCContext &Ctx;
-  Mangler *Mang;
+  Mangler &Mang;
   X86AsmPrinter &AsmPrinter;
 
   const X86Subtarget &getSubtarget() const;
 public:
-  X86MCInstLower(MCContext &ctx, Mangler *mang, X86AsmPrinter &asmprinter)
+  X86MCInstLower(MCContext &ctx, Mangler &mang, X86AsmPrinter &asmprinter)
     : Ctx(ctx), Mang(mang), AsmPrinter(asmprinter) {}
   
   void Lower(const MachineInstr *MI, MCInst &OutMI) const;




From vhernandez at apple.com  Mon Oct 19 16:47:22 2009
From: vhernandez at apple.com (Victor Hernandez)
Date: Mon, 19 Oct 2009 21:47:22 -0000
Subject: [llvm-commits] [llvm] r84541 -
	/llvm/trunk/lib/Transforms/IPO/FunctionAttrs.cpp
Message-ID: <200910192147.n9JLlMNu016134@zion.cs.uiuc.edu>

Author: hernande
Date: Mon Oct 19 16:47:22 2009
New Revision: 84541

URL: http://llvm.org/viewvc/llvm-project?rev=84541&view=rev
Log:
Malloc calls are marked NoAlias, so the code below the isMalloc() check makes it redundant.  Removing the isMalloc() check.


Modified:
    llvm/trunk/lib/Transforms/IPO/FunctionAttrs.cpp

Modified: llvm/trunk/lib/Transforms/IPO/FunctionAttrs.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/FunctionAttrs.cpp?rev=84541&r1=84540&r2=84541&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/IPO/FunctionAttrs.cpp (original)
+++ llvm/trunk/lib/Transforms/IPO/FunctionAttrs.cpp Mon Oct 19 16:47:22 2009
@@ -269,8 +269,6 @@
         case Instruction::Alloca:
           break;
         case Instruction::Call:
-          if (isMalloc(RVI))
-            break;
         case Instruction::Invoke: {
           CallSite CS(RVI);
           if (CS.paramHasAttr(0, Attribute::NoAlias))




From vhernandez at apple.com  Mon Oct 19 16:47:39 2009
From: vhernandez at apple.com (Victor Hernandez)
Date: Mon, 19 Oct 2009 14:47:39 -0700
Subject: [llvm-commits] [llvm] r82300 - in /llvm/trunk/lib/Transforms:
	IPO/FunctionAttrs.cpp IPO/GlobalOpt.cpp Scalar/GVN.cpp
	Scalar/InstructionCombining.cpp Scalar/Reassociate.cpp
	Scalar/SCCP.cpp Scalar/SimplifyLibCalls.cpp
	Scalar/TailDuplication.cpp Utils/InlineCost.cpp
In-Reply-To: <D862AEFC-B486-40E3-ACB1-060A00DD8179@apple.com>
References: <200909182235.n8IMZnYJ001246@zion.cs.uiuc.edu>
	<D862AEFC-B486-40E3-ACB1-060A00DD8179@apple.com>
Message-ID: <91BFC0CD-3854-4237-B9B5-B30F539FD472@apple.com>

Chris,
This code review got ignored while I investigated the nightly  
regressions.

>> +          // fall through
>> +        case Instruction::GetElementPtr:
>>          FlowsToReturn.insert(RVI->getOperand(0));
>>          continue;
>>        case Instruction::Select: {
>> @@ -267,6 +271,8 @@
>>        case Instruction::Malloc:
>>          break;
>>        case Instruction::Call:
>> +          if (isMalloc(RVI))
>> +            break;
>
>
> This *shouldn't* be needed.  The result of the malloc prototype  
> should already be marked NoAlias.  Please verify that this is  
> happening (likewise for calloc, valloc, memalign, etc).  If not,  
> llvm-gcc/clang and simplifylibcalls should be updated to make this  
> happen.  One bonus of this that more stuff will benefit.
>
> I see that you already modified SimplifyLibCalls.cpp to handle  
> malloc, so this can probably be removed.

Just verified that malloc prototype is being marked NoAlias.  And  
SimplifyLibCalls is marking all of those functions NoAlias also.  I  
just deleted the isMalloc() check in rev 84541.


>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> =====================================================================
>> --- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp  
>> (original)
>> @@ -5891,9 +5893,9 @@
>>
>>  // icmp <global/alloca*/null>, <global/alloca*/null> - Global/ 
>> Stack value
>>  // addresses never equal each other!  We already know that Op0 !=  
>> Op1.
>> -  if ((isa<GlobalValue>(Op0) || isa<AllocaInst>(Op0) ||
>> +  if ((isa<GlobalValue>(Op0) || isa<AllocaInst>(Op0) || isMalloc 
>> (Op0) ||
>>       isa<ConstantPointerNull>(Op0)) &&
>> -      (isa<GlobalValue>(Op1) || isa<AllocaInst>(Op1) ||
>> +      (isa<GlobalValue>(Op1) || isa<AllocaInst>(Op1) || isMalloc 
>> (Op1) ||
>>       isa<ConstantPointerNull>(Op1)))
>
> Instead of special casing malloc here, please handle all "noalias"  
> functions the same.

I have already stopped special-casing malloc here.

Victor
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091019/27b6a503/attachment.html 

From sabre at nondot.org  Mon Oct 19 16:53:01 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 21:53:01 -0000
Subject: [llvm-commits] [llvm] r84542 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMAsmPrinter.cpp ARMInstPrinter.cpp
 ARMMCInstLower.cpp ARMMCInstLower.h
Message-ID: <200910192153.n9JLr1ob016348@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 16:53:00 2009
New Revision: 84542

URL: http://llvm.org/viewvc/llvm-project?rev=84542&view=rev
Log:
add jump tables, constant pools and some trivial global
lowering stuff.  We can now compile hello world to:

_main:
	stm , 
	mov r7, sp
	sub sp, sp, #4
	mov r0, #0
	str r0, 
	ldr r0, 
	bl _printf
	ldr r0, 
	mov sp, r7
	ldm , 

Almost looks like arm code :)


Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84542&r1=84541&r2=84542&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 16:53:00 2009
@@ -1311,7 +1311,7 @@
 //===----------------------------------------------------------------------===//
 
 void ARMAsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI) {
-  ARMMCInstLower MCInstLowering(OutContext, Mang);
+  ARMMCInstLower MCInstLowering(OutContext, *Mang, getFunctionNumber(), *MAI);
   switch (MI->getOpcode()) {
   case TargetInstrInfo::DBG_LABEL:
   case TargetInstrInfo::EH_LABEL:

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84542&r1=84541&r2=84542&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 16:53:00 2009
@@ -16,9 +16,8 @@
 #include "ARMAddressingModes.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCAsmInfo.h"
-//#include "llvm/MC/MCExpr.h"
-//#include "llvm/Support/ErrorHandling.h"
-#include "llvm/Support/FormattedStream.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/Support/raw_ostream.h"
 #include "ARMGenInstrNames.inc"
 using namespace llvm;
 
@@ -34,7 +33,8 @@
 
 void ARMInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
                                   const char *Modifier) {
-  assert((Modifier == 0 || Modifier[0] == 0) && "Cannot print modifiers");
+  // FIXME: TURN ASSERT ON.
+  //assert((Modifier == 0 || Modifier[0] == 0) && "Cannot print modifiers");
   
   const MCOperand &Op = MI->getOperand(OpNo);
   if (Op.isReg()) {
@@ -43,9 +43,7 @@
     O << '#' << Op.getImm();
   } else {
     assert(Op.isExpr() && "unknown operand kind in printOperand");
-    assert(0 && "UNIMP");
-    //O << '$';
-    //Op.getExpr()->print(O, &MAI);
+    Op.getExpr()->print(O, &MAI);
   }
 }
 

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp?rev=84542&r1=84541&r2=84542&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp Mon Oct 19 16:53:00 2009
@@ -13,16 +13,16 @@
 //===----------------------------------------------------------------------===//
 
 #include "ARMMCInstLower.h"
-//#include "ARMMCAsmInfo.h"
 //#include "llvm/CodeGen/MachineModuleInfoImpls.h"
 #include "llvm/CodeGen/MachineInstr.h"
-//#include "llvm/MC/MCContext.h"
-//#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 //#include "llvm/MC/MCStreamer.h"
-//#include "llvm/Support/FormattedStream.h"
-//#include "llvm/Support/Mangler.h"
-//#include "llvm/ADT/SmallString.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/Mangler.h"
+#include "llvm/ADT/SmallString.h"
 using namespace llvm;
 
 
@@ -37,6 +37,74 @@
 }
 #endif
 
+MCSymbol *ARMMCInstLower::
+GetGlobalAddressSymbol(const MachineOperand &MO) const {
+  const GlobalValue *GV = MO.getGlobal();
+  
+  SmallString<128> Name;
+  Mang.getNameWithPrefix(Name, GV, false);
+  
+  // FIXME: HANDLE PLT references how??
+  switch (MO.getTargetFlags()) {
+  default: assert(0 && "Unknown target flag on GV operand");
+  case 0: break;
+  }
+  
+  return Ctx.GetOrCreateSymbol(Name.str());
+}
+
+
+MCSymbol *ARMMCInstLower::
+GetJumpTableSymbol(const MachineOperand &MO) const {
+  SmallString<256> Name;
+  raw_svector_ostream(Name) << MAI.getPrivateGlobalPrefix() << "JTI"
+    << CurFunctionNumber << '_' << MO.getIndex();
+  
+#if 0
+  switch (MO.getTargetFlags()) {
+    default: llvm_unreachable("Unknown target flag on GV operand");
+  }
+#endif
+  
+  // Create a symbol for the name.
+  return Ctx.GetOrCreateSymbol(Name.str());
+}
+
+MCSymbol *ARMMCInstLower::
+GetConstantPoolIndexSymbol(const MachineOperand &MO) const {
+  SmallString<256> Name;
+  raw_svector_ostream(Name) << MAI.getPrivateGlobalPrefix() << "CPI"
+    << CurFunctionNumber << '_' << MO.getIndex();
+  
+#if 0
+  switch (MO.getTargetFlags()) {
+  default: llvm_unreachable("Unknown target flag on GV operand");
+  }
+#endif
+  
+  // Create a symbol for the name.
+  return Ctx.GetOrCreateSymbol(Name.str());
+}
+  
+MCOperand ARMMCInstLower::
+LowerSymbolOperand(const MachineOperand &MO, MCSymbol *Sym) const {
+  // FIXME: We would like an efficient form for this, so we don't have to do a
+  // lot of extra uniquing.
+  const MCExpr *Expr = MCSymbolRefExpr::Create(Sym, Ctx);
+  
+#if 0
+  switch (MO.getTargetFlags()) {
+  default: llvm_unreachable("Unknown target flag on GV operand");
+  }
+#endif
+  
+  if (!MO.isJTI() && MO.getOffset())
+    Expr = MCBinaryExpr::CreateAdd(Expr,
+                                   MCConstantExpr::Create(MO.getOffset(), Ctx),
+                                   Ctx);
+  return MCOperand::CreateExpr(Expr);
+}
+
 
 void ARMMCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
   OutMI.setOpcode(MI->getOpcode());
@@ -60,19 +128,21 @@
       MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
                        AsmPrinter.GetMBBSymbol(MO.getMBB()->getNumber()), Ctx));
       break;
+#endif
     case MachineOperand::MO_GlobalAddress:
       MCOp = LowerSymbolOperand(MO, GetGlobalAddressSymbol(MO));
       break;
+#if 0
     case MachineOperand::MO_ExternalSymbol:
       MCOp = LowerSymbolOperand(MO, GetExternalSymbolSymbol(MO));
       break;
+#endif
     case MachineOperand::MO_JumpTableIndex:
       MCOp = LowerSymbolOperand(MO, GetJumpTableSymbol(MO));
       break;
     case MachineOperand::MO_ConstantPoolIndex:
       MCOp = LowerSymbolOperand(MO, GetConstantPoolIndexSymbol(MO));
       break;
-#endif
     }
     
     OutMI.addOperand(MCOp);

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h?rev=84542&r1=84541&r2=84542&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h Mon Oct 19 16:53:00 2009
@@ -13,6 +13,7 @@
 #include "llvm/Support/Compiler.h"
 
 namespace llvm {
+  class MCAsmInfo;
   class MCContext;
   class MCInst;
   class MCOperand;
@@ -26,24 +27,27 @@
 /// ARMMCInstLower - This class is used to lower an MachineInstr into an MCInst.
 class VISIBILITY_HIDDEN ARMMCInstLower {
   MCContext &Ctx;
-  Mangler *Mang;
+  Mangler &Mang;
+  
+  const unsigned CurFunctionNumber;
+  const MCAsmInfo &MAI;
 
   //const ARMSubtarget &getSubtarget() const;
 public:
-  ARMMCInstLower(MCContext &ctx, Mangler *mang)
-    : Ctx(ctx), Mang(mang) {}
+  ARMMCInstLower(MCContext &ctx, Mangler &mang, unsigned FuncNum,
+                 const MCAsmInfo &mai)
+    : Ctx(ctx), Mang(mang), CurFunctionNumber(FuncNum), MAI(mai) {}
   
   void Lower(const MachineInstr *MI, MCInst &OutMI) const;
 
-/*
- MCSymbol *GetPICBaseSymbol() const;
-  
+  //MCSymbol *GetPICBaseSymbol() const;
   MCSymbol *GetGlobalAddressSymbol(const MachineOperand &MO) const;
-  MCSymbol *GetExternalSymbolSymbol(const MachineOperand &MO) const;
+  //MCSymbol *GetExternalSymbolSymbol(const MachineOperand &MO) const;
   MCSymbol *GetJumpTableSymbol(const MachineOperand &MO) const;
   MCSymbol *GetConstantPoolIndexSymbol(const MachineOperand &MO) const;
   MCOperand LowerSymbolOperand(const MachineOperand &MO, MCSymbol *Sym) const;
   
+/*
 private:
   MachineModuleInfoMachO &getMachOMMI() const;
  */




From sabre at nondot.org  Mon Oct 19 16:57:05 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 21:57:05 -0000
Subject: [llvm-commits] [llvm] r84543 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMInstPrinter.cpp ARMInstPrinter.h
Message-ID: <200910192157.n9JLv5eo016531@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 16:57:05 2009
New Revision: 84543

URL: http://llvm.org/viewvc/llvm-project?rev=84543&view=rev
Log:
add addrmode2 support, getting us up to:

_main:
	stm , 
	mov r7, sp
	sub sp, sp, #4
	mov r0, #0
	str r0, [sp]
	ldr r0, LCPI1_0
	bl _printf
	ldr r0, [sp]
	mov sp, r7
	ldm , 


Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84543&r1=84542&r2=84543&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 16:57:05 2009
@@ -77,3 +77,37 @@
   assert(MO.isImm() && "Not a valid so_imm value!");
   printSOImm(O, MO.getImm(), VerboseAsm, &MAI);
 }
+
+
+
+void ARMInstPrinter::printAddrMode2Operand(const MCInst *MI, unsigned Op) {
+  const MCOperand &MO1 = MI->getOperand(Op);
+  const MCOperand &MO2 = MI->getOperand(Op+1);
+  const MCOperand &MO3 = MI->getOperand(Op+2);
+  
+  if (!MO1.isReg()) {   // FIXME: This is for CP entries, but isn't right.
+    printOperand(MI, Op);
+    return;
+  }
+  
+  O << "[" << getRegisterName(MO1.getReg());
+  
+  if (!MO2.getReg()) {
+    if (ARM_AM::getAM2Offset(MO3.getImm()))  // Don't print +0.
+      O << ", #"
+      << (char)ARM_AM::getAM2Op(MO3.getImm())
+      << ARM_AM::getAM2Offset(MO3.getImm());
+    O << "]";
+    return;
+  }
+  
+  O << ", "
+  << (char)ARM_AM::getAM2Op(MO3.getImm())
+  << getRegisterName(MO2.getReg());
+  
+  if (unsigned ShImm = ARM_AM::getAM2Offset(MO3.getImm()))
+    O << ", "
+    << ARM_AM::getShiftOpcStr(ARM_AM::getAM2ShiftOpc(MO3.getImm()))
+    << " #" << ShImm;
+  O << "]";
+}  

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84543&r1=84542&r2=84543&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 16:57:05 2009
@@ -40,7 +40,7 @@
   
   void printSOImm2PartOperand(const MCInst *MI, unsigned OpNum) {}
   void printSORegOperand(const MCInst *MI, unsigned OpNum) {}
-  void printAddrMode2Operand(const MCInst *MI, unsigned OpNum) {}
+  void printAddrMode2Operand(const MCInst *MI, unsigned OpNum);
   void printAddrMode2OffsetOperand(const MCInst *MI, unsigned OpNum) {}
   void printAddrMode3Operand(const MCInst *MI, unsigned OpNum) {}
   void printAddrMode3OffsetOperand(const MCInst *MI, unsigned OpNum) {}




From sabre at nondot.org  Mon Oct 19 16:59:25 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 21:59:25 -0000
Subject: [llvm-commits] [llvm] r84544 - in
 /llvm/trunk/lib/Target/X86/AsmPrinter: X86MCInstLower.cpp X86MCInstLower.h
Message-ID: <200910192159.n9JLxQM4016721@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 16:59:25 2009
New Revision: 84544

URL: http://llvm.org/viewvc/llvm-project?rev=84544&view=rev
Log:
revert r84540, fixing build breakage I didn't see because of
broken makefile deps :(

Modified:
    llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp
    llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp?rev=84544&r1=84543&r2=84544&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp Mon Oct 19 16:59:25 2009
@@ -59,7 +59,7 @@
     isImplicitlyPrivate = true;
   
   SmallString<128> Name;
-  Mang.getNameWithPrefix(Name, GV, isImplicitlyPrivate);
+  Mang->getNameWithPrefix(Name, GV, isImplicitlyPrivate);
   
   if (getSubtarget().isTargetCygMing()) {
     X86COFFMachineModuleInfo &COFFMMI = 
@@ -86,7 +86,7 @@
     const MCSymbol *&StubSym = getMachOMMI().getGVStubEntry(Sym);
     if (StubSym == 0) {
       Name.clear();
-      Mang.getNameWithPrefix(Name, GV, false);
+      Mang->getNameWithPrefix(Name, GV, false);
       StubSym = Ctx.GetOrCreateSymbol(Name.str());
     }
     return Sym;
@@ -97,7 +97,7 @@
     const MCSymbol *&StubSym = getMachOMMI().getHiddenGVStubEntry(Sym);
     if (StubSym == 0) {
       Name.clear();
-      Mang.getNameWithPrefix(Name, GV, false);
+      Mang->getNameWithPrefix(Name, GV, false);
       StubSym = Ctx.GetOrCreateSymbol(Name.str());
     }
     return Sym;
@@ -108,7 +108,7 @@
     const MCSymbol *&StubSym = getMachOMMI().getFnStubEntry(Sym);
     if (StubSym == 0) {
       Name.clear();
-      Mang.getNameWithPrefix(Name, GV, false);
+      Mang->getNameWithPrefix(Name, GV, false);
       StubSym = Ctx.GetOrCreateSymbol(Name.str());
     }
     return Sym;
@@ -393,7 +393,7 @@
 
 
 void X86AsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI) {
-  X86MCInstLower MCInstLowering(OutContext, *Mang, *this);
+  X86MCInstLower MCInstLowering(OutContext, Mang, *this);
   switch (MI->getOpcode()) {
   case TargetInstrInfo::DBG_LABEL:
   case TargetInstrInfo::EH_LABEL:

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h?rev=84544&r1=84543&r2=84544&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.h Mon Oct 19 16:59:25 2009
@@ -27,12 +27,12 @@
 /// X86MCInstLower - This class is used to lower an MachineInstr into an MCInst.
 class VISIBILITY_HIDDEN X86MCInstLower {
   MCContext &Ctx;
-  Mangler &Mang;
+  Mangler *Mang;
   X86AsmPrinter &AsmPrinter;
 
   const X86Subtarget &getSubtarget() const;
 public:
-  X86MCInstLower(MCContext &ctx, Mangler &mang, X86AsmPrinter &asmprinter)
+  X86MCInstLower(MCContext &ctx, Mangler *mang, X86AsmPrinter &asmprinter)
     : Ctx(ctx), Mang(mang), AsmPrinter(asmprinter) {}
   
   void Lower(const MachineInstr *MI, MCInst &OutMI) const;




From sabre at nondot.org  Mon Oct 19 17:09:23 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 22:09:23 -0000
Subject: [llvm-commits] [llvm] r84546 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMInstPrinter.cpp ARMInstPrinter.h
Message-ID: <200910192209.n9JM9NPE017071@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 17:09:23 2009
New Revision: 84546

URL: http://llvm.org/viewvc/llvm-project?rev=84546&view=rev
Log:
add register list and hacked up addrmode #4 support, we now get this:

_main:
	stmsp! sp!, {r7, lr}
	mov r7, sp
	sub sp, sp, #4
	mov r0, #0
	str r0, [sp]
	ldr r0, LCPI1_0
	bl _printf
	ldr r0, [sp]
	mov sp, r7
	ldmsp! sp!, {r7, pc}

Note the unhappy ldm/stm because of modifiers being ignored.


Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84546&r1=84545&r2=84546&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 17:09:23 2009
@@ -19,6 +19,7 @@
 #include "llvm/MC/MCExpr.h"
 #include "llvm/Support/raw_ostream.h"
 #include "ARMGenInstrNames.inc"
+#include "ARMGenRegisterNames.inc"
 using namespace llvm;
 
 // Include the auto-generated portion of the assembly writer.
@@ -111,3 +112,48 @@
     << " #" << ShImm;
   O << "]";
 }  
+
+
+void ARMInstPrinter::printAddrMode4Operand(const MCInst *MI, unsigned OpNum,
+                                           const char *Modifier) {
+  // FIXME: ENABLE assert.
+  //assert((Modifier == 0 || Modifier[0] == 0) && "Cannot print modifiers");
+  
+  const MCOperand &MO1 = MI->getOperand(OpNum);
+  const MCOperand &MO2 = MI->getOperand(OpNum+1);
+  ARM_AM::AMSubMode Mode = ARM_AM::getAM4SubMode(MO2.getImm());
+  if (0 && Modifier && strcmp(Modifier, "submode") == 0) {
+    if (MO1.getReg() == ARM::SP) {
+      // FIXME
+      bool isLDM = (MI->getOpcode() == ARM::LDM ||
+                    MI->getOpcode() == ARM::LDM_RET ||
+                    MI->getOpcode() == ARM::t2LDM ||
+                    MI->getOpcode() == ARM::t2LDM_RET);
+      O << ARM_AM::getAMSubModeAltStr(Mode, isLDM);
+    } else
+      O << ARM_AM::getAMSubModeStr(Mode);
+  } else if (0 && Modifier && strcmp(Modifier, "wide") == 0) {
+    ARM_AM::AMSubMode Mode = ARM_AM::getAM4SubMode(MO2.getImm());
+    if (Mode == ARM_AM::ia)
+      O << ".w";
+  } else {
+    printOperand(MI, OpNum);
+    if (ARM_AM::getAM4WBFlag(MO2.getImm()))
+      O << "!";
+  }
+}
+
+void ARMInstPrinter::printRegisterList(const MCInst *MI, unsigned OpNum) {
+  O << "{";
+  // Always skip the first operand, it's the optional (and implicit writeback).
+  for (unsigned i = OpNum+1, e = MI->getNumOperands(); i != e; ++i) {
+#if 0 // FIXME: HANDLE WHEN LOWERING??
+    if (MI->getOperand(i).isImplicit())
+      continue;
+#endif
+    if (i != OpNum+1) O << ", ";
+    
+    O << getRegisterName(MI->getOperand(i).getReg());
+  }
+  O << "}";
+}

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84546&r1=84545&r2=84546&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 17:09:23 2009
@@ -45,7 +45,7 @@
   void printAddrMode3Operand(const MCInst *MI, unsigned OpNum) {}
   void printAddrMode3OffsetOperand(const MCInst *MI, unsigned OpNum) {}
   void printAddrMode4Operand(const MCInst *MI, unsigned OpNum,
-                             const char *Modifier = 0) {}
+                             const char *Modifier = 0);
   void printAddrMode5Operand(const MCInst *MI, unsigned OpNum,
                              const char *Modifier = 0) {}
   void printAddrMode6Operand(const MCInst *MI, unsigned OpNum) {}
@@ -72,7 +72,7 @@
   void printPredicateOperand(const MCInst *MI, unsigned OpNum) {}
   void printSBitModifierOperand(const MCInst *MI, unsigned OpNum) {}
   void printPCLabel(const MCInst *MI, unsigned OpNum) {}
-  void printRegisterList(const MCInst *MI, unsigned OpNum) {}
+  void printRegisterList(const MCInst *MI, unsigned OpNum);
   void printCPInstOperand(const MCInst *MI, unsigned OpNum,
                           const char *Modifier) {}
   void printJTBlockOperand(const MCInst *MI, unsigned OpNum) {}




From resistor at mac.com  Mon Oct 19 17:14:22 2009
From: resistor at mac.com (Owen Anderson)
Date: Mon, 19 Oct 2009 22:14:22 -0000
Subject: [llvm-commits] [llvm] r84547 -
	/llvm/trunk/lib/Transforms/Scalar/GVN.cpp
Message-ID: <200910192214.n9JMEM6r017222@zion.cs.uiuc.edu>

Author: resistor
Date: Mon Oct 19 17:14:22 2009
New Revision: 84547

URL: http://llvm.org/viewvc/llvm-project?rev=84547&view=rev
Log:
Refactor lookup_or_add to contain _MUCH_ less duplicated code.  Add support for
numbering first class aggregate instructions while we're at it.

Modified:
    llvm/trunk/lib/Transforms/Scalar/GVN.cpp

Modified: llvm/trunk/lib/Transforms/Scalar/GVN.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/GVN.cpp?rev=84547&r1=84546&r2=84547&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/GVN.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/GVN.cpp Mon Oct 19 17:14:22 2009
@@ -78,13 +78,10 @@
                             SHUFFLE, SELECT, TRUNC, ZEXT, SEXT, FPTOUI,
                             FPTOSI, UITOFP, SITOFP, FPTRUNC, FPEXT,
                             PTRTOINT, INTTOPTR, BITCAST, GEP, CALL, CONSTANT,
-                            EMPTY, TOMBSTONE };
+                            INSERTVALUE, EXTRACTVALUE, EMPTY, TOMBSTONE };
 
     ExpressionOpcode opcode;
     const Type* type;
-    uint32_t firstVN;
-    uint32_t secondVN;
-    uint32_t thirdVN;
     SmallVector<uint32_t, 4> varargs;
     Value *function;
 
@@ -100,12 +97,6 @@
         return false;
       else if (function != other.function)
         return false;
-      else if (firstVN != other.firstVN)
-        return false;
-      else if (secondVN != other.secondVN)
-        return false;
-      else if (thirdVN != other.thirdVN)
-        return false;
       else {
         if (varargs.size() != other.varargs.size())
           return false;
@@ -146,6 +137,10 @@
       Expression create_expression(GetElementPtrInst* G);
       Expression create_expression(CallInst* C);
       Expression create_expression(Constant* C);
+      Expression create_expression(ExtractValueInst* C);
+      Expression create_expression(InsertValueInst* C);
+      
+      uint32_t lookup_or_add_call(CallInst* C);
     public:
       ValueTable() : nextValueNumber(1) { }
       uint32_t lookup_or_add(Value *V);
@@ -176,13 +171,8 @@
   static unsigned getHashValue(const Expression e) {
     unsigned hash = e.opcode;
 
-    hash = e.firstVN + hash * 37;
-    hash = e.secondVN + hash * 37;
-    hash = e.thirdVN + hash * 37;
-
     hash = ((unsigned)((uintptr_t)e.type >> 4) ^
-            (unsigned)((uintptr_t)e.type >> 9)) +
-           hash * 37;
+            (unsigned)((uintptr_t)e.type >> 9));
 
     for (SmallVector<uint32_t, 4>::const_iterator I = e.varargs.begin(),
          E = e.varargs.end(); I != E; ++I)
@@ -290,9 +280,6 @@
   Expression e;
 
   e.type = C->getType();
-  e.firstVN = 0;
-  e.secondVN = 0;
-  e.thirdVN = 0;
   e.function = C->getCalledFunction();
   e.opcode = Expression::CALL;
 
@@ -305,10 +292,8 @@
 
 Expression ValueTable::create_expression(BinaryOperator* BO) {
   Expression e;
-
-  e.firstVN = lookup_or_add(BO->getOperand(0));
-  e.secondVN = lookup_or_add(BO->getOperand(1));
-  e.thirdVN = 0;
+  e.varargs.push_back(lookup_or_add(BO->getOperand(0)));
+  e.varargs.push_back(lookup_or_add(BO->getOperand(1)));
   e.function = 0;
   e.type = BO->getType();
   e.opcode = getOpcode(BO);
@@ -319,9 +304,8 @@
 Expression ValueTable::create_expression(CmpInst* C) {
   Expression e;
 
-  e.firstVN = lookup_or_add(C->getOperand(0));
-  e.secondVN = lookup_or_add(C->getOperand(1));
-  e.thirdVN = 0;
+  e.varargs.push_back(lookup_or_add(C->getOperand(0)));
+  e.varargs.push_back(lookup_or_add(C->getOperand(1)));
   e.function = 0;
   e.type = C->getType();
   e.opcode = getOpcode(C);
@@ -332,9 +316,7 @@
 Expression ValueTable::create_expression(CastInst* C) {
   Expression e;
 
-  e.firstVN = lookup_or_add(C->getOperand(0));
-  e.secondVN = 0;
-  e.thirdVN = 0;
+  e.varargs.push_back(lookup_or_add(C->getOperand(0)));
   e.function = 0;
   e.type = C->getType();
   e.opcode = getOpcode(C);
@@ -345,9 +327,9 @@
 Expression ValueTable::create_expression(ShuffleVectorInst* S) {
   Expression e;
 
-  e.firstVN = lookup_or_add(S->getOperand(0));
-  e.secondVN = lookup_or_add(S->getOperand(1));
-  e.thirdVN = lookup_or_add(S->getOperand(2));
+  e.varargs.push_back(lookup_or_add(S->getOperand(0)));
+  e.varargs.push_back(lookup_or_add(S->getOperand(1)));
+  e.varargs.push_back(lookup_or_add(S->getOperand(2)));
   e.function = 0;
   e.type = S->getType();
   e.opcode = Expression::SHUFFLE;
@@ -358,9 +340,8 @@
 Expression ValueTable::create_expression(ExtractElementInst* E) {
   Expression e;
 
-  e.firstVN = lookup_or_add(E->getOperand(0));
-  e.secondVN = lookup_or_add(E->getOperand(1));
-  e.thirdVN = 0;
+  e.varargs.push_back(lookup_or_add(E->getOperand(0)));
+  e.varargs.push_back(lookup_or_add(E->getOperand(1)));
   e.function = 0;
   e.type = E->getType();
   e.opcode = Expression::EXTRACT;
@@ -371,9 +352,9 @@
 Expression ValueTable::create_expression(InsertElementInst* I) {
   Expression e;
 
-  e.firstVN = lookup_or_add(I->getOperand(0));
-  e.secondVN = lookup_or_add(I->getOperand(1));
-  e.thirdVN = lookup_or_add(I->getOperand(2));
+  e.varargs.push_back(lookup_or_add(I->getOperand(0)));
+  e.varargs.push_back(lookup_or_add(I->getOperand(1)));
+  e.varargs.push_back(lookup_or_add(I->getOperand(2)));
   e.function = 0;
   e.type = I->getType();
   e.opcode = Expression::INSERT;
@@ -384,9 +365,9 @@
 Expression ValueTable::create_expression(SelectInst* I) {
   Expression e;
 
-  e.firstVN = lookup_or_add(I->getCondition());
-  e.secondVN = lookup_or_add(I->getTrueValue());
-  e.thirdVN = lookup_or_add(I->getFalseValue());
+  e.varargs.push_back(lookup_or_add(I->getCondition()));
+  e.varargs.push_back(lookup_or_add(I->getTrueValue()));
+  e.varargs.push_back(lookup_or_add(I->getFalseValue()));
   e.function = 0;
   e.type = I->getType();
   e.opcode = Expression::SELECT;
@@ -397,9 +378,7 @@
 Expression ValueTable::create_expression(GetElementPtrInst* G) {
   Expression e;
 
-  e.firstVN = lookup_or_add(G->getPointerOperand());
-  e.secondVN = 0;
-  e.thirdVN = 0;
+  e.varargs.push_back(lookup_or_add(G->getPointerOperand()));
   e.function = 0;
   e.type = G->getType();
   e.opcode = Expression::GEP;
@@ -411,6 +390,35 @@
   return e;
 }
 
+Expression ValueTable::create_expression(ExtractValueInst* E) {
+  Expression e;
+
+  e.varargs.push_back(lookup_or_add(E->getAggregateOperand()));
+  for (ExtractValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end();
+       II != IE; ++II)
+    e.varargs.push_back(*II);
+  e.function = 0;
+  e.type = E->getType();
+  e.opcode = Expression::EXTRACTVALUE;
+
+  return e;
+}
+
+Expression ValueTable::create_expression(InsertValueInst* E) {
+  Expression e;
+
+  e.varargs.push_back(lookup_or_add(E->getAggregateOperand()));
+  e.varargs.push_back(lookup_or_add(E->getInsertedValueOperand()));
+  for (InsertValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end();
+       II != IE; ++II)
+    e.varargs.push_back(*II);
+  e.function = 0;
+  e.type = E->getType();
+  e.opcode = Expression::INSERTVALUE;
+
+  return e;
+}
+
 //===----------------------------------------------------------------------===//
 //                     ValueTable External Functions
 //===----------------------------------------------------------------------===//
@@ -420,169 +428,197 @@
   valueNumbering.insert(std::make_pair(V, num));
 }
 
-/// lookup_or_add - Returns the value number for the specified value, assigning
-/// it a new number if it did not have one before.
-uint32_t ValueTable::lookup_or_add(Value *V) {
-  DenseMap<Value*, uint32_t>::iterator VI = valueNumbering.find(V);
-  if (VI != valueNumbering.end())
-    return VI->second;
-
-  if (CallInst* C = dyn_cast<CallInst>(V)) {
-    if (AA->doesNotAccessMemory(C)) {
-      Expression exp = create_expression(C);
-      uint32_t& e = expressionNumbering[exp];
-      if (!e) e = nextValueNumber++;
-      valueNumbering[V] = e;
+uint32_t ValueTable::lookup_or_add_call(CallInst* C) {
+  if (AA->doesNotAccessMemory(C)) {
+    Expression exp = create_expression(C);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) e = nextValueNumber++;
+    valueNumbering[C] = e;
+    return e;
+  } else if (AA->onlyReadsMemory(C)) {
+    Expression exp = create_expression(C);
+    uint32_t& e = expressionNumbering[exp];
+    if (!e) {
+      e = nextValueNumber++;
+      valueNumbering[C] = e;
       return e;
-    } else if (AA->onlyReadsMemory(C)) {
-      Expression exp = create_expression(C);
-      uint32_t& e = expressionNumbering[exp];
-      if (!e) {
-        e = nextValueNumber++;
-        valueNumbering[V] = e;
-        return e;
-      }
+    }
 
-      MemDepResult local_dep = MD->getDependency(C);
+    MemDepResult local_dep = MD->getDependency(C);
 
-      if (!local_dep.isDef() && !local_dep.isNonLocal()) {
-        valueNumbering[V] =  nextValueNumber;
+    if (!local_dep.isDef() && !local_dep.isNonLocal()) {
+      valueNumbering[C] =  nextValueNumber;
+      return nextValueNumber++;
+    }
+
+    if (local_dep.isDef()) {
+      CallInst* local_cdep = cast<CallInst>(local_dep.getInst());
+
+      if (local_cdep->getNumOperands() != C->getNumOperands()) {
+        valueNumbering[C] = nextValueNumber;
         return nextValueNumber++;
       }
 
-      if (local_dep.isDef()) {
-        CallInst* local_cdep = cast<CallInst>(local_dep.getInst());
-
-        if (local_cdep->getNumOperands() != C->getNumOperands()) {
-          valueNumbering[V] = nextValueNumber;
+      for (unsigned i = 1; i < C->getNumOperands(); ++i) {
+        uint32_t c_vn = lookup_or_add(C->getOperand(i));
+        uint32_t cd_vn = lookup_or_add(local_cdep->getOperand(i));
+        if (c_vn != cd_vn) {
+          valueNumbering[C] = nextValueNumber;
           return nextValueNumber++;
         }
-
-        for (unsigned i = 1; i < C->getNumOperands(); ++i) {
-          uint32_t c_vn = lookup_or_add(C->getOperand(i));
-          uint32_t cd_vn = lookup_or_add(local_cdep->getOperand(i));
-          if (c_vn != cd_vn) {
-            valueNumbering[V] = nextValueNumber;
-            return nextValueNumber++;
-          }
-        }
-
-        uint32_t v = lookup_or_add(local_cdep);
-        valueNumbering[V] = v;
-        return v;
       }
 
-      // Non-local case.
-      const MemoryDependenceAnalysis::NonLocalDepInfo &deps =
-        MD->getNonLocalCallDependency(CallSite(C));
-      // FIXME: call/call dependencies for readonly calls should return def, not
-      // clobber!  Move the checking logic to MemDep!
-      CallInst* cdep = 0;
-
-      // Check to see if we have a single dominating call instruction that is
-      // identical to C.
-      for (unsigned i = 0, e = deps.size(); i != e; ++i) {
-        const MemoryDependenceAnalysis::NonLocalDepEntry *I = &deps[i];
-        // Ignore non-local dependencies.
-        if (I->second.isNonLocal())
-          continue;
-
-        // We don't handle non-depedencies.  If we already have a call, reject
-        // instruction dependencies.
-        if (I->second.isClobber() || cdep != 0) {
-          cdep = 0;
-          break;
-        }
+      uint32_t v = lookup_or_add(local_cdep);
+      valueNumbering[C] = v;
+      return v;
+    }
 
-        CallInst *NonLocalDepCall = dyn_cast<CallInst>(I->second.getInst());
-        // FIXME: All duplicated with non-local case.
-        if (NonLocalDepCall && DT->properlyDominates(I->first, C->getParent())){
-          cdep = NonLocalDepCall;
-          continue;
-        }
+    // Non-local case.
+    const MemoryDependenceAnalysis::NonLocalDepInfo &deps =
+      MD->getNonLocalCallDependency(CallSite(C));
+    // FIXME: call/call dependencies for readonly calls should return def, not
+    // clobber!  Move the checking logic to MemDep!
+    CallInst* cdep = 0;
+
+    // Check to see if we have a single dominating call instruction that is
+    // identical to C.
+    for (unsigned i = 0, e = deps.size(); i != e; ++i) {
+      const MemoryDependenceAnalysis::NonLocalDepEntry *I = &deps[i];
+      // Ignore non-local dependencies.
+      if (I->second.isNonLocal())
+        continue;
 
+      // We don't handle non-depedencies.  If we already have a call, reject
+      // instruction dependencies.
+      if (I->second.isClobber() || cdep != 0) {
         cdep = 0;
         break;
       }
 
-      if (!cdep) {
-        valueNumbering[V] = nextValueNumber;
-        return nextValueNumber++;
+      CallInst *NonLocalDepCall = dyn_cast<CallInst>(I->second.getInst());
+      // FIXME: All duplicated with non-local case.
+      if (NonLocalDepCall && DT->properlyDominates(I->first, C->getParent())){
+        cdep = NonLocalDepCall;
+        continue;
       }
 
-      if (cdep->getNumOperands() != C->getNumOperands()) {
-        valueNumbering[V] = nextValueNumber;
-        return nextValueNumber++;
-      }
-      for (unsigned i = 1; i < C->getNumOperands(); ++i) {
-        uint32_t c_vn = lookup_or_add(C->getOperand(i));
-        uint32_t cd_vn = lookup_or_add(cdep->getOperand(i));
-        if (c_vn != cd_vn) {
-          valueNumbering[V] = nextValueNumber;
-          return nextValueNumber++;
-        }
-      }
+      cdep = 0;
+      break;
+    }
 
-      uint32_t v = lookup_or_add(cdep);
-      valueNumbering[V] = v;
-      return v;
+    if (!cdep) {
+      valueNumbering[C] = nextValueNumber;
+      return nextValueNumber++;
+    }
 
-    } else {
-      valueNumbering[V] = nextValueNumber;
+    if (cdep->getNumOperands() != C->getNumOperands()) {
+      valueNumbering[C] = nextValueNumber;
       return nextValueNumber++;
     }
-  } else if (BinaryOperator* BO = dyn_cast<BinaryOperator>(V)) {
-    Expression exp = create_expression(BO);
-    uint32_t& e = expressionNumbering[exp];
-    if (!e) e = nextValueNumber++;
-    valueNumbering[V] = e;
-    return e;
-  } else if (CmpInst* C = dyn_cast<CmpInst>(V)) {
-    Expression exp = create_expression(C);
-    uint32_t& e = expressionNumbering[exp];
-    if (!e) e = nextValueNumber++;
-    valueNumbering[V] = e;
-    return e;
-  } else if (ShuffleVectorInst* U = dyn_cast<ShuffleVectorInst>(V)) {
-    Expression exp = create_expression(U);
-    uint32_t& e = expressionNumbering[exp];
-    if (!e) e = nextValueNumber++;
-    valueNumbering[V] = e;
-    return e;
-  } else if (ExtractElementInst* U = dyn_cast<ExtractElementInst>(V)) {
-    Expression exp = create_expression(U);
-    uint32_t& e = expressionNumbering[exp];
-    if (!e) e = nextValueNumber++;
-    valueNumbering[V] = e;
-    return e;
-  } else if (InsertElementInst* U = dyn_cast<InsertElementInst>(V)) {
-    Expression exp = create_expression(U);
-    uint32_t& e = expressionNumbering[exp];
-    if (!e) e = nextValueNumber++;
-    valueNumbering[V] = e;
-    return e;
-  } else if (SelectInst* U = dyn_cast<SelectInst>(V)) {
-    Expression exp = create_expression(U);
-    uint32_t& e = expressionNumbering[exp];
-    if (!e) e = nextValueNumber++;
-    valueNumbering[V] = e;
-    return e;
-  } else if (CastInst* U = dyn_cast<CastInst>(V)) {
-    Expression exp = create_expression(U);
-    uint32_t& e = expressionNumbering[exp];
-    if (!e) e = nextValueNumber++;
-    valueNumbering[V] = e;
-    return e;
-  } else if (GetElementPtrInst* U = dyn_cast<GetElementPtrInst>(V)) {
-    Expression exp = create_expression(U);
-    uint32_t& e = expressionNumbering[exp];
-    if (!e) e = nextValueNumber++;
-    valueNumbering[V] = e;
-    return e;
+    for (unsigned i = 1; i < C->getNumOperands(); ++i) {
+      uint32_t c_vn = lookup_or_add(C->getOperand(i));
+      uint32_t cd_vn = lookup_or_add(cdep->getOperand(i));
+      if (c_vn != cd_vn) {
+        valueNumbering[C] = nextValueNumber;
+        return nextValueNumber++;
+      }
+    }
+
+    uint32_t v = lookup_or_add(cdep);
+    valueNumbering[C] = v;
+    return v;
+
   } else {
+    valueNumbering[C] = nextValueNumber;
+    return nextValueNumber++;
+  }
+}
+
+/// lookup_or_add - Returns the value number for the specified value, assigning
+/// it a new number if it did not have one before.
+uint32_t ValueTable::lookup_or_add(Value *V) {
+  DenseMap<Value*, uint32_t>::iterator VI = valueNumbering.find(V);
+  if (VI != valueNumbering.end())
+    return VI->second;
+
+  if (!isa<Instruction>(V)) {
     valueNumbering[V] = nextValueNumber;
     return nextValueNumber++;
   }
+  
+  Instruction* I = cast<Instruction>(V);
+  Expression exp;
+  switch (I->getOpcode()) {
+    case Instruction::Call:
+      return lookup_or_add_call(cast<CallInst>(I));
+    case Instruction::Add:
+    case Instruction::FAdd:
+    case Instruction::Sub:
+    case Instruction::FSub:
+    case Instruction::Mul:
+    case Instruction::FMul:
+    case Instruction::UDiv:
+    case Instruction::SDiv:
+    case Instruction::FDiv:
+    case Instruction::URem:
+    case Instruction::SRem:
+    case Instruction::FRem:
+    case Instruction::Shl:
+    case Instruction::LShr:
+    case Instruction::AShr:
+    case Instruction::And:
+    case Instruction::Or :
+    case Instruction::Xor:
+      exp = create_expression(cast<BinaryOperator>(I));
+      break;
+    case Instruction::ICmp:
+    case Instruction::FCmp:
+      exp = create_expression(cast<CmpInst>(I));
+      break;
+    case Instruction::Trunc:
+    case Instruction::ZExt:
+    case Instruction::SExt:
+    case Instruction::FPToUI:
+    case Instruction::FPToSI:
+    case Instruction::UIToFP:
+    case Instruction::SIToFP:
+    case Instruction::FPTrunc:
+    case Instruction::FPExt:
+    case Instruction::PtrToInt:
+    case Instruction::IntToPtr:
+    case Instruction::BitCast:
+      exp = create_expression(cast<CastInst>(I));
+      break;
+    case Instruction::Select:
+      exp = create_expression(cast<SelectInst>(I));
+      break;
+    case Instruction::ExtractElement:
+      exp = create_expression(cast<ExtractElementInst>(I));
+      break;
+    case Instruction::InsertElement:
+      exp = create_expression(cast<InsertElementInst>(I));
+      break;
+    case Instruction::ShuffleVector:
+      exp = create_expression(cast<ShuffleVectorInst>(I));
+      break;
+    case Instruction::ExtractValue:
+      exp = create_expression(cast<ExtractValueInst>(I));
+      break;
+    case Instruction::InsertValue:
+      exp = create_expression(cast<InsertValueInst>(I));
+      break;      
+    case Instruction::GetElementPtr:
+      exp = create_expression(cast<GetElementPtrInst>(I));
+      break;
+    default:
+      valueNumbering[V] = nextValueNumber;
+      return nextValueNumber++;
+  }
+
+  uint32_t& e = expressionNumbering[exp];
+  if (!e) e = nextValueNumber++;
+  valueNumbering[V] = e;
+  return e;
 }
 
 /// lookup - Returns the value number of the specified value. Fails if




From sabre at nondot.org  Mon Oct 19 17:23:05 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 22:23:05 -0000
Subject: [llvm-commits] [llvm] r84553 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMAsmPrinter.cpp ARMInstPrinter.cpp
 ARMInstPrinter.h
Message-ID: <200910192223.n9JMN5SW017690@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 17:23:04 2009
New Revision: 84553

URL: http://llvm.org/viewvc/llvm-project?rev=84553&view=rev
Log:
add MCInstLower support for lowering ARM::PICADD, a pseudo op for pic stuffola.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84553&r1=84552&r2=84553&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 17:23:04 2009
@@ -1313,6 +1313,7 @@
 void ARMAsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI) {
   ARMMCInstLower MCInstLowering(OutContext, *Mang, getFunctionNumber(), *MAI);
   switch (MI->getOpcode()) {
+  default: break;
   case TargetInstrInfo::DBG_LABEL:
   case TargetInstrInfo::EH_LABEL:
   case TargetInstrInfo::GC_LABEL:
@@ -1327,7 +1328,29 @@
   case TargetInstrInfo::IMPLICIT_DEF:
     printImplicitDef(MI);
     return;
-  default: break;
+  case ARM::PICADD: { // FIXME: Remove asm string from td file.
+    // This is a pseudo op for a label + instruction sequence, which looks like:
+    // LPC0:
+    //     add r0, pc, r0
+    // This adds the address of LPC0 to r0.
+    
+    // Emit the label.
+    // FIXME: MOVE TO SHARED PLACE.
+    SmallString<60> Name;
+    int Id = (int)MI->getOperand(2).getImm();
+    raw_svector_ostream(Name) << MAI->getPrivateGlobalPrefix() << "PC" << Id;
+    OutStreamer.EmitLabel(OutContext.GetOrCreateSymbol(Name.str()));
+    
+    
+    // Form and emit tha dd.
+    MCInst AddInst;
+    AddInst.setOpcode(ARM::ADDrr);
+    AddInst.addOperand(MCOperand::CreateReg(MI->getOperand(0).getReg()));
+    AddInst.addOperand(MCOperand::CreateReg(ARM::PC));
+    AddInst.addOperand(MCOperand::CreateReg(MI->getOperand(1).getReg()));
+    printMCInst(&AddInst);
+    return;
+  }
   }
       
   MCInst TmpInst;

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84553&r1=84552&r2=84553&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 17:23:04 2009
@@ -157,3 +157,10 @@
   }
   O << "}";
 }
+
+
+
+void ARMInstPrinter::printPCLabel(const MCInst *MI, unsigned OpNum) {
+  // FIXME: remove this.
+  abort();
+}

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84553&r1=84552&r2=84553&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 17:23:04 2009
@@ -71,7 +71,6 @@
   
   void printPredicateOperand(const MCInst *MI, unsigned OpNum) {}
   void printSBitModifierOperand(const MCInst *MI, unsigned OpNum) {}
-  void printPCLabel(const MCInst *MI, unsigned OpNum) {}
   void printRegisterList(const MCInst *MI, unsigned OpNum);
   void printCPInstOperand(const MCInst *MI, unsigned OpNum,
                           const char *Modifier) {}
@@ -79,9 +78,9 @@
   void printJT2BlockOperand(const MCInst *MI, unsigned OpNum) {}
   void printTBAddrMode(const MCInst *MI, unsigned OpNum) {}
   void printNoHashImmediate(const MCInst *MI, unsigned OpNum) {}
-  
-  
-  // FIXME:
+
+  void printPCLabel(const MCInst *MI, unsigned OpNum);  
+  // FIXME: Implement.
   void PrintSpecial(const MCInst *MI, const char *Kind) {}
 };
   




From grosbach at apple.com  Mon Oct 19 17:27:30 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Mon, 19 Oct 2009 22:27:30 -0000
Subject: [llvm-commits] [llvm] r84554 - in /llvm/trunk:
 include/llvm/Target/TargetRegisterInfo.h lib/CodeGen/RegisterScavenging.cpp
 lib/Target/ARM/Thumb1RegisterInfo.cpp lib/Target/ARM/Thumb1RegisterInfo.h
Message-ID: <200910192227.n9JMRUJ6017834@zion.cs.uiuc.edu>

Author: grosbach
Date: Mon Oct 19 17:27:30 2009
New Revision: 84554

URL: http://llvm.org/viewvc/llvm-project?rev=84554&view=rev
Log:
Adjust the scavenge register spilling to allow the target to choose an
appropriate restore location for the spill as well as perform the actual
save and restore.

The Thumb1 target uses this to make sure R12 is not clobbered while a spilled
scavenger register is live there.


Modified:
    llvm/trunk/include/llvm/Target/TargetRegisterInfo.h
    llvm/trunk/lib/CodeGen/RegisterScavenging.cpp
    llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
    llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h

Modified: llvm/trunk/include/llvm/Target/TargetRegisterInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Target/TargetRegisterInfo.h?rev=84554&r1=84553&r2=84554&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Target/TargetRegisterInfo.h (original)
+++ llvm/trunk/include/llvm/Target/TargetRegisterInfo.h Mon Oct 19 17:27:30 2009
@@ -641,24 +641,17 @@
   virtual void processFunctionBeforeFrameFinalized(MachineFunction &MF) const {
   }
 
-  /// saveScavengerRegister - Save the register so it can be used by the
-  /// register scavenger. Return true if the register was saved, false
-  /// otherwise. If this function does not save the register, the scavenger
+  /// saveScavengerRegister - Spill the register so it can be used by the
+  /// register scavenger. Return true if the register was spilled, false
+  /// otherwise. If this function does not spill the register, the scavenger
   /// will instead spill it to the emergency spill slot.
   ///
   virtual bool saveScavengerRegister(MachineBasicBlock &MBB,
                                      MachineBasicBlock::iterator I,
+                                     MachineBasicBlock::iterator &UseMI,
                                      const TargetRegisterClass *RC,
                                      unsigned Reg) const {return false;}
 
-  /// restoreScavengerRegister - Restore a register saved by
-  /// saveScavengerRegister().
-  ///
-  virtual void restoreScavengerRegister(MachineBasicBlock &MBB,
-                                        MachineBasicBlock::iterator I,
-                                        const TargetRegisterClass *RC,
-                                        unsigned Reg) const {}
-
   /// eliminateFrameIndex - This method must be overriden to eliminate abstract
   /// frame indices from instructions which may use them.  The instruction
   /// referenced by the iterator contains an MO_FrameIndex operand which must be

Modified: llvm/trunk/lib/CodeGen/RegisterScavenging.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/RegisterScavenging.cpp?rev=84554&r1=84553&r2=84554&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/RegisterScavenging.cpp (original)
+++ llvm/trunk/lib/CodeGen/RegisterScavenging.cpp Mon Oct 19 17:27:30 2009
@@ -300,7 +300,7 @@
 
   // If the target knows how to save/restore the register, let it do so;
   // otherwise, use the emergency stack spill slot.
-  if (!TRI->saveScavengerRegister(*MBB, I, RC, SReg)) {
+  if (!TRI->saveScavengerRegister(*MBB, I, UseMI, RC, SReg)) {
     // Spill the scavenged register before I.
     assert(ScavengingFrameIndex >= 0 &&
            "Cannot scavenge register without an emergency spill slot!");
@@ -310,8 +310,7 @@
 
     // Restore the scavenged register before its use (or first terminator).
     TII->loadRegFromStackSlot(*MBB, UseMI, SReg, ScavengingFrameIndex, RC);
-  } else
-    TRI->restoreScavengerRegister(*MBB, UseMI, RC, SReg);
+  }
 
   ScavengeRestore = prior(UseMI);
 

Modified: llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp?rev=84554&r1=84553&r2=84554&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp Mon Oct 19 17:27:30 2009
@@ -394,31 +394,48 @@
   return 0;
 }
 
-/// saveScavengerRegister - Save the register so it can be used by the
+/// saveScavengerRegister - Spill the register so it can be used by the
 /// register scavenger. Return true.
-bool Thumb1RegisterInfo::saveScavengerRegister(MachineBasicBlock &MBB,
-                                               MachineBasicBlock::iterator I,
-                                               const TargetRegisterClass *RC,
-                                               unsigned Reg) const {
+bool
+Thumb1RegisterInfo::saveScavengerRegister(MachineBasicBlock &MBB,
+                                          MachineBasicBlock::iterator I,
+                                          MachineBasicBlock::iterator &UseMI,
+                                          const TargetRegisterClass *RC,
+                                          unsigned Reg) const {
   // Thumb1 can't use the emergency spill slot on the stack because
   // ldr/str immediate offsets must be positive, and if we're referencing
   // off the frame pointer (if, for example, there are alloca() calls in
   // the function, the offset will be negative. Use R12 instead since that's
   // a call clobbered register that we know won't be used in Thumb1 mode.
+  DebugLoc DL = DebugLoc::getUnknownLoc();
+  BuildMI(MBB, I, DL, TII.get(ARM::tMOVtgpr2gpr)).
+    addReg(ARM::R12, RegState::Define).addReg(Reg, RegState::Kill);
+
+  // The UseMI is where we would like to restore the register. If there's
+  // interference with R12 before then, however, we'll need to restore it
+  // before that instead and adjust the UseMI.
+  bool done = false;
+  for (MachineBasicBlock::iterator II = I; !done && II != UseMI ; ++II) {
+    // If this instruction affects R12, adjust our restore point.
+    for (unsigned i = 0, e = II->getNumOperands(); i != e; ++i) {
+      const MachineOperand &MO = II->getOperand(i);
+      if (!MO.isReg() || MO.isUndef() || !MO.getReg() ||
+          TargetRegisterInfo::isVirtualRegister(MO.getReg()))
+        continue;
+      if (MO.getReg() == ARM::R12) {
+        UseMI = II;
+        done = true;
+        break;
+      }
+    }
+  }
+  // Restore the register from R12
+  BuildMI(MBB, UseMI, DL, TII.get(ARM::tMOVgpr2tgpr)).
+    addReg(Reg, RegState::Define).addReg(ARM::R12, RegState::Kill);
 
-  TII.copyRegToReg(MBB, I, ARM::R12, Reg, ARM::GPRRegisterClass, RC);
   return true;
 }
 
-/// restoreScavengerRegister - restore a registers saved by
-// saveScavengerRegister().
-void Thumb1RegisterInfo::restoreScavengerRegister(MachineBasicBlock &MBB,
-                                               MachineBasicBlock::iterator I,
-                                               const TargetRegisterClass *RC,
-                                               unsigned Reg) const {
-  TII.copyRegToReg(MBB, I, Reg, ARM::R12, RC, ARM::GPRRegisterClass);
-}
-
 unsigned
 Thumb1RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
                                         int SPAdj, int *Value,

Modified: llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h?rev=84554&r1=84553&r2=84554&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h (original)
+++ llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h Mon Oct 19 17:27:30 2009
@@ -57,12 +57,9 @@
 
   bool saveScavengerRegister(MachineBasicBlock &MBB,
                              MachineBasicBlock::iterator I,
+                             MachineBasicBlock::iterator &UseMI,
                              const TargetRegisterClass *RC,
                              unsigned Reg) const;
-  void restoreScavengerRegister(MachineBasicBlock &MBB,
-                                MachineBasicBlock::iterator I,
-                                const TargetRegisterClass *RC,
-                                unsigned Reg) const;
   unsigned eliminateFrameIndex(MachineBasicBlock::iterator II,
                                int SPAdj, int *Value = NULL,
                                RegScavenger *RS = NULL) const;




From sabre at nondot.org  Mon Oct 19 17:33:06 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 22:33:06 -0000
Subject: [llvm-commits] [llvm] r84558 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMAsmPrinter.cpp ARMInstPrinter.cpp
 ARMInstPrinter.h
Message-ID: <200910192233.n9JMX6F2018151@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 17:33:05 2009
New Revision: 84558

URL: http://llvm.org/viewvc/llvm-project?rev=84558&view=rev
Log:
lower the ARM::CONSTPOOL_ENTRY pseudo op, giving us constant pool entries
like:

@ BB#1:
	.align	2
LCPI1_0:
	.long	L_.str-(LPC0+8)

Note that proper indentation of the label :)


Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84558&r1=84557&r2=84558&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 17:33:05 2009
@@ -1029,16 +1029,16 @@
 void ARMAsmPrinter::printMachineInstruction(const MachineInstr *MI) {
   ++EmittedInsts;
 
-  int Opc = MI->getOpcode();
-  if (Opc == ARM::CONSTPOOL_ENTRY)
-    EmitAlignment(2);
-
   // Call the autogenerated instruction printer routines.
   processDebugLoc(MI, true);
   
   if (EnableMCInst) {
     printInstructionThroughMCStreamer(MI);
   } else {
+    int Opc = MI->getOpcode();
+    if (Opc == ARM::CONSTPOOL_ENTRY)
+      EmitAlignment(2);
+    
     printInstruction(MI);
   }
   
@@ -1337,7 +1337,7 @@
     // Emit the label.
     // FIXME: MOVE TO SHARED PLACE.
     SmallString<60> Name;
-    int Id = (int)MI->getOperand(2).getImm();
+    unsigned Id = (unsigned)MI->getOperand(2).getImm();
     raw_svector_ostream(Name) << MAI->getPrivateGlobalPrefix() << "PC" << Id;
     OutStreamer.EmitLabel(OutContext.GetOrCreateSymbol(Name.str()));
     
@@ -1351,6 +1351,27 @@
     printMCInst(&AddInst);
     return;
   }
+  case ARM::CONSTPOOL_ENTRY: { // FIXME: Remove asm string from td file.
+    /// CONSTPOOL_ENTRY - This instruction represents a floating constant pool
+    /// in the function.  The first operand is the ID# for this instruction, the
+    /// second is the index into the MachineConstantPool that this is, the third
+    /// is the size in bytes of this constant pool entry.
+    unsigned LabelId = (unsigned)MI->getOperand(0).getImm();
+    unsigned CPIdx   = (unsigned)MI->getOperand(1).getIndex();
+
+    EmitAlignment(2);
+
+    O << MAI->getPrivateGlobalPrefix() << "CPI" << getFunctionNumber()
+      << '_' << LabelId << ":\n";
+
+    const MachineConstantPoolEntry &MCPE = MCP->getConstants()[CPIdx];
+    if (MCPE.isMachineConstantPoolEntry())
+      EmitMachineConstantPoolValue(MCPE.Val.MachineCPVal);
+    else
+      EmitGlobalConstant(MCPE.Val.ConstVal);
+    
+    return;
+  }
   }
       
   MCInst TmpInst;

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84558&r1=84557&r2=84558&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 17:33:05 2009
@@ -159,6 +159,11 @@
 }
 
 
+void ARMInstPrinter::printCPInstOperand(const MCInst *MI, unsigned OpNum,
+                                        const char *Modifier) {
+  // FIXME: remove this.
+  abort();
+}
 
 void ARMInstPrinter::printPCLabel(const MCInst *MI, unsigned OpNum) {
   // FIXME: remove this.

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84558&r1=84557&r2=84558&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 17:33:05 2009
@@ -73,7 +73,7 @@
   void printSBitModifierOperand(const MCInst *MI, unsigned OpNum) {}
   void printRegisterList(const MCInst *MI, unsigned OpNum);
   void printCPInstOperand(const MCInst *MI, unsigned OpNum,
-                          const char *Modifier) {}
+                          const char *Modifier);
   void printJTBlockOperand(const MCInst *MI, unsigned OpNum) {}
   void printJT2BlockOperand(const MCInst *MI, unsigned OpNum) {}
   void printTBAddrMode(const MCInst *MI, unsigned OpNum) {}




From sabre at nondot.org  Mon Oct 19 17:49:00 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 22:49:00 -0000
Subject: [llvm-commits] [llvm] r84561 - in /llvm/trunk:
 include/llvm/MC/MCContext.h lib/MC/MCContext.cpp
 lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
 lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp
Message-ID: <200910192249.n9JMn1Jk018643@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 17:49:00 2009
New Revision: 84561

URL: http://llvm.org/viewvc/llvm-project?rev=84561&view=rev
Log:
add a twine version of MCContext::GetOrCreateSymbol.

Modified:
    llvm/trunk/include/llvm/MC/MCContext.h
    llvm/trunk/lib/MC/MCContext.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp

Modified: llvm/trunk/include/llvm/MC/MCContext.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/MC/MCContext.h?rev=84561&r1=84560&r2=84561&view=diff

==============================================================================
--- llvm/trunk/include/llvm/MC/MCContext.h (original)
+++ llvm/trunk/include/llvm/MC/MCContext.h Mon Oct 19 17:49:00 2009
@@ -19,6 +19,7 @@
   class MCSection;
   class MCSymbol;
   class StringRef;
+  class Twine;
 
   /// MCContext - Context object for machine code objects.  This class owns all
   /// of the sections that it creates.
@@ -58,6 +59,7 @@
     /// @param IsTemporary - Whether this symbol is an assembler temporary,
     /// which should not survive into the symbol table for the translation unit.
     MCSymbol *GetOrCreateSymbol(const StringRef &Name);
+    MCSymbol *GetOrCreateSymbol(const Twine &Name);
 
     /// CreateTemporarySymbol - Create a new temporary symbol with the specified
     /// @param Name.

Modified: llvm/trunk/lib/MC/MCContext.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/MC/MCContext.cpp?rev=84561&r1=84560&r2=84561&view=diff

==============================================================================
--- llvm/trunk/lib/MC/MCContext.cpp (original)
+++ llvm/trunk/lib/MC/MCContext.cpp Mon Oct 19 17:49:00 2009
@@ -8,10 +8,11 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/MC/MCContext.h"
-
 #include "llvm/MC/MCSection.h"
 #include "llvm/MC/MCSymbol.h"
 #include "llvm/MC/MCValue.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/Twine.h"
 using namespace llvm;
 
 MCContext::MCContext() {
@@ -38,6 +39,13 @@
   return Entry = new (*this) MCSymbol(Name, false);
 }
 
+MCSymbol *MCContext::GetOrCreateSymbol(const Twine &Name) {
+  SmallString<128> NameSV;
+  Name.toVector(NameSV);
+  return GetOrCreateSymbol(NameSV.str());
+}
+
+
 MCSymbol *MCContext::CreateTemporarySymbol(const StringRef &Name) {
   // If unnamed, just create a symbol.
   if (Name.empty())

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84561&r1=84560&r2=84561&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 17:49:00 2009
@@ -173,7 +173,7 @@
         else {
           // FIXME: Remove this when Darwin transition to @GOT like syntax.
           Name = Mang->getMangledName(GV, "$non_lazy_ptr", true);
-          MCSymbol *Sym = OutContext.GetOrCreateSymbol(Name.c_str());
+          MCSymbol *Sym = OutContext.GetOrCreateSymbol(StringRef(Name));
           
           MachineModuleInfoMachO &MMIMachO =
             MMI->getObjFileInfo<MachineModuleInfoMachO>();
@@ -1336,10 +1336,11 @@
     
     // Emit the label.
     // FIXME: MOVE TO SHARED PLACE.
-    SmallString<60> Name;
     unsigned Id = (unsigned)MI->getOperand(2).getImm();
-    raw_svector_ostream(Name) << MAI->getPrivateGlobalPrefix() << "PC" << Id;
-    OutStreamer.EmitLabel(OutContext.GetOrCreateSymbol(Name.str()));
+    const char *Prefix = MAI->getPrivateGlobalPrefix();
+    MCSymbol *Label =
+      OutContext.GetOrCreateSymbol(Twine(Prefix)+"PC"+Twine(Id));
+    OutStreamer.EmitLabel(Label);
     
     
     // Form and emit tha dd.

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp?rev=84561&r1=84560&r2=84561&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp Mon Oct 19 17:49:00 2009
@@ -288,12 +288,12 @@
     std::string Name = Mang->makeNameProper(MO.getSymbolName());
     if (MO.getTargetFlags() == X86II::MO_DARWIN_STUB) {
       Name += "$stub";
-      MCSymbol *Sym = OutContext.GetOrCreateSymbol(Name);
+      MCSymbol *Sym = OutContext.GetOrCreateSymbol(StringRef(Name));
       const MCSymbol *&StubSym =
         MMI->getObjFileInfo<MachineModuleInfoMachO>().getFnStubEntry(Sym);
       if (StubSym == 0) {
         Name.erase(Name.end()-5, Name.end());
-        StubSym = OutContext.GetOrCreateSymbol(Name);
+        StubSym = OutContext.GetOrCreateSymbol(StringRef(Name));
       }
     }
     




From sabre at nondot.org  Mon Oct 19 17:51:16 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 22:51:16 -0000
Subject: [llvm-commits] [llvm] r84562 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910192251.n9JMpH6e018716@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 17:51:16 2009
New Revision: 84562

URL: http://llvm.org/viewvc/llvm-project?rev=84562&view=rev
Log:
use EmitLabel instead of text emission

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84562&r1=84561&r2=84562&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 17:51:16 2009
@@ -1338,8 +1338,7 @@
     // FIXME: MOVE TO SHARED PLACE.
     unsigned Id = (unsigned)MI->getOperand(2).getImm();
     const char *Prefix = MAI->getPrivateGlobalPrefix();
-    MCSymbol *Label =
-      OutContext.GetOrCreateSymbol(Twine(Prefix)+"PC"+Twine(Id));
+    MCSymbol *Label =OutContext.GetOrCreateSymbol(Twine(Prefix)+"PC"+Twine(Id));
     OutStreamer.EmitLabel(Label);
     
     
@@ -1362,8 +1361,11 @@
 
     EmitAlignment(2);
 
-    O << MAI->getPrivateGlobalPrefix() << "CPI" << getFunctionNumber()
-      << '_' << LabelId << ":\n";
+    const char *Prefix = MAI->getPrivateGlobalPrefix();
+    MCSymbol *Label = OutContext.GetOrCreateSymbol(Twine(Prefix)+"CPI"+
+                                                   Twine(getFunctionNumber())+
+                                                   "_"+ Twine(LabelId));
+    OutStreamer.EmitLabel(Label);
 
     const MachineConstantPoolEntry &MCPE = MCP->getConstants()[CPIdx];
     if (MCPE.isMachineConstantPoolEntry())




From grosbach at apple.com  Mon Oct 19 17:57:03 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Mon, 19 Oct 2009 22:57:03 -0000
Subject: [llvm-commits] [llvm] r84563 - in /llvm/trunk:
 lib/Target/ARM/ARMBaseRegisterInfo.cpp lib/Target/ARM/ARMRegisterInfo.td
 lib/Target/ARM/README-Thumb.txt lib/Target/ARM/Thumb1RegisterInfo.cpp
 test/CodeGen/Thumb/2009-08-20-ISelBug.ll
Message-ID: <200910192257.n9JMv32b018931@zion.cs.uiuc.edu>

Author: grosbach
Date: Mon Oct 19 17:57:03 2009
New Revision: 84563

URL: http://llvm.org/viewvc/llvm-project?rev=84563&view=rev
Log:
Enable allocation of R3 in Thumb1

Modified:
    llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
    llvm/trunk/lib/Target/ARM/ARMRegisterInfo.td
    llvm/trunk/lib/Target/ARM/README-Thumb.txt
    llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
    llvm/trunk/test/CodeGen/Thumb/2009-08-20-ISelBug.ll

Modified: llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp?rev=84563&r1=84562&r2=84563&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp Mon Oct 19 17:57:03 2009
@@ -740,8 +740,7 @@
   case ARM::R1:
     return ARM::R0;
   case ARM::R3:
-    // FIXME!
-    return STI.isThumb1Only() ? 0 : ARM::R2;
+    return ARM::R2;
   case ARM::R5:
     return ARM::R4;
   case ARM::R7:
@@ -830,8 +829,7 @@
   case ARM::R0:
     return ARM::R1;
   case ARM::R2:
-    // FIXME!
-    return STI.isThumb1Only() ? 0 : ARM::R3;
+    return ARM::R3;
   case ARM::R4:
     return ARM::R5;
   case ARM::R6:

Modified: llvm/trunk/lib/Target/ARM/ARMRegisterInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMRegisterInfo.td?rev=84563&r1=84562&r2=84563&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMRegisterInfo.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMRegisterInfo.td Mon Oct 19 17:57:03 2009
@@ -222,12 +222,9 @@
     iterator allocation_order_begin(const MachineFunction &MF) const;
     iterator allocation_order_end(const MachineFunction &MF) const;
   }];
-  // FIXME: We are reserving r3 in Thumb mode in case the PEI needs to use it
-  // to generate large stack offset. Make it available once we have register
-  // scavenging.
   let MethodBodies = [{
     static const unsigned THUMB_tGPR_AO[] = {
-      ARM::R0, ARM::R1, ARM::R2,
+      ARM::R0, ARM::R1, ARM::R2, ARM::R3,
       ARM::R4, ARM::R5, ARM::R6, ARM::R7 };
 
     // FP is R7, only low registers available.

Modified: llvm/trunk/lib/Target/ARM/README-Thumb.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/README-Thumb.txt?rev=84563&r1=84562&r2=84563&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/README-Thumb.txt (original)
+++ llvm/trunk/lib/Target/ARM/README-Thumb.txt Mon Oct 19 17:57:03 2009
@@ -196,14 +196,6 @@
 objects are referenced off the frame pointer with negative offsets. See
 oggenc for an example.
 
-//===---------------------------------------------------------------------===//
-
-We are reserving R3 as a scratch register under thumb mode. So if it is live in
-to the function, we save / restore R3 to / from R12. Until register scavenging
-is done, we should save R3 to a high callee saved reg at emitPrologue time
-(when hasFP is true or stack size is large) and restore R3 from that register
-instead. This allows us to at least get rid of the save to r12 everytime it is
-used.
 
 //===---------------------------------------------------------------------===//
 

Modified: llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp?rev=84563&r1=84562&r2=84563&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp Mon Oct 19 17:57:03 2009
@@ -845,7 +845,6 @@
 
   if (VARegSaveSize) {
     // Epilogue for vararg functions: pop LR to R3 and branch off it.
-    // FIXME: Verify this is still ok when R3 is no longer being reserved.
     AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tPOP)))
       .addReg(0) // No write back.
       .addReg(ARM::R3, RegState::Define);

Modified: llvm/trunk/test/CodeGen/Thumb/2009-08-20-ISelBug.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Thumb/2009-08-20-ISelBug.ll?rev=84563&r1=84562&r2=84563&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/Thumb/2009-08-20-ISelBug.ll (original)
+++ llvm/trunk/test/CodeGen/Thumb/2009-08-20-ISelBug.ll Mon Oct 19 17:57:03 2009
@@ -11,7 +11,7 @@
 
 define arm_apcscc i32 @t(%struct.asl_file_t* %s, i64 %off, i64* %out) nounwind optsize {
 ; CHECK: t:
-; CHECK: adds r4, #8
+; CHECK: adds r3, #8
 entry:
   %val = alloca i64, align 4                      ; <i64*> [#uses=3]
   %0 = icmp eq %struct.asl_file_t* %s, null       ; <i1> [#uses=1]




From ofv at wanadoo.es  Mon Oct 19 18:00:00 2009
From: ofv at wanadoo.es (Oscar Fuentes)
Date: Mon, 19 Oct 2009 23:00:00 -0000
Subject: [llvm-commits] [llvm] r84564 -
	/llvm/trunk/cmake/modules/LLVMLibDeps.cmake
Message-ID: <200910192300.n9JN00OK019259@zion.cs.uiuc.edu>

Author: ofv
Date: Mon Oct 19 18:00:00 2009
New Revision: 84564

URL: http://llvm.org/viewvc/llvm-project?rev=84564&view=rev
Log:
Updated cmake library dependencies.

Modified:
    llvm/trunk/cmake/modules/LLVMLibDeps.cmake

Modified: llvm/trunk/cmake/modules/LLVMLibDeps.cmake
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/cmake/modules/LLVMLibDeps.cmake?rev=84564&r1=84563&r2=84564&view=diff

==============================================================================
--- llvm/trunk/cmake/modules/LLVMLibDeps.cmake (original)
+++ llvm/trunk/cmake/modules/LLVMLibDeps.cmake Mon Oct 19 18:00:00 2009
@@ -34,7 +34,7 @@
 set(MSVC_LIB_DEPS_LLVMMSIL LLVMAnalysis LLVMCodeGen LLVMCore LLVMMSILInfo LLVMScalarOpts LLVMSupport LLVMSystem LLVMTarget LLVMTransformUtils LLVMipa)
 set(MSVC_LIB_DEPS_LLVMMSILInfo LLVMSupport)
 set(MSVC_LIB_DEPS_LLVMMSP430AsmPrinter LLVMAsmPrinter LLVMCodeGen LLVMCore LLVMMC LLVMMSP430Info LLVMSupport LLVMSystem LLVMTarget)
-set(MSVC_LIB_DEPS_LLVMMSP430CodeGen LLVMCodeGen LLVMCore LLVMMC LLVMMSP430Info LLVMSelectionDAG LLVMSupport LLVMTarget)
+set(MSVC_LIB_DEPS_LLVMMSP430CodeGen LLVMCodeGen LLVMCore LLVMMC LLVMMSP430Info LLVMSelectionDAG LLVMSupport LLVMSystem LLVMTarget)
 set(MSVC_LIB_DEPS_LLVMMSP430Info LLVMSupport)
 set(MSVC_LIB_DEPS_LLVMMipsAsmPrinter LLVMAsmPrinter LLVMCodeGen LLVMCore LLVMMC LLVMMipsCodeGen LLVMMipsInfo LLVMSupport LLVMSystem LLVMTarget)
 set(MSVC_LIB_DEPS_LLVMMipsCodeGen LLVMCodeGen LLVMCore LLVMMC LLVMMipsInfo LLVMSelectionDAG LLVMSupport LLVMSystem LLVMTarget)




From sabre at nondot.org  Mon Oct 19 18:05:23 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 23:05:23 -0000
Subject: [llvm-commits] [llvm] r84565 -
	/llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp
Message-ID: <200910192305.n9JN5Ok6019681@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 18:05:23 2009
New Revision: 84565

URL: http://llvm.org/viewvc/llvm-project?rev=84565&view=rev
Log:
simplify by using the twine form of GetOrCreateSymbol

Modified:
    llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp?rev=84565&r1=84564&r2=84565&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp Mon Oct 19 18:05:23 2009
@@ -38,10 +38,8 @@
 
 
 MCSymbol *X86MCInstLower::GetPICBaseSymbol() const {
-  SmallString<60> Name;
-  raw_svector_ostream(Name) << AsmPrinter.MAI->getPrivateGlobalPrefix()
-    << AsmPrinter.getFunctionNumber() << "$pb";
-  return Ctx.GetOrCreateSymbol(Name.str());
+  return Ctx.GetOrCreateSymbol(Twine(AsmPrinter.MAI->getPrivateGlobalPrefix())+
+                               Twine(AsmPrinter.getFunctionNumber())+"$pb");
 }
 
 
@@ -449,10 +447,9 @@
     //   MYGLOBAL + (. - PICBASE)
     // However, we can't generate a ".", so just emit a new label here and refer
     // to it.  We know that this operand flag occurs at most once per function.
-    SmallString<64> Name;
-    raw_svector_ostream(Name) << MAI->getPrivateGlobalPrefix()
-      << "picbaseref" << getFunctionNumber();
-    MCSymbol *DotSym = OutContext.GetOrCreateSymbol(Name.str());
+    const char *Prefix = MAI->getPrivateGlobalPrefix();
+    MCSymbol *DotSym = OutContext.GetOrCreateSymbol(Twine(Prefix)+"picbaseref"+
+                                                    Twine(getFunctionNumber()));
     OutStreamer.EmitLabel(DotSym);
     
     // Now that we have emitted the label, lower the complex operand expression.




From sabre at nondot.org  Mon Oct 19 18:31:43 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 23:31:43 -0000
Subject: [llvm-commits] [llvm] r84566 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMInstPrinter.cpp
 ARMMCInstLower.cpp
Message-ID: <200910192331.n9JNVhZw020740@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 18:31:43 2009
New Revision: 84566

URL: http://llvm.org/viewvc/llvm-project?rev=84566&view=rev
Log:
handle addmode4 modifiers, fix a fixme in printRegisterList
by ignoring all implicit regs when lowering.


Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84566&r1=84565&r2=84566&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 18:31:43 2009
@@ -116,13 +116,10 @@
 
 void ARMInstPrinter::printAddrMode4Operand(const MCInst *MI, unsigned OpNum,
                                            const char *Modifier) {
-  // FIXME: ENABLE assert.
-  //assert((Modifier == 0 || Modifier[0] == 0) && "Cannot print modifiers");
-  
   const MCOperand &MO1 = MI->getOperand(OpNum);
   const MCOperand &MO2 = MI->getOperand(OpNum+1);
   ARM_AM::AMSubMode Mode = ARM_AM::getAM4SubMode(MO2.getImm());
-  if (0 && Modifier && strcmp(Modifier, "submode") == 0) {
+  if (Modifier && strcmp(Modifier, "submode") == 0) {
     if (MO1.getReg() == ARM::SP) {
       // FIXME
       bool isLDM = (MI->getOpcode() == ARM::LDM ||
@@ -132,7 +129,7 @@
       O << ARM_AM::getAMSubModeAltStr(Mode, isLDM);
     } else
       O << ARM_AM::getAMSubModeStr(Mode);
-  } else if (0 && Modifier && strcmp(Modifier, "wide") == 0) {
+  } else if (Modifier && strcmp(Modifier, "wide") == 0) {
     ARM_AM::AMSubMode Mode = ARM_AM::getAM4SubMode(MO2.getImm());
     if (Mode == ARM_AM::ia)
       O << ".w";
@@ -147,12 +144,7 @@
   O << "{";
   // Always skip the first operand, it's the optional (and implicit writeback).
   for (unsigned i = OpNum+1, e = MI->getNumOperands(); i != e; ++i) {
-#if 0 // FIXME: HANDLE WHEN LOWERING??
-    if (MI->getOperand(i).isImplicit())
-      continue;
-#endif
     if (i != OpNum+1) O << ", ";
-    
     O << getRegisterName(MI->getOperand(i).getReg());
   }
   O << "}";

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp?rev=84566&r1=84565&r2=84566&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp Mon Oct 19 18:31:43 2009
@@ -118,6 +118,8 @@
       MI->dump();
       assert(0 && "unknown operand type");
     case MachineOperand::MO_Register:
+      // Ignore all implicit register operands.
+      if (MO.isImplicit()) continue;
       MCOp = MCOperand::CreateReg(MO.getReg());
       break;
     case MachineOperand::MO_Immediate:




From sabre at nondot.org  Mon Oct 19 18:35:57 2009
From: sabre at nondot.org (Chris Lattner)
Date: Mon, 19 Oct 2009 23:35:57 -0000
Subject: [llvm-commits] [llvm] r84567 -
	/llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp
Message-ID: <200910192335.n9JNZvfW020893@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 18:35:57 2009
New Revision: 84567

URL: http://llvm.org/viewvc/llvm-project?rev=84567&view=rev
Log:
X86 should ignore implicit regs when lowering to MCInst also,
no functionality change.

Modified:
    llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp

Modified: llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp?rev=84567&r1=84566&r2=84567&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/X86/AsmPrinter/X86MCInstLower.cpp Mon Oct 19 18:35:57 2009
@@ -306,6 +306,8 @@
       MI->dump();
       llvm_unreachable("unknown operand type");
     case MachineOperand::MO_Register:
+      // Ignore all implicit register operands.
+      if (MO.isImplicit()) continue;
       MCOp = MCOperand::CreateReg(MO.getReg());
       break;
     case MachineOperand::MO_Immediate:




From johnny.chen at apple.com  Mon Oct 19 18:46:50 2009
From: johnny.chen at apple.com (Johnny Chen)
Date: Mon, 19 Oct 2009 16:46:50 -0700
Subject: [llvm-commits] ARM NEON patch for ARMInstrFormats.td
Message-ID: <2568D436-F301-49A2-ACBC-763032CD2D91@apple.com>

Hi,

I manually edited the patch file to remove some hunks which represent WIP
to not over-specify some immediate fields.  But this patch should be easy
to review so I send it first.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: ARM-NEON-NLdSt.patch
Type: application/octet-stream
Size: 1096 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091019/1e869b37/attachment.obj 
-------------- next part --------------



All the following NEON instructions are under-specified without this patch:


    VLD1d16         11110100________________________
    VLD1d32         11110100________________________
    VLD1d64         11110100________________________
    VLD1d8          11110100________________________
    VLD1df          11110100________________________
    VLD1q16         11110100________________________
    VLD1q32         11110100________________________
    VLD1q64         11110100________________________
    VLD1q8          11110100________________________
    VLD1qf          11110100________________________
    VLD2LNd16       11110100________________________
    VLD2LNd32       11110100________________________
    VLD2LNd8        11110100________________________
    VLD2LNq16a      11110100________________________
    VLD2LNq16b      11110100________________________
    VLD2LNq32a      11110100________________________
    VLD2LNq32b      11110100________________________
    VLD2d16         11110100________________________
    VLD2d32         11110100________________________
    VLD2d64         11110100________________________
    VLD2d8          11110100________________________
    VLD2q16         11110100________________________
    VLD2q32         11110100________________________
    VLD2q8          11110100________________________
    VLD3LNd16       11110100________________________
    VLD3LNd32       11110100________________________
    VLD3LNd8        11110100________________________
    VLD3LNq16a      11110100________________________
    VLD3LNq16b      11110100________________________
    VLD3LNq32a      11110100________________________
    VLD3LNq32b      11110100________________________
    VLD3d16         11110100________________________
    VLD3d32         11110100________________________
    VLD3d64         11110100________________________
    VLD3d8          11110100________________________
    VLD3q16a        11110100________________________
    VLD3q16b        11110100________________________
    VLD3q32a        11110100________________________
    VLD3q32b        11110100________________________
    VLD3q8a         11110100________________________
    VLD3q8b         11110100________________________
    VLD4LNd16       11110100________________________
    VLD4LNd32       11110100________________________
    VLD4LNd8        11110100________________________
    VLD4LNq16a      11110100________________________
    VLD4LNq16b      11110100________________________
    VLD4LNq32a      11110100________________________
    VLD4LNq32b      11110100________________________
    VLD4d16         11110100________________________
    VLD4d32         11110100________________________
    VLD4d64         11110100________________________
    VLD4d8          11110100________________________
    VLD4q16a        11110100________________________
    VLD4q16b        11110100________________________
    VLD4q32a        11110100________________________
    VLD4q32b        11110100________________________
    VLD4q8a         11110100________________________
    VLD4q8b         11110100________________________
    VST1d16         11110100________________________
    VST1d32         11110100________________________
    VST1d64         11110100________________________
    VST1d8          11110100________________________
    VST1df          11110100________________________
    VST1q16         11110100________________________
    VST1q32         11110100________________________
    VST1q64         11110100________________________
    VST1q8          11110100________________________
    VST1qf          11110100________________________
    VST2LNd16       11110100________________________
    VST2LNd32       11110100________________________
    VST2LNd8        11110100________________________
    VST2LNq16a      11110100________________________
    VST2LNq16b      11110100________________________
    VST2LNq32a      11110100________________________
    VST2LNq32b      11110100________________________
    VST2d16         11110100________________________
    VST2d32         11110100________________________
    VST2d64         11110100________________________
    VST2d8          11110100________________________
    VST2q16         11110100________________________
    VST2q32         11110100________________________
    VST2q8          11110100________________________
    VST3LNd16       11110100________________________
    VST3LNd32       11110100________________________
    VST3LNd8        11110100________________________
    VST3LNq16a      11110100________________________
    VST3LNq16b      11110100________________________
    VST3LNq32a      11110100________________________
    VST3LNq32b      11110100________________________
    VST3d16         11110100________________________
    VST3d32         11110100________________________
    VST3d64         11110100________________________
    VST3d8          11110100________________________
    VST3q16a        11110100________________________
    VST3q16b        11110100________________________
    VST3q32a        11110100________________________
    VST3q32b        11110100________________________
    VST3q8a         11110100________________________
    VST3q8b         11110100________________________
    VST4LNd16       11110100________________________
    VST4LNd32       11110100________________________
    VST4LNd8        11110100________________________
    VST4LNq16a      11110100________________________
    VST4LNq16b      11110100________________________
    VST4LNq32a      11110100________________________
    VST4LNq32b      11110100________________________
    VST4d16         11110100________________________
    VST4d32         11110100________________________
    VST4d64         11110100________________________
    VST4d8          11110100________________________
    VST4q16a        11110100________________________
    VST4q16b        11110100________________________
    VST4q32a        11110100________________________
    VST4q32b        11110100________________________
    VST4q8a         11110100________________________
    VST4q8b         11110100________________________


From grosbach at apple.com  Mon Oct 19 19:19:08 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Tue, 20 Oct 2009 00:19:08 -0000
Subject: [llvm-commits] [llvm] r84570 -
	/llvm/trunk/lib/Target/ARM/ARMInstrFormats.td
Message-ID: <200910200019.n9K0J8Cl022403@zion.cs.uiuc.edu>

Author: grosbach
Date: Mon Oct 19 19:19:08 2009
New Revision: 84570

URL: http://llvm.org/viewvc/llvm-project?rev=84570&view=rev
Log:
Add missing encoding bits to NLdSt class of instructions.

Patch by Johnny Chen.


Modified:
    llvm/trunk/lib/Target/ARM/ARMInstrFormats.td

Modified: llvm/trunk/lib/Target/ARM/ARMInstrFormats.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrFormats.td?rev=84570&r1=84569&r2=84570&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrFormats.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrFormats.td Mon Oct 19 19:19:08 2009
@@ -1220,6 +1220,10 @@
             string asm, string cstr, list<dag> pattern>
   : NeonI<oops, iops, AddrMode6, IndexModeNone, itin, asm, cstr, pattern> {
   let Inst{31-24} = 0b11110100;
+  let Inst{23} = op23;
+  let Inst{21-20} = op21_20;
+  let Inst{11-8} = op11_8;
+  let Inst{7-4} = op7_4;
 }
 
 class NDataI<dag oops, dag iops, InstrItinClass itin,




From grosbach at apple.com  Mon Oct 19 19:19:52 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Mon, 19 Oct 2009 17:19:52 -0700
Subject: [llvm-commits] ARM NEON patch for ARMInstrFormats.td
In-Reply-To: <2568D436-F301-49A2-ACBC-763032CD2D91@apple.com>
References: <2568D436-F301-49A2-ACBC-763032CD2D91@apple.com>
Message-ID: <509F0CED-04F8-43C9-8A11-D425531CEC12@apple.com>

Applied as r84570.

Thanks again for these patches.

-Jim

On Oct 19, 2009, at 4:46 PM, Johnny Chen wrote:

> Hi,
>
> I manually edited the patch file to remove some hunks which  
> represent WIP
> to not over-specify some immediate fields.  But this patch should be  
> easy
> to review so I send it first.
>
> <ARM-NEON-NLdSt.patch>
>
>
> All the following NEON instructions are under-specified without this  
> patch:
>
>
>    VLD1d16         11110100________________________
>    VLD1d32         11110100________________________
>    VLD1d64         11110100________________________
>    VLD1d8          11110100________________________
>    VLD1df          11110100________________________
>    VLD1q16         11110100________________________
>    VLD1q32         11110100________________________
>    VLD1q64         11110100________________________
>    VLD1q8          11110100________________________
>    VLD1qf          11110100________________________
>    VLD2LNd16       11110100________________________
>    VLD2LNd32       11110100________________________
>    VLD2LNd8        11110100________________________
>    VLD2LNq16a      11110100________________________
>    VLD2LNq16b      11110100________________________
>    VLD2LNq32a      11110100________________________
>    VLD2LNq32b      11110100________________________
>    VLD2d16         11110100________________________
>    VLD2d32         11110100________________________
>    VLD2d64         11110100________________________
>    VLD2d8          11110100________________________
>    VLD2q16         11110100________________________
>    VLD2q32         11110100________________________
>    VLD2q8          11110100________________________
>    VLD3LNd16       11110100________________________
>    VLD3LNd32       11110100________________________
>    VLD3LNd8        11110100________________________
>    VLD3LNq16a      11110100________________________
>    VLD3LNq16b      11110100________________________
>    VLD3LNq32a      11110100________________________
>    VLD3LNq32b      11110100________________________
>    VLD3d16         11110100________________________
>    VLD3d32         11110100________________________
>    VLD3d64         11110100________________________
>    VLD3d8          11110100________________________
>    VLD3q16a        11110100________________________
>    VLD3q16b        11110100________________________
>    VLD3q32a        11110100________________________
>    VLD3q32b        11110100________________________
>    VLD3q8a         11110100________________________
>    VLD3q8b         11110100________________________
>    VLD4LNd16       11110100________________________
>    VLD4LNd32       11110100________________________
>    VLD4LNd8        11110100________________________
>    VLD4LNq16a      11110100________________________
>    VLD4LNq16b      11110100________________________
>    VLD4LNq32a      11110100________________________
>    VLD4LNq32b      11110100________________________
>    VLD4d16         11110100________________________
>    VLD4d32         11110100________________________
>    VLD4d64         11110100________________________
>    VLD4d8          11110100________________________
>    VLD4q16a        11110100________________________
>    VLD4q16b        11110100________________________
>    VLD4q32a        11110100________________________
>    VLD4q32b        11110100________________________
>    VLD4q8a         11110100________________________
>    VLD4q8b         11110100________________________
>    VST1d16         11110100________________________
>    VST1d32         11110100________________________
>    VST1d64         11110100________________________
>    VST1d8          11110100________________________
>    VST1df          11110100________________________
>    VST1q16         11110100________________________
>    VST1q32         11110100________________________
>    VST1q64         11110100________________________
>    VST1q8          11110100________________________
>    VST1qf          11110100________________________
>    VST2LNd16       11110100________________________
>    VST2LNd32       11110100________________________
>    VST2LNd8        11110100________________________
>    VST2LNq16a      11110100________________________
>    VST2LNq16b      11110100________________________
>    VST2LNq32a      11110100________________________
>    VST2LNq32b      11110100________________________
>    VST2d16         11110100________________________
>    VST2d32         11110100________________________
>    VST2d64         11110100________________________
>    VST2d8          11110100________________________
>    VST2q16         11110100________________________
>    VST2q32         11110100________________________
>    VST2q8          11110100________________________
>    VST3LNd16       11110100________________________
>    VST3LNd32       11110100________________________
>    VST3LNd8        11110100________________________
>    VST3LNq16a      11110100________________________
>    VST3LNq16b      11110100________________________
>    VST3LNq32a      11110100________________________
>    VST3LNq32b      11110100________________________
>    VST3d16         11110100________________________
>    VST3d32         11110100________________________
>    VST3d64         11110100________________________
>    VST3d8          11110100________________________
>    VST3q16a        11110100________________________
>    VST3q16b        11110100________________________
>    VST3q32a        11110100________________________
>    VST3q32b        11110100________________________
>    VST3q8a         11110100________________________
>    VST3q8b         11110100________________________
>    VST4LNd16       11110100________________________
>    VST4LNd32       11110100________________________
>    VST4LNd8        11110100________________________
>    VST4LNq16a      11110100________________________
>    VST4LNq16b      11110100________________________
>    VST4LNq32a      11110100________________________
>    VST4LNq32b      11110100________________________
>    VST4d16         11110100________________________
>    VST4d32         11110100________________________
>    VST4d64         11110100________________________
>    VST4d8          11110100________________________
>    VST4q16a        11110100________________________
>    VST4q16b        11110100________________________
>    VST4q32a        11110100________________________
>    VST4q32b        11110100________________________
>    VST4q8a         11110100________________________
>    VST4q8b         11110100________________________
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From johnny.chen at apple.com  Mon Oct 19 19:33:53 2009
From: johnny.chen at apple.com (Johnny Chen)
Date: Mon, 19 Oct 2009 17:33:53 -0700
Subject: [llvm-commits] ARM NEON patch for ARMInstrFormats.td
In-Reply-To: <509F0CED-04F8-43C9-8A11-D425531CEC12@apple.com>
References: <2568D436-F301-49A2-ACBC-763032CD2D91@apple.com>
	<509F0CED-04F8-43C9-8A11-D425531CEC12@apple.com>
Message-ID: <300516B4-D2D1-4E59-8706-C6BF5DAAD31E@apple.com>

No problem.
BTW, is my previous patch concerning VEXT (Vector Extract) still being reviewed? :-)

On Oct 19, 2009, at 5:19 PM, Jim Grosbach wrote:

> Applied as r84570.
> 
> Thanks again for these patches.
> 
> -Jim
> 
> On Oct 19, 2009, at 4:46 PM, Johnny Chen wrote:
> 
>> Hi,
>> 
>> I manually edited the patch file to remove some hunks which represent WIP
>> to not over-specify some immediate fields.  But this patch should be easy
>> to review so I send it first.
>> 
>> <ARM-NEON-NLdSt.patch>
>> 
>> 
>> All the following NEON instructions are under-specified without this patch:
>> 
>> 
>>   VLD1d16         11110100________________________
>>   VLD1d32         11110100________________________
>>   VLD1d64         11110100________________________
>>   VLD1d8          11110100________________________
>>   VLD1df          11110100________________________
>>   VLD1q16         11110100________________________
>>   VLD1q32         11110100________________________
>>   VLD1q64         11110100________________________
>>   VLD1q8          11110100________________________
>>   VLD1qf          11110100________________________
>>   VLD2LNd16       11110100________________________
>>   VLD2LNd32       11110100________________________
>>   VLD2LNd8        11110100________________________
>>   VLD2LNq16a      11110100________________________
>>   VLD2LNq16b      11110100________________________
>>   VLD2LNq32a      11110100________________________
>>   VLD2LNq32b      11110100________________________
>>   VLD2d16         11110100________________________
>>   VLD2d32         11110100________________________
>>   VLD2d64         11110100________________________
>>   VLD2d8          11110100________________________
>>   VLD2q16         11110100________________________
>>   VLD2q32         11110100________________________
>>   VLD2q8          11110100________________________
>>   VLD3LNd16       11110100________________________
>>   VLD3LNd32       11110100________________________
>>   VLD3LNd8        11110100________________________
>>   VLD3LNq16a      11110100________________________
>>   VLD3LNq16b      11110100________________________
>>   VLD3LNq32a      11110100________________________
>>   VLD3LNq32b      11110100________________________
>>   VLD3d16         11110100________________________
>>   VLD3d32         11110100________________________
>>   VLD3d64         11110100________________________
>>   VLD3d8          11110100________________________
>>   VLD3q16a        11110100________________________
>>   VLD3q16b        11110100________________________
>>   VLD3q32a        11110100________________________
>>   VLD3q32b        11110100________________________
>>   VLD3q8a         11110100________________________
>>   VLD3q8b         11110100________________________
>>   VLD4LNd16       11110100________________________
>>   VLD4LNd32       11110100________________________
>>   VLD4LNd8        11110100________________________
>>   VLD4LNq16a      11110100________________________
>>   VLD4LNq16b      11110100________________________
>>   VLD4LNq32a      11110100________________________
>>   VLD4LNq32b      11110100________________________
>>   VLD4d16         11110100________________________
>>   VLD4d32         11110100________________________
>>   VLD4d64         11110100________________________
>>   VLD4d8          11110100________________________
>>   VLD4q16a        11110100________________________
>>   VLD4q16b        11110100________________________
>>   VLD4q32a        11110100________________________
>>   VLD4q32b        11110100________________________
>>   VLD4q8a         11110100________________________
>>   VLD4q8b         11110100________________________
>>   VST1d16         11110100________________________
>>   VST1d32         11110100________________________
>>   VST1d64         11110100________________________
>>   VST1d8          11110100________________________
>>   VST1df          11110100________________________
>>   VST1q16         11110100________________________
>>   VST1q32         11110100________________________
>>   VST1q64         11110100________________________
>>   VST1q8          11110100________________________
>>   VST1qf          11110100________________________
>>   VST2LNd16       11110100________________________
>>   VST2LNd32       11110100________________________
>>   VST2LNd8        11110100________________________
>>   VST2LNq16a      11110100________________________
>>   VST2LNq16b      11110100________________________
>>   VST2LNq32a      11110100________________________
>>   VST2LNq32b      11110100________________________
>>   VST2d16         11110100________________________
>>   VST2d32         11110100________________________
>>   VST2d64         11110100________________________
>>   VST2d8          11110100________________________
>>   VST2q16         11110100________________________
>>   VST2q32         11110100________________________
>>   VST2q8          11110100________________________
>>   VST3LNd16       11110100________________________
>>   VST3LNd32       11110100________________________
>>   VST3LNd8        11110100________________________
>>   VST3LNq16a      11110100________________________
>>   VST3LNq16b      11110100________________________
>>   VST3LNq32a      11110100________________________
>>   VST3LNq32b      11110100________________________
>>   VST3d16         11110100________________________
>>   VST3d32         11110100________________________
>>   VST3d64         11110100________________________
>>   VST3d8          11110100________________________
>>   VST3q16a        11110100________________________
>>   VST3q16b        11110100________________________
>>   VST3q32a        11110100________________________
>>   VST3q32b        11110100________________________
>>   VST3q8a         11110100________________________
>>   VST3q8b         11110100________________________
>>   VST4LNd16       11110100________________________
>>   VST4LNd32       11110100________________________
>>   VST4LNd8        11110100________________________
>>   VST4LNq16a      11110100________________________
>>   VST4LNq16b      11110100________________________
>>   VST4LNq32a      11110100________________________
>>   VST4LNq32b      11110100________________________
>>   VST4d16         11110100________________________
>>   VST4d32         11110100________________________
>>   VST4d64         11110100________________________
>>   VST4d8          11110100________________________
>>   VST4q16a        11110100________________________
>>   VST4q16b        11110100________________________
>>   VST4q32a        11110100________________________
>>   VST4q32b        11110100________________________
>>   VST4q8a         11110100________________________
>>   VST4q8b         11110100________________________
>> 
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 




From grosbach at apple.com  Mon Oct 19 19:38:20 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Tue, 20 Oct 2009 00:38:20 -0000
Subject: [llvm-commits] [llvm] r84572 - in /llvm/trunk/lib/Target/ARM:
 ARMInstrFormats.td ARMInstrNEON.td
Message-ID: <200910200038.n9K0cKI2023038@zion.cs.uiuc.edu>

Author: grosbach
Date: Mon Oct 19 19:38:19 2009
New Revision: 84572

URL: http://llvm.org/viewvc/llvm-project?rev=84572&view=rev
Log:
Refs: A8-598.
Leave Inst{11-8}, which represents the starting byte index of the extracted
result in the concatenation of the operands and is left unspecified.

Patch by Johnny Chen.

Modified:
    llvm/trunk/lib/Target/ARM/ARMInstrFormats.td
    llvm/trunk/lib/Target/ARM/ARMInstrNEON.td

Modified: llvm/trunk/lib/Target/ARM/ARMInstrFormats.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrFormats.td?rev=84572&r1=84571&r2=84572&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrFormats.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrFormats.td Mon Oct 19 19:38:19 2009
@@ -1290,6 +1290,20 @@
   let Inst{4} = op4;
 }
 
+// NEON 3 vector register with immediate.  This is only used for VEXT where
+// op11_8 represents the starting byte index of the extracted result in the
+// concatenation of the operands and is left unspecified.
+class N3VImm<bit op24, bit op23, bits<2> op21_20, bit op6, bit op4,
+             dag oops, dag iops, InstrItinClass itin,
+             string asm, string cstr, list<dag> pattern>
+  : NDataI<oops, iops, itin, asm, cstr, pattern> {
+  let Inst{24} = op24;
+  let Inst{23} = op23;
+  let Inst{21-20} = op21_20;
+  let Inst{6} = op6;
+  let Inst{4} = op4;
+}
+
 // NEON VMOVs between scalar and core registers.
 class NVLaneOp<bits<8> opcod1, bits<4> opcod2, bits<2> opcod3,
                dag oops, dag iops, Format f, InstrItinClass itin,

Modified: llvm/trunk/lib/Target/ARM/ARMInstrNEON.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrNEON.td?rev=84572&r1=84571&r2=84572&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrNEON.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrNEON.td Mon Oct 19 19:38:19 2009
@@ -2670,18 +2670,18 @@
 //   VEXT     : Vector Extract
 
 class VEXTd<string OpcodeStr, ValueType Ty>
-  : N3V<0,1,0b11,0b0000,0,0, (outs DPR:$dst),
-        (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
-        !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
-        [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
-                                      (Ty DPR:$rhs), imm:$index)))]>;
+  : N3VImm<0,1,0b11,0,0, (outs DPR:$dst),
+           (ins DPR:$lhs, DPR:$rhs, i32imm:$index), IIC_VEXTD,
+           !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
+           [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
+                                         (Ty DPR:$rhs), imm:$index)))]>;
 
 class VEXTq<string OpcodeStr, ValueType Ty>
-  : N3V<0,1,0b11,0b0000,1,0, (outs QPR:$dst),
-        (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
-        !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
-        [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
-                                      (Ty QPR:$rhs), imm:$index)))]>;
+  : N3VImm<0,1,0b11,1,0, (outs QPR:$dst),
+           (ins QPR:$lhs, QPR:$rhs, i32imm:$index), IIC_VEXTQ,
+           !strconcat(OpcodeStr, "\t$dst, $lhs, $rhs, $index"), "",
+           [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
+                                         (Ty QPR:$rhs), imm:$index)))]>;
 
 def VEXTd8  : VEXTd<"vext.8",  v8i8>;
 def VEXTd16 : VEXTd<"vext.16", v4i16>;




From grosbach at apple.com  Mon Oct 19 19:39:45 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Mon, 19 Oct 2009 17:39:45 -0700
Subject: [llvm-commits] ARM NEON patch for ARMInstrFormats.td
In-Reply-To: <300516B4-D2D1-4E59-8706-C6BF5DAAD31E@apple.com>
References: <2568D436-F301-49A2-ACBC-763032CD2D91@apple.com>
	<509F0CED-04F8-43C9-8A11-D425531CEC12@apple.com>
	<300516B4-D2D1-4E59-8706-C6BF5DAAD31E@apple.com>
Message-ID: <32632376-3F4B-418A-BCF6-EE94BDFE8412@apple.com>

That one's in, too, now. Thanks for the reminder. :)

-Jim

On Oct 19, 2009, at 5:33 PM, Johnny Chen wrote:

> No problem.
> BTW, is my previous patch concerning VEXT (Vector Extract) still  
> being reviewed? :-)
>
> On Oct 19, 2009, at 5:19 PM, Jim Grosbach wrote:
>
>> Applied as r84570.
>>
>> Thanks again for these patches.
>>
>> -Jim
>>
>> On Oct 19, 2009, at 4:46 PM, Johnny Chen wrote:
>>
>>> Hi,
>>>
>>> I manually edited the patch file to remove some hunks which  
>>> represent WIP
>>> to not over-specify some immediate fields.  But this patch should  
>>> be easy
>>> to review so I send it first.
>>>
>>> <ARM-NEON-NLdSt.patch>
>>>
>>>
>>> All the following NEON instructions are under-specified without  
>>> this patch:
>>>
>>>
>>>  VLD1d16         11110100________________________
>>>  VLD1d32         11110100________________________
>>>  VLD1d64         11110100________________________
>>>  VLD1d8          11110100________________________
>>>  VLD1df          11110100________________________
>>>  VLD1q16         11110100________________________
>>>  VLD1q32         11110100________________________
>>>  VLD1q64         11110100________________________
>>>  VLD1q8          11110100________________________
>>>  VLD1qf          11110100________________________
>>>  VLD2LNd16       11110100________________________
>>>  VLD2LNd32       11110100________________________
>>>  VLD2LNd8        11110100________________________
>>>  VLD2LNq16a      11110100________________________
>>>  VLD2LNq16b      11110100________________________
>>>  VLD2LNq32a      11110100________________________
>>>  VLD2LNq32b      11110100________________________
>>>  VLD2d16         11110100________________________
>>>  VLD2d32         11110100________________________
>>>  VLD2d64         11110100________________________
>>>  VLD2d8          11110100________________________
>>>  VLD2q16         11110100________________________
>>>  VLD2q32         11110100________________________
>>>  VLD2q8          11110100________________________
>>>  VLD3LNd16       11110100________________________
>>>  VLD3LNd32       11110100________________________
>>>  VLD3LNd8        11110100________________________
>>>  VLD3LNq16a      11110100________________________
>>>  VLD3LNq16b      11110100________________________
>>>  VLD3LNq32a      11110100________________________
>>>  VLD3LNq32b      11110100________________________
>>>  VLD3d16         11110100________________________
>>>  VLD3d32         11110100________________________
>>>  VLD3d64         11110100________________________
>>>  VLD3d8          11110100________________________
>>>  VLD3q16a        11110100________________________
>>>  VLD3q16b        11110100________________________
>>>  VLD3q32a        11110100________________________
>>>  VLD3q32b        11110100________________________
>>>  VLD3q8a         11110100________________________
>>>  VLD3q8b         11110100________________________
>>>  VLD4LNd16       11110100________________________
>>>  VLD4LNd32       11110100________________________
>>>  VLD4LNd8        11110100________________________
>>>  VLD4LNq16a      11110100________________________
>>>  VLD4LNq16b      11110100________________________
>>>  VLD4LNq32a      11110100________________________
>>>  VLD4LNq32b      11110100________________________
>>>  VLD4d16         11110100________________________
>>>  VLD4d32         11110100________________________
>>>  VLD4d64         11110100________________________
>>>  VLD4d8          11110100________________________
>>>  VLD4q16a        11110100________________________
>>>  VLD4q16b        11110100________________________
>>>  VLD4q32a        11110100________________________
>>>  VLD4q32b        11110100________________________
>>>  VLD4q8a         11110100________________________
>>>  VLD4q8b         11110100________________________
>>>  VST1d16         11110100________________________
>>>  VST1d32         11110100________________________
>>>  VST1d64         11110100________________________
>>>  VST1d8          11110100________________________
>>>  VST1df          11110100________________________
>>>  VST1q16         11110100________________________
>>>  VST1q32         11110100________________________
>>>  VST1q64         11110100________________________
>>>  VST1q8          11110100________________________
>>>  VST1qf          11110100________________________
>>>  VST2LNd16       11110100________________________
>>>  VST2LNd32       11110100________________________
>>>  VST2LNd8        11110100________________________
>>>  VST2LNq16a      11110100________________________
>>>  VST2LNq16b      11110100________________________
>>>  VST2LNq32a      11110100________________________
>>>  VST2LNq32b      11110100________________________
>>>  VST2d16         11110100________________________
>>>  VST2d32         11110100________________________
>>>  VST2d64         11110100________________________
>>>  VST2d8          11110100________________________
>>>  VST2q16         11110100________________________
>>>  VST2q32         11110100________________________
>>>  VST2q8          11110100________________________
>>>  VST3LNd16       11110100________________________
>>>  VST3LNd32       11110100________________________
>>>  VST3LNd8        11110100________________________
>>>  VST3LNq16a      11110100________________________
>>>  VST3LNq16b      11110100________________________
>>>  VST3LNq32a      11110100________________________
>>>  VST3LNq32b      11110100________________________
>>>  VST3d16         11110100________________________
>>>  VST3d32         11110100________________________
>>>  VST3d64         11110100________________________
>>>  VST3d8          11110100________________________
>>>  VST3q16a        11110100________________________
>>>  VST3q16b        11110100________________________
>>>  VST3q32a        11110100________________________
>>>  VST3q32b        11110100________________________
>>>  VST3q8a         11110100________________________
>>>  VST3q8b         11110100________________________
>>>  VST4LNd16       11110100________________________
>>>  VST4LNd32       11110100________________________
>>>  VST4LNd8        11110100________________________
>>>  VST4LNq16a      11110100________________________
>>>  VST4LNq16b      11110100________________________
>>>  VST4LNq32a      11110100________________________
>>>  VST4LNq32b      11110100________________________
>>>  VST4d16         11110100________________________
>>>  VST4d32         11110100________________________
>>>  VST4d64         11110100________________________
>>>  VST4d8          11110100________________________
>>>  VST4q16a        11110100________________________
>>>  VST4q16b        11110100________________________
>>>  VST4q32a        11110100________________________
>>>  VST4q32b        11110100________________________
>>>  VST4q8a         11110100________________________
>>>  VST4q8b         11110100________________________
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>



From sabre at nondot.org  Mon Oct 19 19:40:56 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 00:40:56 -0000
Subject: [llvm-commits] [llvm] r84573 - in /llvm/trunk/lib/Target/ARM:
 ARMInstrInfo.td AsmPrinter/ARMAsmPrinter.cpp AsmPrinter/ARMInstPrinter.cpp
 AsmPrinter/ARMInstPrinter.h
Message-ID: <200910200040.n9K0euJp023169@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 19:40:56 2009
New Revision: 84573

URL: http://llvm.org/viewvc/llvm-project?rev=84573&view=rev
Log:
implement printSORegOperand, add lowering for the nasty and despicable MOVi2pieces :)

Modified:
    llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrInfo.td?rev=84573&r1=84572&r2=84573&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrInfo.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrInfo.td Mon Oct 19 19:40:56 2009
@@ -1582,8 +1582,9 @@
                     (so_imm2part_2 imm:$RHS))>;
 
 // 32-bit immediate using movw + movt.
-// This is a single pseudo instruction to make it re-materializable. Remove
-// when we can do generalized remat.
+// This is a single pseudo instruction, the benefit is that it can be remat'd
+// as a single unit instead of having to handle reg inputs.
+// FIXME: Remove this when we can do generalized remat.
 let isReMaterializable = 1 in
 def MOVi32imm : AI1x2<(outs GPR:$dst), (ins i32imm:$src), Pseudo, IIC_iMOVi,
                      "movw", " $dst, ${src:lo16}\n\tmovt${p} $dst, ${src:hi16}",

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84573&r1=84572&r2=84573&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 19:40:56 2009
@@ -1375,6 +1375,45 @@
     
     return;
   }
+  case ARM::MOVi2pieces: { // FIXME: Remove asmstring from td file.
+    // This is a hack that lowers as a two instruction sequence.
+    unsigned DstReg = MI->getOperand(0).getReg();
+    unsigned ImmVal = (unsigned)MI->getOperand(1).getImm();
+
+    unsigned SOImmValV1 = ARM_AM::getSOImmTwoPartFirst(ImmVal);
+    unsigned SOImmValV2 = ARM_AM::getSOImmTwoPartSecond(ImmVal);
+    
+    {
+      MCInst TmpInst;
+      TmpInst.setOpcode(ARM::MOVi);
+      TmpInst.addOperand(MCOperand::CreateReg(DstReg));
+      TmpInst.addOperand(MCOperand::CreateImm(SOImmValV1));
+      
+      // Predicate.
+      TmpInst.addOperand(MCOperand::CreateImm(MI->getOperand(2).getImm()));
+      TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(3).getReg()));
+      printMCInst(&TmpInst);
+      O << '\n';
+    }
+
+    {
+      MCInst TmpInst;
+      TmpInst.setOpcode(ARM::ORRri);
+      TmpInst.addOperand(MCOperand::CreateReg(DstReg));     // dstreg
+      TmpInst.addOperand(MCOperand::CreateReg(DstReg));     // inreg
+      TmpInst.addOperand(MCOperand::CreateImm(SOImmValV2)); // so_imm
+      // Predicate.
+      TmpInst.addOperand(MCOperand::CreateImm(MI->getOperand(2).getImm()));
+      TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(3).getReg()));
+      
+      TmpInst.addOperand(MCOperand::CreateReg(0));          // cc_out
+      printMCInst(&TmpInst);
+    }
+    return; 
+  }
+      // FIXME: Also MOVi32imm.
+
+      
   }
       
   MCInst TmpInst;

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84573&r1=84572&r2=84573&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 19:40:56 2009
@@ -79,6 +79,37 @@
   printSOImm(O, MO.getImm(), VerboseAsm, &MAI);
 }
 
+/// printSOImm2PartOperand - SOImm is broken into two pieces using a 'mov'
+/// followed by an 'orr' to materialize.
+void ARMInstPrinter::printSOImm2PartOperand(const MCInst *MI, unsigned OpNum) {
+  // FIXME: REMOVE this method.
+  abort();
+}
+
+// so_reg is a 4-operand unit corresponding to register forms of the A5.1
+// "Addressing Mode 1 - Data-processing operands" forms.  This includes:
+//    REG 0   0           - e.g. R5
+//    REG REG 0,SH_OPC    - e.g. R5, ROR R3
+//    REG 0   IMM,SH_OPC  - e.g. R5, LSL #3
+void ARMInstPrinter::printSORegOperand(const MCInst *MI, unsigned OpNum) {
+  const MCOperand &MO1 = MI->getOperand(OpNum);
+  const MCOperand &MO2 = MI->getOperand(OpNum+1);
+  const MCOperand &MO3 = MI->getOperand(OpNum+2);
+  
+  O << getRegisterName(MO1.getReg());
+  
+  // Print the shift opc.
+  O << ", "
+    << ARM_AM::getShiftOpcStr(ARM_AM::getSORegShOp(MO3.getImm()))
+    << ' ';
+  
+  if (MO2.getReg()) {
+    O << getRegisterName(MO2.getReg());
+    assert(ARM_AM::getSORegOffset(MO3.getImm()) == 0);
+  } else {
+    O << "#" << ARM_AM::getSORegOffset(MO3.getImm());
+  }
+}
 
 
 void ARMInstPrinter::printAddrMode2Operand(const MCInst *MI, unsigned Op) {

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84573&r1=84572&r2=84573&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 19:40:56 2009
@@ -36,10 +36,9 @@
                     const char *Modifier = 0);
     
   void printSOImmOperand(const MCInst *MI, unsigned OpNum);
+  void printSOImm2PartOperand(const MCInst *MI, unsigned OpNum);
   
-  
-  void printSOImm2PartOperand(const MCInst *MI, unsigned OpNum) {}
-  void printSORegOperand(const MCInst *MI, unsigned OpNum) {}
+  void printSORegOperand(const MCInst *MI, unsigned OpNum);
   void printAddrMode2Operand(const MCInst *MI, unsigned OpNum);
   void printAddrMode2OffsetOperand(const MCInst *MI, unsigned OpNum) {}
   void printAddrMode3Operand(const MCInst *MI, unsigned OpNum) {}




From grosbach at apple.com  Mon Oct 19 19:40:33 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Mon, 19 Oct 2009 17:40:33 -0700
Subject: [llvm-commits] Patch for encodings of VEXT (Vector Extract)
In-Reply-To: <AD31FE2C-054B-424B-A0DC-27C3F3A6BC4A@apple.com>
References: <AD31FE2C-054B-424B-A0DC-27C3F3A6BC4A@apple.com>
Message-ID: <6BF01A53-6F79-40B9-B0F7-290B30272867@apple.com>

Checked in as 84572.

Thanks!

On Oct 16, 2009, at 11:17 AM, Johnny Chen wrote:

> Hi,
>
> Refs: A8-598.
> Leave Inst{11-8}, which represents the starting byte index of the  
> extracted result in the
> concatenation of the operands and is left unspecified.
>
> <ARMInstrNEON.patch5>_______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From sabre at nondot.org  Mon Oct 19 19:42:49 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 00:42:49 -0000
Subject: [llvm-commits] [llvm] r84575 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMInstPrinter.cpp ARMInstPrinter.h
Message-ID: <200910200042.n9K0gn0o023291@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 19:42:49 2009
New Revision: 84575

URL: http://llvm.org/viewvc/llvm-project?rev=84575&view=rev
Log:
add support for instruction predicates.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84575&r1=84574&r2=84575&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 19:42:49 2009
@@ -12,14 +12,13 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "asm-printer"
+#include "ARM.h" // FIXME: FACTOR ENUMS BETTER.
 #include "ARMInstPrinter.h"
 #include "ARMAddressingModes.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/Support/raw_ostream.h"
-#include "ARMGenInstrNames.inc"
-#include "ARMGenRegisterNames.inc"
 using namespace llvm;
 
 // Include the auto-generated portion of the assembly writer.
@@ -181,6 +180,12 @@
   O << "}";
 }
 
+void ARMInstPrinter::printPredicateOperand(const MCInst *MI, unsigned OpNum) {
+  ARMCC::CondCodes CC = (ARMCC::CondCodes)MI->getOperand(OpNum).getImm();
+  if (CC != ARMCC::AL)
+    O << ARMCondCodeToString(CC);
+}
+
 
 void ARMInstPrinter::printCPInstOperand(const MCInst *MI, unsigned OpNum,
                                         const char *Modifier) {

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84575&r1=84574&r2=84575&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 19:42:49 2009
@@ -68,7 +68,7 @@
   void printT2AddrModeImm8OffsetOperand(const MCInst *MI, unsigned OpNum) {}
   void printT2AddrModeSoRegOperand(const MCInst *MI, unsigned OpNum) {}
   
-  void printPredicateOperand(const MCInst *MI, unsigned OpNum) {}
+  void printPredicateOperand(const MCInst *MI, unsigned OpNum);
   void printSBitModifierOperand(const MCInst *MI, unsigned OpNum) {}
   void printRegisterList(const MCInst *MI, unsigned OpNum);
   void printCPInstOperand(const MCInst *MI, unsigned OpNum,




From sabre at nondot.org  Mon Oct 19 19:46:11 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 00:46:11 -0000
Subject: [llvm-commits] [llvm] r84577 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMAsmPrinter.cpp ARMInstPrinter.cpp
 ARMInstPrinter.h
Message-ID: <200910200046.n9K0kBMO023642@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 19:46:11 2009
New Revision: 84577

URL: http://llvm.org/viewvc/llvm-project?rev=84577&view=rev
Log:
add supprort for the 'sbit' operand, MOVi apparently has one.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84577&r1=84576&r2=84577&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 19:46:11 2009
@@ -1392,6 +1392,8 @@
       // Predicate.
       TmpInst.addOperand(MCOperand::CreateImm(MI->getOperand(2).getImm()));
       TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(3).getReg()));
+
+      TmpInst.addOperand(MCOperand::CreateReg(0));          // cc_out
       printMCInst(&TmpInst);
       O << '\n';
     }

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84577&r1=84576&r2=84577&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Mon Oct 19 19:46:11 2009
@@ -186,6 +186,14 @@
     O << ARMCondCodeToString(CC);
 }
 
+void ARMInstPrinter::printSBitModifierOperand(const MCInst *MI, unsigned OpNum){
+  if (unsigned Reg = MI->getOperand(OpNum).getReg()) {
+    assert(Reg == ARM::CPSR && "Expect ARM CPSR register!");
+    O << 's';
+  }
+}
+
+
 
 void ARMInstPrinter::printCPInstOperand(const MCInst *MI, unsigned OpNum,
                                         const char *Modifier) {

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84577&r1=84576&r2=84577&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Mon Oct 19 19:46:11 2009
@@ -69,7 +69,7 @@
   void printT2AddrModeSoRegOperand(const MCInst *MI, unsigned OpNum) {}
   
   void printPredicateOperand(const MCInst *MI, unsigned OpNum);
-  void printSBitModifierOperand(const MCInst *MI, unsigned OpNum) {}
+  void printSBitModifierOperand(const MCInst *MI, unsigned OpNum);
   void printRegisterList(const MCInst *MI, unsigned OpNum);
   void printCPInstOperand(const MCInst *MI, unsigned OpNum,
                           const char *Modifier);




From sabre at nondot.org  Mon Oct 19 19:52:47 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 00:52:47 -0000
Subject: [llvm-commits] [llvm] r84579 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMAsmPrinter.cpp ARMMCInstLower.cpp
 ARMMCInstLower.h
Message-ID: <200910200052.n9K0qlQE023974@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 19:52:47 2009
New Revision: 84579

URL: http://llvm.org/viewvc/llvm-project?rev=84579&view=rev
Log:
get fancy: support basic block operands.  Yay for jumps.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84579&r1=84578&r2=84579&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 19:52:47 2009
@@ -1311,7 +1311,7 @@
 //===----------------------------------------------------------------------===//
 
 void ARMAsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI) {
-  ARMMCInstLower MCInstLowering(OutContext, *Mang, getFunctionNumber(), *MAI);
+  ARMMCInstLower MCInstLowering(OutContext, *Mang, *this);
   switch (MI->getOpcode()) {
   default: break;
   case TargetInstrInfo::DBG_LABEL:

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp?rev=84579&r1=84578&r2=84579&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp Mon Oct 19 19:52:47 2009
@@ -14,7 +14,8 @@
 
 #include "ARMMCInstLower.h"
 //#include "llvm/CodeGen/MachineModuleInfoImpls.h"
-#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/AsmPrinter.h"
+#include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCExpr.h"
@@ -57,8 +58,8 @@
 MCSymbol *ARMMCInstLower::
 GetJumpTableSymbol(const MachineOperand &MO) const {
   SmallString<256> Name;
-  raw_svector_ostream(Name) << MAI.getPrivateGlobalPrefix() << "JTI"
-    << CurFunctionNumber << '_' << MO.getIndex();
+  raw_svector_ostream(Name) << Printer.MAI->getPrivateGlobalPrefix() << "JTI"
+    << Printer.getFunctionNumber() << '_' << MO.getIndex();
   
 #if 0
   switch (MO.getTargetFlags()) {
@@ -73,8 +74,8 @@
 MCSymbol *ARMMCInstLower::
 GetConstantPoolIndexSymbol(const MachineOperand &MO) const {
   SmallString<256> Name;
-  raw_svector_ostream(Name) << MAI.getPrivateGlobalPrefix() << "CPI"
-    << CurFunctionNumber << '_' << MO.getIndex();
+  raw_svector_ostream(Name) << Printer.MAI->getPrivateGlobalPrefix() << "CPI"
+    << Printer.getFunctionNumber() << '_' << MO.getIndex();
   
 #if 0
   switch (MO.getTargetFlags()) {
@@ -125,12 +126,10 @@
     case MachineOperand::MO_Immediate:
       MCOp = MCOperand::CreateImm(MO.getImm());
       break;
-#if 0
     case MachineOperand::MO_MachineBasicBlock:
       MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
-                       AsmPrinter.GetMBBSymbol(MO.getMBB()->getNumber()), Ctx));
+                       Printer.GetMBBSymbol(MO.getMBB()->getNumber()), Ctx));
       break;
-#endif
     case MachineOperand::MO_GlobalAddress:
       MCOp = LowerSymbolOperand(MO, GetGlobalAddressSymbol(MO));
       break;

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h?rev=84579&r1=84578&r2=84579&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h Mon Oct 19 19:52:47 2009
@@ -13,6 +13,7 @@
 #include "llvm/Support/Compiler.h"
 
 namespace llvm {
+  class AsmPrinter;
   class MCAsmInfo;
   class MCContext;
   class MCInst;
@@ -28,15 +29,12 @@
 class VISIBILITY_HIDDEN ARMMCInstLower {
   MCContext &Ctx;
   Mangler &Mang;
-  
-  const unsigned CurFunctionNumber;
-  const MCAsmInfo &MAI;
+  AsmPrinter &Printer;
 
   //const ARMSubtarget &getSubtarget() const;
 public:
-  ARMMCInstLower(MCContext &ctx, Mangler &mang, unsigned FuncNum,
-                 const MCAsmInfo &mai)
-    : Ctx(ctx), Mang(mang), CurFunctionNumber(FuncNum), MAI(mai) {}
+  ARMMCInstLower(MCContext &ctx, Mangler &mang, AsmPrinter &printer)
+    : Ctx(ctx), Mang(mang), Printer(printer) {}
   
   void Lower(const MachineInstr *MI, MCInst &OutMI) const;
 




From sabre at nondot.org  Mon Oct 19 19:56:16 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 00:56:16 -0000
Subject: [llvm-commits] [llvm] r84582 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMMCInstLower.cpp ARMMCInstLower.h
Message-ID: <200910200056.n9K0uGHO024138@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 19:56:16 2009
New Revision: 84582

URL: http://llvm.org/viewvc/llvm-project?rev=84582&view=rev
Log:
add support for external symbols.  The mc instprinter  can now handle
reasonable code like Codegen/ARM/2009-02-27-SpillerBug.ll, producing 
identical output except for superior formatting of constant pool entries.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp?rev=84582&r1=84581&r2=84582&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.cpp Mon Oct 19 19:56:16 2009
@@ -54,6 +54,22 @@
   return Ctx.GetOrCreateSymbol(Name.str());
 }
 
+MCSymbol *ARMMCInstLower::
+GetExternalSymbolSymbol(const MachineOperand &MO) const {
+  SmallString<128> Name;
+  Name += Printer.MAI->getGlobalPrefix();
+  Name += MO.getSymbolName();
+  
+  // FIXME: HANDLE PLT references how??
+  switch (MO.getTargetFlags()) {
+  default: assert(0 && "Unknown target flag on GV operand");
+  case 0: break;
+  }
+  
+  return Ctx.GetOrCreateSymbol(Name.str());
+}
+
+
 
 MCSymbol *ARMMCInstLower::
 GetJumpTableSymbol(const MachineOperand &MO) const {
@@ -133,11 +149,9 @@
     case MachineOperand::MO_GlobalAddress:
       MCOp = LowerSymbolOperand(MO, GetGlobalAddressSymbol(MO));
       break;
-#if 0
     case MachineOperand::MO_ExternalSymbol:
       MCOp = LowerSymbolOperand(MO, GetExternalSymbolSymbol(MO));
       break;
-#endif
     case MachineOperand::MO_JumpTableIndex:
       MCOp = LowerSymbolOperand(MO, GetJumpTableSymbol(MO));
       break;

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h?rev=84582&r1=84581&r2=84582&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMMCInstLower.h Mon Oct 19 19:56:16 2009
@@ -40,7 +40,7 @@
 
   //MCSymbol *GetPICBaseSymbol() const;
   MCSymbol *GetGlobalAddressSymbol(const MachineOperand &MO) const;
-  //MCSymbol *GetExternalSymbolSymbol(const MachineOperand &MO) const;
+  MCSymbol *GetExternalSymbolSymbol(const MachineOperand &MO) const;
   MCSymbol *GetJumpTableSymbol(const MachineOperand &MO) const;
   MCSymbol *GetConstantPoolIndexSymbol(const MachineOperand &MO) const;
   MCOperand LowerSymbolOperand(const MachineOperand &MO, MCSymbol *Sym) const;




From sabre at nondot.org  Mon Oct 19 20:11:38 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 01:11:38 -0000
Subject: [llvm-commits] [llvm] r84583 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910200111.n9K1BcjJ024853@zion.cs.uiuc.edu>

Author: lattner
Date: Mon Oct 19 20:11:37 2009
New Revision: 84583

URL: http://llvm.org/viewvc/llvm-project?rev=84583&view=rev
Log:
lower ARM::MOVi32imm properly.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84583&r1=84582&r2=84583&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Mon Oct 19 20:11:37 2009
@@ -1413,9 +1413,43 @@
     }
     return; 
   }
-      // FIXME: Also MOVi32imm.
-
+  case ARM::MOVi32imm: { // FIXME: Remove asmstring from td file.
+    // This is a hack that lowers as a two instruction sequence.
+    unsigned DstReg = MI->getOperand(0).getReg();
+    unsigned ImmVal = (unsigned)MI->getOperand(1).getImm();
+    
+    {
+      MCInst TmpInst;
+      TmpInst.setOpcode(ARM::MOVi16);
+      TmpInst.addOperand(MCOperand::CreateReg(DstReg));         // dstreg
+      TmpInst.addOperand(MCOperand::CreateImm(ImmVal & 65535)); // lower16(imm)
       
+      // Predicate.
+      TmpInst.addOperand(MCOperand::CreateImm(MI->getOperand(2).getImm()));
+      TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(3).getReg()));
+      
+      printMCInst(&TmpInst);
+      O << '\n';
+    }
+    
+    {
+      MCInst TmpInst;
+      TmpInst.setOpcode(ARM::MOVTi16);
+      TmpInst.addOperand(MCOperand::CreateReg(DstReg));         // dstreg
+      TmpInst.addOperand(MCOperand::CreateReg(DstReg));         // srcreg
+      TmpInst.addOperand(MCOperand::CreateImm(ImmVal >> 16));   // upper16(imm)
+      
+      // Predicate.
+      TmpInst.addOperand(MCOperand::CreateImm(MI->getOperand(2).getImm()));
+      TmpInst.addOperand(MCOperand::CreateReg(MI->getOperand(3).getReg()));
+      
+      printMCInst(&TmpInst);
+    }
+    
+    return;
+  }
+      
+  // FIXME: Handle t2MOVi32imm also.
   }
       
   MCInst TmpInst;




From grosbach at apple.com  Mon Oct 19 20:26:59 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Tue, 20 Oct 2009 01:26:59 -0000
Subject: [llvm-commits] [llvm] r84585 - in /llvm/trunk/lib:
 CodeGen/PrologEpilogInserter.cpp Target/ARM/ARMBaseRegisterInfo.cpp
 Target/ARM/ARMBaseRegisterInfo.h
Message-ID: <200910200126.n9K1QxOY025311@zion.cs.uiuc.edu>

Author: grosbach
Date: Mon Oct 19 20:26:58 2009
New Revision: 84585

URL: http://llvm.org/viewvc/llvm-project?rev=84585&view=rev
Log:
Enable post-pass frame index register scavenging for ARM and Thumb2

Modified:
    llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp
    llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
    llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.h

Modified: llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp?rev=84585&r1=84584&r2=84585&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp (original)
+++ llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp Mon Oct 19 20:26:58 2009
@@ -785,11 +785,6 @@
           if (Reg == 0)
             continue;
           if (!TargetRegisterInfo::isVirtualRegister(Reg)) {
-            // If we have an active scavenged register, we shouldn't be
-            // seeing any references to it.
-            assert (Reg != CurrentScratchReg
-                    && "overlapping use of scavenged frame index register!");
-
             // If we have a previous scratch reg, check and see if anything
             // here kills whatever value is in there.
             if (Reg == PrevScratchReg) {

Modified: llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp?rev=84585&r1=84584&r2=84585&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp Mon Oct 19 20:26:58 2009
@@ -934,6 +934,10 @@
 requiresRegisterScavenging(const MachineFunction &MF) const {
   return true;
 }
+bool ARMBaseRegisterInfo::
+requiresFrameIndexScavenging(const MachineFunction &MF) const {
+  return true;
+}
 
 // hasReservedCallFrame - Under normal circumstances, when a frame pointer is
 // not required, we reserve argument space for call sites in the function
@@ -1010,17 +1014,6 @@
   MBB.erase(I);
 }
 
-/// findScratchRegister - Find a 'free' ARM register. If register scavenger
-/// is not being used, R12 is available. Otherwise, try for a call-clobbered
-/// register first and then a spilled callee-saved register if that fails.
-static
-unsigned findScratchRegister(RegScavenger *RS, const TargetRegisterClass *RC,
-                             ARMFunctionInfo *AFI) {
-  unsigned Reg = RS ? RS->FindUnusedReg(RC) : (unsigned) ARM::R12;
-  assert(!AFI->isThumb1OnlyFunction());
-  return Reg;
-}
-
 unsigned
 ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
                                          int SPAdj, int *Value,
@@ -1075,14 +1068,7 @@
           (MI.getDesc().TSFlags & ARMII::AddrModeMask) == ARMII::AddrMode4) &&
          "This code isn't needed if offset already handled!");
 
-  // Insert a set of r12 with the full address: r12 = sp + offset
-  // If the offset we have is too large to fit into the instruction, we need
-  // to form it with a series of ADDri's.  Do this by taking 8-bit chunks
-  // out of 'Offset'.
-  unsigned ScratchReg = findScratchRegister(RS, ARM::GPRRegisterClass, AFI);
-  if (ScratchReg == 0)
-    // No register is "free". Scavenge a register.
-    ScratchReg = RS->scavengeRegister(ARM::GPRRegisterClass, II, SPAdj);
+  unsigned ScratchReg = 0;
   int PIdx = MI.findFirstPredOperandIdx();
   ARMCC::CondCodes Pred = (PIdx == -1)
     ? ARMCC::AL : (ARMCC::CondCodes)MI.getOperand(PIdx).getImm();
@@ -1091,6 +1077,8 @@
     // Must be addrmode4.
     MI.getOperand(i).ChangeToRegister(FrameReg, false, false, false);
   else {
+    ScratchReg = MF.getRegInfo().createVirtualRegister(ARM::GPRRegisterClass);
+    *Value = Offset;
     if (!AFI->isThumbFunction())
       emitARMRegPlusImmediate(MBB, II, MI.getDebugLoc(), ScratchReg, FrameReg,
                               Offset, Pred, PredReg, TII);
@@ -1101,7 +1089,7 @@
     }
     MI.getOperand(i).ChangeToRegister(ScratchReg, false, false, true);
   }
-  return 0;
+  return ScratchReg;
 }
 
 /// Move iterator pass the next bunch of callee save load / store ops for

Modified: llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.h?rev=84585&r1=84584&r2=84585&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.h (original)
+++ llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.h Mon Oct 19 20:26:58 2009
@@ -122,6 +122,8 @@
 
   virtual bool requiresRegisterScavenging(const MachineFunction &MF) const;
 
+  virtual bool requiresFrameIndexScavenging(const MachineFunction &MF) const;
+
   virtual bool hasReservedCallFrame(MachineFunction &MF) const;
 
   virtual void eliminateCallFramePseudoInstr(MachineFunction &MF,




From evan.cheng at apple.com  Mon Oct 19 20:31:09 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Tue, 20 Oct 2009 01:31:09 -0000
Subject: [llvm-commits] [llvm] r84586 - in /llvm/trunk:
 lib/CodeGen/LiveIntervalAnalysis.cpp
 test/CodeGen/X86/2009-10-19-EmergencySpill.ll
Message-ID: <200910200131.n9K1V9fY025440@zion.cs.uiuc.edu>

Author: evancheng
Date: Mon Oct 19 20:31:09 2009
New Revision: 84586

URL: http://llvm.org/viewvc/llvm-project?rev=84586&view=rev
Log:
If the physical register being spilled does not have an interval, spill its sub-registers instead.

Added:
    llvm/trunk/test/CodeGen/X86/2009-10-19-EmergencySpill.ll
Modified:
    llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp

Modified: llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp?rev=84586&r1=84585&r2=84586&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp (original)
+++ llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp Mon Oct 19 20:31:09 2009
@@ -2603,7 +2603,19 @@
            tri_->isSuperRegister(*AS, SpillReg));
 
   bool Cut = false;
-  LiveInterval &pli = getInterval(SpillReg);
+  SmallVector<unsigned, 4> PRegs;
+  if (hasInterval(SpillReg))
+    PRegs.push_back(SpillReg);
+  else {
+    SmallSet<unsigned, 4> Added;
+    for (const unsigned* AS = tri_->getSubRegisters(SpillReg); *AS; ++AS)
+      if (Added.insert(*AS) && hasInterval(*AS)) {
+        PRegs.push_back(*AS);
+        for (const unsigned* ASS = tri_->getSubRegisters(*AS); *ASS; ++ASS)
+          Added.insert(*ASS);
+      }
+  }
+
   SmallPtrSet<MachineInstr*, 8> SeenMIs;
   for (MachineRegisterInfo::reg_iterator I = mri_->reg_begin(li.reg),
          E = mri_->reg_end(); I != E; ++I) {
@@ -2613,8 +2625,12 @@
       continue;
     SeenMIs.insert(MI);
     LiveIndex Index = getInstructionIndex(MI);
-    if (pli.liveAt(Index)) {
-      vrm.addEmergencySpill(SpillReg, MI);
+    for (unsigned i = 0, e = PRegs.size(); i != e; ++i) {
+      unsigned PReg = PRegs[i];
+      LiveInterval &pli = getInterval(PReg);
+      if (!pli.liveAt(Index))
+        continue;
+      vrm.addEmergencySpill(PReg, MI);
       LiveIndex StartIdx = getLoadIndex(Index);
       LiveIndex EndIdx = getNextSlot(getStoreIndex(Index));
       if (pli.isInOneLiveRange(StartIdx, EndIdx)) {
@@ -2626,12 +2642,12 @@
         Msg << "Ran out of registers during register allocation!";
         if (MI->getOpcode() == TargetInstrInfo::INLINEASM) {
           Msg << "\nPlease check your inline asm statement for invalid "
-               << "constraints:\n";
+              << "constraints:\n";
           MI->print(Msg, tm_);
         }
         llvm_report_error(Msg.str());
       }
-      for (const unsigned* AS = tri_->getSubRegisters(SpillReg); *AS; ++AS) {
+      for (const unsigned* AS = tri_->getSubRegisters(PReg); *AS; ++AS) {
         if (!hasInterval(*AS))
           continue;
         LiveInterval &spli = getInterval(*AS);

Added: llvm/trunk/test/CodeGen/X86/2009-10-19-EmergencySpill.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-10-19-EmergencySpill.ll?rev=84586&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/X86/2009-10-19-EmergencySpill.ll (added)
+++ llvm/trunk/test/CodeGen/X86/2009-10-19-EmergencySpill.ll Mon Oct 19 20:31:09 2009
@@ -0,0 +1,54 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -disable-fp-elim
+; rdar://7291624
+
+%union.RtreeCoord = type { float }
+%struct.RtreeCell = type { i64, [10 x %union.RtreeCoord] }
+%struct.Rtree = type { i32, i32*, i32, i32, i32, i32, i8*, i8* }
+%struct.RtreeNode = type { i32*, i64, i32, i32, i8*, i32* }
+
+define fastcc void @nodeOverwriteCell(%struct.Rtree* nocapture %pRtree, %struct.RtreeNode* nocapture %pNode, %struct.RtreeCell* nocapture %pCell, i32 %iCell) nounwind ssp {
+entry:
+  %0 = load i8** undef, align 8                   ; <i8*> [#uses=2]
+  %1 = load i32* undef, align 8                   ; <i32> [#uses=1]
+  %2 = mul i32 %1, %iCell                         ; <i32> [#uses=1]
+  %3 = add nsw i32 %2, 4                          ; <i32> [#uses=1]
+  %4 = sext i32 %3 to i64                         ; <i64> [#uses=2]
+  %5 = load i64* null, align 8                    ; <i64> [#uses=2]
+  %6 = lshr i64 %5, 48                            ; <i64> [#uses=1]
+  %7 = trunc i64 %6 to i8                         ; <i8> [#uses=1]
+  store i8 %7, i8* undef, align 1
+  %8 = lshr i64 %5, 8                             ; <i64> [#uses=1]
+  %9 = trunc i64 %8 to i8                         ; <i8> [#uses=1]
+  %.sum4 = add i64 %4, 6                          ; <i64> [#uses=1]
+  %10 = getelementptr inbounds i8* %0, i64 %.sum4 ; <i8*> [#uses=1]
+  store i8 %9, i8* %10, align 1
+  %11 = getelementptr inbounds %struct.Rtree* %pRtree, i64 0, i32 3 ; <i32*> [#uses=1]
+  br i1 undef, label %bb.nph, label %bb2
+
+bb.nph:                                           ; preds = %entry
+  %tmp25 = add i64 %4, 11                         ; <i64> [#uses=1]
+  br label %bb
+
+bb:                                               ; preds = %bb, %bb.nph
+  %indvar = phi i64 [ 0, %bb.nph ], [ %indvar.next, %bb ] ; <i64> [#uses=3]
+  %scevgep = getelementptr %struct.RtreeCell* %pCell, i64 0, i32 1, i64 %indvar ; <%union.RtreeCoord*> [#uses=1]
+  %scevgep12 = bitcast %union.RtreeCoord* %scevgep to i32* ; <i32*> [#uses=1]
+  %tmp = shl i64 %indvar, 2                       ; <i64> [#uses=1]
+  %tmp26 = add i64 %tmp, %tmp25                   ; <i64> [#uses=1]
+  %scevgep27 = getelementptr i8* %0, i64 %tmp26   ; <i8*> [#uses=1]
+  %12 = load i32* %scevgep12, align 4             ; <i32> [#uses=1]
+  %13 = lshr i32 %12, 24                          ; <i32> [#uses=1]
+  %14 = trunc i32 %13 to i8                       ; <i8> [#uses=1]
+  store i8 %14, i8* undef, align 1
+  store i8 undef, i8* %scevgep27, align 1
+  %15 = load i32* %11, align 4                    ; <i32> [#uses=1]
+  %16 = shl i32 %15, 1                            ; <i32> [#uses=1]
+  %17 = icmp sgt i32 %16, undef                   ; <i1> [#uses=1]
+  %indvar.next = add i64 %indvar, 1               ; <i64> [#uses=1]
+  br i1 %17, label %bb, label %bb2
+
+bb2:                                              ; preds = %bb, %entry
+  %18 = getelementptr inbounds %struct.RtreeNode* %pNode, i64 0, i32 3 ; <i32*> [#uses=1]
+  store i32 1, i32* %18, align 4
+  ret void
+}




From grosbach at apple.com  Mon Oct 19 20:32:47 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Tue, 20 Oct 2009 01:32:47 -0000
Subject: [llvm-commits] [llvm] r84587 - in /llvm/trunk/lib/Target/ARM:
 Thumb1RegisterInfo.cpp Thumb1RegisterInfo.h
Message-ID: <200910200132.n9K1Wl5M025501@zion.cs.uiuc.edu>

Author: grosbach
Date: Mon Oct 19 20:32:47 2009
New Revision: 84587

URL: http://llvm.org/viewvc/llvm-project?rev=84587&view=rev
Log:
Now that all ARM subtargets use frame index scavenging, the Thumb1 requires*
functions are not needed.

Modified:
    llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
    llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h

Modified: llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp?rev=84587&r1=84586&r2=84587&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp Mon Oct 19 20:32:47 2009
@@ -77,18 +77,6 @@
   return TargetRegisterInfo::getPhysicalRegisterRegClass(Reg, VT);
 }
 
-bool
-Thumb1RegisterInfo::requiresRegisterScavenging(const MachineFunction &MF) const {
-  return true;
-}
-
-bool
-Thumb1RegisterInfo::requiresFrameIndexScavenging(const MachineFunction &MF)
-  const {
-  return true;
-}
-
-
 bool Thumb1RegisterInfo::hasReservedCallFrame(MachineFunction &MF) const {
   const MachineFrameInfo *FFI = MF.getFrameInfo();
   unsigned CFSize = FFI->getMaxCallFrameSize();

Modified: llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h?rev=84587&r1=84586&r2=84587&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h (original)
+++ llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h Mon Oct 19 20:32:47 2009
@@ -40,9 +40,6 @@
   const TargetRegisterClass *
     getPhysicalRegisterRegClass(unsigned Reg, EVT VT = MVT::Other) const;
 
-  bool requiresRegisterScavenging(const MachineFunction &MF) const;
-  bool requiresFrameIndexScavenging(const MachineFunction &MF) const;
-
   bool hasReservedCallFrame(MachineFunction &MF) const;
 
   void eliminateCallFramePseudoInstr(MachineFunction &MF,




From isanbard at gmail.com  Mon Oct 19 20:40:14 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Tue, 20 Oct 2009 01:40:14 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84588 - in /llvm-gcc-4.2/trunk:
 include/libiberty.h libiberty/make-relative-prefix.c
Message-ID: <200910200140.n9K1eEkJ025737@zion.cs.uiuc.edu>

Author: void
Date: Mon Oct 19 20:40:14 2009
New Revision: 84588

URL: http://llvm.org/viewvc/llvm-project?rev=84588&view=rev
Log:
Add missing LLVM markers.

Modified:
    llvm-gcc-4.2/trunk/include/libiberty.h
    llvm-gcc-4.2/trunk/libiberty/make-relative-prefix.c

Modified: llvm-gcc-4.2/trunk/include/libiberty.h
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/include/libiberty.h?rev=84588&r1=84587&r2=84588&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/include/libiberty.h (original)
+++ llvm-gcc-4.2/trunk/include/libiberty.h Mon Oct 19 20:40:14 2009
@@ -197,12 +197,14 @@
 extern char *make_relative_prefix (const char *, const char *,
                                    const char *) ATTRIBUTE_MALLOC;
 
+/* LLVM LOCAL begin */
 /* Generate a relocated path to some installation directory without
    attempting to follow any soft links.  Allocates
    return value using malloc.  */
 
 extern char *make_relative_prefix_ignore_links (const char *, const char *,
 						const char *) ATTRIBUTE_MALLOC;
+/* LLVM LOCAL end */
 
 /* Choose a temporary directory to use for scratch files.  */
 

Modified: llvm-gcc-4.2/trunk/libiberty/make-relative-prefix.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/libiberty/make-relative-prefix.c?rev=84588&r1=84587&r2=84588&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/libiberty/make-relative-prefix.c (original)
+++ llvm-gcc-4.2/trunk/libiberty/make-relative-prefix.c Mon Oct 19 20:40:14 2009
@@ -216,10 +216,11 @@
    function will return /red/green/blue/../../omega/.
 
    If no relative prefix can be found, return NULL.  */
-
+/* LLVM LOCAL begin */
 static char *
 make_relative_prefix_1 (const char *progname, const char *bin_prefix,
 			const char *prefix, const int resolve_links)
+/* LLVM LOCAL end */
 {
   char **prog_dirs, **bin_dirs, **prefix_dirs;
   int prog_num, bin_num, prefix_num;
@@ -289,6 +290,7 @@
 	}
     }
 
+  /* LLVM LOCAL begin */
   if ( resolve_links )
     {
       full_progname = lrealpath (progname);
@@ -297,6 +299,7 @@
     }
   else
     full_progname = strdup(progname);
+  /* LLVM LOCAL end */
 
   prog_dirs = split_directories (full_progname, &prog_num);
   bin_dirs = split_directories (bin_prefix, &bin_num);
@@ -393,7 +396,7 @@
   return ret;
 }
 
-
+/* LLVM LOCAL begin */
 /* Do the full job, including symlink resolution.
    This path will find files installed in the same place as the
    program even when a soft link has been made to the program
@@ -421,4 +424,4 @@
 {
   return make_relative_prefix_1 (progname, bin_prefix, prefix, 0);
 }
-
+/* LLVM LOCAL end */




From daniel at zuster.org  Mon Oct 19 21:23:05 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Tue, 20 Oct 2009 02:23:05 -0000
Subject: [llvm-commits] [llvm] r84591 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910200223.n9K2N54F027473@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 21:23:05 2009
New Revision: 84591

URL: http://llvm.org/viewvc/llvm-project?rev=84591&view=rev
Log:
NNT: Remove duplicate verbose print.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84591&r1=84590&r2=84591&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Mon Oct 19 21:23:05 2009
@@ -571,10 +571,6 @@
 
   # Run the programs tests... creating a report.nightly.csv file.
   my $LLCBetaOpts = "";
-  if( $VERBOSE) {
-    print "$MAKECMD -k $MAKEOPTS $PROGTESTOPTS report.nightly.csv ".
-          "$TESTFLAGS TEST=nightly > $ProgramTestLog 2>&1\n";
-  }
   RunLoggedCommand("$MAKECMD -k $MAKEOPTS $PROGTESTOPTS report.nightly.csv ".
                    "$TESTFLAGS TEST=nightly",
                    $ProgramTestLog, "TEST DIRECTORY $SubDir");




From daniel at zuster.org  Mon Oct 19 21:23:14 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Tue, 20 Oct 2009 02:23:14 -0000
Subject: [llvm-commits] [llvm] r84592 - in /llvm/trunk: Makefile
 utils/buildit/GNUmakefile utils/buildit/build_llvm
Message-ID: <200910200223.n9K2NE4h027490@zion.cs.uiuc.edu>

Author: ddunbar
Date: Mon Oct 19 21:23:13 2009
New Revision: 84592

URL: http://llvm.org/viewvc/llvm-project?rev=84592&view=rev
Log:
Revert "Tweak top-level Makefile to facilitate Apple-style build.", this is
breaking Clang's Apple-style build.

Modified:
    llvm/trunk/Makefile
    llvm/trunk/utils/buildit/GNUmakefile
    llvm/trunk/utils/buildit/build_llvm

Modified: llvm/trunk/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/Makefile?rev=84592&r1=84591&r2=84592&view=diff

==============================================================================
--- llvm/trunk/Makefile (original)
+++ llvm/trunk/Makefile Mon Oct 19 21:23:13 2009
@@ -19,7 +19,6 @@
 #
 # When cross-compiling, there are some things (tablegen) that need to
 # be build for the build system first.
-ifndef RC_ARCHS  # Normal build (not "Apple-style").
 ifeq ($(BUILD_DIRS_ONLY),1)
   DIRS := lib/System lib/Support utils
   OPTIONAL_DIRS :=
@@ -212,8 +211,3 @@
 
 .NOTPARALLEL:
 
-else # Building "Apple-style."
-# N.B. In an Apple-style build, once configuration is done, lines
-# marked "Apple-style" are removed with sed!  See utils/buildit/build_llvm.
-include utils/buildit/GNUmakefile # Building "Apple-style."
-endif # Building "Apple-style."

Modified: llvm/trunk/utils/buildit/GNUmakefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/GNUmakefile?rev=84592&r1=84591&r2=84592&view=diff

==============================================================================
--- llvm/trunk/utils/buildit/GNUmakefile (original)
+++ llvm/trunk/utils/buildit/GNUmakefile Mon Oct 19 21:23:13 2009
@@ -59,7 +59,7 @@
 # NOTE : Always put version numbers at the end because they are optional.
 install: $(OBJROOT) $(SYMROOT) $(DSTROOT)
 	cd $(OBJROOT) && \
-	  $(SRC)/utils/buildit/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
+	  $(SRC)/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
 	    $(SRC) $(PREFIX) $(DSTROOT) $(SYMROOT) \
 	    $(LLVM_ASSERTIONS) $(LLVM_OPTIMIZED) \
 	    $(RC_ProjectSourceVersion) $(RC_ProjectSourceSubversion) 

Modified: llvm/trunk/utils/buildit/build_llvm
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/build_llvm?rev=84592&r1=84591&r2=84592&view=diff

==============================================================================
--- llvm/trunk/utils/buildit/build_llvm (original)
+++ llvm/trunk/utils/buildit/build_llvm Mon Oct 19 21:23:13 2009
@@ -82,10 +82,6 @@
 rm -rf $SRC_DIR || exit 1
 mkdir $SRC_DIR || exit 1
 ln -s $ORIG_SRC_DIR/* $SRC_DIR/ || exit 1
-# We can't use the top-level Makefile as-is.  Remove the soft link:
-rm $SRC_DIR/Makefile || exit 1
-# Now create our own by editing the top-level Makefile, deleting every line marked "Apple-style":
-sed -e '/[Aa]pple-style/d' -e '/include.*GNUmakefile/d' $ORIG_SRC_DIR/Makefile > $SRC_DIR/Makefile || exit 1
 
 # Build the LLVM tree universal.
 mkdir -p $DIR/obj-llvm || exit 1




From daniel at zuster.org  Mon Oct 19 21:28:47 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Mon, 19 Oct 2009 19:28:47 -0700
Subject: [llvm-commits] [llvm] r84507 - in /llvm/trunk: Makefile
	utils/buildit/GNUmakefile utils/buildit/build_llvm
In-Reply-To: <6a8523d60910191412t4e63186cre37e8db21587e349@mail.gmail.com>
References: <200910191753.n9JHrwL2006834@zion.cs.uiuc.edu>
	<6a8523d60910191412t4e63186cre37e8db21587e349@mail.gmail.com>
Message-ID: <6a8523d60910191928q3aaf02a9h3ee4fc282b1cf935@mail.gmail.com>

I went ahead and reverted this for now.

 - Daniel

On Mon, Oct 19, 2009 at 2:12 PM, Daniel Dunbar <daniel at zuster.org> wrote:
> Hi Stuart,
>
> This is breaking Clang's Apple style build, see our projects builder.
>
> Can you find another solution, or perhaps limit this to only affect
> the LLVM build?
>
> ?- Daniel
>
> On Mon, Oct 19, 2009 at 10:53 AM, Stuart Hastings <stuart at apple.com> wrote:
>> Author: stuart
>> Date: Mon Oct 19 12:53:54 2009
>> New Revision: 84507
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=84507&view=rev
>> Log:
>> Tweak top-level Makefile to facilitate Apple-style build.
>>
>> Modified:
>> ? ?llvm/trunk/Makefile
>> ? ?llvm/trunk/utils/buildit/GNUmakefile
>> ? ?llvm/trunk/utils/buildit/build_llvm
>>
>> Modified: llvm/trunk/Makefile
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/Makefile?rev=84507&r1=84506&r2=84507&view=diff
>>
>> ==============================================================================
>> --- llvm/trunk/Makefile (original)
>> +++ llvm/trunk/Makefile Mon Oct 19 12:53:54 2009
>> @@ -19,6 +19,7 @@
>> ?#
>> ?# When cross-compiling, there are some things (tablegen) that need to
>> ?# be build for the build system first.
>> +ifndef RC_ARCHS ?# Normal build (not "Apple-style").
>> ?ifeq ($(BUILD_DIRS_ONLY),1)
>> ? DIRS := lib/System lib/Support utils
>> ? OPTIONAL_DIRS :=
>> @@ -211,3 +212,8 @@
>>
>> ?.NOTPARALLEL:
>>
>> +else # Building "Apple-style."
>> +# N.B. In an Apple-style build, once configuration is done, lines
>> +# marked "Apple-style" are removed with sed! ?See utils/buildit/build_llvm.
>> +include utils/buildit/GNUmakefile # Building "Apple-style."
>> +endif # Building "Apple-style."
>>
>> Modified: llvm/trunk/utils/buildit/GNUmakefile
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/GNUmakefile?rev=84507&r1=84506&r2=84507&view=diff
>>
>> ==============================================================================
>> --- llvm/trunk/utils/buildit/GNUmakefile (original)
>> +++ llvm/trunk/utils/buildit/GNUmakefile Mon Oct 19 12:53:54 2009
>> @@ -59,7 +59,7 @@
>> ?# NOTE : Always put version numbers at the end because they are optional.
>> ?install: $(OBJROOT) $(SYMROOT) $(DSTROOT)
>> ? ? ? ?cd $(OBJROOT) && \
>> - ? ? ? ? $(SRC)/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
>> + ? ? ? ? $(SRC)/utils/buildit/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
>> ? ? ? ? ? ?$(SRC) $(PREFIX) $(DSTROOT) $(SYMROOT) \
>> ? ? ? ? ? ?$(LLVM_ASSERTIONS) $(LLVM_OPTIMIZED) \
>> ? ? ? ? ? ?$(RC_ProjectSourceVersion) $(RC_ProjectSourceSubversion)
>>
>> Modified: llvm/trunk/utils/buildit/build_llvm
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/build_llvm?rev=84507&r1=84506&r2=84507&view=diff
>>
>> ==============================================================================
>> --- llvm/trunk/utils/buildit/build_llvm (original)
>> +++ llvm/trunk/utils/buildit/build_llvm Mon Oct 19 12:53:54 2009
>> @@ -82,6 +82,10 @@
>> ?rm -rf $SRC_DIR || exit 1
>> ?mkdir $SRC_DIR || exit 1
>> ?ln -s $ORIG_SRC_DIR/* $SRC_DIR/ || exit 1
>> +# We can't use the top-level Makefile as-is. ?Remove the soft link:
>> +rm $SRC_DIR/Makefile || exit 1
>> +# Now create our own by editing the top-level Makefile, deleting every line marked "Apple-style":
>> +sed -e '/[Aa]pple-style/d' -e '/include.*GNUmakefile/d' $ORIG_SRC_DIR/Makefile > $SRC_DIR/Makefile || exit 1
>>
>> ?# Build the LLVM tree universal.
>> ?mkdir -p $DIR/obj-llvm || exit 1
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>



From nicholas at mxc.ca  Mon Oct 19 23:09:51 2009
From: nicholas at mxc.ca (Nick Lewycky)
Date: Tue, 20 Oct 2009 04:09:51 -0000
Subject: [llvm-commits] [llvm] r84595 -
	/llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp
Message-ID: <200910200409.n9K49pW0031125@zion.cs.uiuc.edu>

Author: nicholas
Date: Mon Oct 19 23:09:50 2009
New Revision: 84595

URL: http://llvm.org/viewvc/llvm-project?rev=84595&view=rev
Log:
Correct test for PowerPC.

Modified:
    llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp

Modified: llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp?rev=84595&r1=84594&r2=84595&view=diff

==============================================================================
--- llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp (original)
+++ llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp Mon Oct 19 23:09:50 2009
@@ -166,7 +166,7 @@
   EXPECT_EQ(8, TestFunctionPtr());
 }
 
-#if !defined(__arm__) && !defined(__POWERPC__) && !defined(__ppc__)
+#if !defined(__arm__) && !defined(__powerpc__)
 // Test a function C which calls A and B which call each other.
 TEST_F(JITTest, NonLazyCompilationStillNeedsStubs) {
   TheJIT->DisableLazyCompilation();




From gohman at apple.com  Mon Oct 19 23:16:37 2009
From: gohman at apple.com (Dan Gohman)
Date: Tue, 20 Oct 2009 04:16:37 -0000
Subject: [llvm-commits] [llvm] r84596 - in /llvm/trunk:
 include/llvm/CodeGen/MachineLoopInfo.h lib/CodeGen/MachineLoopInfo.cpp
Message-ID: <200910200416.n9K4Gbbq031325@zion.cs.uiuc.edu>

Author: djg
Date: Mon Oct 19 23:16:37 2009
New Revision: 84596

URL: http://llvm.org/viewvc/llvm-project?rev=84596&view=rev
Log:
Add getTopBlock and getBottomBlock member functions to MachineLoopInfo.

Modified:
    llvm/trunk/include/llvm/CodeGen/MachineLoopInfo.h
    llvm/trunk/lib/CodeGen/MachineLoopInfo.cpp

Modified: llvm/trunk/include/llvm/CodeGen/MachineLoopInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/CodeGen/MachineLoopInfo.h?rev=84596&r1=84595&r2=84596&view=diff

==============================================================================
--- llvm/trunk/include/llvm/CodeGen/MachineLoopInfo.h (original)
+++ llvm/trunk/include/llvm/CodeGen/MachineLoopInfo.h Mon Oct 19 23:16:37 2009
@@ -38,6 +38,17 @@
 class MachineLoop : public LoopBase<MachineBasicBlock, MachineLoop> {
 public:
   MachineLoop();
+
+  /// getTopBlock - Return the "top" block in the loop, which is the first
+  /// block in the linear layout, ignoring any parts of the loop not
+  /// contiguous with the part the contains the header.
+  MachineBasicBlock *getTopBlock();
+
+  /// getBottomBlock - Return the "bottom" block in the loop, which is the last
+  /// block in the linear layout, ignoring any parts of the loop not
+  /// contiguous with the part the contains the header.
+  MachineBasicBlock *getBottomBlock();
+
 private:
   friend class LoopInfoBase<MachineBasicBlock, MachineLoop>;
   explicit MachineLoop(MachineBasicBlock *MBB)

Modified: llvm/trunk/lib/CodeGen/MachineLoopInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/MachineLoopInfo.cpp?rev=84596&r1=84595&r2=84596&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/MachineLoopInfo.cpp (original)
+++ llvm/trunk/lib/CodeGen/MachineLoopInfo.cpp Mon Oct 19 23:16:37 2009
@@ -43,3 +43,31 @@
   AU.addRequired<MachineDominatorTree>();
   MachineFunctionPass::getAnalysisUsage(AU);
 }
+
+MachineBasicBlock *MachineLoop::getTopBlock() {
+  MachineBasicBlock *TopMBB = getHeader();
+  MachineFunction::iterator Begin = TopMBB->getParent()->begin();
+  if (TopMBB != Begin) {
+    MachineBasicBlock *PriorMBB = prior(MachineFunction::iterator(TopMBB));
+    while (contains(PriorMBB)) {
+      TopMBB = PriorMBB;
+      if (TopMBB == Begin) break;
+      PriorMBB = prior(MachineFunction::iterator(TopMBB));
+    }
+  }
+  return TopMBB;
+}
+
+MachineBasicBlock *MachineLoop::getBottomBlock() {
+  MachineBasicBlock *BotMBB = getHeader();
+  MachineFunction::iterator End = BotMBB->getParent()->end();
+  if (BotMBB != prior(End)) {
+    MachineBasicBlock *NextMBB = next(MachineFunction::iterator(BotMBB));
+    while (contains(NextMBB)) {
+      BotMBB = NextMBB;
+      if (BotMBB == next(MachineFunction::iterator(BotMBB))) break;
+      NextMBB = next(MachineFunction::iterator(BotMBB));
+    }
+  }
+  return BotMBB;
+}




From evan.cheng at apple.com  Mon Oct 19 23:23:20 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Tue, 20 Oct 2009 04:23:20 -0000
Subject: [llvm-commits] [llvm] r84597 - in /llvm/trunk/lib/CodeGen:
 LiveIntervalAnalysis.cpp ScheduleDAGPrinter.cpp
Message-ID: <200910200423.n9K4NKok031678@zion.cs.uiuc.edu>

Author: evancheng
Date: Mon Oct 19 23:23:20 2009
New Revision: 84597

URL: http://llvm.org/viewvc/llvm-project?rev=84597&view=rev
Log:
Trim unnecessary includes.

Modified:
    llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp
    llvm/trunk/lib/CodeGen/ScheduleDAGPrinter.cpp

Modified: llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp?rev=84597&r1=84596&r2=84597&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp (original)
+++ llvm/trunk/lib/CodeGen/LiveIntervalAnalysis.cpp Mon Oct 19 23:23:20 2009
@@ -28,7 +28,6 @@
 #include "llvm/CodeGen/MachineMemOperand.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/CodeGen/PseudoSourceValue.h"
 #include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Target/TargetInstrInfo.h"
 #include "llvm/Target/TargetMachine.h"

Modified: llvm/trunk/lib/CodeGen/ScheduleDAGPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/ScheduleDAGPrinter.cpp?rev=84597&r1=84596&r2=84597&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/ScheduleDAGPrinter.cpp (original)
+++ llvm/trunk/lib/CodeGen/ScheduleDAGPrinter.cpp Mon Oct 19 23:23:20 2009
@@ -18,7 +18,6 @@
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
-#include "llvm/CodeGen/PseudoSourceValue.h"
 #include "llvm/Target/TargetRegisterInfo.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Support/Debug.h"




From gohman at apple.com  Mon Oct 19 23:50:38 2009
From: gohman at apple.com (Dan Gohman)
Date: Tue, 20 Oct 2009 04:50:38 -0000
Subject: [llvm-commits] [llvm] r84598 - in /llvm/trunk:
 lib/CodeGen/CodePlacementOpt.cpp
 test/CodeGen/X86/2009-04-20-LinearScanOpt.ll
 test/CodeGen/X86/discontiguous-loops.ll test/CodeGen/X86/loop-blocks.ll
Message-ID: <200910200450.n9K4ocjP032734@zion.cs.uiuc.edu>

Author: djg
Date: Mon Oct 19 23:50:37 2009
New Revision: 84598

URL: http://llvm.org/viewvc/llvm-project?rev=84598&view=rev
Log:
Re-apply r84295, with fixes to how the loop "top" and "bottom" blocks are
tracked. Instead of trying to manually keep track of these locations
while doing complex modifications, just recompute them when they're needed.
This fixes a bug in which the TopMBB and BotMBB were not correctly updated,
leading to invalid transformations.

Added:
    llvm/trunk/test/CodeGen/X86/discontiguous-loops.ll
    llvm/trunk/test/CodeGen/X86/loop-blocks.ll
Modified:
    llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp
    llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll

Modified: llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp?rev=84598&r1=84597&r2=84598&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp (original)
+++ llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp Mon Oct 19 23:50:37 2009
@@ -34,14 +34,6 @@
     const TargetInstrInfo *TII;
     const TargetLowering  *TLI;
 
-    /// ChangedMBBs - BBs which are modified by OptimizeIntraLoopEdges.
-    SmallPtrSet<MachineBasicBlock*, 8> ChangedMBBs;
-
-    /// UncondJmpMBBs - A list of BBs which are in loops and end with
-    /// unconditional branches.
-    SmallVector<std::pair<MachineBasicBlock*,MachineBasicBlock*>, 4>
-    UncondJmpMBBs;
-
   public:
     static char ID;
     CodePlacementOpt() : MachineFunctionPass(&ID) {}
@@ -58,7 +50,19 @@
     }
 
   private:
-    bool OptimizeIntraLoopEdges();
+    bool HasFallthrough(MachineBasicBlock *MBB);
+    bool HasAnalyzableTerminator(MachineBasicBlock *MBB);
+    void Splice(MachineFunction &MF,
+                MachineFunction::iterator InsertPt,
+                MachineFunction::iterator Begin,
+                MachineFunction::iterator End);
+    void UpdateTerminator(MachineBasicBlock *MBB);
+    bool EliminateUnconditionalJumpsToTop(MachineFunction &MF,
+                                          MachineLoop *L);
+    bool MoveDiscontiguousLoopBlocks(MachineFunction &MF,
+                                     MachineLoop *L);
+    bool OptimizeIntraLoopEdgesInLoopNest(MachineFunction &MF, MachineLoop *L);
+    bool OptimizeIntraLoopEdges(MachineFunction &MF);
     bool AlignLoops(MachineFunction &MF);
     bool AlignLoop(MachineFunction &MF, MachineLoop *L, unsigned Align);
   };
@@ -70,168 +74,354 @@
   return new CodePlacementOpt();
 }
 
-/// OptimizeBackEdges - Place loop back edges to move unconditional branches
-/// out of the loop.
-///
-///       A:
-///       ...
-///       <fallthrough to B>
+/// HasFallthrough - Test whether the given branch has a fallthrough, either as
+/// a plain fallthrough or as a fallthrough case of a conditional branch.
 ///
-///       B:  --> loop header
-///       ...
-///       jcc <cond> C, [exit]
-///
-///       C:
-///       ...
-///       jmp B
+bool CodePlacementOpt::HasFallthrough(MachineBasicBlock *MBB) {
+  MachineBasicBlock *TBB = 0, *FBB = 0;
+  SmallVector<MachineOperand, 4> Cond;
+  if (TII->AnalyzeBranch(*MBB, TBB, FBB, Cond))
+    return false;
+  // This conditional branch has no fallthrough.
+  if (FBB)
+    return false;
+  // An unconditional branch has no fallthrough.
+  if (Cond.empty() && TBB)
+    return false;
+  // It has a fallthrough.
+  return true;
+}
+
+/// HasAnalyzableTerminator - Test whether AnalyzeBranch will succeed on MBB.
+/// This is called before major changes are begun to test whether it will be
+/// possible to complete the changes.
 ///
-/// ==>
+/// Target-specific code is hereby encouraged to make AnalyzeBranch succeed
+/// whenever possible.
 ///
-///       A:
-///       ...
-///       jmp B
+bool CodePlacementOpt::HasAnalyzableTerminator(MachineBasicBlock *MBB) {
+  // Conservatively ignore EH landing pads.
+  if (MBB->isLandingPad()) return false;
+
+  // Ignore blocks which look like they might have EH-related control flow.
+  // At the time of this writing, there are blocks which AnalyzeBranch
+  // thinks end in single uncoditional branches, yet which have two CFG
+  // successors. Code in this file is not prepared to reason about such things.
+  if (!MBB->empty() && MBB->back().getOpcode() == TargetInstrInfo::EH_LABEL)
+    return false;
+
+  // Aggressively handle return blocks and similar constructs.
+  if (MBB->succ_empty()) return true;
+
+  // Ask the target's AnalyzeBranch if it can handle this block.
+  MachineBasicBlock *TBB = 0, *FBB = 0;
+  SmallVector<MachineOperand, 4> Cond;
+  // Make the the terminator is understood.
+  if (TII->AnalyzeBranch(*MBB, TBB, FBB, Cond))
+    return false;
+  // Make sure we have the option of reversing the condition.
+  if (!Cond.empty() && TII->ReverseBranchCondition(Cond))
+    return false;
+  return true;
+}
+
+/// Splice - Move the sequence of instructions [Begin,End) to just before
+/// InsertPt. Update branch instructions as needed to account for broken
+/// fallthrough edges and to take advantage of newly exposed fallthrough
+/// opportunities.
 ///
-///       C:
-///       ...
-///       <fallthough to B>
-///       
-///       B:  --> loop header
-///       ...
-///       jcc <cond> C, [exit]
+void CodePlacementOpt::Splice(MachineFunction &MF,
+                              MachineFunction::iterator InsertPt,
+                              MachineFunction::iterator Begin,
+                              MachineFunction::iterator End) {
+  assert(Begin != MF.begin() && End != MF.begin() && InsertPt != MF.begin() &&
+         "Splice can't change the entry block!");
+  MachineFunction::iterator OldBeginPrior = prior(Begin);
+  MachineFunction::iterator OldEndPrior = prior(End);
+
+  MF.splice(InsertPt, Begin, End);
+
+  UpdateTerminator(prior(Begin));
+  UpdateTerminator(OldBeginPrior);
+  UpdateTerminator(OldEndPrior);
+}
+
+/// UpdateTerminator - Update the terminator instructions in MBB to account
+/// for changes to the layout. If the block previously used a fallthrough,
+/// it may now need a branch, and if it previously used branching it may now
+/// be able to use a fallthrough.
 ///
-bool CodePlacementOpt::OptimizeIntraLoopEdges() {
-  if (!TLI->shouldOptimizeCodePlacement())
-    return false;
+void CodePlacementOpt::UpdateTerminator(MachineBasicBlock *MBB) {
+  // A block with no successors has no concerns with fall-through edges.
+  if (MBB->succ_empty()) return;
+
+  MachineBasicBlock *TBB = 0, *FBB = 0;
+  SmallVector<MachineOperand, 4> Cond;
+  bool B = TII->AnalyzeBranch(*MBB, TBB, FBB, Cond);
+  (void) B;
+  assert(!B && "UpdateTerminators requires analyzable predecessors!");
+  if (Cond.empty()) {
+    if (TBB) {
+      // The block has an unconditional branch. If its successor is now
+      // its layout successor, delete the branch.
+      if (MBB->isLayoutSuccessor(TBB))
+        TII->RemoveBranch(*MBB);
+    } else {
+      // The block has an unconditional fallthrough. If its successor is not
+      // its layout successor, insert a branch.
+      TBB = *MBB->succ_begin();
+      if (!MBB->isLayoutSuccessor(TBB))
+        TII->InsertBranch(*MBB, TBB, 0, Cond);
+    }
+  } else {
+    if (FBB) {
+      // The block has a non-fallthrough conditional branch. If one of its
+      // successors is its layout successor, rewrite it to a fallthrough
+      // conditional branch.
+      if (MBB->isLayoutSuccessor(TBB)) {
+        TII->RemoveBranch(*MBB);
+        TII->ReverseBranchCondition(Cond);
+        TII->InsertBranch(*MBB, FBB, 0, Cond);
+      } else if (MBB->isLayoutSuccessor(FBB)) {
+        TII->RemoveBranch(*MBB);
+        TII->InsertBranch(*MBB, TBB, 0, Cond);
+      }
+    } else {
+      // The block has a fallthrough conditional branch.
+      MachineBasicBlock *MBBA = *MBB->succ_begin();
+      MachineBasicBlock *MBBB = *next(MBB->succ_begin());
+      if (MBBA == TBB) std::swap(MBBB, MBBA);
+      if (MBB->isLayoutSuccessor(TBB)) {
+        TII->RemoveBranch(*MBB);
+        TII->ReverseBranchCondition(Cond);
+        TII->InsertBranch(*MBB, MBBA, 0, Cond);
+      } else if (!MBB->isLayoutSuccessor(MBBA)) {
+        TII->RemoveBranch(*MBB);
+        TII->InsertBranch(*MBB, TBB, MBBA, Cond);
+      }
+    }
+  }
+}
 
+/// EliminateUnconditionalJumpsToTop - Move blocks which unconditionally jump
+/// to the loop top to the top of the loop so that they have a fall through.
+/// This can introduce a branch on entry to the loop, but it can eliminate a
+/// branch within the loop. See the @simple case in
+/// test/CodeGen/X86/loop_blocks.ll for an example of this.
+bool CodePlacementOpt::EliminateUnconditionalJumpsToTop(MachineFunction &MF,
+                                                        MachineLoop *L) {
   bool Changed = false;
-  for (unsigned i = 0, e = UncondJmpMBBs.size(); i != e; ++i) {
-    MachineBasicBlock *MBB = UncondJmpMBBs[i].first;
-    MachineBasicBlock *SuccMBB = UncondJmpMBBs[i].second;
-    MachineLoop *L = MLI->getLoopFor(MBB);
-    assert(L && "BB is expected to be in a loop!");
-
-    if (ChangedMBBs.count(MBB)) {
-      // BB has been modified, re-analyze.
-      MachineBasicBlock *TBB = 0, *FBB = 0;
-      SmallVector<MachineOperand, 4> Cond;
-      if (TII->AnalyzeBranch(*MBB, TBB, FBB, Cond) || !Cond.empty())
+  MachineBasicBlock *TopMBB = L->getTopBlock();
+
+  bool BotHasFallthrough = HasFallthrough(L->getBottomBlock());
+
+  if (TopMBB == MF.begin() ||
+      HasAnalyzableTerminator(prior(MachineFunction::iterator(TopMBB)))) {
+  new_top:
+    for (MachineBasicBlock::pred_iterator PI = TopMBB->pred_begin(),
+         PE = TopMBB->pred_end(); PI != PE; ++PI) {
+      MachineBasicBlock *Pred = *PI;
+      if (Pred == TopMBB) continue;
+      if (HasFallthrough(Pred)) continue;
+      if (!L->contains(Pred)) continue;
+
+      // Verify that we can analyze all the loop entry edges before beginning
+      // any changes which will require us to be able to analyze them.
+      if (Pred == MF.begin())
         continue;
-      if (MLI->getLoopFor(TBB) != L || TBB->isLandingPad())
+      if (!HasAnalyzableTerminator(Pred))
+        continue;
+      if (!HasAnalyzableTerminator(prior(MachineFunction::iterator(Pred))))
         continue;
-      SuccMBB = TBB;
-    } else {
-      assert(MLI->getLoopFor(SuccMBB) == L &&
-             "Successor is not in the same loop!");
-    }
 
-    if (MBB->isLayoutSuccessor(SuccMBB)) {
-      // Successor is right after MBB, just eliminate the unconditional jmp.
-      // Can this happen?
-      TII->RemoveBranch(*MBB);
-      ChangedMBBs.insert(MBB);
-      ++NumIntraElim;
+      // Move the block.
       Changed = true;
-      continue;
-    }
 
-    // Now check if the predecessor is fallthrough from any BB. If there is,
-    // that BB should be from outside the loop since edge will become a jmp.
-    bool OkToMove = true;
-    MachineBasicBlock *FtMBB = 0, *FtTBB = 0, *FtFBB = 0;
-    SmallVector<MachineOperand, 4> FtCond;    
-    for (MachineBasicBlock::pred_iterator PI = SuccMBB->pred_begin(),
-           PE = SuccMBB->pred_end(); PI != PE; ++PI) {
-      MachineBasicBlock *PredMBB = *PI;
-      if (PredMBB->isLayoutSuccessor(SuccMBB)) {
-        if (TII->AnalyzeBranch(*PredMBB, FtTBB, FtFBB, FtCond)) {
-          OkToMove = false;
+      // Move it and all the blocks that can reach it via fallthrough edges
+      // exclusively, to keep existing fallthrough edges intact.
+      MachineFunction::iterator Begin = Pred;
+      MachineFunction::iterator End = next(Begin);
+      while (Begin != MF.begin()) {
+        MachineFunction::iterator Prior = prior(Begin);
+        if (Prior == MF.begin())
+          break;
+        // Stop when a non-fallthrough edge is found.
+        if (!HasFallthrough(Prior))
+          break;
+        // Stop if a block which could fall-through out of the loop is found.
+        if (Prior->isSuccessor(End))
+          break;
+        // If we've reached the top, stop scanning.
+        if (Prior == MachineFunction::iterator(TopMBB)) {
+          // We know top currently has a fall through (because we just checked
+          // it) which would be lost if we do the transformation, so it isn't
+          // worthwhile to do the transformation unless it would expose a new
+          // fallthrough edge.
+          if (!Prior->isSuccessor(End))
+            goto next_pred;
+          // Otherwise we can stop scanning and procede to move the blocks.
           break;
         }
-        if (!FtTBB)
-          FtTBB = SuccMBB;
-        else if (!FtFBB) {
-          assert(FtFBB != SuccMBB && "Unexpected control flow!");
-          FtFBB = SuccMBB;
-        }
-        
-        // A fallthrough.
-        FtMBB = PredMBB;
-        MachineLoop *PL = MLI->getLoopFor(PredMBB);
-        if (PL && (PL == L || PL->getLoopDepth() >= L->getLoopDepth()))
-          OkToMove = false;
-
-        break;
+        // If we hit a switch or something complicated, don't move anything
+        // for this predecessor.
+        if (!HasAnalyzableTerminator(prior(MachineFunction::iterator(Prior))))
+          break;
+        // Ok, the block prior to Begin will be moved along with the rest.
+        // Extend the range to include it.
+        Begin = Prior;
+        ++NumIntraMoved;
       }
+
+      // Move the blocks.
+      Splice(MF, TopMBB, Begin, End);
+
+      // Update TopMBB.
+      TopMBB = L->getTopBlock();
+
+      // We have a new loop top. Iterate on it. We shouldn't have to do this
+      // too many times if BranchFolding has done a reasonable job.
+      goto new_top;
+    next_pred:;
     }
+  }
+
+  // If the loop previously didn't exit with a fall-through and it now does,
+  // we eliminated a branch.
+  if (Changed &&
+      !BotHasFallthrough &&
+      HasFallthrough(L->getBottomBlock())) {
+    ++NumIntraElim;
+    BotHasFallthrough = true;
+  }
 
-    if (!OkToMove)
-      continue;
+  return Changed;
+}
+
+/// MoveDiscontiguousLoopBlocks - Move any loop blocks that are not in the
+/// portion of the loop contiguous with the header. This usually makes the loop
+/// contiguous, provided that AnalyzeBranch can handle all the relevant
+/// branching. See the @cfg_islands case in test/CodeGen/X86/loop_blocks.ll
+/// for an example of this.
+bool CodePlacementOpt::MoveDiscontiguousLoopBlocks(MachineFunction &MF,
+                                                   MachineLoop *L) {
+  bool Changed = false;
+  MachineBasicBlock *TopMBB = L->getTopBlock();
+  MachineBasicBlock *BotMBB = L->getBottomBlock();
+
+  // Determine a position to move orphaned loop blocks to. If TopMBB is not
+  // entered via fallthrough and BotMBB is exited via fallthrough, prepend them
+  // to the top of the loop to avoid loosing that fallthrough. Otherwise append
+  // them to the bottom, even if it previously had a fallthrough, on the theory
+  // that it's worth an extra branch to keep the loop contiguous.
+  MachineFunction::iterator InsertPt = next(MachineFunction::iterator(BotMBB));
+  bool InsertAtTop = false;
+  if (TopMBB != MF.begin() &&
+      !HasFallthrough(prior(MachineFunction::iterator(TopMBB))) &&
+      HasFallthrough(BotMBB)) {
+    InsertPt = TopMBB;
+    InsertAtTop = true;
+  }
 
-    // Is it profitable? If SuccMBB can fallthrough itself, that can be changed
-    // into a jmp.
-    MachineBasicBlock *TBB = 0, *FBB = 0;
-    SmallVector<MachineOperand, 4> Cond;
-    if (TII->AnalyzeBranch(*SuccMBB, TBB, FBB, Cond))
-      continue;
-    if (!TBB && Cond.empty())
-      TBB = next(MachineFunction::iterator(SuccMBB));
-    else if (!FBB && !Cond.empty())
-      FBB = next(MachineFunction::iterator(SuccMBB));
-
-    // This calculate the cost of the transformation. Also, it finds the *only*
-    // intra-loop edge if there is one.
-    int Cost = 0;
-    bool HasOneIntraSucc = true;
-    MachineBasicBlock *IntraSucc = 0;
-    for (MachineBasicBlock::succ_iterator SI = SuccMBB->succ_begin(),
-           SE = SuccMBB->succ_end(); SI != SE; ++SI) {
-      MachineBasicBlock *SSMBB = *SI;
-      if (MLI->getLoopFor(SSMBB) == L) {
-        if (!IntraSucc)
-          IntraSucc = SSMBB;
-        else
-          HasOneIntraSucc = false;
+  // Keep a record of which blocks are in the portion of the loop contiguous
+  // with the loop header.
+  SmallPtrSet<MachineBasicBlock *, 8> ContiguousBlocks;
+  for (MachineFunction::iterator I = TopMBB,
+       E = next(MachineFunction::iterator(BotMBB)); I != E; ++I)
+    ContiguousBlocks.insert(I);
+
+  // Find non-contigous blocks and fix them.
+  if (InsertPt != MF.begin() && HasAnalyzableTerminator(prior(InsertPt)))
+    for (MachineLoop::block_iterator BI = L->block_begin(), BE = L->block_end();
+         BI != BE; ++BI) {
+      MachineBasicBlock *BB = *BI;
+
+      // Verify that we can analyze all the loop entry edges before beginning
+      // any changes which will require us to be able to analyze them.
+      if (!HasAnalyzableTerminator(BB))
+        continue;
+      if (!HasAnalyzableTerminator(prior(MachineFunction::iterator(BB))))
+        continue;
+
+      // If the layout predecessor is part of the loop, this block will be
+      // processed along with it. This keeps them in their relative order.
+      if (BB != MF.begin() &&
+          L->contains(prior(MachineFunction::iterator(BB))))
+        continue;
+
+      // Check to see if this block is already contiguous with the main
+      // portion of the loop.
+      if (!ContiguousBlocks.insert(BB))
+        continue;
+
+      // Move the block.
+      Changed = true;
+
+      // Process this block and all loop blocks contiguous with it, to keep
+      // them in their relative order.
+      MachineFunction::iterator Begin = BB;
+      MachineFunction::iterator End = next(MachineFunction::iterator(BB));
+      for (; End != MF.end(); ++End) {
+        if (!L->contains(End)) break;
+        if (!HasAnalyzableTerminator(End)) break;
+        ContiguousBlocks.insert(End);
+        ++NumIntraMoved;
       }
 
-      if (SuccMBB->isLayoutSuccessor(SSMBB))
-        // This will become a jmp.
-        ++Cost;
-      else if (MBB->isLayoutSuccessor(SSMBB)) {
-        // One of the successor will become the new fallthrough.
-        if (SSMBB == FBB) {
-          FBB = 0;
-          --Cost;
-        } else if (!FBB && SSMBB == TBB && Cond.empty()) {
-          TBB = 0;
-          --Cost;
-        } else if (!Cond.empty() && !TII->ReverseBranchCondition(Cond)) {
-          assert(SSMBB == TBB);
-          TBB = FBB;
-          FBB = 0;
-          --Cost;
+      // If we're inserting at the bottom of the loop, and the code we're
+      // moving originally had fall-through successors, bring the sucessors
+      // up with the loop blocks to preserve the fall-through edges.
+      if (!InsertAtTop)
+        for (; End != MF.end(); ++End) {
+          if (L->contains(End)) break;
+          if (!HasAnalyzableTerminator(End)) break;
+          if (!HasFallthrough(prior(End))) break;
         }
-      }
-    }
-    if (Cost)
-      continue;
 
-    // Now, let's move the successor to below the BB to eliminate the jmp.
-    SuccMBB->moveAfter(MBB);
-    TII->RemoveBranch(*MBB);
-    TII->RemoveBranch(*SuccMBB);
-    if (TBB)
-      TII->InsertBranch(*SuccMBB, TBB, FBB, Cond);
-    ChangedMBBs.insert(MBB);
-    ChangedMBBs.insert(SuccMBB);
-    if (FtMBB) {
-      TII->RemoveBranch(*FtMBB);
-      TII->InsertBranch(*FtMBB, FtTBB, FtFBB, FtCond);
-      ChangedMBBs.insert(FtMBB);
+      // Move the blocks. This may invalidate TopMBB and/or BotMBB, but
+      // we don't need them anymore at this point.
+      Splice(MF, InsertPt, Begin, End);
     }
-    Changed = true;
-  }
 
-  ++NumIntraMoved;
+  return Changed;
+}
+
+/// OptimizeIntraLoopEdgesInLoopNest - Reposition loop blocks to minimize
+/// intra-loop branching and to form contiguous loops.
+///
+/// This code takes the approach of making minor changes to the existing
+/// layout to fix specific loop-oriented problems. Also, it depends on
+/// AnalyzeBranch, which can't understand complex control instructions.
+///
+bool CodePlacementOpt::OptimizeIntraLoopEdgesInLoopNest(MachineFunction &MF,
+                                                        MachineLoop *L) {
+  bool Changed = false;
+
+  // Do optimization for nested loops.
+  for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I)
+    Changed |= OptimizeIntraLoopEdgesInLoopNest(MF, *I);
+
+  // Do optimization for this loop.
+  Changed |= EliminateUnconditionalJumpsToTop(MF, L);
+  Changed |= MoveDiscontiguousLoopBlocks(MF, L);
+
+  return Changed;
+}
+
+/// OptimizeIntraLoopEdges - Reposition loop blocks to minimize
+/// intra-loop branching and to form contiguous loops.
+///
+bool CodePlacementOpt::OptimizeIntraLoopEdges(MachineFunction &MF) {
+  bool Changed = false;
+
+  if (!TLI->shouldOptimizeCodePlacement())
+    return Changed;
+
+  // Do optimization for each loop in the function.
+  for (MachineLoopInfo::iterator I = MLI->begin(), E = MLI->end();
+       I != E; ++I)
+    if (!(*I)->getParentLoop())
+      Changed |= OptimizeIntraLoopEdgesInLoopNest(MF, *I);
+
   return Changed;
 }
 
@@ -255,6 +445,8 @@
   return Changed;
 }
 
+/// AlignLoop - Align loop headers to target preferred alignments.
+///
 bool CodePlacementOpt::AlignLoop(MachineFunction &MF, MachineLoop *L,
                                  unsigned Align) {
   bool Changed = false;
@@ -263,17 +455,7 @@
   for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I)
     Changed |= AlignLoop(MF, *I, Align);
 
-  MachineBasicBlock *TopMBB = L->getHeader();
-  if (TopMBB == MF.begin()) return Changed;
-
-  MachineBasicBlock *PredMBB = prior(MachineFunction::iterator(TopMBB));
-  while (MLI->getLoopFor(PredMBB) == L) {
-    TopMBB = PredMBB;
-    if (TopMBB == MF.begin()) return Changed;
-    PredMBB = prior(MachineFunction::iterator(TopMBB));
-  }
-
-  TopMBB->setAlignment(Align);
+  L->getTopBlock()->setAlignment(Align);
   Changed = true;
   ++NumLoopsAligned;
 
@@ -288,30 +470,9 @@
   TLI = MF.getTarget().getTargetLowering();
   TII = MF.getTarget().getInstrInfo();
 
-  // Analyze the BBs first and keep track of BBs that
-  // end with an unconditional jmp to another block in the same loop.
-  for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
-    MachineBasicBlock *MBB = I;
-    if (MBB->isLandingPad())
-      continue;
-    MachineLoop *L = MLI->getLoopFor(MBB);
-    if (!L)
-      continue;
-
-    MachineBasicBlock *TBB = 0, *FBB = 0;
-    SmallVector<MachineOperand, 4> Cond;
-    if (TII->AnalyzeBranch(*MBB, TBB, FBB, Cond) || !Cond.empty())
-      continue;
-    if (MLI->getLoopFor(TBB) == L && !TBB->isLandingPad())
-      UncondJmpMBBs.push_back(std::make_pair(MBB, TBB));
-  }
-
-  bool Changed = OptimizeIntraLoopEdges();
+  bool Changed = OptimizeIntraLoopEdges(MF);
 
   Changed |= AlignLoops(MF);
 
-  ChangedMBBs.clear();
-  UncondJmpMBBs.clear();
-
   return Changed;
 }

Modified: llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll?rev=84598&r1=84597&r2=84598&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll (original)
+++ llvm/trunk/test/CodeGen/X86/2009-04-20-LinearScanOpt.ll Mon Oct 19 23:50:37 2009
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=x86_64-apple-darwin10.0 -relocation-model=pic -disable-fp-elim -stats |& grep asm-printer | grep 84
+; RUN: llc < %s -mtriple=x86_64-apple-darwin10.0 -relocation-model=pic -disable-fp-elim -stats |& grep asm-printer | grep 83
 ; rdar://6802189
 
 ; Test if linearscan is unfavoring registers for allocation to allow more reuse

Added: llvm/trunk/test/CodeGen/X86/discontiguous-loops.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/discontiguous-loops.ll?rev=84598&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/X86/discontiguous-loops.ll (added)
+++ llvm/trunk/test/CodeGen/X86/discontiguous-loops.ll Mon Oct 19 23:50:37 2009
@@ -0,0 +1,72 @@
+; RUN: llc -verify-loop-info -verify-dom-info -march=x86-64 < %s
+; PR5243
+
+ at .str96 = external constant [37 x i8], align 8    ; <[37 x i8]*> [#uses=1]
+
+define void @foo() nounwind {
+bb:
+  br label %ybb1
+
+ybb1:                                              ; preds = %yybb13, %xbb6, %bb
+  switch i32 undef, label %bb18 [
+    i32 150, label %ybb2
+    i32 151, label %bb17
+    i32 152, label %bb19
+    i32 157, label %ybb8
+  ]
+
+ybb2:                                              ; preds = %ybb1
+  %tmp = icmp eq i8** undef, null                 ; <i1> [#uses=1]
+  br i1 %tmp, label %bb3, label %xbb6
+
+bb3:                                              ; preds = %ybb2
+  unreachable
+
+xbb4:                                              ; preds = %xbb6
+  store i32 0, i32* undef, align 8
+  br i1 undef, label %xbb6, label %bb5
+
+bb5:                                              ; preds = %xbb4
+  call fastcc void @decl_mode_check_failed() nounwind
+  unreachable
+
+xbb6:                                              ; preds = %xbb4, %ybb2
+  %tmp7 = icmp slt i32 undef, 0                   ; <i1> [#uses=1]
+  br i1 %tmp7, label %xbb4, label %ybb1
+
+ybb8:                                              ; preds = %ybb1
+  %tmp9 = icmp eq i8** undef, null                ; <i1> [#uses=1]
+  br i1 %tmp9, label %bb10, label %ybb12
+
+bb10:                                             ; preds = %ybb8
+  %tmp11 = load i8** undef, align 8               ; <i8*> [#uses=1]
+  call void (i8*, ...)* @fatal(i8* getelementptr inbounds ([37 x i8]* @.str96, i64 0, i64 0), i8* %tmp11) nounwind
+  unreachable
+
+ybb12:                                             ; preds = %ybb8
+  br i1 undef, label %bb15, label %ybb13
+
+ybb13:                                             ; preds = %ybb12
+  %tmp14 = icmp sgt i32 undef, 0                  ; <i1> [#uses=1]
+  br i1 %tmp14, label %bb16, label %ybb1
+
+bb15:                                             ; preds = %ybb12
+  call void (i8*, ...)* @fatal(i8* getelementptr inbounds ([37 x i8]* @.str96, i64 0, i64 0), i8* undef) nounwind
+  unreachable
+
+bb16:                                             ; preds = %ybb13
+  unreachable
+
+bb17:                                             ; preds = %ybb1
+  unreachable
+
+bb18:                                             ; preds = %ybb1
+  unreachable
+
+bb19:                                             ; preds = %ybb1
+  unreachable
+}
+
+declare void @fatal(i8*, ...)
+
+declare fastcc void @decl_mode_check_failed() nounwind

Added: llvm/trunk/test/CodeGen/X86/loop-blocks.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/loop-blocks.ll?rev=84598&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/X86/loop-blocks.ll (added)
+++ llvm/trunk/test/CodeGen/X86/loop-blocks.ll Mon Oct 19 23:50:37 2009
@@ -0,0 +1,207 @@
+; RUN: llc < %s -march=x86-64 -mtriple=x86_64-unknown-linux-gnu -asm-verbose=false | FileCheck %s
+
+; These tests check for loop branching structure, and that the loop align
+; directive is placed in the expected place.
+
+; CodeGen should insert a branch into the middle of the loop in
+; order to avoid a branch within the loop.
+
+; CHECK: simple:
+;      CHECK:   jmp   .LBB1_1
+; CHECK-NEXT:   align
+; CHECK-NEXT: .LBB1_2:
+; CHECK-NEXT:   call loop_latch
+; CHECK-NEXT: .LBB1_1:
+; CHECK-NEXT:   call loop_header
+
+define void @simple() nounwind {
+entry:
+  br label %loop
+
+loop:
+  call void @loop_header()
+  %t0 = tail call i32 @get()
+  %t1 = icmp slt i32 %t0, 0
+  br i1 %t1, label %done, label %bb
+
+bb:
+  call void @loop_latch()
+  br label %loop
+
+done:
+  call void @exit()
+  ret void
+}
+
+; CodeGen should move block_a to the top of the loop so that it
+; falls through into the loop, avoiding a branch within the loop.
+
+; CHECK: slightly_more_involved:
+;      CHECK:   jmp .LBB2_1
+; CHECK-NEXT:   align
+; CHECK-NEXT: .LBB2_4:
+; CHECK-NEXT:   call bar99
+; CHECK-NEXT: .LBB2_1:
+; CHECK-NEXT:   call body
+
+define void @slightly_more_involved() nounwind {
+entry:
+  br label %loop
+
+loop:
+  call void @body()
+  %t0 = call i32 @get()
+  %t1 = icmp slt i32 %t0, 2
+  br i1 %t1, label %block_a, label %bb
+
+bb:
+  %t2 = call i32 @get()
+  %t3 = icmp slt i32 %t2, 99
+  br i1 %t3, label %exit, label %loop
+
+block_a:
+  call void @bar99()
+  br label %loop
+
+exit:
+  call void @exit()
+  ret void
+}
+
+; Same as slightly_more_involved, but block_a is now a CFG diamond with
+; fallthrough edges which should be preserved.
+
+; CHECK: yet_more_involved:
+;      CHECK:   jmp .LBB3_1
+; CHECK-NEXT:   align
+; CHECK-NEXT: .LBB3_7:
+; CHECK-NEXT:   call block_a_true_func
+; CHECK-NEXT:   jmp .LBB3_4
+; CHECK-NEXT: .LBB3_2:
+; CHECK-NEXT:   call bar99
+; CHECK-NEXT:   call get
+; CHECK-NEXT:   cmpl $2999, %eax
+; CHECK-NEXT:   jle .LBB3_7
+; CHECK-NEXT:   call block_a_false_func
+; CHECK-NEXT: .LBB3_4:
+; CHECK-NEXT:   call block_a_merge_func
+; CHECK-NEXT: .LBB3_1:
+; CHECK-NEXT:   call body
+
+define void @yet_more_involved() nounwind {
+entry:
+  br label %loop
+
+loop:
+  call void @body()
+  %t0 = call i32 @get()
+  %t1 = icmp slt i32 %t0, 2
+  br i1 %t1, label %block_a, label %bb
+
+bb:
+  %t2 = call i32 @get()
+  %t3 = icmp slt i32 %t2, 99
+  br i1 %t3, label %exit, label %loop
+
+block_a:
+  call void @bar99()
+  %z0 = call i32 @get()
+  %z1 = icmp slt i32 %z0, 3000
+  br i1 %z1, label %block_a_true, label %block_a_false
+
+block_a_true:
+  call void @block_a_true_func()
+  br label %block_a_merge
+
+block_a_false:
+  call void @block_a_false_func()
+  br label %block_a_merge
+
+block_a_merge:
+  call void @block_a_merge_func()
+  br label %loop
+
+exit:
+  call void @exit()
+  ret void
+}
+
+; CodeGen should move the CFG islands that are part of the loop but don't
+; conveniently fit anywhere so that they are at least contiguous with the
+; loop.
+
+; CHECK: cfg_islands:
+;      CHECK:   jmp     .LBB4_1
+; CHECK-NEXT:   align
+; CHECK-NEXT: .LBB4_7:
+; CHECK-NEXT:   call    bar100
+; CHECK-NEXT:   jmp     .LBB4_1
+; CHECK-NEXT: .LBB4_8:
+; CHECK-NEXT:   call    bar101
+; CHECK-NEXT:   jmp     .LBB4_1
+; CHECK-NEXT: .LBB4_9:
+; CHECK-NEXT:   call    bar102
+; CHECK-NEXT:   jmp     .LBB4_1
+; CHECK-NEXT: .LBB4_5:
+; CHECK-NEXT:   call    loop_latch
+; CHECK-NEXT: .LBB4_1:
+; CHECK-NEXT:   call    loop_header
+
+define void @cfg_islands() nounwind {
+entry:
+  br label %loop
+
+loop:
+  call void @loop_header()
+  %t0 = call i32 @get()
+  %t1 = icmp slt i32 %t0, 100
+  br i1 %t1, label %block100, label %bb
+
+bb:
+  %t2 = call i32 @get()
+  %t3 = icmp slt i32 %t2, 101
+  br i1 %t3, label %block101, label %bb1
+
+bb1:
+  %t4 = call i32 @get()
+  %t5 = icmp slt i32 %t4, 102
+  br i1 %t5, label %block102, label %bb2
+
+bb2:
+  %t6 = call i32 @get()
+  %t7 = icmp slt i32 %t6, 103
+  br i1 %t7, label %exit, label %bb3
+
+bb3:
+  call void @loop_latch()
+  br label %loop
+
+exit:
+  call void @exit()
+  ret void
+
+block100:
+  call void @bar100()
+  br label %loop
+
+block101:
+  call void @bar101()
+  br label %loop
+
+block102:
+  call void @bar102()
+  br label %loop
+}
+
+declare void @bar99() nounwind
+declare void @bar100() nounwind
+declare void @bar101() nounwind
+declare void @bar102() nounwind
+declare void @body() nounwind
+declare void @exit() nounwind
+declare void @loop_header() nounwind
+declare void @loop_latch() nounwind
+declare i32 @get() nounwind
+declare void @block_a_true_func() nounwind
+declare void @block_a_false_func() nounwind
+declare void @block_a_merge_func() nounwind




From daniel at zuster.org  Tue Oct 20 00:15:36 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Tue, 20 Oct 2009 05:15:36 -0000
Subject: [llvm-commits] [llvm] r84600 - in /llvm/trunk:
 include/llvm/Target/TargetRegistry.h
 lib/Target/ARM/AsmParser/ARMAsmParser.cpp
 lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910200515.n9K5FaKk001243@zion.cs.uiuc.edu>

Author: ddunbar
Date: Tue Oct 20 00:15:36 2009
New Revision: 84600

URL: http://llvm.org/viewvc/llvm-project?rev=84600&view=rev
Log:
Wire up the ARM MCInst printer, for llvm-mc.

Modified:
    llvm/trunk/include/llvm/Target/TargetRegistry.h
    llvm/trunk/lib/Target/ARM/AsmParser/ARMAsmParser.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/include/llvm/Target/TargetRegistry.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Target/TargetRegistry.h?rev=84600&r1=84599&r2=84600&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Target/TargetRegistry.h (original)
+++ llvm/trunk/include/llvm/Target/TargetRegistry.h Tue Oct 20 00:15:36 2009
@@ -387,6 +387,15 @@
         T.MCDisassemblerCtorFn = Fn;
     }
 
+    /// RegisterMCInstPrinter - Register a MCInstPrinter implementation for the
+    /// given target.
+    /// 
+    /// Clients are responsible for ensuring that registration doesn't occur
+    /// while another thread is attempting to access the registry. Typically
+    /// this is done by initializing all targets at program startup.
+    ///
+    /// @param T - The target being registered.
+    /// @param Fn - A function to construct an MCInstPrinter for the target.
     static void RegisterMCInstPrinter(Target &T,
                                       Target::MCInstPrinterCtorTy Fn) {
       if (!T.MCInstPrinterCtorFn)
@@ -395,7 +404,7 @@
     
     /// RegisterCodeEmitter - Register a MCCodeEmitter implementation for the
     /// given target.
-    /// 
+    ///
     /// Clients are responsible for ensuring that registration doesn't occur
     /// while another thread is attempting to access the registry. Typically
     /// this is done by initializing all targets at program startup.

Modified: llvm/trunk/lib/Target/ARM/AsmParser/ARMAsmParser.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmParser/ARMAsmParser.cpp?rev=84600&r1=84599&r2=84600&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmParser/ARMAsmParser.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmParser/ARMAsmParser.cpp Tue Oct 20 00:15:36 2009
@@ -534,8 +534,17 @@
       Mnemonic == "bl" ||
       Mnemonic == "push" ||
       Mnemonic == "blx" ||
-      Mnemonic == "pop")
+      Mnemonic == "pop") {
+    // Hard-coded to a valid instruction, till we have a real matcher.
+    Inst = MCInst();
+    Inst.setOpcode(ARM::MOVr);
+    Inst.addOperand(MCOperand::CreateReg(2));
+    Inst.addOperand(MCOperand::CreateReg(2));
+    Inst.addOperand(MCOperand::CreateImm(0));
+    Inst.addOperand(MCOperand::CreateImm(0));
+    Inst.addOperand(MCOperand::CreateReg(0));
     return false;
+  }
 
   return true;
 }

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84600&r1=84599&r2=84600&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Tue Oct 20 00:15:36 2009
@@ -1301,13 +1301,6 @@
   }
 }
 
-// Force static initialization.
-extern "C" void LLVMInitializeARMAsmPrinter() {
-  RegisterAsmPrinter<ARMAsmPrinter> X(TheARMTarget);
-  RegisterAsmPrinter<ARMAsmPrinter> Y(TheThumbTarget);
-}
-
-
 //===----------------------------------------------------------------------===//
 
 void ARMAsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI) {
@@ -1457,3 +1450,26 @@
   
   printMCInst(&TmpInst);
 }
+
+//===----------------------------------------------------------------------===//
+// Target Registry Stuff
+//===----------------------------------------------------------------------===//
+
+static MCInstPrinter *createARMMCInstPrinter(const Target &T,
+                                             unsigned SyntaxVariant,
+                                             const MCAsmInfo &MAI,
+                                             raw_ostream &O) {
+  if (SyntaxVariant == 0)
+    return new ARMInstPrinter(O, MAI, false);
+  return 0;
+}
+
+// Force static initialization.
+extern "C" void LLVMInitializeARMAsmPrinter() {
+  RegisterAsmPrinter<ARMAsmPrinter> X(TheARMTarget);
+  RegisterAsmPrinter<ARMAsmPrinter> Y(TheThumbTarget);
+
+  TargetRegistry::RegisterMCInstPrinter(TheARMTarget, createARMMCInstPrinter);
+  TargetRegistry::RegisterMCInstPrinter(TheThumbTarget, createARMMCInstPrinter);
+}
+




From daniel at zuster.org  Tue Oct 20 00:33:23 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Tue, 20 Oct 2009 05:33:23 -0000
Subject: [llvm-commits] [llvm] r84603 -
	/llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp
Message-ID: <200910200533.n9K5XNAa001945@zion.cs.uiuc.edu>

Author: ddunbar
Date: Tue Oct 20 00:33:23 2009
New Revision: 84603

URL: http://llvm.org/viewvc/llvm-project?rev=84603&view=rev
Log:
PowerPC ifdef'ing considered more complicated than one might like.

Modified:
    llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp

Modified: llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp?rev=84603&r1=84602&r2=84603&view=diff

==============================================================================
--- llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp (original)
+++ llvm/trunk/unittests/ExecutionEngine/JIT/JITTest.cpp Tue Oct 20 00:33:23 2009
@@ -166,7 +166,7 @@
   EXPECT_EQ(8, TestFunctionPtr());
 }
 
-#if !defined(__arm__) && !defined(__powerpc__)
+#if !defined(__arm__) && !defined(__powerpc__) && !defined(__ppc__)
 // Test a function C which calls A and B which call each other.
 TEST_F(JITTest, NonLazyCompilationStillNeedsStubs) {
   TheJIT->DisableLazyCompilation();




From sabre at nondot.org  Tue Oct 20 00:58:02 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 05:58:02 -0000
Subject: [llvm-commits] [llvm] r84611 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
Message-ID: <200910200558.n9K5w2fp002781@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 00:58:02 2009
New Revision: 84611

URL: http://llvm.org/viewvc/llvm-project?rev=84611&view=rev
Log:
t2MOVi32imm is currently always lowered by the Thumb2ITBlockPass.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp?rev=84611&r1=84610&r2=84611&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMAsmPrinter.cpp Tue Oct 20 00:58:02 2009
@@ -1306,6 +1306,8 @@
 void ARMAsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI) {
   ARMMCInstLower MCInstLowering(OutContext, *Mang, *this);
   switch (MI->getOpcode()) {
+  case ARM::t2MOVi32imm:
+    assert(0 && "Should be lowered by thumb2it pass");
   default: break;
   case TargetInstrInfo::DBG_LABEL:
   case TargetInstrInfo::EH_LABEL:
@@ -1441,8 +1443,6 @@
     
     return;
   }
-      
-  // FIXME: Handle t2MOVi32imm also.
   }
       
   MCInst TmpInst;




From sabre at nondot.org  Tue Oct 20 01:15:28 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 06:15:28 -0000
Subject: [llvm-commits] [llvm] r84613 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMInstPrinter.cpp ARMInstPrinter.h
Message-ID: <200910200615.n9K6FSHM003458@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 01:15:28 2009
New Revision: 84613

URL: http://llvm.org/viewvc/llvm-project?rev=84613&view=rev
Log:
Implement some hooks, make printOperand abort if unknown modifiers are
present.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84613&r1=84612&r2=84613&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Tue Oct 20 01:15:28 2009
@@ -33,15 +33,34 @@
 
 void ARMInstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
                                   const char *Modifier) {
-  // FIXME: TURN ASSERT ON.
-  //assert((Modifier == 0 || Modifier[0] == 0) && "Cannot print modifiers");
-  
   const MCOperand &Op = MI->getOperand(OpNo);
   if (Op.isReg()) {
-    O << getRegisterName(Op.getReg());
+    unsigned Reg = Op.getReg();
+    if (Modifier && strcmp(Modifier, "dregpair") == 0) {
+      // FIXME: Breaks e.g. ARM/vmul.ll.
+      assert(0);
+      /*
+      unsigned DRegLo = TRI->getSubReg(Reg, 5); // arm_dsubreg_0
+      unsigned DRegHi = TRI->getSubReg(Reg, 6); // arm_dsubreg_1
+      O << '{'
+      << getRegisterName(DRegLo) << ',' << getRegisterName(DRegHi)
+      << '}';*/
+    } else if (Modifier && strcmp(Modifier, "lane") == 0) {
+      assert(0);
+      /*
+      unsigned RegNum = ARMRegisterInfo::getRegisterNumbering(Reg);
+      unsigned DReg = TRI->getMatchingSuperReg(Reg, RegNum & 1 ? 2 : 1,
+                                               &ARM::DPR_VFP2RegClass);
+      O << getRegisterName(DReg) << '[' << (RegNum & 1) << ']';
+       */
+    } else {
+      O << getRegisterName(Reg);
+    }
   } else if (Op.isImm()) {
+    assert((Modifier == 0 || Modifier[0] == 0) && "No modifiers supported");
     O << '#' << Op.getImm();
   } else {
+    assert((Modifier == 0 || Modifier[0] == 0) && "No modifiers supported");
     assert(Op.isExpr() && "unknown operand kind in printOperand");
     Op.getExpr()->print(O, &MAI);
   }
@@ -143,6 +162,64 @@
   O << "]";
 }  
 
+void ARMInstPrinter::printAddrMode2OffsetOperand(const MCInst *MI,
+                                                 unsigned OpNum) {
+  const MCOperand &MO1 = MI->getOperand(OpNum);
+  const MCOperand &MO2 = MI->getOperand(OpNum+1);
+  
+  if (!MO1.getReg()) {
+    unsigned ImmOffs = ARM_AM::getAM2Offset(MO2.getImm());
+    assert(ImmOffs && "Malformed indexed load / store!");
+    O << '#' << (char)ARM_AM::getAM2Op(MO2.getImm()) << ImmOffs;
+    return;
+  }
+  
+  O << (char)ARM_AM::getAM2Op(MO2.getImm()) << getRegisterName(MO1.getReg());
+  
+  if (unsigned ShImm = ARM_AM::getAM2Offset(MO2.getImm()))
+    O << ", "
+    << ARM_AM::getShiftOpcStr(ARM_AM::getAM2ShiftOpc(MO2.getImm()))
+    << " #" << ShImm;
+}
+
+void ARMInstPrinter::printAddrMode3Operand(const MCInst *MI, unsigned OpNum) {
+  const MCOperand &MO1 = MI->getOperand(OpNum);
+  const MCOperand &MO2 = MI->getOperand(OpNum+1);
+  const MCOperand &MO3 = MI->getOperand(OpNum+2);
+  
+  O << '[' << getRegisterName(MO1.getReg());
+  
+  if (MO2.getReg()) {
+    O << ", " << (char)ARM_AM::getAM3Op(MO3.getImm())
+      << getRegisterName(MO2.getReg()) << ']';
+    return;
+  }
+  
+  if (unsigned ImmOffs = ARM_AM::getAM3Offset(MO3.getImm()))
+    O << ", #"
+    << (char)ARM_AM::getAM3Op(MO3.getImm())
+    << ImmOffs;
+  O << ']';
+}
+
+void ARMInstPrinter::printAddrMode3OffsetOperand(const MCInst *MI,
+                                                 unsigned OpNum) {
+  const MCOperand &MO1 = MI->getOperand(OpNum);
+  const MCOperand &MO2 = MI->getOperand(OpNum+1);
+  
+  if (MO1.getReg()) {
+    O << (char)ARM_AM::getAM3Op(MO2.getImm())
+    << getRegisterName(MO1.getReg());
+    return;
+  }
+  
+  unsigned ImmOffs = ARM_AM::getAM3Offset(MO2.getImm());
+  assert(ImmOffs && "Malformed indexed load / store!");
+  O << "#"
+  << (char)ARM_AM::getAM3Op(MO2.getImm())
+  << ImmOffs;
+}
+
 
 void ARMInstPrinter::printAddrMode4Operand(const MCInst *MI, unsigned OpNum,
                                            const char *Modifier) {
@@ -170,6 +247,44 @@
   }
 }
 
+void ARMInstPrinter::printAddrMode5Operand(const MCInst *MI, unsigned OpNum,
+                                           const char *Modifier) {
+  const MCOperand &MO1 = MI->getOperand(OpNum);
+  const MCOperand &MO2 = MI->getOperand(OpNum+1);
+  
+  if (!MO1.isReg()) {   // FIXME: This is for CP entries, but isn't right.
+    printOperand(MI, OpNum);
+    return;
+  }
+  
+  if (Modifier && strcmp(Modifier, "submode") == 0) {
+    ARM_AM::AMSubMode Mode = ARM_AM::getAM5SubMode(MO2.getImm());
+    if (MO1.getReg() == ARM::SP) {
+      bool isFLDM = (MI->getOpcode() == ARM::FLDMD ||
+                     MI->getOpcode() == ARM::FLDMS);
+      O << ARM_AM::getAMSubModeAltStr(Mode, isFLDM);
+    } else
+      O << ARM_AM::getAMSubModeStr(Mode);
+    return;
+  } else if (Modifier && strcmp(Modifier, "base") == 0) {
+    // Used for FSTM{D|S} and LSTM{D|S} operations.
+    O << getRegisterName(MO1.getReg());
+    if (ARM_AM::getAM5WBFlag(MO2.getImm()))
+      O << "!";
+    return;
+  }
+  
+  O << "[" << getRegisterName(MO1.getReg());
+  
+  if (unsigned ImmOffs = ARM_AM::getAM5Offset(MO2.getImm())) {
+    O << ", #"
+      << (char)ARM_AM::getAM5Op(MO2.getImm())
+      << ImmOffs*4;
+  }
+  O << "]";
+}
+
+
 void ARMInstPrinter::printRegisterList(const MCInst *MI, unsigned OpNum) {
   O << "{";
   // Always skip the first operand, it's the optional (and implicit writeback).
@@ -201,6 +316,11 @@
   abort();
 }
 
+void ARMInstPrinter::printNoHashImmediate(const MCInst *MI, unsigned OpNum) {
+  O << MI->getOperand(OpNum).getImm();
+}
+
+
 void ARMInstPrinter::printPCLabel(const MCInst *MI, unsigned OpNum) {
   // FIXME: remove this.
   abort();

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84613&r1=84612&r2=84613&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Tue Oct 20 01:15:28 2009
@@ -40,13 +40,13 @@
   
   void printSORegOperand(const MCInst *MI, unsigned OpNum);
   void printAddrMode2Operand(const MCInst *MI, unsigned OpNum);
-  void printAddrMode2OffsetOperand(const MCInst *MI, unsigned OpNum) {}
-  void printAddrMode3Operand(const MCInst *MI, unsigned OpNum) {}
-  void printAddrMode3OffsetOperand(const MCInst *MI, unsigned OpNum) {}
+  void printAddrMode2OffsetOperand(const MCInst *MI, unsigned OpNum);
+  void printAddrMode3Operand(const MCInst *MI, unsigned OpNum);
+  void printAddrMode3OffsetOperand(const MCInst *MI, unsigned OpNum);
   void printAddrMode4Operand(const MCInst *MI, unsigned OpNum,
                              const char *Modifier = 0);
   void printAddrMode5Operand(const MCInst *MI, unsigned OpNum,
-                             const char *Modifier = 0) {}
+                             const char *Modifier = 0);
   void printAddrMode6Operand(const MCInst *MI, unsigned OpNum) {}
   void printAddrModePCOperand(const MCInst *MI, unsigned OpNum,
                               const char *Modifier = 0) {}
@@ -76,7 +76,7 @@
   void printJTBlockOperand(const MCInst *MI, unsigned OpNum) {}
   void printJT2BlockOperand(const MCInst *MI, unsigned OpNum) {}
   void printTBAddrMode(const MCInst *MI, unsigned OpNum) {}
-  void printNoHashImmediate(const MCInst *MI, unsigned OpNum) {}
+  void printNoHashImmediate(const MCInst *MI, unsigned OpNum);
 
   void printPCLabel(const MCInst *MI, unsigned OpNum);  
   // FIXME: Implement.




From sabre at nondot.org  Tue Oct 20 01:22:34 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 06:22:34 -0000
Subject: [llvm-commits] [llvm] r84614 - in
 /llvm/trunk/lib/Target/ARM/AsmPrinter: ARMInstPrinter.cpp ARMInstPrinter.h
Message-ID: <200910200622.n9K6MYA0003744@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 01:22:33 2009
New Revision: 84614

URL: http://llvm.org/viewvc/llvm-project?rev=84614&view=rev
Log:
implement some more easy hooks.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84614&r1=84613&r2=84614&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Tue Oct 20 01:22:33 2009
@@ -284,6 +284,36 @@
   O << "]";
 }
 
+void ARMInstPrinter::printAddrMode6Operand(const MCInst *MI, unsigned OpNum) {
+  const MCOperand &MO1 = MI->getOperand(OpNum);
+  const MCOperand &MO2 = MI->getOperand(OpNum+1);
+  const MCOperand &MO3 = MI->getOperand(OpNum+2);
+  
+  // FIXME: No support yet for specifying alignment.
+  O << '[' << getRegisterName(MO1.getReg()) << ']';
+  
+  if (ARM_AM::getAM6WBFlag(MO3.getImm())) {
+    if (MO2.getReg() == 0)
+      O << '!';
+    else
+      O << ", " << getRegisterName(MO2.getReg());
+  }
+}
+
+void ARMInstPrinter::printAddrModePCOperand(const MCInst *MI, unsigned OpNum,
+                                            const char *Modifier) {
+  assert(0 && "FIXME: Implement printAddrModePCOperand");
+}
+
+void ARMInstPrinter::printBitfieldInvMaskImmOperand (const MCInst *MI,
+                                                     unsigned OpNum) {
+  const MCOperand &MO = MI->getOperand(OpNum);
+  uint32_t v = ~MO.getImm();
+  int32_t lsb = CountTrailingZeros_32(v);
+  int32_t width = (32 - CountLeadingZeros_32 (v)) - lsb;
+  assert(MO.isImm() && "Not a valid bf_inv_mask_imm value!");
+  O << '#' << lsb << ", #" << width;
+}
 
 void ARMInstPrinter::printRegisterList(const MCInst *MI, unsigned OpNum) {
   O << "{";

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84614&r1=84613&r2=84614&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Tue Oct 20 01:22:33 2009
@@ -47,10 +47,11 @@
                              const char *Modifier = 0);
   void printAddrMode5Operand(const MCInst *MI, unsigned OpNum,
                              const char *Modifier = 0);
-  void printAddrMode6Operand(const MCInst *MI, unsigned OpNum) {}
+  void printAddrMode6Operand(const MCInst *MI, unsigned OpNum);
   void printAddrModePCOperand(const MCInst *MI, unsigned OpNum,
-                              const char *Modifier = 0) {}
-  void printBitfieldInvMaskImmOperand (const MCInst *MI, unsigned OpNum) {}
+                              const char *Modifier = 0);
+    
+  void printBitfieldInvMaskImmOperand (const MCInst *MI, unsigned OpNum);
   
   void printThumbITMask(const MCInst *MI, unsigned OpNum) {}
   void printThumbAddrModeRROperand(const MCInst *MI, unsigned OpNum) {}




From jyasskin at gmail.com  Tue Oct 20 02:11:09 2009
From: jyasskin at gmail.com (jyasskin at gmail.com)
Date: Tue, 20 Oct 2009 07:11:09 +0000
Subject: [llvm-commits] [PATCH] Move the Function*->allocated blocks map
	from the JITMemoryManager to the JITEmitter
Message-ID: <001636283fda0ee7fb0476589186@google.com>

Reviewers: natebegeman_mac.com,

Message:
Please take a look.

Description:
I'm making Functions auto-remove themselves from the JIT when they're
destroyed. In this case, the Function needs to be removed from the
JITEmitter, but the map recording which Functions need to be removed
lived behind the JITMemoryManager interface, which made things
difficult.

This patch replaces the deallocateMemForFunction(Function*) method with
a pair of methods deallocateFunctionBody(void *) and
deallocateExceptionTable(void *) corresponding to the two
startFoo/endFoo pairs.

Patch at http://codereview.appspot.com/download/issue135055_1.diff. This
slightly depends on http://codereview.appspot.com/132055 for ValueMap<>,
but I can remove that if this gets approved first.

Please review this at http://codereview.appspot.com/135055

Affected files:
   M include/llvm/ExecutionEngine/JITMemoryManager.h
   M lib/ExecutionEngine/JIT/JITEmitter.cpp
   M lib/ExecutionEngine/JIT/JITMemoryManager.cpp
   M unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp




From daniel at zuster.org  Tue Oct 20 02:30:46 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Tue, 20 Oct 2009 07:30:46 -0000
Subject: [llvm-commits] [llvm] r84619 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910200730.n9K7Ullt006074@zion.cs.uiuc.edu>

Author: ddunbar
Date: Tue Oct 20 02:30:46 2009
New Revision: 84619

URL: http://llvm.org/viewvc/llvm-project?rev=84619&view=rev
Log:
NNT: Remove unused BUILDTYPE argument.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84619&r1=84618&r2=84619&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Tue Oct 20 02:30:46 2009
@@ -164,11 +164,10 @@
   if (/^-parallel-test$/)  { $PROGTESTOPTS .= " ENABLE_PARALLEL_REPORT=1"; next; }
   if (/^-with-clang$/)     { $WITHCLANG = 1; next; }
   if (/^-release$/)        { $MAKEOPTS = "$MAKEOPTS ENABLE_OPTIMIZED=1 ".
-                             "OPTIMIZE_OPTION=-O2"; $BUILDTYPE="release"; next;}
+                             "OPTIMIZE_OPTION=-O2"; next;}
   if (/^-release-asserts$/){ $MAKEOPTS = "$MAKEOPTS ENABLE_OPTIMIZED=1 ".
                              "DISABLE_ASSERTIONS=1 ".
-                             "OPTIMIZE_OPTION=-O2";
-                             $BUILDTYPE="release-asserts"; next;}
+                             "OPTIMIZE_OPTION=-O2"; next;}
   if (/^-enable-llcbeta$/) { $PROGTESTOPTS .= " ENABLE_LLCBETA=1"; next; }
   if (/^-disable-pic$/)    { $CONFIGUREARGS .= " --enable-pic=no"; next; }
   if (/^-enable-lli$/)     { $PROGTESTOPTS .= " ENABLE_LLI=1";
@@ -257,10 +256,6 @@
        "\"-nickname <nickname>\"");
 }
 
-if ($BUILDTYPE ne "release" && $BUILDTYPE ne "release-asserts") {
-  $BUILDTYPE = "debug";
-}
-
 if ($CONFIG_PATH ne "") {
   die "error: -config mode is not yet implemented,";
 }




From daniel at zuster.org  Tue Oct 20 02:30:55 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Tue, 20 Oct 2009 07:30:55 -0000
Subject: [llvm-commits] [llvm] r84620 - /llvm/trunk/utils/NewNightlyTest.pl
Message-ID: <200910200730.n9K7UuIk006089@zion.cs.uiuc.edu>

Author: ddunbar
Date: Tue Oct 20 02:30:54 2009
New Revision: 84620

URL: http://llvm.org/viewvc/llvm-project?rev=84620&view=rev
Log:
NNT: Implement "config mode", use -config path/to/llvm-config

 - This runs the nightly test and does all the submission logic, but using the
   LLVM build specified by the llvm-config.

 - Useful for, among other things, testing NNT itself.

Modified:
    llvm/trunk/utils/NewNightlyTest.pl

Modified: llvm/trunk/utils/NewNightlyTest.pl
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/NewNightlyTest.pl?rev=84620&r1=84619&r2=84620&view=diff

==============================================================================
--- llvm/trunk/utils/NewNightlyTest.pl (original)
+++ llvm/trunk/utils/NewNightlyTest.pl Tue Oct 20 02:30:54 2009
@@ -246,51 +246,41 @@
   $WebDir     = $ARGV[2];
 }
 
-if ($BuildDir   eq "" or
-    $WebDir     eq "") {
-  die("please specify a build directory, and a web directory");
- }
+if ($CONFIG_PATH ne "") {
+  $BuildDir = "";
+  $SVNURL = $TestSVNURL = "";
+  if ($WebDir     eq "") {
+    die("please specify a web directory");
+  }
+} else {
+  if ($BuildDir   eq "" or
+      $WebDir     eq "") {
+    die("please specify a build directory, and a web directory");
+  }
+}
 
 if ($nickname eq "") {
   die ("Please invoke NewNightlyTest.pl with command line option " .
        "\"-nickname <nickname>\"");
 }
 
-if ($CONFIG_PATH ne "") {
-  die "error: -config mode is not yet implemented,";
-}
-
 ##############################################################
 #
 # Define the file names we'll use
 #
 ##############################################################
+
 my $Prefix = "$WebDir/$DATE";
-my $ConfigureLog = "$Prefix-Configure-Log.txt";
-my $BuildLog = "$Prefix-Build-Log.txt";
-my $COLog = "$Prefix-CVS-Log.txt";
 my $SingleSourceLog = "$Prefix-SingleSource-ProgramTest.txt.gz";
 my $MultiSourceLog = "$Prefix-MultiSource-ProgramTest.txt.gz";
 my $ExternalLog = "$Prefix-External-ProgramTest.txt.gz";
-my $DejagnuLog = "$Prefix-Dejagnu-testrun.log";
-my $DejagnuSum = "$Prefix-Dejagnu-testrun.sum";
-my $DejagnuLog = "$Prefix-DejagnuTests-Log.txt";
-if (! -d $WebDir) {
-  mkdir $WebDir, 0777 or die "Unable to create web directory: '$WebDir'.";
-  if($VERBOSE){
-    warn "$WebDir did not exist; creating it.\n";
-  }
-}
 
-if ($VERBOSE) {
-  print "INITIALIZED\n";
-  print "SVN URL  = $SVNURL\n";
-  print "COLog    = $COLog\n";
-  print "BuildDir = $BuildDir\n";
-  print "WebDir   = $WebDir\n";
-  print "Prefix   = $Prefix\n";
-  print "BuildLog = $BuildLog\n";
-}
+# These are only valid in non-config mode.
+my $ConfigureLog = "", $BuildLog = "", $COLog = "";
+my $DejagnuLog = "", $DejagnuSum = "", $DejagnuLog = "";
+
+# Are we in config mode?
+my $ConfigMode = 0;
 
 ##############################################################
 #
@@ -496,6 +486,7 @@
 
 # Create the source repository directory.
 sub CheckoutSource {
+  die "Invalid call!" unless $ConfigMode == 0;
   if (-d $BuildDir) {
     if (!$NOREMOVE) {
       if ( $VERBOSE ) {
@@ -527,6 +518,7 @@
 # Build the entire tree, saving build messages to the build log. Returns false
 # on build failure.
 sub BuildLLVM {
+  die "Invalid call!" unless $ConfigMode == 0;
   my $EXTRAFLAGS = "--enable-spec --with-objroot=.";
   RunLoggedCommand("(time -p $NICE ./configure $CONFIGUREARGS $EXTRAFLAGS) ",
                    $ConfigureLog, "CONFIGURE");
@@ -544,6 +536,7 @@
 
 # Running dejagnu tests and save results to log.
 sub RunDejaGNUTests {
+  die "Invalid call!" unless $ConfigMode == 0;
   # Run the feature and regression tests, results are put into testrun.sum and
   # the full log in testrun.log.
   system "rm -f test/testrun.log test/testrun.sum";
@@ -564,6 +557,12 @@
 
   my $ProgramTestLog = "$Prefix-$SubDir-ProgramTest.txt";
 
+  # Make sure to clean things if in non-config mode.
+  if ($ConfigMode == 1) {
+    RunLoggedCommand("$MAKECMD -k $MAKEOPTS $PROGTESTOPTS clean $TESTFLAGS",
+                     $ProgramTestLog, "TEST DIRECTORY $SubDir");
+  }
+
   # Run the programs tests... creating a report.nightly.csv file.
   my $LLCBetaOpts = "";
   RunLoggedCommand("$MAKECMD -k $MAKEOPTS $PROGTESTOPTS report.nightly.csv ".
@@ -640,34 +639,83 @@
 
 ##############################################################
 #
-# The actual NewNightlyTest logic.
+# Initialize filenames
 #
 ##############################################################
 
-$starttime = `date "+20%y-%m-%d %H:%M:%S"`;
-
-if (!$NOCHECKOUT) {
-  CheckoutSource();
+if (! -d $WebDir) {
+  mkdir $WebDir, 0777 or die "Unable to create web directory: '$WebDir'.";
+  if($VERBOSE){
+    warn "$WebDir did not exist; creating it.\n";
+  }
 }
 
-# Build LLVM.
-my $BuildError = 0, $BuildStatus = "OK";
-ChangeDir( $LLVMSrcDir , "llvm source directory") ;
-if ($NOCHECKOUT || $NOBUILD) {
-  $BuildStatus = "Skipped by user";
+if ($CONFIG_PATH ne "") {
+  $ConfigMode = 1;
+  $LLVMSrcDir = GetRegex "^(.*)\\s+", `$CONFIG_PATH --src-root`;
+  $LLVMObjDir = GetRegex "^(.*)\\s+", `$CONFIG_PATH --obj-root`;
+  # FIXME: Add llvm-config hook for this?
+  $LLVMTestDir = $LLVMObjDir . "/projects/test-suite";
 } else {
-  if (!BuildLLVM()) {
-    if( $VERBOSE) { print  "\n***ERROR BUILDING TREE\n\n"; }
-    $BuildError = 1;
-    $BuildStatus = "Error: compilation aborted";
-    $NODEJAGNU=1;
+  $ConfigureLog = "$Prefix-Configure-Log.txt";
+  $BuildLog = "$Prefix-Build-Log.txt";
+  $COLog = "$Prefix-CVS-Log.txt";
+  $DejagnuLog = "$Prefix-Dejagnu-testrun.log";
+  $DejagnuSum = "$Prefix-Dejagnu-testrun.sum";
+  $DejagnuLog = "$Prefix-DejagnuTests-Log.txt";
+}
+
+if ($VERBOSE) {
+  if ($CONFIG_PATH ne "") {
+    print "INITIALIZED (config mode)\n";
+    print "WebDir    = $WebDir\n";
+    print "Prefix    = $Prefix\n";
+    print "LLVM Src  = $LLVMSrcDir\n";
+    print "LLVM Obj  = $LLVMObjDir\n";
+    print "LLVM Test = $LLVMTestDir\n";
+  } else {
+    print "INITIALIZED\n";
+    print "SVN URL  = $SVNURL\n";
+    print "COLog    = $COLog\n";
+    print "BuildDir = $BuildDir\n";
+    print "WebDir   = $WebDir\n";
+    print "Prefix   = $Prefix\n";
+    print "BuildLog = $BuildLog\n";
   }
 }
 
-# Run DejaGNU.
+##############################################################
+#
+# The actual NewNightlyTest logic.
+#
+##############################################################
+
+$starttime = `date "+20%y-%m-%d %H:%M:%S"`;
+
+my $BuildError = 0, $BuildStatus = "OK";
 my $DejagnuTestResults = "Dejagnu skipped by user choice.";
-if (!$NODEJAGNU && !$BuildError) {
-  $DejagnuTestResults = RunDejaGNUTests();
+if ($ConfigMode == 0) {
+  if (!$NOCHECKOUT) {
+    CheckoutSource();
+  }
+
+  # Build LLVM.
+  ChangeDir( $LLVMSrcDir , "llvm source directory") ;
+  if ($NOCHECKOUT || $NOBUILD) {
+    $BuildStatus = "Skipped by user";
+  } else {
+    if (!BuildLLVM()) {
+      if( $VERBOSE) { print  "\n***ERROR BUILDING TREE\n\n"; }
+      $BuildError = 1;
+      $BuildStatus = "Error: compilation aborted";
+      $NODEJAGNU=1;
+    }
+  }
+
+  # Run DejaGNU.
+  if (!$NODEJAGNU && !$BuildError) {
+    $DejagnuTestResults = RunDejaGNUTests();
+  }
 }
 
 # Run the llvm-test tests.
@@ -720,10 +768,13 @@
 my $targetTriple = $1;
 
 # Logs.
-my $ConfigureLogData = ReadFile $ConfigureLog;
-my $BuildLogData = ReadFile $BuildLog;
-my $DejagnuLogData = ReadFile $DejagnuLog;
-my $CheckoutLogData = ReadFile $COLog;
+my ($ConfigureLogData, $BuildLogData, $DejagnuLogData, $CheckoutLogData) = "";
+if ($ConfigMode == 0) {
+  $ConfigureLogData = ReadFile $ConfigureLog;
+  $BuildLogData = ReadFile $BuildLog;
+  $DejagnuLogData = ReadFile $DejagnuLog;
+  $CheckoutLogData = ReadFile $COLog;
+}
 
 # Checkout info.
 my $CheckoutTime_Wall = GetRegex "^real ([0-9.]+)", $CheckoutLogData;




From sanjiv.gupta at microchip.com  Tue Oct 20 04:16:33 2009
From: sanjiv.gupta at microchip.com (Sanjiv Gupta)
Date: Tue, 20 Oct 2009 09:16:33 -0000
Subject: [llvm-commits] [llvm] r84628 -
	/llvm/trunk/lib/Target/PIC16/MCSectionPIC16.h
Message-ID: <200910200916.n9K9GX8j023519@zion.cs.uiuc.edu>

Author: sgupta
Date: Tue Oct 20 04:16:32 2009
New Revision: 84628

URL: http://llvm.org/viewvc/llvm-project?rev=84628&view=rev
Log:
This file is replaeced by PIC16Section.h.

Removed:
    llvm/trunk/lib/Target/PIC16/MCSectionPIC16.h

Removed: llvm/trunk/lib/Target/PIC16/MCSectionPIC16.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/MCSectionPIC16.h?rev=84627&view=auto

==============================================================================
--- llvm/trunk/lib/Target/PIC16/MCSectionPIC16.h (original)
+++ llvm/trunk/lib/Target/PIC16/MCSectionPIC16.h (removed)
@@ -1,88 +0,0 @@
-//===- MCSectionPIC16.h - PIC16-specific section representation -*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file declares the MCSectionPIC16 class.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef LLVM_PIC16SECTION_H
-#define LLVM_PIC16SECTION_H
-
-#include "llvm/MC/MCSection.h"
-
-namespace llvm {
-
-  /// MCSectionPIC16 - Represents a physical section in PIC16 COFF.
-  /// Contains data objects.
-  ///
-  class MCSectionPIC16 : public MCSection {
-    /// Name of the section to uniquely identify it.
-    std::string Name;
-
-    /// User can specify an address at which a section should be placed. 
-    /// Negative value here means user hasn't specified any. 
-    int Address; 
-
-    /// Overlay information - Sections with same color can be overlaid on
-    /// one another.
-    int Color; 
-
-    /// Conatined data objects.
-    std::vector<const GlobalVariable *>Items;
-
-    /// Total size of all data objects contained here.
-    unsigned Size;
-    
-    MCSectionPIC16(const StringRef &name, SectionKind K, int addr, int color)
-      : MCSection(K), Name(name), Address(addr), Color(color) {
-    }
-    
-  public:
-    /// Return the name of the section.
-    const std::string &getName() const { return Name; }
-
-    /// Return the Address of the section.
-    int getAddress() const { return Address; }
-
-    /// Return the Color of the section.
-    int getColor() const { return Color; }
-
-    /// PIC16 Terminology for section kinds is as below.
-    /// UDATA - BSS
-    /// IDATA - initialized data (equiv to Metadata) 
-    /// ROMDATA - ReadOnly.
-    /// UDATA_OVR - Sections that can be overlaid. Section of such type is
-    ///             used to contain function autos an frame. We can think of
-    ///             it as equiv to llvm ThreadBSS)
-    /// So, let's have some convenience functions to Map PIC16 Section types 
-    /// to SectionKind just for the sake of better readability.
-    static SectionKind UDATA_Kind() { return SectionKind::getBSS(); } 
-    static SectionKind IDATA_Kind() { return SectionKind::getMetadata(); }
-    static SectionKind ROMDATA_Kind() { return SectionKind::getReadOnly(); }
-    static SectionKind UDATA_OVR_Kind() { return SectionKind::getThreadBSS(); }
-
-    // If we could just do getKind() == UDATA_Kind() ?
-    bool isUDATA_Kind() { return getKind().isBSS(); }
-    bool isIDATA_Kind() { return getKind().isMetadata(); }
-    bool isROMDATA_Kind() { return getKind().isMetadata(); }
-    bool isUDATA_OVR_Kind() { return getKind().isThreadBSS(); }
-
-    /// This would be the only way to create a section. 
-    static MCSectionPIC16 *Create(const StringRef &Name, SectionKind K, 
-                                  int Address, int Color, MCContext &Ctx);
-    
-    /// Override this as PIC16 has its own way of printing switching
-    /// to a section.
-    virtual void PrintSwitchToSection(const MCAsmInfo &MAI,
-                                      raw_ostream &OS) const;
-  };
-
-} // end namespace llvm
-
-#endif




From benny.kra at googlemail.com  Tue Oct 20 06:44:45 2009
From: benny.kra at googlemail.com (Benjamin Kramer)
Date: Tue, 20 Oct 2009 11:44:45 -0000
Subject: [llvm-commits] [llvm] r84632 - in /llvm/trunk:
 include/llvm/Constants.h include/llvm/Metadata.h
 lib/Target/ARM/ARMISelLowering.cpp
Message-ID: <200910201144.n9KBikoK028482@zion.cs.uiuc.edu>

Author: d0k
Date: Tue Oct 20 06:44:38 2009
New Revision: 84632

URL: http://llvm.org/viewvc/llvm-project?rev=84632&view=rev
Log:
Random #include pruning.

Modified:
    llvm/trunk/include/llvm/Constants.h
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp

Modified: llvm/trunk/include/llvm/Constants.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Constants.h?rev=84632&r1=84631&r2=84632&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Constants.h (original)
+++ llvm/trunk/include/llvm/Constants.h Tue Oct 20 06:44:38 2009
@@ -22,15 +22,16 @@
 #define LLVM_CONSTANTS_H
 
 #include "llvm/Constant.h"
-#include "llvm/Type.h"
 #include "llvm/OperandTraits.h"
 #include "llvm/ADT/APInt.h"
 #include "llvm/ADT/APFloat.h"
 #include "llvm/ADT/SmallVector.h"
+#include <vector>
 
 namespace llvm {
 
 class ArrayType;
+class IntegerType;
 class StructType;
 class PointerType;
 class VectorType;

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84632&r1=84631&r2=84632&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Tue Oct 20 06:44:38 2009
@@ -13,15 +13,13 @@
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef LLVM_MDNODE_H
-#define LLVM_MDNODE_H
+#ifndef LLVM_METADATA_H
+#define LLVM_METADATA_H
 
 #include "llvm/User.h"
 #include "llvm/Type.h"
-#include "llvm/OperandTraits.h"
 #include "llvm/ADT/FoldingSet.h"
 #include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/StringMap.h"
 #include "llvm/ADT/ilist_node.h"

Modified: llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp?rev=84632&r1=84631&r2=84632&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp Tue Oct 20 06:44:38 2009
@@ -25,9 +25,10 @@
 #include "llvm/CallingConv.h"
 #include "llvm/Constants.h"
 #include "llvm/Function.h"
+#include "llvm/GlobalValue.h"
 #include "llvm/Instruction.h"
 #include "llvm/Intrinsics.h"
-#include "llvm/GlobalValue.h"
+#include "llvm/Type.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"




From devang.patel at gmail.com  Tue Oct 20 09:08:42 2009
From: devang.patel at gmail.com (Devang Patel)
Date: Tue, 20 Oct 2009 07:08:42 -0700
Subject: [llvm-commits] metadata.h/cpp review stuff
In-Reply-To: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>
References: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>
Message-ID: <352a1fb20910200708n4336cdaax503fce6480996830@mail.gmail.com>

On Mon, Oct 19, 2009 at 12:06 AM, Chris Lattner <clattner at apple.com> wrote:
> Hi Devang,
>
> I'm just now getting caught up on your work. ?Here are some thoughts/
> questions on Metadata.h/cpp:
>
>
> The metadata hierarchy is:
>
> User
> ? -> MetadataBase
> ? ? ?-> MDString
> ? ? ?-> MDNode
> ? ? ?-> NamedMDNode
>
> Why not:
>
> Value
> ? -> MetadataBase
> ? ? ?-> MDString
> ? ? ?-> MDNode
> ? ? ?-> NamedMDNode
>
> This would shrink them significantly and reduce the games you have to
> play with hiding getNumOperands, overloading operator new, etc. ?Also,
> MetadataBase doesn't have much common state, it isn't clear it needs
> to exist at all.
>
> There is some seriously confused code circling around
> MetadataBase::resizeOperands. ?It looks like half the code thinks that
> operands can happen, but half doesn't. ?Please remove all this stuff.

It is trying to maintain following use counts

!1 = { i32 1} ; [uses 1]
!2 = {  !1 }   ; [uses 0]

...
call @llvm.dbg.blah(!1, !2)
...

Metadata used by another metadata is counted as regular uses. Metadata
used by non-metadata values is not a normal use (it is invisible to
the non-metadata value). This is why the hierarchy is based on User
and operand list is only counting metadata uses.

-
Devang



From devang.patel at gmail.com  Tue Oct 20 09:21:37 2009
From: devang.patel at gmail.com (Devang Patel)
Date: Tue, 20 Oct 2009 07:21:37 -0700
Subject: [llvm-commits] metadata.h/cpp review stuff
In-Reply-To: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>
References: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>
Message-ID: <352a1fb20910200721j5bb5743ex1cd8ee56e66efa63@mail.gmail.com>

On Mon, Oct 19, 2009 at 12:06 AM, Chris Lattner <clattner at apple.com> wrote:
> Hi Devang,
>
> I'm just now getting caught up on your work. ?Here are some thoughts/
> questions on Metadata.h/cpp:

Thanks!

>
> In Metadata.h:
>
>
> #include "llvm/User.h"
> #include "llvm/Type.h"
> #include "llvm/OperandTraits.h"
> #include "llvm/ADT/FoldingSet.h"
> #include "llvm/ADT/SmallVector.h"
> #include "llvm/ADT/SmallPtrSet.h"
> #include "llvm/ADT/DenseMap.h"
> #include "llvm/ADT/StringMap.h"
> #include "llvm/ADT/ilist_node.h"
> #include "llvm/Support/ErrorHandling.h"
> #include "llvm/Support/ValueHandle.h"
>
> This is including way too much. ?Please reduce this as aggressively as
> possible.

Yes, this is on my list.

>
>
> class MDString : public MetadataBase {
> ...
> ? explicit MDString(LLVMContext &C, const char *begin, unsigned l)
>
> This ctor should just take a StringRef.
>

ok

>
> ...
> ? static MDString *get(LLVMContext &Context, const StringRef &Str);
>
> Please take StringRef's by value, not by const&.
>

ok

>
> class MDNode : public MetadataBase, public FoldingSetNode {
> ...
> ? friend class ElementVH;
> ? // Use CallbackVH to hold MDNOde elements.
> ? struct ElementVH : public CallbackVH {
> ? ? MDNode *Parent;
> ? ? ElementVH(Value *V, MDNode *P) : CallbackVH(V), Parent(P) {}
> ? ? ~ElementVH() {}
>
> ? ? virtual void deleted() {
> ? ? ? Parent->replaceElement(this->operator Value*(), 0);
> ? ? }
>
> ? ? virtual void allUsesReplacedWith(Value *NV) {
> ? ? ? Parent->replaceElement(this->operator Value*(), NV);
> ? ? }
> ? };
> ? // Replace each instance of F from the element list of this node
> with T.
> ? void replaceElement(Value *F, Value *T);
>
> ? SmallVector<ElementVH, 4> Node;
>
>
> Please move all of this stuff out of line to Metadata.cpp. ?If needbe,
> Node should be declared as a 'void*' in the header. ?Also, there is no
> reason to use a SmallVector here: the size is fixed on node
> construction. ?Just use a pointer to array and new[]/delete[] the
> array. ?Keeping a SmallVector in NamedMDNode (but not MDNode) is
> completely fine.

ok

>
> Why does MDNode::dropAllReferences need to exist? ?It seems *highly*
> unsafe because it clears 'Node' but doesn't update the FoldingSet.

FoldingSet should be updated properly otherwise it is a bug.

>
> ? Value *getElement(unsigned i) const {
> ? ? assert(getNumElements() > i && "Invalid element number!");
>
> This is fine, but please use "i < getNumElements()" because that is
> more idiomatic. ?This happens multiple times in the file.

ok

>
> /// WeakMetadataVH - a weak value handle for metadata.
> class WeakMetadataVH : public WeakVH {
> public:
>
> I think this should be removed and the use replaced with TrackingVH??

I am also using this to make sure that NamedMDNode collects only
metadata values.

> NamedMDNode::setParent shouldn't be public. ?Does
> NamedMDNode::dropAllReferences need to exist?

It is using MDNodes.

> I don't understand NamedMDNode::Create(const NamedMDNode *NMD, Module
> *M). ?Is this a weird copy ctor? ?Why does it need to exist?

This is used for transferring NamedMDNode from one module to final
linked module during linking.

>
> ? void addElement(MetadataBase *M) {
> ? ? resizeOperands(0);
> ? ? OperandList[NumOperands++] = M;
> ? ? Node.push_back(WeakMetadataVH(M));
> ? }
>
> How can this work? ?Isn't this an invalid access into OperandList?

It is wrong. I'll fix it.

>
> /// MetadataContext -
> /// MetadataContext handles uniquing and assignment of IDs for custom
> metadata
> /// types. Custom metadata handler names do not contain spaces. And
> the name
> /// must start with an alphabet. The regular expression used to check
> name
>
> "start with an alphabet" is missing something.
>
> class MetadataContext {
> public:
> ? typedef std::pair<unsigned, WeakVH> MDPairTy;
> ? typedef SmallVector<MDPairTy, 2> MDMapTy;
> ? typedef DenseMap<const Instruction *, MDMapTy> MDStoreTy;
> ? friend class BitcodeReader;
> private:
>
> ? /// MetadataStore - Collection of metadata used in this context.
> ? MDStoreTy MetadataStore;
>
> ? /// MDHandlerNames - Map to hold metadata handler names.
> ? StringMap<unsigned> MDHandlerNames;
>
>
> These datastructures are all implementation details. ?They should be
> hidden behind a void*, allowing you to avoid #include of DenseMap.h,
> StringMap.h etc. ?MDPairTy should be replaced with TrackingVH<MDNode>.
>

ok

>
> ? /// RegisterMDKind - Register a new metadata kind and return its ID.
> ? /// A metadata kind can be registered only once.
> ? unsigned RegisterMDKind(const char *Name);
>
> ? /// getMDKind - Return metadata kind. If the requested metadata kind
> ? /// is not registered then return 0.
> ? unsigned getMDKind(const char *Name);
>
> ? /// validName - Return true if Name is a valid custom metadata
> handler name.
> ? bool validName(const char *Name);
>
> These methods should all take StringRef for consistency. ?The second
> should be const, the third should be static. ?validName ->
> isValidName. ?RegisterMDKind -> registerMDKind.

ok

>
> ? /// getMD - Get the metadata of given kind attached with an
> Instruction.
> ? /// If the metadata is not found then return 0.
> ? MDNode *getMD(unsigned Kind, const Instruction *Inst);
>
> "with an instruction" -> "to an instruction". ? This occurs several
> times in both files.
>

ok

> ? /// removeMDs - Remove all metadata attached with an instruction.
> ? void removeMDs(const Instruction *Inst);
>
> removeAllMetadata()?

ok :)

>
>
> ? /// getHandlerNames - Get handler names. This is used by bitcode
> ? /// writer.
> ? const StringMap<unsigned> *getHandlerNames();
>
> This is gross for several reasons: at the least, it should be const.
> It also exposes implementation details. ?Finally, the bitcode writer
> is currently writing these out in iterator order which is not at all
> deterministic. ?It would be better to have something like:
>
> ? void getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef>
> ?> &) const;
>
> and have the implementation of getHandlerNames do the sort (with
> array_pod_sort).

ok

[I'll reply to rest of the comments in separate email :) ]
Thanks for the review!
-
Devang



From edwintorok at gmail.com  Tue Oct 20 10:15:09 2009
From: edwintorok at gmail.com (Torok Edwin)
Date: Tue, 20 Oct 2009 15:15:09 -0000
Subject: [llvm-commits] [llvm] r84637 - in /llvm/trunk:
 lib/Transforms/Scalar/SCCP.cpp
 test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll
Message-ID: <200910201515.n9KFF994003926@zion.cs.uiuc.edu>

Author: edwin
Date: Tue Oct 20 10:15:09 2009
New Revision: 84637

URL: http://llvm.org/viewvc/llvm-project?rev=84637&view=rev
Log:
Fix PR4313: IPSCCP was not setting the lattice value for the invoke instruction
when the invoke had multiple return values: it set the lattice value only on the
extractvalue.
This caused the invoke's lattice value to remain the default (undefined), and
later propagated to extractvalue's operand, which incorrectly introduces
undefined behavior.

Added:
    llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll
Modified:
    llvm/trunk/lib/Transforms/Scalar/SCCP.cpp

Modified: llvm/trunk/lib/Transforms/Scalar/SCCP.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/SCCP.cpp?rev=84637&r1=84636&r2=84637&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/SCCP.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/SCCP.cpp Tue Oct 20 10:15:09 2009
@@ -1229,7 +1229,10 @@
     TMRVI = TrackedMultipleRetVals.find(std::make_pair(F, 0));
     if (TMRVI == TrackedMultipleRetVals.end())
       goto CallOverdefined;
-    
+
+    // Need to mark as overdefined, otherwise it stays undefined which
+    // creates extractvalue undef, <idx>
+    markOverdefined(I);
     // If we are tracking this callee, propagate the return values of the call
     // into this call site.  We do this by walking all the uses. Single-index
     // ExtractValueInst uses can be tracked; anything more complicated is
@@ -1271,7 +1274,6 @@
   }
 }
 
-
 void SCCPSolver::Solve() {
   // Process the work lists until they are empty!
   while (!BBWorkList.empty() || !InstWorkList.empty() ||

Added: llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll?rev=84637&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll (added)
+++ llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll Tue Oct 20 10:15:09 2009
@@ -0,0 +1,716 @@
+; RUN: opt -ipsccp <%s -S | FileCheck %s
+; PR4313
+; the return value of a multiple-return value invoke must not be left undefined
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%0 = type <{ %1, %2, [4 x i8], %3 }>
+%1 = type { i32 (...)**, i32, i32, i32, i32, %1*, i8*, i8*, i8* }
+%2 = type { i32 }
+%3 = type <{ %4, %5 }>
+%4 = type { i32 (...)**, i8 }
+%5 = type <{ [33 x i8], [191 x i8] }>
+%6 = type { %1, %7, %10 }
+%7 = type { %8 }
+%8 = type { %9 }
+%9 = type { i8* }
+%10 = type { %11 }
+%11 = type { %12 }
+%12 = type { i32 (...)** }
+%13 = type { %1, %14, %18, %10 }
+%14 = type { %15 }
+%15 = type { %16 }
+%16 = type { %17 }
+%17 = type { %8*, %8*, %8* }
+%18 = type { %19 }
+%19 = type { %20 }
+%20 = type { i32*, i32*, i32* }
+%21 = type { %22, %22 }
+%22 = type { %23 }
+%23 = type { %24 }
+%24 = type { %25, %26, i64 }
+%25 = type <{ i8 }>
+%26 = type { i32, %26*, %26*, %26* }
+%27 = type { %28, %15, i8* }
+%28 = type { %29 }
+%29 = type { %30 }
+%30 = type { %31*, %31*, %31* }
+%31 = type { %32*, %9 }
+%32 = type { i32 (...)**, i8*, i8*, i8 }
+%33 = type { i64, [12 x i32] }
+%34 = type { %35 }
+%35 = type { i32, i32, i32, i32, i32, i32, %36 }
+%36 = type { %36*, %36* }
+%37 = type { i32, %8, %9, %15, %38, %42 }
+%38 = type { %39 }
+%39 = type { %40 }
+%40 = type { %41*, %41*, %41* }
+%41 = type { %8, %12*, i32, %12* }
+%42 = type { %43 }
+%43 = type { %44 }
+%44 = type { %37**, %37**, %37** }
+%45 = type { i32 (...)**, i8*, i8*, i8*, i32, i8 }
+%46 = type { %47, %37*, %12*, %8, %15, %37*, %50 }
+%47 = type { %48 }
+%48 = type { %49 }
+%49 = type { i8*, i8*, i8* }
+%50 = type { %51 }
+%51 = type { %52 }
+%52 = type { %46**, %46**, %46** }
+%53 = type { %21*, %54, %63, %63, %22, %22, %22, %22, %22, %22, %37*, %37*, %37*, %72 }
+%54 = type { %37*, %22, %55, %59, %18 }
+%55 = type { %56 }
+%56 = type { %57 }
+%57 = type { %58*, %58*, %58* }
+%58 = type { %37*, i32, i32 }
+%59 = type { %60 }
+%60 = type { %61 }
+%61 = type { %62*, %62*, %62* }
+%62 = type { %37*, %8, %42, %18, i32, i32, i32, %42, %8, %8 }
+%63 = type { %64 }
+%64 = type { %65 }
+%65 = type { %66*, %66*, %66* }
+%66 = type { %37*, %8, %8, %8, %8, %67, i32, i8, i8, %68 }
+%67 = type { %18, %42, %18, %42 }
+%68 = type { %69 }
+%69 = type { %70 }
+%70 = type { %71*, %71*, %71* }
+%71 = type { i32, i32 }
+%72 = type { %73 }
+%73 = type { %74 }
+%74 = type { %75*, %75*, %75* }
+%75 = type { %76*, %46*, %46*, %42, i32 }
+%76 = type { %77, %78 }
+%77 = type { %12, %12* }
+%78 = type { %79 }
+%79 = type { %80 }
+%80 = type { %12**, %12**, %12** }
+%81 = type { %12, %21*, %53 }
+%82 = type { %22, %8 }
+%83 = type { %26, %84 }
+%84 = type { i32, %22 }
+%s2i64 = type { i64, i64 }
+%85 = type { %26* }
+%86 = type { %27*, i8*, %32*, i8*, i32, %8, i64, i32 }
+%87 = type { %86, %88, %22, %95* }
+%88 = type { %89 }
+%89 = type { %90 }
+%90 = type { %91*, %91*, %91* }
+%91 = type { %92 }
+%92 = type { %93 }
+%93 = type { %94*, %94*, %94* }
+%94 = type { %8, %18, %12*, %9 }
+%95 = type { %37, %42 }
+
+ at _ZNSs4_Rep20_S_empty_rep_storageE = external global [4 x i64] ; <[4 x i64]*> [#uses=1]
+ at .str111723 = external constant [1 x i8], align 1 ; <[1 x i8]*> [#uses=1]
+ at .str181730 = external constant [4 x i8], align 1 ; <[4 x i8]*> [#uses=1]
+ at .str721784 = external constant [37 x i8], align 8 ; <[37 x i8]*> [#uses=1]
+ at _ZN12_GLOBAL__N_16ActionE = external global %0, align 32 ; <%0*> [#uses=1]
+ at _ZN12_GLOBAL__N_114OutputFilenameE = external global %6, align 32 ; <%6*> [#uses=1]
+ at _ZN12_GLOBAL__N_111IncludeDirsE = external global %13, align 32 ; <%13*> [#uses=1]
+ at .str533653 = external constant [2 x i8], align 1 ; <[2 x i8]*> [#uses=1]
+ at _ZN4llvm7RecordsE = external global %21, align 32 ; <%21*> [#uses=2]
+ at _ZL6SrcMgr = external global %27, align 32       ; <%27*> [#uses=2]
+ at .str3723 = external constant [88 x i8], align 8  ; <[88 x i8]*> [#uses=1]
+ at .str13724 = external constant [136 x i8], align 8 ; <[136 x i8]*> [#uses=1]
+
+ at _ZL20__gthrw_pthread_oncePiPFvvE = alias weak i32 (i32*, void ()*)* @pthread_once ; <i32 (i32*, void ()*)*> [#uses=0]
+ at _ZL27__gthrw_pthread_getspecificj = alias weak i8* (i32)* @pthread_getspecific ; <i8* (i32)*> [#uses=0]
+ at _ZL27__gthrw_pthread_setspecificjPKv = alias weak i32 (i32, i8*)* @pthread_setspecific ; <i32 (i32, i8*)*> [#uses=0]
+ at _ZL22__gthrw_pthread_createPmPK14pthread_attr_tPFPvS3_ES3_ = alias weak i32 (i64*, %33*, i8* (i8*)*, i8*)* @pthread_create ; <i32 (i64*, %33*, i8* (i8*)*, i8*)*> [#uses=0]
+ at _ZL22__gthrw_pthread_cancelm = alias weak i32 (i64)* @pthread_cancel ; <i32 (i64)*> [#uses=0]
+ at _ZL26__gthrw_pthread_mutex_lockP15pthread_mutex_t = alias weak i32 (%34*)* @pthread_mutex_lock ; <i32 (%34*)*> [#uses=0]
+ at _ZL29__gthrw_pthread_mutex_trylockP15pthread_mutex_t = alias weak i32 (%34*)* @pthread_mutex_trylock ; <i32 (%34*)*> [#uses=0]
+ at _ZL28__gthrw_pthread_mutex_unlockP15pthread_mutex_t = alias weak i32 (%34*)* @pthread_mutex_unlock ; <i32 (%34*)*> [#uses=0]
+ at _ZL26__gthrw_pthread_mutex_initP15pthread_mutex_tPK19pthread_mutexattr_t = alias weak i32 (%34*, %2*)* @pthread_mutex_init ; <i32 (%34*, %2*)*> [#uses=0]
+ at _ZL26__gthrw_pthread_key_createPjPFvPvE = alias weak i32 (i32*, void (i8*)*)* @pthread_key_create ; <i32 (i32*, void (i8*)*)*> [#uses=0]
+ at _ZL26__gthrw_pthread_key_deletej = alias weak i32 (i32)* @pthread_key_delete ; <i32 (i32)*> [#uses=0]
+ at _ZL30__gthrw_pthread_mutexattr_initP19pthread_mutexattr_t = alias weak i32 (%2*)* @pthread_mutexattr_init ; <i32 (%2*)*> [#uses=0]
+ at _ZL33__gthrw_pthread_mutexattr_settypeP19pthread_mutexattr_ti = alias weak i32 (%2*, i32)* @pthread_mutexattr_settype ; <i32 (%2*, i32)*> [#uses=0]
+ at _ZL33__gthrw_pthread_mutexattr_destroyP19pthread_mutexattr_t = alias weak i32 (%2*)* @pthread_mutexattr_destroy ; <i32 (%2*)*> [#uses=0]
+
+declare void @_ZNSsC1EPKcRKSaIcE(%8*, i8*, %25*)
+
+declare i8* @_Znwm(i64)
+
+declare zeroext i8 @_ZNK4llvm6Record12isSubClassOfENS_9StringRefE(%37*, i64, i64) align 2
+
+declare i32 @_ZNKSs7compareEPKc(%8*, i8*)
+
+declare %45* @_ZN4llvm11raw_ostreamlsEPKc(%45*, i8*) align 2
+
+declare void @_ZNSsC1ERKSs(%8*, %8*)
+
+declare %26* @_ZSt18_Rb_tree_incrementPSt18_Rb_tree_node_base(%26*)
+
+declare void @_ZStplIcSt11char_traitsIcESaIcEESbIT_T0_T1_EPKS3_RKS6_(%8* noalias sret, i8*, %8*)
+
+declare %15* @_ZNSt6vectorISsSaISsEEaSERKS1_(%15*, %15*) align 2
+
+declare void @_ZNSt6vectorISsSaISsEE9push_backERKSs(%15*, %8*) align 2
+
+declare i32 @_ZNK4llvm15TreePatternNode10getTypeNumEj(%46*, i32) align 2
+
+declare void @_ZN4llvm18CodeGenDAGPatternsD1Ev(%53*) align 2
+
+declare void @_ZN4llvm18CodeGenDAGPatternsC1ERNS_12RecordKeeperE(%53*, %21*) align 2
+
+declare void @_ZNK4llvm14PatternToMatch17getPredicateCheckEv(%8* noalias sret, %75*) align 2
+
+define internal void @0(%81*, %45*) align 2 {
+  invoke void @_ZNSsC1ERKSs(%8* undef, %8* null)
+          to label %3 unwind label %28
+
+; <label>:3                                       ; preds = %2
+  %4 = getelementptr inbounds i8* null, i64 -24   ; <i8*> [#uses=1]
+  %5 = icmp eq i8* %4, bitcast ([4 x i64]* @_ZNSs4_Rep20_S_empty_rep_storageE to i8*) ; <i1> [#uses=1]
+  br i1 %5, label %7, label %6
+
+; <label>:6                                       ; preds = %3
+  unreachable
+
+; <label>:7                                       ; preds = %3
+  br i1 undef, label %9, label %8
+
+; <label>:8                                       ; preds = %7
+  invoke void @_ZStplIcSt11char_traitsIcESaIcEESbIT_T0_T1_EPKS3_RKS6_(%8* noalias sret null, i8* getelementptr inbounds ([37 x i8]* @.str721784, i64 0, i64 0), %8* undef)
+          to label %10 unwind label %29
+
+; <label>:9                                       ; preds = %7
+  unreachable
+
+; <label>:10                                      ; preds = %8
+  invoke void @_ZNSsC1ERKSs(%8* null, %8* null)
+          to label %11 unwind label %30
+
+; <label>:11                                      ; preds = %10
+  %12 = invoke %45* @_ZN4llvm11raw_ostreamlsEPKc(%45* %1, i8* getelementptr inbounds ([88 x i8]* @.str3723, i64 0, i64 0))
+          to label %13 unwind label %31           ; <%45*> [#uses=1]
+
+; <label>:13                                      ; preds = %11
+  %14 = invoke %45* @_ZN4llvm11raw_ostreamlsEPKc(%45* %12, i8* getelementptr inbounds ([136 x i8]* @.str13724, i64 0, i64 0))
+          to label %15 unwind label %31           ; <%45*> [#uses=0]
+
+; <label>:15                                      ; preds = %13
+  %16 = getelementptr inbounds i8* null, i64 -24  ; <i8*> [#uses=1]
+  %17 = icmp eq i8* %16, bitcast ([4 x i64]* @_ZNSs4_Rep20_S_empty_rep_storageE to i8*) ; <i1> [#uses=1]
+  br i1 %17, label %19, label %18
+
+; <label>:18                                      ; preds = %15
+  unreachable
+
+; <label>:19                                      ; preds = %15
+  %20 = getelementptr inbounds i8* null, i64 -24  ; <i8*> [#uses=1]
+  %21 = icmp eq i8* %20, bitcast ([4 x i64]* @_ZNSs4_Rep20_S_empty_rep_storageE to i8*) ; <i1> [#uses=1]
+  br i1 %21, label %23, label %22
+
+; <label>:22                                      ; preds = %19
+  unreachable
+
+; <label>:23                                      ; preds = %19
+  invoke void @_ZNSsC1ERKSs(%8* undef, %8* undef)
+          to label %24 unwind label %29
+
+; <label>:24                                      ; preds = %23
+  invoke void @_ZNSsC1ERKSs(%8* undef, %8* undef)
+          to label %26 unwind label %25
+
+; <label>:25                                      ; preds = %24
+  unreachable
+
+; <label>:26                                      ; preds = %24
+  invoke void @f4(%82* undef, %53* undef)
+          to label %27 unwind label %32
+
+; <label>:27                                      ; preds = %26
+  unreachable
+
+; <label>:28                                      ; preds = %2
+  unreachable
+
+; <label>:29                                      ; preds = %23, %8
+  unreachable
+
+; <label>:30                                      ; preds = %10
+  unreachable
+
+; <label>:31                                      ; preds = %13, %11
+  unreachable
+
+; <label>:32                                      ; preds = %26
+  unreachable
+}
+
+declare void @_ZNSt3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS1_EEED1Ev(%22*) align 2
+
+declare void @_ZNSt3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS1_EEEC1ERKS8_(%22*, %22*) align 2
+
+declare %83* @_ZNSt8_Rb_treeIN4llvm3MVT15SimpleValueTypeESt4pairIKS2_St3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaIS3_IKSsS7_EEEESt10_Select1stISE_ES8_IS2_ESaISE_EE14_M_create_nodeERKSE_(%23*, %84*) align 2
+
+declare %22* @_ZNSt3mapIN12_GLOBAL__N_117OperandsSignatureES_ISsS_IN4llvm3MVT15SimpleValueTypeES_IS4_S_ISsNS0_15InstructionMemoESt4lessISsESaISt4pairIKSsS5_EEES6_IS4_ESaIS8_IKS4_SC_EEESD_SaIS8_ISE_SH_EEES7_SaIS8_IS9_SK_EEES6_IS1_ESaIS8_IKS1_SN_EEEixERSP_(%22*, %14*) align 2
+
+define internal %22* @1(%22*, i32*) align 2 {
+  unreachable
+}
+
+declare void @test(i64);
+
+define internal %22* @f3(%22*, i32*) align 2 {
+  %3 = alloca %85, align 8                        ; <%85*> [#uses=1]
+  %4 = alloca %84, align 8                        ; <%84*> [#uses=5]
+  invoke void @_ZNSt3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS1_EEEC1ERKS8_(%22* undef, %22* undef)
+          to label %5 unwind label %30
+
+; <label>:5                                       ; preds = %2
+  %6 = getelementptr inbounds %26* null, i64 1, i32 0 ; <i32*> [#uses=1]
+  %7 = load i32* %6, align 4                      ; <i32> [#uses=0]
+  br i1 false, label %8, label %11
+
+; <label>:8                                       ; preds = %5
+  %9 = invoke %83* @_ZNSt8_Rb_treeIN4llvm3MVT15SimpleValueTypeESt4pairIKS2_St3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaIS3_IKSsS7_EEEESt10_Select1stISE_ES8_IS2_ESaISE_EE14_M_create_nodeERKSE_(%23* undef, %84* %4)
+          to label %10 unwind label %31           ; <%83*> [#uses=0]
+
+; <label>:10                                      ; preds = %8
+  unreachable
+
+; <label>:11                                      ; preds = %5
+  %12 = getelementptr inbounds %85* %3, i64 0, i32 0 ; <%26**> [#uses=1]
+  %13 = load %26** %12, align 8                   ; <%26*> [#uses=0]
+  br i1 false, label %14, label %17
+
+; <label>:14                                      ; preds = %11
+  %15 = invoke %83* @_ZNSt8_Rb_treeIN4llvm3MVT15SimpleValueTypeESt4pairIKS2_St3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaIS3_IKSsS7_EEEESt10_Select1stISE_ES8_IS2_ESaISE_EE14_M_create_nodeERKSE_(%23* undef, %84* %4)
+          to label %16 unwind label %31           ; <%83*> [#uses=0]
+
+; <label>:16                                      ; preds = %14
+  unreachable
+
+; <label>:17                                      ; preds = %11
+  %18 = invoke %26* @_ZSt18_Rb_tree_incrementPSt18_Rb_tree_node_base(%26* undef)
+          to label %19 unwind label %31           ; <%26*> [#uses=0]
+
+; <label>:19                                      ; preds = %17
+  %20 = getelementptr inbounds %84* %4, i64 0, i32 0 ; <i32*> [#uses=1]
+  %21 = load i32* %20, align 8                    ; <i32> [#uses=0]
+  br i1 false, label %22, label %25
+
+; <label>:22                                      ; preds = %19
+  %23 = invoke %83* @_ZNSt8_Rb_treeIN4llvm3MVT15SimpleValueTypeESt4pairIKS2_St3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaIS3_IKSsS7_EEEESt10_Select1stISE_ES8_IS2_ESaISE_EE14_M_create_nodeERKSE_(%23* undef, %84* %4)
+          to label %24 unwind label %31           ; <%83*> [#uses=0]
+
+; <label>:24                                      ; preds = %22
+  unreachable
+
+; <label>:25                                      ; preds = %19
+  %26 = invoke %s2i64 @f_3(%23* undef, %84* %4)
+          to label %l1 unwind label %31
+
+; <label>:27                                      ; preds = %25
+  %eval = extractvalue %s2i64 %inv, 0                   ; <i64> [#uses=0]
+  call void @test(i64 %eval)
+; CHECK: = invoke %s2i64 @f_3
+; CHECK: %eval = extractvalue %s2i64 %inv, 0
+; CHECK-NEXT: call void @test(i64 %eval)
+  invoke void @_ZNSt3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS1_EEED1Ev(%22* undef)
+          to label %29 unwind label %30
+
+l1:
+  %28 = extractvalue %s2i64 %26, 0
+  call void @test(i64 %28)
+;CHECK: call void @test(i64 5)
+;CHECK-NEXT: %inv = invoke %s2i64 @f2
+  %inv = invoke %s2i64 @f2(%23* undef, %84* %4)
+          to label %27 unwind label %31           ; <%s2i64> [#uses=1]
+
+
+; <label>:29                                      ; preds = %27
+  unreachable
+
+; <label>:30                                      ; preds = %27, %2
+  unreachable
+
+; <label>:31                                      ; preds = %25, %22, %17, %14, %8
+  unreachable
+}
+define internal %s2i64 @f2(%23*, %84*) align 2 {
+  br i1 undef, label %3, label %4
+
+; <label>:3                                       ; preds = %2
+  br label %4
+
+; <label>:4                                       ; preds = %3, %2
+  %5 = insertvalue %s2i64 undef, i64 4, 1        ; <%s2i64> [#uses=1]
+  %6 = ptrtoint %84* %1 to i64
+  %7 = insertvalue %s2i64 %5, i64 %6, 0        ; <%s2i64> [#uses=1]
+  ret %s2i64 %7
+}
+define internal %s2i64 @f_3(%23*, %84*) align 2 {
+  br i1 undef, label %3, label %4
+
+; <label>:3                                       ; preds = %2
+  br label %4
+
+; <label>:4                                       ; preds = %3, %2
+  %5 = insertvalue %s2i64 undef, i64 4, 1        ; <%s2i64> [#uses=1]
+  %6 = insertvalue %s2i64 %5, i64 5, 0        ; <%s2i64> [#uses=1]
+  ret %s2i64 %6
+}
+
+
+declare %22* @_ZNSt3mapISsS_IN4llvm3MVT15SimpleValueTypeES_IS2_S_ISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS4_EEES5_IS2_ESaIS7_IKS2_SB_EEESC_SaIS7_ISD_SG_EEES6_SaIS7_IS8_SJ_EEEixERS8_(%22*, %8*) align 2
+
+define internal void @f4(%82*, %53*) align 2 {
+  invoke void @_ZNSsC1ERKSs(%8* undef, %8* undef)
+          to label %3 unwind label %57
+
+; <label>:3                                       ; preds = %2
+  br i1 undef, label %6, label %4
+
+; <label>:4                                       ; preds = %3
+  br i1 undef, label %5, label %6
+
+; <label>:5                                       ; preds = %4
+  br label %6
+
+; <label>:6                                       ; preds = %5, %4, %3
+  br i1 undef, label %8, label %7
+
+; <label>:7                                       ; preds = %6
+  br i1 undef, label %56, label %9
+
+; <label>:8                                       ; preds = %6
+  unreachable
+
+; <label>:9                                       ; preds = %7
+  %10 = icmp eq %12* null, null                   ; <i1> [#uses=1]
+  br i1 %10, label %12, label %11
+
+; <label>:11                                      ; preds = %9
+  unreachable
+
+; <label>:12                                      ; preds = %9
+  %13 = getelementptr inbounds %46* null, i64 0, i32 6, i32 0, i32 0, i32 0 ; <%46***> [#uses=0]
+  %14 = icmp eq i32 0, 0                          ; <i1> [#uses=1]
+  br i1 %14, label %16, label %15
+
+; <label>:15                                      ; preds = %12
+  unreachable
+
+; <label>:16                                      ; preds = %12
+  br i1 undef, label %19, label %17
+
+; <label>:17                                      ; preds = %16
+  %18 = icmp eq %12* null, null                   ; <i1> [#uses=1]
+  br i1 %18, label %21, label %20
+
+; <label>:19                                      ; preds = %16
+  unreachable
+
+; <label>:20                                      ; preds = %17
+  unreachable
+
+; <label>:21                                      ; preds = %17
+  %22 = invoke i32 @_ZNK4llvm15TreePatternNode10getTypeNumEj(%46* null, i32 0)
+          to label %23 unwind label %58           ; <i32> [#uses=0]
+
+; <label>:23                                      ; preds = %21
+  %24 = invoke zeroext i8 @_ZNK4llvm6Record12isSubClassOfENS_9StringRefE(%37* undef, i64 undef, i64 undef)
+          to label %25 unwind label %58           ; <i8> [#uses=0]
+
+; <label>:25                                      ; preds = %23
+  br i1 undef, label %26, label %27
+
+; <label>:26                                      ; preds = %25
+  unreachable
+
+; <label>:27                                      ; preds = %25
+  %28 = invoke i8* @_Znwm(i64 24)
+          to label %29 unwind label %59           ; <i8*> [#uses=0]
+
+; <label>:29                                      ; preds = %27
+  %30 = icmp eq %12* null, null                   ; <i1> [#uses=1]
+  br i1 %30, label %32, label %31
+
+; <label>:31                                      ; preds = %29
+  unreachable
+
+; <label>:32                                      ; preds = %29
+  %33 = invoke i32 @_ZNKSs7compareEPKc(%8* undef, i8* getelementptr inbounds ([4 x i8]* @.str181730, i64 0, i64 0))
+          to label %34 unwind label %59           ; <i32> [#uses=1]
+
+; <label>:34                                      ; preds = %32
+  %35 = icmp eq i32 %33, 0                        ; <i1> [#uses=1]
+  br i1 %35, label %37, label %36
+
+; <label>:36                                      ; preds = %34
+  unreachable
+
+; <label>:37                                      ; preds = %34
+  invoke void @_ZNSsC1EPKcRKSaIcE(%8* undef, i8* getelementptr inbounds ([1 x i8]* @.str111723, i64 0, i64 0), %25* undef)
+          to label %38 unwind label %60
+
+; <label>:38                                      ; preds = %37
+  invoke void @_ZNSt6vectorISsSaISsEE9push_backERKSs(%15* undef, %8* undef)
+          to label %39 unwind label %61
+
+; <label>:39                                      ; preds = %38
+  br i1 undef, label %42, label %40
+
+; <label>:40                                      ; preds = %39
+  br i1 undef, label %41, label %42
+
+; <label>:41                                      ; preds = %40
+  unreachable
+
+; <label>:42                                      ; preds = %40, %39
+  invoke void @_ZNK4llvm14PatternToMatch17getPredicateCheckEv(%8* noalias sret undef, %75* null)
+          to label %43 unwind label %59
+
+; <label>:43                                      ; preds = %42
+  %44 = icmp eq %12* null, null                   ; <i1> [#uses=1]
+  br i1 %44, label %46, label %45
+
+; <label>:45                                      ; preds = %43
+  unreachable
+
+; <label>:46                                      ; preds = %43
+  invoke void @_ZNSsC1ERKSs(%8* undef, %8* undef)
+          to label %47 unwind label %62
+
+; <label>:47                                      ; preds = %46
+  %48 = invoke %22* @_ZNSt3mapIN12_GLOBAL__N_117OperandsSignatureES_ISsS_IN4llvm3MVT15SimpleValueTypeES_IS4_S_ISsNS0_15InstructionMemoESt4lessISsESaISt4pairIKSsS5_EEES6_IS4_ESaIS8_IKS4_SC_EEESD_SaIS8_ISE_SH_EEES7_SaIS8_IS9_SK_EEES6_IS1_ESaIS8_IKS1_SN_EEEixERSP_(%22* undef, %14* undef)
+          to label %49 unwind label %63           ; <%22*> [#uses=1]
+
+; <label>:49                                      ; preds = %47
+  %50 = invoke %22* @_ZNSt3mapISsS_IN4llvm3MVT15SimpleValueTypeES_IS2_S_ISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS4_EEES5_IS2_ESaIS7_IKS2_SB_EEESC_SaIS7_ISD_SG_EEES6_SaIS7_IS8_SJ_EEEixERS8_(%22* %48, %8* undef)
+          to label %51 unwind label %63           ; <%22*> [#uses=1]
+
+; <label>:51                                      ; preds = %49
+  %52 = invoke %22* @1(%22* %50, i32* undef)
+          to label %53 unwind label %63           ; <%22*> [#uses=1]
+
+; <label>:53                                      ; preds = %51
+  %54 = invoke %22* @f3(%22* %52, i32* undef)
+          to label %55 unwind label %63           ; <%22*> [#uses=0]
+
+; <label>:55                                      ; preds = %53
+  unreachable
+
+; <label>:56                                      ; preds = %7
+  ret void
+
+; <label>:57                                      ; preds = %2
+  unreachable
+
+; <label>:58                                      ; preds = %23, %21
+  unreachable
+
+; <label>:59                                      ; preds = %42, %32, %27
+  unreachable
+
+; <label>:60                                      ; preds = %37
+  unreachable
+
+; <label>:61                                      ; preds = %38
+  unreachable
+
+; <label>:62                                      ; preds = %46
+  unreachable
+
+; <label>:63                                      ; preds = %53, %51, %49, %47
+  unreachable
+}
+
+declare i32 @_ZN4llvm7TGLexer8LexTokenEv(%86*) align 2
+
+declare void @_ZN4llvm8TGParserD1Ev(%87*) align 2
+
+declare i32 @_ZN4llvm9SourceMgr18AddNewSourceBufferEPNS_12MemoryBufferENS_5SMLocE(%27*, %32*, %9* noalias) align 2
+
+define i32 @main(i32, i8**) {
+  br i1 undef, label %3, label %4
+
+; <label>:3                                       ; preds = %2
+  unreachable
+
+; <label>:4                                       ; preds = %2
+  %5 = icmp eq i32 0, 0                           ; <i1> [#uses=1]
+  br i1 %5, label %7, label %6
+
+; <label>:6                                       ; preds = %4
+  unreachable
+
+; <label>:7                                       ; preds = %4
+  br i1 undef, label %8, label %9
+
+; <label>:8                                       ; preds = %7
+  br i1 undef, label %10, label %9
+
+; <label>:9                                       ; preds = %8, %7
+  br i1 undef, label %14, label %18
+
+; <label>:10                                      ; preds = %8
+  br label %11
+
+; <label>:11                                      ; preds = %12, %10
+  br i1 undef, label %12, label %13
+
+; <label>:12                                      ; preds = %11
+  br label %11
+
+; <label>:13                                      ; preds = %11
+  unreachable
+
+; <label>:14                                      ; preds = %9
+  br i1 undef, label %15, label %17
+
+; <label>:15                                      ; preds = %14
+  br i1 undef, label %17, label %16
+
+; <label>:16                                      ; preds = %15
+  unreachable
+
+; <label>:17                                      ; preds = %15, %14
+  unreachable
+
+; <label>:18                                      ; preds = %9
+  %19 = invoke i32 @_ZN4llvm9SourceMgr18AddNewSourceBufferEPNS_12MemoryBufferENS_5SMLocE(%27* @_ZL6SrcMgr, %32* undef, %9* noalias undef)
+          to label %20 unwind label %34           ; <i32> [#uses=0]
+
+; <label>:20                                      ; preds = %18
+  %21 = invoke %15* @_ZNSt6vectorISsSaISsEEaSERKS1_(%15* getelementptr inbounds (%27* @_ZL6SrcMgr, i64 0, i32 1), %15* getelementptr inbounds (%13* @_ZN12_GLOBAL__N_111IncludeDirsE, i64 0, i32 1, i32 0))
+          to label %22 unwind label %34           ; <%15*> [#uses=0]
+
+; <label>:22                                      ; preds = %20
+  %23 = getelementptr inbounds %27* null, i64 0, i32 0, i32 0, i32 0, i32 1 ; <%31**> [#uses=1]
+  %24 = load %31** %23, align 8                   ; <%31*> [#uses=1]
+  %25 = ptrtoint %31* %24 to i64                  ; <i64> [#uses=1]
+  %26 = sub i64 %25, 0                            ; <i64> [#uses=1]
+  %27 = icmp ugt i64 %26, 15                      ; <i1> [#uses=1]
+  br i1 %27, label %29, label %28
+
+; <label>:28                                      ; preds = %22
+  unreachable
+
+; <label>:29                                      ; preds = %22
+  %30 = invoke i32 @_ZN4llvm7TGLexer8LexTokenEv(%86* undef)
+          to label %31 unwind label %35           ; <i32> [#uses=0]
+
+; <label>:31                                      ; preds = %29
+  invoke void @_ZN4llvm8TGParserD1Ev(%87* null)
+          to label %32 unwind label %34
+
+; <label>:32                                      ; preds = %31
+  %33 = icmp eq i8 0, 0                           ; <i1> [#uses=1]
+  br i1 %33, label %36, label %56
+
+; <label>:34                                      ; preds = %31, %20, %18
+  unreachable
+
+; <label>:35                                      ; preds = %29
+  unreachable
+
+; <label>:36                                      ; preds = %32
+  %37 = invoke i32 @_ZNKSs7compareEPKc(%8* getelementptr inbounds (%6* @_ZN12_GLOBAL__N_114OutputFilenameE, i64 0, i32 1, i32 0), i8* getelementptr inbounds ([2 x i8]* @.str533653, i64 0, i64 0))
+          to label %38 unwind label %57           ; <i32> [#uses=1]
+
+; <label>:38                                      ; preds = %36
+  %39 = icmp eq i32 %37, 0                        ; <i1> [#uses=1]
+  br i1 %39, label %43, label %40
+
+; <label>:40                                      ; preds = %38
+  %41 = invoke i8* @_Znwm(i64 56)
+          to label %42 unwind label %58           ; <i8*> [#uses=0]
+
+; <label>:42                                      ; preds = %40
+  unreachable
+
+; <label>:43                                      ; preds = %38
+  %44 = load i32* getelementptr inbounds (%0* @_ZN12_GLOBAL__N_16ActionE, i64 0, i32 1, i32 0), align 8 ; <i32> [#uses=1]
+  switch i32 %44, label %56 [
+    i32 0, label %45
+    i32 12, label %48
+    i32 13, label %51
+  ]
+
+; <label>:45                                      ; preds = %43
+  br i1 undef, label %46, label %47
+
+; <label>:46                                      ; preds = %45
+  unreachable
+
+; <label>:47                                      ; preds = %45
+  unreachable
+
+; <label>:48                                      ; preds = %43
+  invoke void @_ZN4llvm18CodeGenDAGPatternsC1ERNS_12RecordKeeperE(%53* undef, %21* @_ZN4llvm7RecordsE)
+          to label %50 unwind label %49
+
+; <label>:49                                      ; preds = %48
+  unreachable
+
+; <label>:50                                      ; preds = %48
+  unreachable
+
+; <label>:51                                      ; preds = %43
+  invoke void @_ZN4llvm18CodeGenDAGPatternsC1ERNS_12RecordKeeperE(%53* undef, %21* @_ZN4llvm7RecordsE)
+          to label %53 unwind label %52
+
+; <label>:52                                      ; preds = %51
+  unreachable
+
+; <label>:53                                      ; preds = %51
+  invoke void @0(%81* undef, %45* null)
+          to label %54 unwind label %60
+
+; <label>:54                                      ; preds = %53
+  invoke void @_ZN4llvm18CodeGenDAGPatternsD1Ev(%53* undef)
+          to label %55 unwind label %59
+
+; <label>:55                                      ; preds = %54
+  unreachable
+
+; <label>:56                                      ; preds = %43, %32
+  ret i32 1
+
+; <label>:57                                      ; preds = %36
+  unreachable
+
+; <label>:58                                      ; preds = %40
+  unreachable
+
+; <label>:59                                      ; preds = %54
+  unreachable
+
+; <label>:60                                      ; preds = %53
+  unreachable
+}
+
+declare i8* @pthread_getspecific(i32) nounwind
+
+declare i32 @pthread_setspecific(i32, i8*) nounwind
+
+declare i32 @pthread_key_delete(i32) nounwind
+
+declare i32 @pthread_key_create(i32*, void (i8*)*) nounwind
+
+declare i32 @pthread_mutex_trylock(%34*) nounwind
+
+declare i32 @pthread_mutex_unlock(%34*) nounwind
+
+declare i32 @pthread_mutex_lock(%34*) nounwind
+
+declare i32 @pthread_mutexattr_init(%2*) nounwind
+
+declare i32 @pthread_mutexattr_settype(%2*, i32) nounwind
+
+declare i32 @pthread_mutex_init(%34*, %2*) nounwind
+
+declare i32 @pthread_mutexattr_destroy(%2*) nounwind
+
+declare extern_weak i32 @pthread_once(i32*, void ()*)
+
+declare extern_weak i32 @pthread_create(i64*, %33*, i8* (i8*)*, i8*)
+
+declare extern_weak i32 @pthread_cancel(i64)




From edwintorok at gmail.com  Tue Oct 20 10:42:00 2009
From: edwintorok at gmail.com (Torok Edwin)
Date: Tue, 20 Oct 2009 15:42:00 -0000
Subject: [llvm-commits] [llvm] r84638 - in /llvm/trunk:
 lib/Transforms/Utils/SSAUpdater.cpp
 test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll
Message-ID: <200910201542.n9KFg1HB005101@zion.cs.uiuc.edu>

Author: edwin
Date: Tue Oct 20 10:42:00 2009
New Revision: 84638

URL: http://llvm.org/viewvc/llvm-project?rev=84638&view=rev
Log:
Fix PR5258, jump-threading creating invalid PHIs.
When an incoming value for a PHI is updated, we must also updated all other
incoming values for the same BB to match, otherwise we create invalid PHIs.

Added:
    llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll
Modified:
    llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp

Modified: llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp?rev=84638&r1=84637&r2=84638&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp (original)
+++ llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp Tue Oct 20 10:42:00 2009
@@ -178,10 +178,18 @@
 void SSAUpdater::RewriteUse(Use &U) {
   Instruction *User = cast<Instruction>(U.getUser());
   BasicBlock *UseBB = User->getParent();
-  if (PHINode *UserPN = dyn_cast<PHINode>(User))
+  PHINode *UserPN = dyn_cast<PHINode>(User);
+  if (UserPN)
     UseBB = UserPN->getIncomingBlock(U);
 
-  U.set(GetValueInMiddleOfBlock(UseBB));
+  Value *V = GetValueInMiddleOfBlock(UseBB);
+  U.set(V);
+  if (UserPN) {
+    // Incoming value from the same BB must be consistent
+    for (unsigned i=0;i<UserPN->getNumIncomingValues();i++)
+      if (UserPN->getIncomingBlock(i) == UseBB)
+        UserPN->setIncomingValue(i, V);
+  }
 }
 
 

Added: llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll?rev=84638&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll (added)
+++ llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll Tue Oct 20 10:42:00 2009
@@ -0,0 +1,348 @@
+; RUN: opt -jump-threading -verify %s -disable-output
+; PR5258
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%0 = type { i64, [12 x i32] }
+%1 = type { %2 }
+%2 = type { i32, i32, i32, i32, i32, i32, %3 }
+%3 = type { %3*, %3* }
+%4 = type { i32 }
+%5 = type { %6*, %84, %87, %89, %92, %95, %69, %79*, %102*, %69, %69, %69 }
+%6 = type { %7* }
+%7 = type { %8, %10, %28, %34, %38, %40, %40, %40, %40, %40, %40, %40, %21*, %21*, %45, %46, %45, %45, %15, %15, %15, %15, %15, %15, %15, %15, %49, %49, %49, %49, %49, %9, %9, %51, %51, %51, %51, %51, %51, %53, %56 }
+%8 = type { %9, i32, i32 }
+%9 = type { i8* }
+%10 = type { i32, %11*, i32, i32 }
+%11 = type { %12, %21* }
+%12 = type { %13, %15* }
+%13 = type { i32, %14 }
+%14 = type { i64 }
+%15 = type { %16, i8, [3 x i8], i32, %6*, %15*, %17, i32, %20* }
+%16 = type { i32 (...)** }
+%17 = type { %18 }
+%18 = type { %19 }
+%19 = type { %16**, %16**, %16** }
+%20 = type { %15*, %16* }
+%21 = type { %22, %13 }
+%22 = type { %23 }
+%23 = type { %24, %26*, i32 }
+%24 = type { i32 (...)**, i8, i8, i8, i16, %25, %26*, %27* }
+%25 = type { %15* }
+%26 = type { %24*, %26*, %14 }
+%27 = type { %4, %24* }
+%28 = type { i32, %29*, i32, i32 }
+%29 = type { %30, %33* }
+%30 = type { %31 }
+%31 = type { %32*, %14, i16, i16 }
+%32 = type { i16, i16, i32, i32 }
+%33 = type { %22, %31 }
+%34 = type { %35, %37 }
+%35 = type { %36*, i32, i32, i32, i32 }
+%36 = type { i32, %4* }
+%37 = type <{ i8 }>
+%38 = type { %39 }
+%39 = type { i32 (...)**, i8**, i32, i32 }
+%40 = type { %16, %41, %41, %41, %45 }
+%41 = type { %42 }
+%42 = type { %43 }
+%43 = type { %37, %44, i64 }
+%44 = type { i32, %44*, %44*, %44* }
+%45 = type { %9, i32, i8 }
+%46 = type { %47, %24*, i8* }
+%47 = type { %48, [8 x i8*] }
+%48 = type { i8**, i32, i32, i32, [1 x i8*] }
+%49 = type { %50 }
+%50 = type { %15 }
+%51 = type { %52, %41 }
+%52 = type { %41 }
+%53 = type { i32, %54*, i32, i32 }
+%54 = type { %24*, %55* }
+%55 = type { %14, %55*, %24* }
+%56 = type { %57, %34 }
+%57 = type { i32, %58*, i32, i32 }
+%58 = type <{ %59*, [8 x i8], %83 }>
+%59 = type { %23, %60, %62* }
+%60 = type { %61, %59* }
+%61 = type { %59* }
+%62 = type { %24, %63, %65, %67* }
+%63 = type { %64, %62* }
+%64 = type { %62* }
+%65 = type { %66, %59* }
+%66 = type { %61 }
+%67 = type { %68, %70, %72, %74, %79*, %80 }
+%68 = type { %22, %5*, i32, %69 }
+%69 = type { %9 }
+%70 = type { %71, %67* }
+%71 = type { %67* }
+%72 = type { %73, %62* }
+%73 = type { %64 }
+%74 = type { %75, %77* }
+%75 = type { %76 }
+%76 = type { %77* }
+%77 = type { %24, %78, %67* }
+%78 = type { %76, %77* }
+%79 = type { %34, i32 }
+%80 = type { %81* }
+%81 = type <{ %9, i32, [4 x i8], %82 }>
+%82 = type <{ [33 x i8], [31 x i8] }>
+%83 = type <{ [33 x i8], [63 x i8] }>
+%84 = type { %85* }
+%85 = type { %68, %86, i8 }
+%86 = type { %84, %85* }
+%87 = type { %88, %67* }
+%88 = type { %70 }
+%89 = type { %90* }
+%90 = type { %68, %91 }
+%91 = type { %89, %90* }
+%92 = type { %93 }
+%93 = type { %94 }
+%94 = type { %69*, %69*, %69* }
+%95 = type { %96, %99* }
+%96 = type { %97 }
+%97 = type { %98, %99* }
+%98 = type { %99* }
+%99 = type <{ %100, %97, %5*, %101 }>
+%100 = type { [52 x i8], i32 }
+%101 = type <{ [33 x i8], [95 x i8] }>
+%102 = type { %16, %41, i32 }
+
+ at _ZL20__gthrw_pthread_oncePiPFvvE = alias weak i32 (i32*, void ()*)* @pthread_once ; <i32 (i32*, void ()*)*> [#uses=0]
+ at _ZL27__gthrw_pthread_getspecificj = alias weak i8* (i32)* @pthread_getspecific ; <i8* (i32)*> [#uses=0]
+ at _ZL27__gthrw_pthread_setspecificjPKv = alias weak i32 (i32, i8*)* @pthread_setspecific ; <i32 (i32, i8*)*> [#uses=0]
+ at _ZL22__gthrw_pthread_createPmPK14pthread_attr_tPFPvS3_ES3_ = alias weak i32 (i64*, %0*, i8* (i8*)*, i8*)* @pthread_create ; <i32 (i64*, %0*, i8* (i8*)*, i8*)*> [#uses=0]
+ at _ZL22__gthrw_pthread_cancelm = alias weak i32 (i64)* @pthread_cancel ; <i32 (i64)*> [#uses=0]
+ at _ZL26__gthrw_pthread_mutex_lockP15pthread_mutex_t = alias weak i32 (%1*)* @pthread_mutex_lock ; <i32 (%1*)*> [#uses=0]
+ at _ZL29__gthrw_pthread_mutex_trylockP15pthread_mutex_t = alias weak i32 (%1*)* @pthread_mutex_trylock ; <i32 (%1*)*> [#uses=0]
+ at _ZL28__gthrw_pthread_mutex_unlockP15pthread_mutex_t = alias weak i32 (%1*)* @pthread_mutex_unlock ; <i32 (%1*)*> [#uses=0]
+ at _ZL26__gthrw_pthread_mutex_initP15pthread_mutex_tPK19pthread_mutexattr_t = alias weak i32 (%1*, %4*)* @pthread_mutex_init ; <i32 (%1*, %4*)*> [#uses=0]
+ at _ZL26__gthrw_pthread_key_createPjPFvPvE = alias weak i32 (i32*, void (i8*)*)* @pthread_key_create ; <i32 (i32*, void (i8*)*)*> [#uses=0]
+ at _ZL26__gthrw_pthread_key_deletej = alias weak i32 (i32)* @pthread_key_delete ; <i32 (i32)*> [#uses=0]
+ at _ZL30__gthrw_pthread_mutexattr_initP19pthread_mutexattr_t = alias weak i32 (%4*)* @pthread_mutexattr_init ; <i32 (%4*)*> [#uses=0]
+ at _ZL33__gthrw_pthread_mutexattr_settypeP19pthread_mutexattr_ti = alias weak i32 (%4*, i32)* @pthread_mutexattr_settype ; <i32 (%4*, i32)*> [#uses=0]
+ at _ZL33__gthrw_pthread_mutexattr_destroyP19pthread_mutexattr_t = alias weak i32 (%4*)* @pthread_mutexattr_destroy ; <i32 (%4*)*> [#uses=0]
+
+define fastcc zeroext i8 @_ZN4llvm6Linker11LinkModulesEPNS_6ModuleES2_PSs(%5*, %5*, %69*) nounwind align 2 {
+  br i1 false, label %4, label %5
+
+; <label>:4                                       ; preds = %3
+  unreachable
+
+; <label>:5                                       ; preds = %3
+  br i1 false, label %6, label %7
+
+; <label>:6                                       ; preds = %5
+  unreachable
+
+; <label>:7                                       ; preds = %5
+  br i1 false, label %8, label %11
+
+; <label>:8                                       ; preds = %7
+  br i1 undef, label %10, label %9
+
+; <label>:9                                       ; preds = %8
+  unreachable
+
+; <label>:10                                      ; preds = %8
+  unreachable
+
+; <label>:11                                      ; preds = %7
+  br i1 undef, label %13, label %12
+
+; <label>:12                                      ; preds = %11
+  unreachable
+
+; <label>:13                                      ; preds = %11
+  br i1 undef, label %15, label %14
+
+; <label>:14                                      ; preds = %13
+  br label %15
+
+; <label>:15                                      ; preds = %14, %13
+  br i1 undef, label %17, label %16
+
+; <label>:16                                      ; preds = %15
+  unreachable
+
+; <label>:17                                      ; preds = %15
+  br i1 undef, label %19, label %18
+
+; <label>:18                                      ; preds = %17
+  unreachable
+
+; <label>:19                                      ; preds = %17
+  br i1 false, label %20, label %21
+
+; <label>:20                                      ; preds = %19
+  unreachable
+
+; <label>:21                                      ; preds = %19
+  br i1 undef, label %22, label %23
+
+; <label>:22                                      ; preds = %21
+  br label %23
+
+; <label>:23                                      ; preds = %22, %21
+  br i1 false, label %24, label %25
+
+; <label>:24                                      ; preds = %23
+  unreachable
+
+; <label>:25                                      ; preds = %23
+  br i1 undef, label %29, label %26
+
+; <label>:26                                      ; preds = %25
+  br i1 undef, label %28, label %27
+
+; <label>:27                                      ; preds = %26
+  unreachable
+
+; <label>:28                                      ; preds = %26
+  unreachable
+
+; <label>:29                                      ; preds = %25
+  br i1 undef, label %31, label %30
+
+; <label>:30                                      ; preds = %29
+  unreachable
+
+; <label>:31                                      ; preds = %29
+  br i1 undef, label %32, label %33
+
+; <label>:32                                      ; preds = %31
+  br label %33
+
+; <label>:33                                      ; preds = %32, %31
+  br i1 false, label %34, label %35
+
+; <label>:34                                      ; preds = %33
+  unreachable
+
+; <label>:35                                      ; preds = %33
+  br i1 undef, label %36, label %37
+
+; <label>:36                                      ; preds = %35
+  br label %37
+
+; <label>:37                                      ; preds = %36, %35
+  br i1 undef, label %39, label %38
+
+; <label>:38                                      ; preds = %37
+  br i1 false, label %39, label %40
+
+; <label>:39                                      ; preds = %38, %37
+  unreachable
+
+; <label>:40                                      ; preds = %38
+  %41 = load i8* undef, align 8                   ; <i8> [#uses=1]
+  switch i8 %41, label %42 [
+    i8 4, label %43
+    i8 2, label %43
+    i8 3, label %43
+  ]
+
+; <label>:42                                      ; preds = %40
+  unreachable
+
+; <label>:43                                      ; preds = %40, %40, %40
+  %44 = trunc i32 undef to i5                     ; <i5> [#uses=1]
+  switch i5 %44, label %45 [
+    i5 7, label %50
+    i5 9, label %50
+  ]
+
+; <label>:45                                      ; preds = %43
+  br i1 undef, label %47, label %46
+
+; <label>:46                                      ; preds = %45
+  br label %47
+
+; <label>:47                                      ; preds = %46, %45
+  %48 = icmp eq %85* null, null                   ; <i1> [#uses=1]
+  br i1 %48, label %50, label %49
+
+; <label>:49                                      ; preds = %47
+  unreachable
+
+; <label>:50                                      ; preds = %47, %43, %43
+  %51 = phi %68* [ null, %43 ], [ undef, %47 ], [ null, %43 ] ; <%68*> [#uses=4]
+  %52 = phi %68* [ null, %43 ], [ undef, %47 ], [ null, %43 ] ; <%68*> [#uses=1]
+  %53 = icmp eq %68* %52, null                    ; <i1> [#uses=1]
+  br i1 %53, label %54, label %59
+
+; <label>:54                                      ; preds = %50
+  %55 = trunc i32 undef to i5                     ; <i5> [#uses=1]
+  switch i5 %55, label %56 [
+    i5 7, label %59
+    i5 8, label %59
+    i5 9, label %59
+  ]
+
+; <label>:56                                      ; preds = %54
+  br i1 undef, label %58, label %57
+
+; <label>:57                                      ; preds = %56
+  br label %58
+
+; <label>:58                                      ; preds = %57, %56
+  br label %59
+
+; <label>:59                                      ; preds = %58, %54, %54, %54, %50
+  %60 = phi %68* [ %51, %50 ], [ %51, %54 ], [ %51, %54 ], [ %51, %54 ], [ undef, %58 ] ; <%68*> [#uses=0]
+  br i1 undef, label %62, label %61
+
+; <label>:61                                      ; preds = %59
+  br label %62
+
+; <label>:62                                      ; preds = %61, %59
+  switch i8 undef, label %64 [
+    i8 3, label %63
+    i8 2, label %65
+  ]
+
+; <label>:63                                      ; preds = %62
+  unreachable
+
+; <label>:64                                      ; preds = %62
+  unreachable
+
+; <label>:65                                      ; preds = %62
+  switch i8 undef, label %67 [
+    i8 0, label %66
+    i8 1, label %66
+  ]
+
+; <label>:66                                      ; preds = %65, %65
+  unreachable
+
+; <label>:67                                      ; preds = %65
+  unreachable
+}
+
+declare i32 @pthread_mutex_trylock(%1*) nounwind
+
+declare i32 @pthread_mutex_unlock(%1*) nounwind
+
+declare i32 @pthread_mutex_lock(%1*) nounwind
+
+declare i32 @pthread_mutexattr_init(%4*) nounwind
+
+declare i32 @pthread_mutexattr_settype(%4*, i32) nounwind
+
+declare i32 @pthread_mutex_init(%1*, %4*) nounwind
+
+declare i32 @pthread_mutexattr_destroy(%4*) nounwind
+
+declare i8* @pthread_getspecific(i32) nounwind
+
+declare i32 @pthread_setspecific(i32, i8*) nounwind
+
+declare i32 @pthread_key_delete(i32) nounwind
+
+declare i32 @pthread_key_create(i32*, void (i8*)*) nounwind
+
+declare i32 @pthread_once(i32*, void ()*)
+
+declare i32 @pthread_create(i64*, %0*, i8* (i8*)*, i8*)
+
+declare i32 @pthread_cancel(i64)




From natebegeman at mac.com  Tue Oct 20 10:51:52 2009
From: natebegeman at mac.com (Nate Begeman)
Date: Tue, 20 Oct 2009 08:51:52 -0700
Subject: [llvm-commits] [PATCH] Move the Function*->allocated blocks map
 from the JITMemoryManager to the JITEmitter
In-Reply-To: <001636283fda0ee7fb0476589186@google.com>
References: <001636283fda0ee7fb0476589186@google.com>
Message-ID: <F7F04496-C1EE-40E3-B610-1854B22C86DC@mac.com>

Looks good to me.

On Oct 20, 2009, at 12:11 AM, jyasskin at gmail.com wrote:

> Reviewers: natebegeman_mac.com,
>
> Message:
> Please take a look.
>
> Description:
> I'm making Functions auto-remove themselves from the JIT when they're
> destroyed. In this case, the Function needs to be removed from the
> JITEmitter, but the map recording which Functions need to be removed
> lived behind the JITMemoryManager interface, which made things
> difficult.
>
> This patch replaces the deallocateMemForFunction(Function*) method  
> with
> a pair of methods deallocateFunctionBody(void *) and
> deallocateExceptionTable(void *) corresponding to the two
> startFoo/endFoo pairs.
>
> Patch at http://codereview.appspot.com/download/issue135055_1.diff.  
> This
> slightly depends on http://codereview.appspot.com/132055 for  
> ValueMap<>,
> but I can remove that if this gets approved first.
>
> Please review this at http://codereview.appspot.com/135055
>
> Affected files:
>  M include/llvm/ExecutionEngine/JITMemoryManager.h
>  M lib/ExecutionEngine/JIT/JITEmitter.cpp
>  M lib/ExecutionEngine/JIT/JITMemoryManager.cpp
>  M unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
>
>



From gohman at apple.com  Tue Oct 20 11:22:37 2009
From: gohman at apple.com (Dan Gohman)
Date: Tue, 20 Oct 2009 16:22:37 -0000
Subject: [llvm-commits] [llvm] r84640 - in /llvm/trunk:
 lib/Target/X86/X86ISelLowering.cpp
 test/CodeGen/X86/legalize-fmp-oeq-vector-select.ll
Message-ID: <200910201622.n9KGMcMG006764@zion.cs.uiuc.edu>

Author: djg
Date: Tue Oct 20 11:22:37 2009
New Revision: 84640

URL: http://llvm.org/viewvc/llvm-project?rev=84640&view=rev
Log:
Make TranslateX86CC return COND_INVALID instead of aborting when it
encounters an OEQ or UNE comparison, and update its callers to check
for this return status and recover. This fixes a problem resulting from
the LowerOperation hooks being called from LegalizeVectorOps, because
LegalizeVectorOps only lowers vectors, so OEQ and UNE comparisons may
still be at large. This fixes PR5092.

Added:
    llvm/trunk/test/CodeGen/X86/legalize-fmp-oeq-vector-select.ll
Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=84640&r1=84639&r2=84640&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Tue Oct 20 11:22:37 2009
@@ -2286,6 +2286,8 @@
   case ISD::SETNE:   return X86::COND_NE;
   case ISD::SETUO:   return X86::COND_P;
   case ISD::SETO:    return X86::COND_NP;
+  case ISD::SETOEQ:
+  case ISD::SETUNE:  return X86::COND_INVALID;
   }
 }
 
@@ -5566,6 +5568,8 @@
 
   bool isFP = Op.getOperand(1).getValueType().isFloatingPoint();
   unsigned X86CC = TranslateX86CC(CC, isFP, Op0, Op1, DAG);
+  if (X86CC == X86::COND_INVALID)
+    return SDValue();
 
   SDValue Cond = EmitCmp(Op0, Op1, X86CC, DAG);
   return DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
@@ -5714,8 +5718,11 @@
   DebugLoc dl = Op.getDebugLoc();
   SDValue CC;
 
-  if (Cond.getOpcode() == ISD::SETCC)
-    Cond = LowerSETCC(Cond, DAG);
+  if (Cond.getOpcode() == ISD::SETCC) {
+    SDValue NewCond = LowerSETCC(Cond, DAG);
+    if (NewCond.getNode())
+      Cond = NewCond;
+  }
 
   // If condition flag is set by a X86ISD::CMP, then use it as the condition
   // setting operand in place of the X86ISD::SETCC.
@@ -5788,8 +5795,11 @@
   DebugLoc dl = Op.getDebugLoc();
   SDValue CC;
 
-  if (Cond.getOpcode() == ISD::SETCC)
-    Cond = LowerSETCC(Cond, DAG);
+  if (Cond.getOpcode() == ISD::SETCC) {
+    SDValue NewCond = LowerSETCC(Cond, DAG);
+    if (NewCond.getNode())
+      Cond = NewCond;
+  }
 #if 0
   // FIXME: LowerXALUO doesn't handle these!!
   else if (Cond.getOpcode() == X86ISD::ADD  ||
@@ -6338,6 +6348,7 @@
     SDValue LHS = Op.getOperand(1);
     SDValue RHS = Op.getOperand(2);
     unsigned X86CC = TranslateX86CC(CC, true, LHS, RHS, DAG);
+    assert(X86CC != X86::COND_INVALID && "Unexpected illegal condition!");
     SDValue Cond = DAG.getNode(Opc, dl, MVT::i32, LHS, RHS);
     SDValue SetCC = DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
                                 DAG.getConstant(X86CC, MVT::i8), Cond);

Added: llvm/trunk/test/CodeGen/X86/legalize-fmp-oeq-vector-select.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/legalize-fmp-oeq-vector-select.ll?rev=84640&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/X86/legalize-fmp-oeq-vector-select.ll (added)
+++ llvm/trunk/test/CodeGen/X86/legalize-fmp-oeq-vector-select.ll Tue Oct 20 11:22:37 2009
@@ -0,0 +1,11 @@
+; RUN: llc -march=x86-64 -enable-legalize-types-checking < %s
+; PR5092
+
+define <4 x float> @bug(float %a) nounwind {
+entry:
+  %cmp = fcmp oeq float %a, 0.000000e+00          ; <i1> [#uses=1]
+  %temp = select i1 %cmp, <4 x float> <float 1.000000e+00, float 0.000000e+00,
+float 0.000000e+00, float 0.000000e+00>, <4 x float> zeroinitializer
+  ret <4 x float> %temp
+}
+




From grosbach at apple.com  Tue Oct 20 11:33:57 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Tue, 20 Oct 2009 16:33:57 -0000
Subject: [llvm-commits] [llvm] r84641 -
	/llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp
Message-ID: <200910201633.n9KGXv3N007261@zion.cs.uiuc.edu>

Author: grosbach
Date: Tue Oct 20 11:33:57 2009
New Revision: 84641

URL: http://llvm.org/viewvc/llvm-project?rev=84641&view=rev
Log:
Register re-use for scavenged frame indices must check for re-deginition
of the register in the instruction which kills the scavenged value.

Modified:
    llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp

Modified: llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp?rev=84641&r1=84640&r2=84641&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp (original)
+++ llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp Tue Oct 20 11:33:57 2009
@@ -873,6 +873,10 @@
             PrevLastUseOp = i;
             CurrentScratchReg = CurrentVirtReg = 0;
             havePrevValue = trackingCurrentValue;
+            // Re-scan the operands of this instruction to catch definitions
+            // of the scratch register we're using. This is to handle things
+            // like ldr "r2, [scratch]" where scratch is r2.
+            i = 0;
           }
         }
       RS->forward(MI);




From johnny.chen at apple.com  Tue Oct 20 12:49:23 2009
From: johnny.chen at apple.com (Johnny Chen)
Date: Tue, 20 Oct 2009 10:49:23 -0700
Subject: [llvm-commits] ARM .td patch to specify only the well-known
	encoding bits for imm6 and imm4 fields
Message-ID: <E0C31726-F049-46CB-B443-8C01B7FF9AE0@apple.com>

Hi,

Refs: A8-754 (VSHLL and friends), A8-580 (VCVT between floating-point and fixed-point, Advanced SIMD), A8-592 (VDUP scalar)

This patch attempts to specify just the encoding bits described in the manual for several vector shift, vector convert, and vector dupe
operations.

One question I have is whether we should explicitly specify that VCVT's imm6 is of the pattern '1xxxxx'.
In the submitted patch, I left the imm6 (Inst{21-16}) of VCVT unspecified.

Thanks.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: ARM-NEON-IMM.patch
Type: application/octet-stream
Size: 41500 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091020/da05a2ff/attachment.obj 

From tonic at nondot.org  Tue Oct 20 13:04:18 2009
From: tonic at nondot.org (Tanya Lattner)
Date: Tue, 20 Oct 2009 13:04:18 -0500
Subject: [llvm-commits] CVS: llvm-www/www-index.html
Message-ID: <200910201804.n9KI4IFK010773@zion.cs.uiuc.edu>



Changes in directory llvm-www:

www-index.html updated: 1.182 -> 1.183
---
Log message:

Update release schedule.


---
Diffs of the changes:  (+3 -3)

 www-index.html |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)


Index: llvm-www/www-index.html
diff -u llvm-www/www-index.html:1.182 llvm-www/www-index.html:1.183
--- llvm-www/www-index.html:1.182	Wed Sep 30 11:44:47 2009
+++ llvm-www/www-index.html	Tue Oct 20 13:01:54 2009
@@ -108,9 +108,9 @@
 <li>Aug  21 - Code Freeze</li>
 <li>Aug  28 - Pre-release1 testing begins</li>
 <li>Sept 04 - Pre-release1 testing ends</li>
-<li>TBD - Pre-release2 testing begins</li>
-<li>TBD  - Pre-release2 testing ends</li>
-<li>TBD  - Release!</li>
+<li>Oct  17 - Pre-release2 testing begins</li>
+<li>Oct  21  - Pre-release2 testing ends</li>
+<li>Oct  23  - Release!</li>
 </ul>
 </div>
 





From jyasskin at google.com  Tue Oct 20 13:13:21 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Tue, 20 Oct 2009 18:13:21 -0000
Subject: [llvm-commits] [llvm] r84651 - in /llvm/trunk:
 include/llvm/ExecutionEngine/JITMemoryManager.h
 lib/ExecutionEngine/JIT/JITEmitter.cpp
 lib/ExecutionEngine/JIT/JITMemoryManager.cpp
 unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
Message-ID: <200910201813.n9KIDLbV011114@zion.cs.uiuc.edu>

Author: jyasskin
Date: Tue Oct 20 13:13:21 2009
New Revision: 84651

URL: http://llvm.org/viewvc/llvm-project?rev=84651&view=rev
Log:
Move the Function*->allocated blocks map from the JITMemoryManager to the
JITEmitter.

I'm gradually making Functions auto-remove themselves from the JIT when they're
destroyed. In this case, the Function needs to be removed from the JITEmitter,
but the map recording which Functions need to be removed lived behind the
JITMemoryManager interface, which made things difficult.

This patch replaces the deallocateMemForFunction(Function*) method with a pair
of methods deallocateFunctionBody(void *) and deallocateExceptionTable(void *)
corresponding to the two startFoo/endFoo pairs.


Modified:
    llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h
    llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp
    llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
    llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp

Modified: llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h?rev=84651&r1=84650&r2=84651&view=diff

==============================================================================
--- llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h (original)
+++ llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h Tue Oct 20 13:13:21 2009
@@ -132,9 +132,11 @@
   ///
   virtual uint8_t *allocateGlobal(uintptr_t Size, unsigned Alignment) = 0;
 
-  /// deallocateMemForFunction - Free JIT memory for the specified function.
-  /// This is never called when the JIT is currently emitting a function.
-  virtual void deallocateMemForFunction(const Function *F) = 0;
+  /// deallocateFunctionBody - Free the specified function body.  The argument
+  /// must be the return value from a call to startFunctionBody() that hasn't
+  /// been deallocated yet.  This is never called when the JIT is currently
+  /// emitting a function.
+  virtual void deallocateFunctionBody(void *Body) = 0;
   
   /// startExceptionTable - When we finished JITing the function, if exception
   /// handling is set, we emit the exception table.
@@ -146,6 +148,12 @@
   virtual void endExceptionTable(const Function *F, uint8_t *TableStart,
                                  uint8_t *TableEnd, uint8_t* FrameRegister) = 0;
 
+  /// deallocateExceptionTable - Free the specified exception table's memory.
+  /// The argument must be the return value from a call to startExceptionTable()
+  /// that hasn't been deallocated yet.  This is never called when the JIT is
+  /// currently emitting an exception table.
+  virtual void deallocateExceptionTable(void *ET) = 0;
+
   /// CheckInvariants - For testing only.  Return true if all internal
   /// invariants are preserved, or return false and set ErrorStr to a helpful
   /// error message.

Modified: llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp?rev=84651&r1=84650&r2=84651&view=diff

==============================================================================
--- llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp (original)
+++ llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp Tue Oct 20 13:13:21 2009
@@ -42,6 +42,7 @@
 #include "llvm/System/Disassembler.h"
 #include "llvm/System/Memory.h"
 #include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
@@ -548,6 +549,13 @@
     /// finishFunction.
     JITEvent_EmittedFunctionDetails EmissionDetails;
 
+    struct EmittedCode {
+      void *FunctionBody;
+      void *ExceptionTable;
+      EmittedCode() : FunctionBody(0), ExceptionTable(0) {}
+    };
+    DenseMap<const Function *, EmittedCode> EmittedFunctions;
+
     // CurFnStubUses - For a given Function, a vector of stubs that it
     // references.  This facilitates the JIT detecting that a stub is no
     // longer used, so that it may be deallocated.
@@ -1011,7 +1019,8 @@
   BufferBegin = CurBufferPtr = MemMgr->startFunctionBody(F.getFunction(),
                                                          ActualSize);
   BufferEnd = BufferBegin+ActualSize;
-  
+  EmittedFunctions[F.getFunction()].FunctionBody = BufferBegin;
+
   // Ensure the constant pool/jump table info is at least 4-byte aligned.
   emitAlignment(16);
 
@@ -1201,6 +1210,7 @@
     BufferBegin = CurBufferPtr = MemMgr->startExceptionTable(F.getFunction(),
                                                              ActualSize);
     BufferEnd = BufferBegin+ActualSize;
+    EmittedFunctions[F.getFunction()].ExceptionTable = BufferBegin;
     uint8_t *EhStart;
     uint8_t *FrameRegister = DE->EmitDwarfTable(F, *this, FnStart, FnEnd,
                                                 EhStart);
@@ -1244,7 +1254,13 @@
 /// deallocateMemForFunction - Deallocate all memory for the specified
 /// function body.  Also drop any references the function has to stubs.
 void JITEmitter::deallocateMemForFunction(const Function *F) {
-  MemMgr->deallocateMemForFunction(F);
+  DenseMap<const Function *, EmittedCode>::iterator Emitted =
+    EmittedFunctions.find(F);
+  if (Emitted != EmittedFunctions.end()) {
+    MemMgr->deallocateFunctionBody(Emitted->second.FunctionBody);
+    MemMgr->deallocateExceptionTable(Emitted->second.ExceptionTable);
+    EmittedFunctions.erase(Emitted);
+  }
 
   // TODO: Do we need to unregister exception handling information from libgcc
   // here?

Modified: llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp?rev=84651&r1=84650&r2=84651&view=diff

==============================================================================
--- llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp (original)
+++ llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp Tue Oct 20 13:13:21 2009
@@ -297,9 +297,6 @@
 
     uint8_t *GOTBase;     // Target Specific reserved memory
     void *DlsymTable;     // Stub external symbol information
-
-    std::map<const Function*, MemoryRangeHeader*> FunctionBlocks;
-    std::map<const Function*, MemoryRangeHeader*> TableBlocks;
   public:
     DefaultJITMemoryManager();
     ~DefaultJITMemoryManager();
@@ -414,7 +411,6 @@
              "Mismatched function start/end!");
 
       uintptr_t BlockSize = FunctionEnd - (uint8_t *)CurBlock;
-      FunctionBlocks[F] = CurBlock;
 
       // Release the memory at the end of this block that isn't needed.
       FreeMemoryList =CurBlock->TrimAllocationToSize(FreeMemoryList, BlockSize);
@@ -464,7 +460,6 @@
              "Mismatched table start/end!");
       
       uintptr_t BlockSize = TableEnd - (uint8_t *)CurBlock;
-      TableBlocks[F] = CurBlock;
 
       // Release the memory at the end of this block that isn't needed.
       FreeMemoryList =CurBlock->TrimAllocationToSize(FreeMemoryList, BlockSize);
@@ -478,15 +473,9 @@
       return DlsymTable;
     }
     
-    /// deallocateMemForFunction - Deallocate all memory for the specified
-    /// function body.
-    void deallocateMemForFunction(const Function *F) {
-      std::map<const Function*, MemoryRangeHeader*>::iterator
-        I = FunctionBlocks.find(F);
-      if (I == FunctionBlocks.end()) return;
-      
+    void deallocateBlock(void *Block) {
       // Find the block that is allocated for this function.
-      MemoryRangeHeader *MemRange = I->second;
+      MemoryRangeHeader *MemRange = static_cast<MemoryRangeHeader*>(Block) - 1;
       assert(MemRange->ThisAllocated && "Block isn't allocated!");
 
       // Fill the buffer with garbage!
@@ -496,27 +485,18 @@
 
       // Free the memory.
       FreeMemoryList = MemRange->FreeBlock(FreeMemoryList);
-      
-      // Finally, remove this entry from FunctionBlocks.
-      FunctionBlocks.erase(I);
-      
-      I = TableBlocks.find(F);
-      if (I == TableBlocks.end()) return;
-      
-      // Find the block that is allocated for this function.
-      MemRange = I->second;
-      assert(MemRange->ThisAllocated && "Block isn't allocated!");
+    }
 
-      // Fill the buffer with garbage!
-      if (PoisonMemory) {
-        memset(MemRange+1, 0xCD, MemRange->BlockSize-sizeof(*MemRange));
-      }
+    /// deallocateFunctionBody - Deallocate all memory for the specified
+    /// function body.
+    void deallocateFunctionBody(void *Body) {
+      deallocateBlock(Body);
+    }
 
-      // Free the memory.
-      FreeMemoryList = MemRange->FreeBlock(FreeMemoryList);
-      
-      // Finally, remove this entry from TableBlocks.
-      TableBlocks.erase(I);
+    /// deallocateExceptionTable - Deallocate memory for the specified
+    /// exception table.
+    void deallocateExceptionTable(void *ET) {
+      deallocateBlock(ET);
     }
 
     /// setMemoryWritable - When code generation is in progress,

Modified: llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp?rev=84651&r1=84650&r2=84651&view=diff

==============================================================================
--- llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp (original)
+++ llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp Tue Oct 20 13:13:21 2009
@@ -32,37 +32,36 @@
   OwningPtr<JITMemoryManager> MemMgr(
       JITMemoryManager::CreateDefaultMemManager());
   uintptr_t size;
-  uint8_t *start;
   std::string Error;
 
   // Allocate the functions.
   OwningPtr<Function> F1(makeFakeFunction());
   size = 1024;
-  start = MemMgr->startFunctionBody(F1.get(), size);
-  memset(start, 0xFF, 1024);
-  MemMgr->endFunctionBody(F1.get(), start, start + 1024);
+  uint8_t *FunctionBody1 = MemMgr->startFunctionBody(F1.get(), size);
+  memset(FunctionBody1, 0xFF, 1024);
+  MemMgr->endFunctionBody(F1.get(), FunctionBody1, FunctionBody1 + 1024);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 
   OwningPtr<Function> F2(makeFakeFunction());
   size = 1024;
-  start = MemMgr->startFunctionBody(F2.get(), size);
-  memset(start, 0xFF, 1024);
-  MemMgr->endFunctionBody(F2.get(), start, start + 1024);
+  uint8_t *FunctionBody2 = MemMgr->startFunctionBody(F2.get(), size);
+  memset(FunctionBody2, 0xFF, 1024);
+  MemMgr->endFunctionBody(F2.get(), FunctionBody2, FunctionBody2 + 1024);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 
   OwningPtr<Function> F3(makeFakeFunction());
   size = 1024;
-  start = MemMgr->startFunctionBody(F3.get(), size);
-  memset(start, 0xFF, 1024);
-  MemMgr->endFunctionBody(F3.get(), start, start + 1024);
+  uint8_t *FunctionBody3 = MemMgr->startFunctionBody(F3.get(), size);
+  memset(FunctionBody3, 0xFF, 1024);
+  MemMgr->endFunctionBody(F3.get(), FunctionBody3, FunctionBody3 + 1024);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 
   // Deallocate them out of order, in case that matters.
-  MemMgr->deallocateMemForFunction(F2.get());
+  MemMgr->deallocateFunctionBody(FunctionBody2);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
-  MemMgr->deallocateMemForFunction(F1.get());
+  MemMgr->deallocateFunctionBody(FunctionBody1);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
-  MemMgr->deallocateMemForFunction(F3.get());
+  MemMgr->deallocateFunctionBody(FunctionBody3);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 }
 
@@ -72,7 +71,6 @@
   OwningPtr<JITMemoryManager> MemMgr(
       JITMemoryManager::CreateDefaultMemManager());
   uintptr_t size;
-  uint8_t *start;
   std::string Error;
 
   // Big functions are a little less than the largest block size.
@@ -83,26 +81,26 @@
   // Allocate big functions
   OwningPtr<Function> F1(makeFakeFunction());
   size = bigFuncSize;
-  start = MemMgr->startFunctionBody(F1.get(), size);
+  uint8_t *FunctionBody1 = MemMgr->startFunctionBody(F1.get(), size);
   ASSERT_LE(bigFuncSize, size);
-  memset(start, 0xFF, bigFuncSize);
-  MemMgr->endFunctionBody(F1.get(), start, start + bigFuncSize);
+  memset(FunctionBody1, 0xFF, bigFuncSize);
+  MemMgr->endFunctionBody(F1.get(), FunctionBody1, FunctionBody1 + bigFuncSize);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 
   OwningPtr<Function> F2(makeFakeFunction());
   size = bigFuncSize;
-  start = MemMgr->startFunctionBody(F2.get(), size);
+  uint8_t *FunctionBody2 = MemMgr->startFunctionBody(F2.get(), size);
   ASSERT_LE(bigFuncSize, size);
-  memset(start, 0xFF, bigFuncSize);
-  MemMgr->endFunctionBody(F2.get(), start, start + bigFuncSize);
+  memset(FunctionBody2, 0xFF, bigFuncSize);
+  MemMgr->endFunctionBody(F2.get(), FunctionBody2, FunctionBody2 + bigFuncSize);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 
   OwningPtr<Function> F3(makeFakeFunction());
   size = bigFuncSize;
-  start = MemMgr->startFunctionBody(F3.get(), size);
+  uint8_t *FunctionBody3 = MemMgr->startFunctionBody(F3.get(), size);
   ASSERT_LE(bigFuncSize, size);
-  memset(start, 0xFF, bigFuncSize);
-  MemMgr->endFunctionBody(F3.get(), start, start + bigFuncSize);
+  memset(FunctionBody3, 0xFF, bigFuncSize);
+  MemMgr->endFunctionBody(F3.get(), FunctionBody3, FunctionBody3 + bigFuncSize);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 
   // Check that each large function took it's own slab.
@@ -111,43 +109,46 @@
   // Allocate small functions
   OwningPtr<Function> F4(makeFakeFunction());
   size = smallFuncSize;
-  start = MemMgr->startFunctionBody(F4.get(), size);
+  uint8_t *FunctionBody4 = MemMgr->startFunctionBody(F4.get(), size);
   ASSERT_LE(smallFuncSize, size);
-  memset(start, 0xFF, smallFuncSize);
-  MemMgr->endFunctionBody(F4.get(), start, start + smallFuncSize);
+  memset(FunctionBody4, 0xFF, smallFuncSize);
+  MemMgr->endFunctionBody(F4.get(), FunctionBody4,
+                          FunctionBody4 + smallFuncSize);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 
   OwningPtr<Function> F5(makeFakeFunction());
   size = smallFuncSize;
-  start = MemMgr->startFunctionBody(F5.get(), size);
+  uint8_t *FunctionBody5 = MemMgr->startFunctionBody(F5.get(), size);
   ASSERT_LE(smallFuncSize, size);
-  memset(start, 0xFF, smallFuncSize);
-  MemMgr->endFunctionBody(F5.get(), start, start + smallFuncSize);
+  memset(FunctionBody5, 0xFF, smallFuncSize);
+  MemMgr->endFunctionBody(F5.get(), FunctionBody5,
+                          FunctionBody5 + smallFuncSize);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 
   OwningPtr<Function> F6(makeFakeFunction());
   size = smallFuncSize;
-  start = MemMgr->startFunctionBody(F6.get(), size);
+  uint8_t *FunctionBody6 = MemMgr->startFunctionBody(F6.get(), size);
   ASSERT_LE(smallFuncSize, size);
-  memset(start, 0xFF, smallFuncSize);
-  MemMgr->endFunctionBody(F6.get(), start, start + smallFuncSize);
+  memset(FunctionBody6, 0xFF, smallFuncSize);
+  MemMgr->endFunctionBody(F6.get(), FunctionBody6,
+                          FunctionBody6 + smallFuncSize);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 
   // Check that the small functions didn't allocate any new slabs.
   EXPECT_EQ(3U, MemMgr->GetNumCodeSlabs());
 
   // Deallocate them out of order, in case that matters.
-  MemMgr->deallocateMemForFunction(F2.get());
+  MemMgr->deallocateFunctionBody(FunctionBody2);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
-  MemMgr->deallocateMemForFunction(F1.get());
+  MemMgr->deallocateFunctionBody(FunctionBody1);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
-  MemMgr->deallocateMemForFunction(F4.get());
+  MemMgr->deallocateFunctionBody(FunctionBody4);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
-  MemMgr->deallocateMemForFunction(F3.get());
+  MemMgr->deallocateFunctionBody(FunctionBody3);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
-  MemMgr->deallocateMemForFunction(F5.get());
+  MemMgr->deallocateFunctionBody(FunctionBody5);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
-  MemMgr->deallocateMemForFunction(F6.get());
+  MemMgr->deallocateFunctionBody(FunctionBody6);
   EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
 }
 




From gohman at apple.com  Tue Oct 20 13:14:49 2009
From: gohman at apple.com (Dan Gohman)
Date: Tue, 20 Oct 2009 18:14:49 -0000
Subject: [llvm-commits] [llvm] r84652 -
	/llvm/trunk/lib/Target/X86/X86InstrInfo.td
Message-ID: <200910201814.n9KIEnrN011180@zion.cs.uiuc.edu>

Author: djg
Date: Tue Oct 20 13:14:49 2009
New Revision: 84652

URL: http://llvm.org/viewvc/llvm-project?rev=84652&view=rev
Log:
Following r84485, add Defs = [EFLAGS] to the 32-bit lock instructions too.

Modified:
    llvm/trunk/lib/Target/X86/X86InstrInfo.td

Modified: llvm/trunk/lib/Target/X86/X86InstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrInfo.td?rev=84652&r1=84651&r2=84652&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrInfo.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrInfo.td Tue Oct 20 13:14:49 2009
@@ -3593,6 +3593,7 @@
 
 // Optimized codegen when the non-memory output is not used.
 // FIXME: Use normal add / sub instructions and add lock prefix dynamically.
+let Defs = [EFLAGS] in {
 def LOCK_ADD8mr  : I<0x00, MRMDestMem, (outs), (ins i8mem:$dst, GR8:$src2),
                     "lock\n\t"
                     "add{b}\t{$src2, $dst|$dst, $src2}", []>, LOCK;
@@ -3662,6 +3663,7 @@
 def LOCK_DEC32m : I<0xFF, MRM1m, (outs), (ins i32mem:$dst),
                     "lock\n\t"
                     "dec{l}\t$dst", []>, LOCK;
+}
 
 // Atomic exchange, and, or, xor
 let Constraints = "$val = $dst", Defs = [EFLAGS],




From bob.wilson at apple.com  Tue Oct 20 13:57:55 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Tue, 20 Oct 2009 11:57:55 -0700
Subject: [llvm-commits] Patch for encodings of VEXT (Vector Extract)
In-Reply-To: <AD31FE2C-054B-424B-A0DC-27C3F3A6BC4A@apple.com>
References: <AD31FE2C-054B-424B-A0DC-27C3F3A6BC4A@apple.com>
Message-ID: <BCD8351B-64D7-416B-8780-E787553E359D@apple.com>

Is this really the direction we want to go?  There are a fair number  
of NEON instruction encodings where it is very useful to have some  
default bits set to zero by the .td file, assuming that the code  
emitter may overwrite them later.  For example, the vector shift- 
immediate instructions have the "imm6" field that specifies both the  
vector element size and the shift amount, where the number of bits  
used for the shift amount varies depending on the element size.  See  
VSHR, for example.  It is useful to specify the element size in  
the .td file and it is a lot more convenient to do that by stuffing  
zeros
into the shift amount.

Another example of this is the set of NEON instructions using the "One  
register and a modified immediate value" format (A7.4.6).  Currently  
we've only got the VMOV variant implemented in LLVM.  Some of the bits  
in the "cmode" field can be set in the .td file but some depend on the  
immediate value.

It's not a big problem to have a special N3VImm format for VEXT, but  
it seems like we might need a different solution for the general  
problem, regardless of what we do for VEXT.  Is it important that you  
leave those bits unspecified?  You didn't say why the patch was needed.

On Oct 16, 2009, at 11:17 AM, Johnny Chen wrote:

> Hi,
>
> Refs: A8-598.
> Leave Inst{11-8}, which represents the starting byte index of the  
> extracted result in the
> concatenation of the operands and is left unspecified.
>
> <ARMInstrNEON.patch5>_______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From bob.wilson at apple.com  Tue Oct 20 14:01:44 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Tue, 20 Oct 2009 12:01:44 -0700
Subject: [llvm-commits] ARM .td patch to specify only the
	well-known	encoding bits for imm6 and imm4 fields
In-Reply-To: <E0C31726-F049-46CB-B443-8C01B7FF9AE0@apple.com>
References: <E0C31726-F049-46CB-B443-8C01B7FF9AE0@apple.com>
Message-ID: <FA34BFD6-78D3-45A0-8D9B-B77613EC44B5@apple.com>


On Oct 20, 2009, at 10:49 AM, Johnny Chen wrote:

> Hi,
>
> Refs: A8-754 (VSHLL and friends), A8-580 (VCVT between floating- 
> point and fixed-point, Advanced SIMD), A8-592 (VDUP scalar)
>
> This patch attempts to specify just the encoding bits described in  
> the manual for several vector shift, vector convert, and vector dupe
> operations.
>
> One question I have is whether we should explicitly specify that  
> VCVT's imm6 is of the pattern '1xxxxx'.
> In the submitted patch, I left the imm6 (Inst{21-16}) of VCVT  
> unspecified.
>
> Thanks.

This is exactly the problem I was hoping to avoid.  Please see my  
response to your VEXT patch before we decide whether to use this  
approach.


From grosbach at apple.com  Tue Oct 20 14:52:35 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Tue, 20 Oct 2009 19:52:35 -0000
Subject: [llvm-commits] [llvm] r84657 -
	/llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp
Message-ID: <200910201952.n9KJqaFu016208@zion.cs.uiuc.edu>

Author: grosbach
Date: Tue Oct 20 14:52:35 2009
New Revision: 84657

URL: http://llvm.org/viewvc/llvm-project?rev=84657&view=rev
Log:
Better handle instructions that re-def a scratch register

Modified:
    llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp

Modified: llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp?rev=84657&r1=84656&r2=84657&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp (original)
+++ llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp Tue Oct 20 14:52:35 2009
@@ -778,6 +778,8 @@
     // directly.
     for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) {
       MachineInstr *MI = I;
+      bool isDefInsn = false;
+      bool isKillInsn = false;
       for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i)
         if (MI->getOperand(i).isReg()) {
           MachineOperand &MO = MI->getOperand(i);
@@ -802,6 +804,12 @@
             }
             continue;
           }
+          // If this is a def, remember that this insn defines the value.
+          // This lets us properly consider insns which re-use the scratch
+          // register, such as r2 = sub r2, #imm, in the middle of the
+          // scratch range.
+          if (MO.isDef())
+            isDefInsn = true;
 
           // Have we already allocated a scratch register for this virtual?
           if (Reg != CurrentVirtReg) {
@@ -866,19 +874,20 @@
           assert (CurrentScratchReg && "Missing scratch register!");
           MI->getOperand(i).setReg(CurrentScratchReg);
 
-          // If this is the last use of the register, stop tracking it.
           if (MI->getOperand(i).isKill()) {
-            PrevScratchReg = CurrentScratchReg;
-            PrevLastUseMI = MI;
+            isKillInsn = true;
             PrevLastUseOp = i;
-            CurrentScratchReg = CurrentVirtReg = 0;
-            havePrevValue = trackingCurrentValue;
-            // Re-scan the operands of this instruction to catch definitions
-            // of the scratch register we're using. This is to handle things
-            // like ldr "r2, [scratch]" where scratch is r2.
-            i = 0;
+            PrevLastUseMI = MI;
           }
         }
+      // If this is the last use of the scratch, stop tracking it. The
+      // last use will be a kill operand in an instruction that does
+      // not also define the scratch register.
+      if (isKillInsn && !isDefInsn) {
+        PrevScratchReg = CurrentScratchReg;
+        CurrentScratchReg = CurrentVirtReg = 0;
+        havePrevValue = trackingCurrentValue;
+      }
       RS->forward(MI);
     }
   }




From david_goodwin at apple.com  Tue Oct 20 14:54:44 2009
From: david_goodwin at apple.com (David Goodwin)
Date: Tue, 20 Oct 2009 19:54:44 -0000
Subject: [llvm-commits] [llvm] r84658 - in /llvm/trunk:
 lib/CodeGen/PostRASchedulerList.cpp
 test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll
 test/CodeGen/X86/break-anti-dependencies.ll test/CodeGen/X86/sse3.ll
Message-ID: <200910201954.n9KJsjO6016298@zion.cs.uiuc.edu>

Author: david_goodwin
Date: Tue Oct 20 14:54:44 2009
New Revision: 84658

URL: http://llvm.org/viewvc/llvm-project?rev=84658&view=rev
Log:
Checkpoint more aggressive anti-dependency breaking for post-ra scheduler.

Modified:
    llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
    llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll
    llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll
    llvm/trunk/test/CodeGen/X86/sse3.ll

Modified: llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp?rev=84658&r1=84657&r2=84658&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp (original)
+++ llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp Tue Oct 20 14:54:44 2009
@@ -48,6 +48,7 @@
 
 STATISTIC(NumNoops, "Number of noops inserted");
 STATISTIC(NumStalls, "Number of pipeline stalls");
+STATISTIC(NumFixedAnti, "Number of fixed anti-dependencies");
 
 // Post-RA scheduling is enabled with
 // TargetSubtarget.enablePostRAScheduler(). This flag can be used to
@@ -56,10 +57,11 @@
 EnablePostRAScheduler("post-RA-scheduler",
                        cl::desc("Enable scheduling after register allocation"),
                        cl::init(false), cl::Hidden);
-static cl::opt<bool>
+static cl::opt<std::string>
 EnableAntiDepBreaking("break-anti-dependencies",
-                      cl::desc("Break post-RA scheduling anti-dependencies"),
-                      cl::init(true), cl::Hidden);
+                      cl::desc("Break post-RA scheduling anti-dependencies: "
+                               "\"critical\", \"all\", or \"none\""),
+                      cl::init("critical"), cl::Hidden);
 static cl::opt<bool>
 EnablePostRAHazardAvoidance("avoid-hazards",
                       cl::desc("Enable exact hazard avoidance"),
@@ -104,8 +106,16 @@
   char PostRAScheduler::ID = 0;
 
   class VISIBILITY_HIDDEN SchedulePostRATDList : public ScheduleDAGInstrs {
+    /// RegisterReference - Information about a register reference
+    /// within a liverange
+    typedef struct {
+      /// Operand - The registers operand
+      MachineOperand *Operand;
+      /// RC - The register class
+      const TargetRegisterClass *RC;
+    } RegisterReference;
+
     /// AvailableQueue - The priority queue to use for the available SUnits.
-    ///
     LatencyPriorityQueue AvailableQueue;
   
     /// PendingQueue - This contains all of the instructions whose operands have
@@ -117,40 +127,43 @@
     /// Topo - A topological ordering for SUnits.
     ScheduleDAGTopologicalSort Topo;
 
-    /// AllocatableSet - The set of allocatable registers.
-    /// We'll be ignoring anti-dependencies on non-allocatable registers,
-    /// because they may not be safe to break.
-    const BitVector AllocatableSet;
-
     /// HazardRec - The hazard recognizer to use.
     ScheduleHazardRecognizer *HazardRec;
 
     /// AA - AliasAnalysis for making memory reference queries.
     AliasAnalysis *AA;
 
-    /// Classes - For live regs that are only used in one register class in a
-    /// live range, the register class. If the register is not live, the
-    /// corresponding value is null. If the register is live but used in
-    /// multiple register classes, the corresponding value is -1 casted to a
-    /// pointer.
-    const TargetRegisterClass *
-      Classes[TargetRegisterInfo::FirstVirtualRegister];
+    /// AllocatableSet - The set of allocatable registers.
+    /// We'll be ignoring anti-dependencies on non-allocatable registers,
+    /// because they may not be safe to break.
+    const BitVector AllocatableSet;
+
+    /// GroupNodes - Implements a disjoint-union data structure to
+    /// form register groups. A node is represented by an index into
+    /// the vector. A node can "point to" itself to indicate that it
+    /// is the parent of a group, or point to another node to indicate
+    /// that it is a member of the same group as that node.
+    std::vector<unsigned> GroupNodes;
+
+    /// GroupNodeIndices - For each register, the index of the GroupNode
+    /// currently representing the group that the register belongs to.
+    /// Register 0 is always represented by the 0 group, a group
+    /// composed of registers that are not eligible for anti-aliasing.
+    unsigned GroupNodeIndices[TargetRegisterInfo::FirstVirtualRegister];
 
     /// RegRegs - Map registers to all their references within a live range.
-    std::multimap<unsigned, MachineOperand *> RegRefs;
+    std::multimap<unsigned, RegisterReference> RegRefs;
 
-    /// KillIndices - The index of the most recent kill (proceding bottom-up),
-    /// or ~0u if the register is not live.
+    /// KillIndices - The index of the most recent kill (proceding
+    /// bottom-up), or ~0u if no kill of the register has been
+    /// seen. The register is live if this index != ~0u and DefIndices
+    /// == ~0u.
     unsigned KillIndices[TargetRegisterInfo::FirstVirtualRegister];
 
     /// DefIndices - The index of the most recent complete def (proceding bottom
     /// up), or ~0u if the register is live.
     unsigned DefIndices[TargetRegisterInfo::FirstVirtualRegister];
 
-    /// KeepRegs - A set of registers which are live and cannot be changed to
-    /// break anti-dependencies.
-    SmallSet<unsigned, 4> KeepRegs;
-
   public:
     SchedulePostRATDList(MachineFunction &MF,
                          const MachineLoopInfo &MLI,
@@ -158,8 +171,9 @@
                          ScheduleHazardRecognizer *HR,
                          AliasAnalysis *aa)
       : ScheduleDAGInstrs(MF, MLI, MDT), Topo(SUnits),
-        AllocatableSet(TRI->getAllocatableSet(MF)),
-        HazardRec(HR), AA(aa) {}
+      HazardRec(HR), AA(aa),
+      AllocatableSet(TRI->getAllocatableSet(MF)),
+      GroupNodes(TargetRegisterInfo::FirstVirtualRegister, 0) {}
 
     ~SchedulePostRATDList() {
       delete HazardRec;
@@ -170,41 +184,63 @@
     ///
     void StartBlock(MachineBasicBlock *BB);
 
-    /// Schedule - Schedule the instruction range using list scheduling.
-    ///
-    void Schedule();
-    
-    /// FixupKills - Fix register kill flags that have been made
-    /// invalid due to scheduling
+    /// FinishBlock - Clean up register live-range state.
     ///
-    void FixupKills(MachineBasicBlock *MBB);
+    void FinishBlock();
 
     /// Observe - Update liveness information to account for the current
     /// instruction, which will not be scheduled.
     ///
     void Observe(MachineInstr *MI, unsigned Count);
 
-    /// FinishBlock - Clean up register live-range state.
+    /// Schedule - Schedule the instruction range using list scheduling.
     ///
-    void FinishBlock();
+    void Schedule();
+    
+    /// FixupKills - Fix register kill flags that have been made
+    /// invalid due to scheduling
+    ///
+    void FixupKills(MachineBasicBlock *MBB);
 
   private:
-    void PrescanInstruction(MachineInstr *MI);
+    /// IsLive - Return true if Reg is live
+    bool IsLive(unsigned Reg);
+
+    void PrescanInstruction(MachineInstr *MI, unsigned Count);
     void ScanInstruction(MachineInstr *MI, unsigned Count);
+    bool BreakAntiDependencies(bool CriticalPathOnly);
+    unsigned FindSuitableFreeRegister(unsigned AntiDepReg,
+                                      unsigned LastNewReg);
+
     void ReleaseSucc(SUnit *SU, SDep *SuccEdge);
     void ReleaseSuccessors(SUnit *SU);
     void ScheduleNodeTopDown(SUnit *SU, unsigned CurCycle);
     void ListScheduleTopDown();
-    bool BreakAntiDependencies();
-    unsigned findSuitableFreeRegister(unsigned AntiDepReg,
-                                      unsigned LastNewReg,
-                                      const TargetRegisterClass *);
+
     void StartBlockForKills(MachineBasicBlock *BB);
     
     // ToggleKillFlag - Toggle a register operand kill flag. Other
     // adjustments may be made to the instruction if necessary. Return
     // true if the operand has been deleted, false if not.
     bool ToggleKillFlag(MachineInstr *MI, MachineOperand &MO);
+    
+    // GetGroup - Get the group for a register. The returned value is
+    // the index of the GroupNode representing the group.
+    unsigned GetGroup(unsigned Reg);
+    
+    // GetGroupRegs - Return a vector of the registers belonging to a
+    // group.
+    void GetGroupRegs(unsigned Group, std::vector<unsigned> &Regs);
+
+    // UnionGroups - Union Reg1's and Reg2's groups to form a new
+    // group. Return the index of the GroupNode representing the
+    // group.
+    unsigned UnionGroups(unsigned Reg1, unsigned Reg2);
+
+    // LeaveGroup - Remove a register from its current group and place
+    // it alone in its own group. Return the index of the GroupNode
+    // representing the registers new group.
+    unsigned LeaveGroup(unsigned Reg);
   };
 }
 
@@ -303,7 +339,58 @@
 
   return true;
 }
+
+unsigned SchedulePostRATDList::GetGroup(unsigned Reg)
+{
+  unsigned Node = GroupNodeIndices[Reg];
+  while (GroupNodes[Node] != Node)
+    Node = GroupNodes[Node];
+
+  return Node;
+}
+
+void SchedulePostRATDList::GetGroupRegs(unsigned Group, std::vector<unsigned> &Regs)
+{
+  for (unsigned Reg = 0; Reg != TargetRegisterInfo::FirstVirtualRegister; ++Reg) {
+    if (GetGroup(Reg) == Group)
+      Regs.push_back(Reg);
+  }
+}
+
+unsigned SchedulePostRATDList::UnionGroups(unsigned Reg1, unsigned Reg2)
+{
+  assert(GroupNodes[0] == 0 && "GroupNode 0 not parent!");
+  assert(GroupNodeIndices[0] == 0 && "Reg 0 not in Group 0!");
+  
+  // find group for each register
+  unsigned Group1 = GetGroup(Reg1);
+  unsigned Group2 = GetGroup(Reg2);
   
+  // if either group is 0, then that must become the parent
+  unsigned Parent = (Group1 == 0) ? Group1 : Group2;
+  unsigned Other = (Parent == Group1) ? Group2 : Group1;
+  GroupNodes.at(Other) = Parent;
+  return Parent;
+}
+  
+unsigned SchedulePostRATDList::LeaveGroup(unsigned Reg)
+{
+  // Create a new GroupNode for Reg. Reg's existing GroupNode must
+  // stay as is because there could be other GroupNodes referring to
+  // it.
+  unsigned idx = GroupNodes.size();
+  GroupNodes.push_back(idx);
+  GroupNodeIndices[Reg] = idx;
+  return idx;
+}
+
+bool SchedulePostRATDList::IsLive(unsigned Reg)
+{
+  // KillIndex must be defined and DefIndex not defined for a register
+  // to be live.
+  return((KillIndices[Reg] != ~0u) && (DefIndices[Reg] == ~0u));
+}
+
 /// StartBlock - Initialize register live-range state for scheduling in
 /// this block.
 ///
@@ -314,17 +401,15 @@
   // Reset the hazard recognizer.
   HazardRec->Reset();
 
-  // Clear out the register class data.
-  std::fill(Classes, array_endof(Classes),
-            static_cast<const TargetRegisterClass *>(0));
+  // Initialize all registers to be in their own group. Initially we
+  // assign the register to the same-indexed GroupNode.
+  for (unsigned i = 0; i < TargetRegisterInfo::FirstVirtualRegister; ++i)
+    GroupNodeIndices[i] = i;
 
   // Initialize the indices to indicate that no registers are live.
   std::fill(KillIndices, array_endof(KillIndices), ~0u);
   std::fill(DefIndices, array_endof(DefIndices), BB->size());
 
-  // Clear "do not change" set.
-  KeepRegs.clear();
-
   bool IsReturnBlock = (!BB->empty() && BB->back().getDesc().isReturn());
 
   // Determine the live-out physregs for this block.
@@ -333,13 +418,13 @@
     for (MachineRegisterInfo::liveout_iterator I = MRI.liveout_begin(),
          E = MRI.liveout_end(); I != E; ++I) {
       unsigned Reg = *I;
-      Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
+      UnionGroups(Reg, 0);
       KillIndices[Reg] = BB->size();
       DefIndices[Reg] = ~0u;
       // Repeat, for all aliases.
       for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
         unsigned AliasReg = *Alias;
-        Classes[AliasReg] = reinterpret_cast<TargetRegisterClass *>(-1);
+        UnionGroups(AliasReg, 0);
         KillIndices[AliasReg] = BB->size();
         DefIndices[AliasReg] = ~0u;
       }
@@ -351,13 +436,13 @@
       for (MachineBasicBlock::livein_iterator I = (*SI)->livein_begin(),
            E = (*SI)->livein_end(); I != E; ++I) {
         unsigned Reg = *I;
-        Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
+        UnionGroups(Reg, 0);
         KillIndices[Reg] = BB->size();
         DefIndices[Reg] = ~0u;
         // Repeat, for all aliases.
         for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
           unsigned AliasReg = *Alias;
-          Classes[AliasReg] = reinterpret_cast<TargetRegisterClass *>(-1);
+          UnionGroups(AliasReg, 0);
           KillIndices[AliasReg] = BB->size();
           DefIndices[AliasReg] = ~0u;
         }
@@ -372,13 +457,13 @@
   for (const unsigned *I = TRI->getCalleeSavedRegs(); *I; ++I) {
     unsigned Reg = *I;
     if (!IsReturnBlock && !Pristine.test(Reg)) continue;
-    Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
+    UnionGroups(Reg, 0);
     KillIndices[Reg] = BB->size();
     DefIndices[Reg] = ~0u;
     // Repeat, for all aliases.
     for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
       unsigned AliasReg = *Alias;
-      Classes[AliasReg] = reinterpret_cast<TargetRegisterClass *>(-1);
+      UnionGroups(AliasReg, 0);
       KillIndices[AliasReg] = BB->size();
       DefIndices[AliasReg] = ~0u;
     }
@@ -393,8 +478,8 @@
   // Build the scheduling graph.
   BuildSchedGraph(AA);
 
-  if (EnableAntiDepBreaking) {
-    if (BreakAntiDependencies()) {
+  if (EnableAntiDepBreaking != "none") {
+    if (BreakAntiDependencies((EnableAntiDepBreaking == "all") ? false : true)) {
       // We made changes. Update the dependency graph.
       // Theoretically we could update the graph in place:
       // When a live range is changed to use a different register, remove
@@ -424,21 +509,27 @@
 void SchedulePostRATDList::Observe(MachineInstr *MI, unsigned Count) {
   assert(Count < InsertPosIndex && "Instruction index out of expected range!");
 
-  // Any register which was defined within the previous scheduling region
-  // may have been rescheduled and its lifetime may overlap with registers
-  // in ways not reflected in our current liveness state. For each such
-  // register, adjust the liveness state to be conservatively correct.
-  for (unsigned Reg = 0; Reg != TargetRegisterInfo::FirstVirtualRegister; ++Reg)
-    if (DefIndices[Reg] < InsertPosIndex && DefIndices[Reg] >= Count) {
-      assert(KillIndices[Reg] == ~0u && "Clobbered register is live!");
-      // Mark this register to be non-renamable.
-      Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
-      // Move the def index to the end of the previous region, to reflect
-      // that the def could theoretically have been scheduled at the end.
-      DefIndices[Reg] = InsertPosIndex;
+  DEBUG(errs() << "Observe: ");
+  DEBUG(MI->dump());
+
+  for (unsigned Reg = 0; Reg != TargetRegisterInfo::FirstVirtualRegister; ++Reg) {
+    // If Reg is current live, then mark that it can't be renamed as
+    // we don't know the extent of its live-range anymore (now that it
+    // has been scheduled). If it is not live but was defined in the
+    // previous schedule region, then set its def index to the most
+    // conservative location (i.e. the beginning of the previous
+    // schedule region).
+    if (IsLive(Reg)) {
+      DEBUG(if (GetGroup(Reg) != 0)
+              errs() << " " << TRI->getName(Reg) << "=g" << 
+                GetGroup(Reg) << "->g0(region live-out)");
+      UnionGroups(Reg, 0);
+    } else if ((DefIndices[Reg] < InsertPosIndex) && (DefIndices[Reg] >= Count)) {
+      DefIndices[Reg] = Count;
     }
+  }
 
-  PrescanInstruction(MI);
+  PrescanInstruction(MI, Count);
   ScanInstruction(MI, Count);
 }
 
@@ -473,156 +564,232 @@
   return Next;
 }
 
-void SchedulePostRATDList::PrescanInstruction(MachineInstr *MI) {
-  // Scan the register operands for this instruction and update
-  // Classes and RegRefs.
-  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
-    MachineOperand &MO = MI->getOperand(i);
-    if (!MO.isReg()) continue;
-    unsigned Reg = MO.getReg();
-    if (Reg == 0) continue;
-    const TargetRegisterClass *NewRC = 0;
-    
-    if (i < MI->getDesc().getNumOperands())
-      NewRC = MI->getDesc().OpInfo[i].getRegClass(TRI);
-
-    // For now, only allow the register to be changed if its register
-    // class is consistent across all uses.
-    if (!Classes[Reg] && NewRC)
-      Classes[Reg] = NewRC;
-    else if (!NewRC || Classes[Reg] != NewRC)
-      Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
-
-    // Now check for aliases.
-    for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
-      // If an alias of the reg is used during the live range, give up.
-      // Note that this allows us to skip checking if AntiDepReg
-      // overlaps with any of the aliases, among other things.
-      unsigned AliasReg = *Alias;
-      if (Classes[AliasReg]) {
-        Classes[AliasReg] = reinterpret_cast<TargetRegisterClass *>(-1);
-        Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
-      }
-    }
-
-    // If we're still willing to consider this register, note the reference.
-    if (Classes[Reg] != reinterpret_cast<TargetRegisterClass *>(-1))
-      RegRefs.insert(std::make_pair(Reg, &MO));
-
-    // It's not safe to change register allocation for source operands of
-    // that have special allocation requirements.
-    if (MO.isUse() && MI->getDesc().hasExtraSrcRegAllocReq()) {
-      if (KeepRegs.insert(Reg)) {
-        for (const unsigned *Subreg = TRI->getSubRegisters(Reg);
-             *Subreg; ++Subreg)
-          KeepRegs.insert(*Subreg);
-      }
+/// AntiDepPathStep - Return SUnit that SU has an anti-dependence on.
+static SDep *AntiDepPathStep(SUnit *SU) {
+  for (SUnit::pred_iterator P = SU->Preds.begin(), PE = SU->Preds.end();
+       P != PE; ++P) {
+    if (P->getKind() == SDep::Anti) {
+      return &*P;
     }
   }
+  return 0;
 }
 
-void SchedulePostRATDList::ScanInstruction(MachineInstr *MI,
-                                           unsigned Count) {
-  // Update liveness.
-  // Proceding upwards, registers that are defed but not used in this
-  // instruction are now dead.
+void SchedulePostRATDList::PrescanInstruction(MachineInstr *MI, unsigned Count) {
+  // Scan the register defs for this instruction and update
+  // live-ranges, groups and RegRefs.
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     MachineOperand &MO = MI->getOperand(i);
-    if (!MO.isReg()) continue;
+    if (!MO.isReg() || !MO.isDef()) continue;
     unsigned Reg = MO.getReg();
     if (Reg == 0) continue;
-    if (!MO.isDef()) continue;
-    // Ignore two-addr defs.
+    // Ignore two-addr defs for liveness...
     if (MI->isRegTiedToUseOperand(i)) continue;
 
+    // Update Def for Reg and subregs.
     DefIndices[Reg] = Count;
-    KillIndices[Reg] = ~0u;
-    assert(((KillIndices[Reg] == ~0u) !=
-            (DefIndices[Reg] == ~0u)) &&
-           "Kill and Def maps aren't consistent for Reg!");
-    KeepRegs.erase(Reg);
-    Classes[Reg] = 0;
-    RegRefs.erase(Reg);
-    // Repeat, for all subregs.
     for (const unsigned *Subreg = TRI->getSubRegisters(Reg);
          *Subreg; ++Subreg) {
       unsigned SubregReg = *Subreg;
       DefIndices[SubregReg] = Count;
-      KillIndices[SubregReg] = ~0u;
-      KeepRegs.erase(SubregReg);
-      Classes[SubregReg] = 0;
-      RegRefs.erase(SubregReg);
-    }
-    // Conservatively mark super-registers as unusable.
-    for (const unsigned *Super = TRI->getSuperRegisters(Reg);
-         *Super; ++Super) {
-      unsigned SuperReg = *Super;
-      Classes[SuperReg] = reinterpret_cast<TargetRegisterClass *>(-1);
     }
   }
+
+  DEBUG(errs() << "\tGroups:");
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     MachineOperand &MO = MI->getOperand(i);
-    if (!MO.isReg()) continue;
+    if (!MO.isReg() || !MO.isDef()) continue;
     unsigned Reg = MO.getReg();
     if (Reg == 0) continue;
-    if (!MO.isUse()) continue;
 
-    const TargetRegisterClass *NewRC = 0;
-    if (i < MI->getDesc().getNumOperands())
-      NewRC = MI->getDesc().OpInfo[i].getRegClass(TRI);
+    DEBUG(errs() << " " << TRI->getName(Reg) << "=g" << GetGroup(Reg)); 
 
-    // For now, only allow the register to be changed if its register
-    // class is consistent across all uses.
-    if (!Classes[Reg] && NewRC)
-      Classes[Reg] = NewRC;
-    else if (!NewRC || Classes[Reg] != NewRC)
-      Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
+    // If MI's defs have special allocation requirement, don't allow
+    // any def registers to be changed. Also assume all registers
+    // defined in a call must not be changed (ABI).
+    if (MI->getDesc().isCall() || MI->getDesc().hasExtraDefRegAllocReq()) {
+      DEBUG(if (GetGroup(Reg) != 0) errs() << "->g0(alloc-req)");
+      UnionGroups(Reg, 0);
+    }
 
-    RegRefs.insert(std::make_pair(Reg, &MO));
+    // Any subregisters that are live at this point are defined here,
+    // so group those subregisters with Reg.
+    for (const unsigned *Subreg = TRI->getSubRegisters(Reg);
+         *Subreg; ++Subreg) {
+      unsigned SubregReg = *Subreg;
+      if (IsLive(SubregReg)) {
+        UnionGroups(Reg, SubregReg);
+        DEBUG(errs() << "->g" << GetGroup(Reg) << "(via " << 
+              TRI->getName(SubregReg) << ")");
+      }
+    }
+    
+    // Note register reference...
+    const TargetRegisterClass *RC = NULL;
+    if (i < MI->getDesc().getNumOperands())
+      RC = MI->getDesc().OpInfo[i].getRegClass(TRI);
+    RegisterReference RR = { &MO, RC };
+    RegRefs.insert(std::make_pair(Reg, RR));
+  }
+
+  DEBUG(errs() << '\n');
+}
 
-    // It wasn't previously live but now it is, this is a kill.
-    if (KillIndices[Reg] == ~0u) {
+void SchedulePostRATDList::ScanInstruction(MachineInstr *MI,
+                                           unsigned Count) {
+  // Scan the register uses for this instruction and update
+  // live-ranges, groups and RegRefs.
+  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+    MachineOperand &MO = MI->getOperand(i);
+    if (!MO.isReg() || !MO.isUse()) continue;
+    unsigned Reg = MO.getReg();
+    if (Reg == 0) continue;
+    
+    // It wasn't previously live but now it is, this is a kill. Forget
+    // the previous live-range information and start a new live-range
+    // for the register.
+    if (!IsLive(Reg)) {
       KillIndices[Reg] = Count;
       DefIndices[Reg] = ~0u;
-          assert(((KillIndices[Reg] == ~0u) !=
-                  (DefIndices[Reg] == ~0u)) &&
-               "Kill and Def maps aren't consistent for Reg!");
+      RegRefs.erase(Reg);
+      LeaveGroup(Reg);
     }
-    // Repeat, for all aliases.
-    for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
-      unsigned AliasReg = *Alias;
-      if (KillIndices[AliasReg] == ~0u) {
-        KillIndices[AliasReg] = Count;
-        DefIndices[AliasReg] = ~0u;
+    // Repeat, for subregisters.
+    for (const unsigned *Subreg = TRI->getSubRegisters(Reg);
+         *Subreg; ++Subreg) {
+      unsigned SubregReg = *Subreg;
+      if (!IsLive(SubregReg)) {
+        KillIndices[SubregReg] = Count;
+        DefIndices[SubregReg] = ~0u;
+        RegRefs.erase(SubregReg);
+        LeaveGroup(SubregReg);
       }
     }
+
+    // Note register reference...
+    const TargetRegisterClass *RC = NULL;
+    if (i < MI->getDesc().getNumOperands())
+      RC = MI->getDesc().OpInfo[i].getRegClass(TRI);
+    RegisterReference RR = { &MO, RC };
+    RegRefs.insert(std::make_pair(Reg, RR));
+  }
+  
+  // Form a group of all defs and uses of a KILL instruction to ensure
+  // that all registers are renamed as a group.
+  if (MI->getOpcode() == TargetInstrInfo::KILL) {
+    unsigned FirstReg = 0;
+    for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+      MachineOperand &MO = MI->getOperand(i);
+      if (!MO.isReg()) continue;
+      unsigned Reg = MO.getReg();
+      if (Reg == 0) continue;
+      
+      if (FirstReg != 0)
+        UnionGroups(FirstReg, Reg);
+      FirstReg = Reg;
+    }
+
+    DEBUG(if (FirstReg != 0) errs() << "\tKill Group: g" << 
+                               GetGroup(FirstReg) << '\n'); 
   }
 }
 
-unsigned
-SchedulePostRATDList::findSuitableFreeRegister(unsigned AntiDepReg,
-                                               unsigned LastNewReg,
-                                               const TargetRegisterClass *RC) {
-  for (TargetRegisterClass::iterator R = RC->allocation_order_begin(MF),
-       RE = RC->allocation_order_end(MF); R != RE; ++R) {
-    unsigned NewReg = *R;
-    // Don't replace a register with itself.
-    if (NewReg == AntiDepReg) continue;
-    // Don't replace a register with one that was recently used to repair
-    // an anti-dependence with this AntiDepReg, because that would
-    // re-introduce that anti-dependence.
-    if (NewReg == LastNewReg) continue;
-    // If NewReg is dead and NewReg's most recent def is not before
-    // AntiDepReg's kill, it's safe to replace AntiDepReg with NewReg.
-    assert(((KillIndices[AntiDepReg] == ~0u) != (DefIndices[AntiDepReg] == ~0u)) &&
-           "Kill and Def maps aren't consistent for AntiDepReg!");
-    assert(((KillIndices[NewReg] == ~0u) != (DefIndices[NewReg] == ~0u)) &&
-           "Kill and Def maps aren't consistent for NewReg!");
-    if (KillIndices[NewReg] != ~0u ||
-        Classes[NewReg] == reinterpret_cast<TargetRegisterClass *>(-1) ||
-        KillIndices[AntiDepReg] > DefIndices[NewReg])
-      continue;
-    return NewReg;
+unsigned SchedulePostRATDList::FindSuitableFreeRegister(unsigned AntiDepReg,
+                                                        unsigned LastNewReg) {
+  // Collect all registers in the same group as AntiDepReg. These all
+  // need to be renamed together if we are to break the
+  // anti-dependence.
+  std::vector<unsigned> Regs;
+  GetGroupRegs(GetGroup(AntiDepReg), Regs);
+
+  DEBUG(errs() << "\tRename Register Group:");
+  DEBUG(for (unsigned i = 0, e = Regs.size(); i != e; ++i)
+          DEBUG(errs() << " " << TRI->getName(Regs[i])));
+  DEBUG(errs() << "\n");
+
+  // If there is a single register that needs to be renamed then we
+  // can do it ourselves.
+  if (Regs.size() == 1) {
+    assert(Regs[0] == AntiDepReg && "Register group does not contain register!");
+
+    // Check all references that need rewriting. Gather up all the
+    // register classes for the register references.
+    const TargetRegisterClass *FirstRC = NULL;
+    std::set<const TargetRegisterClass *> RCs;
+    std::pair<std::multimap<unsigned, RegisterReference>::iterator,
+      std::multimap<unsigned, RegisterReference>::iterator>
+      Range = RegRefs.equal_range(AntiDepReg);
+    for (std::multimap<unsigned, RegisterReference>::iterator
+           Q = Range.first, QE = Range.second; Q != QE; ++Q) {
+      const TargetRegisterClass *RC = Q->second.RC;
+      if (RC == NULL) continue;
+      if (FirstRC == NULL)
+        FirstRC = RC;
+      else if (FirstRC != RC)
+        RCs.insert(RC);
+    }
+    
+    if (FirstRC == NULL)
+      return 0;
+
+    DEBUG(errs() << "\tChecking Regclasses: " << FirstRC->getName());
+    DEBUG(for (std::set<const TargetRegisterClass *>::iterator S = 
+                 RCs.begin(), E = RCs.end(); S != E; ++S)
+            errs() << " " << (*S)->getName());
+    DEBUG(errs() << '\n');
+
+    // Using the allocation order for one of the register classes,
+    // find the first register that belongs to all the register
+    // classes that is available over the liverange of the register.
+    DEBUG(errs() << "\tFind Register:");
+    for (TargetRegisterClass::iterator R = FirstRC->allocation_order_begin(MF),
+           RE = FirstRC->allocation_order_end(MF); R != RE; ++R) {
+      unsigned NewReg = *R;
+      
+      // Don't replace a register with itself.
+      if (NewReg == AntiDepReg) continue;
+
+      DEBUG(errs() << " " << TRI->getName(NewReg));
+      
+      // Make sure NewReg is in all required register classes.
+      for (std::set<const TargetRegisterClass *>::iterator S = 
+             RCs.begin(), E = RCs.end(); S != E; ++S) {
+        const TargetRegisterClass *RC = *S;
+        if (!RC->contains(NewReg)) {
+          DEBUG(errs() << "(not in " << RC->getName() << ")");
+          NewReg = 0;
+          break;
+        }
+      }
+
+      // If NewReg is dead and NewReg's most recent def is not before
+      // AntiDepReg's kill, it's safe to replace AntiDepReg with
+      // NewReg. We must also check all subregisters of NewReg.
+      if (IsLive(NewReg) || (KillIndices[AntiDepReg] > DefIndices[NewReg])) {
+        DEBUG(errs() << "(live)");
+        continue;
+      }
+      {
+        bool found = false;
+        for (const unsigned *Subreg = TRI->getSubRegisters(NewReg);
+             *Subreg; ++Subreg) {
+          unsigned SubregReg = *Subreg;
+          if (IsLive(SubregReg) || (KillIndices[AntiDepReg] > DefIndices[SubregReg])) {
+            DEBUG(errs() << "(subreg " << TRI->getName(SubregReg) << " live)");
+            found = true;
+          }
+        }
+        if (found)
+          continue;
+      }
+      
+      if (NewReg != 0) { 
+        DEBUG(errs() << '\n');
+        return NewReg;
+      }
+    }
+
+    DEBUG(errs() << '\n');
   }
 
   // No registers are free and available!
@@ -632,139 +799,159 @@
 /// BreakAntiDependencies - Identifiy anti-dependencies along the critical path
 /// of the ScheduleDAG and break them by renaming registers.
 ///
-bool SchedulePostRATDList::BreakAntiDependencies() {
+bool SchedulePostRATDList::BreakAntiDependencies(bool CriticalPathOnly) {
   // The code below assumes that there is at least one instruction,
   // so just duck out immediately if the block is empty.
   if (SUnits.empty()) return false;
 
+  // If breaking anti-dependencies only along the critical path, track
+  // progress along the critical path through the SUnit graph as we
+  // walk the instructions.
+  SUnit *CriticalPathSU = 0;
+  MachineInstr *CriticalPathMI = 0;
+  
+  // If breaking all anti-dependencies need a map from MI to SUnit.
+  std::map<MachineInstr *, SUnit *> MISUnitMap;
+
   // Find the node at the bottom of the critical path.
-  SUnit *Max = 0;
-  for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
-    SUnit *SU = &SUnits[i];
-    if (!Max || SU->getDepth() + SU->Latency > Max->getDepth() + Max->Latency)
-      Max = SU;
-  }
+  if (CriticalPathOnly) {
+    SUnit *Max = 0;
+    for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
+      SUnit *SU = &SUnits[i];
+      if (!Max || SU->getDepth() + SU->Latency > Max->getDepth() + Max->Latency)
+        Max = SU;
+    }
 
-#ifndef NDEBUG
-  {
     DEBUG(errs() << "Critical path has total latency "
           << (Max->getDepth() + Max->Latency) << "\n");
+    CriticalPathSU = Max;
+    CriticalPathMI = CriticalPathSU->getInstr();
+  } else {
+    for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
+      SUnit *SU = &SUnits[i];
+      MISUnitMap.insert(std::pair<MachineInstr *, SUnit *>(SU->getInstr(), SU));
+    }
+    DEBUG(errs() << "Breaking all anti-dependencies\n");
+  }
+
+#ifndef NDEBUG 
+  {
     DEBUG(errs() << "Available regs:");
     for (unsigned Reg = 0; Reg < TRI->getNumRegs(); ++Reg) {
-      if (KillIndices[Reg] == ~0u)
+      if (!IsLive(Reg))
         DEBUG(errs() << " " << TRI->getName(Reg));
     }
     DEBUG(errs() << '\n');
   }
+  std::string dbgStr;
 #endif
 
-  // Track progress along the critical path through the SUnit graph as we walk
-  // the instructions.
-  SUnit *CriticalPathSU = Max;
-  MachineInstr *CriticalPathMI = CriticalPathSU->getInstr();
-
-  // Consider this pattern:
-  //   A = ...
-  //   ... = A
-  //   A = ...
-  //   ... = A
-  //   A = ...
-  //   ... = A
-  //   A = ...
-  //   ... = A
-  // There are three anti-dependencies here, and without special care,
-  // we'd break all of them using the same register:
-  //   A = ...
-  //   ... = A
-  //   B = ...
-  //   ... = B
-  //   B = ...
-  //   ... = B
-  //   B = ...
-  //   ... = B
-  // because at each anti-dependence, B is the first register that
-  // isn't A which is free.  This re-introduces anti-dependencies
-  // at all but one of the original anti-dependencies that we were
-  // trying to break.  To avoid this, keep track of the most recent
-  // register that each register was replaced with, avoid
-  // using it to repair an anti-dependence on the same register.
-  // This lets us produce this:
-  //   A = ...
-  //   ... = A
-  //   B = ...
-  //   ... = B
-  //   C = ...
-  //   ... = C
-  //   B = ...
-  //   ... = B
-  // This still has an anti-dependence on B, but at least it isn't on the
-  // original critical path.
-  //
   // TODO: If we tracked more than one register here, we could potentially
   // fix that remaining critical edge too. This is a little more involved,
   // because unlike the most recent register, less recent registers should
   // still be considered, though only if no other registers are available.
   unsigned LastNewReg[TargetRegisterInfo::FirstVirtualRegister] = {};
 
-  // Attempt to break anti-dependence edges on the critical path. Walk the
-  // instructions from the bottom up, tracking information about liveness
-  // as we go to help determine which registers are available.
+  // Attempt to break anti-dependence edges. Walk the instructions
+  // from the bottom up, tracking information about liveness as we go
+  // to help determine which registers are available.
   bool Changed = false;
   unsigned Count = InsertPosIndex - 1;
   for (MachineBasicBlock::iterator I = InsertPos, E = Begin;
        I != E; --Count) {
     MachineInstr *MI = --I;
 
-    // Check if this instruction has a dependence on the critical path that
-    // is an anti-dependence that we may be able to break. If it is, set
-    // AntiDepReg to the non-zero register associated with the anti-dependence.
+    DEBUG(errs() << "Anti: ");
+    DEBUG(MI->dump());
+
+    // Process the defs in MI...
+    PrescanInstruction(MI, Count);
+
+    // Check if this instruction has an anti-dependence that we may be
+    // able to break. If it is, set AntiDepReg to the non-zero
+    // register associated with the anti-dependence.
     //
-    // We limit our attention to the critical path as a heuristic to avoid
+    unsigned AntiDepReg = 0;
+  
+    // Limiting our attention to the critical path is a heuristic to avoid
     // breaking anti-dependence edges that aren't going to significantly
     // impact the overall schedule. There are a limited number of registers
     // and we want to save them for the important edges.
     // 
+    // We can also break all anti-dependencies because they can
+    // occur along the non-critical path but are still detrimental for
+    // scheduling.
+    // 
     // TODO: Instructions with multiple defs could have multiple
     // anti-dependencies. The current code here only knows how to break one
     // edge per instruction. Note that we'd have to be able to break all of
     // the anti-dependencies in an instruction in order to be effective.
-    unsigned AntiDepReg = 0;
-    if (MI == CriticalPathMI) {
-      if (SDep *Edge = CriticalPathStep(CriticalPathSU)) {
+    if (!CriticalPathOnly || (MI == CriticalPathMI)) {
+      DEBUG(dbgStr.clear());
+
+      SUnit *PathSU;
+      SDep *Edge;
+      if (CriticalPathOnly) {
+        PathSU = CriticalPathSU;
+        Edge = CriticalPathStep(PathSU);
+      } else {
+        PathSU = MISUnitMap[MI];
+        Edge = (PathSU) ? AntiDepPathStep(PathSU) : 0;
+      }
+      
+      if (Edge) {
         SUnit *NextSU = Edge->getSUnit();
 
-        // Only consider anti-dependence edges.
-        if (Edge->getKind() == SDep::Anti) {
+        // Only consider anti-dependence edges, and ignore KILL
+        // instructions (they form a group in ScanInstruction but
+        // don't cause any anti-dependence breaking themselves)
+        if ((Edge->getKind() == SDep::Anti) &&
+            (MI->getOpcode() != TargetInstrInfo::KILL)) {
           AntiDepReg = Edge->getReg();
+          DEBUG(dbgStr += "\tAntidep reg: ");
+          DEBUG(dbgStr += TRI->getName(AntiDepReg));
           assert(AntiDepReg != 0 && "Anti-dependence on reg0?");
-          if (!AllocatableSet.test(AntiDepReg))
+          if (!AllocatableSet.test(AntiDepReg)) {
             // Don't break anti-dependencies on non-allocatable registers.
+            DEBUG(dbgStr += " (non-allocatable)");
             AntiDepReg = 0;
-          else if (KeepRegs.count(AntiDepReg))
-            // Don't break anti-dependencies if an use down below requires
-            // this exact register.
-            AntiDepReg = 0;
-          else {
-            // If the SUnit has other dependencies on the SUnit that it
-            // anti-depends on, don't bother breaking the anti-dependency
-            // since those edges would prevent such units from being
-            // scheduled past each other regardless.
-            //
-            // Also, if there are dependencies on other SUnits with the
-            // same register as the anti-dependency, don't attempt to
-            // break it.
-            for (SUnit::pred_iterator P = CriticalPathSU->Preds.begin(),
-                 PE = CriticalPathSU->Preds.end(); P != PE; ++P)
-              if (P->getSUnit() == NextSU ?
+          } else {
+            int OpIdx = MI->findRegisterDefOperandIdx(AntiDepReg);
+            assert(OpIdx != -1 && "Can't find index for defined register operand");
+            if (MI->isRegTiedToUseOperand(OpIdx)) {
+              // If the anti-dep register is tied to a use, then don't try to
+              // change it. It will be changed along with the use if required
+              // to break an earlier antidep.
+              DEBUG(dbgStr += " (tied-to-use)");
+              AntiDepReg = 0;
+            } else {
+              // If the SUnit has other dependencies on the SUnit that
+              // it anti-depends on, don't bother breaking the
+              // anti-dependency since those edges would prevent such
+              // units from being scheduled past each other
+              // regardless.
+              //
+              // Also, if there are dependencies on other SUnits with
+              // the same register as the anti-dependency, don't
+              // attempt to break it.
+              for (SUnit::pred_iterator P = PathSU->Preds.begin(),
+                     PE = PathSU->Preds.end(); P != PE; ++P) {
+                if (P->getSUnit() == NextSU ?
                     (P->getKind() != SDep::Anti || P->getReg() != AntiDepReg) :
                     (P->getKind() == SDep::Data && P->getReg() == AntiDepReg)) {
-                AntiDepReg = 0;
-                break;
+                  DEBUG(dbgStr += " (real dependency)");
+                  AntiDepReg = 0;
+                  break;
+                }
               }
+            }
           }
         }
-        CriticalPathSU = NextSU;
-        CriticalPathMI = CriticalPathSU->getInstr();
+        
+        if (CriticalPathOnly) {
+          CriticalPathSU = NextSU;
+          CriticalPathMI = CriticalPathSU->getInstr();
+        }
       } else {
         // We've reached the end of the critical path.
         CriticalPathSU = 0;
@@ -772,77 +959,57 @@
       }
     }
 
-    PrescanInstruction(MI);
-
-    if (MI->getDesc().hasExtraDefRegAllocReq())
-      // If this instruction's defs have special allocation requirement, don't
-      // break this anti-dependency.
+    // Determine AntiDepReg's register group.
+    const unsigned GroupIndex = AntiDepReg != 0 ? GetGroup(AntiDepReg) : 0;
+    if (GroupIndex == 0) {
+      DEBUG(if (AntiDepReg != 0) dbgStr += " (zero group)");
       AntiDepReg = 0;
-    else if (AntiDepReg) {
-      // If this instruction has a use of AntiDepReg, breaking it
-      // is invalid.
-      for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
-        MachineOperand &MO = MI->getOperand(i);
-        if (!MO.isReg()) continue;
-        unsigned Reg = MO.getReg();
-        if (Reg == 0) continue;
-        if (MO.isUse() && AntiDepReg == Reg) {
-          AntiDepReg = 0;
-          break;
-        }
-      }
     }
 
-    // Determine AntiDepReg's register class, if it is live and is
-    // consistently used within a single class.
-    const TargetRegisterClass *RC = AntiDepReg != 0 ? Classes[AntiDepReg] : 0;
-    assert((AntiDepReg == 0 || RC != NULL) &&
-           "Register should be live if it's causing an anti-dependence!");
-    if (RC == reinterpret_cast<TargetRegisterClass *>(-1))
-      AntiDepReg = 0;
+    DEBUG(if (!dbgStr.empty()) errs() << dbgStr << '\n');
 
-    // Look for a suitable register to use to break the anti-depenence.
+    // Look for a suitable register to use to break the anti-dependence.
     //
     // TODO: Instead of picking the first free register, consider which might
     // be the best.
     if (AntiDepReg != 0) {
-      if (unsigned NewReg = findSuitableFreeRegister(AntiDepReg,
-                                                     LastNewReg[AntiDepReg],
-                                                     RC)) {
-        DEBUG(errs() << "Breaking anti-dependence edge on "
+      if (unsigned NewReg = FindSuitableFreeRegister(AntiDepReg,
+                                                     LastNewReg[AntiDepReg])) {
+        DEBUG(errs() << "\tBreaking anti-dependence edge on "
               << TRI->getName(AntiDepReg)
               << " with " << RegRefs.count(AntiDepReg) << " references"
               << " using " << TRI->getName(NewReg) << "!\n");
 
         // Update the references to the old register to refer to the new
         // register.
-        std::pair<std::multimap<unsigned, MachineOperand *>::iterator,
-                  std::multimap<unsigned, MachineOperand *>::iterator>
+        std::pair<std::multimap<unsigned, RegisterReference>::iterator,
+                  std::multimap<unsigned, RegisterReference>::iterator>
            Range = RegRefs.equal_range(AntiDepReg);
-        for (std::multimap<unsigned, MachineOperand *>::iterator
+        for (std::multimap<unsigned, RegisterReference>::iterator
              Q = Range.first, QE = Range.second; Q != QE; ++Q)
-          Q->second->setReg(NewReg);
+          Q->second.Operand->setReg(NewReg);
 
         // We just went back in time and modified history; the
-        // liveness information for the anti-depenence reg is now
+        // liveness information for the anti-dependence reg is now
         // inconsistent. Set the state as if it were dead.
-        Classes[NewReg] = Classes[AntiDepReg];
+        // FIXME forall in group
+        UnionGroups(NewReg, 0);
+        RegRefs.erase(NewReg);
         DefIndices[NewReg] = DefIndices[AntiDepReg];
         KillIndices[NewReg] = KillIndices[AntiDepReg];
-        assert(((KillIndices[NewReg] == ~0u) !=
-                (DefIndices[NewReg] == ~0u)) &&
-             "Kill and Def maps aren't consistent for NewReg!");
 
-        Classes[AntiDepReg] = 0;
+        // FIXME forall in group
+        UnionGroups(AntiDepReg, 0);
+        RegRefs.erase(AntiDepReg);
         DefIndices[AntiDepReg] = KillIndices[AntiDepReg];
         KillIndices[AntiDepReg] = ~0u;
         assert(((KillIndices[AntiDepReg] == ~0u) !=
                 (DefIndices[AntiDepReg] == ~0u)) &&
              "Kill and Def maps aren't consistent for AntiDepReg!");
 
-        RegRefs.erase(AntiDepReg);
         Changed = true;
         LastNewReg[AntiDepReg] = NewReg;
+        ++NumFixedAnti;
       }
     }
 
@@ -921,7 +1088,7 @@
     }
   }
 
-  if(AllDead)
+  if (AllDead)
     MO.setIsKill(true);
   return false;
 }

Modified: llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll?rev=84658&r1=84657&r2=84658&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll (original)
+++ llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll Tue Oct 20 14:54:44 2009
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=thumbv7-apple-darwin -mcpu=cortex-a8 | FileCheck %s
+; RUN: llc < %s -mtriple=thumbv7-apple-darwin -mcpu=cortex-a8 -break-anti-dependencies=none | FileCheck %s
 
 %struct.pix_pos = type { i32, i32, i32, i32, i32, i32 }
 

Modified: llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll?rev=84658&r1=84657&r2=84658&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll (original)
+++ llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll Tue Oct 20 14:54:44 2009
@@ -1,7 +1,7 @@
-; RUN: llc < %s -march=x86-64 -post-RA-scheduler -break-anti-dependencies=false > %t
+; RUN: llc < %s -march=x86-64 -post-RA-scheduler -break-anti-dependencies=none > %t
 ; RUN:   grep {%xmm0} %t | count 14
 ; RUN:   not grep {%xmm1} %t
-; RUN: llc < %s -march=x86-64 -post-RA-scheduler -break-anti-dependencies > %t
+; RUN: llc < %s -march=x86-64 -post-RA-scheduler -break-anti-dependencies=critical > %t
 ; RUN:   grep {%xmm0} %t | count 7
 ; RUN:   grep {%xmm1} %t | count 7
 

Modified: llvm/trunk/test/CodeGen/X86/sse3.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse3.ll?rev=84658&r1=84657&r2=84658&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/sse3.ll (original)
+++ llvm/trunk/test/CodeGen/X86/sse3.ll Tue Oct 20 14:54:44 2009
@@ -51,9 +51,9 @@
 	%tmp = shufflevector <8 x i16> %A, <8 x i16> %A, <8 x i32> < i32 8, i32 3, i32 2, i32 13, i32 7, i32 6, i32 5, i32 4 >
 	ret <8 x i16> %tmp
 ; X64: t3:
+; X64: 	pshuflw	$44, %xmm0, %xmm1
 ; X64: 	pextrw	$5, %xmm0, %eax
-; X64: 	pshuflw	$44, %xmm0, %xmm0
-; X64: 	pshufhw	$27, %xmm0, %xmm0
+; X64: 	pshufhw	$27, %xmm1, %xmm0
 ; X64: 	pinsrw	$3, %eax, %xmm0
 ; X64: 	ret
 }
@@ -168,11 +168,11 @@
         ret void
 ; X64: 	t10:
 ; X64: 		pextrw	$4, %xmm0, %eax
-; X64: 		pextrw	$6, %xmm0, %edx
+; X64: 		pextrw	$6, %xmm0, %ecx
 ; X64: 		movlhps	%xmm1, %xmm1
 ; X64: 		pshuflw	$8, %xmm1, %xmm1
 ; X64: 		pinsrw	$2, %eax, %xmm1
-; X64: 		pinsrw	$3, %edx, %xmm1
+; X64: 		pinsrw	$3, %ecx, %xmm1
 }
 
 
@@ -250,9 +250,9 @@
         %tmp9 = shufflevector <16 x i8> %tmp8, <16 x i8> %T0,  <16 x i32> < i32 0, i32 1, i32 2, i32 17,  i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef , i32 undef >
         ret <16 x i8> %tmp9
 ; X64: 	t16:
+; X64: 		pextrw	$8, %xmm0, %ecx
 ; X64: 		pinsrw	$0, %eax, %xmm1
-; X64: 		pextrw	$8, %xmm0, %eax
-; X64: 		pinsrw	$1, %eax, %xmm1
+; X64: 		pinsrw	$1, %ecx, %xmm1
 ; X64: 		pextrw	$1, %xmm1, %ecx
 ; X64: 		movd	%xmm1, %edx
 ; X64: 		pinsrw	$0, %edx, %xmm1




From gohman at apple.com  Tue Oct 20 15:06:09 2009
From: gohman at apple.com (Dan Gohman)
Date: Tue, 20 Oct 2009 20:06:09 -0000
Subject: [llvm-commits] [llvm] r84661 -
	/llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp
Message-ID: <200910202006.n9KK69rS016820@zion.cs.uiuc.edu>

Author: djg
Date: Tue Oct 20 15:06:09 2009
New Revision: 84661

URL: http://llvm.org/viewvc/llvm-project?rev=84661&view=rev
Log:
Restore LoopUnswitch's block-oriented threshold. LoopUnswitch now checks both
the estimated code size and the number of blocks when deciding whether to
do a non-trivial unswitch. This protects it from some very undesirable
worst-case behavior on large numbers of loop-unswitchable conditions, such
as in the testcase in PR5259.

Modified:
    llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp

Modified: llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp?rev=84661&r1=84660&r2=84661&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/LoopUnswitch.cpp Tue Oct 20 15:06:09 2009
@@ -138,7 +138,6 @@
     void SplitExitEdges(Loop *L, const SmallVector<BasicBlock *, 8> &ExitBlocks);
 
     bool UnswitchIfProfitable(Value *LoopCond, Constant *Val);
-    unsigned getLoopUnswitchCost(Value *LIC);
     void UnswitchTrivialCondition(Loop *L, Value *Cond, Constant *Val,
                                   BasicBlock *ExitBlock);
     void UnswitchNontrivialCondition(Value *LIC, Constant *OnVal, Loop *L);
@@ -400,25 +399,6 @@
   return true;
 }
 
-/// getLoopUnswitchCost - Return the cost (code size growth) that will happen if
-/// we choose to unswitch current loop on the specified value.
-///
-unsigned LoopUnswitch::getLoopUnswitchCost(Value *LIC) {
-  // If the condition is trivial, always unswitch.  There is no code growth for
-  // this case.
-  if (IsTrivialUnswitchCondition(LIC))
-    return 0;
-  
-  // FIXME: This is overly conservative because it does not take into
-  // consideration code simplification opportunities.
-  CodeMetrics Metrics;
-  for (Loop::block_iterator I = currentLoop->block_begin(), 
-         E = currentLoop->block_end();
-       I != E; ++I)
-    Metrics.analyzeBasicBlock(*I);
-  return Metrics.NumInsts;
-}
-
 /// UnswitchIfProfitable - We have found that we can unswitch currentLoop when
 /// LoopCond == Val to simplify the loop.  If we decide that this is profitable,
 /// unswitch the loop, reprocess the pieces, then return true.
@@ -427,24 +407,35 @@
   initLoopData();
   Function *F = loopHeader->getParent();
 
+  // If the condition is trivial, always unswitch.  There is no code growth for
+  // this case.
+  if (!IsTrivialUnswitchCondition(LoopCond)) {
+    // Check to see if it would be profitable to unswitch current loop.
 
-  // Check to see if it would be profitable to unswitch current loop.
-  unsigned Cost = getLoopUnswitchCost(LoopCond);
+    // Do not do non-trivial unswitch while optimizing for size.
+    if (OptimizeForSize || F->hasFnAttr(Attribute::OptimizeForSize))
+      return false;
 
-  // Do not do non-trivial unswitch while optimizing for size.
-  if (Cost && OptimizeForSize)
-    return false;
-  if (Cost && !F->isDeclaration() && F->hasFnAttr(Attribute::OptimizeForSize))
-    return false;
-
-  if (Cost > Threshold) {
-    // FIXME: this should estimate growth by the amount of code shared by the
-    // resultant unswitched loops.
-    //
-    DEBUG(errs() << "NOT unswitching loop %"
-          << currentLoop->getHeader()->getName() << ", cost too high: "
-          << currentLoop->getBlocks().size() << "\n");
-    return false;
+    // FIXME: This is overly conservative because it does not take into
+    // consideration code simplification opportunities and code that can
+    // be shared by the resultant unswitched loops.
+    CodeMetrics Metrics;
+    for (Loop::block_iterator I = currentLoop->block_begin(), 
+           E = currentLoop->block_end();
+         I != E; ++I)
+      Metrics.analyzeBasicBlock(*I);
+
+    // Limit the number of instructions to avoid causing significant code
+    // expansion, and the number of basic blocks, to avoid loops with
+    // large numbers of branches which cause loop unswitching to go crazy.
+    // This is a very ad-hoc heuristic.
+    if (Metrics.NumInsts > Threshold ||
+        Metrics.NumBlocks * 5 > Threshold) {
+      DEBUG(errs() << "NOT unswitching loop %"
+            << currentLoop->getHeader()->getName() << ", cost too high: "
+            << currentLoop->getBlocks().size() << "\n");
+      return false;
+    }
   }
 
   Constant *CondVal;




From grosbach at apple.com  Tue Oct 20 15:19:50 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Tue, 20 Oct 2009 20:19:50 -0000
Subject: [llvm-commits] [llvm] r84664 -
	/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
Message-ID: <200910202019.n9KKJpXc017351@zion.cs.uiuc.edu>

Author: grosbach
Date: Tue Oct 20 15:19:50 2009
New Revision: 84664

URL: http://llvm.org/viewvc/llvm-project?rev=84664&view=rev
Log:
add cmd line opt to disable frame index reuse for ARM and T2. debug aid.

Modified:
    llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp

Modified: llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp?rev=84664&r1=84663&r2=84664&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp Tue Oct 20 15:19:50 2009
@@ -36,8 +36,13 @@
 #include "llvm/Target/TargetOptions.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/SmallVector.h"
+#include "llvm/Support/CommandLine.h"
 using namespace llvm;
 
+static cl::opt<bool>
+ReuseFrameIndexVals("arm-reuse-frame-index-vals", cl::Hidden, cl::init(true),
+          cl::desc("Reuse repeated frame index values"));
+
 unsigned ARMBaseRegisterInfo::getRegisterNumbering(unsigned RegEnum,
                                                    bool *isSPVFP) {
   if (isSPVFP)
@@ -1088,6 +1093,8 @@
                              Offset, Pred, PredReg, TII);
     }
     MI.getOperand(i).ChangeToRegister(ScratchReg, false, false, true);
+    if (!ReuseFrameIndexVals)
+      ScratchReg = 0;
   }
   return ScratchReg;
 }




From dpatel at apple.com  Tue Oct 20 15:26:17 2009
From: dpatel at apple.com (Devang Patel)
Date: Tue, 20 Oct 2009 20:26:17 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84666 -
	/llvm-gcc-4.2/trunk/gcc/llvm-debug.cpp
Message-ID: <200910202026.n9KKQIWE017678@zion.cs.uiuc.edu>

Author: dpatel
Date: Tue Oct 20 15:26:17 2009
New Revision: 84666

URL: http://llvm.org/viewvc/llvm-project?rev=84666&view=rev
Log:
Do not cache forward struct types.

Modified:
    llvm-gcc-4.2/trunk/gcc/llvm-debug.cpp

Modified: llvm-gcc-4.2/trunk/gcc/llvm-debug.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/llvm-debug.cpp?rev=84666&r1=84665&r2=84666&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/llvm-debug.cpp (original)
+++ llvm-gcc-4.2/trunk/gcc/llvm-debug.cpp Tue Oct 20 15:26:17 2009
@@ -865,7 +865,9 @@
     
     case POINTER_TYPE:
     case REFERENCE_TYPE:
-      Ty = createPointerType(type);
+      // Do not cache pointer type. The pointer may point to forward declared
+      // struct.
+      return createPointerType(type);
       break;
 
     case BLOCK_POINTER_TYPE: {
@@ -900,7 +902,7 @@
     case RECORD_TYPE:
     case QUAL_UNION_TYPE:
     case UNION_TYPE: 
-      Ty = createStructType(type);
+      return createStructType(type);
       break;
 
     case INTEGER_TYPE:




From sabre at nondot.org  Tue Oct 20 15:27:50 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 20:27:50 -0000
Subject: [llvm-commits] [llvm] r84667 - in /llvm/trunk:
 lib/Transforms/Utils/SSAUpdater.cpp
 test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll
 test/Transforms/JumpThreading/crash.ll
Message-ID: <200910202027.n9KKRo1u017760@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 15:27:49 2009
New Revision: 84667

URL: http://llvm.org/viewvc/llvm-project?rev=84667&view=rev
Log:
alternate fix for PR5258 which avoids worklist problems, with reduced testcase.


Removed:
    llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll
Modified:
    llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp
    llvm/trunk/test/Transforms/JumpThreading/crash.ll

Modified: llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp?rev=84667&r1=84666&r2=84667&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp (original)
+++ llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp Tue Oct 20 15:27:49 2009
@@ -177,19 +177,14 @@
 /// which use their value in the corresponding predecessor.
 void SSAUpdater::RewriteUse(Use &U) {
   Instruction *User = cast<Instruction>(U.getUser());
-  BasicBlock *UseBB = User->getParent();
-  PHINode *UserPN = dyn_cast<PHINode>(User);
-  if (UserPN)
-    UseBB = UserPN->getIncomingBlock(U);
+  
+  Value *V;
+  if (PHINode *UserPN = dyn_cast<PHINode>(User))
+    V = GetValueAtEndOfBlock(UserPN->getIncomingBlock(U));
+  else
+    V = GetValueInMiddleOfBlock(User->getParent());
 
-  Value *V = GetValueInMiddleOfBlock(UseBB);
   U.set(V);
-  if (UserPN) {
-    // Incoming value from the same BB must be consistent
-    for (unsigned i=0;i<UserPN->getNumIncomingValues();i++)
-      if (UserPN->getIncomingBlock(i) == UseBB)
-        UserPN->setIncomingValue(i, V);
-  }
 }
 
 

Removed: llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll?rev=84666&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll (original)
+++ llvm/trunk/test/Transforms/JumpThreading/2009-10-20-jumpthreading-phi.ll (removed)
@@ -1,348 +0,0 @@
-; RUN: opt -jump-threading -verify %s -disable-output
-; PR5258
-target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
-target triple = "x86_64-unknown-linux-gnu"
-
-%0 = type { i64, [12 x i32] }
-%1 = type { %2 }
-%2 = type { i32, i32, i32, i32, i32, i32, %3 }
-%3 = type { %3*, %3* }
-%4 = type { i32 }
-%5 = type { %6*, %84, %87, %89, %92, %95, %69, %79*, %102*, %69, %69, %69 }
-%6 = type { %7* }
-%7 = type { %8, %10, %28, %34, %38, %40, %40, %40, %40, %40, %40, %40, %21*, %21*, %45, %46, %45, %45, %15, %15, %15, %15, %15, %15, %15, %15, %49, %49, %49, %49, %49, %9, %9, %51, %51, %51, %51, %51, %51, %53, %56 }
-%8 = type { %9, i32, i32 }
-%9 = type { i8* }
-%10 = type { i32, %11*, i32, i32 }
-%11 = type { %12, %21* }
-%12 = type { %13, %15* }
-%13 = type { i32, %14 }
-%14 = type { i64 }
-%15 = type { %16, i8, [3 x i8], i32, %6*, %15*, %17, i32, %20* }
-%16 = type { i32 (...)** }
-%17 = type { %18 }
-%18 = type { %19 }
-%19 = type { %16**, %16**, %16** }
-%20 = type { %15*, %16* }
-%21 = type { %22, %13 }
-%22 = type { %23 }
-%23 = type { %24, %26*, i32 }
-%24 = type { i32 (...)**, i8, i8, i8, i16, %25, %26*, %27* }
-%25 = type { %15* }
-%26 = type { %24*, %26*, %14 }
-%27 = type { %4, %24* }
-%28 = type { i32, %29*, i32, i32 }
-%29 = type { %30, %33* }
-%30 = type { %31 }
-%31 = type { %32*, %14, i16, i16 }
-%32 = type { i16, i16, i32, i32 }
-%33 = type { %22, %31 }
-%34 = type { %35, %37 }
-%35 = type { %36*, i32, i32, i32, i32 }
-%36 = type { i32, %4* }
-%37 = type <{ i8 }>
-%38 = type { %39 }
-%39 = type { i32 (...)**, i8**, i32, i32 }
-%40 = type { %16, %41, %41, %41, %45 }
-%41 = type { %42 }
-%42 = type { %43 }
-%43 = type { %37, %44, i64 }
-%44 = type { i32, %44*, %44*, %44* }
-%45 = type { %9, i32, i8 }
-%46 = type { %47, %24*, i8* }
-%47 = type { %48, [8 x i8*] }
-%48 = type { i8**, i32, i32, i32, [1 x i8*] }
-%49 = type { %50 }
-%50 = type { %15 }
-%51 = type { %52, %41 }
-%52 = type { %41 }
-%53 = type { i32, %54*, i32, i32 }
-%54 = type { %24*, %55* }
-%55 = type { %14, %55*, %24* }
-%56 = type { %57, %34 }
-%57 = type { i32, %58*, i32, i32 }
-%58 = type <{ %59*, [8 x i8], %83 }>
-%59 = type { %23, %60, %62* }
-%60 = type { %61, %59* }
-%61 = type { %59* }
-%62 = type { %24, %63, %65, %67* }
-%63 = type { %64, %62* }
-%64 = type { %62* }
-%65 = type { %66, %59* }
-%66 = type { %61 }
-%67 = type { %68, %70, %72, %74, %79*, %80 }
-%68 = type { %22, %5*, i32, %69 }
-%69 = type { %9 }
-%70 = type { %71, %67* }
-%71 = type { %67* }
-%72 = type { %73, %62* }
-%73 = type { %64 }
-%74 = type { %75, %77* }
-%75 = type { %76 }
-%76 = type { %77* }
-%77 = type { %24, %78, %67* }
-%78 = type { %76, %77* }
-%79 = type { %34, i32 }
-%80 = type { %81* }
-%81 = type <{ %9, i32, [4 x i8], %82 }>
-%82 = type <{ [33 x i8], [31 x i8] }>
-%83 = type <{ [33 x i8], [63 x i8] }>
-%84 = type { %85* }
-%85 = type { %68, %86, i8 }
-%86 = type { %84, %85* }
-%87 = type { %88, %67* }
-%88 = type { %70 }
-%89 = type { %90* }
-%90 = type { %68, %91 }
-%91 = type { %89, %90* }
-%92 = type { %93 }
-%93 = type { %94 }
-%94 = type { %69*, %69*, %69* }
-%95 = type { %96, %99* }
-%96 = type { %97 }
-%97 = type { %98, %99* }
-%98 = type { %99* }
-%99 = type <{ %100, %97, %5*, %101 }>
-%100 = type { [52 x i8], i32 }
-%101 = type <{ [33 x i8], [95 x i8] }>
-%102 = type { %16, %41, i32 }
-
- at _ZL20__gthrw_pthread_oncePiPFvvE = alias weak i32 (i32*, void ()*)* @pthread_once ; <i32 (i32*, void ()*)*> [#uses=0]
- at _ZL27__gthrw_pthread_getspecificj = alias weak i8* (i32)* @pthread_getspecific ; <i8* (i32)*> [#uses=0]
- at _ZL27__gthrw_pthread_setspecificjPKv = alias weak i32 (i32, i8*)* @pthread_setspecific ; <i32 (i32, i8*)*> [#uses=0]
- at _ZL22__gthrw_pthread_createPmPK14pthread_attr_tPFPvS3_ES3_ = alias weak i32 (i64*, %0*, i8* (i8*)*, i8*)* @pthread_create ; <i32 (i64*, %0*, i8* (i8*)*, i8*)*> [#uses=0]
- at _ZL22__gthrw_pthread_cancelm = alias weak i32 (i64)* @pthread_cancel ; <i32 (i64)*> [#uses=0]
- at _ZL26__gthrw_pthread_mutex_lockP15pthread_mutex_t = alias weak i32 (%1*)* @pthread_mutex_lock ; <i32 (%1*)*> [#uses=0]
- at _ZL29__gthrw_pthread_mutex_trylockP15pthread_mutex_t = alias weak i32 (%1*)* @pthread_mutex_trylock ; <i32 (%1*)*> [#uses=0]
- at _ZL28__gthrw_pthread_mutex_unlockP15pthread_mutex_t = alias weak i32 (%1*)* @pthread_mutex_unlock ; <i32 (%1*)*> [#uses=0]
- at _ZL26__gthrw_pthread_mutex_initP15pthread_mutex_tPK19pthread_mutexattr_t = alias weak i32 (%1*, %4*)* @pthread_mutex_init ; <i32 (%1*, %4*)*> [#uses=0]
- at _ZL26__gthrw_pthread_key_createPjPFvPvE = alias weak i32 (i32*, void (i8*)*)* @pthread_key_create ; <i32 (i32*, void (i8*)*)*> [#uses=0]
- at _ZL26__gthrw_pthread_key_deletej = alias weak i32 (i32)* @pthread_key_delete ; <i32 (i32)*> [#uses=0]
- at _ZL30__gthrw_pthread_mutexattr_initP19pthread_mutexattr_t = alias weak i32 (%4*)* @pthread_mutexattr_init ; <i32 (%4*)*> [#uses=0]
- at _ZL33__gthrw_pthread_mutexattr_settypeP19pthread_mutexattr_ti = alias weak i32 (%4*, i32)* @pthread_mutexattr_settype ; <i32 (%4*, i32)*> [#uses=0]
- at _ZL33__gthrw_pthread_mutexattr_destroyP19pthread_mutexattr_t = alias weak i32 (%4*)* @pthread_mutexattr_destroy ; <i32 (%4*)*> [#uses=0]
-
-define fastcc zeroext i8 @_ZN4llvm6Linker11LinkModulesEPNS_6ModuleES2_PSs(%5*, %5*, %69*) nounwind align 2 {
-  br i1 false, label %4, label %5
-
-; <label>:4                                       ; preds = %3
-  unreachable
-
-; <label>:5                                       ; preds = %3
-  br i1 false, label %6, label %7
-
-; <label>:6                                       ; preds = %5
-  unreachable
-
-; <label>:7                                       ; preds = %5
-  br i1 false, label %8, label %11
-
-; <label>:8                                       ; preds = %7
-  br i1 undef, label %10, label %9
-
-; <label>:9                                       ; preds = %8
-  unreachable
-
-; <label>:10                                      ; preds = %8
-  unreachable
-
-; <label>:11                                      ; preds = %7
-  br i1 undef, label %13, label %12
-
-; <label>:12                                      ; preds = %11
-  unreachable
-
-; <label>:13                                      ; preds = %11
-  br i1 undef, label %15, label %14
-
-; <label>:14                                      ; preds = %13
-  br label %15
-
-; <label>:15                                      ; preds = %14, %13
-  br i1 undef, label %17, label %16
-
-; <label>:16                                      ; preds = %15
-  unreachable
-
-; <label>:17                                      ; preds = %15
-  br i1 undef, label %19, label %18
-
-; <label>:18                                      ; preds = %17
-  unreachable
-
-; <label>:19                                      ; preds = %17
-  br i1 false, label %20, label %21
-
-; <label>:20                                      ; preds = %19
-  unreachable
-
-; <label>:21                                      ; preds = %19
-  br i1 undef, label %22, label %23
-
-; <label>:22                                      ; preds = %21
-  br label %23
-
-; <label>:23                                      ; preds = %22, %21
-  br i1 false, label %24, label %25
-
-; <label>:24                                      ; preds = %23
-  unreachable
-
-; <label>:25                                      ; preds = %23
-  br i1 undef, label %29, label %26
-
-; <label>:26                                      ; preds = %25
-  br i1 undef, label %28, label %27
-
-; <label>:27                                      ; preds = %26
-  unreachable
-
-; <label>:28                                      ; preds = %26
-  unreachable
-
-; <label>:29                                      ; preds = %25
-  br i1 undef, label %31, label %30
-
-; <label>:30                                      ; preds = %29
-  unreachable
-
-; <label>:31                                      ; preds = %29
-  br i1 undef, label %32, label %33
-
-; <label>:32                                      ; preds = %31
-  br label %33
-
-; <label>:33                                      ; preds = %32, %31
-  br i1 false, label %34, label %35
-
-; <label>:34                                      ; preds = %33
-  unreachable
-
-; <label>:35                                      ; preds = %33
-  br i1 undef, label %36, label %37
-
-; <label>:36                                      ; preds = %35
-  br label %37
-
-; <label>:37                                      ; preds = %36, %35
-  br i1 undef, label %39, label %38
-
-; <label>:38                                      ; preds = %37
-  br i1 false, label %39, label %40
-
-; <label>:39                                      ; preds = %38, %37
-  unreachable
-
-; <label>:40                                      ; preds = %38
-  %41 = load i8* undef, align 8                   ; <i8> [#uses=1]
-  switch i8 %41, label %42 [
-    i8 4, label %43
-    i8 2, label %43
-    i8 3, label %43
-  ]
-
-; <label>:42                                      ; preds = %40
-  unreachable
-
-; <label>:43                                      ; preds = %40, %40, %40
-  %44 = trunc i32 undef to i5                     ; <i5> [#uses=1]
-  switch i5 %44, label %45 [
-    i5 7, label %50
-    i5 9, label %50
-  ]
-
-; <label>:45                                      ; preds = %43
-  br i1 undef, label %47, label %46
-
-; <label>:46                                      ; preds = %45
-  br label %47
-
-; <label>:47                                      ; preds = %46, %45
-  %48 = icmp eq %85* null, null                   ; <i1> [#uses=1]
-  br i1 %48, label %50, label %49
-
-; <label>:49                                      ; preds = %47
-  unreachable
-
-; <label>:50                                      ; preds = %47, %43, %43
-  %51 = phi %68* [ null, %43 ], [ undef, %47 ], [ null, %43 ] ; <%68*> [#uses=4]
-  %52 = phi %68* [ null, %43 ], [ undef, %47 ], [ null, %43 ] ; <%68*> [#uses=1]
-  %53 = icmp eq %68* %52, null                    ; <i1> [#uses=1]
-  br i1 %53, label %54, label %59
-
-; <label>:54                                      ; preds = %50
-  %55 = trunc i32 undef to i5                     ; <i5> [#uses=1]
-  switch i5 %55, label %56 [
-    i5 7, label %59
-    i5 8, label %59
-    i5 9, label %59
-  ]
-
-; <label>:56                                      ; preds = %54
-  br i1 undef, label %58, label %57
-
-; <label>:57                                      ; preds = %56
-  br label %58
-
-; <label>:58                                      ; preds = %57, %56
-  br label %59
-
-; <label>:59                                      ; preds = %58, %54, %54, %54, %50
-  %60 = phi %68* [ %51, %50 ], [ %51, %54 ], [ %51, %54 ], [ %51, %54 ], [ undef, %58 ] ; <%68*> [#uses=0]
-  br i1 undef, label %62, label %61
-
-; <label>:61                                      ; preds = %59
-  br label %62
-
-; <label>:62                                      ; preds = %61, %59
-  switch i8 undef, label %64 [
-    i8 3, label %63
-    i8 2, label %65
-  ]
-
-; <label>:63                                      ; preds = %62
-  unreachable
-
-; <label>:64                                      ; preds = %62
-  unreachable
-
-; <label>:65                                      ; preds = %62
-  switch i8 undef, label %67 [
-    i8 0, label %66
-    i8 1, label %66
-  ]
-
-; <label>:66                                      ; preds = %65, %65
-  unreachable
-
-; <label>:67                                      ; preds = %65
-  unreachable
-}
-
-declare i32 @pthread_mutex_trylock(%1*) nounwind
-
-declare i32 @pthread_mutex_unlock(%1*) nounwind
-
-declare i32 @pthread_mutex_lock(%1*) nounwind
-
-declare i32 @pthread_mutexattr_init(%4*) nounwind
-
-declare i32 @pthread_mutexattr_settype(%4*, i32) nounwind
-
-declare i32 @pthread_mutex_init(%1*, %4*) nounwind
-
-declare i32 @pthread_mutexattr_destroy(%4*) nounwind
-
-declare i8* @pthread_getspecific(i32) nounwind
-
-declare i32 @pthread_setspecific(i32, i8*) nounwind
-
-declare i32 @pthread_key_delete(i32) nounwind
-
-declare i32 @pthread_key_create(i32*, void (i8*)*) nounwind
-
-declare i32 @pthread_once(i32*, void ()*)
-
-declare i32 @pthread_create(i64*, %0*, i8* (i8*)*, i8*)
-
-declare i32 @pthread_cancel(i64)

Modified: llvm/trunk/test/Transforms/JumpThreading/crash.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/JumpThreading/crash.ll?rev=84667&r1=84666&r2=84667&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/JumpThreading/crash.ll (original)
+++ llvm/trunk/test/Transforms/JumpThreading/crash.ll Tue Oct 20 15:27:49 2009
@@ -54,3 +54,38 @@
 bb61:                                            
   ret void
 }
+
+
+; PR5258
+define i32 @test(i1 %cond, i1 %cond2, i32 %a) {
+A:
+  br i1 %cond, label %F, label %A1
+F:
+  br label %A1
+
+A1:  
+  %d = phi i1 [false, %A], [true, %F]
+  %e = add i32 %a, %a
+  br i1 %d, label %B, label %G
+  
+G:
+  br i1 %cond2, label %B, label %D
+  
+B:
+  %f = phi i32 [%e, %G], [%e, %A1]
+  %b = add i32 0, 0
+  switch i32 %a, label %C [
+    i32 7, label %D
+    i32 8, label %D
+    i32 9, label %D
+  ]
+
+C:
+  br label %D
+  
+D:
+  %c = phi i32 [%e, %B], [%e, %B], [%e, %B], [%f, %C], [%e, %G]
+  ret i32 %c
+E:
+  ret i32 412
+}
\ No newline at end of file




From grosbach at apple.com  Tue Oct 20 15:31:31 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Tue, 20 Oct 2009 20:31:31 -0000
Subject: [llvm-commits] [llvm] r84669 -
	/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
Message-ID: <200910202031.n9KKVVad017925@zion.cs.uiuc.edu>

Author: grosbach
Date: Tue Oct 20 15:31:31 2009
New Revision: 84669

URL: http://llvm.org/viewvc/llvm-project?rev=84669&view=rev
Log:
Disable by default while debugging

Modified:
    llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp

Modified: llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp?rev=84669&r1=84668&r2=84669&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp Tue Oct 20 15:31:31 2009
@@ -40,7 +40,7 @@
 using namespace llvm;
 
 static cl::opt<bool>
-ReuseFrameIndexVals("arm-reuse-frame-index-vals", cl::Hidden, cl::init(true),
+ReuseFrameIndexVals("arm-reuse-frame-index-vals", cl::Hidden, cl::init(false),
           cl::desc("Reuse repeated frame index values"));
 
 unsigned ARMBaseRegisterInfo::getRegisterNumbering(unsigned RegEnum,




From sabre at nondot.org  Tue Oct 20 15:33:46 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 20:33:46 -0000
Subject: [llvm-commits] [llvm] r84671 - in /llvm/trunk/test/Transforms/SCCP:
	basictest.ll sccptest.ll
Message-ID: <200910202033.n9KKXkGM018039@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 15:33:46 2009
New Revision: 84671

URL: http://llvm.org/viewvc/llvm-project?rev=84671&view=rev
Log:
merge two tests and convert to filecheck.

Removed:
    llvm/trunk/test/Transforms/SCCP/basictest.ll
Modified:
    llvm/trunk/test/Transforms/SCCP/sccptest.ll

Removed: llvm/trunk/test/Transforms/SCCP/basictest.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SCCP/basictest.ll?rev=84670&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/SCCP/basictest.ll (original)
+++ llvm/trunk/test/Transforms/SCCP/basictest.ll (removed)
@@ -1,17 +0,0 @@
-; This is a basic sanity check for constant propogation.  The add instruction 
-; should be eliminated.
-
-; RUN: opt < %s -sccp -S | not grep add
-
-define i32 @test(i1 %B) {
-	br i1 %B, label %BB1, label %BB2
-BB1:		; preds = %0
-	%Val = add i32 0, 0		; <i32> [#uses=1]
-	br label %BB3
-BB2:		; preds = %0
-	br label %BB3
-BB3:		; preds = %BB2, %BB1
-	%Ret = phi i32 [ %Val, %BB1 ], [ 1, %BB2 ]		; <i32> [#uses=1]
-	ret i32 %Ret
-}
-

Modified: llvm/trunk/test/Transforms/SCCP/sccptest.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SCCP/sccptest.ll?rev=84671&r1=84670&r2=84671&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/SCCP/sccptest.ll (original)
+++ llvm/trunk/test/Transforms/SCCP/sccptest.ll Tue Oct 20 15:33:46 2009
@@ -1,31 +1,58 @@
+; RUN: opt < %s -sccp -S | FileCheck %s
+
+; This is a basic sanity check for constant propagation.  The add instruction 
+; should be eliminated.
+
+define i32 @test1(i1 %B) {
+	br i1 %B, label %BB1, label %BB2
+BB1:		; preds = %0
+	%Val = add i32 0, 0		; <i32> [#uses=1]
+	br label %BB3
+BB2:		; preds = %0
+	br label %BB3
+BB3:		; preds = %BB2, %BB1
+	%Ret = phi i32 [ %Val, %BB1 ], [ 1, %BB2 ]		; <i32> [#uses=1]
+	ret i32 %Ret
+        
+; CHECK: @test1
+; CHECK: %Ret = phi i32 [ 0, %BB1 ], [ 1, %BB2 ]
+}
+
 ; This is the test case taken from appel's book that illustrates a hard case
-; that SCCP gets right. BB3 should be completely eliminated.
+; that SCCP gets right.
 ;
-; RUN: opt < %s -sccp -constprop -dce -simplifycfg | \
-; RUN:   llvm-dis | not grep BB3
-
-define i32 @testfunction(i32 %i0, i32 %j0) {
+define i32 @test2(i32 %i0, i32 %j0) {
+; CHECK: @test2
 BB1:
 	br label %BB2
-BB2:		; preds = %BB7, %BB1
-	%j2 = phi i32 [ %j4, %BB7 ], [ 1, %BB1 ]		; <i32> [#uses=2]
-	%k2 = phi i32 [ %k4, %BB7 ], [ 0, %BB1 ]		; <i32> [#uses=4]
-	%kcond = icmp slt i32 %k2, 100		; <i1> [#uses=1]
+BB2:
+	%j2 = phi i32 [ %j4, %BB7 ], [ 1, %BB1 ]
+	%k2 = phi i32 [ %k4, %BB7 ], [ 0, %BB1 ]
+	%kcond = icmp slt i32 %k2, 100
 	br i1 %kcond, label %BB3, label %BB4
-BB3:		; preds = %BB2
-	%jcond = icmp slt i32 %j2, 20		; <i1> [#uses=1]
+BB3:
+	%jcond = icmp slt i32 %j2, 20
 	br i1 %jcond, label %BB5, label %BB6
-BB4:		; preds = %BB2
+; CHECK: BB3:
+; CHECK-NEXT: br i1 true, label %BB5, label %BB6
+BB4:
 	ret i32 %j2
-BB5:		; preds = %BB3
-	%k3 = add i32 %k2, 1		; <i32> [#uses=1]
+; CHECK: BB4:
+; CHECK-NEXT: ret i32 1
+BB5:
+	%k3 = add i32 %k2, 1
 	br label %BB7
-BB6:		; preds = %BB3
-	%k5 = add i32 %k2, 1		; <i32> [#uses=1]
+BB6:
+	%k5 = add i32 %k2, 1
 	br label %BB7
-BB7:		; preds = %BB6, %BB5
-	%j4 = phi i32 [ 1, %BB5 ], [ %k2, %BB6 ]		; <i32> [#uses=1]
-	%k4 = phi i32 [ %k3, %BB5 ], [ %k5, %BB6 ]		; <i32> [#uses=1]
+; CHECK: BB6:
+; CHECK-NEXT: br label %BB7
+BB7:
+	%j4 = phi i32 [ 1, %BB5 ], [ %k2, %BB6 ]
+	%k4 = phi i32 [ %k3, %BB5 ], [ %k5, %BB6 ]
 	br label %BB2
+; CHECK: BB7:
+; CHECK-NEXT: %k4 = phi i32 [ %k3, %BB5 ], [ undef, %BB6 ]
+; CHECK-NEXT: br label %BB2
 }
 




From sabre at nondot.org  Tue Oct 20 15:39:44 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 20:39:44 -0000
Subject: [llvm-commits] [llvm] r84672 - in /llvm/trunk/test/Transforms/SCCP:
 ipsccp-basic.ll ipsccp-conditional.ll ipsccp-gvar.ll
Message-ID: <200910202039.n9KKdiJl018323@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 15:39:43 2009
New Revision: 84672

URL: http://llvm.org/viewvc/llvm-project?rev=84672&view=rev
Log:
merge and filecheckize

Removed:
    llvm/trunk/test/Transforms/SCCP/ipsccp-conditional.ll
    llvm/trunk/test/Transforms/SCCP/ipsccp-gvar.ll
Modified:
    llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll

Modified: llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll?rev=84672&r1=84671&r2=84672&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll (original)
+++ llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll Tue Oct 20 15:39:43 2009
@@ -1,13 +1,74 @@
-; RUN: opt < %s -ipsccp -S | \
-; RUN:   grep -v {ret i32 17} | grep -v {ret i32 undef} | not grep ret
+; RUN: opt < %s -ipsccp -S | FileCheck %s
 
-define internal i32 @bar(i32 %A) {
-	%X = add i32 1, 2		; <i32> [#uses=0]
+;;======================== test1
+
+define internal i32 @test1a(i32 %A) {
+	%X = add i32 1, 2
 	ret i32 %A
 }
+; CHECK: define internal i32 @test1a
+; CHECK: ret i32 undef
+
+define i32 @test1b() {
+	%X = call i32 @test1a( i32 17 )
+	ret i32 %X
+
+; CHECK: define i32 @test1b
+; CHECK: ret i32 17
+}
+
+
+
+;;======================== test2
+
+define internal i32 @test2a(i32 %A) {
+	%C = icmp eq i32 %A, 0	
+	br i1 %C, label %T, label %F
+T:
+	%B = call i32 @test2a( i32 0 )
+	ret i32 0
+F:
+	%C.upgrd.1 = call i32 @test2a(i32 1)
+	ret i32 %C.upgrd.1
+}
+; CHECK: define internal i32 @test2a
+; CHECK-NEXT: br label %T
+; CHECK: ret i32 undef
+
 
-define i32 @foo() {
-	%X = call i32 @bar( i32 17 )		; <i32> [#uses=1]
+define i32 @test2b() {
+	%X = call i32 @test2a(i32 0)
 	ret i32 %X
 }
+; CHECK: define i32 @test2b
+; CHECK-NEXT: %X = call i32 @test2a(i32 0)
+; CHECK-NEXT: ret i32 0
+
+
+;;======================== test3
+
+ at G = internal global i32 undef
 
+define void @test3a() {
+	%X = load i32* @G
+	store i32 %X, i32* @G
+	ret void
+}
+; CHECK: define void @test3a
+; CHECK-NEXT: ret void
+
+
+define i32 @test3b() {
+	%V = load i32* @G
+	%C = icmp eq i32 %V, 17
+	br i1 %C, label %T, label %F
+T:
+	store i32 17, i32* @G
+	ret i32 %V
+F:	
+	store i32 123, i32* @G
+	ret i32 0
+}
+; CHECK: define i32 @test3b
+; CHECK-NOT: store
+; CHECK: ret i32 0

Removed: llvm/trunk/test/Transforms/SCCP/ipsccp-conditional.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SCCP/ipsccp-conditional.ll?rev=84671&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/SCCP/ipsccp-conditional.ll (original)
+++ llvm/trunk/test/Transforms/SCCP/ipsccp-conditional.ll (removed)
@@ -1,19 +0,0 @@
-; RUN: opt < %s -ipsccp -S | \
-; RUN:   grep -v {ret i32 0} | grep -v {ret i32 undef} | not grep ret
-
-define internal i32 @bar(i32 %A) {
-	%C = icmp eq i32 %A, 0		; <i1> [#uses=1]
-	br i1 %C, label %T, label %F
-T:		; preds = %0
-	%B = call i32 @bar( i32 0 )		; <i32> [#uses=0]
-	ret i32 0
-F:		; preds = %0
-	%C.upgrd.1 = call i32 @bar( i32 1 )		; <i32> [#uses=1]
-	ret i32 %C.upgrd.1
-}
-
-define i32 @foo() {
-	%X = call i32 @bar( i32 0 )		; <i32> [#uses=1]
-	ret i32 %X
-}
-

Removed: llvm/trunk/test/Transforms/SCCP/ipsccp-gvar.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SCCP/ipsccp-gvar.ll?rev=84671&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/SCCP/ipsccp-gvar.ll (original)
+++ llvm/trunk/test/Transforms/SCCP/ipsccp-gvar.ll (removed)
@@ -1,22 +0,0 @@
-; RUN: opt < %s -ipsccp -S | not grep global
-
- at G = internal global i32 undef		; <i32*> [#uses=5]
-
-define void @foo() {
-	%X = load i32* @G		; <i32> [#uses=1]
-	store i32 %X, i32* @G
-	ret void
-}
-
-define i32 @bar() {
-	%V = load i32* @G		; <i32> [#uses=2]
-	%C = icmp eq i32 %V, 17		; <i1> [#uses=1]
-	br i1 %C, label %T, label %F
-T:		; preds = %0
-	store i32 17, i32* @G
-	ret i32 %V
-F:		; preds = %0
-	store i32 123, i32* @G
-	ret i32 0
-}
-




From gohman at apple.com  Tue Oct 20 15:41:13 2009
From: gohman at apple.com (Dan Gohman)
Date: Tue, 20 Oct 2009 20:41:13 -0000
Subject: [llvm-commits] [llvm] r84673 - /llvm/trunk/lib/Analysis/LoopInfo.cpp
Message-ID: <200910202041.n9KKfDvl018481@zion.cs.uiuc.edu>

Author: djg
Date: Tue Oct 20 15:41:13 2009
New Revision: 84673

URL: http://llvm.org/viewvc/llvm-project?rev=84673&view=rev
Log:
Fix another place that calls Loop::contains a lot to construct a sorted
container of the blocks and do efficient lookups. This makes
isLoopSimplifyForm much faster on large loops, fixing a significant
compile-time issue in builds with assertions enabled.

Modified:
    llvm/trunk/lib/Analysis/LoopInfo.cpp

Modified: llvm/trunk/lib/Analysis/LoopInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/LoopInfo.cpp?rev=84673&r1=84672&r2=84673&view=diff

==============================================================================
--- llvm/trunk/lib/Analysis/LoopInfo.cpp (original)
+++ llvm/trunk/lib/Analysis/LoopInfo.cpp Tue Oct 20 15:41:13 2009
@@ -292,6 +292,9 @@
   // Normal-form loops have a single backedge.
   if (!getLoopLatch())
     return false;
+  // Sort the blocks vector so that we can use binary search to do quick
+  // lookups.
+  SmallPtrSet<BasicBlock *, 16> LoopBBs(block_begin(), block_end());
   // Each predecessor of each exit block of a normal loop is contained
   // within the loop.
   SmallVector<BasicBlock *, 4> ExitBlocks;
@@ -299,7 +302,7 @@
   for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i)
     for (pred_iterator PI = pred_begin(ExitBlocks[i]),
          PE = pred_end(ExitBlocks[i]); PI != PE; ++PI)
-      if (!contains(*PI))
+      if (!LoopBBs.count(*PI))
         return false;
   // All the requirements are met.
   return true;




From sabre at nondot.org  Tue Oct 20 15:57:59 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 20:57:59 -0000
Subject: [llvm-commits] [llvm] r84674 -
 /llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll
Message-ID: <200910202057.n9KKvxO3019145@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 15:57:58 2009
New Revision: 84674

URL: http://llvm.org/viewvc/llvm-project?rev=84674&view=rev
Log:
the date on this testcase is wrong, it is unreduced, and it passes without the fix for PR4313.

Removed:
    llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll

Removed: llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll?rev=84673&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll (original)
+++ llvm/trunk/test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll (removed)
@@ -1,716 +0,0 @@
-; RUN: opt -ipsccp <%s -S | FileCheck %s
-; PR4313
-; the return value of a multiple-return value invoke must not be left undefined
-target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
-target triple = "x86_64-unknown-linux-gnu"
-
-%0 = type <{ %1, %2, [4 x i8], %3 }>
-%1 = type { i32 (...)**, i32, i32, i32, i32, %1*, i8*, i8*, i8* }
-%2 = type { i32 }
-%3 = type <{ %4, %5 }>
-%4 = type { i32 (...)**, i8 }
-%5 = type <{ [33 x i8], [191 x i8] }>
-%6 = type { %1, %7, %10 }
-%7 = type { %8 }
-%8 = type { %9 }
-%9 = type { i8* }
-%10 = type { %11 }
-%11 = type { %12 }
-%12 = type { i32 (...)** }
-%13 = type { %1, %14, %18, %10 }
-%14 = type { %15 }
-%15 = type { %16 }
-%16 = type { %17 }
-%17 = type { %8*, %8*, %8* }
-%18 = type { %19 }
-%19 = type { %20 }
-%20 = type { i32*, i32*, i32* }
-%21 = type { %22, %22 }
-%22 = type { %23 }
-%23 = type { %24 }
-%24 = type { %25, %26, i64 }
-%25 = type <{ i8 }>
-%26 = type { i32, %26*, %26*, %26* }
-%27 = type { %28, %15, i8* }
-%28 = type { %29 }
-%29 = type { %30 }
-%30 = type { %31*, %31*, %31* }
-%31 = type { %32*, %9 }
-%32 = type { i32 (...)**, i8*, i8*, i8 }
-%33 = type { i64, [12 x i32] }
-%34 = type { %35 }
-%35 = type { i32, i32, i32, i32, i32, i32, %36 }
-%36 = type { %36*, %36* }
-%37 = type { i32, %8, %9, %15, %38, %42 }
-%38 = type { %39 }
-%39 = type { %40 }
-%40 = type { %41*, %41*, %41* }
-%41 = type { %8, %12*, i32, %12* }
-%42 = type { %43 }
-%43 = type { %44 }
-%44 = type { %37**, %37**, %37** }
-%45 = type { i32 (...)**, i8*, i8*, i8*, i32, i8 }
-%46 = type { %47, %37*, %12*, %8, %15, %37*, %50 }
-%47 = type { %48 }
-%48 = type { %49 }
-%49 = type { i8*, i8*, i8* }
-%50 = type { %51 }
-%51 = type { %52 }
-%52 = type { %46**, %46**, %46** }
-%53 = type { %21*, %54, %63, %63, %22, %22, %22, %22, %22, %22, %37*, %37*, %37*, %72 }
-%54 = type { %37*, %22, %55, %59, %18 }
-%55 = type { %56 }
-%56 = type { %57 }
-%57 = type { %58*, %58*, %58* }
-%58 = type { %37*, i32, i32 }
-%59 = type { %60 }
-%60 = type { %61 }
-%61 = type { %62*, %62*, %62* }
-%62 = type { %37*, %8, %42, %18, i32, i32, i32, %42, %8, %8 }
-%63 = type { %64 }
-%64 = type { %65 }
-%65 = type { %66*, %66*, %66* }
-%66 = type { %37*, %8, %8, %8, %8, %67, i32, i8, i8, %68 }
-%67 = type { %18, %42, %18, %42 }
-%68 = type { %69 }
-%69 = type { %70 }
-%70 = type { %71*, %71*, %71* }
-%71 = type { i32, i32 }
-%72 = type { %73 }
-%73 = type { %74 }
-%74 = type { %75*, %75*, %75* }
-%75 = type { %76*, %46*, %46*, %42, i32 }
-%76 = type { %77, %78 }
-%77 = type { %12, %12* }
-%78 = type { %79 }
-%79 = type { %80 }
-%80 = type { %12**, %12**, %12** }
-%81 = type { %12, %21*, %53 }
-%82 = type { %22, %8 }
-%83 = type { %26, %84 }
-%84 = type { i32, %22 }
-%s2i64 = type { i64, i64 }
-%85 = type { %26* }
-%86 = type { %27*, i8*, %32*, i8*, i32, %8, i64, i32 }
-%87 = type { %86, %88, %22, %95* }
-%88 = type { %89 }
-%89 = type { %90 }
-%90 = type { %91*, %91*, %91* }
-%91 = type { %92 }
-%92 = type { %93 }
-%93 = type { %94*, %94*, %94* }
-%94 = type { %8, %18, %12*, %9 }
-%95 = type { %37, %42 }
-
- at _ZNSs4_Rep20_S_empty_rep_storageE = external global [4 x i64] ; <[4 x i64]*> [#uses=1]
- at .str111723 = external constant [1 x i8], align 1 ; <[1 x i8]*> [#uses=1]
- at .str181730 = external constant [4 x i8], align 1 ; <[4 x i8]*> [#uses=1]
- at .str721784 = external constant [37 x i8], align 8 ; <[37 x i8]*> [#uses=1]
- at _ZN12_GLOBAL__N_16ActionE = external global %0, align 32 ; <%0*> [#uses=1]
- at _ZN12_GLOBAL__N_114OutputFilenameE = external global %6, align 32 ; <%6*> [#uses=1]
- at _ZN12_GLOBAL__N_111IncludeDirsE = external global %13, align 32 ; <%13*> [#uses=1]
- at .str533653 = external constant [2 x i8], align 1 ; <[2 x i8]*> [#uses=1]
- at _ZN4llvm7RecordsE = external global %21, align 32 ; <%21*> [#uses=2]
- at _ZL6SrcMgr = external global %27, align 32       ; <%27*> [#uses=2]
- at .str3723 = external constant [88 x i8], align 8  ; <[88 x i8]*> [#uses=1]
- at .str13724 = external constant [136 x i8], align 8 ; <[136 x i8]*> [#uses=1]
-
- at _ZL20__gthrw_pthread_oncePiPFvvE = alias weak i32 (i32*, void ()*)* @pthread_once ; <i32 (i32*, void ()*)*> [#uses=0]
- at _ZL27__gthrw_pthread_getspecificj = alias weak i8* (i32)* @pthread_getspecific ; <i8* (i32)*> [#uses=0]
- at _ZL27__gthrw_pthread_setspecificjPKv = alias weak i32 (i32, i8*)* @pthread_setspecific ; <i32 (i32, i8*)*> [#uses=0]
- at _ZL22__gthrw_pthread_createPmPK14pthread_attr_tPFPvS3_ES3_ = alias weak i32 (i64*, %33*, i8* (i8*)*, i8*)* @pthread_create ; <i32 (i64*, %33*, i8* (i8*)*, i8*)*> [#uses=0]
- at _ZL22__gthrw_pthread_cancelm = alias weak i32 (i64)* @pthread_cancel ; <i32 (i64)*> [#uses=0]
- at _ZL26__gthrw_pthread_mutex_lockP15pthread_mutex_t = alias weak i32 (%34*)* @pthread_mutex_lock ; <i32 (%34*)*> [#uses=0]
- at _ZL29__gthrw_pthread_mutex_trylockP15pthread_mutex_t = alias weak i32 (%34*)* @pthread_mutex_trylock ; <i32 (%34*)*> [#uses=0]
- at _ZL28__gthrw_pthread_mutex_unlockP15pthread_mutex_t = alias weak i32 (%34*)* @pthread_mutex_unlock ; <i32 (%34*)*> [#uses=0]
- at _ZL26__gthrw_pthread_mutex_initP15pthread_mutex_tPK19pthread_mutexattr_t = alias weak i32 (%34*, %2*)* @pthread_mutex_init ; <i32 (%34*, %2*)*> [#uses=0]
- at _ZL26__gthrw_pthread_key_createPjPFvPvE = alias weak i32 (i32*, void (i8*)*)* @pthread_key_create ; <i32 (i32*, void (i8*)*)*> [#uses=0]
- at _ZL26__gthrw_pthread_key_deletej = alias weak i32 (i32)* @pthread_key_delete ; <i32 (i32)*> [#uses=0]
- at _ZL30__gthrw_pthread_mutexattr_initP19pthread_mutexattr_t = alias weak i32 (%2*)* @pthread_mutexattr_init ; <i32 (%2*)*> [#uses=0]
- at _ZL33__gthrw_pthread_mutexattr_settypeP19pthread_mutexattr_ti = alias weak i32 (%2*, i32)* @pthread_mutexattr_settype ; <i32 (%2*, i32)*> [#uses=0]
- at _ZL33__gthrw_pthread_mutexattr_destroyP19pthread_mutexattr_t = alias weak i32 (%2*)* @pthread_mutexattr_destroy ; <i32 (%2*)*> [#uses=0]
-
-declare void @_ZNSsC1EPKcRKSaIcE(%8*, i8*, %25*)
-
-declare i8* @_Znwm(i64)
-
-declare zeroext i8 @_ZNK4llvm6Record12isSubClassOfENS_9StringRefE(%37*, i64, i64) align 2
-
-declare i32 @_ZNKSs7compareEPKc(%8*, i8*)
-
-declare %45* @_ZN4llvm11raw_ostreamlsEPKc(%45*, i8*) align 2
-
-declare void @_ZNSsC1ERKSs(%8*, %8*)
-
-declare %26* @_ZSt18_Rb_tree_incrementPSt18_Rb_tree_node_base(%26*)
-
-declare void @_ZStplIcSt11char_traitsIcESaIcEESbIT_T0_T1_EPKS3_RKS6_(%8* noalias sret, i8*, %8*)
-
-declare %15* @_ZNSt6vectorISsSaISsEEaSERKS1_(%15*, %15*) align 2
-
-declare void @_ZNSt6vectorISsSaISsEE9push_backERKSs(%15*, %8*) align 2
-
-declare i32 @_ZNK4llvm15TreePatternNode10getTypeNumEj(%46*, i32) align 2
-
-declare void @_ZN4llvm18CodeGenDAGPatternsD1Ev(%53*) align 2
-
-declare void @_ZN4llvm18CodeGenDAGPatternsC1ERNS_12RecordKeeperE(%53*, %21*) align 2
-
-declare void @_ZNK4llvm14PatternToMatch17getPredicateCheckEv(%8* noalias sret, %75*) align 2
-
-define internal void @0(%81*, %45*) align 2 {
-  invoke void @_ZNSsC1ERKSs(%8* undef, %8* null)
-          to label %3 unwind label %28
-
-; <label>:3                                       ; preds = %2
-  %4 = getelementptr inbounds i8* null, i64 -24   ; <i8*> [#uses=1]
-  %5 = icmp eq i8* %4, bitcast ([4 x i64]* @_ZNSs4_Rep20_S_empty_rep_storageE to i8*) ; <i1> [#uses=1]
-  br i1 %5, label %7, label %6
-
-; <label>:6                                       ; preds = %3
-  unreachable
-
-; <label>:7                                       ; preds = %3
-  br i1 undef, label %9, label %8
-
-; <label>:8                                       ; preds = %7
-  invoke void @_ZStplIcSt11char_traitsIcESaIcEESbIT_T0_T1_EPKS3_RKS6_(%8* noalias sret null, i8* getelementptr inbounds ([37 x i8]* @.str721784, i64 0, i64 0), %8* undef)
-          to label %10 unwind label %29
-
-; <label>:9                                       ; preds = %7
-  unreachable
-
-; <label>:10                                      ; preds = %8
-  invoke void @_ZNSsC1ERKSs(%8* null, %8* null)
-          to label %11 unwind label %30
-
-; <label>:11                                      ; preds = %10
-  %12 = invoke %45* @_ZN4llvm11raw_ostreamlsEPKc(%45* %1, i8* getelementptr inbounds ([88 x i8]* @.str3723, i64 0, i64 0))
-          to label %13 unwind label %31           ; <%45*> [#uses=1]
-
-; <label>:13                                      ; preds = %11
-  %14 = invoke %45* @_ZN4llvm11raw_ostreamlsEPKc(%45* %12, i8* getelementptr inbounds ([136 x i8]* @.str13724, i64 0, i64 0))
-          to label %15 unwind label %31           ; <%45*> [#uses=0]
-
-; <label>:15                                      ; preds = %13
-  %16 = getelementptr inbounds i8* null, i64 -24  ; <i8*> [#uses=1]
-  %17 = icmp eq i8* %16, bitcast ([4 x i64]* @_ZNSs4_Rep20_S_empty_rep_storageE to i8*) ; <i1> [#uses=1]
-  br i1 %17, label %19, label %18
-
-; <label>:18                                      ; preds = %15
-  unreachable
-
-; <label>:19                                      ; preds = %15
-  %20 = getelementptr inbounds i8* null, i64 -24  ; <i8*> [#uses=1]
-  %21 = icmp eq i8* %20, bitcast ([4 x i64]* @_ZNSs4_Rep20_S_empty_rep_storageE to i8*) ; <i1> [#uses=1]
-  br i1 %21, label %23, label %22
-
-; <label>:22                                      ; preds = %19
-  unreachable
-
-; <label>:23                                      ; preds = %19
-  invoke void @_ZNSsC1ERKSs(%8* undef, %8* undef)
-          to label %24 unwind label %29
-
-; <label>:24                                      ; preds = %23
-  invoke void @_ZNSsC1ERKSs(%8* undef, %8* undef)
-          to label %26 unwind label %25
-
-; <label>:25                                      ; preds = %24
-  unreachable
-
-; <label>:26                                      ; preds = %24
-  invoke void @f4(%82* undef, %53* undef)
-          to label %27 unwind label %32
-
-; <label>:27                                      ; preds = %26
-  unreachable
-
-; <label>:28                                      ; preds = %2
-  unreachable
-
-; <label>:29                                      ; preds = %23, %8
-  unreachable
-
-; <label>:30                                      ; preds = %10
-  unreachable
-
-; <label>:31                                      ; preds = %13, %11
-  unreachable
-
-; <label>:32                                      ; preds = %26
-  unreachable
-}
-
-declare void @_ZNSt3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS1_EEED1Ev(%22*) align 2
-
-declare void @_ZNSt3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS1_EEEC1ERKS8_(%22*, %22*) align 2
-
-declare %83* @_ZNSt8_Rb_treeIN4llvm3MVT15SimpleValueTypeESt4pairIKS2_St3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaIS3_IKSsS7_EEEESt10_Select1stISE_ES8_IS2_ESaISE_EE14_M_create_nodeERKSE_(%23*, %84*) align 2
-
-declare %22* @_ZNSt3mapIN12_GLOBAL__N_117OperandsSignatureES_ISsS_IN4llvm3MVT15SimpleValueTypeES_IS4_S_ISsNS0_15InstructionMemoESt4lessISsESaISt4pairIKSsS5_EEES6_IS4_ESaIS8_IKS4_SC_EEESD_SaIS8_ISE_SH_EEES7_SaIS8_IS9_SK_EEES6_IS1_ESaIS8_IKS1_SN_EEEixERSP_(%22*, %14*) align 2
-
-define internal %22* @1(%22*, i32*) align 2 {
-  unreachable
-}
-
-declare void @test(i64);
-
-define internal %22* @f3(%22*, i32*) align 2 {
-  %3 = alloca %85, align 8                        ; <%85*> [#uses=1]
-  %4 = alloca %84, align 8                        ; <%84*> [#uses=5]
-  invoke void @_ZNSt3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS1_EEEC1ERKS8_(%22* undef, %22* undef)
-          to label %5 unwind label %30
-
-; <label>:5                                       ; preds = %2
-  %6 = getelementptr inbounds %26* null, i64 1, i32 0 ; <i32*> [#uses=1]
-  %7 = load i32* %6, align 4                      ; <i32> [#uses=0]
-  br i1 false, label %8, label %11
-
-; <label>:8                                       ; preds = %5
-  %9 = invoke %83* @_ZNSt8_Rb_treeIN4llvm3MVT15SimpleValueTypeESt4pairIKS2_St3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaIS3_IKSsS7_EEEESt10_Select1stISE_ES8_IS2_ESaISE_EE14_M_create_nodeERKSE_(%23* undef, %84* %4)
-          to label %10 unwind label %31           ; <%83*> [#uses=0]
-
-; <label>:10                                      ; preds = %8
-  unreachable
-
-; <label>:11                                      ; preds = %5
-  %12 = getelementptr inbounds %85* %3, i64 0, i32 0 ; <%26**> [#uses=1]
-  %13 = load %26** %12, align 8                   ; <%26*> [#uses=0]
-  br i1 false, label %14, label %17
-
-; <label>:14                                      ; preds = %11
-  %15 = invoke %83* @_ZNSt8_Rb_treeIN4llvm3MVT15SimpleValueTypeESt4pairIKS2_St3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaIS3_IKSsS7_EEEESt10_Select1stISE_ES8_IS2_ESaISE_EE14_M_create_nodeERKSE_(%23* undef, %84* %4)
-          to label %16 unwind label %31           ; <%83*> [#uses=0]
-
-; <label>:16                                      ; preds = %14
-  unreachable
-
-; <label>:17                                      ; preds = %11
-  %18 = invoke %26* @_ZSt18_Rb_tree_incrementPSt18_Rb_tree_node_base(%26* undef)
-          to label %19 unwind label %31           ; <%26*> [#uses=0]
-
-; <label>:19                                      ; preds = %17
-  %20 = getelementptr inbounds %84* %4, i64 0, i32 0 ; <i32*> [#uses=1]
-  %21 = load i32* %20, align 8                    ; <i32> [#uses=0]
-  br i1 false, label %22, label %25
-
-; <label>:22                                      ; preds = %19
-  %23 = invoke %83* @_ZNSt8_Rb_treeIN4llvm3MVT15SimpleValueTypeESt4pairIKS2_St3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaIS3_IKSsS7_EEEESt10_Select1stISE_ES8_IS2_ESaISE_EE14_M_create_nodeERKSE_(%23* undef, %84* %4)
-          to label %24 unwind label %31           ; <%83*> [#uses=0]
-
-; <label>:24                                      ; preds = %22
-  unreachable
-
-; <label>:25                                      ; preds = %19
-  %26 = invoke %s2i64 @f_3(%23* undef, %84* %4)
-          to label %l1 unwind label %31
-
-; <label>:27                                      ; preds = %25
-  %eval = extractvalue %s2i64 %inv, 0                   ; <i64> [#uses=0]
-  call void @test(i64 %eval)
-; CHECK: = invoke %s2i64 @f_3
-; CHECK: %eval = extractvalue %s2i64 %inv, 0
-; CHECK-NEXT: call void @test(i64 %eval)
-  invoke void @_ZNSt3mapISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS1_EEED1Ev(%22* undef)
-          to label %29 unwind label %30
-
-l1:
-  %28 = extractvalue %s2i64 %26, 0
-  call void @test(i64 %28)
-;CHECK: call void @test(i64 5)
-;CHECK-NEXT: %inv = invoke %s2i64 @f2
-  %inv = invoke %s2i64 @f2(%23* undef, %84* %4)
-          to label %27 unwind label %31           ; <%s2i64> [#uses=1]
-
-
-; <label>:29                                      ; preds = %27
-  unreachable
-
-; <label>:30                                      ; preds = %27, %2
-  unreachable
-
-; <label>:31                                      ; preds = %25, %22, %17, %14, %8
-  unreachable
-}
-define internal %s2i64 @f2(%23*, %84*) align 2 {
-  br i1 undef, label %3, label %4
-
-; <label>:3                                       ; preds = %2
-  br label %4
-
-; <label>:4                                       ; preds = %3, %2
-  %5 = insertvalue %s2i64 undef, i64 4, 1        ; <%s2i64> [#uses=1]
-  %6 = ptrtoint %84* %1 to i64
-  %7 = insertvalue %s2i64 %5, i64 %6, 0        ; <%s2i64> [#uses=1]
-  ret %s2i64 %7
-}
-define internal %s2i64 @f_3(%23*, %84*) align 2 {
-  br i1 undef, label %3, label %4
-
-; <label>:3                                       ; preds = %2
-  br label %4
-
-; <label>:4                                       ; preds = %3, %2
-  %5 = insertvalue %s2i64 undef, i64 4, 1        ; <%s2i64> [#uses=1]
-  %6 = insertvalue %s2i64 %5, i64 5, 0        ; <%s2i64> [#uses=1]
-  ret %s2i64 %6
-}
-
-
-declare %22* @_ZNSt3mapISsS_IN4llvm3MVT15SimpleValueTypeES_IS2_S_ISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS4_EEES5_IS2_ESaIS7_IKS2_SB_EEESC_SaIS7_ISD_SG_EEES6_SaIS7_IS8_SJ_EEEixERS8_(%22*, %8*) align 2
-
-define internal void @f4(%82*, %53*) align 2 {
-  invoke void @_ZNSsC1ERKSs(%8* undef, %8* undef)
-          to label %3 unwind label %57
-
-; <label>:3                                       ; preds = %2
-  br i1 undef, label %6, label %4
-
-; <label>:4                                       ; preds = %3
-  br i1 undef, label %5, label %6
-
-; <label>:5                                       ; preds = %4
-  br label %6
-
-; <label>:6                                       ; preds = %5, %4, %3
-  br i1 undef, label %8, label %7
-
-; <label>:7                                       ; preds = %6
-  br i1 undef, label %56, label %9
-
-; <label>:8                                       ; preds = %6
-  unreachable
-
-; <label>:9                                       ; preds = %7
-  %10 = icmp eq %12* null, null                   ; <i1> [#uses=1]
-  br i1 %10, label %12, label %11
-
-; <label>:11                                      ; preds = %9
-  unreachable
-
-; <label>:12                                      ; preds = %9
-  %13 = getelementptr inbounds %46* null, i64 0, i32 6, i32 0, i32 0, i32 0 ; <%46***> [#uses=0]
-  %14 = icmp eq i32 0, 0                          ; <i1> [#uses=1]
-  br i1 %14, label %16, label %15
-
-; <label>:15                                      ; preds = %12
-  unreachable
-
-; <label>:16                                      ; preds = %12
-  br i1 undef, label %19, label %17
-
-; <label>:17                                      ; preds = %16
-  %18 = icmp eq %12* null, null                   ; <i1> [#uses=1]
-  br i1 %18, label %21, label %20
-
-; <label>:19                                      ; preds = %16
-  unreachable
-
-; <label>:20                                      ; preds = %17
-  unreachable
-
-; <label>:21                                      ; preds = %17
-  %22 = invoke i32 @_ZNK4llvm15TreePatternNode10getTypeNumEj(%46* null, i32 0)
-          to label %23 unwind label %58           ; <i32> [#uses=0]
-
-; <label>:23                                      ; preds = %21
-  %24 = invoke zeroext i8 @_ZNK4llvm6Record12isSubClassOfENS_9StringRefE(%37* undef, i64 undef, i64 undef)
-          to label %25 unwind label %58           ; <i8> [#uses=0]
-
-; <label>:25                                      ; preds = %23
-  br i1 undef, label %26, label %27
-
-; <label>:26                                      ; preds = %25
-  unreachable
-
-; <label>:27                                      ; preds = %25
-  %28 = invoke i8* @_Znwm(i64 24)
-          to label %29 unwind label %59           ; <i8*> [#uses=0]
-
-; <label>:29                                      ; preds = %27
-  %30 = icmp eq %12* null, null                   ; <i1> [#uses=1]
-  br i1 %30, label %32, label %31
-
-; <label>:31                                      ; preds = %29
-  unreachable
-
-; <label>:32                                      ; preds = %29
-  %33 = invoke i32 @_ZNKSs7compareEPKc(%8* undef, i8* getelementptr inbounds ([4 x i8]* @.str181730, i64 0, i64 0))
-          to label %34 unwind label %59           ; <i32> [#uses=1]
-
-; <label>:34                                      ; preds = %32
-  %35 = icmp eq i32 %33, 0                        ; <i1> [#uses=1]
-  br i1 %35, label %37, label %36
-
-; <label>:36                                      ; preds = %34
-  unreachable
-
-; <label>:37                                      ; preds = %34
-  invoke void @_ZNSsC1EPKcRKSaIcE(%8* undef, i8* getelementptr inbounds ([1 x i8]* @.str111723, i64 0, i64 0), %25* undef)
-          to label %38 unwind label %60
-
-; <label>:38                                      ; preds = %37
-  invoke void @_ZNSt6vectorISsSaISsEE9push_backERKSs(%15* undef, %8* undef)
-          to label %39 unwind label %61
-
-; <label>:39                                      ; preds = %38
-  br i1 undef, label %42, label %40
-
-; <label>:40                                      ; preds = %39
-  br i1 undef, label %41, label %42
-
-; <label>:41                                      ; preds = %40
-  unreachable
-
-; <label>:42                                      ; preds = %40, %39
-  invoke void @_ZNK4llvm14PatternToMatch17getPredicateCheckEv(%8* noalias sret undef, %75* null)
-          to label %43 unwind label %59
-
-; <label>:43                                      ; preds = %42
-  %44 = icmp eq %12* null, null                   ; <i1> [#uses=1]
-  br i1 %44, label %46, label %45
-
-; <label>:45                                      ; preds = %43
-  unreachable
-
-; <label>:46                                      ; preds = %43
-  invoke void @_ZNSsC1ERKSs(%8* undef, %8* undef)
-          to label %47 unwind label %62
-
-; <label>:47                                      ; preds = %46
-  %48 = invoke %22* @_ZNSt3mapIN12_GLOBAL__N_117OperandsSignatureES_ISsS_IN4llvm3MVT15SimpleValueTypeES_IS4_S_ISsNS0_15InstructionMemoESt4lessISsESaISt4pairIKSsS5_EEES6_IS4_ESaIS8_IKS4_SC_EEESD_SaIS8_ISE_SH_EEES7_SaIS8_IS9_SK_EEES6_IS1_ESaIS8_IKS1_SN_EEEixERSP_(%22* undef, %14* undef)
-          to label %49 unwind label %63           ; <%22*> [#uses=1]
-
-; <label>:49                                      ; preds = %47
-  %50 = invoke %22* @_ZNSt3mapISsS_IN4llvm3MVT15SimpleValueTypeES_IS2_S_ISsN12_GLOBAL__N_115InstructionMemoESt4lessISsESaISt4pairIKSsS4_EEES5_IS2_ESaIS7_IKS2_SB_EEESC_SaIS7_ISD_SG_EEES6_SaIS7_IS8_SJ_EEEixERS8_(%22* %48, %8* undef)
-          to label %51 unwind label %63           ; <%22*> [#uses=1]
-
-; <label>:51                                      ; preds = %49
-  %52 = invoke %22* @1(%22* %50, i32* undef)
-          to label %53 unwind label %63           ; <%22*> [#uses=1]
-
-; <label>:53                                      ; preds = %51
-  %54 = invoke %22* @f3(%22* %52, i32* undef)
-          to label %55 unwind label %63           ; <%22*> [#uses=0]
-
-; <label>:55                                      ; preds = %53
-  unreachable
-
-; <label>:56                                      ; preds = %7
-  ret void
-
-; <label>:57                                      ; preds = %2
-  unreachable
-
-; <label>:58                                      ; preds = %23, %21
-  unreachable
-
-; <label>:59                                      ; preds = %42, %32, %27
-  unreachable
-
-; <label>:60                                      ; preds = %37
-  unreachable
-
-; <label>:61                                      ; preds = %38
-  unreachable
-
-; <label>:62                                      ; preds = %46
-  unreachable
-
-; <label>:63                                      ; preds = %53, %51, %49, %47
-  unreachable
-}
-
-declare i32 @_ZN4llvm7TGLexer8LexTokenEv(%86*) align 2
-
-declare void @_ZN4llvm8TGParserD1Ev(%87*) align 2
-
-declare i32 @_ZN4llvm9SourceMgr18AddNewSourceBufferEPNS_12MemoryBufferENS_5SMLocE(%27*, %32*, %9* noalias) align 2
-
-define i32 @main(i32, i8**) {
-  br i1 undef, label %3, label %4
-
-; <label>:3                                       ; preds = %2
-  unreachable
-
-; <label>:4                                       ; preds = %2
-  %5 = icmp eq i32 0, 0                           ; <i1> [#uses=1]
-  br i1 %5, label %7, label %6
-
-; <label>:6                                       ; preds = %4
-  unreachable
-
-; <label>:7                                       ; preds = %4
-  br i1 undef, label %8, label %9
-
-; <label>:8                                       ; preds = %7
-  br i1 undef, label %10, label %9
-
-; <label>:9                                       ; preds = %8, %7
-  br i1 undef, label %14, label %18
-
-; <label>:10                                      ; preds = %8
-  br label %11
-
-; <label>:11                                      ; preds = %12, %10
-  br i1 undef, label %12, label %13
-
-; <label>:12                                      ; preds = %11
-  br label %11
-
-; <label>:13                                      ; preds = %11
-  unreachable
-
-; <label>:14                                      ; preds = %9
-  br i1 undef, label %15, label %17
-
-; <label>:15                                      ; preds = %14
-  br i1 undef, label %17, label %16
-
-; <label>:16                                      ; preds = %15
-  unreachable
-
-; <label>:17                                      ; preds = %15, %14
-  unreachable
-
-; <label>:18                                      ; preds = %9
-  %19 = invoke i32 @_ZN4llvm9SourceMgr18AddNewSourceBufferEPNS_12MemoryBufferENS_5SMLocE(%27* @_ZL6SrcMgr, %32* undef, %9* noalias undef)
-          to label %20 unwind label %34           ; <i32> [#uses=0]
-
-; <label>:20                                      ; preds = %18
-  %21 = invoke %15* @_ZNSt6vectorISsSaISsEEaSERKS1_(%15* getelementptr inbounds (%27* @_ZL6SrcMgr, i64 0, i32 1), %15* getelementptr inbounds (%13* @_ZN12_GLOBAL__N_111IncludeDirsE, i64 0, i32 1, i32 0))
-          to label %22 unwind label %34           ; <%15*> [#uses=0]
-
-; <label>:22                                      ; preds = %20
-  %23 = getelementptr inbounds %27* null, i64 0, i32 0, i32 0, i32 0, i32 1 ; <%31**> [#uses=1]
-  %24 = load %31** %23, align 8                   ; <%31*> [#uses=1]
-  %25 = ptrtoint %31* %24 to i64                  ; <i64> [#uses=1]
-  %26 = sub i64 %25, 0                            ; <i64> [#uses=1]
-  %27 = icmp ugt i64 %26, 15                      ; <i1> [#uses=1]
-  br i1 %27, label %29, label %28
-
-; <label>:28                                      ; preds = %22
-  unreachable
-
-; <label>:29                                      ; preds = %22
-  %30 = invoke i32 @_ZN4llvm7TGLexer8LexTokenEv(%86* undef)
-          to label %31 unwind label %35           ; <i32> [#uses=0]
-
-; <label>:31                                      ; preds = %29
-  invoke void @_ZN4llvm8TGParserD1Ev(%87* null)
-          to label %32 unwind label %34
-
-; <label>:32                                      ; preds = %31
-  %33 = icmp eq i8 0, 0                           ; <i1> [#uses=1]
-  br i1 %33, label %36, label %56
-
-; <label>:34                                      ; preds = %31, %20, %18
-  unreachable
-
-; <label>:35                                      ; preds = %29
-  unreachable
-
-; <label>:36                                      ; preds = %32
-  %37 = invoke i32 @_ZNKSs7compareEPKc(%8* getelementptr inbounds (%6* @_ZN12_GLOBAL__N_114OutputFilenameE, i64 0, i32 1, i32 0), i8* getelementptr inbounds ([2 x i8]* @.str533653, i64 0, i64 0))
-          to label %38 unwind label %57           ; <i32> [#uses=1]
-
-; <label>:38                                      ; preds = %36
-  %39 = icmp eq i32 %37, 0                        ; <i1> [#uses=1]
-  br i1 %39, label %43, label %40
-
-; <label>:40                                      ; preds = %38
-  %41 = invoke i8* @_Znwm(i64 56)
-          to label %42 unwind label %58           ; <i8*> [#uses=0]
-
-; <label>:42                                      ; preds = %40
-  unreachable
-
-; <label>:43                                      ; preds = %38
-  %44 = load i32* getelementptr inbounds (%0* @_ZN12_GLOBAL__N_16ActionE, i64 0, i32 1, i32 0), align 8 ; <i32> [#uses=1]
-  switch i32 %44, label %56 [
-    i32 0, label %45
-    i32 12, label %48
-    i32 13, label %51
-  ]
-
-; <label>:45                                      ; preds = %43
-  br i1 undef, label %46, label %47
-
-; <label>:46                                      ; preds = %45
-  unreachable
-
-; <label>:47                                      ; preds = %45
-  unreachable
-
-; <label>:48                                      ; preds = %43
-  invoke void @_ZN4llvm18CodeGenDAGPatternsC1ERNS_12RecordKeeperE(%53* undef, %21* @_ZN4llvm7RecordsE)
-          to label %50 unwind label %49
-
-; <label>:49                                      ; preds = %48
-  unreachable
-
-; <label>:50                                      ; preds = %48
-  unreachable
-
-; <label>:51                                      ; preds = %43
-  invoke void @_ZN4llvm18CodeGenDAGPatternsC1ERNS_12RecordKeeperE(%53* undef, %21* @_ZN4llvm7RecordsE)
-          to label %53 unwind label %52
-
-; <label>:52                                      ; preds = %51
-  unreachable
-
-; <label>:53                                      ; preds = %51
-  invoke void @0(%81* undef, %45* null)
-          to label %54 unwind label %60
-
-; <label>:54                                      ; preds = %53
-  invoke void @_ZN4llvm18CodeGenDAGPatternsD1Ev(%53* undef)
-          to label %55 unwind label %59
-
-; <label>:55                                      ; preds = %54
-  unreachable
-
-; <label>:56                                      ; preds = %43, %32
-  ret i32 1
-
-; <label>:57                                      ; preds = %36
-  unreachable
-
-; <label>:58                                      ; preds = %40
-  unreachable
-
-; <label>:59                                      ; preds = %54
-  unreachable
-
-; <label>:60                                      ; preds = %53
-  unreachable
-}
-
-declare i8* @pthread_getspecific(i32) nounwind
-
-declare i32 @pthread_setspecific(i32, i8*) nounwind
-
-declare i32 @pthread_key_delete(i32) nounwind
-
-declare i32 @pthread_key_create(i32*, void (i8*)*) nounwind
-
-declare i32 @pthread_mutex_trylock(%34*) nounwind
-
-declare i32 @pthread_mutex_unlock(%34*) nounwind
-
-declare i32 @pthread_mutex_lock(%34*) nounwind
-
-declare i32 @pthread_mutexattr_init(%2*) nounwind
-
-declare i32 @pthread_mutexattr_settype(%2*, i32) nounwind
-
-declare i32 @pthread_mutex_init(%34*, %2*) nounwind
-
-declare i32 @pthread_mutexattr_destroy(%2*) nounwind
-
-declare extern_weak i32 @pthread_once(i32*, void ()*)
-
-declare extern_weak i32 @pthread_create(i64*, %33*, i8* (i8*)*, i8*)
-
-declare extern_weak i32 @pthread_cancel(i64)




From sabre at nondot.org  Tue Oct 20 16:00:48 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 21:00:48 -0000
Subject: [llvm-commits] [llvm] r84675 -
	/llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll
Message-ID: <200910202100.n9KL0mVq019268@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 16:00:47 2009
New Revision: 84675

URL: http://llvm.org/viewvc/llvm-project?rev=84675&view=rev
Log:
add a test similar to that needed for PR4313, but that doesn't
fail without the patch.

Modified:
    llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll

Modified: llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll?rev=84675&r1=84674&r2=84675&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll (original)
+++ llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll Tue Oct 20 16:00:47 2009
@@ -72,3 +72,33 @@
 ; CHECK: define i32 @test3b
 ; CHECK-NOT: store
 ; CHECK: ret i32 0
+
+
+;;======================== test4
+
+define internal {i64,i64} @test4a() {
+  %a = insertvalue {i64,i64} undef, i64 4, 1
+  %b = insertvalue {i64,i64} %a, i64 5, 0
+  ret {i64,i64} %b
+}
+
+define i64 @test4b() {
+  %a = invoke {i64,i64} @test4a()
+          to label %A unwind label %B
+A:
+  %b = extractvalue {i64,i64} %a, 0
+  %c = call i64 @test4c(i64 %b)
+  ret i64 %c
+B:
+  ret i64 0
+}
+; CHECK: define i64 @test4b()
+; CHECK:   %c = call i64 @test4c(i64 5)
+; CHECK-NEXT:  ret i64 5
+
+
+define internal i64 @test4c(i64 %a) {
+  ret i64 %a
+}
+; CHECK: define internal i64 @test4c
+; CHECK: ret i64 undef




From sabre at nondot.org  Tue Oct 20 16:04:26 2009
From: sabre at nondot.org (Chris Lattner)
Date: Tue, 20 Oct 2009 21:04:26 -0000
Subject: [llvm-commits] [llvm] r84676 -
	/llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll
Message-ID: <200910202104.n9KL4Q0L019400@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 16:04:26 2009
New Revision: 84676

URL: http://llvm.org/viewvc/llvm-project?rev=84676&view=rev
Log:
add a real testcase for PR4313

Modified:
    llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll

Modified: llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll?rev=84676&r1=84675&r2=84676&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll (original)
+++ llvm/trunk/test/Transforms/SCCP/ipsccp-basic.ll Tue Oct 20 16:04:26 2009
@@ -102,3 +102,35 @@
 }
 ; CHECK: define internal i64 @test4c
 ; CHECK: ret i64 undef
+
+
+
+;;======================== test5
+
+; PR4313
+define internal {i64,i64} @test5a() {
+  %a = insertvalue {i64,i64} undef, i64 4, 1
+  %b = insertvalue {i64,i64} %a, i64 5, 0
+  ret {i64,i64} %b
+}
+
+define i64 @test5b() {
+  %a = invoke {i64,i64} @test5a()
+          to label %A unwind label %B
+A:
+  %c = call i64 @test5c({i64,i64} %a)
+  ret i64 %c
+B:
+  ret i64 0
+}
+
+; CHECK: define i64 @test5b()
+; CHECK:     A:
+; CHECK-NEXT:  %c = call i64 @test5c(%0 %a)
+; CHECK-NEXT:  ret i64 %c
+
+define internal i64 @test5c({i64,i64} %a) {
+  %b = extractvalue {i64,i64} %a, 0
+  ret i64 %b
+}
+




From clattner at apple.com  Tue Oct 20 16:05:23 2009
From: clattner at apple.com (Chris Lattner)
Date: Tue, 20 Oct 2009 14:05:23 -0700
Subject: [llvm-commits] [llvm] r84637 - in /llvm/trunk:
	lib/Transforms/Scalar/SCCP.cpp
	test/Transforms/SCCP/2009-06-03-sccp-structret-undef.ll
In-Reply-To: <200910201515.n9KFF994003926@zion.cs.uiuc.edu>
References: <200910201515.n9KFF994003926@zion.cs.uiuc.edu>
Message-ID: <BD236468-63D1-4F95-BD28-5312DA0085DB@apple.com>


On Oct 20, 2009, at 8:15 AM, Torok Edwin wrote:

> Author: edwin
> Date: Tue Oct 20 10:15:09 2009
> New Revision: 84637
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84637&view=rev
> Log:
> Fix PR4313: IPSCCP was not setting the lattice value for the invoke  
> instruction
> when the invoke had multiple return values: it set the lattice value  
> only on the
> extractvalue.
> This caused the invoke's lattice value to remain the default  
> (undefined), and
> later propagated to extractvalue's operand, which incorrectly  
> introduces
> undefined behavior.

Hi Edwin,

Your patch didn't affect the testcase you committed.  I removed the  
test and added a new testcase in r84676.  Thanks for tracking down  
this bug!

-Chris


From bob.wilson at apple.com  Tue Oct 20 16:05:42 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Tue, 20 Oct 2009 21:05:42 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84677 -
	/llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp
Message-ID: <200910202105.n9KL5hLY019459@zion.cs.uiuc.edu>

Author: bwilson
Date: Tue Oct 20 16:05:42 2009
New Revision: 84677

URL: http://llvm.org/viewvc/llvm-project?rev=84677&view=rev
Log:
Add a function to build int32 constants.  Remove some redundant "Context"s.

Modified:
    llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp

Modified: llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp?rev=84677&r1=84676&r2=84677&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp (original)
+++ llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp Tue Oct 20 16:05:42 2009
@@ -47,6 +47,10 @@
   (int fcode, neon_itype *itype, enum neon_builtins *neon_code);
 }
 
+static ConstantInt *getInt32Const(int c) {
+  return ConstantInt::get(Type::getInt32Ty(Context), c);
+}
+
 /// UnexpectedError - Report errors about unexpected uses of builtins.  The
 /// msg argument should begin with a "%H" so that the location of the
 /// expression is printed in the error message.
@@ -131,8 +135,6 @@
 /// element of a vector.
 static Value *BuildDup(const Type *ResultType, Value *Val,
                        LLVMBuilder &Builder) {
-  LLVMContext &Context = getGlobalContext();
-
   // GCC may promote the scalar argument; cast it back.
   const VectorType *VTy = dyn_cast<const VectorType>(ResultType);
   assert(VTy && "expected a vector type");
@@ -145,16 +147,14 @@
 
   // Insert the value into lane 0 of an undef vector.
   Value *Undef = UndefValue::get(ResultType);
-  Value *Result =
-    Builder.CreateInsertElement(Undef, Val,
-                                ConstantInt::get(Type::getInt32Ty(Context), 0));
+  Value *Result = Builder.CreateInsertElement(Undef, Val, getInt32Const(0));
 
   // Use a shuffle to move the value into the other lanes.
   unsigned NUnits = VTy->getNumElements();
   if (NUnits > 1) {
     std::vector<Constant*> Idxs;
     for (unsigned i = 0; i != NUnits; ++i)
-      Idxs.push_back(ConstantInt::get(Type::getInt32Ty(Context), 0));
+      Idxs.push_back(getInt32Const(0));
     Result = Builder.CreateShuffleVector(Result, Undef,
                                          ConstantVector::get(Idxs));
   }
@@ -167,9 +167,8 @@
                            LLVMBuilder &Builder) {
   // Translate this to a vector shuffle.
   std::vector<Constant*> Idxs;
-  LLVMContext &Context = getGlobalContext();
   for (unsigned i = 0; i != NUnits; ++i)
-    Idxs.push_back(ConstantInt::get(Type::getInt32Ty(Context), LaneVal));
+    Idxs.push_back(getInt32Const(LaneVal));
   return Builder.CreateShuffleVector(Vec, UndefValue::get(Vec->getType()),
                                      ConstantVector::get(Idxs));
 }
@@ -217,7 +216,6 @@
     return false;
 
   // Right shifts are represented in NEON intrinsics by a negative shift count.
-  LLVMContext &Context = getGlobalContext();
   Cnt = ConstantInt::get(IntegerType::get(Context, ElemBits),
                          NegateRightShift ? -CntVal : CntVal);
   Op = BuildConstantSplatVector(GET_MODE_NUNITS(Mode), Cnt);
@@ -259,8 +257,6 @@
   if (FnCode < ARM_BUILTIN_NEON_BASE)
     return false;
 
-  LLVMContext &Context = getGlobalContext();
-
   neon_builtins neon_code;
   enum insn_code icode = locate_neon_builtin_icode (FnCode, 0, &neon_code);
 
@@ -1686,7 +1682,7 @@
     unsigned NUnits = GET_MODE_NUNITS(insn_data[icode].operand[0].mode);
     std::vector<Constant*> Idxs;
     for (unsigned i = 0; i != NUnits; ++i)
-      Idxs.push_back(ConstantInt::get(Type::getInt32Ty(Context), i));
+      Idxs.push_back(getInt32Const(i));
     Result = Builder.CreateShuffleVector(Ops[0], Ops[1],
                                          ConstantVector::get(Idxs));
     break;
@@ -1698,7 +1694,7 @@
     std::vector<Constant*> Idxs;
     unsigned Idx = (neon_code == NEON_BUILTIN_vget_low ? 0 : NUnits);
     for (unsigned i = 0; i != NUnits; ++i)
-      Idxs.push_back(ConstantInt::get(Type::getInt32Ty(Context), Idx++));
+      Idxs.push_back(getInt32Const(Idx++));
     Result = Builder.CreateShuffleVector(Ops[0],
                                          UndefValue::get(Ops[0]->getType()),
                                          ConstantVector::get(Idxs));
@@ -1762,7 +1758,7 @@
     // Translate to a vector shuffle.
     std::vector<Constant*> Idxs;
     for (unsigned i = 0; i != NUnits; ++i)
-      Idxs.push_back(ConstantInt::get(Type::getInt32Ty(Context), i + ImmVal));
+      Idxs.push_back(getInt32Const(i + ImmVal));
     Result = Builder.CreateShuffleVector(Ops[0], Ops[1],
                                          ConstantVector::get(Idxs));
     break;
@@ -1788,7 +1784,7 @@
     unsigned NUnits = VTy->getNumElements();
     for (unsigned c = ChunkElts; c <= NUnits; c += ChunkElts) {
       for (unsigned i = 0; i != ChunkElts; ++i) {
-        Idxs.push_back(ConstantInt::get(Type::getInt32Ty(Context), c - i - 1));
+        Idxs.push_back(getInt32Const(c - i - 1));
       }
     }
     Result = Builder.CreateShuffleVector(Ops[0], UndefValue::get(ResultType),
@@ -1934,12 +1930,11 @@
   case NEON_BUILTIN_vtrn: {
     // Translate this to a pair of vector shuffles.
     unsigned NUnits = GET_MODE_NUNITS(insn_data[icode].operand[1].mode);
-    const Type *Int32Ty = Type::getInt32Ty(Context);
     for (unsigned Elt = 0; Elt != 2; ++Elt) {
       std::vector<Constant*> Idxs;
       for (unsigned i = 0; i < NUnits; i += 2) {
-        Idxs.push_back(ConstantInt::get(Int32Ty, i + Elt));
-        Idxs.push_back(ConstantInt::get(Int32Ty, i + NUnits + Elt));
+        Idxs.push_back(getInt32Const(i + Elt));
+        Idxs.push_back(getInt32Const(i + NUnits + Elt));
       }
       Result = Builder.CreateShuffleVector(Ops[0], Ops[1],
                                            ConstantVector::get(Idxs));
@@ -1953,13 +1948,12 @@
   case NEON_BUILTIN_vzip: {
     // Translate this to a pair of vector shuffles.
     unsigned NUnits = GET_MODE_NUNITS(insn_data[icode].operand[1].mode);
-    const Type *Int32Ty = Type::getInt32Ty(Context);
     unsigned Idx = 0;
     for (unsigned Elt = 0; Elt != 2; ++Elt) {
       std::vector<Constant*> Idxs;
       for (unsigned i = 0; i != NUnits; i += 2) {
-        Idxs.push_back(ConstantInt::get(Int32Ty, Idx));
-        Idxs.push_back(ConstantInt::get(Int32Ty, Idx + NUnits));
+        Idxs.push_back(getInt32Const(Idx));
+        Idxs.push_back(getInt32Const(Idx + NUnits));
         Idx += 1;
       }
       Result = Builder.CreateShuffleVector(Ops[0], Ops[1],
@@ -1974,11 +1968,10 @@
   case NEON_BUILTIN_vuzp: {
     // Translate this to a pair of vector shuffles.
     unsigned NUnits = GET_MODE_NUNITS(insn_data[icode].operand[1].mode);
-    const Type *Int32Ty = Type::getInt32Ty(Context);
     for (unsigned Elt = 0; Elt != 2; ++Elt) {
       std::vector<Constant*> Idxs;
       for (unsigned i = 0; i != NUnits; ++i)
-        Idxs.push_back(ConstantInt::get(Int32Ty, 2 * i + Elt));
+        Idxs.push_back(getInt32Const(2 * i + Elt));
       Result = Builder.CreateShuffleVector(Ops[0], Ops[1],
                                            ConstantVector::get(Idxs));
       Value *Addr = Builder.CreateConstInBoundsGEP2_32(DestLoc->Ptr, 0, Elt);
@@ -2109,7 +2102,7 @@
     for (unsigned n = 0; n != NumVecs; ++n) {
       Args.push_back(UndefValue::get(VTy));
     }
-    Args.push_back(ConstantInt::get(Type::getInt32Ty(Context), 0));
+    Args.push_back(getInt32Const(0));
     Result = Builder.CreateCall(intFn, Args.begin(), Args.end());
 
     // Now splat the values in lane 0 to the rest of the elements.




From bob.wilson at apple.com  Tue Oct 20 16:08:21 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Tue, 20 Oct 2009 21:08:21 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84678 -
	/llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp
Message-ID: <200910202108.n9KL8LRN019588@zion.cs.uiuc.edu>

Author: bwilson
Date: Tue Oct 20 16:08:21 2009
New Revision: 84678

URL: http://llvm.org/viewvc/llvm-project?rev=84678&view=rev
Log:
Translate NEON vget_high and vget_low builtins to extract_element with a
v2f64 type instead of using vector shuffles.  The code generator does not
produce good code for a shuffle where the inputs have more elements than
the result.

Modified:
    llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp

Modified: llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp?rev=84678&r1=84677&r2=84678&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp (original)
+++ llvm-gcc-4.2/trunk/gcc/config/arm/llvm-arm.cpp Tue Oct 20 16:08:21 2009
@@ -1690,14 +1690,11 @@
 
   case NEON_BUILTIN_vget_high:
   case NEON_BUILTIN_vget_low: {
-    unsigned NUnits = GET_MODE_NUNITS(insn_data[icode].operand[0].mode);
-    std::vector<Constant*> Idxs;
-    unsigned Idx = (neon_code == NEON_BUILTIN_vget_low ? 0 : NUnits);
-    for (unsigned i = 0; i != NUnits; ++i)
-      Idxs.push_back(getInt32Const(Idx++));
-    Result = Builder.CreateShuffleVector(Ops[0],
-                                         UndefValue::get(Ops[0]->getType()),
-                                         ConstantVector::get(Idxs));
+    const Type *v2f64Ty = VectorType::get(Type::getDoubleTy(Context), 2);
+    unsigned Idx = (neon_code == NEON_BUILTIN_vget_low ? 0 : 1);
+    Result = Builder.CreateBitCast(Ops[0], v2f64Ty);
+    Result = Builder.CreateExtractElement(Result, getInt32Const(Idx));
+    Result = Builder.CreateBitCast(Result, ResultType);
     break;
   }
 




From lhames at gmail.com  Tue Oct 20 16:25:13 2009
From: lhames at gmail.com (Lang Hames)
Date: Tue, 20 Oct 2009 21:25:13 -0000
Subject: [llvm-commits] [llvm] r84681 -
	/llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp
Message-ID: <200910202125.n9KLPEfI020298@zion.cs.uiuc.edu>

Author: lhames
Date: Tue Oct 20 16:25:13 2009
New Revision: 84681

URL: http://llvm.org/viewvc/llvm-project?rev=84681&view=rev
Log:
Added some debugging output to pre-alloc splitting.

Modified:
    llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp

Modified: llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp?rev=84681&r1=84680&r2=84681&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp (original)
+++ llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp Tue Oct 20 16:25:13 2009
@@ -93,6 +93,8 @@
 
     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
       AU.setPreservesCFG();
+      AU.addRequired<LiveIndexes>();
+      AU.addPreserved<LiveIndexes>();
       AU.addRequired<LiveIntervals>();
       AU.addPreserved<LiveIntervals>();
       AU.addRequired<LiveStacks>();
@@ -399,7 +401,7 @@
 
   SmallPtrSet<MachineBasicBlock*, 4> Processed;
   LiveIndex EndIdx = LIs->getMBBEndIdx(MBB);
-  LiveRange SLR(SpillIndex, LIs->getNextSlot(EndIdx), CurrSValNo);
+  LiveRange SLR(SpillIndex, EndIdx.getNextSlot(), CurrSValNo);
   CurrSLI->addRange(SLR);
   Processed.insert(MBB);
 
@@ -428,7 +430,7 @@
         CurrSLI->addRange(SLR);
       } else if (LR->end > EndIdx) {
         // Live range extends beyond end of mbb, process successors.
-        LiveRange SLR(Idx, LIs->getNextIndex(EndIdx), CurrSValNo);
+        LiveRange SLR(Idx, EndIdx.getNextIndex(), CurrSValNo);
         CurrSLI->addRange(SLR);
         for (MachineBasicBlock::succ_iterator SI = MBB->succ_begin(),
                SE = MBB->succ_end(); SI != SE; ++SI)
@@ -492,11 +494,11 @@
     
     // Once we've found it, extend its VNInfo to our instruction.
     LiveIndex DefIndex = LIs->getInstructionIndex(Walker);
-    DefIndex = LIs->getDefIndex(DefIndex);
+    DefIndex = DefIndex.getDefIndex();
     LiveIndex EndIndex = LIs->getMBBEndIdx(MBB);
     
     RetVNI = NewVNs[Walker];
-    LI->addRange(LiveRange(DefIndex, LIs->getNextSlot(EndIndex), RetVNI));
+    LI->addRange(LiveRange(DefIndex, EndIndex.getNextSlot(), RetVNI));
   } else if (!ContainsDefs && ContainsUses) {
     SmallPtrSet<MachineInstr*, 2>& BlockUses = Uses[MBB];
     
@@ -529,11 +531,11 @@
     }
 
     LiveIndex UseIndex = LIs->getInstructionIndex(Walker);
-    UseIndex = LIs->getUseIndex(UseIndex);
+    UseIndex = UseIndex.getUseIndex();
     LiveIndex EndIndex;
     if (IsIntraBlock) {
       EndIndex = LIs->getInstructionIndex(UseI);
-      EndIndex = LIs->getUseIndex(EndIndex);
+      EndIndex = EndIndex.getUseIndex();
     } else
       EndIndex = LIs->getMBBEndIdx(MBB);
 
@@ -542,7 +544,7 @@
     RetVNI = PerformPHIConstruction(Walker, MBB, LI, Visited, Defs, Uses,
                                     NewVNs, LiveOut, Phis, false, true);
     
-    LI->addRange(LiveRange(UseIndex, LIs->getNextSlot(EndIndex), RetVNI));
+    LI->addRange(LiveRange(UseIndex, EndIndex.getNextSlot(), RetVNI));
     
     // FIXME: Need to set kills properly for inter-block stuff.
     if (RetVNI->isKill(UseIndex)) RetVNI->removeKill(UseIndex);
@@ -589,12 +591,11 @@
     }
 
     LiveIndex StartIndex = LIs->getInstructionIndex(Walker);
-    StartIndex = foundDef ? LIs->getDefIndex(StartIndex) :
-                            LIs->getUseIndex(StartIndex);
+    StartIndex = foundDef ? StartIndex.getDefIndex() : StartIndex.getUseIndex();
     LiveIndex EndIndex;
     if (IsIntraBlock) {
       EndIndex = LIs->getInstructionIndex(UseI);
-      EndIndex = LIs->getUseIndex(EndIndex);
+      EndIndex = EndIndex.getUseIndex();
     } else
       EndIndex = LIs->getMBBEndIdx(MBB);
 
@@ -604,7 +605,7 @@
       RetVNI = PerformPHIConstruction(Walker, MBB, LI, Visited, Defs, Uses,
                                       NewVNs, LiveOut, Phis, false, true);
 
-    LI->addRange(LiveRange(StartIndex, LIs->getNextSlot(EndIndex), RetVNI));
+    LI->addRange(LiveRange(StartIndex, EndIndex.getNextSlot(), RetVNI));
     
     if (foundUse && RetVNI->isKill(StartIndex))
       RetVNI->removeKill(StartIndex);
@@ -694,10 +695,10 @@
   LiveIndex EndIndex;
   if (IsIntraBlock) {
     EndIndex = LIs->getInstructionIndex(UseI);
-    EndIndex = LIs->getUseIndex(EndIndex);
+    EndIndex = EndIndex.getUseIndex();
   } else
     EndIndex = LIs->getMBBEndIdx(MBB);
-  LI->addRange(LiveRange(StartIndex, LIs->getNextSlot(EndIndex), RetVNI));
+  LI->addRange(LiveRange(StartIndex, EndIndex.getNextSlot(), RetVNI));
   if (IsIntraBlock)
     RetVNI->addKill(EndIndex);
 
@@ -734,7 +735,7 @@
     Defs[(*DI).getParent()].insert(&*DI);
     
     LiveIndex DefIdx = LIs->getInstructionIndex(&*DI);
-    DefIdx = LIs->getDefIndex(DefIdx);
+    DefIdx = DefIdx.getDefIndex();
     
     assert(DI->getOpcode() != TargetInstrInfo::PHI &&
            "Following NewVN isPHIDef flag incorrect. Fix me!");
@@ -770,12 +771,12 @@
   for (MachineRegisterInfo::def_iterator DI = MRI->def_begin(LI->reg),
        DE = MRI->def_end(); DI != DE; ++DI) {
     LiveIndex DefIdx = LIs->getInstructionIndex(&*DI);
-    DefIdx = LIs->getDefIndex(DefIdx);
+    DefIdx = DefIdx.getDefIndex();
     
     if (LI->liveAt(DefIdx)) continue;
     
     VNInfo* DeadVN = NewVNs[&*DI];
-    LI->addRange(LiveRange(DefIdx, LIs->getNextSlot(DefIdx), DeadVN));
+    LI->addRange(LiveRange(DefIdx, DefIdx.getNextSlot(), DeadVN));
     DeadVN->addKill(DefIdx);
   }
 
@@ -785,7 +786,7 @@
     VNInfo* VNI = *VI;
     for (unsigned i = 0, e = VNI->kills.size(); i != e; ++i) {
       LiveIndex KillIdx = VNI->kills[i];
-      if (KillIdx.isPHIIndex())
+      if (KillIdx.isPHI())
         continue;
       MachineInstr *KillMI = LIs->getInstructionFromIndex(KillIdx);
       if (KillMI) {
@@ -826,14 +827,14 @@
     // Locate two-address redefinitions
     for (VNInfo::KillSet::iterator KI = OldVN->kills.begin(),
          KE = OldVN->kills.end(); KI != KE; ++KI) {
-      assert(!KI->isPHIIndex() &&
+      assert(!KI->isPHI() &&
              "VN previously reported having no PHI kills.");
       MachineInstr* MI = LIs->getInstructionFromIndex(*KI);
       unsigned DefIdx = MI->findRegisterDefOperandIdx(CurrLI->reg);
       if (DefIdx == ~0U) continue;
       if (MI->isRegTiedToUseOperand(DefIdx)) {
         VNInfo* NextVN =
-          CurrLI->findDefinedVNInfoForRegInt(LIs->getDefIndex(*KI));
+          CurrLI->findDefinedVNInfoForRegInt(KI->getDefIndex());
         if (NextVN == OldVN) continue;
         Stack.push_back(NextVN);
       }
@@ -867,8 +868,8 @@
     MachineOperand& MO = I.getOperand();
     LiveIndex InstrIdx = LIs->getInstructionIndex(&*I);
     
-    if ((MO.isUse() && NewLI.liveAt(LIs->getUseIndex(InstrIdx))) ||
-        (MO.isDef() && NewLI.liveAt(LIs->getDefIndex(InstrIdx))))
+    if ((MO.isUse() && NewLI.liveAt(InstrIdx.getUseIndex())) ||
+        (MO.isDef() && NewLI.liveAt(InstrIdx.getDefIndex())))
       OpsToChange.push_back(std::make_pair(&*I, I.getOperandNo()));
   }
   
@@ -912,7 +913,7 @@
   
   ReconstructLiveInterval(CurrLI);
   LiveIndex RematIdx = LIs->getInstructionIndex(prior(RestorePt));
-  RematIdx = LIs->getDefIndex(RematIdx);
+  RematIdx = RematIdx.getDefIndex();
   RenumberValno(CurrLI->findDefinedVNInfoForRegInt(RematIdx));
   
   ++NumSplits;
@@ -1052,11 +1053,14 @@
 /// so it would not cross the barrier that's being processed. Shrink wrap
 /// (minimize) the live interval to the last uses.
 bool PreAllocSplitting::SplitRegLiveInterval(LiveInterval *LI) {
+  DEBUG(errs() << "Pre-alloc splitting " << LI->reg << " for " << *Barrier
+               << "  result: ");
+
   CurrLI = LI;
 
   // Find live range where current interval cross the barrier.
   LiveInterval::iterator LR =
-    CurrLI->FindLiveRangeContaining(LIs->getUseIndex(BarrierIdx));
+    CurrLI->FindLiveRangeContaining(BarrierIdx.getUseIndex());
   VNInfo *ValNo = LR->valno;
 
   assert(!ValNo->isUnused() && "Val# is defined by a dead def?");
@@ -1065,8 +1069,10 @@
     ? LIs->getInstructionFromIndex(ValNo->def) : NULL;
 
   // If this would create a new join point, do not split.
-  if (DefMI && createsNewJoin(LR, DefMI->getParent(), Barrier->getParent()))
+  if (DefMI && createsNewJoin(LR, DefMI->getParent(), Barrier->getParent())) {
+    DEBUG(errs() << "FAILED (would create a new join point).\n");
     return false;
+  }
 
   // Find all references in the barrier mbb.
   SmallPtrSet<MachineInstr*, 4> RefsInMBB;
@@ -1081,13 +1087,17 @@
   LiveIndex RestoreIndex;
   MachineBasicBlock::iterator RestorePt =
     findRestorePoint(BarrierMBB, Barrier, LR->end, RefsInMBB, RestoreIndex);
-  if (RestorePt == BarrierMBB->end())
+  if (RestorePt == BarrierMBB->end()) {
+    DEBUG(errs() << "FAILED (could not find a suitable restore point).\n");
     return false;
+  }
 
   if (DefMI && LIs->isReMaterializable(*LI, ValNo, DefMI))
     if (Rematerialize(LI->reg, ValNo, DefMI, RestorePt,
-                      RestoreIndex, RefsInMBB))
-    return true;
+                      RestoreIndex, RefsInMBB)) {
+      DEBUG(errs() << "success (remat).\n");
+      return true;
+    }
 
   // Add a spill either before the barrier or after the definition.
   MachineBasicBlock *DefMBB = DefMI ? DefMI->getParent() : NULL;
@@ -1103,8 +1113,10 @@
     } else {
       MachineBasicBlock::iterator SpillPt = 
         findSpillPoint(BarrierMBB, Barrier, NULL, RefsInMBB, SpillIndex);
-      if (SpillPt == BarrierMBB->begin())
+      if (SpillPt == BarrierMBB->begin()) {
+        DEBUG(errs() << "FAILED (could not find a suitable spill point).\n");
         return false; // No gap to insert spill.
+      }
       // Add spill.
     
       SS = CreateSpillStackSlot(CurrLI->reg, RC);
@@ -1116,8 +1128,10 @@
                                  RestoreIndex, SpillIndex, SS)) {
     // If it's already split, just restore the value. There is no need to spill
     // the def again.
-    if (!DefMI)
+    if (!DefMI) {
+      DEBUG(errs() << "FAILED (def is dead).\n");
       return false; // Def is dead. Do nothing.
+    }
     
     if ((SpillMI = FoldSpill(LI->reg, RC, DefMI, Barrier,
                              BarrierMBB, SS, RefsInMBB))) {
@@ -1129,12 +1143,16 @@
         // Add spill after the def and the last use before the barrier.
         SpillPt = findSpillPoint(BarrierMBB, Barrier, DefMI,
                                  RefsInMBB, SpillIndex);
-        if (SpillPt == DefMBB->begin())
+        if (SpillPt == DefMBB->begin()) {
+          DEBUG(errs() << "FAILED (could not find a suitable spill point).\n");
           return false; // No gap to insert spill.
+        }
       } else {
         SpillPt = findNextEmptySlot(DefMBB, DefMI, SpillIndex);
-        if (SpillPt == DefMBB->end())
+        if (SpillPt == DefMBB->end()) {
+          DEBUG(errs() << "FAILED (could not find a suitable spill point).\n");
           return false; // No gap to insert spill.
+        }
       }
       // Add spill. 
       SS = CreateSpillStackSlot(CurrLI->reg, RC);
@@ -1162,18 +1180,19 @@
   }
 
   // Update spill stack slot live interval.
-  UpdateSpillSlotInterval(ValNo, LIs->getNextSlot(LIs->getUseIndex(SpillIndex)),
-                          LIs->getDefIndex(RestoreIndex));
+  UpdateSpillSlotInterval(ValNo, SpillIndex.getUseIndex().getNextSlot(),
+                          RestoreIndex.getDefIndex());
 
   ReconstructLiveInterval(CurrLI);
 
   if (!FoldedRestore) {
     LiveIndex RestoreIdx = LIs->getInstructionIndex(prior(RestorePt));
-    RestoreIdx = LIs->getDefIndex(RestoreIdx);
+    RestoreIdx = RestoreIdx.getDefIndex();
     RenumberValno(CurrLI->findDefinedVNInfoForRegInt(RestoreIdx));
   }
   
   ++NumSplits;
+  DEBUG(errs() << "success.\n");
   return true;
 }
 
@@ -1255,7 +1274,7 @@
     for (MachineRegisterInfo::use_iterator UI = MRI->use_begin((*LI)->reg),
          UE = MRI->use_end(); UI != UE; ++UI) {
       LiveIndex index = LIs->getInstructionIndex(&*UI);
-      index = LIs->getUseIndex(index);
+      index = index.getUseIndex();
       
       const LiveRange* LR = (*LI)->getLiveRangeContaining(index);
       VNUseCount[LR->valno].insert(&*UI);




From lhames at gmail.com  Tue Oct 20 16:28:22 2009
From: lhames at gmail.com (Lang Hames)
Date: Tue, 20 Oct 2009 21:28:22 -0000
Subject: [llvm-commits] [llvm] r84682 -
	/llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp
Message-ID: <200910202128.n9KLSMxm020415@zion.cs.uiuc.edu>

Author: lhames
Date: Tue Oct 20 16:28:22 2009
New Revision: 84682

URL: http://llvm.org/viewvc/llvm-project?rev=84682&view=rev
Log:
Oops. Backing out 84681 - needs to wait for the indexing patch.

Modified:
    llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp

Modified: llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp?rev=84682&r1=84681&r2=84682&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp (original)
+++ llvm/trunk/lib/CodeGen/PreAllocSplitting.cpp Tue Oct 20 16:28:22 2009
@@ -93,8 +93,6 @@
 
     virtual void getAnalysisUsage(AnalysisUsage &AU) const {
       AU.setPreservesCFG();
-      AU.addRequired<LiveIndexes>();
-      AU.addPreserved<LiveIndexes>();
       AU.addRequired<LiveIntervals>();
       AU.addPreserved<LiveIntervals>();
       AU.addRequired<LiveStacks>();
@@ -401,7 +399,7 @@
 
   SmallPtrSet<MachineBasicBlock*, 4> Processed;
   LiveIndex EndIdx = LIs->getMBBEndIdx(MBB);
-  LiveRange SLR(SpillIndex, EndIdx.getNextSlot(), CurrSValNo);
+  LiveRange SLR(SpillIndex, LIs->getNextSlot(EndIdx), CurrSValNo);
   CurrSLI->addRange(SLR);
   Processed.insert(MBB);
 
@@ -430,7 +428,7 @@
         CurrSLI->addRange(SLR);
       } else if (LR->end > EndIdx) {
         // Live range extends beyond end of mbb, process successors.
-        LiveRange SLR(Idx, EndIdx.getNextIndex(), CurrSValNo);
+        LiveRange SLR(Idx, LIs->getNextIndex(EndIdx), CurrSValNo);
         CurrSLI->addRange(SLR);
         for (MachineBasicBlock::succ_iterator SI = MBB->succ_begin(),
                SE = MBB->succ_end(); SI != SE; ++SI)
@@ -494,11 +492,11 @@
     
     // Once we've found it, extend its VNInfo to our instruction.
     LiveIndex DefIndex = LIs->getInstructionIndex(Walker);
-    DefIndex = DefIndex.getDefIndex();
+    DefIndex = LIs->getDefIndex(DefIndex);
     LiveIndex EndIndex = LIs->getMBBEndIdx(MBB);
     
     RetVNI = NewVNs[Walker];
-    LI->addRange(LiveRange(DefIndex, EndIndex.getNextSlot(), RetVNI));
+    LI->addRange(LiveRange(DefIndex, LIs->getNextSlot(EndIndex), RetVNI));
   } else if (!ContainsDefs && ContainsUses) {
     SmallPtrSet<MachineInstr*, 2>& BlockUses = Uses[MBB];
     
@@ -531,11 +529,11 @@
     }
 
     LiveIndex UseIndex = LIs->getInstructionIndex(Walker);
-    UseIndex = UseIndex.getUseIndex();
+    UseIndex = LIs->getUseIndex(UseIndex);
     LiveIndex EndIndex;
     if (IsIntraBlock) {
       EndIndex = LIs->getInstructionIndex(UseI);
-      EndIndex = EndIndex.getUseIndex();
+      EndIndex = LIs->getUseIndex(EndIndex);
     } else
       EndIndex = LIs->getMBBEndIdx(MBB);
 
@@ -544,7 +542,7 @@
     RetVNI = PerformPHIConstruction(Walker, MBB, LI, Visited, Defs, Uses,
                                     NewVNs, LiveOut, Phis, false, true);
     
-    LI->addRange(LiveRange(UseIndex, EndIndex.getNextSlot(), RetVNI));
+    LI->addRange(LiveRange(UseIndex, LIs->getNextSlot(EndIndex), RetVNI));
     
     // FIXME: Need to set kills properly for inter-block stuff.
     if (RetVNI->isKill(UseIndex)) RetVNI->removeKill(UseIndex);
@@ -591,11 +589,12 @@
     }
 
     LiveIndex StartIndex = LIs->getInstructionIndex(Walker);
-    StartIndex = foundDef ? StartIndex.getDefIndex() : StartIndex.getUseIndex();
+    StartIndex = foundDef ? LIs->getDefIndex(StartIndex) :
+                            LIs->getUseIndex(StartIndex);
     LiveIndex EndIndex;
     if (IsIntraBlock) {
       EndIndex = LIs->getInstructionIndex(UseI);
-      EndIndex = EndIndex.getUseIndex();
+      EndIndex = LIs->getUseIndex(EndIndex);
     } else
       EndIndex = LIs->getMBBEndIdx(MBB);
 
@@ -605,7 +604,7 @@
       RetVNI = PerformPHIConstruction(Walker, MBB, LI, Visited, Defs, Uses,
                                       NewVNs, LiveOut, Phis, false, true);
 
-    LI->addRange(LiveRange(StartIndex, EndIndex.getNextSlot(), RetVNI));
+    LI->addRange(LiveRange(StartIndex, LIs->getNextSlot(EndIndex), RetVNI));
     
     if (foundUse && RetVNI->isKill(StartIndex))
       RetVNI->removeKill(StartIndex);
@@ -695,10 +694,10 @@
   LiveIndex EndIndex;
   if (IsIntraBlock) {
     EndIndex = LIs->getInstructionIndex(UseI);
-    EndIndex = EndIndex.getUseIndex();
+    EndIndex = LIs->getUseIndex(EndIndex);
   } else
     EndIndex = LIs->getMBBEndIdx(MBB);
-  LI->addRange(LiveRange(StartIndex, EndIndex.getNextSlot(), RetVNI));
+  LI->addRange(LiveRange(StartIndex, LIs->getNextSlot(EndIndex), RetVNI));
   if (IsIntraBlock)
     RetVNI->addKill(EndIndex);
 
@@ -735,7 +734,7 @@
     Defs[(*DI).getParent()].insert(&*DI);
     
     LiveIndex DefIdx = LIs->getInstructionIndex(&*DI);
-    DefIdx = DefIdx.getDefIndex();
+    DefIdx = LIs->getDefIndex(DefIdx);
     
     assert(DI->getOpcode() != TargetInstrInfo::PHI &&
            "Following NewVN isPHIDef flag incorrect. Fix me!");
@@ -771,12 +770,12 @@
   for (MachineRegisterInfo::def_iterator DI = MRI->def_begin(LI->reg),
        DE = MRI->def_end(); DI != DE; ++DI) {
     LiveIndex DefIdx = LIs->getInstructionIndex(&*DI);
-    DefIdx = DefIdx.getDefIndex();
+    DefIdx = LIs->getDefIndex(DefIdx);
     
     if (LI->liveAt(DefIdx)) continue;
     
     VNInfo* DeadVN = NewVNs[&*DI];
-    LI->addRange(LiveRange(DefIdx, DefIdx.getNextSlot(), DeadVN));
+    LI->addRange(LiveRange(DefIdx, LIs->getNextSlot(DefIdx), DeadVN));
     DeadVN->addKill(DefIdx);
   }
 
@@ -786,7 +785,7 @@
     VNInfo* VNI = *VI;
     for (unsigned i = 0, e = VNI->kills.size(); i != e; ++i) {
       LiveIndex KillIdx = VNI->kills[i];
-      if (KillIdx.isPHI())
+      if (KillIdx.isPHIIndex())
         continue;
       MachineInstr *KillMI = LIs->getInstructionFromIndex(KillIdx);
       if (KillMI) {
@@ -827,14 +826,14 @@
     // Locate two-address redefinitions
     for (VNInfo::KillSet::iterator KI = OldVN->kills.begin(),
          KE = OldVN->kills.end(); KI != KE; ++KI) {
-      assert(!KI->isPHI() &&
+      assert(!KI->isPHIIndex() &&
              "VN previously reported having no PHI kills.");
       MachineInstr* MI = LIs->getInstructionFromIndex(*KI);
       unsigned DefIdx = MI->findRegisterDefOperandIdx(CurrLI->reg);
       if (DefIdx == ~0U) continue;
       if (MI->isRegTiedToUseOperand(DefIdx)) {
         VNInfo* NextVN =
-          CurrLI->findDefinedVNInfoForRegInt(KI->getDefIndex());
+          CurrLI->findDefinedVNInfoForRegInt(LIs->getDefIndex(*KI));
         if (NextVN == OldVN) continue;
         Stack.push_back(NextVN);
       }
@@ -868,8 +867,8 @@
     MachineOperand& MO = I.getOperand();
     LiveIndex InstrIdx = LIs->getInstructionIndex(&*I);
     
-    if ((MO.isUse() && NewLI.liveAt(InstrIdx.getUseIndex())) ||
-        (MO.isDef() && NewLI.liveAt(InstrIdx.getDefIndex())))
+    if ((MO.isUse() && NewLI.liveAt(LIs->getUseIndex(InstrIdx))) ||
+        (MO.isDef() && NewLI.liveAt(LIs->getDefIndex(InstrIdx))))
       OpsToChange.push_back(std::make_pair(&*I, I.getOperandNo()));
   }
   
@@ -913,7 +912,7 @@
   
   ReconstructLiveInterval(CurrLI);
   LiveIndex RematIdx = LIs->getInstructionIndex(prior(RestorePt));
-  RematIdx = RematIdx.getDefIndex();
+  RematIdx = LIs->getDefIndex(RematIdx);
   RenumberValno(CurrLI->findDefinedVNInfoForRegInt(RematIdx));
   
   ++NumSplits;
@@ -1053,14 +1052,11 @@
 /// so it would not cross the barrier that's being processed. Shrink wrap
 /// (minimize) the live interval to the last uses.
 bool PreAllocSplitting::SplitRegLiveInterval(LiveInterval *LI) {
-  DEBUG(errs() << "Pre-alloc splitting " << LI->reg << " for " << *Barrier
-               << "  result: ");
-
   CurrLI = LI;
 
   // Find live range where current interval cross the barrier.
   LiveInterval::iterator LR =
-    CurrLI->FindLiveRangeContaining(BarrierIdx.getUseIndex());
+    CurrLI->FindLiveRangeContaining(LIs->getUseIndex(BarrierIdx));
   VNInfo *ValNo = LR->valno;
 
   assert(!ValNo->isUnused() && "Val# is defined by a dead def?");
@@ -1069,10 +1065,8 @@
     ? LIs->getInstructionFromIndex(ValNo->def) : NULL;
 
   // If this would create a new join point, do not split.
-  if (DefMI && createsNewJoin(LR, DefMI->getParent(), Barrier->getParent())) {
-    DEBUG(errs() << "FAILED (would create a new join point).\n");
+  if (DefMI && createsNewJoin(LR, DefMI->getParent(), Barrier->getParent()))
     return false;
-  }
 
   // Find all references in the barrier mbb.
   SmallPtrSet<MachineInstr*, 4> RefsInMBB;
@@ -1087,17 +1081,13 @@
   LiveIndex RestoreIndex;
   MachineBasicBlock::iterator RestorePt =
     findRestorePoint(BarrierMBB, Barrier, LR->end, RefsInMBB, RestoreIndex);
-  if (RestorePt == BarrierMBB->end()) {
-    DEBUG(errs() << "FAILED (could not find a suitable restore point).\n");
+  if (RestorePt == BarrierMBB->end())
     return false;
-  }
 
   if (DefMI && LIs->isReMaterializable(*LI, ValNo, DefMI))
     if (Rematerialize(LI->reg, ValNo, DefMI, RestorePt,
-                      RestoreIndex, RefsInMBB)) {
-      DEBUG(errs() << "success (remat).\n");
-      return true;
-    }
+                      RestoreIndex, RefsInMBB))
+    return true;
 
   // Add a spill either before the barrier or after the definition.
   MachineBasicBlock *DefMBB = DefMI ? DefMI->getParent() : NULL;
@@ -1113,10 +1103,8 @@
     } else {
       MachineBasicBlock::iterator SpillPt = 
         findSpillPoint(BarrierMBB, Barrier, NULL, RefsInMBB, SpillIndex);
-      if (SpillPt == BarrierMBB->begin()) {
-        DEBUG(errs() << "FAILED (could not find a suitable spill point).\n");
+      if (SpillPt == BarrierMBB->begin())
         return false; // No gap to insert spill.
-      }
       // Add spill.
     
       SS = CreateSpillStackSlot(CurrLI->reg, RC);
@@ -1128,10 +1116,8 @@
                                  RestoreIndex, SpillIndex, SS)) {
     // If it's already split, just restore the value. There is no need to spill
     // the def again.
-    if (!DefMI) {
-      DEBUG(errs() << "FAILED (def is dead).\n");
+    if (!DefMI)
       return false; // Def is dead. Do nothing.
-    }
     
     if ((SpillMI = FoldSpill(LI->reg, RC, DefMI, Barrier,
                              BarrierMBB, SS, RefsInMBB))) {
@@ -1143,16 +1129,12 @@
         // Add spill after the def and the last use before the barrier.
         SpillPt = findSpillPoint(BarrierMBB, Barrier, DefMI,
                                  RefsInMBB, SpillIndex);
-        if (SpillPt == DefMBB->begin()) {
-          DEBUG(errs() << "FAILED (could not find a suitable spill point).\n");
+        if (SpillPt == DefMBB->begin())
           return false; // No gap to insert spill.
-        }
       } else {
         SpillPt = findNextEmptySlot(DefMBB, DefMI, SpillIndex);
-        if (SpillPt == DefMBB->end()) {
-          DEBUG(errs() << "FAILED (could not find a suitable spill point).\n");
+        if (SpillPt == DefMBB->end())
           return false; // No gap to insert spill.
-        }
       }
       // Add spill. 
       SS = CreateSpillStackSlot(CurrLI->reg, RC);
@@ -1180,19 +1162,18 @@
   }
 
   // Update spill stack slot live interval.
-  UpdateSpillSlotInterval(ValNo, SpillIndex.getUseIndex().getNextSlot(),
-                          RestoreIndex.getDefIndex());
+  UpdateSpillSlotInterval(ValNo, LIs->getNextSlot(LIs->getUseIndex(SpillIndex)),
+                          LIs->getDefIndex(RestoreIndex));
 
   ReconstructLiveInterval(CurrLI);
 
   if (!FoldedRestore) {
     LiveIndex RestoreIdx = LIs->getInstructionIndex(prior(RestorePt));
-    RestoreIdx = RestoreIdx.getDefIndex();
+    RestoreIdx = LIs->getDefIndex(RestoreIdx);
     RenumberValno(CurrLI->findDefinedVNInfoForRegInt(RestoreIdx));
   }
   
   ++NumSplits;
-  DEBUG(errs() << "success.\n");
   return true;
 }
 
@@ -1274,7 +1255,7 @@
     for (MachineRegisterInfo::use_iterator UI = MRI->use_begin((*LI)->reg),
          UE = MRI->use_end(); UI != UE; ++UI) {
       LiveIndex index = LIs->getInstructionIndex(&*UI);
-      index = index.getUseIndex();
+      index = LIs->getUseIndex(index);
       
       const LiveRange* LR = (*LI)->getLiveRangeContaining(index);
       VNUseCount[LR->valno].insert(&*UI);




From asl at math.spbu.ru  Tue Oct 20 16:37:45 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Tue, 20 Oct 2009 21:37:45 -0000
Subject: [llvm-commits] [llvm] r84683 - in /llvm/trunk:
 lib/CodeGen/SelectionDAG/DAGCombiner.cpp
 test/CodeGen/ARM/2009-10-21-InvalidFNeg.ll
Message-ID: <200910202137.n9KLbkPN020793@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 16:37:45 2009
New Revision: 84683

URL: http://llvm.org/viewvc/llvm-project?rev=84683&view=rev
Log:
Fix invalid for vector types fneg(bitconvert(x)) => bitconvert(x ^ sign)
transform.

Added:
    llvm/trunk/test/CodeGen/ARM/2009-10-21-InvalidFNeg.ll
Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=84683&r1=84682&r2=84683&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Tue Oct 20 16:37:45 2009
@@ -4381,15 +4381,17 @@
 
 SDValue DAGCombiner::visitFNEG(SDNode *N) {
   SDValue N0 = N->getOperand(0);
+  EVT VT = N->getValueType(0);
 
   if (isNegatibleForFree(N0, LegalOperations))
     return GetNegatedExpression(N0, DAG, LegalOperations);
 
   // Transform fneg(bitconvert(x)) -> bitconvert(x^sign) to avoid loading
   // constant pool values.
-  if (N0.getOpcode() == ISD::BIT_CONVERT && N0.getNode()->hasOneUse() &&
-      N0.getOperand(0).getValueType().isInteger() &&
-      !N0.getOperand(0).getValueType().isVector()) {
+  if (N0.getOpcode() == ISD::BIT_CONVERT && 
+      !VT.isVector() &&
+      N0.getNode()->hasOneUse() &&
+      N0.getOperand(0).getValueType().isInteger()) {
     SDValue Int = N0.getOperand(0);
     EVT IntVT = Int.getValueType();
     if (IntVT.isInteger() && !IntVT.isVector()) {
@@ -4397,7 +4399,7 @@
               DAG.getConstant(APInt::getSignBit(IntVT.getSizeInBits()), IntVT));
       AddToWorkList(Int.getNode());
       return DAG.getNode(ISD::BIT_CONVERT, N->getDebugLoc(),
-                         N->getValueType(0), Int);
+                         VT, Int);
     }
   }
 

Added: llvm/trunk/test/CodeGen/ARM/2009-10-21-InvalidFNeg.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/2009-10-21-InvalidFNeg.ll?rev=84683&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/ARM/2009-10-21-InvalidFNeg.ll (added)
+++ llvm/trunk/test/CodeGen/ARM/2009-10-21-InvalidFNeg.ll Tue Oct 20 16:37:45 2009
@@ -0,0 +1,48 @@
+; RUN: llc -mcpu=cortex-a8 -mattr=+neon < %s | grep vneg
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64"
+target triple = "armv7-eabi"
+
+%aaa = type { %fff, %fff }
+%bbb = type { [6 x %ddd] }
+%ccc = type { %eee, %fff }
+%ddd = type { %fff }
+%eee = type { %fff, %fff, %fff, %fff }
+%fff = type { %struct.vec_float4 }
+%struct.vec_float4 = type { <4 x float> }
+
+define linkonce_odr arm_aapcs_vfpcc void @foo(%eee* noalias sret %agg.result, i64 %tfrm.0.0, i64 %tfrm.0.1, i64 %tfrm.0.2, i64 %tfrm.0.3, i64 %tfrm.0.4, i64 %tfrm.0.5, i64 %tfrm.0.6, i64 %tfrm.0.7) nounwind noinline {
+entry:
+  %tmp104 = zext i64 %tfrm.0.2 to i512            ; <i512> [#uses=1]
+  %tmp105 = shl i512 %tmp104, 128                 ; <i512> [#uses=1]
+  %tmp118 = zext i64 %tfrm.0.3 to i512            ; <i512> [#uses=1]
+  %tmp119 = shl i512 %tmp118, 192                 ; <i512> [#uses=1]
+  %ins121 = or i512 %tmp119, %tmp105              ; <i512> [#uses=1]
+  %tmp99 = zext i64 %tfrm.0.4 to i512             ; <i512> [#uses=1]
+  %tmp100 = shl i512 %tmp99, 256                  ; <i512> [#uses=1]
+  %tmp123 = zext i64 %tfrm.0.5 to i512            ; <i512> [#uses=1]
+  %tmp124 = shl i512 %tmp123, 320                 ; <i512> [#uses=1]
+  %tmp96 = zext i64 %tfrm.0.6 to i512             ; <i512> [#uses=1]
+  %tmp97 = shl i512 %tmp96, 384                   ; <i512> [#uses=1]
+  %tmp128 = zext i64 %tfrm.0.7 to i512            ; <i512> [#uses=1]
+  %tmp129 = shl i512 %tmp128, 448                 ; <i512> [#uses=1]
+  %mask.masked = or i512 %tmp124, %tmp100         ; <i512> [#uses=1]
+  %ins131 = or i512 %tmp129, %tmp97               ; <i512> [#uses=1]
+  %tmp109132 = zext i64 %tfrm.0.0 to i128         ; <i128> [#uses=1]
+  %tmp113134 = zext i64 %tfrm.0.1 to i128         ; <i128> [#uses=1]
+  %tmp114133 = shl i128 %tmp113134, 64            ; <i128> [#uses=1]
+  %tmp94 = or i128 %tmp114133, %tmp109132         ; <i128> [#uses=1]
+  %tmp95 = bitcast i128 %tmp94 to <4 x float>     ; <<4 x float>> [#uses=0]
+  %tmp82 = lshr i512 %ins121, 128                 ; <i512> [#uses=1]
+  %tmp83 = trunc i512 %tmp82 to i128              ; <i128> [#uses=1]
+  %tmp84 = bitcast i128 %tmp83 to <4 x float>     ; <<4 x float>> [#uses=0]
+  %tmp86 = lshr i512 %mask.masked, 256            ; <i512> [#uses=1]
+  %tmp87 = trunc i512 %tmp86 to i128              ; <i128> [#uses=1]
+  %tmp88 = bitcast i128 %tmp87 to <4 x float>     ; <<4 x float>> [#uses=0]
+  %tmp90 = lshr i512 %ins131, 384                 ; <i512> [#uses=1]
+  %tmp91 = trunc i512 %tmp90 to i128              ; <i128> [#uses=1]
+  %tmp92 = bitcast i128 %tmp91 to <4 x float>     ; <<4 x float>> [#uses=1]
+  %tmp = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %tmp92 ; <<4 x float>> [#uses=1]
+  %tmp28 = getelementptr inbounds %eee* %agg.result, i32 0, i32 3, i32 0, i32 0 ; <<4 x float>*> [#uses=1]
+  store <4 x float> %tmp, <4 x float>* %tmp28, align 16
+  ret void
+}




From gohman at apple.com  Tue Oct 20 16:37:49 2009
From: gohman at apple.com (Dan Gohman)
Date: Tue, 20 Oct 2009 14:37:49 -0700
Subject: [llvm-commits] [llvm] r84327 -
	/llvm/trunk/lib/CodeGen/CodePlacementOpt.cpp
In-Reply-To: <6a8523d60910190522w2b8bc18eu8af2b5c2b0bee681@mail.gmail.com>
References: <200910170933.n9H9XeSH030932@zion.cs.uiuc.edu>
	<B4FE2DBB-405E-419B-8C8C-255D4203BEA4@apple.com>
	<5671CCF8-74F5-4544-893B-711524124B41@apple.com>
	<6a8523d60910190522w2b8bc18eu8af2b5c2b0bee681@mail.gmail.com>
Message-ID: <813AD5EF-58EF-4E24-9480-6A58FAE72A01@apple.com>


On Oct 19, 2009, at 5:22 AM, Daniel Dunbar wrote:

> On Sun, Oct 18, 2009 at 8:36 PM, Evan Cheng <evan.cheng at apple.com> wrote:
>> 
>> On Oct 17, 2009, at 10:44 AM, Chris Lattner wrote:
>> 
>>> 
>>> On Oct 17, 2009, at 2:33 AM, Daniel Dunbar wrote:
>>> 
>>>> Author: ddunbar
>>>> Date: Sat Oct 17 04:33:00 2009
>>>> New Revision: 84327
>>>> 
>>>> URL: http://llvm.org/viewvc/llvm-project?rev=84327&view=rev
>>>> Log:
>>>> Suppress -Asserts warning.
>>> 
>>> Despite the name, I think that AnalyzeBranch does mutate the code in
>>> some cases on some targets.  Is this true?
>>> 
>> 
>> Not in this case since AllowModify is false.
>> 
>> virtual bool AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
>>                             MachineBasicBlock *&FBB,
>>                             SmallVectorImpl<MachineOperand> &Cond,
>>                             bool AllowModify = false) const {
> 
> Ok. Please put it back inside the assert if it doesn't do anything
> else and you don't want it called in a -Asserts build, I just wanted
> the build warning to go away.

We do want it called in a -Asserts build. It sets TBB, FBB, and Cond
(by-reference). If the current code "(void) B;" is sufficient to
avoid the warning, then it's fine.

Dan




From asl at math.spbu.ru  Tue Oct 20 16:42:20 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Tue, 20 Oct 2009 21:42:20 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84685 -
	/llvm-gcc-4.2/trunk/gcc/config/arm/arm.h
Message-ID: <200910202142.n9KLgKrw020978@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 16:42:20 2009
New Revision: 84685

URL: http://llvm.org/viewvc/llvm-project?rev=84685&view=rev
Log:
Propagate vfp3 subtarget features to backend

Modified:
    llvm-gcc-4.2/trunk/gcc/config/arm/arm.h

Modified: llvm-gcc-4.2/trunk/gcc/config/arm/arm.h
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/arm/arm.h?rev=84685&r1=84684&r2=84685&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/arm/arm.h (original)
+++ llvm-gcc-4.2/trunk/gcc/config/arm/arm.h Tue Oct 20 16:42:20 2009
@@ -3480,6 +3480,10 @@
       F.setCPU("arm7tdmi"); \
       break; \
     } \
+    if (TARGET_VFP3)						\
+      F.AddFeature("vfp3");					\
+    else							\
+      F.AddFeature("vfp3", false);				\
     if (TARGET_NEON)						\
       F.AddFeature("neon");					\
     else							\




From bob.wilson at apple.com  Tue Oct 20 16:44:20 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Tue, 20 Oct 2009 14:44:20 -0700
Subject: [llvm-commits] Patch for encodings of VEXT (Vector Extract)
In-Reply-To: <BCD8351B-64D7-416B-8780-E787553E359D@apple.com>
References: <AD31FE2C-054B-424B-A0DC-27C3F3A6BC4A@apple.com>
	<BCD8351B-64D7-416B-8780-E787553E359D@apple.com>
Message-ID: <37A6DD2D-3CF4-4DD2-B62F-F03428E55C86@apple.com>

Nevermind.  I had been thinking about this only from the perspective  
of encoding.  I can see why you want this for decoding.

Johnny suggested to me off-line that we could avoid this by having  
TableGen recognize "x" bits, but I haven't yet looked into that.  For  
now, at least, it makes sense to go ahead and add the new instruction  
formats.

On Oct 20, 2009, at 11:57 AM, Bob Wilson wrote:

> Is this really the direction we want to go?  There are a fair number
> of NEON instruction encodings where it is very useful to have some
> default bits set to zero by the .td file, assuming that the code
> emitter may overwrite them later.  For example, the vector shift-
> immediate instructions have the "imm6" field that specifies both the
> vector element size and the shift amount, where the number of bits
> used for the shift amount varies depending on the element size.  See
> VSHR, for example.  It is useful to specify the element size in
> the .td file and it is a lot more convenient to do that by stuffing
> zeros
> into the shift amount.
>
> Another example of this is the set of NEON instructions using the "One
> register and a modified immediate value" format (A7.4.6).  Currently
> we've only got the VMOV variant implemented in LLVM.  Some of the bits
> in the "cmode" field can be set in the .td file but some depend on the
> immediate value.
>
> It's not a big problem to have a special N3VImm format for VEXT, but
> it seems like we might need a different solution for the general
> problem, regardless of what we do for VEXT.  Is it important that you
> leave those bits unspecified?  You didn't say why the patch was  
> needed.
>
> On Oct 16, 2009, at 11:17 AM, Johnny Chen wrote:
>
>> Hi,
>>
>> Refs: A8-598.
>> Leave Inst{11-8}, which represents the starting byte index of the
>> extracted result in the
>> concatenation of the operands and is left unspecified.
>>
>> <ARMInstrNEON.patch5>_______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From clattner at apple.com  Tue Oct 20 16:48:29 2009
From: clattner at apple.com (Chris Lattner)
Date: Tue, 20 Oct 2009 14:48:29 -0700
Subject: [llvm-commits] metadata.h/cpp review stuff
In-Reply-To: <352a1fb20910200708n4336cdaax503fce6480996830@mail.gmail.com>
References: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>
	<352a1fb20910200708n4336cdaax503fce6480996830@mail.gmail.com>
Message-ID: <DD3781CA-F185-4E23-BB49-4D464DA11B22@apple.com>

On Oct 20, 2009, at 7:08 AM, Devang Patel wrote:
>> This would shrink them significantly and reduce the games you have to
>> play with hiding getNumOperands, overloading operator new, etc.   
>> Also,
>> MetadataBase doesn't have much common state, it isn't clear it needs
>> to exist at all.
>>
>> There is some seriously confused code circling around
>> MetadataBase::resizeOperands.  It looks like half the code thinks  
>> that
>> operands can happen, but half doesn't.  Please remove all this stuff.
>
> It is trying to maintain following use counts
>
> !1 = { i32 1} ; [uses 1]
> !2 = {  !1 }   ; [uses 0]
>
> ...
> call @llvm.dbg.blah(!1, !2)
> ...
>
> Metadata used by another metadata is counted as regular uses. Metadata
> used by non-metadata values is not a normal use (it is invisible to
> the non-metadata value). This is why the hierarchy is based on User
> and operand list is only counting metadata uses.

Ok, I see now that you're doing this.  But why? :)  What capability  
does this provide?  Is there another way to get it?  This capability  
costs a lot of complexity, memory and compile time.

-Chris


From daniel at zuster.org  Tue Oct 20 17:10:05 2009
From: daniel at zuster.org (Daniel Dunbar)
Date: Tue, 20 Oct 2009 22:10:05 -0000
Subject: [llvm-commits] [llvm] r84687 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
Message-ID: <200910202210.n9KMA5mf022219@zion.cs.uiuc.edu>

Author: ddunbar
Date: Tue Oct 20 17:10:05 2009
New Revision: 84687

URL: http://llvm.org/viewvc/llvm-project?rev=84687&view=rev
Log:
Fix -Asserts warning.

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp?rev=84687&r1=84686&r2=84687&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.cpp Tue Oct 20 17:10:05 2009
@@ -332,8 +332,9 @@
 }
 
 void ARMInstPrinter::printSBitModifierOperand(const MCInst *MI, unsigned OpNum){
-  if (unsigned Reg = MI->getOperand(OpNum).getReg()) {
-    assert(Reg == ARM::CPSR && "Expect ARM CPSR register!");
+  if (MI->getOperand(OpNum).getReg()) {
+    assert(MI->getOperand(OpNum).getReg() == ARM::CPSR &&
+           "Expect ARM CPSR register!");
     O << 's';
   }
 }




From dpatel at apple.com  Tue Oct 20 17:10:17 2009
From: dpatel at apple.com (Devang Patel)
Date: Tue, 20 Oct 2009 15:10:17 -0700
Subject: [llvm-commits] metadata.h/cpp review stuff
In-Reply-To: <DD3781CA-F185-4E23-BB49-4D464DA11B22@apple.com>
References: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>
	<352a1fb20910200708n4336cdaax503fce6480996830@mail.gmail.com>
	<DD3781CA-F185-4E23-BB49-4D464DA11B22@apple.com>
Message-ID: <F85930CE-3A4B-478B-B0C6-E20EE110B47E@apple.com>


On Oct 20, 2009, at 2:48 PM, Chris Lattner wrote:

> On Oct 20, 2009, at 7:08 AM, Devang Patel wrote:
>>> This would shrink them significantly and reduce the games you have  
>>> to
>>> play with hiding getNumOperands, overloading operator new, etc.   
>>> Also,
>>> MetadataBase doesn't have much common state, it isn't clear it needs
>>> to exist at all.
>>>
>>> There is some seriously confused code circling around
>>> MetadataBase::resizeOperands.  It looks like half the code thinks  
>>> that
>>> operands can happen, but half doesn't.  Please remove all this  
>>> stuff.
>>
>> It is trying to maintain following use counts
>>
>> !1 = { i32 1} ; [uses 1]
>> !2 = {  !1 }   ; [uses 0]
>>
>> ...
>> call @llvm.dbg.blah(!1, !2)
>> ...
>>
>> Metadata used by another metadata is counted as regular uses.  
>> Metadata
>> used by non-metadata values is not a normal use (it is invisible to
>> the non-metadata value). This is why the hierarchy is based on User
>> and operand list is only counting metadata uses.
>
> Ok, I see now that you're doing this.  But why? :)  What capability  
> does this provide?  Is there another way to get it?  This capability  
> costs a lot of complexity, memory and compile time.


This may be useful to reduce memory pressure by allowing metadata user  
to delete MDNode and its MDNode users. After stumbling couple of times  
to get this, I am also wondering whether this is worth it or not or is  
this the right approach.

MDNodes are uniqued, just like Constant. However Constants become  
zoombie if its use count is reduced to zero. This may be too expensive  
for metadata because of possibility of high volume of metadata in the  
IR due to various reasons.

-
Devang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091020/50249b08/attachment.html 

From johnny.chen at apple.com  Tue Oct 20 17:23:17 2009
From: johnny.chen at apple.com (Johnny Chen)
Date: Tue, 20 Oct 2009 15:23:17 -0700
Subject: [llvm-commits] small patch for utils/emacs/tablegen-mode.el
Message-ID: <B6735BF7-0D29-4527-92BA-DE8954B800E8@apple.com>

For setting the comment-start string.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: tablegen-mode.patch
Type: application/octet-stream
Size: 511 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091020/1203052a/attachment.obj 

From isanbard at gmail.com  Tue Oct 20 17:35:00 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Tue, 20 Oct 2009 22:35:00 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84688 - in /llvm-gcc-4.2/trunk/gcc:
 ChangeLog.apple config/arm/arm-protos.h config/arm/linux-eabi.h
 config/darwin-c.c cp/parser.c doc/invoke.texi gcc.c tree-eh.c version.c
Message-ID: <200910202235.n9KMZ0MC023283@zion.cs.uiuc.edu>

Author: void
Date: Tue Oct 20 17:34:59 2009
New Revision: 84688

URL: http://llvm.org/viewvc/llvm-project?rev=84688&view=rev
Log:
Sync to Apple gcc 5654.

Modified:
    llvm-gcc-4.2/trunk/gcc/ChangeLog.apple
    llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h
    llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h
    llvm-gcc-4.2/trunk/gcc/config/darwin-c.c
    llvm-gcc-4.2/trunk/gcc/cp/parser.c
    llvm-gcc-4.2/trunk/gcc/doc/invoke.texi
    llvm-gcc-4.2/trunk/gcc/gcc.c
    llvm-gcc-4.2/trunk/gcc/tree-eh.c
    llvm-gcc-4.2/trunk/gcc/version.c

Modified: llvm-gcc-4.2/trunk/gcc/ChangeLog.apple
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/ChangeLog.apple?rev=84688&r1=84687&r2=84688&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/ChangeLog.apple (original)
+++ llvm-gcc-4.2/trunk/gcc/ChangeLog.apple Tue Oct 20 17:34:59 2009
@@ -1,3 +1,18 @@
+2009-10-15  Jim Grosbach <grosbach at apple.com>
+
+	Radar 7295521
+	* config/arm/arm.c (arm_split_compare_and_swap): Call insns for
+	arm_memory_barrier* and arm_memory_sync* directory rather than using
+	the expander.
+	* config/arm/sync.md (memory_barrier, memory_sync): Allocate a scratch
+	register rather than forcing IP.
+	(arm_memory_barrier_v6): Match the scratch register from the expander
+	rather than forcing IP. Rename to arm_memory_barrier_v6_explicit.
+	(arm_memory_barrier_v6_explicit): New.
+	(arm_memory_sync_v6): Match the scratch register from the expander
+	rather than forcing IP. Rename to arm_memory_sync_v6_explicit.
+	(arm_memory_sync_v6_explicit): New.
+
 2009-09-11  Caroline Tice  <ctice at apple.com>
 
 	Radar 6419781
@@ -9,8 +24,8 @@
 
 2009-09-08  Jim Grosbach <grosbach at apple.com>
 
-       Radar 7174451
-       * config/arm/arm.c (arm_select_dominance_cc_mode): Unordered floats.
+	Radar 7174451
+	* config/arm/arm.c (arm_select_dominance_cc_mode): Unordered floats.
 
 2009-09-08  Jim Grosbach <grosbach at apple.com>
 

Modified: llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h?rev=84688&r1=84687&r2=84688&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h (original)
+++ llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h Tue Oct 20 17:34:59 2009
@@ -258,7 +258,14 @@
 /* APPLE LOCAL 5946347 ms_struct support */
 extern int arm_field_ms_struct_align (tree);
 
-/* LLVM LOCAL pr5037 removed arm_mangle_type */
+/* LLVM LOCAL begin pr5037 removed arm_mangle_type */
+#ifndef ENABLE_LLVM
+/* APPLE LOCAL begin v7 support. Merge from Codesourcery */
+extern const char *arm_mangle_type (tree);
+
+/* APPLE LOCAL end v7 support. Merge from Codesourcery */
+#endif
+/* LLVM LOCAL end pr5037 removed arm_mangle_type */
 
 /* APPLE LOCAL v7 support. Fix compact switch tables */
 extern void arm_asm_output_addr_diff_vec (FILE *file, rtx LABEL, rtx BODY);

Modified: llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h?rev=84688&r1=84687&r2=84688&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h (original)
+++ llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h Tue Oct 20 17:34:59 2009
@@ -79,7 +79,7 @@
   register unsigned long _end __asm ("a2") = (unsigned long) (END);	\
   register unsigned long _flg __asm ("a3") = 0;				\
   register unsigned long _scno __asm ("r7") = 0xf0002;			\
-  __asm __volatile ("swi 0		@ sys_cacheflush"		\
+  __asm __volatile ("swi 0x9f0002		@ sys_cacheflush"	\
 		    : "=r" (_beg)					\
 		    : "0" (_beg), "r" (_end), "r" (_flg), "r" (_scno));	\
 }

Modified: llvm-gcc-4.2/trunk/gcc/config/darwin-c.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/darwin-c.c?rev=84688&r1=84687&r2=84688&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/darwin-c.c (original)
+++ llvm-gcc-4.2/trunk/gcc/config/darwin-c.c Tue Oct 20 17:34:59 2009
@@ -412,7 +412,7 @@
 /* APPLE LOCAL end pragma reverse_bitfields */
 
 /* APPLE LOCAL begin optimization pragmas 3124235/3420242 */
-static GTY(()) varray_type va_opt;
+varray_type va_opt;
 
 static void
 push_opt_level (int level, int size)

Modified: llvm-gcc-4.2/trunk/gcc/cp/parser.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/cp/parser.c?rev=84688&r1=84687&r2=84688&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/cp/parser.c (original)
+++ llvm-gcc-4.2/trunk/gcc/cp/parser.c Tue Oct 20 17:34:59 2009
@@ -21078,12 +21078,11 @@
   CONSTRUCTOR_ELTS (constructor) = build_block_struct_initlist (block_struct_type,
                                                                 block_impl);
   /* Temporary representing a global block is made global static.  */
-  /* APPLE LOCAL begin radar 6230297 */
+  /* APPLE LOCAL radar 6230297 */
   if (staticBlockTmp || global_bindings_p ()) {
     TREE_PUBLIC (block_holder_tmp_decl) = 0;
     TREE_STATIC (block_holder_tmp_decl) = 1;
   }
-  /* APPLE LOCAL end radar 6230297 */
   cp_finish_decl (block_holder_tmp_decl, constructor, 0, 0, LOOKUP_ONLYCONVERTING);
   return block_holder_tmp_decl;
 }

Modified: llvm-gcc-4.2/trunk/gcc/doc/invoke.texi
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/doc/invoke.texi?rev=84688&r1=84687&r2=84688&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/doc/invoke.texi (original)
+++ llvm-gcc-4.2/trunk/gcc/doc/invoke.texi Tue Oct 20 17:34:59 2009
@@ -175,8 +175,7 @@
 @table @emph
 @item Overall Options
 @xref{Overall Options,,Options Controlling the Kind of Output}.
- at gccoptlist{-c  -S  -E  -o @var{file}  -combine  -no-canonical-prefixes  @gol
--pipe  -pass-exit-codes  @gol
+ at gccoptlist{-c  -S  -E  -o @var{file}  -combine -pipe  -pass-exit-codes  @gol
 @c APPLE LOCAL -ObjC 2001-08-03 --sts **
 -ObjC (APPLE ONLY) -ObjC++ (APPLE ONLY) @gol
 @c APPLE LOCAL begin fat builds
@@ -1239,12 +1238,6 @@
 Print (on the standard output) a description of target specific command
 line options for each tool.
 
- at item -no-canonical-prefixes
- at opindex no-canonical-prefixes
-Do not expand any symbolic links, resolve references to @samp{/../}
-or @samp{/./}, or make the path absolute when generating a relative
-prefix.
-
 @item --version
 @opindex version
 Display the version number and copyrights of the invoked GCC at .

Modified: llvm-gcc-4.2/trunk/gcc/gcc.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/gcc.c?rev=84688&r1=84687&r2=84688&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/gcc.c (original)
+++ llvm-gcc-4.2/trunk/gcc/gcc.c Tue Oct 20 17:34:59 2009
@@ -1190,7 +1190,6 @@
    /* LLVM LOCAL end */
    {"--machine", "-m", "aj"},
    {"--machine-", "-m", "*j"},
-   {"--no-canonical-prefixes", "-no-canonical-prefixes", 0},
    {"--no-integrated-cpp", "-no-integrated-cpp", 0},
    {"--no-line-commands", "-P", 0},
    {"--no-precompiled-includes", "-noprecomp", 0},
@@ -3342,10 +3341,6 @@
   fputs (_("  -Xlinker <arg>           Pass <arg> on to the linker\n"), stdout);
   fputs (_("  -combine                 Pass multiple source files to compiler at once\n"), stdout);
   fputs (_("  -save-temps              Do not delete intermediate files\n"), stdout);
-  fputs (_("\
-  -no-canonical-prefixes   Do not canonicalize paths when building relative\n\
-                           prefixes to other gcc components\n"), stdout);
-
   fputs (_("  -pipe                    Use pipes rather than intermediate files\n"), stdout);
   fputs (_("  -time                    Time the execution of each subprocess\n"), stdout);
   fputs (_("  -specs=<file>            Override built-in specs with the contents of <file>\n"), stdout);
@@ -3438,8 +3433,6 @@
   int is_modify_target_name;
   unsigned int j;
 #endif
-  char *(*get_relative_prefix) (const char *, const char *,
-				const char *) = NULL;
 
   GET_ENVIRONMENT (gcc_exec_prefix, "GCC_EXEC_PREFIX");
 
@@ -3526,28 +3519,6 @@
       fatal ("couldn't run '%s': %s", new_argv0, xstrerror (errno));
     }
 
-  /* Convert new-style -- options to old-style.  */
-  translate_options (&argc, (const char * const **) &argv);
-
-  /* Do language-specific adjustment/addition of flags.  */
-  lang_specific_driver (&argc, (const char * const **) &argv,
-			&added_libraries);
-
-  /* Handle any -no-canonical-prefixes flag early, to assign the function
-     that builds relative prefixes.  This function creates default search
-     paths that are needed later in normal option handling.  */
-
-  for (i = 1; i < argc; i++)
-    {
-      if (! strcmp (argv[i], "-no-canonical-prefixes"))
-	{
-	  get_relative_prefix = make_relative_prefix_ignore_links;
-	  break;
-	}
-    }
-  if (! get_relative_prefix)
-    get_relative_prefix = make_relative_prefix;
-
   /* Set up the default search paths.  If there is no GCC_EXEC_PREFIX,
      see if we can create it from the pathname specified in argv[0].  */
 
@@ -3556,12 +3527,11 @@
   /* FIXME: make_relative_prefix doesn't yet work for VMS.  */
   if (!gcc_exec_prefix)
     {
-      gcc_exec_prefix = get_relative_prefix (argv[0],
-					     standard_bindir_prefix,
-					     standard_exec_prefix);
-      gcc_libexec_prefix = get_relative_prefix (argv[0],
-					     standard_bindir_prefix,
-					     standard_libexec_prefix);
+      gcc_exec_prefix = make_relative_prefix (argv[0], standard_bindir_prefix,
+					      standard_exec_prefix);
+      gcc_libexec_prefix = make_relative_prefix (argv[0],
+						 standard_bindir_prefix,
+						 standard_libexec_prefix);
       if (gcc_exec_prefix)
 	putenv (concat ("GCC_EXEC_PREFIX=", gcc_exec_prefix, NULL));
     }
@@ -3572,9 +3542,9 @@
 	 / (which is ignored by make_relative_prefix), so append a
 	 program name.  */
       char *tmp_prefix = concat (gcc_exec_prefix, "gcc", NULL);
-      gcc_libexec_prefix = get_relative_prefix (tmp_prefix,
-						standard_exec_prefix,
-						standard_libexec_prefix);
+      gcc_libexec_prefix = make_relative_prefix (tmp_prefix,
+						 standard_exec_prefix,
+						 standard_libexec_prefix);
       free (tmp_prefix);
     }
 #else
@@ -3704,6 +3674,12 @@
 	}
     }
 
+  /* Convert new-style -- options to old-style.  */
+  translate_options (&argc, (const char *const **) &argv);
+
+  /* Do language-specific adjustment/addition of flags.  */
+  lang_specific_driver (&argc, (const char *const **) &argv, &added_libraries);
+
   /* Scan argv twice.  Here, the first time, just count how many switches
      there will be in their vector, and how many input files in theirs.
      Here we also parse the switches that cc itself uses (e.g. -v).  */
@@ -3915,9 +3891,6 @@
 	  n_infiles += 2;
 	  i++;
 	}
-      else if (strcmp (argv[i], "-no-canonical-prefixes") == 0)
-	/* Already handled as a special case, so ignored here.  */
-	;
       /* APPLE LOCAL end -weak_* (radar 3235250) */
       else if (strcmp (argv[i], "-combine") == 0)
 	{
@@ -4251,9 +4224,9 @@
      ``make_relative_prefix'' is not compiled for VMS, so don't call it.  */
   if (target_system_root && gcc_exec_prefix)
     {
-      char *tmp_prefix = get_relative_prefix (argv[0],
-					      standard_bindir_prefix,
-					      target_system_root);
+      char *tmp_prefix = make_relative_prefix (argv[0],
+					       standard_bindir_prefix,
+					       target_system_root);
       if (tmp_prefix && access_check (tmp_prefix, F_OK) == 0)
 	{
 	  target_system_root = tmp_prefix;
@@ -4295,8 +4268,6 @@
 	;
       else if (! strncmp (argv[i], "-Wp,", 4))
 	;
-      else if (! strcmp (argv[i], "-no-canonical-prefixes"))
-	;
       else if (! strcmp (argv[i], "-pass-exit-codes"))
 	;
       else if (! strcmp (argv[i], "-print-search-dirs"))

Modified: llvm-gcc-4.2/trunk/gcc/tree-eh.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/tree-eh.c?rev=84688&r1=84687&r2=84688&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/tree-eh.c (original)
+++ llvm-gcc-4.2/trunk/gcc/tree-eh.c Tue Oct 20 17:34:59 2009
@@ -839,10 +839,12 @@
 
       save_eptr = create_tmp_var (ptr_type_node, "save_eptr");
       save_filt = create_tmp_var (integer_type_node, "save_filt");
+
       i = tsi_start (finally);
       x = build0 (EXC_PTR_EXPR, ptr_type_node);
       x = build2 (MODIFY_EXPR, void_type_node, save_eptr, x);
       tsi_link_before (&i, x, TSI_CONTINUE_LINKING);
+
       x = build0 (FILTER_EXPR, integer_type_node);
       x = build2 (MODIFY_EXPR, void_type_node, save_filt, x);
       tsi_link_before (&i, x, TSI_CONTINUE_LINKING);

Modified: llvm-gcc-4.2/trunk/gcc/version.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/version.c?rev=84688&r1=84687&r2=84688&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/version.c (original)
+++ llvm-gcc-4.2/trunk/gcc/version.c Tue Oct 20 17:34:59 2009
@@ -11,12 +11,12 @@
 /* APPLE LOCAL begin Apple version */
 #ifdef ENABLE_LLVM
 #ifdef LLVM_VERSION_INFO
-#define VERSUFFIX " (Based on Apple Inc. build 5653) (LLVM build " LLVM_VERSION_INFO ")"
+#define VERSUFFIX " (Based on Apple Inc. build 5654) (LLVM build " LLVM_VERSION_INFO ")"
 #else
-#define VERSUFFIX " (Based on Apple Inc. build 5653) (LLVM build)"
+#define VERSUFFIX " (Based on Apple Inc. build 5654) (LLVM build)"
 #endif
 #else
-#define VERSUFFIX " (Based on Apple Inc. build 5653)"
+#define VERSUFFIX " (Based on Apple Inc. build 5654)"
 #endif
 /* APPLE LOCAL end Apple version */
 




From dpatel at apple.com  Tue Oct 20 17:50:27 2009
From: dpatel at apple.com (Devang Patel)
Date: Tue, 20 Oct 2009 22:50:27 -0000
Subject: [llvm-commits] [llvm] r84689 - in /llvm/trunk:
 include/llvm/Metadata.h include/llvm/Support/IRBuilder.h
 lib/AsmParser/LLParser.cpp lib/VMCore/Metadata.cpp
Message-ID: <200910202250.n9KMoSuS023873@zion.cs.uiuc.edu>

Author: dpatel
Date: Tue Oct 20 17:50:27 2009
New Revision: 84689

URL: http://llvm.org/viewvc/llvm-project?rev=84689&view=rev
Log:
Cosmetic changes.

s/validName/isValidName/g
s/with an Instruction/to an Instruction/g
s/RegisterMDKind/registerMDKind/g

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/include/llvm/Support/IRBuilder.h
    llvm/trunk/lib/AsmParser/LLParser.cpp
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84689&r1=84688&r2=84689&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Tue Oct 20 17:50:27 2009
@@ -295,25 +295,25 @@
   StringMap<unsigned> MDHandlerNames;
 
 public:
-  /// RegisterMDKind - Register a new metadata kind and return its ID.
+  /// registerMDKind - Register a new metadata kind and return its ID.
   /// A metadata kind can be registered only once. 
-  unsigned RegisterMDKind(const char *Name);
+  unsigned registerMDKind(const char *Name);
 
   /// getMDKind - Return metadata kind. If the requested metadata kind
   /// is not registered then return 0.
   unsigned getMDKind(const char *Name);
 
-  /// validName - Return true if Name is a valid custom metadata handler name.
-  bool validName(const char *Name);
+  /// isValidName - Return true if Name is a valid custom metadata handler name.
+  bool isValidName(const char *Name);
 
-  /// getMD - Get the metadata of given kind attached with an Instruction.
+  /// getMD - Get the metadata of given kind attached to an Instruction.
   /// If the metadata is not found then return 0.
   MDNode *getMD(unsigned Kind, const Instruction *Inst);
 
-  /// getMDs - Get the metadata attached with an Instruction.
+  /// getMDs - Get the metadata attached to an Instruction.
   const MDMapTy *getMDs(const Instruction *Inst);
 
-  /// addMD - Attach the metadata of given kind with an Instruction.
+  /// addMD - Attach the metadata of given kind to an Instruction.
   void addMD(unsigned Kind, MDNode *Node, Instruction *Inst);
   
   /// removeMD - Remove metadata of given kind attached with an instuction.

Modified: llvm/trunk/include/llvm/Support/IRBuilder.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/IRBuilder.h?rev=84689&r1=84688&r2=84689&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Support/IRBuilder.h (original)
+++ llvm/trunk/include/llvm/Support/IRBuilder.h Tue Oct 20 17:50:27 2009
@@ -139,7 +139,7 @@
     if (MDKind == 0) 
       MDKind = Context.getMetadata().getMDKind("dbg");
     if (MDKind == 0)
-      MDKind = Context.getMetadata().RegisterMDKind("dbg");
+      MDKind = Context.getMetadata().registerMDKind("dbg");
     CurDbgLocation = L;
   }
 

Modified: llvm/trunk/lib/AsmParser/LLParser.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/AsmParser/LLParser.cpp?rev=84689&r1=84688&r2=84689&view=diff

==============================================================================
--- llvm/trunk/lib/AsmParser/LLParser.cpp (original)
+++ llvm/trunk/lib/AsmParser/LLParser.cpp Tue Oct 20 17:50:27 2009
@@ -1067,7 +1067,7 @@
   MetadataContext &TheMetadata = M->getContext().getMetadata();
   unsigned MDK = TheMetadata.getMDKind(Name.c_str());
   if (!MDK)
-    MDK = TheMetadata.RegisterMDKind(Name.c_str());
+    MDK = TheMetadata.registerMDKind(Name.c_str());
   MDsOnInst.push_back(std::make_pair(MDK, cast<MDNode>(Node)));
 
   return false;

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84689&r1=84688&r2=84689&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Tue Oct 20 17:50:27 2009
@@ -245,10 +245,10 @@
 // MetadataContext implementation.
 //
 
-/// RegisterMDKind - Register a new metadata kind and return its ID.
+/// registerMDKind - Register a new metadata kind and return its ID.
 /// A metadata kind can be registered only once. 
-unsigned MetadataContext::RegisterMDKind(const char *Name) {
-  assert(validName(Name) && "Invalid custome metadata name!");
+unsigned MetadataContext::registerMDKind(const char *Name) {
+  assert(isValidName(Name) && "Invalid custome metadata name!");
   unsigned Count = MDHandlerNames.size();
   assert(MDHandlerNames.find(Name) == MDHandlerNames.end() 
          && "Already registered MDKind!");
@@ -256,8 +256,8 @@
   return Count + 1;
 }
 
-/// validName - Return true if Name is a valid custom metadata handler name.
-bool MetadataContext::validName(const char *Name) {
+/// isValidName - Return true if Name is a valid custom metadata handler name.
+bool MetadataContext::isValidName(const char *Name) {
   if (!Name)
     return false;
 
@@ -280,7 +280,7 @@
 /// getMDKind - Return metadata kind. If the requested metadata kind
 /// is not registered then return 0.
 unsigned MetadataContext::getMDKind(const char *Name) {
-  assert(validName(Name) && "Invalid custome metadata name!");
+  assert(isValidName(Name) && "Invalid custome metadata name!");
   StringMap<unsigned>::iterator I = MDHandlerNames.find(Name);
   if (I == MDHandlerNames.end())
     return 0;
@@ -288,7 +288,7 @@
   return I->getValue();
 }
 
-/// addMD - Attach the metadata of given kind with an Instruction.
+/// addMD - Attach the metadata of given kind to an Instruction.
 void MetadataContext::addMD(unsigned MDKind, MDNode *Node, Instruction *Inst) {
   assert(Node && "Invalid null MDNode");
   Inst->HasMetadata = true;
@@ -362,7 +362,7 @@
       addMD(I->first, MD, In2);
 }
 
-/// getMD - Get the metadata of given kind attached with an Instruction.
+/// getMD - Get the metadata of given kind attached to an Instruction.
 /// If the metadata is not found then return 0.
 MDNode *MetadataContext::getMD(unsigned MDKind, const Instruction *Inst) {
   MDStoreTy::iterator I = MetadataStore.find(Inst);
@@ -376,7 +376,7 @@
   return NULL;
 }
 
-/// getMDs - Get the metadata attached with an Instruction.
+/// getMDs - Get the metadata attached to an Instruction.
 const MetadataContext::MDMapTy *
 MetadataContext::getMDs(const Instruction *Inst) {
   MDStoreTy::iterator I = MetadataStore.find(Inst);




From david_goodwin at apple.com  Tue Oct 20 17:50:43 2009
From: david_goodwin at apple.com (David Goodwin)
Date: Tue, 20 Oct 2009 22:50:43 -0000
Subject: [llvm-commits] [llvm] r84691 -
	/llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
Message-ID: <200910202250.n9KMohf8023906@zion.cs.uiuc.edu>

Author: david_goodwin
Date: Tue Oct 20 17:50:43 2009
New Revision: 84691

URL: http://llvm.org/viewvc/llvm-project?rev=84691&view=rev
Log:
Respect src register allocation requirements when breaking anti-dependencies. Remove some dead code.

Modified:
    llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp

Modified: llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp?rev=84691&r1=84690&r2=84691&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp (original)
+++ llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp Tue Oct 20 17:50:43 2009
@@ -209,8 +209,7 @@
     void PrescanInstruction(MachineInstr *MI, unsigned Count);
     void ScanInstruction(MachineInstr *MI, unsigned Count);
     bool BreakAntiDependencies(bool CriticalPathOnly);
-    unsigned FindSuitableFreeRegister(unsigned AntiDepReg,
-                                      unsigned LastNewReg);
+    unsigned FindSuitableFreeRegister(unsigned AntiDepReg);
 
     void ReleaseSucc(SUnit *SU, SDep *SuccEdge);
     void ReleaseSuccessors(SUnit *SU);
@@ -595,7 +594,7 @@
     }
   }
 
-  DEBUG(errs() << "\tGroups:");
+  DEBUG(errs() << "\tDef Groups:");
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     MachineOperand &MO = MI->getOperand(i);
     if (!MO.isReg() || !MO.isDef()) continue;
@@ -637,6 +636,8 @@
 
 void SchedulePostRATDList::ScanInstruction(MachineInstr *MI,
                                            unsigned Count) {
+  DEBUG(errs() << "\tUse Groups:");
+
   // Scan the register uses for this instruction and update
   // live-ranges, groups and RegRefs.
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
@@ -645,6 +646,8 @@
     unsigned Reg = MO.getReg();
     if (Reg == 0) continue;
     
+    DEBUG(errs() << " " << TRI->getName(Reg) << "=g" << GetGroup(Reg)); 
+
     // It wasn't previously live but now it is, this is a kill. Forget
     // the previous live-range information and start a new live-range
     // for the register.
@@ -653,6 +656,7 @@
       DefIndices[Reg] = ~0u;
       RegRefs.erase(Reg);
       LeaveGroup(Reg);
+      DEBUG(errs() << "->g" << GetGroup(Reg) << "(last-use)");
     }
     // Repeat, for subregisters.
     for (const unsigned *Subreg = TRI->getSubRegisters(Reg);
@@ -663,9 +667,18 @@
         DefIndices[SubregReg] = ~0u;
         RegRefs.erase(SubregReg);
         LeaveGroup(SubregReg);
+        DEBUG(errs() << "->g" << GetGroup(SubregReg) << "(last-use)");
       }
     }
 
+    // If MI's uses have special allocation requirement, don't allow
+    // any use registers to be changed. Also assume all registers
+    // used in a call must not be changed (ABI).
+    if (MI->getDesc().isCall() || MI->getDesc().hasExtraSrcRegAllocReq()) {
+      DEBUG(if (GetGroup(Reg) != 0) errs() << "->g0(alloc-req)");
+      UnionGroups(Reg, 0);
+    }
+
     // Note register reference...
     const TargetRegisterClass *RC = NULL;
     if (i < MI->getDesc().getNumOperands())
@@ -674,6 +687,8 @@
     RegRefs.insert(std::make_pair(Reg, RR));
   }
   
+  DEBUG(errs() << '\n');
+
   // Form a group of all defs and uses of a KILL instruction to ensure
   // that all registers are renamed as a group.
   if (MI->getOpcode() == TargetInstrInfo::KILL) {
@@ -694,8 +709,7 @@
   }
 }
 
-unsigned SchedulePostRATDList::FindSuitableFreeRegister(unsigned AntiDepReg,
-                                                        unsigned LastNewReg) {
+unsigned SchedulePostRATDList::FindSuitableFreeRegister(unsigned AntiDepReg) {
   // Collect all registers in the same group as AntiDepReg. These all
   // need to be renamed together if we are to break the
   // anti-dependence.
@@ -846,12 +860,6 @@
   std::string dbgStr;
 #endif
 
-  // TODO: If we tracked more than one register here, we could potentially
-  // fix that remaining critical edge too. This is a little more involved,
-  // because unlike the most recent register, less recent registers should
-  // still be considered, though only if no other registers are available.
-  unsigned LastNewReg[TargetRegisterInfo::FirstVirtualRegister] = {};
-
   // Attempt to break anti-dependence edges. Walk the instructions
   // from the bottom up, tracking information about liveness as we go
   // to help determine which registers are available.
@@ -969,12 +977,8 @@
     DEBUG(if (!dbgStr.empty()) errs() << dbgStr << '\n');
 
     // Look for a suitable register to use to break the anti-dependence.
-    //
-    // TODO: Instead of picking the first free register, consider which might
-    // be the best.
     if (AntiDepReg != 0) {
-      if (unsigned NewReg = FindSuitableFreeRegister(AntiDepReg,
-                                                     LastNewReg[AntiDepReg])) {
+      if (unsigned NewReg = FindSuitableFreeRegister(AntiDepReg)) {
         DEBUG(errs() << "\tBreaking anti-dependence edge on "
               << TRI->getName(AntiDepReg)
               << " with " << RegRefs.count(AntiDepReg) << " references"
@@ -1008,7 +1012,6 @@
              "Kill and Def maps aren't consistent for AntiDepReg!");
 
         Changed = true;
-        LastNewReg[AntiDepReg] = NewReg;
         ++NumFixedAnti;
       }
     }




From isanbard at gmail.com  Tue Oct 20 19:06:40 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Wed, 21 Oct 2009 00:06:40 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84699 - in
 /llvm-gcc-4.2/trunk/gcc/config: darwin-c.c t-darwin
Message-ID: <200910210006.n9L06eWV027444@zion.cs.uiuc.edu>

Author: void
Date: Tue Oct 20 19:06:40 2009
New Revision: 84699

URL: http://llvm.org/viewvc/llvm-project?rev=84699&view=rev
Log:
Forgot these files with the recent merge.

Modified:
    llvm-gcc-4.2/trunk/gcc/config/darwin-c.c
    llvm-gcc-4.2/trunk/gcc/config/t-darwin

Modified: llvm-gcc-4.2/trunk/gcc/config/darwin-c.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/darwin-c.c?rev=84699&r1=84698&r2=84699&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/darwin-c.c (original)
+++ llvm-gcc-4.2/trunk/gcc/config/darwin-c.c Tue Oct 20 19:06:40 2009
@@ -1245,4 +1245,3 @@
   return decl;
 }
 /* APPLE LOCAL end radar 2996215 - 6068877 */
-#include "gt-darwin-c.h"

Modified: llvm-gcc-4.2/trunk/gcc/config/t-darwin
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/t-darwin?rev=84699&r1=84698&r2=84699&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/t-darwin (original)
+++ llvm-gcc-4.2/trunk/gcc/config/t-darwin Tue Oct 20 19:06:40 2009
@@ -16,11 +16,10 @@
 
 darwin-c.o: $(srcdir)/config/darwin-c.c $(CONFIG_H) $(SYSTEM_H) coretypes.h \
   $(TM_H) $(CPPLIB_H) $(TREE_H) c-pragma.h $(C_TREE_H) toplev.h $(TM_P_H) \
-  c-incpath.h flags.h $(C_COMMON_H) gt-darwin-c.h
+  c-incpath.h flags.h $(C_COMMON_H)
 	$(CC) -c $(ALL_CFLAGS) $(ALL_CPPFLAGS) $(INCLUDES) $(srcdir)/config/darwin-c.c $(PREPROCESSOR_DEFINES)
 
 gt-darwin.h : s-gtype ; @true
-gt-darwin-c.h : s-gtype ; @true
 
 # APPLE LOCAL begin mainline 2007-06-14 5235474
 darwin-driver.o: $(srcdir)/config/darwin-driver.c \




From asl at math.spbu.ru  Tue Oct 20 19:10:01 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:10:01 -0000
Subject: [llvm-commits] [llvm] r84700 -
	/llvm/trunk/lib/Target/MSP430/MSP430TargetMachine.cpp
Message-ID: <200910210010.n9L0A1Co027646@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:10:00 2009
New Revision: 84700

URL: http://llvm.org/viewvc/llvm-project?rev=84700&view=rev
Log:
Use proper target data

Modified:
    llvm/trunk/lib/Target/MSP430/MSP430TargetMachine.cpp

Modified: llvm/trunk/lib/Target/MSP430/MSP430TargetMachine.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430TargetMachine.cpp?rev=84700&r1=84699&r2=84700&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430TargetMachine.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430TargetMachine.cpp Tue Oct 20 19:10:00 2009
@@ -32,7 +32,7 @@
   LLVMTargetMachine(T, TT),
   Subtarget(TT, FS),
   // FIXME: Check TargetData string.
-  DataLayout("e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8"),
+  DataLayout("e-p:16:16:16-i8:8:8-i16:16:16-i32:16:32"),
   InstrInfo(*this), TLInfo(*this),
   FrameInfo(TargetFrameInfo::StackGrowsDown, 2, -2) { }
 




From asl at math.spbu.ru  Tue Oct 20 19:10:30 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:10:30 -0000
Subject: [llvm-commits] [llvm] r84701 - in
 /llvm/trunk/lib/Target/MSP430/AsmPrinter: CMakeLists.txt
 MSP430InstPrinter.cpp MSP430InstPrinter.h
Message-ID: <200910210010.n9L0AVou027719@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:10:30 2009
New Revision: 84701

URL: http://llvm.org/viewvc/llvm-project?rev=84701&view=rev
Log:
Add MSP430 InstPrinter stub

Added:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h
Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt?rev=84701&r1=84700&r2=84701&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt Tue Oct 20 19:10:30 2009
@@ -1,6 +1,7 @@
 include_directories( ${CMAKE_CURRENT_BINARY_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/.. )
 
 add_llvm_library(LLVMMSP430AsmPrinter
+  MSP430InstPrinter.cpp
   MSP430AsmPrinter.cpp
   )
 add_dependencies(LLVMMSP430AsmPrinter MSP430CodeGenTable_gen)

Added: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp?rev=84701&view=auto

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp (added)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp Tue Oct 20 19:10:30 2009
@@ -0,0 +1,35 @@
+//===-- MSP430InstPrinter.cpp - Convert MSP430 MCInst to assembly syntax --===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class prints an MSP430 MCInst to a .s file.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "asm-printer"
+#include "MSP430InstPrinter.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/FormattedStream.h"
+#include "MSP430GenInstrNames.inc"
+using namespace llvm;
+
+
+// Include the auto-generated portion of the assembly writer.
+#define MachineInstr MCInst
+#define MSP430AsmPrinter MSP430InstPrinter  // FIXME: REMOVE.
+#define NO_ASM_WRITER_BOILERPLATE
+#include "MSP430GenAsmWriter.inc"
+#undef MachineInstr
+#undef MSP430AsmPrinter
+
+void MSP430InstPrinter::printInst(const MCInst *MI) {
+  printInstruction(MI);
+}

Added: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h?rev=84701&view=auto

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h (added)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h Tue Oct 20 19:10:30 2009
@@ -0,0 +1,48 @@
+//===-- MSP430InstPrinter.h - Convert MSP430 MCInst to assembly syntax ----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This class prints a MSP430 MCInst to a .s file.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MSP430INSTPRINTER_H
+#define MSP430INSTPRINTER_H
+
+#include "llvm/MC/MCInstPrinter.h"
+
+namespace llvm 
+{
+
+  class MCOperand;
+
+  class MSP430InstPrinter : public MCInstPrinter {
+  public:
+    MSP430InstPrinter(raw_ostream &O, const MCAsmInfo &MAI) :
+      MCInstPrinter(O, MAI){
+    }
+
+    virtual void printInst(const MCInst *MI);
+
+    // Autogenerated by tblgen.
+    void printInstruction(const MCInst *MI);
+    static const char *getRegisterName(unsigned RegNo);
+
+    void printOperand(const MCInst *MI, unsigned OpNo,
+                      const char *Modifier = 0) {
+    }
+    void printSrcMemOperand(const MCInst *MI, unsigned OpNo,
+                           const char *Modifier = 0) {
+    }
+    void printCCOperand(const MCInst *MI, unsigned OpNo) {
+    }
+
+  };
+}
+
+#endif




From asl at math.spbu.ru  Tue Oct 20 19:10:47 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:10:47 -0000
Subject: [llvm-commits] [llvm] r84702 -
	/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
Message-ID: <200910210010.n9L0AmIC027738@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:10:47 2009
New Revision: 84702

URL: http://llvm.org/viewvc/llvm-project?rev=84702&view=rev
Log:
Wire up MSP430 printMCInst() method

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp?rev=84702&r1=84701&r2=84702&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp Tue Oct 20 19:10:47 2009
@@ -15,6 +15,7 @@
 #define DEBUG_TYPE "asm-printer"
 #include "MSP430.h"
 #include "MSP430InstrInfo.h"
+#include "MSP430InstPrinter.h"
 #include "MSP430MCAsmInfo.h"
 #include "MSP430TargetMachine.h"
 #include "llvm/Constants.h"
@@ -52,6 +53,9 @@
       return "MSP430 Assembly Printer";
     }
 
+    void printMCInst(const MCInst *MI) {
+      MSP430InstPrinter(O, *MAI).printInstruction(MI);
+    }
     void printOperand(const MachineInstr *MI, int OpNum,
                       const char* Modifier = 0);
     void printSrcMemOperand(const MachineInstr *MI, int OpNum,




From asl at math.spbu.ru  Tue Oct 20 19:11:09 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:11:09 -0000
Subject: [llvm-commits] [llvm] r84703 - in
 /llvm/trunk/lib/Target/MSP430/AsmPrinter: CMakeLists.txt
 MSP430AsmPrinter.cpp MSP430MCInstLower.cpp MSP430MCInstLower.h
Message-ID: <200910210011.n9L0B9a3027786@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:11:08 2009
New Revision: 84703

URL: http://llvm.org/viewvc/llvm-project?rev=84703&view=rev
Log:
Add experimental MSP430 MCInstLowering stuff

Added:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h
Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt?rev=84703&r1=84702&r2=84703&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/CMakeLists.txt Tue Oct 20 19:11:08 2009
@@ -3,5 +3,6 @@
 add_llvm_library(LLVMMSP430AsmPrinter
   MSP430InstPrinter.cpp
   MSP430AsmPrinter.cpp
+  MSP430MCInstLower.cpp
   )
 add_dependencies(LLVMMSP430AsmPrinter MSP430CodeGenTable_gen)

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp?rev=84703&r1=84702&r2=84703&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp Tue Oct 20 19:11:08 2009
@@ -17,6 +17,7 @@
 #include "MSP430InstrInfo.h"
 #include "MSP430InstPrinter.h"
 #include "MSP430MCAsmInfo.h"
+#include "MSP430MCInstLower.h"
 #include "MSP430TargetMachine.h"
 #include "llvm/Constants.h"
 #include "llvm/DerivedTypes.h"
@@ -27,12 +28,14 @@
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
 #include "llvm/Target/TargetData.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
 #include "llvm/Target/TargetRegistry.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/FormattedStream.h"
 #include "llvm/Support/Mangler.h"
@@ -42,6 +45,10 @@
 
 STATISTIC(EmittedInsts, "Number of machine instrs printed");
 
+static cl::opt<bool>
+EnableMCInst("enable-msp430-mcinst-printer", cl::Hidden,
+             cl::desc("enable experimental mcinst gunk in the msp430 backend"));
+
 namespace {
   class VISIBILITY_HIDDEN MSP430AsmPrinter : public AsmPrinter {
   public:
@@ -71,6 +78,7 @@
     bool PrintAsmMemoryOperand(const MachineInstr *MI,
                                unsigned OpNo, unsigned AsmVariant,
                                const char *ExtraCode);
+    void printInstructionThroughMCStreamer(const MachineInstr *MI);
 
     void emitFunctionHeader(const MachineFunction &MF);
     bool runOnMachineFunction(MachineFunction &F);
@@ -152,7 +160,11 @@
   processDebugLoc(MI, true);
 
   // Call the autogenerated instruction printer routines.
-  printInstruction(MI);
+  if (EnableMCInst) {
+    printInstructionThroughMCStreamer(MI);
+  } else {
+    printInstruction(MI);
+  }
 
   if (VerboseAsm && !MI->getDebugLoc().isUnknown())
     EmitComments(*MI);
@@ -279,6 +291,36 @@
   return false;
 }
 
+//===----------------------------------------------------------------------===//
+void MSP430AsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI)
+{
+
+  MSP430MCInstLower MCInstLowering(OutContext, Mang);
+
+  switch (MI->getOpcode()) {
+  case TargetInstrInfo::DBG_LABEL:
+  case TargetInstrInfo::EH_LABEL:
+  case TargetInstrInfo::GC_LABEL:
+    printLabel(MI);
+    return;
+  case TargetInstrInfo::KILL:
+    return;
+  case TargetInstrInfo::INLINEASM:
+    O << '\t';
+    printInlineAsm(MI);
+    return;
+  case TargetInstrInfo::IMPLICIT_DEF:
+    printImplicitDef(MI);
+    return;
+  default: break;
+  }
+
+  MCInst TmpInst;
+  MCInstLowering.Lower(MI, TmpInst);
+
+  printMCInst(&TmpInst);
+}
+
 // Force static initialization.
 extern "C" void LLVMInitializeMSP430AsmPrinter() {
   RegisterAsmPrinter<MSP430AsmPrinter> X(TheMSP430Target);

Added: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp?rev=84703&view=auto

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp (added)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp Tue Oct 20 19:11:08 2009
@@ -0,0 +1,66 @@
+//===-- MSP430MCInstLower.cpp - Convert MSP430 MachineInstr to an MCInst---===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains code to lower MSP430 MachineInstrs to their corresponding
+// MCInst records.
+//
+//===----------------------------------------------------------------------===//
+
+#include "MSP430MCInstLower.h"
+#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCExpr.h"
+#include "llvm/MC/MCInst.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/Mangler.h"
+#include "llvm/ADT/SmallString.h"
+using namespace llvm;
+
+void MSP430MCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
+  OutMI.setOpcode(MI->getOpcode());
+
+  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+    const MachineOperand &MO = MI->getOperand(i);
+
+    MCOperand MCOp;
+    switch (MO.getType()) {
+    default:
+      MI->dump();
+      assert(0 && "unknown operand type");
+    case MachineOperand::MO_Register:
+      MCOp = MCOperand::CreateReg(MO.getReg());
+      break;
+    case MachineOperand::MO_Immediate:
+      MCOp = MCOperand::CreateImm(MO.getImm());
+      break;
+#if 0
+    case MachineOperand::MO_MachineBasicBlock:
+      MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
+                       AsmPrinter.GetMBBSymbol(MO.getMBB()->getNumber()), Ctx));
+      break;
+    case MachineOperand::MO_GlobalAddress:
+      MCOp = LowerSymbolOperand(MO, GetGlobalAddressSymbol(MO));
+      break;
+    case MachineOperand::MO_ExternalSymbol:
+      MCOp = LowerSymbolOperand(MO, GetExternalSymbolSymbol(MO));
+      break;
+    case MachineOperand::MO_JumpTableIndex:
+      MCOp = LowerSymbolOperand(MO, GetJumpTableSymbol(MO));
+      break;
+    case MachineOperand::MO_ConstantPoolIndex:
+      MCOp = LowerSymbolOperand(MO, GetConstantPoolIndexSymbol(MO));
+      break;
+#endif
+    }
+
+    OutMI.addOperand(MCOp);
+  }
+
+}

Added: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h?rev=84703&view=auto

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h (added)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h Tue Oct 20 19:11:08 2009
@@ -0,0 +1,45 @@
+//===-- MSP430MCInstLower.h - Lower MachineInstr to MCInst ----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MSP430_MCINSTLOWER_H
+#define MSP430_MCINSTLOWER_H
+
+#include "llvm/Support/Compiler.h"
+
+namespace llvm {
+  class MCAsmInfo;
+  class MCContext;
+  class MCInst;
+  class MCOperand;
+  class MCSymbol;
+  class MachineInstr;
+  class MachineModuleInfoMachO;
+  class MachineOperand;
+  class Mangler;
+
+  /// MSP430MCInstLower - This class is used to lower an MachineInstr
+  /// into an MCInst.
+class VISIBILITY_HIDDEN MSP430MCInstLower {
+  MCContext &Ctx;
+  Mangler *Mang;
+
+  #if 0
+  const unsigned CurFunctionNumber;
+  const MCAsmInfo &MAI;
+  #endif
+
+public:
+  MSP430MCInstLower(MCContext &ctx, Mangler *mang) : Ctx(ctx), Mang(mang) {}
+
+  void Lower(const MachineInstr *MI, MCInst &OutMI) const;
+};
+
+}
+
+#endif




From asl at math.spbu.ru  Tue Oct 20 19:11:27 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:11:27 -0000
Subject: [llvm-commits] [llvm] r84704 - in
 /llvm/trunk/lib/Target/MSP430/AsmPrinter: MSP430InstPrinter.cpp
 MSP430InstPrinter.h
Message-ID: <200910210011.n9L0BReY027813@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:11:27 2009
New Revision: 84704

URL: http://llvm.org/viewvc/llvm-project?rev=84704&view=rev
Log:
Add simple operand printing stuff

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp?rev=84704&r1=84703&r2=84704&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp Tue Oct 20 19:11:27 2009
@@ -33,3 +33,18 @@
 void MSP430InstPrinter::printInst(const MCInst *MI) {
   printInstruction(MI);
 }
+
+void MSP430InstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
+                                     const char *Modifier) {
+  assert((Modifier == 0 || Modifier[0] == 0) && "Cannot print modifiers");
+
+  const MCOperand &Op = MI->getOperand(OpNo);
+  if (Op.isReg()) {
+    O << getRegisterName(Op.getReg());
+  } else if (Op.isImm()) {
+    O << '#' << Op.getImm();
+  } else {
+    assert(Op.isExpr() && "unknown operand kind in printOperand");
+    assert(0 && "Unimplemented!");
+  }
+}

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h?rev=84704&r1=84703&r2=84704&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h Tue Oct 20 19:11:27 2009
@@ -16,7 +16,7 @@
 
 #include "llvm/MC/MCInstPrinter.h"
 
-namespace llvm 
+namespace llvm
 {
 
   class MCOperand;
@@ -34,10 +34,10 @@
     static const char *getRegisterName(unsigned RegNo);
 
     void printOperand(const MCInst *MI, unsigned OpNo,
-                      const char *Modifier = 0) {
-    }
+                      const char *Modifier = 0);
+
     void printSrcMemOperand(const MCInst *MI, unsigned OpNo,
-                           const char *Modifier = 0) {
+                            const char *Modifier = 0) {
     }
     void printCCOperand(const MCInst *MI, unsigned OpNo) {
     }




From asl at math.spbu.ru  Tue Oct 20 19:11:44 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:11:44 -0000
Subject: [llvm-commits] [llvm] r84705 - in /llvm/trunk/test/CodeGen/MSP430:
	Inst16ri.ll Inst8ri.ll
Message-ID: <200910210011.n9L0BiPg027839@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:11:44 2009
New Revision: 84705

URL: http://llvm.org/viewvc/llvm-project?rev=84705&view=rev
Log:
Add reg-imm tests

Added:
    llvm/trunk/test/CodeGen/MSP430/Inst16ri.ll
    llvm/trunk/test/CodeGen/MSP430/Inst8ri.ll

Added: llvm/trunk/test/CodeGen/MSP430/Inst16ri.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/MSP430/Inst16ri.ll?rev=84705&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/MSP430/Inst16ri.ll (added)
+++ llvm/trunk/test/CodeGen/MSP430/Inst16ri.ll Tue Oct 20 19:11:44 2009
@@ -0,0 +1,37 @@
+; RUN: llc -march=msp430 < %s | FileCheck %s
+target datalayout = "e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8"
+target triple = "msp430-generic-generic"
+
+define i16 @mov() nounwind {
+; CHECK: mov:
+; CHECK: mov.w	#1, r15
+	ret i16 1
+}
+
+define i16 @add(i16 %a, i16 %b) nounwind {
+; CHECK: add:
+; CHECK: add.w	#1, r15
+	%1 = add i16 %a, 1
+	ret i16 %1
+}
+
+define i16 @and(i16 %a, i16 %b) nounwind {
+; CHECK: and:
+; CHECK: and.w	#1, r15
+	%1 = and i16 %a, 1
+	ret i16 %1
+}
+
+define i16 @bis(i16 %a, i16 %b) nounwind {
+; CHECK: bis:
+; CHECK: bis.w	#1, r15
+	%1 = or i16 %a, 1
+	ret i16 %1
+}
+
+define i16 @xor(i16 %a, i16 %b) nounwind {
+; CHECK: xor:
+; CHECK: xor.w	#1, r15
+	%1 = xor i16 %a, 1
+	ret i16 %1
+}

Added: llvm/trunk/test/CodeGen/MSP430/Inst8ri.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/MSP430/Inst8ri.ll?rev=84705&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/MSP430/Inst8ri.ll (added)
+++ llvm/trunk/test/CodeGen/MSP430/Inst8ri.ll Tue Oct 20 19:11:44 2009
@@ -0,0 +1,37 @@
+; RUN: llc -march=msp430 < %s | FileCheck %s
+target datalayout = "e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8"
+target triple = "msp430-generic-generic"
+
+define i8 @mov() nounwind {
+; CHECK: mov:
+; CHECK: mov.b	#1, r15
+	ret i8 1
+}
+
+define i8 @add(i8 %a, i8 %b) nounwind {
+; CHECK: add:
+; CHECK: add.b	#1, r15
+	%1 = add i8 %a, 1
+	ret i8 %1
+}
+
+define i8 @and(i8 %a, i8 %b) nounwind {
+; CHECK: and:
+; CHECK: and.b	#1, r15
+	%1 = and i8 %a, 1
+	ret i8 %1
+}
+
+define i8 @bis(i8 %a, i8 %b) nounwind {
+; CHECK: bis:
+; CHECK: bis.b	#1, r15
+	%1 = or i8 %a, 1
+	ret i8 %1
+}
+
+define i8 @xor(i8 %a, i8 %b) nounwind {
+; CHECK: xor:
+; CHECK: xor.b	#1, r15
+	%1 = xor i8 %a, 1
+	ret i8 %1
+}




From asl at math.spbu.ru  Tue Oct 20 19:12:08 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:12:08 -0000
Subject: [llvm-commits] [llvm] r84706 - in
 /llvm/trunk/lib/Target/MSP430/AsmPrinter: MSP430AsmPrinter.cpp
 MSP430InstPrinter.cpp MSP430InstPrinter.h MSP430MCInstLower.cpp
 MSP430MCInstLower.h
Message-ID: <200910210012.n9L0C9jU027881@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:12:08 2009
New Revision: 84706

URL: http://llvm.org/viewvc/llvm-project?rev=84706&view=rev
Log:
Checkpoint MCInst printer. We (almostly) able to print global / JT / constpool entries

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp?rev=84706&r1=84705&r2=84706&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp Tue Oct 20 19:12:08 2009
@@ -295,7 +295,7 @@
 void MSP430AsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI)
 {
 
-  MSP430MCInstLower MCInstLowering(OutContext, Mang);
+  MSP430MCInstLower MCInstLowering(OutContext, *Mang, getFunctionNumber(), *MAI);
 
   switch (MI->getOpcode()) {
   case TargetInstrInfo::DBG_LABEL:

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp?rev=84706&r1=84705&r2=84706&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp Tue Oct 20 19:12:08 2009
@@ -45,6 +45,30 @@
     O << '#' << Op.getImm();
   } else {
     assert(Op.isExpr() && "unknown operand kind in printOperand");
-    assert(0 && "Unimplemented!");
+    Op.getExpr()->print(O, &MAI);
   }
 }
+
+void MSP430InstPrinter::printSrcMemOperand(const MCInst *MI, unsigned OpNo,
+                                           const char *Modifier) {
+  const MCOperand &Base = MI->getOperand(OpNo);
+  const MCOperand &Disp = MI->getOperand(OpNo+1);
+
+  if (Disp.isImm() && !Base.isReg())
+    printOperand(MI, OpNo);
+  else if (Base.isReg()) {
+    if (Disp.getImm()) {
+      O << Disp.getImm() << '(';
+      printOperand(MI, OpNo);
+      O << ')';
+    } else {
+      O << '@';
+      printOperand(MI, OpNo);
+    }
+  } else {
+    Base.dump();
+    Disp.dump();
+    llvm_unreachable("Unsupported memory operand");
+  }
+
+}

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h?rev=84706&r1=84705&r2=84706&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h Tue Oct 20 19:12:08 2009
@@ -37,8 +37,8 @@
                       const char *Modifier = 0);
 
     void printSrcMemOperand(const MCInst *MI, unsigned OpNo,
-                            const char *Modifier = 0) {
-    }
+                            const char *Modifier = 0);
+
     void printCCOperand(const MCInst *MI, unsigned OpNo) {
     }
 

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp?rev=84706&r1=84705&r2=84706&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp Tue Oct 20 19:12:08 2009
@@ -19,10 +19,75 @@
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/Mangler.h"
 #include "llvm/ADT/SmallString.h"
 using namespace llvm;
 
+MCSymbol *MSP430MCInstLower::
+GetGlobalAddressSymbol(const MachineOperand &MO) const {
+  const GlobalValue *GV = MO.getGlobal();
+
+  SmallString<128> Name;
+  Mang.getNameWithPrefix(Name, GV, false);
+
+  switch (MO.getTargetFlags()) {
+  default: llvm_unreachable(0 && "Unknown target flag on GV operand");
+  case 0: break;
+  }
+
+  return Ctx.GetOrCreateSymbol(Name.str());
+}
+
+
+MCSymbol *MSP430MCInstLower::
+GetJumpTableSymbol(const MachineOperand &MO) const {
+  SmallString<256> Name;
+  raw_svector_ostream(Name) << MAI.getPrivateGlobalPrefix() << "JTI"
+    << CurFunctionNumber << '_' << MO.getIndex();
+
+  switch (MO.getTargetFlags()) {
+  default: llvm_unreachable("Unknown target flag on GV operand");
+  case 0: break;
+  }
+
+  // Create a symbol for the name.
+  return Ctx.GetOrCreateSymbol(Name.str());
+}
+
+MCSymbol *MSP430MCInstLower::
+GetConstantPoolIndexSymbol(const MachineOperand &MO) const {
+  SmallString<256> Name;
+  raw_svector_ostream(Name) << MAI.getPrivateGlobalPrefix() << "CPI"
+    << CurFunctionNumber << '_' << MO.getIndex();
+
+  switch (MO.getTargetFlags()) {
+  default: llvm_unreachable("Unknown target flag on GV operand");
+  case 0: break;
+  }
+
+  // Create a symbol for the name.
+  return Ctx.GetOrCreateSymbol(Name.str());
+}
+
+MCOperand MSP430MCInstLower::
+LowerSymbolOperand(const MachineOperand &MO, MCSymbol *Sym) const {
+  // FIXME: We would like an efficient form for this, so we don't have to do a
+  // lot of extra uniquing.
+  const MCExpr *Expr = MCSymbolRefExpr::Create(Sym, Ctx);
+
+  switch (MO.getTargetFlags()) {
+  default: llvm_unreachable("Unknown target flag on GV operand");
+  case 0: break;
+  }
+
+  if (!MO.isJTI() && MO.getOffset())
+    Expr = MCBinaryExpr::CreateAdd(Expr,
+                                   MCConstantExpr::Create(MO.getOffset(), Ctx),
+                                   Ctx);
+  return MCOperand::CreateExpr(Expr);
+}
+
 void MSP430MCInstLower::Lower(const MachineInstr *MI, MCInst &OutMI) const {
   OutMI.setOpcode(MI->getOpcode());
 
@@ -45,22 +110,23 @@
       MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
                        AsmPrinter.GetMBBSymbol(MO.getMBB()->getNumber()), Ctx));
       break;
+#endif
     case MachineOperand::MO_GlobalAddress:
       MCOp = LowerSymbolOperand(MO, GetGlobalAddressSymbol(MO));
       break;
+#if 0
     case MachineOperand::MO_ExternalSymbol:
       MCOp = LowerSymbolOperand(MO, GetExternalSymbolSymbol(MO));
       break;
+#endif
     case MachineOperand::MO_JumpTableIndex:
       MCOp = LowerSymbolOperand(MO, GetJumpTableSymbol(MO));
       break;
     case MachineOperand::MO_ConstantPoolIndex:
       MCOp = LowerSymbolOperand(MO, GetConstantPoolIndexSymbol(MO));
       break;
-#endif
     }
 
     OutMI.addOperand(MCOp);
   }
-
 }

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h?rev=84706&r1=84705&r2=84706&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h Tue Oct 20 19:12:08 2009
@@ -27,17 +27,22 @@
   /// into an MCInst.
 class VISIBILITY_HIDDEN MSP430MCInstLower {
   MCContext &Ctx;
-  Mangler *Mang;
+  Mangler &Mang;
 
-  #if 0
   const unsigned CurFunctionNumber;
   const MCAsmInfo &MAI;
-  #endif
 
 public:
-  MSP430MCInstLower(MCContext &ctx, Mangler *mang) : Ctx(ctx), Mang(mang) {}
-
+  MSP430MCInstLower(MCContext &ctx, Mangler &mang, unsigned FuncNum,
+                    const MCAsmInfo &mai)
+    : Ctx(ctx), Mang(mang), CurFunctionNumber(FuncNum), MAI(mai) {}
   void Lower(const MachineInstr *MI, MCInst &OutMI) const;
+
+  MCOperand LowerSymbolOperand(const MachineOperand &MO, MCSymbol *Sym) const;
+
+  MCSymbol *GetGlobalAddressSymbol(const MachineOperand &MO) const;
+  MCSymbol *GetJumpTableSymbol(const MachineOperand &MO) const;
+  MCSymbol *GetConstantPoolIndexSymbol(const MachineOperand &MO) const;
 };
 
 }




From asl at math.spbu.ru  Tue Oct 20 19:12:27 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:12:27 -0000
Subject: [llvm-commits] [llvm] r84707 -
 /llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
Message-ID: <200910210012.n9L0CRLk027957@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:12:27 2009
New Revision: 84707

URL: http://llvm.org/viewvc/llvm-project?rev=84707&view=rev
Log:
Add a workaround for different memops prefixes

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp?rev=84707&r1=84706&r2=84707&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp Tue Oct 20 19:12:27 2009
@@ -36,8 +36,6 @@
 
 void MSP430InstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
                                      const char *Modifier) {
-  assert((Modifier == 0 || Modifier[0] == 0) && "Cannot print modifiers");
-
   const MCOperand &Op = MI->getOperand(OpNo);
   if (Op.isReg()) {
     O << getRegisterName(Op.getReg());
@@ -45,6 +43,8 @@
     O << '#' << Op.getImm();
   } else {
     assert(Op.isExpr() && "unknown operand kind in printOperand");
+    bool isMemOp  = Modifier && !strcmp(Modifier, "mem");
+    O << (isMemOp ? '&' : '#');
     Op.getExpr()->print(O, &MAI);
   }
 }
@@ -54,7 +54,10 @@
   const MCOperand &Base = MI->getOperand(OpNo);
   const MCOperand &Disp = MI->getOperand(OpNo+1);
 
-  if (Disp.isImm() && !Base.isReg())
+  // FIXME: move global to displacement field!
+  if (Base.isExpr())
+    printOperand(MI, OpNo, "mem");
+  else if (Disp.isImm() && !Base.isReg())
     printOperand(MI, OpNo);
   else if (Base.isReg()) {
     if (Disp.getImm()) {




From asl at math.spbu.ru  Tue Oct 20 19:12:44 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:12:44 -0000
Subject: [llvm-commits] [llvm] r84708 -
 /llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
Message-ID: <200910210012.n9L0Ci2I027975@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:12:44 2009
New Revision: 84708

URL: http://llvm.org/viewvc/llvm-project?rev=84708&view=rev
Log:
Ignore all implicit reg operands

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp?rev=84708&r1=84707&r2=84708&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp Tue Oct 20 19:12:44 2009
@@ -100,6 +100,8 @@
       MI->dump();
       assert(0 && "unknown operand type");
     case MachineOperand::MO_Register:
+      // Ignore all implicit register operands.
+      if (MO.isImplicit()) continue;
       MCOp = MCOperand::CreateReg(MO.getReg());
       break;
     case MachineOperand::MO_Immediate:




From asl at math.spbu.ru  Tue Oct 20 19:13:05 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:13:05 -0000
Subject: [llvm-commits] [llvm] r84709 - in
 /llvm/trunk/lib/Target/MSP430/AsmPrinter: MSP430AsmPrinter.cpp
 MSP430InstPrinter.cpp MSP430InstPrinter.h MSP430MCInstLower.cpp
 MSP430MCInstLower.h
Message-ID: <200910210013.n9L0D558027997@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:13:05 2009
New Revision: 84709

URL: http://llvm.org/viewvc/llvm-project?rev=84709&view=rev
Log:
Add basic block operands & jump kinds

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp?rev=84709&r1=84708&r2=84709&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp Tue Oct 20 19:13:05 2009
@@ -295,7 +295,7 @@
 void MSP430AsmPrinter::printInstructionThroughMCStreamer(const MachineInstr *MI)
 {
 
-  MSP430MCInstLower MCInstLowering(OutContext, *Mang, getFunctionNumber(), *MAI);
+  MSP430MCInstLower MCInstLowering(OutContext, *Mang, *this);
 
   switch (MI->getOpcode()) {
   case TargetInstrInfo::DBG_LABEL:

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp?rev=84709&r1=84708&r2=84709&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp Tue Oct 20 19:13:05 2009
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "asm-printer"
+#include "MSP430InstrInfo.h"
 #include "MSP430InstPrinter.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCAsmInfo.h"
@@ -75,3 +76,30 @@
   }
 
 }
+void MSP430InstPrinter::printCCOperand(const MCInst *MI, unsigned OpNo) {
+  unsigned CC = MI->getOperand(OpNo).getImm();
+
+  switch (CC) {
+  default:
+   llvm_unreachable("Unsupported CC code");
+   break;
+  case MSP430::COND_E:
+   O << "eq";
+   break;
+  case MSP430::COND_NE:
+   O << "ne";
+   break;
+  case MSP430::COND_HS:
+   O << "hs";
+   break;
+  case MSP430::COND_LO:
+   O << "lo";
+   break;
+  case MSP430::COND_GE:
+   O << "ge";
+   break;
+  case MSP430::COND_L:
+   O << 'l';
+   break;
+  }
+}

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h?rev=84709&r1=84708&r2=84709&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h Tue Oct 20 19:13:05 2009
@@ -39,8 +39,7 @@
     void printSrcMemOperand(const MCInst *MI, unsigned OpNo,
                             const char *Modifier = 0);
 
-    void printCCOperand(const MCInst *MI, unsigned OpNo) {
-    }
+    void printCCOperand(const MCInst *MI, unsigned OpNo);
 
   };
 }

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp?rev=84709&r1=84708&r2=84709&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp Tue Oct 20 19:13:05 2009
@@ -13,6 +13,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "MSP430MCInstLower.h"
+#include "llvm/CodeGen/AsmPrinter.h"
+#include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCContext.h"
@@ -43,8 +45,9 @@
 MCSymbol *MSP430MCInstLower::
 GetJumpTableSymbol(const MachineOperand &MO) const {
   SmallString<256> Name;
-  raw_svector_ostream(Name) << MAI.getPrivateGlobalPrefix() << "JTI"
-    << CurFunctionNumber << '_' << MO.getIndex();
+  raw_svector_ostream(Name) << Printer.MAI->getPrivateGlobalPrefix() << "JTI"
+                            << Printer.getFunctionNumber() << '_'
+                            << MO.getIndex();
 
   switch (MO.getTargetFlags()) {
   default: llvm_unreachable("Unknown target flag on GV operand");
@@ -58,8 +61,9 @@
 MCSymbol *MSP430MCInstLower::
 GetConstantPoolIndexSymbol(const MachineOperand &MO) const {
   SmallString<256> Name;
-  raw_svector_ostream(Name) << MAI.getPrivateGlobalPrefix() << "CPI"
-    << CurFunctionNumber << '_' << MO.getIndex();
+  raw_svector_ostream(Name) << Printer.MAI->getPrivateGlobalPrefix() << "CPI"
+                            << Printer.getFunctionNumber() << '_'
+                            << MO.getIndex();
 
   switch (MO.getTargetFlags()) {
   default: llvm_unreachable("Unknown target flag on GV operand");
@@ -107,12 +111,10 @@
     case MachineOperand::MO_Immediate:
       MCOp = MCOperand::CreateImm(MO.getImm());
       break;
-#if 0
     case MachineOperand::MO_MachineBasicBlock:
       MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
-                       AsmPrinter.GetMBBSymbol(MO.getMBB()->getNumber()), Ctx));
+                         Printer.GetMBBSymbol(MO.getMBB()->getNumber()), Ctx));
       break;
-#endif
     case MachineOperand::MO_GlobalAddress:
       MCOp = LowerSymbolOperand(MO, GetGlobalAddressSymbol(MO));
       break;

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h?rev=84709&r1=84708&r2=84709&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h Tue Oct 20 19:13:05 2009
@@ -13,6 +13,7 @@
 #include "llvm/Support/Compiler.h"
 
 namespace llvm {
+  class AsmPrinter;
   class MCAsmInfo;
   class MCContext;
   class MCInst;
@@ -29,13 +30,10 @@
   MCContext &Ctx;
   Mangler &Mang;
 
-  const unsigned CurFunctionNumber;
-  const MCAsmInfo &MAI;
-
+  AsmPrinter &Printer;
 public:
-  MSP430MCInstLower(MCContext &ctx, Mangler &mang, unsigned FuncNum,
-                    const MCAsmInfo &mai)
-    : Ctx(ctx), Mang(mang), CurFunctionNumber(FuncNum), MAI(mai) {}
+  MSP430MCInstLower(MCContext &ctx, Mangler &mang, AsmPrinter &printer)
+    : Ctx(ctx), Mang(mang), Printer(printer) {}
   void Lower(const MachineInstr *MI, MCInst &OutMI) const;
 
   MCOperand LowerSymbolOperand(const MachineOperand &MO, MCSymbol *Sym) const;




From asl at math.spbu.ru  Tue Oct 20 19:13:25 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:13:25 -0000
Subject: [llvm-commits] [llvm] r84710 - in /llvm/trunk/lib/Target/MSP430:
 AsmPrinter/MSP430AsmPrinter.cpp AsmPrinter/MSP430InstPrinter.cpp
 AsmPrinter/MSP430InstPrinter.h MSP430InstrInfo.td
Message-ID: <200910210013.n9L0DPAq028027@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:13:25 2009
New Revision: 84710

URL: http://llvm.org/viewvc/llvm-project?rev=84710&view=rev
Log:
Distinguish between pcrel imm operands and 'normal' ones. Fix fixes gross weirdness of asmprinting.

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h
    llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.td

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp?rev=84710&r1=84709&r2=84710&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp Tue Oct 20 19:13:25 2009
@@ -65,6 +65,9 @@
     }
     void printOperand(const MachineInstr *MI, int OpNum,
                       const char* Modifier = 0);
+    void printPCRelImmOperand(const MachineInstr *MI, int OpNum) {
+      printOperand(MI, OpNum);
+    }
     void printSrcMemOperand(const MachineInstr *MI, int OpNum,
                             const char* Modifier = 0);
     void printCCOperand(const MachineInstr *MI, int OpNum);

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp?rev=84710&r1=84709&r2=84710&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp Tue Oct 20 19:13:25 2009
@@ -35,6 +35,16 @@
   printInstruction(MI);
 }
 
+void MSP430InstPrinter::printPCRelImmOperand(const MCInst *MI, unsigned OpNo) {
+  const MCOperand &Op = MI->getOperand(OpNo);
+  if (Op.isImm())
+    O << Op.getImm();
+  else {
+    assert(Op.isExpr() && "unknown pcrel immediate operand");
+    Op.getExpr()->print(O, &MAI);
+  }
+}
+
 void MSP430InstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
                                      const char *Modifier) {
   const MCOperand &Op = MI->getOperand(OpNo);
@@ -44,8 +54,7 @@
     O << '#' << Op.getImm();
   } else {
     assert(Op.isExpr() && "unknown operand kind in printOperand");
-    bool isMemOp  = Modifier && !strcmp(Modifier, "mem");
-    O << (isMemOp ? '&' : '#');
+    O << '#';
     Op.getExpr()->print(O, &MAI);
   }
 }
@@ -56,9 +65,10 @@
   const MCOperand &Disp = MI->getOperand(OpNo+1);
 
   // FIXME: move global to displacement field!
-  if (Base.isExpr())
-    printOperand(MI, OpNo, "mem");
-  else if (Disp.isImm() && !Base.isReg())
+  if (Base.isExpr()) {
+    O << '&';
+    Base.getExpr()->print(O, &MAI);
+  } else if (Disp.isImm() && !Base.isReg())
     printOperand(MI, OpNo);
   else if (Base.isReg()) {
     if (Disp.getImm()) {
@@ -74,8 +84,8 @@
     Disp.dump();
     llvm_unreachable("Unsupported memory operand");
   }
-
 }
+
 void MSP430InstPrinter::printCCOperand(const MCInst *MI, unsigned OpNo) {
   unsigned CC = MI->getOperand(OpNo).getImm();
 

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h?rev=84710&r1=84709&r2=84710&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.h Tue Oct 20 19:13:25 2009
@@ -35,10 +35,9 @@
 
     void printOperand(const MCInst *MI, unsigned OpNo,
                       const char *Modifier = 0);
-
+    void printPCRelImmOperand(const MCInst *MI, unsigned OpNo);
     void printSrcMemOperand(const MCInst *MI, unsigned OpNo,
                             const char *Modifier = 0);
-
     void printCCOperand(const MCInst *MI, unsigned OpNo);
 
   };

Modified: llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.td?rev=84710&r1=84709&r2=84710&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.td (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.td Tue Oct 20 19:13:25 2009
@@ -71,7 +71,9 @@
 }
 
 // Branch targets have OtherVT type.
-def brtarget : Operand<OtherVT>;
+def brtarget : Operand<OtherVT> {
+  let PrintMethod = "printPCRelImmOperand";
+}
 
 // Operand for printing out a condition code.
 def cc : Operand<i8> {




From asl at math.spbu.ru  Tue Oct 20 19:13:42 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:13:42 -0000
Subject: [llvm-commits] [llvm] r84711 - in
 /llvm/trunk/lib/Target/MSP430/AsmPrinter: MSP430MCInstLower.cpp
 MSP430MCInstLower.h
Message-ID: <200910210013.n9L0Dh71028050@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:13:42 2009
New Revision: 84711

URL: http://llvm.org/viewvc/llvm-project?rev=84711&view=rev
Log:
Handle external symbols

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp?rev=84711&r1=84710&r2=84711&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.cpp Tue Oct 20 19:13:42 2009
@@ -41,6 +41,19 @@
   return Ctx.GetOrCreateSymbol(Name.str());
 }
 
+MCSymbol *MSP430MCInstLower::
+GetExternalSymbolSymbol(const MachineOperand &MO) const {
+  SmallString<128> Name;
+  Name += Printer.MAI->getGlobalPrefix();
+  Name += MO.getSymbolName();
+
+  switch (MO.getTargetFlags()) {
+  default: assert(0 && "Unknown target flag on GV operand");
+  case 0: break;
+  }
+
+  return Ctx.GetOrCreateSymbol(Name.str());
+}
 
 MCSymbol *MSP430MCInstLower::
 GetJumpTableSymbol(const MachineOperand &MO) const {
@@ -118,11 +131,9 @@
     case MachineOperand::MO_GlobalAddress:
       MCOp = LowerSymbolOperand(MO, GetGlobalAddressSymbol(MO));
       break;
-#if 0
     case MachineOperand::MO_ExternalSymbol:
       MCOp = LowerSymbolOperand(MO, GetExternalSymbolSymbol(MO));
       break;
-#endif
     case MachineOperand::MO_JumpTableIndex:
       MCOp = LowerSymbolOperand(MO, GetJumpTableSymbol(MO));
       break;

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h?rev=84711&r1=84710&r2=84711&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430MCInstLower.h Tue Oct 20 19:13:42 2009
@@ -39,6 +39,7 @@
   MCOperand LowerSymbolOperand(const MachineOperand &MO, MCSymbol *Sym) const;
 
   MCSymbol *GetGlobalAddressSymbol(const MachineOperand &MO) const;
+  MCSymbol *GetExternalSymbolSymbol(const MachineOperand &MO) const;
   MCSymbol *GetJumpTableSymbol(const MachineOperand &MO) const;
   MCSymbol *GetConstantPoolIndexSymbol(const MachineOperand &MO) const;
 };




From asl at math.spbu.ru  Tue Oct 20 19:13:59 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:13:59 -0000
Subject: [llvm-commits] [llvm] r84712 -
 /llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
Message-ID: <200910210013.n9L0Dxer028068@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:13:58 2009
New Revision: 84712

URL: http://llvm.org/viewvc/llvm-project?rev=84712&view=rev
Log:
Be crazy and assert in case of unsupported modifier passed.

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp?rev=84712&r1=84711&r2=84712&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp Tue Oct 20 19:13:58 2009
@@ -47,6 +47,7 @@
 
 void MSP430InstPrinter::printOperand(const MCInst *MI, unsigned OpNo,
                                      const char *Modifier) {
+  assert((Modifier == 0 || Modifier[0] == 0) && "No modifiers supported");
   const MCOperand &Op = MI->getOperand(OpNo);
   if (Op.isReg()) {
     O << getRegisterName(Op.getReg());




From asl at math.spbu.ru  Tue Oct 20 19:14:15 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 00:14:15 -0000
Subject: [llvm-commits] [llvm] r84713 -
	/llvm/trunk/lib/Target/MSP430/README.txt
Message-ID: <200910210014.n9L0EFk7028096@zion.cs.uiuc.edu>

Author: asl
Date: Tue Oct 20 19:14:15 2009
New Revision: 84713

URL: http://llvm.org/viewvc/llvm-project?rev=84713&view=rev
Log:
Add note

Modified:
    llvm/trunk/lib/Target/MSP430/README.txt

Modified: llvm/trunk/lib/Target/MSP430/README.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/README.txt?rev=84713&r1=84712&r2=84713&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/README.txt (original)
+++ llvm/trunk/lib/Target/MSP430/README.txt Tue Oct 20 19:14:15 2009
@@ -40,3 +40,5 @@
 10. Handle imm in comparisons in better way (see comment in MSP430InstrInfo.td)
 
 11. Implement hooks for better memory op folding, etc.
+
+12. Implement branch folding.




From dalej at apple.com  Tue Oct 20 19:23:17 2009
From: dalej at apple.com (Dale Johannesen)
Date: Tue, 20 Oct 2009 17:23:17 -0700
Subject: [llvm-commits] (half precision) [llvm] r84239 - in /llvm/trunk:
	include/llvm/ADT/APFloat.h lib/Support/APFloat.cpp
In-Reply-To: <200910160213.n9G2Dp0C003020@zion.cs.uiuc.edu>
References: <200910160213.n9G2Dp0C003020@zion.cs.uiuc.edu>
Message-ID: <3E890328-4BC7-47D4-B3BE-58CE46E21262@apple.com>

Some testcases to exercise basic functionality would be nice.  I think  
there is a bug here that should have shown up in testing.

On Oct 15, 2009, at 7:13 PM, Chris Lattner wrote:
> +APInt
> +APFloat::convertHalfAPFloatToAPInt() const
> +{
> +  assert(semantics == (const llvm::fltSemantics*)&IEEEhalf);
> +  assert (partCount()==1);
> +
> +  uint32_t myexponent, mysignificand;
> +
> +  if (category==fcNormal) {
> +    myexponent = exponent+15; //bias
> +    mysignificand = (uint32_t)*significandParts();
> +    if (myexponent == 1 && !(mysignificand & 0x400))
> +      myexponent = 0;   // denormal
> +  } else if (category==fcZero) {
> +    myexponent = 0;
> +    mysignificand = 0;
> +  } else if (category==fcInfinity) {
> +    myexponent = 0xff;

0x1f (doesn't matter functionally, but cleaner)

> +    mysignificand = 0;
> +  } else {
> +    assert(category == fcNaN && "Unknown category!");
> +    myexponent = 0xff;

0x1f

> +    mysignificand = (uint32_t)*significandParts();
> +  }
> +
> +  return APInt(16, (((sign&1) << 15) | ((myexponent&0x1f) << 10) |
> +                    (mysignificand & 0x3ff)));
> +}
> +



> +void
> +APFloat::initFromHalfAPInt(const APInt & api)
> +{
> +  assert(api.getBitWidth()==16);
> +  uint32_t i = (uint32_t)*api.getRawData();
> +  uint32_t myexponent = (i >> 15) & 0x1f;

Bug: should be >> 10

> +  uint32_t mysignificand = i & 0x3ff;
> +
> +  initialize(&APFloat::IEEEhalf);
> +  assert(partCount()==1);
> +
> +  sign = i >> 15;
> +  if (myexponent==0 && mysignificand==0) {
> +    // exponent, significand meaningless
> +    category = fcZero;
> +  } else if (myexponent==0x1f && mysignificand==0) {
> +    // exponent, significand meaningless
> +    category = fcInfinity;
> +  } else if (myexponent==0x1f && mysignificand!=0) {
> +    // sign, exponent, significand meaningless
> +    category = fcNaN;
> +    *significandParts() = mysignificand;
> +  } else {
> +    category = fcNormal;
> +    exponent = myexponent - 15;  //bias
> +    *significandParts() = mysignificand;
> +    if (myexponent==0)    // denormal
> +      exponent = -14;
> +    else
> +      *significandParts() |= 0x400; // integer bit
> +  }
> +}
> +
> /// Treat api as containing the bits of a floating point number.   
> Currently
> /// we infer the floating point type from the size of the APInt.  The
> /// isIEEE argument distinguishes between PPC128 and IEEE128 (not  
> meaningful
> @@ -3058,7 +3124,9 @@



From isanbard at gmail.com  Tue Oct 20 19:40:59 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Wed, 21 Oct 2009 00:40:59 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84717 - in /llvm-gcc-4.2/trunk/gcc:
 config/darwin.h llvm-backend.cpp
Message-ID: <200910210040.n9L0ex0w029601@zion.cs.uiuc.edu>

Author: void
Date: Tue Oct 20 19:40:59 2009
New Revision: 84717

URL: http://llvm.org/viewvc/llvm-project?rev=84717&view=rev
Log:
Internal and CFString strings are aligned on a 16-byte boundary. This is
overkill.

Set an explicit alignment on internal and CFString strings. This alignment is 8
on 64-bit targets and 4 on 32-bit ones. Make this conditional so that each
platform may set their own alignment (or none, and keep the original llvm
behavior).

Modified:
    llvm-gcc-4.2/trunk/gcc/config/darwin.h
    llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp

Modified: llvm-gcc-4.2/trunk/gcc/config/darwin.h
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/darwin.h?rev=84717&r1=84716&r2=84717&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/darwin.h (original)
+++ llvm-gcc-4.2/trunk/gcc/config/darwin.h Tue Oct 20 19:40:59 2009
@@ -698,10 +698,21 @@
       char *N = (char *)alloca(strlen(fmt) + 37);                       \
       sprintf(N, fmt, i++);                                             \
       GV->setName(N);                                                   \
+      GV->setAlignment(TARGET_64BIT ? 8 : 4);                           \
     }                                                                   \
   } while (0)
 /* LLVM LOCAL - end radar 6389998 */
 
+/* LLVM LOCAL - begin radar 7291825 */
+/* Give a constant string a sufficient alignment for the platform.  */
+#define TARGET_ADJUST_CSTRING_ALIGN(GV)                                 \
+  do {                                                                  \
+    if (GV->hasInternalLinkage()) {                                     \
+      GV->setAlignment(TARGET_64BIT ? 8 : 4);                           \
+    }                                                                   \
+  } while (0)
+/* LLVM LOCAL - end radar 7291825 */
+
 #endif
 /* LLVM LOCAL end */
 

Modified: llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp?rev=84717&r1=84716&r2=84717&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp (original)
+++ llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp Tue Oct 20 19:40:59 2009
@@ -1406,8 +1406,16 @@
       unsigned TargetAlign =
         getTargetData().getABITypeAlignment(GV->getType()->getElementType());
       if (DECL_USER_ALIGN(decl) ||
-          8 * TargetAlign < (unsigned)DECL_ALIGN(decl))
+          8 * TargetAlign < (unsigned)DECL_ALIGN(decl)) {
         GV->setAlignment(DECL_ALIGN(decl) / 8);
+      }
+#ifdef TARGET_ADJUST_CSTRING_ALIGN
+      else if (DECL_INITIAL(decl) != error_mark_node && // uninitialized?
+               DECL_INITIAL(decl) &&
+               TREE_CODE(DECL_INITIAL(decl)) == STRING_CST) {
+        TARGET_ADJUST_CSTRING_ALIGN(GV);
+      }
+#endif
     }
 
     // Handle used decls




From jyasskin at google.com  Tue Oct 20 19:43:48 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Wed, 21 Oct 2009 00:43:48 -0000
Subject: [llvm-commits] [llvm] r84720 - in /llvm/trunk:
 include/llvm/ExecutionEngine/JITEventListener.h
 lib/ExecutionEngine/JIT/MacOSJITEventListener.cpp tools/lli/lli.cpp
Message-ID: <200910210043.n9L0hm4B029790@zion.cs.uiuc.edu>

Author: jyasskin
Date: Tue Oct 20 19:43:48 2009
New Revision: 84720

URL: http://llvm.org/viewvc/llvm-project?rev=84720&view=rev
Log:
Delete the MacOSJITEventListener per echristo's request. It was disabled by
default and didn't work anyway.

Removed:
    llvm/trunk/lib/ExecutionEngine/JIT/MacOSJITEventListener.cpp
Modified:
    llvm/trunk/include/llvm/ExecutionEngine/JITEventListener.h
    llvm/trunk/tools/lli/lli.cpp

Modified: llvm/trunk/include/llvm/ExecutionEngine/JITEventListener.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ExecutionEngine/JITEventListener.h?rev=84720&r1=84719&r2=84720&view=diff

==============================================================================
--- llvm/trunk/include/llvm/ExecutionEngine/JITEventListener.h (original)
+++ llvm/trunk/include/llvm/ExecutionEngine/JITEventListener.h Tue Oct 20 19:43:48 2009
@@ -67,8 +67,7 @@
   virtual void NotifyFreeingMachineCode(const Function &F, void *OldPtr) {}
 };
 
-// These return NULL if support isn't available.
-JITEventListener *createMacOSJITEventListener();
+// This returns NULL if support isn't available.
 JITEventListener *createOProfileJITEventListener();
 
 } // end namespace llvm.

Removed: llvm/trunk/lib/ExecutionEngine/JIT/MacOSJITEventListener.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/MacOSJITEventListener.cpp?rev=84719&view=auto

==============================================================================
--- llvm/trunk/lib/ExecutionEngine/JIT/MacOSJITEventListener.cpp (original)
+++ llvm/trunk/lib/ExecutionEngine/JIT/MacOSJITEventListener.cpp (removed)
@@ -1,172 +0,0 @@
-//===-- MacOSJITEventListener.cpp - Save symbol table for OSX perf tools --===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file defines a JITEventListener object that records JITted functions to
-// a global __jitSymbolTable linked list.  Apple's performance tools use this to
-// determine a symbol name and accurate code range for a PC value.  Because
-// performance tools are generally asynchronous, the code below is written with
-// the hope that it could be interrupted at any time and have useful answers.
-// However, we don't go crazy with atomic operations, we just do a "reasonable
-// effort".
-//
-//===----------------------------------------------------------------------===//
-
-#define DEBUG_TYPE "macos-jit-event-listener"
-#include "llvm/Function.h"
-#include "llvm/ExecutionEngine/JITEventListener.h"
-#include <stddef.h>
-using namespace llvm;
-
-#ifdef __APPLE__
-#define ENABLE_JIT_SYMBOL_TABLE 0
-#endif
-
-#if ENABLE_JIT_SYMBOL_TABLE
-
-namespace {
-
-/// JITSymbolEntry - Each function that is JIT compiled results in one of these
-/// being added to an array of symbols.  This indicates the name of the function
-/// as well as the address range it occupies.  This allows the client to map
-/// from a PC value to the name of the function.
-struct JITSymbolEntry {
-  const char *FnName;   // FnName - a strdup'd string.
-  void *FnStart;
-  intptr_t FnSize;
-};
-
-
-struct JITSymbolTable {
-  /// NextPtr - This forms a linked list of JitSymbolTable entries.  This
-  /// pointer is not used right now, but might be used in the future.  Consider
-  /// it reserved for future use.
-  JITSymbolTable *NextPtr;
-  
-  /// Symbols - This is an array of JitSymbolEntry entries.  Only the first
-  /// 'NumSymbols' symbols are valid.
-  JITSymbolEntry *Symbols;
-  
-  /// NumSymbols - This indicates the number entries in the Symbols array that
-  /// are valid.
-  unsigned NumSymbols;
-  
-  /// NumAllocated - This indicates the amount of space we have in the Symbols
-  /// array.  This is a private field that should not be read by external tools.
-  unsigned NumAllocated;
-};
-
-class MacOSJITEventListener : public JITEventListener {
-public:
-  virtual void NotifyFunctionEmitted(const Function &F,
-                                     void *FnStart, size_t FnSize,
-                                     const EmittedFunctionDetails &Details);
-  virtual void NotifyFreeingMachineCode(const Function &F, void *OldPtr);
-};
-
-}  // anonymous namespace.
-
-// This is a public symbol so the performance tools can find it.
-JITSymbolTable *__jitSymbolTable;
-
-namespace llvm {
-JITEventListener *createMacOSJITEventListener() {
-  return new MacOSJITEventListener;
-}
-}
-
-// Adds the just-emitted function to the symbol table.
-void MacOSJITEventListener::NotifyFunctionEmitted(
-    const Function &F, void *FnStart, size_t FnSize,
-    const EmittedFunctionDetails &) {
-  assert(F.hasName() && FnStart != 0 && "Bad symbol to add");
-  JITSymbolTable **SymTabPtrPtr = 0;
-  SymTabPtrPtr = &__jitSymbolTable;
-
-  // If this is the first entry in the symbol table, add the JITSymbolTable
-  // index.
-  if (*SymTabPtrPtr == 0) {
-    JITSymbolTable *New = new JITSymbolTable();
-    New->NextPtr = 0;
-    New->Symbols = 0;
-    New->NumSymbols = 0;
-    New->NumAllocated = 0;
-    *SymTabPtrPtr = New;
-  }
-
-  JITSymbolTable *SymTabPtr = *SymTabPtrPtr;
-
-  // If we have space in the table, reallocate the table.
-  if (SymTabPtr->NumSymbols >= SymTabPtr->NumAllocated) {
-    // If we don't have space, reallocate the table.
-    unsigned NewSize = std::max(64U, SymTabPtr->NumAllocated*2);
-    JITSymbolEntry *NewSymbols = new JITSymbolEntry[NewSize];
-    JITSymbolEntry *OldSymbols = SymTabPtr->Symbols;
-
-    // Copy the old entries over.
-    memcpy(NewSymbols, OldSymbols, SymTabPtr->NumSymbols*sizeof(OldSymbols[0]));
-
-    // Swap the new symbols in, delete the old ones.
-    SymTabPtr->Symbols = NewSymbols;
-    SymTabPtr->NumAllocated = NewSize;
-    delete [] OldSymbols;
-  }
-
-  // Otherwise, we have enough space, just tack it onto the end of the array.
-  JITSymbolEntry &Entry = SymTabPtr->Symbols[SymTabPtr->NumSymbols];
-  Entry.FnName = strdup(F.getName().data());
-  Entry.FnStart = FnStart;
-  Entry.FnSize = FnSize;
-  ++SymTabPtr->NumSymbols;
-}
-
-// Removes the to-be-deleted function from the symbol table.
-void MacOSJITEventListener::NotifyFreeingMachineCode(
-    const Function &, void *FnStart) {
-  assert(FnStart && "Invalid function pointer");
-  JITSymbolTable **SymTabPtrPtr = 0;
-  SymTabPtrPtr = &__jitSymbolTable;
-
-  JITSymbolTable *SymTabPtr = *SymTabPtrPtr;
-  JITSymbolEntry *Symbols = SymTabPtr->Symbols;
-
-  // Scan the table to find its index.  The table is not sorted, so do a linear
-  // scan.
-  unsigned Index;
-  for (Index = 0; Symbols[Index].FnStart != FnStart; ++Index)
-    assert(Index != SymTabPtr->NumSymbols && "Didn't find function!");
-
-  // Once we have an index, we know to nuke this entry, overwrite it with the
-  // entry at the end of the array, making the last entry redundant.
-  const char *OldName = Symbols[Index].FnName;
-  Symbols[Index] = Symbols[SymTabPtr->NumSymbols-1];
-  free((void*)OldName);
-
-  // Drop the number of symbols in the table.
-  --SymTabPtr->NumSymbols;
-
-  // Finally, if we deleted the final symbol, deallocate the table itself.
-  if (SymTabPtr->NumSymbols != 0)
-    return;
-
-  *SymTabPtrPtr = 0;
-  delete [] Symbols;
-  delete SymTabPtr;
-}
-
-#else  // !ENABLE_JIT_SYMBOL_TABLE
-
-namespace llvm {
-// By defining this to return NULL, we can let clients call it unconditionally,
-// even if they aren't on an Apple system.
-JITEventListener *createMacOSJITEventListener() {
-  return NULL;
-}
-}  // namespace llvm
-
-#endif  // ENABLE_JIT_SYMBOL_TABLE

Modified: llvm/trunk/tools/lli/lli.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/lli/lli.cpp?rev=84720&r1=84719&r2=84720&view=diff

==============================================================================
--- llvm/trunk/tools/lli/lli.cpp (original)
+++ llvm/trunk/tools/lli/lli.cpp Tue Oct 20 19:43:48 2009
@@ -163,7 +163,6 @@
     exit(1);
   }
 
-  EE->RegisterJITEventListener(createMacOSJITEventListener());
   EE->RegisterJITEventListener(createOProfileJITEventListener());
 
   if (NoLazyCompilation)




From isanbard at gmail.com  Tue Oct 20 19:48:14 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Wed, 21 Oct 2009 00:48:14 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84722 - in /llvm-gcc-4.2/trunk/gcc:
 ChangeLog.apple config/arm/arm-protos.h config/arm/linux-eabi.h
 config/darwin-c.c config/t-darwin cp/parser.c doc/invoke.texi gcc.c
 tree-eh.c version.c
Message-ID: <200910210048.n9L0mFSI029999@zion.cs.uiuc.edu>

Author: void
Date: Tue Oct 20 19:48:14 2009
New Revision: 84722

URL: http://llvm.org/viewvc/llvm-project?rev=84722&view=rev
Log:
Revert r84688 and r84699. It was breaking the build.

Modified:
    llvm-gcc-4.2/trunk/gcc/ChangeLog.apple
    llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h
    llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h
    llvm-gcc-4.2/trunk/gcc/config/darwin-c.c
    llvm-gcc-4.2/trunk/gcc/config/t-darwin
    llvm-gcc-4.2/trunk/gcc/cp/parser.c
    llvm-gcc-4.2/trunk/gcc/doc/invoke.texi
    llvm-gcc-4.2/trunk/gcc/gcc.c
    llvm-gcc-4.2/trunk/gcc/tree-eh.c
    llvm-gcc-4.2/trunk/gcc/version.c

Modified: llvm-gcc-4.2/trunk/gcc/ChangeLog.apple
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/ChangeLog.apple?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/ChangeLog.apple (original)
+++ llvm-gcc-4.2/trunk/gcc/ChangeLog.apple Tue Oct 20 19:48:14 2009
@@ -1,18 +1,3 @@
-2009-10-15  Jim Grosbach <grosbach at apple.com>
-
-	Radar 7295521
-	* config/arm/arm.c (arm_split_compare_and_swap): Call insns for
-	arm_memory_barrier* and arm_memory_sync* directory rather than using
-	the expander.
-	* config/arm/sync.md (memory_barrier, memory_sync): Allocate a scratch
-	register rather than forcing IP.
-	(arm_memory_barrier_v6): Match the scratch register from the expander
-	rather than forcing IP. Rename to arm_memory_barrier_v6_explicit.
-	(arm_memory_barrier_v6_explicit): New.
-	(arm_memory_sync_v6): Match the scratch register from the expander
-	rather than forcing IP. Rename to arm_memory_sync_v6_explicit.
-	(arm_memory_sync_v6_explicit): New.
-
 2009-09-11  Caroline Tice  <ctice at apple.com>
 
 	Radar 6419781
@@ -24,8 +9,8 @@
 
 2009-09-08  Jim Grosbach <grosbach at apple.com>
 
-	Radar 7174451
-	* config/arm/arm.c (arm_select_dominance_cc_mode): Unordered floats.
+       Radar 7174451
+       * config/arm/arm.c (arm_select_dominance_cc_mode): Unordered floats.
 
 2009-09-08  Jim Grosbach <grosbach at apple.com>
 

Modified: llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h (original)
+++ llvm-gcc-4.2/trunk/gcc/config/arm/arm-protos.h Tue Oct 20 19:48:14 2009
@@ -258,14 +258,7 @@
 /* APPLE LOCAL 5946347 ms_struct support */
 extern int arm_field_ms_struct_align (tree);
 
-/* LLVM LOCAL begin pr5037 removed arm_mangle_type */
-#ifndef ENABLE_LLVM
-/* APPLE LOCAL begin v7 support. Merge from Codesourcery */
-extern const char *arm_mangle_type (tree);
-
-/* APPLE LOCAL end v7 support. Merge from Codesourcery */
-#endif
-/* LLVM LOCAL end pr5037 removed arm_mangle_type */
+/* LLVM LOCAL pr5037 removed arm_mangle_type */
 
 /* APPLE LOCAL v7 support. Fix compact switch tables */
 extern void arm_asm_output_addr_diff_vec (FILE *file, rtx LABEL, rtx BODY);

Modified: llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h (original)
+++ llvm-gcc-4.2/trunk/gcc/config/arm/linux-eabi.h Tue Oct 20 19:48:14 2009
@@ -79,7 +79,7 @@
   register unsigned long _end __asm ("a2") = (unsigned long) (END);	\
   register unsigned long _flg __asm ("a3") = 0;				\
   register unsigned long _scno __asm ("r7") = 0xf0002;			\
-  __asm __volatile ("swi 0x9f0002		@ sys_cacheflush"	\
+  __asm __volatile ("swi 0		@ sys_cacheflush"		\
 		    : "=r" (_beg)					\
 		    : "0" (_beg), "r" (_end), "r" (_flg), "r" (_scno));	\
 }

Modified: llvm-gcc-4.2/trunk/gcc/config/darwin-c.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/darwin-c.c?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/darwin-c.c (original)
+++ llvm-gcc-4.2/trunk/gcc/config/darwin-c.c Tue Oct 20 19:48:14 2009
@@ -412,7 +412,7 @@
 /* APPLE LOCAL end pragma reverse_bitfields */
 
 /* APPLE LOCAL begin optimization pragmas 3124235/3420242 */
-varray_type va_opt;
+static GTY(()) varray_type va_opt;
 
 static void
 push_opt_level (int level, int size)
@@ -1245,3 +1245,4 @@
   return decl;
 }
 /* APPLE LOCAL end radar 2996215 - 6068877 */
+#include "gt-darwin-c.h"

Modified: llvm-gcc-4.2/trunk/gcc/config/t-darwin
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/config/t-darwin?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/config/t-darwin (original)
+++ llvm-gcc-4.2/trunk/gcc/config/t-darwin Tue Oct 20 19:48:14 2009
@@ -16,10 +16,11 @@
 
 darwin-c.o: $(srcdir)/config/darwin-c.c $(CONFIG_H) $(SYSTEM_H) coretypes.h \
   $(TM_H) $(CPPLIB_H) $(TREE_H) c-pragma.h $(C_TREE_H) toplev.h $(TM_P_H) \
-  c-incpath.h flags.h $(C_COMMON_H)
+  c-incpath.h flags.h $(C_COMMON_H) gt-darwin-c.h
 	$(CC) -c $(ALL_CFLAGS) $(ALL_CPPFLAGS) $(INCLUDES) $(srcdir)/config/darwin-c.c $(PREPROCESSOR_DEFINES)
 
 gt-darwin.h : s-gtype ; @true
+gt-darwin-c.h : s-gtype ; @true
 
 # APPLE LOCAL begin mainline 2007-06-14 5235474
 darwin-driver.o: $(srcdir)/config/darwin-driver.c \

Modified: llvm-gcc-4.2/trunk/gcc/cp/parser.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/cp/parser.c?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/cp/parser.c (original)
+++ llvm-gcc-4.2/trunk/gcc/cp/parser.c Tue Oct 20 19:48:14 2009
@@ -21078,11 +21078,12 @@
   CONSTRUCTOR_ELTS (constructor) = build_block_struct_initlist (block_struct_type,
                                                                 block_impl);
   /* Temporary representing a global block is made global static.  */
-  /* APPLE LOCAL radar 6230297 */
+  /* APPLE LOCAL begin radar 6230297 */
   if (staticBlockTmp || global_bindings_p ()) {
     TREE_PUBLIC (block_holder_tmp_decl) = 0;
     TREE_STATIC (block_holder_tmp_decl) = 1;
   }
+  /* APPLE LOCAL end radar 6230297 */
   cp_finish_decl (block_holder_tmp_decl, constructor, 0, 0, LOOKUP_ONLYCONVERTING);
   return block_holder_tmp_decl;
 }

Modified: llvm-gcc-4.2/trunk/gcc/doc/invoke.texi
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/doc/invoke.texi?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/doc/invoke.texi (original)
+++ llvm-gcc-4.2/trunk/gcc/doc/invoke.texi Tue Oct 20 19:48:14 2009
@@ -175,7 +175,8 @@
 @table @emph
 @item Overall Options
 @xref{Overall Options,,Options Controlling the Kind of Output}.
- at gccoptlist{-c  -S  -E  -o @var{file}  -combine -pipe  -pass-exit-codes  @gol
+ at gccoptlist{-c  -S  -E  -o @var{file}  -combine  -no-canonical-prefixes  @gol
+-pipe  -pass-exit-codes  @gol
 @c APPLE LOCAL -ObjC 2001-08-03 --sts **
 -ObjC (APPLE ONLY) -ObjC++ (APPLE ONLY) @gol
 @c APPLE LOCAL begin fat builds
@@ -1238,6 +1239,12 @@
 Print (on the standard output) a description of target specific command
 line options for each tool.
 
+ at item -no-canonical-prefixes
+ at opindex no-canonical-prefixes
+Do not expand any symbolic links, resolve references to @samp{/../}
+or @samp{/./}, or make the path absolute when generating a relative
+prefix.
+
 @item --version
 @opindex version
 Display the version number and copyrights of the invoked GCC at .

Modified: llvm-gcc-4.2/trunk/gcc/gcc.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/gcc.c?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/gcc.c (original)
+++ llvm-gcc-4.2/trunk/gcc/gcc.c Tue Oct 20 19:48:14 2009
@@ -1190,6 +1190,7 @@
    /* LLVM LOCAL end */
    {"--machine", "-m", "aj"},
    {"--machine-", "-m", "*j"},
+   {"--no-canonical-prefixes", "-no-canonical-prefixes", 0},
    {"--no-integrated-cpp", "-no-integrated-cpp", 0},
    {"--no-line-commands", "-P", 0},
    {"--no-precompiled-includes", "-noprecomp", 0},
@@ -3341,6 +3342,10 @@
   fputs (_("  -Xlinker <arg>           Pass <arg> on to the linker\n"), stdout);
   fputs (_("  -combine                 Pass multiple source files to compiler at once\n"), stdout);
   fputs (_("  -save-temps              Do not delete intermediate files\n"), stdout);
+  fputs (_("\
+  -no-canonical-prefixes   Do not canonicalize paths when building relative\n\
+                           prefixes to other gcc components\n"), stdout);
+
   fputs (_("  -pipe                    Use pipes rather than intermediate files\n"), stdout);
   fputs (_("  -time                    Time the execution of each subprocess\n"), stdout);
   fputs (_("  -specs=<file>            Override built-in specs with the contents of <file>\n"), stdout);
@@ -3433,6 +3438,8 @@
   int is_modify_target_name;
   unsigned int j;
 #endif
+  char *(*get_relative_prefix) (const char *, const char *,
+				const char *) = NULL;
 
   GET_ENVIRONMENT (gcc_exec_prefix, "GCC_EXEC_PREFIX");
 
@@ -3519,6 +3526,28 @@
       fatal ("couldn't run '%s': %s", new_argv0, xstrerror (errno));
     }
 
+  /* Convert new-style -- options to old-style.  */
+  translate_options (&argc, (const char * const **) &argv);
+
+  /* Do language-specific adjustment/addition of flags.  */
+  lang_specific_driver (&argc, (const char * const **) &argv,
+			&added_libraries);
+
+  /* Handle any -no-canonical-prefixes flag early, to assign the function
+     that builds relative prefixes.  This function creates default search
+     paths that are needed later in normal option handling.  */
+
+  for (i = 1; i < argc; i++)
+    {
+      if (! strcmp (argv[i], "-no-canonical-prefixes"))
+	{
+	  get_relative_prefix = make_relative_prefix_ignore_links;
+	  break;
+	}
+    }
+  if (! get_relative_prefix)
+    get_relative_prefix = make_relative_prefix;
+
   /* Set up the default search paths.  If there is no GCC_EXEC_PREFIX,
      see if we can create it from the pathname specified in argv[0].  */
 
@@ -3527,11 +3556,12 @@
   /* FIXME: make_relative_prefix doesn't yet work for VMS.  */
   if (!gcc_exec_prefix)
     {
-      gcc_exec_prefix = make_relative_prefix (argv[0], standard_bindir_prefix,
-					      standard_exec_prefix);
-      gcc_libexec_prefix = make_relative_prefix (argv[0],
-						 standard_bindir_prefix,
-						 standard_libexec_prefix);
+      gcc_exec_prefix = get_relative_prefix (argv[0],
+					     standard_bindir_prefix,
+					     standard_exec_prefix);
+      gcc_libexec_prefix = get_relative_prefix (argv[0],
+					     standard_bindir_prefix,
+					     standard_libexec_prefix);
       if (gcc_exec_prefix)
 	putenv (concat ("GCC_EXEC_PREFIX=", gcc_exec_prefix, NULL));
     }
@@ -3542,9 +3572,9 @@
 	 / (which is ignored by make_relative_prefix), so append a
 	 program name.  */
       char *tmp_prefix = concat (gcc_exec_prefix, "gcc", NULL);
-      gcc_libexec_prefix = make_relative_prefix (tmp_prefix,
-						 standard_exec_prefix,
-						 standard_libexec_prefix);
+      gcc_libexec_prefix = get_relative_prefix (tmp_prefix,
+						standard_exec_prefix,
+						standard_libexec_prefix);
       free (tmp_prefix);
     }
 #else
@@ -3674,12 +3704,6 @@
 	}
     }
 
-  /* Convert new-style -- options to old-style.  */
-  translate_options (&argc, (const char *const **) &argv);
-
-  /* Do language-specific adjustment/addition of flags.  */
-  lang_specific_driver (&argc, (const char *const **) &argv, &added_libraries);
-
   /* Scan argv twice.  Here, the first time, just count how many switches
      there will be in their vector, and how many input files in theirs.
      Here we also parse the switches that cc itself uses (e.g. -v).  */
@@ -3891,6 +3915,9 @@
 	  n_infiles += 2;
 	  i++;
 	}
+      else if (strcmp (argv[i], "-no-canonical-prefixes") == 0)
+	/* Already handled as a special case, so ignored here.  */
+	;
       /* APPLE LOCAL end -weak_* (radar 3235250) */
       else if (strcmp (argv[i], "-combine") == 0)
 	{
@@ -4224,9 +4251,9 @@
      ``make_relative_prefix'' is not compiled for VMS, so don't call it.  */
   if (target_system_root && gcc_exec_prefix)
     {
-      char *tmp_prefix = make_relative_prefix (argv[0],
-					       standard_bindir_prefix,
-					       target_system_root);
+      char *tmp_prefix = get_relative_prefix (argv[0],
+					      standard_bindir_prefix,
+					      target_system_root);
       if (tmp_prefix && access_check (tmp_prefix, F_OK) == 0)
 	{
 	  target_system_root = tmp_prefix;
@@ -4268,6 +4295,8 @@
 	;
       else if (! strncmp (argv[i], "-Wp,", 4))
 	;
+      else if (! strcmp (argv[i], "-no-canonical-prefixes"))
+	;
       else if (! strcmp (argv[i], "-pass-exit-codes"))
 	;
       else if (! strcmp (argv[i], "-print-search-dirs"))

Modified: llvm-gcc-4.2/trunk/gcc/tree-eh.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/tree-eh.c?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/tree-eh.c (original)
+++ llvm-gcc-4.2/trunk/gcc/tree-eh.c Tue Oct 20 19:48:14 2009
@@ -839,12 +839,10 @@
 
       save_eptr = create_tmp_var (ptr_type_node, "save_eptr");
       save_filt = create_tmp_var (integer_type_node, "save_filt");
-
       i = tsi_start (finally);
       x = build0 (EXC_PTR_EXPR, ptr_type_node);
       x = build2 (MODIFY_EXPR, void_type_node, save_eptr, x);
       tsi_link_before (&i, x, TSI_CONTINUE_LINKING);
-
       x = build0 (FILTER_EXPR, integer_type_node);
       x = build2 (MODIFY_EXPR, void_type_node, save_filt, x);
       tsi_link_before (&i, x, TSI_CONTINUE_LINKING);

Modified: llvm-gcc-4.2/trunk/gcc/version.c
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/version.c?rev=84722&r1=84721&r2=84722&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/version.c (original)
+++ llvm-gcc-4.2/trunk/gcc/version.c Tue Oct 20 19:48:14 2009
@@ -11,12 +11,12 @@
 /* APPLE LOCAL begin Apple version */
 #ifdef ENABLE_LLVM
 #ifdef LLVM_VERSION_INFO
-#define VERSUFFIX " (Based on Apple Inc. build 5654) (LLVM build " LLVM_VERSION_INFO ")"
+#define VERSUFFIX " (Based on Apple Inc. build 5653) (LLVM build " LLVM_VERSION_INFO ")"
 #else
-#define VERSUFFIX " (Based on Apple Inc. build 5654) (LLVM build)"
+#define VERSUFFIX " (Based on Apple Inc. build 5653) (LLVM build)"
 #endif
 #else
-#define VERSUFFIX " (Based on Apple Inc. build 5654)"
+#define VERSUFFIX " (Based on Apple Inc. build 5653)"
 #endif
 /* APPLE LOCAL end Apple version */
 




From isanbard at gmail.com  Tue Oct 20 19:51:41 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Wed, 21 Oct 2009 00:51:41 -0000
Subject: [llvm-commits] [llvm] r84723 - /llvm/trunk/test/FrontendC/msasm.c
Message-ID: <200910210051.n9L0pfFT030148@zion.cs.uiuc.edu>

Author: void
Date: Tue Oct 20 19:51:40 2009
New Revision: 84723

URL: http://llvm.org/viewvc/llvm-project?rev=84723&view=rev
Log:
This is passing on Darwin PPC.

Modified:
    llvm/trunk/test/FrontendC/msasm.c

Modified: llvm/trunk/test/FrontendC/msasm.c
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC/msasm.c?rev=84723&r1=84722&r2=84723&view=diff

==============================================================================
--- llvm/trunk/test/FrontendC/msasm.c (original)
+++ llvm/trunk/test/FrontendC/msasm.c Tue Oct 20 19:51:40 2009
@@ -1,7 +1,7 @@
 // RUN: %llvmgcc %s -fasm-blocks -S -o - | FileCheck %s
 // Complicated expression as jump target
 // XFAIL: *
-// XTARGET: x86,i386,i686
+// XTARGET: x86,i386,i686,darwin
 
 void Method3()
 {




From sabre at nondot.org  Tue Oct 20 20:10:38 2009
From: sabre at nondot.org (Chris Lattner)
Date: Wed, 21 Oct 2009 01:10:38 -0000
Subject: [llvm-commits] [llvm] r84725 - /llvm/trunk/lib/Target/README.txt
Message-ID: <200910210110.n9L1Ackw031052@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 20:10:37 2009
New Revision: 84725

URL: http://llvm.org/viewvc/llvm-project?rev=84725&view=rev
Log:
IPSCCP is missing stuff.

Modified:
    llvm/trunk/lib/Target/README.txt

Modified: llvm/trunk/lib/Target/README.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/README.txt?rev=84725&r1=84724&r2=84725&view=diff

==============================================================================
--- llvm/trunk/lib/Target/README.txt (original)
+++ llvm/trunk/lib/Target/README.txt Tue Oct 20 20:10:37 2009
@@ -1644,3 +1644,10 @@
 See also PR4973
 
 //===---------------------------------------------------------------------===//
+
+IPSCCP is propagating elements of first class aggregates, but is not propagating
+the entire aggregate itself.  This leads it to miss opportunities, for example
+in test/Transforms/SCCP/ipsccp-basic.ll:test5b.
+
+//===---------------------------------------------------------------------===//
+




From kennethuil at gmail.com  Tue Oct 20 20:27:30 2009
From: kennethuil at gmail.com (Kenneth Uildriks)
Date: Tue, 20 Oct 2009 20:27:30 -0500
Subject: [llvm-commits] [PATCH] Fix for GVN::processLoad when no TargetData
	available
Message-ID: <400d33ea0910201827n39fe1517y6ccb789b42e423f3@mail.gmail.com>

In the existing code, if the load and the value to replace it with are
of different types *and* target data is available, it tries to use the
target data to coerce the replacement value to the type of the load.
Otherwise, it skips all effort to handle the type mismatch and just
feeds the wrongly-typed replacement value to replaceAllUsesWith, which
triggers an assertion.

The patch replaces it with an outer if checking for type mismatch, and
an inner if-else that checks whether target data is available and, if
not, returns false rather than trying to replace the load.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gvn.patch
Type: text/x-patch
Size: 2265 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091020/5668c4b5/attachment.bin 

From gohman at apple.com  Tue Oct 20 20:44:44 2009
From: gohman at apple.com (Dan Gohman)
Date: Wed, 21 Oct 2009 01:44:44 -0000
Subject: [llvm-commits] [llvm] r84727 - in /llvm/trunk:
 lib/CodeGen/PostRASchedulerList.cpp
 test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll
 test/CodeGen/X86/break-anti-dependencies.ll test/CodeGen/X86/sse3.ll
Message-ID: <200910210144.n9L1iiBB032608@zion.cs.uiuc.edu>

Author: djg
Date: Tue Oct 20 20:44:44 2009
New Revision: 84727

URL: http://llvm.org/viewvc/llvm-project?rev=84727&view=rev
Log:
Revert r84658 and r84691. They were causing llvm-gcc bootstrap to fail.

Modified:
    llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
    llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll
    llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll
    llvm/trunk/test/CodeGen/X86/sse3.ll

Modified: llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp?rev=84727&r1=84726&r2=84727&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp (original)
+++ llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp Tue Oct 20 20:44:44 2009
@@ -48,7 +48,6 @@
 
 STATISTIC(NumNoops, "Number of noops inserted");
 STATISTIC(NumStalls, "Number of pipeline stalls");
-STATISTIC(NumFixedAnti, "Number of fixed anti-dependencies");
 
 // Post-RA scheduling is enabled with
 // TargetSubtarget.enablePostRAScheduler(). This flag can be used to
@@ -57,11 +56,10 @@
 EnablePostRAScheduler("post-RA-scheduler",
                        cl::desc("Enable scheduling after register allocation"),
                        cl::init(false), cl::Hidden);
-static cl::opt<std::string>
+static cl::opt<bool>
 EnableAntiDepBreaking("break-anti-dependencies",
-                      cl::desc("Break post-RA scheduling anti-dependencies: "
-                               "\"critical\", \"all\", or \"none\""),
-                      cl::init("critical"), cl::Hidden);
+                      cl::desc("Break post-RA scheduling anti-dependencies"),
+                      cl::init(true), cl::Hidden);
 static cl::opt<bool>
 EnablePostRAHazardAvoidance("avoid-hazards",
                       cl::desc("Enable exact hazard avoidance"),
@@ -106,16 +104,8 @@
   char PostRAScheduler::ID = 0;
 
   class VISIBILITY_HIDDEN SchedulePostRATDList : public ScheduleDAGInstrs {
-    /// RegisterReference - Information about a register reference
-    /// within a liverange
-    typedef struct {
-      /// Operand - The registers operand
-      MachineOperand *Operand;
-      /// RC - The register class
-      const TargetRegisterClass *RC;
-    } RegisterReference;
-
     /// AvailableQueue - The priority queue to use for the available SUnits.
+    ///
     LatencyPriorityQueue AvailableQueue;
   
     /// PendingQueue - This contains all of the instructions whose operands have
@@ -127,43 +117,40 @@
     /// Topo - A topological ordering for SUnits.
     ScheduleDAGTopologicalSort Topo;
 
+    /// AllocatableSet - The set of allocatable registers.
+    /// We'll be ignoring anti-dependencies on non-allocatable registers,
+    /// because they may not be safe to break.
+    const BitVector AllocatableSet;
+
     /// HazardRec - The hazard recognizer to use.
     ScheduleHazardRecognizer *HazardRec;
 
     /// AA - AliasAnalysis for making memory reference queries.
     AliasAnalysis *AA;
 
-    /// AllocatableSet - The set of allocatable registers.
-    /// We'll be ignoring anti-dependencies on non-allocatable registers,
-    /// because they may not be safe to break.
-    const BitVector AllocatableSet;
-
-    /// GroupNodes - Implements a disjoint-union data structure to
-    /// form register groups. A node is represented by an index into
-    /// the vector. A node can "point to" itself to indicate that it
-    /// is the parent of a group, or point to another node to indicate
-    /// that it is a member of the same group as that node.
-    std::vector<unsigned> GroupNodes;
-
-    /// GroupNodeIndices - For each register, the index of the GroupNode
-    /// currently representing the group that the register belongs to.
-    /// Register 0 is always represented by the 0 group, a group
-    /// composed of registers that are not eligible for anti-aliasing.
-    unsigned GroupNodeIndices[TargetRegisterInfo::FirstVirtualRegister];
+    /// Classes - For live regs that are only used in one register class in a
+    /// live range, the register class. If the register is not live, the
+    /// corresponding value is null. If the register is live but used in
+    /// multiple register classes, the corresponding value is -1 casted to a
+    /// pointer.
+    const TargetRegisterClass *
+      Classes[TargetRegisterInfo::FirstVirtualRegister];
 
     /// RegRegs - Map registers to all their references within a live range.
-    std::multimap<unsigned, RegisterReference> RegRefs;
+    std::multimap<unsigned, MachineOperand *> RegRefs;
 
-    /// KillIndices - The index of the most recent kill (proceding
-    /// bottom-up), or ~0u if no kill of the register has been
-    /// seen. The register is live if this index != ~0u and DefIndices
-    /// == ~0u.
+    /// KillIndices - The index of the most recent kill (proceding bottom-up),
+    /// or ~0u if the register is not live.
     unsigned KillIndices[TargetRegisterInfo::FirstVirtualRegister];
 
     /// DefIndices - The index of the most recent complete def (proceding bottom
     /// up), or ~0u if the register is live.
     unsigned DefIndices[TargetRegisterInfo::FirstVirtualRegister];
 
+    /// KeepRegs - A set of registers which are live and cannot be changed to
+    /// break anti-dependencies.
+    SmallSet<unsigned, 4> KeepRegs;
+
   public:
     SchedulePostRATDList(MachineFunction &MF,
                          const MachineLoopInfo &MLI,
@@ -171,9 +158,8 @@
                          ScheduleHazardRecognizer *HR,
                          AliasAnalysis *aa)
       : ScheduleDAGInstrs(MF, MLI, MDT), Topo(SUnits),
-      HazardRec(HR), AA(aa),
-      AllocatableSet(TRI->getAllocatableSet(MF)),
-      GroupNodes(TargetRegisterInfo::FirstVirtualRegister, 0) {}
+        AllocatableSet(TRI->getAllocatableSet(MF)),
+        HazardRec(HR), AA(aa) {}
 
     ~SchedulePostRATDList() {
       delete HazardRec;
@@ -184,15 +170,6 @@
     ///
     void StartBlock(MachineBasicBlock *BB);
 
-    /// FinishBlock - Clean up register live-range state.
-    ///
-    void FinishBlock();
-
-    /// Observe - Update liveness information to account for the current
-    /// instruction, which will not be scheduled.
-    ///
-    void Observe(MachineInstr *MI, unsigned Count);
-
     /// Schedule - Schedule the instruction range using list scheduling.
     ///
     void Schedule();
@@ -202,44 +179,32 @@
     ///
     void FixupKills(MachineBasicBlock *MBB);
 
-  private:
-    /// IsLive - Return true if Reg is live
-    bool IsLive(unsigned Reg);
+    /// Observe - Update liveness information to account for the current
+    /// instruction, which will not be scheduled.
+    ///
+    void Observe(MachineInstr *MI, unsigned Count);
 
-    void PrescanInstruction(MachineInstr *MI, unsigned Count);
-    void ScanInstruction(MachineInstr *MI, unsigned Count);
-    bool BreakAntiDependencies(bool CriticalPathOnly);
-    unsigned FindSuitableFreeRegister(unsigned AntiDepReg);
+    /// FinishBlock - Clean up register live-range state.
+    ///
+    void FinishBlock();
 
+  private:
+    void PrescanInstruction(MachineInstr *MI);
+    void ScanInstruction(MachineInstr *MI, unsigned Count);
     void ReleaseSucc(SUnit *SU, SDep *SuccEdge);
     void ReleaseSuccessors(SUnit *SU);
     void ScheduleNodeTopDown(SUnit *SU, unsigned CurCycle);
     void ListScheduleTopDown();
-
+    bool BreakAntiDependencies();
+    unsigned findSuitableFreeRegister(unsigned AntiDepReg,
+                                      unsigned LastNewReg,
+                                      const TargetRegisterClass *);
     void StartBlockForKills(MachineBasicBlock *BB);
     
     // ToggleKillFlag - Toggle a register operand kill flag. Other
     // adjustments may be made to the instruction if necessary. Return
     // true if the operand has been deleted, false if not.
     bool ToggleKillFlag(MachineInstr *MI, MachineOperand &MO);
-    
-    // GetGroup - Get the group for a register. The returned value is
-    // the index of the GroupNode representing the group.
-    unsigned GetGroup(unsigned Reg);
-    
-    // GetGroupRegs - Return a vector of the registers belonging to a
-    // group.
-    void GetGroupRegs(unsigned Group, std::vector<unsigned> &Regs);
-
-    // UnionGroups - Union Reg1's and Reg2's groups to form a new
-    // group. Return the index of the GroupNode representing the
-    // group.
-    unsigned UnionGroups(unsigned Reg1, unsigned Reg2);
-
-    // LeaveGroup - Remove a register from its current group and place
-    // it alone in its own group. Return the index of the GroupNode
-    // representing the registers new group.
-    unsigned LeaveGroup(unsigned Reg);
   };
 }
 
@@ -338,58 +303,7 @@
 
   return true;
 }
-
-unsigned SchedulePostRATDList::GetGroup(unsigned Reg)
-{
-  unsigned Node = GroupNodeIndices[Reg];
-  while (GroupNodes[Node] != Node)
-    Node = GroupNodes[Node];
-
-  return Node;
-}
-
-void SchedulePostRATDList::GetGroupRegs(unsigned Group, std::vector<unsigned> &Regs)
-{
-  for (unsigned Reg = 0; Reg != TargetRegisterInfo::FirstVirtualRegister; ++Reg) {
-    if (GetGroup(Reg) == Group)
-      Regs.push_back(Reg);
-  }
-}
-
-unsigned SchedulePostRATDList::UnionGroups(unsigned Reg1, unsigned Reg2)
-{
-  assert(GroupNodes[0] == 0 && "GroupNode 0 not parent!");
-  assert(GroupNodeIndices[0] == 0 && "Reg 0 not in Group 0!");
-  
-  // find group for each register
-  unsigned Group1 = GetGroup(Reg1);
-  unsigned Group2 = GetGroup(Reg2);
-  
-  // if either group is 0, then that must become the parent
-  unsigned Parent = (Group1 == 0) ? Group1 : Group2;
-  unsigned Other = (Parent == Group1) ? Group2 : Group1;
-  GroupNodes.at(Other) = Parent;
-  return Parent;
-}
   
-unsigned SchedulePostRATDList::LeaveGroup(unsigned Reg)
-{
-  // Create a new GroupNode for Reg. Reg's existing GroupNode must
-  // stay as is because there could be other GroupNodes referring to
-  // it.
-  unsigned idx = GroupNodes.size();
-  GroupNodes.push_back(idx);
-  GroupNodeIndices[Reg] = idx;
-  return idx;
-}
-
-bool SchedulePostRATDList::IsLive(unsigned Reg)
-{
-  // KillIndex must be defined and DefIndex not defined for a register
-  // to be live.
-  return((KillIndices[Reg] != ~0u) && (DefIndices[Reg] == ~0u));
-}
-
 /// StartBlock - Initialize register live-range state for scheduling in
 /// this block.
 ///
@@ -400,15 +314,17 @@
   // Reset the hazard recognizer.
   HazardRec->Reset();
 
-  // Initialize all registers to be in their own group. Initially we
-  // assign the register to the same-indexed GroupNode.
-  for (unsigned i = 0; i < TargetRegisterInfo::FirstVirtualRegister; ++i)
-    GroupNodeIndices[i] = i;
+  // Clear out the register class data.
+  std::fill(Classes, array_endof(Classes),
+            static_cast<const TargetRegisterClass *>(0));
 
   // Initialize the indices to indicate that no registers are live.
   std::fill(KillIndices, array_endof(KillIndices), ~0u);
   std::fill(DefIndices, array_endof(DefIndices), BB->size());
 
+  // Clear "do not change" set.
+  KeepRegs.clear();
+
   bool IsReturnBlock = (!BB->empty() && BB->back().getDesc().isReturn());
 
   // Determine the live-out physregs for this block.
@@ -417,13 +333,13 @@
     for (MachineRegisterInfo::liveout_iterator I = MRI.liveout_begin(),
          E = MRI.liveout_end(); I != E; ++I) {
       unsigned Reg = *I;
-      UnionGroups(Reg, 0);
+      Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
       KillIndices[Reg] = BB->size();
       DefIndices[Reg] = ~0u;
       // Repeat, for all aliases.
       for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
         unsigned AliasReg = *Alias;
-        UnionGroups(AliasReg, 0);
+        Classes[AliasReg] = reinterpret_cast<TargetRegisterClass *>(-1);
         KillIndices[AliasReg] = BB->size();
         DefIndices[AliasReg] = ~0u;
       }
@@ -435,13 +351,13 @@
       for (MachineBasicBlock::livein_iterator I = (*SI)->livein_begin(),
            E = (*SI)->livein_end(); I != E; ++I) {
         unsigned Reg = *I;
-        UnionGroups(Reg, 0);
+        Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
         KillIndices[Reg] = BB->size();
         DefIndices[Reg] = ~0u;
         // Repeat, for all aliases.
         for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
           unsigned AliasReg = *Alias;
-          UnionGroups(AliasReg, 0);
+          Classes[AliasReg] = reinterpret_cast<TargetRegisterClass *>(-1);
           KillIndices[AliasReg] = BB->size();
           DefIndices[AliasReg] = ~0u;
         }
@@ -456,13 +372,13 @@
   for (const unsigned *I = TRI->getCalleeSavedRegs(); *I; ++I) {
     unsigned Reg = *I;
     if (!IsReturnBlock && !Pristine.test(Reg)) continue;
-    UnionGroups(Reg, 0);
+    Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
     KillIndices[Reg] = BB->size();
     DefIndices[Reg] = ~0u;
     // Repeat, for all aliases.
     for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
       unsigned AliasReg = *Alias;
-      UnionGroups(AliasReg, 0);
+      Classes[AliasReg] = reinterpret_cast<TargetRegisterClass *>(-1);
       KillIndices[AliasReg] = BB->size();
       DefIndices[AliasReg] = ~0u;
     }
@@ -477,8 +393,8 @@
   // Build the scheduling graph.
   BuildSchedGraph(AA);
 
-  if (EnableAntiDepBreaking != "none") {
-    if (BreakAntiDependencies((EnableAntiDepBreaking == "all") ? false : true)) {
+  if (EnableAntiDepBreaking) {
+    if (BreakAntiDependencies()) {
       // We made changes. Update the dependency graph.
       // Theoretically we could update the graph in place:
       // When a live range is changed to use a different register, remove
@@ -508,27 +424,21 @@
 void SchedulePostRATDList::Observe(MachineInstr *MI, unsigned Count) {
   assert(Count < InsertPosIndex && "Instruction index out of expected range!");
 
-  DEBUG(errs() << "Observe: ");
-  DEBUG(MI->dump());
-
-  for (unsigned Reg = 0; Reg != TargetRegisterInfo::FirstVirtualRegister; ++Reg) {
-    // If Reg is current live, then mark that it can't be renamed as
-    // we don't know the extent of its live-range anymore (now that it
-    // has been scheduled). If it is not live but was defined in the
-    // previous schedule region, then set its def index to the most
-    // conservative location (i.e. the beginning of the previous
-    // schedule region).
-    if (IsLive(Reg)) {
-      DEBUG(if (GetGroup(Reg) != 0)
-              errs() << " " << TRI->getName(Reg) << "=g" << 
-                GetGroup(Reg) << "->g0(region live-out)");
-      UnionGroups(Reg, 0);
-    } else if ((DefIndices[Reg] < InsertPosIndex) && (DefIndices[Reg] >= Count)) {
-      DefIndices[Reg] = Count;
+  // Any register which was defined within the previous scheduling region
+  // may have been rescheduled and its lifetime may overlap with registers
+  // in ways not reflected in our current liveness state. For each such
+  // register, adjust the liveness state to be conservatively correct.
+  for (unsigned Reg = 0; Reg != TargetRegisterInfo::FirstVirtualRegister; ++Reg)
+    if (DefIndices[Reg] < InsertPosIndex && DefIndices[Reg] >= Count) {
+      assert(KillIndices[Reg] == ~0u && "Clobbered register is live!");
+      // Mark this register to be non-renamable.
+      Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
+      // Move the def index to the end of the previous region, to reflect
+      // that the def could theoretically have been scheduled at the end.
+      DefIndices[Reg] = InsertPosIndex;
     }
-  }
 
-  PrescanInstruction(MI, Count);
+  PrescanInstruction(MI);
   ScanInstruction(MI, Count);
 }
 
@@ -563,247 +473,156 @@
   return Next;
 }
 
-/// AntiDepPathStep - Return SUnit that SU has an anti-dependence on.
-static SDep *AntiDepPathStep(SUnit *SU) {
-  for (SUnit::pred_iterator P = SU->Preds.begin(), PE = SU->Preds.end();
-       P != PE; ++P) {
-    if (P->getKind() == SDep::Anti) {
-      return &*P;
+void SchedulePostRATDList::PrescanInstruction(MachineInstr *MI) {
+  // Scan the register operands for this instruction and update
+  // Classes and RegRefs.
+  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+    MachineOperand &MO = MI->getOperand(i);
+    if (!MO.isReg()) continue;
+    unsigned Reg = MO.getReg();
+    if (Reg == 0) continue;
+    const TargetRegisterClass *NewRC = 0;
+    
+    if (i < MI->getDesc().getNumOperands())
+      NewRC = MI->getDesc().OpInfo[i].getRegClass(TRI);
+
+    // For now, only allow the register to be changed if its register
+    // class is consistent across all uses.
+    if (!Classes[Reg] && NewRC)
+      Classes[Reg] = NewRC;
+    else if (!NewRC || Classes[Reg] != NewRC)
+      Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
+
+    // Now check for aliases.
+    for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
+      // If an alias of the reg is used during the live range, give up.
+      // Note that this allows us to skip checking if AntiDepReg
+      // overlaps with any of the aliases, among other things.
+      unsigned AliasReg = *Alias;
+      if (Classes[AliasReg]) {
+        Classes[AliasReg] = reinterpret_cast<TargetRegisterClass *>(-1);
+        Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
+      }
+    }
+
+    // If we're still willing to consider this register, note the reference.
+    if (Classes[Reg] != reinterpret_cast<TargetRegisterClass *>(-1))
+      RegRefs.insert(std::make_pair(Reg, &MO));
+
+    // It's not safe to change register allocation for source operands of
+    // that have special allocation requirements.
+    if (MO.isUse() && MI->getDesc().hasExtraSrcRegAllocReq()) {
+      if (KeepRegs.insert(Reg)) {
+        for (const unsigned *Subreg = TRI->getSubRegisters(Reg);
+             *Subreg; ++Subreg)
+          KeepRegs.insert(*Subreg);
+      }
     }
   }
-  return 0;
 }
 
-void SchedulePostRATDList::PrescanInstruction(MachineInstr *MI, unsigned Count) {
-  // Scan the register defs for this instruction and update
-  // live-ranges, groups and RegRefs.
+void SchedulePostRATDList::ScanInstruction(MachineInstr *MI,
+                                           unsigned Count) {
+  // Update liveness.
+  // Proceding upwards, registers that are defed but not used in this
+  // instruction are now dead.
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     MachineOperand &MO = MI->getOperand(i);
-    if (!MO.isReg() || !MO.isDef()) continue;
+    if (!MO.isReg()) continue;
     unsigned Reg = MO.getReg();
     if (Reg == 0) continue;
-    // Ignore two-addr defs for liveness...
+    if (!MO.isDef()) continue;
+    // Ignore two-addr defs.
     if (MI->isRegTiedToUseOperand(i)) continue;
 
-    // Update Def for Reg and subregs.
     DefIndices[Reg] = Count;
+    KillIndices[Reg] = ~0u;
+    assert(((KillIndices[Reg] == ~0u) !=
+            (DefIndices[Reg] == ~0u)) &&
+           "Kill and Def maps aren't consistent for Reg!");
+    KeepRegs.erase(Reg);
+    Classes[Reg] = 0;
+    RegRefs.erase(Reg);
+    // Repeat, for all subregs.
     for (const unsigned *Subreg = TRI->getSubRegisters(Reg);
          *Subreg; ++Subreg) {
       unsigned SubregReg = *Subreg;
       DefIndices[SubregReg] = Count;
+      KillIndices[SubregReg] = ~0u;
+      KeepRegs.erase(SubregReg);
+      Classes[SubregReg] = 0;
+      RegRefs.erase(SubregReg);
+    }
+    // Conservatively mark super-registers as unusable.
+    for (const unsigned *Super = TRI->getSuperRegisters(Reg);
+         *Super; ++Super) {
+      unsigned SuperReg = *Super;
+      Classes[SuperReg] = reinterpret_cast<TargetRegisterClass *>(-1);
     }
   }
-
-  DEBUG(errs() << "\tDef Groups:");
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     MachineOperand &MO = MI->getOperand(i);
-    if (!MO.isReg() || !MO.isDef()) continue;
+    if (!MO.isReg()) continue;
     unsigned Reg = MO.getReg();
     if (Reg == 0) continue;
+    if (!MO.isUse()) continue;
 
-    DEBUG(errs() << " " << TRI->getName(Reg) << "=g" << GetGroup(Reg)); 
-
-    // If MI's defs have special allocation requirement, don't allow
-    // any def registers to be changed. Also assume all registers
-    // defined in a call must not be changed (ABI).
-    if (MI->getDesc().isCall() || MI->getDesc().hasExtraDefRegAllocReq()) {
-      DEBUG(if (GetGroup(Reg) != 0) errs() << "->g0(alloc-req)");
-      UnionGroups(Reg, 0);
-    }
-
-    // Any subregisters that are live at this point are defined here,
-    // so group those subregisters with Reg.
-    for (const unsigned *Subreg = TRI->getSubRegisters(Reg);
-         *Subreg; ++Subreg) {
-      unsigned SubregReg = *Subreg;
-      if (IsLive(SubregReg)) {
-        UnionGroups(Reg, SubregReg);
-        DEBUG(errs() << "->g" << GetGroup(Reg) << "(via " << 
-              TRI->getName(SubregReg) << ")");
-      }
-    }
-    
-    // Note register reference...
-    const TargetRegisterClass *RC = NULL;
+    const TargetRegisterClass *NewRC = 0;
     if (i < MI->getDesc().getNumOperands())
-      RC = MI->getDesc().OpInfo[i].getRegClass(TRI);
-    RegisterReference RR = { &MO, RC };
-    RegRefs.insert(std::make_pair(Reg, RR));
-  }
+      NewRC = MI->getDesc().OpInfo[i].getRegClass(TRI);
 
-  DEBUG(errs() << '\n');
-}
+    // For now, only allow the register to be changed if its register
+    // class is consistent across all uses.
+    if (!Classes[Reg] && NewRC)
+      Classes[Reg] = NewRC;
+    else if (!NewRC || Classes[Reg] != NewRC)
+      Classes[Reg] = reinterpret_cast<TargetRegisterClass *>(-1);
 
-void SchedulePostRATDList::ScanInstruction(MachineInstr *MI,
-                                           unsigned Count) {
-  DEBUG(errs() << "\tUse Groups:");
+    RegRefs.insert(std::make_pair(Reg, &MO));
 
-  // Scan the register uses for this instruction and update
-  // live-ranges, groups and RegRefs.
-  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
-    MachineOperand &MO = MI->getOperand(i);
-    if (!MO.isReg() || !MO.isUse()) continue;
-    unsigned Reg = MO.getReg();
-    if (Reg == 0) continue;
-    
-    DEBUG(errs() << " " << TRI->getName(Reg) << "=g" << GetGroup(Reg)); 
-
-    // It wasn't previously live but now it is, this is a kill. Forget
-    // the previous live-range information and start a new live-range
-    // for the register.
-    if (!IsLive(Reg)) {
+    // It wasn't previously live but now it is, this is a kill.
+    if (KillIndices[Reg] == ~0u) {
       KillIndices[Reg] = Count;
       DefIndices[Reg] = ~0u;
-      RegRefs.erase(Reg);
-      LeaveGroup(Reg);
-      DEBUG(errs() << "->g" << GetGroup(Reg) << "(last-use)");
+          assert(((KillIndices[Reg] == ~0u) !=
+                  (DefIndices[Reg] == ~0u)) &&
+               "Kill and Def maps aren't consistent for Reg!");
     }
-    // Repeat, for subregisters.
-    for (const unsigned *Subreg = TRI->getSubRegisters(Reg);
-         *Subreg; ++Subreg) {
-      unsigned SubregReg = *Subreg;
-      if (!IsLive(SubregReg)) {
-        KillIndices[SubregReg] = Count;
-        DefIndices[SubregReg] = ~0u;
-        RegRefs.erase(SubregReg);
-        LeaveGroup(SubregReg);
-        DEBUG(errs() << "->g" << GetGroup(SubregReg) << "(last-use)");
+    // Repeat, for all aliases.
+    for (const unsigned *Alias = TRI->getAliasSet(Reg); *Alias; ++Alias) {
+      unsigned AliasReg = *Alias;
+      if (KillIndices[AliasReg] == ~0u) {
+        KillIndices[AliasReg] = Count;
+        DefIndices[AliasReg] = ~0u;
       }
     }
-
-    // If MI's uses have special allocation requirement, don't allow
-    // any use registers to be changed. Also assume all registers
-    // used in a call must not be changed (ABI).
-    if (MI->getDesc().isCall() || MI->getDesc().hasExtraSrcRegAllocReq()) {
-      DEBUG(if (GetGroup(Reg) != 0) errs() << "->g0(alloc-req)");
-      UnionGroups(Reg, 0);
-    }
-
-    // Note register reference...
-    const TargetRegisterClass *RC = NULL;
-    if (i < MI->getDesc().getNumOperands())
-      RC = MI->getDesc().OpInfo[i].getRegClass(TRI);
-    RegisterReference RR = { &MO, RC };
-    RegRefs.insert(std::make_pair(Reg, RR));
-  }
-  
-  DEBUG(errs() << '\n');
-
-  // Form a group of all defs and uses of a KILL instruction to ensure
-  // that all registers are renamed as a group.
-  if (MI->getOpcode() == TargetInstrInfo::KILL) {
-    unsigned FirstReg = 0;
-    for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
-      MachineOperand &MO = MI->getOperand(i);
-      if (!MO.isReg()) continue;
-      unsigned Reg = MO.getReg();
-      if (Reg == 0) continue;
-      
-      if (FirstReg != 0)
-        UnionGroups(FirstReg, Reg);
-      FirstReg = Reg;
-    }
-
-    DEBUG(if (FirstReg != 0) errs() << "\tKill Group: g" << 
-                               GetGroup(FirstReg) << '\n'); 
   }
 }
 
-unsigned SchedulePostRATDList::FindSuitableFreeRegister(unsigned AntiDepReg) {
-  // Collect all registers in the same group as AntiDepReg. These all
-  // need to be renamed together if we are to break the
-  // anti-dependence.
-  std::vector<unsigned> Regs;
-  GetGroupRegs(GetGroup(AntiDepReg), Regs);
-
-  DEBUG(errs() << "\tRename Register Group:");
-  DEBUG(for (unsigned i = 0, e = Regs.size(); i != e; ++i)
-          DEBUG(errs() << " " << TRI->getName(Regs[i])));
-  DEBUG(errs() << "\n");
-
-  // If there is a single register that needs to be renamed then we
-  // can do it ourselves.
-  if (Regs.size() == 1) {
-    assert(Regs[0] == AntiDepReg && "Register group does not contain register!");
-
-    // Check all references that need rewriting. Gather up all the
-    // register classes for the register references.
-    const TargetRegisterClass *FirstRC = NULL;
-    std::set<const TargetRegisterClass *> RCs;
-    std::pair<std::multimap<unsigned, RegisterReference>::iterator,
-      std::multimap<unsigned, RegisterReference>::iterator>
-      Range = RegRefs.equal_range(AntiDepReg);
-    for (std::multimap<unsigned, RegisterReference>::iterator
-           Q = Range.first, QE = Range.second; Q != QE; ++Q) {
-      const TargetRegisterClass *RC = Q->second.RC;
-      if (RC == NULL) continue;
-      if (FirstRC == NULL)
-        FirstRC = RC;
-      else if (FirstRC != RC)
-        RCs.insert(RC);
-    }
-    
-    if (FirstRC == NULL)
-      return 0;
-
-    DEBUG(errs() << "\tChecking Regclasses: " << FirstRC->getName());
-    DEBUG(for (std::set<const TargetRegisterClass *>::iterator S = 
-                 RCs.begin(), E = RCs.end(); S != E; ++S)
-            errs() << " " << (*S)->getName());
-    DEBUG(errs() << '\n');
-
-    // Using the allocation order for one of the register classes,
-    // find the first register that belongs to all the register
-    // classes that is available over the liverange of the register.
-    DEBUG(errs() << "\tFind Register:");
-    for (TargetRegisterClass::iterator R = FirstRC->allocation_order_begin(MF),
-           RE = FirstRC->allocation_order_end(MF); R != RE; ++R) {
-      unsigned NewReg = *R;
-      
-      // Don't replace a register with itself.
-      if (NewReg == AntiDepReg) continue;
-
-      DEBUG(errs() << " " << TRI->getName(NewReg));
-      
-      // Make sure NewReg is in all required register classes.
-      for (std::set<const TargetRegisterClass *>::iterator S = 
-             RCs.begin(), E = RCs.end(); S != E; ++S) {
-        const TargetRegisterClass *RC = *S;
-        if (!RC->contains(NewReg)) {
-          DEBUG(errs() << "(not in " << RC->getName() << ")");
-          NewReg = 0;
-          break;
-        }
-      }
-
-      // If NewReg is dead and NewReg's most recent def is not before
-      // AntiDepReg's kill, it's safe to replace AntiDepReg with
-      // NewReg. We must also check all subregisters of NewReg.
-      if (IsLive(NewReg) || (KillIndices[AntiDepReg] > DefIndices[NewReg])) {
-        DEBUG(errs() << "(live)");
-        continue;
-      }
-      {
-        bool found = false;
-        for (const unsigned *Subreg = TRI->getSubRegisters(NewReg);
-             *Subreg; ++Subreg) {
-          unsigned SubregReg = *Subreg;
-          if (IsLive(SubregReg) || (KillIndices[AntiDepReg] > DefIndices[SubregReg])) {
-            DEBUG(errs() << "(subreg " << TRI->getName(SubregReg) << " live)");
-            found = true;
-          }
-        }
-        if (found)
-          continue;
-      }
-      
-      if (NewReg != 0) { 
-        DEBUG(errs() << '\n');
-        return NewReg;
-      }
-    }
-
-    DEBUG(errs() << '\n');
+unsigned
+SchedulePostRATDList::findSuitableFreeRegister(unsigned AntiDepReg,
+                                               unsigned LastNewReg,
+                                               const TargetRegisterClass *RC) {
+  for (TargetRegisterClass::iterator R = RC->allocation_order_begin(MF),
+       RE = RC->allocation_order_end(MF); R != RE; ++R) {
+    unsigned NewReg = *R;
+    // Don't replace a register with itself.
+    if (NewReg == AntiDepReg) continue;
+    // Don't replace a register with one that was recently used to repair
+    // an anti-dependence with this AntiDepReg, because that would
+    // re-introduce that anti-dependence.
+    if (NewReg == LastNewReg) continue;
+    // If NewReg is dead and NewReg's most recent def is not before
+    // AntiDepReg's kill, it's safe to replace AntiDepReg with NewReg.
+    assert(((KillIndices[AntiDepReg] == ~0u) != (DefIndices[AntiDepReg] == ~0u)) &&
+           "Kill and Def maps aren't consistent for AntiDepReg!");
+    assert(((KillIndices[NewReg] == ~0u) != (DefIndices[NewReg] == ~0u)) &&
+           "Kill and Def maps aren't consistent for NewReg!");
+    if (KillIndices[NewReg] != ~0u ||
+        Classes[NewReg] == reinterpret_cast<TargetRegisterClass *>(-1) ||
+        KillIndices[AntiDepReg] > DefIndices[NewReg])
+      continue;
+    return NewReg;
   }
 
   // No registers are free and available!
@@ -813,153 +632,139 @@
 /// BreakAntiDependencies - Identifiy anti-dependencies along the critical path
 /// of the ScheduleDAG and break them by renaming registers.
 ///
-bool SchedulePostRATDList::BreakAntiDependencies(bool CriticalPathOnly) {
+bool SchedulePostRATDList::BreakAntiDependencies() {
   // The code below assumes that there is at least one instruction,
   // so just duck out immediately if the block is empty.
   if (SUnits.empty()) return false;
 
-  // If breaking anti-dependencies only along the critical path, track
-  // progress along the critical path through the SUnit graph as we
-  // walk the instructions.
-  SUnit *CriticalPathSU = 0;
-  MachineInstr *CriticalPathMI = 0;
-  
-  // If breaking all anti-dependencies need a map from MI to SUnit.
-  std::map<MachineInstr *, SUnit *> MISUnitMap;
-
   // Find the node at the bottom of the critical path.
-  if (CriticalPathOnly) {
-    SUnit *Max = 0;
-    for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
-      SUnit *SU = &SUnits[i];
-      if (!Max || SU->getDepth() + SU->Latency > Max->getDepth() + Max->Latency)
-        Max = SU;
-    }
-
-    DEBUG(errs() << "Critical path has total latency "
-          << (Max->getDepth() + Max->Latency) << "\n");
-    CriticalPathSU = Max;
-    CriticalPathMI = CriticalPathSU->getInstr();
-  } else {
-    for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
-      SUnit *SU = &SUnits[i];
-      MISUnitMap.insert(std::pair<MachineInstr *, SUnit *>(SU->getInstr(), SU));
-    }
-    DEBUG(errs() << "Breaking all anti-dependencies\n");
+  SUnit *Max = 0;
+  for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
+    SUnit *SU = &SUnits[i];
+    if (!Max || SU->getDepth() + SU->Latency > Max->getDepth() + Max->Latency)
+      Max = SU;
   }
 
-#ifndef NDEBUG 
+#ifndef NDEBUG
   {
+    DEBUG(errs() << "Critical path has total latency "
+          << (Max->getDepth() + Max->Latency) << "\n");
     DEBUG(errs() << "Available regs:");
     for (unsigned Reg = 0; Reg < TRI->getNumRegs(); ++Reg) {
-      if (!IsLive(Reg))
+      if (KillIndices[Reg] == ~0u)
         DEBUG(errs() << " " << TRI->getName(Reg));
     }
     DEBUG(errs() << '\n');
   }
-  std::string dbgStr;
 #endif
 
-  // Attempt to break anti-dependence edges. Walk the instructions
-  // from the bottom up, tracking information about liveness as we go
-  // to help determine which registers are available.
+  // Track progress along the critical path through the SUnit graph as we walk
+  // the instructions.
+  SUnit *CriticalPathSU = Max;
+  MachineInstr *CriticalPathMI = CriticalPathSU->getInstr();
+
+  // Consider this pattern:
+  //   A = ...
+  //   ... = A
+  //   A = ...
+  //   ... = A
+  //   A = ...
+  //   ... = A
+  //   A = ...
+  //   ... = A
+  // There are three anti-dependencies here, and without special care,
+  // we'd break all of them using the same register:
+  //   A = ...
+  //   ... = A
+  //   B = ...
+  //   ... = B
+  //   B = ...
+  //   ... = B
+  //   B = ...
+  //   ... = B
+  // because at each anti-dependence, B is the first register that
+  // isn't A which is free.  This re-introduces anti-dependencies
+  // at all but one of the original anti-dependencies that we were
+  // trying to break.  To avoid this, keep track of the most recent
+  // register that each register was replaced with, avoid
+  // using it to repair an anti-dependence on the same register.
+  // This lets us produce this:
+  //   A = ...
+  //   ... = A
+  //   B = ...
+  //   ... = B
+  //   C = ...
+  //   ... = C
+  //   B = ...
+  //   ... = B
+  // This still has an anti-dependence on B, but at least it isn't on the
+  // original critical path.
+  //
+  // TODO: If we tracked more than one register here, we could potentially
+  // fix that remaining critical edge too. This is a little more involved,
+  // because unlike the most recent register, less recent registers should
+  // still be considered, though only if no other registers are available.
+  unsigned LastNewReg[TargetRegisterInfo::FirstVirtualRegister] = {};
+
+  // Attempt to break anti-dependence edges on the critical path. Walk the
+  // instructions from the bottom up, tracking information about liveness
+  // as we go to help determine which registers are available.
   bool Changed = false;
   unsigned Count = InsertPosIndex - 1;
   for (MachineBasicBlock::iterator I = InsertPos, E = Begin;
        I != E; --Count) {
     MachineInstr *MI = --I;
 
-    DEBUG(errs() << "Anti: ");
-    DEBUG(MI->dump());
-
-    // Process the defs in MI...
-    PrescanInstruction(MI, Count);
-
-    // Check if this instruction has an anti-dependence that we may be
-    // able to break. If it is, set AntiDepReg to the non-zero
-    // register associated with the anti-dependence.
+    // Check if this instruction has a dependence on the critical path that
+    // is an anti-dependence that we may be able to break. If it is, set
+    // AntiDepReg to the non-zero register associated with the anti-dependence.
     //
-    unsigned AntiDepReg = 0;
-  
-    // Limiting our attention to the critical path is a heuristic to avoid
+    // We limit our attention to the critical path as a heuristic to avoid
     // breaking anti-dependence edges that aren't going to significantly
     // impact the overall schedule. There are a limited number of registers
     // and we want to save them for the important edges.
     // 
-    // We can also break all anti-dependencies because they can
-    // occur along the non-critical path but are still detrimental for
-    // scheduling.
-    // 
     // TODO: Instructions with multiple defs could have multiple
     // anti-dependencies. The current code here only knows how to break one
     // edge per instruction. Note that we'd have to be able to break all of
     // the anti-dependencies in an instruction in order to be effective.
-    if (!CriticalPathOnly || (MI == CriticalPathMI)) {
-      DEBUG(dbgStr.clear());
-
-      SUnit *PathSU;
-      SDep *Edge;
-      if (CriticalPathOnly) {
-        PathSU = CriticalPathSU;
-        Edge = CriticalPathStep(PathSU);
-      } else {
-        PathSU = MISUnitMap[MI];
-        Edge = (PathSU) ? AntiDepPathStep(PathSU) : 0;
-      }
-      
-      if (Edge) {
+    unsigned AntiDepReg = 0;
+    if (MI == CriticalPathMI) {
+      if (SDep *Edge = CriticalPathStep(CriticalPathSU)) {
         SUnit *NextSU = Edge->getSUnit();
 
-        // Only consider anti-dependence edges, and ignore KILL
-        // instructions (they form a group in ScanInstruction but
-        // don't cause any anti-dependence breaking themselves)
-        if ((Edge->getKind() == SDep::Anti) &&
-            (MI->getOpcode() != TargetInstrInfo::KILL)) {
+        // Only consider anti-dependence edges.
+        if (Edge->getKind() == SDep::Anti) {
           AntiDepReg = Edge->getReg();
-          DEBUG(dbgStr += "\tAntidep reg: ");
-          DEBUG(dbgStr += TRI->getName(AntiDepReg));
           assert(AntiDepReg != 0 && "Anti-dependence on reg0?");
-          if (!AllocatableSet.test(AntiDepReg)) {
+          if (!AllocatableSet.test(AntiDepReg))
             // Don't break anti-dependencies on non-allocatable registers.
-            DEBUG(dbgStr += " (non-allocatable)");
             AntiDepReg = 0;
-          } else {
-            int OpIdx = MI->findRegisterDefOperandIdx(AntiDepReg);
-            assert(OpIdx != -1 && "Can't find index for defined register operand");
-            if (MI->isRegTiedToUseOperand(OpIdx)) {
-              // If the anti-dep register is tied to a use, then don't try to
-              // change it. It will be changed along with the use if required
-              // to break an earlier antidep.
-              DEBUG(dbgStr += " (tied-to-use)");
-              AntiDepReg = 0;
-            } else {
-              // If the SUnit has other dependencies on the SUnit that
-              // it anti-depends on, don't bother breaking the
-              // anti-dependency since those edges would prevent such
-              // units from being scheduled past each other
-              // regardless.
-              //
-              // Also, if there are dependencies on other SUnits with
-              // the same register as the anti-dependency, don't
-              // attempt to break it.
-              for (SUnit::pred_iterator P = PathSU->Preds.begin(),
-                     PE = PathSU->Preds.end(); P != PE; ++P) {
-                if (P->getSUnit() == NextSU ?
+          else if (KeepRegs.count(AntiDepReg))
+            // Don't break anti-dependencies if an use down below requires
+            // this exact register.
+            AntiDepReg = 0;
+          else {
+            // If the SUnit has other dependencies on the SUnit that it
+            // anti-depends on, don't bother breaking the anti-dependency
+            // since those edges would prevent such units from being
+            // scheduled past each other regardless.
+            //
+            // Also, if there are dependencies on other SUnits with the
+            // same register as the anti-dependency, don't attempt to
+            // break it.
+            for (SUnit::pred_iterator P = CriticalPathSU->Preds.begin(),
+                 PE = CriticalPathSU->Preds.end(); P != PE; ++P)
+              if (P->getSUnit() == NextSU ?
                     (P->getKind() != SDep::Anti || P->getReg() != AntiDepReg) :
                     (P->getKind() == SDep::Data && P->getReg() == AntiDepReg)) {
-                  DEBUG(dbgStr += " (real dependency)");
-                  AntiDepReg = 0;
-                  break;
-                }
+                AntiDepReg = 0;
+                break;
               }
-            }
           }
         }
-        
-        if (CriticalPathOnly) {
-          CriticalPathSU = NextSU;
-          CriticalPathMI = CriticalPathSU->getInstr();
-        }
+        CriticalPathSU = NextSU;
+        CriticalPathMI = CriticalPathSU->getInstr();
       } else {
         // We've reached the end of the critical path.
         CriticalPathSU = 0;
@@ -967,52 +772,77 @@
       }
     }
 
-    // Determine AntiDepReg's register group.
-    const unsigned GroupIndex = AntiDepReg != 0 ? GetGroup(AntiDepReg) : 0;
-    if (GroupIndex == 0) {
-      DEBUG(if (AntiDepReg != 0) dbgStr += " (zero group)");
+    PrescanInstruction(MI);
+
+    if (MI->getDesc().hasExtraDefRegAllocReq())
+      // If this instruction's defs have special allocation requirement, don't
+      // break this anti-dependency.
       AntiDepReg = 0;
+    else if (AntiDepReg) {
+      // If this instruction has a use of AntiDepReg, breaking it
+      // is invalid.
+      for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+        MachineOperand &MO = MI->getOperand(i);
+        if (!MO.isReg()) continue;
+        unsigned Reg = MO.getReg();
+        if (Reg == 0) continue;
+        if (MO.isUse() && AntiDepReg == Reg) {
+          AntiDepReg = 0;
+          break;
+        }
+      }
     }
 
-    DEBUG(if (!dbgStr.empty()) errs() << dbgStr << '\n');
+    // Determine AntiDepReg's register class, if it is live and is
+    // consistently used within a single class.
+    const TargetRegisterClass *RC = AntiDepReg != 0 ? Classes[AntiDepReg] : 0;
+    assert((AntiDepReg == 0 || RC != NULL) &&
+           "Register should be live if it's causing an anti-dependence!");
+    if (RC == reinterpret_cast<TargetRegisterClass *>(-1))
+      AntiDepReg = 0;
 
-    // Look for a suitable register to use to break the anti-dependence.
+    // Look for a suitable register to use to break the anti-depenence.
+    //
+    // TODO: Instead of picking the first free register, consider which might
+    // be the best.
     if (AntiDepReg != 0) {
-      if (unsigned NewReg = FindSuitableFreeRegister(AntiDepReg)) {
-        DEBUG(errs() << "\tBreaking anti-dependence edge on "
+      if (unsigned NewReg = findSuitableFreeRegister(AntiDepReg,
+                                                     LastNewReg[AntiDepReg],
+                                                     RC)) {
+        DEBUG(errs() << "Breaking anti-dependence edge on "
               << TRI->getName(AntiDepReg)
               << " with " << RegRefs.count(AntiDepReg) << " references"
               << " using " << TRI->getName(NewReg) << "!\n");
 
         // Update the references to the old register to refer to the new
         // register.
-        std::pair<std::multimap<unsigned, RegisterReference>::iterator,
-                  std::multimap<unsigned, RegisterReference>::iterator>
+        std::pair<std::multimap<unsigned, MachineOperand *>::iterator,
+                  std::multimap<unsigned, MachineOperand *>::iterator>
            Range = RegRefs.equal_range(AntiDepReg);
-        for (std::multimap<unsigned, RegisterReference>::iterator
+        for (std::multimap<unsigned, MachineOperand *>::iterator
              Q = Range.first, QE = Range.second; Q != QE; ++Q)
-          Q->second.Operand->setReg(NewReg);
+          Q->second->setReg(NewReg);
 
         // We just went back in time and modified history; the
-        // liveness information for the anti-dependence reg is now
+        // liveness information for the anti-depenence reg is now
         // inconsistent. Set the state as if it were dead.
-        // FIXME forall in group
-        UnionGroups(NewReg, 0);
-        RegRefs.erase(NewReg);
+        Classes[NewReg] = Classes[AntiDepReg];
         DefIndices[NewReg] = DefIndices[AntiDepReg];
         KillIndices[NewReg] = KillIndices[AntiDepReg];
+        assert(((KillIndices[NewReg] == ~0u) !=
+                (DefIndices[NewReg] == ~0u)) &&
+             "Kill and Def maps aren't consistent for NewReg!");
 
-        // FIXME forall in group
-        UnionGroups(AntiDepReg, 0);
-        RegRefs.erase(AntiDepReg);
+        Classes[AntiDepReg] = 0;
         DefIndices[AntiDepReg] = KillIndices[AntiDepReg];
         KillIndices[AntiDepReg] = ~0u;
         assert(((KillIndices[AntiDepReg] == ~0u) !=
                 (DefIndices[AntiDepReg] == ~0u)) &&
              "Kill and Def maps aren't consistent for AntiDepReg!");
 
+        RegRefs.erase(AntiDepReg);
         Changed = true;
-        ++NumFixedAnti;
+        LastNewReg[AntiDepReg] = NewReg;
       }
     }
 
@@ -1091,7 +921,7 @@
     }
   }
 
-  if (AllDead)
+  if(AllDead)
     MO.setIsKill(true);
   return false;
 }

Modified: llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll?rev=84727&r1=84726&r2=84727&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll (original)
+++ llvm/trunk/test/CodeGen/Thumb2/2009-09-28-ITBlockBug.ll Tue Oct 20 20:44:44 2009
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=thumbv7-apple-darwin -mcpu=cortex-a8 -break-anti-dependencies=none | FileCheck %s
+; RUN: llc < %s -mtriple=thumbv7-apple-darwin -mcpu=cortex-a8 | FileCheck %s
 
 %struct.pix_pos = type { i32, i32, i32, i32, i32, i32 }
 

Modified: llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll?rev=84727&r1=84726&r2=84727&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll (original)
+++ llvm/trunk/test/CodeGen/X86/break-anti-dependencies.ll Tue Oct 20 20:44:44 2009
@@ -1,7 +1,7 @@
-; RUN: llc < %s -march=x86-64 -post-RA-scheduler -break-anti-dependencies=none > %t
+; RUN: llc < %s -march=x86-64 -post-RA-scheduler -break-anti-dependencies=false > %t
 ; RUN:   grep {%xmm0} %t | count 14
 ; RUN:   not grep {%xmm1} %t
-; RUN: llc < %s -march=x86-64 -post-RA-scheduler -break-anti-dependencies=critical > %t
+; RUN: llc < %s -march=x86-64 -post-RA-scheduler -break-anti-dependencies > %t
 ; RUN:   grep {%xmm0} %t | count 7
 ; RUN:   grep {%xmm1} %t | count 7
 

Modified: llvm/trunk/test/CodeGen/X86/sse3.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse3.ll?rev=84727&r1=84726&r2=84727&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/sse3.ll (original)
+++ llvm/trunk/test/CodeGen/X86/sse3.ll Tue Oct 20 20:44:44 2009
@@ -51,9 +51,9 @@
 	%tmp = shufflevector <8 x i16> %A, <8 x i16> %A, <8 x i32> < i32 8, i32 3, i32 2, i32 13, i32 7, i32 6, i32 5, i32 4 >
 	ret <8 x i16> %tmp
 ; X64: t3:
-; X64: 	pshuflw	$44, %xmm0, %xmm1
 ; X64: 	pextrw	$5, %xmm0, %eax
-; X64: 	pshufhw	$27, %xmm1, %xmm0
+; X64: 	pshuflw	$44, %xmm0, %xmm0
+; X64: 	pshufhw	$27, %xmm0, %xmm0
 ; X64: 	pinsrw	$3, %eax, %xmm0
 ; X64: 	ret
 }
@@ -168,11 +168,11 @@
         ret void
 ; X64: 	t10:
 ; X64: 		pextrw	$4, %xmm0, %eax
-; X64: 		pextrw	$6, %xmm0, %ecx
+; X64: 		pextrw	$6, %xmm0, %edx
 ; X64: 		movlhps	%xmm1, %xmm1
 ; X64: 		pshuflw	$8, %xmm1, %xmm1
 ; X64: 		pinsrw	$2, %eax, %xmm1
-; X64: 		pinsrw	$3, %ecx, %xmm1
+; X64: 		pinsrw	$3, %edx, %xmm1
 }
 
 
@@ -250,9 +250,9 @@
         %tmp9 = shufflevector <16 x i8> %tmp8, <16 x i8> %T0,  <16 x i32> < i32 0, i32 1, i32 2, i32 17,  i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef , i32 undef >
         ret <16 x i8> %tmp9
 ; X64: 	t16:
-; X64: 		pextrw	$8, %xmm0, %ecx
 ; X64: 		pinsrw	$0, %eax, %xmm1
-; X64: 		pinsrw	$1, %ecx, %xmm1
+; X64: 		pextrw	$8, %xmm0, %eax
+; X64: 		pinsrw	$1, %eax, %xmm1
 ; X64: 		pextrw	$1, %xmm1, %ecx
 ; X64: 		movd	%xmm1, %edx
 ; X64: 		pinsrw	$0, %edx, %xmm1




From gohman at apple.com  Tue Oct 20 20:59:09 2009
From: gohman at apple.com (Dan Gohman)
Date: Tue, 20 Oct 2009 18:59:09 -0700
Subject: [llvm-commits] [llvm] r84691 -
	/llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
In-Reply-To: <200910202250.n9KMohf8023906@zion.cs.uiuc.edu>
References: <200910202250.n9KMohf8023906@zion.cs.uiuc.edu>
Message-ID: <C59D7655-F361-4B74-93EB-C9C5D4D680AD@apple.com>

Hi David,

These patches broke several tests, including the llvm-gcc
bootstrap, so I've reverted them for now.

Dan

On Oct 20, 2009, at 3:50 PM, David Goodwin wrote:

> Author: david_goodwin
> Date: Tue Oct 20 17:50:43 2009
> New Revision: 84691
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=84691&view=rev
> Log:
> Respect src register allocation requirements when breaking anti-dependencies. Remove some dead code.
> 
> Modified:
>    llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
> 
> Modified: llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp?rev=84691&r1=84690&r2=84691&view=diff




From foldr at codedgers.com  Tue Oct 20 21:13:14 2009
From: foldr at codedgers.com (Mikhail Glushenkov)
Date: Wed, 21 Oct 2009 02:13:14 -0000
Subject: [llvm-commits] [llvm] r84728 -
	/llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp
Message-ID: <200910210213.n9L2DE0s001964@zion.cs.uiuc.edu>

Author: foldr
Date: Tue Oct 20 21:13:13 2009
New Revision: 84728

URL: http://llvm.org/viewvc/llvm-project?rev=84728&view=rev
Log:
Implement any_[not_]empty and list versions of switch_on and [not_]empty.

Useful for OptionPreprocessor.

Modified:
    llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp

Modified: llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp?rev=84728&r1=84727&r2=84728&view=diff

==============================================================================
--- llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp (original)
+++ llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp Tue Oct 20 21:13:13 2009
@@ -146,6 +146,18 @@
     throw ErrorString;
 }
 
+// apply is needed because C++'s syntax doesn't let us construct a function
+// object and call it in the same statement.
+template<typename F, typename T0>
+void apply(F Fun, T0& Arg0) {
+  return Fun(Arg0);
+}
+
+template<typename F, typename T0, typename T1>
+void apply(F Fun, T0& Arg0, T1& Arg1) {
+  return Fun(Arg0, Arg1);
+}
+
 //===----------------------------------------------------------------------===//
 /// Back-end specific code
 
@@ -157,6 +169,10 @@
   enum OptionType { Alias, Switch, Parameter, ParameterList,
                     Prefix, PrefixList};
 
+  bool IsAlias(OptionType t) {
+    return (t == Alias);
+  }
+
   bool IsList (OptionType t) {
     return (t == ParameterList || t == PrefixList);
   }
@@ -245,12 +261,12 @@
   bool isReallyHidden() const;
   void setReallyHidden();
 
-  bool isParameter() const
-  { return OptionType::IsParameter(this->Type); }
-
   bool isSwitch() const
   { return OptionType::IsSwitch(this->Type); }
 
+  bool isParameter() const
+  { return OptionType::IsParameter(this->Type); }
+
   bool isList() const
   { return OptionType::IsList(this->Type); }
 
@@ -272,7 +288,7 @@
 }
 
 bool OptionDescription::isAlias() const {
-  return Type == OptionType::Alias;
+  return OptionType::IsAlias(this->Type);
 }
 
 bool OptionDescription::isMultiVal() const {
@@ -365,9 +381,14 @@
 public:
   /// FindOption - exception-throwing wrapper for find().
   const OptionDescription& FindOption(const std::string& OptName) const;
-  /// FindSwitch - wrapper for FindOption that throws in case the option is not
-  /// a switch.
+
+  // Wrappers for FindOption that throw an exception in case the option has a
+  // wrong type.
   const OptionDescription& FindSwitch(const std::string& OptName) const;
+  const OptionDescription& FindParameter(const std::string& OptName) const;
+  const OptionDescription& FindList(const std::string& OptName) const;
+  const OptionDescription&
+  FindListOrParameter(const std::string& OptName) const;
 
   /// insertDescription - Insert new OptionDescription into
   /// OptionDescriptions list
@@ -380,8 +401,7 @@
 };
 
 const OptionDescription&
-OptionDescriptions::FindOption(const std::string& OptName) const
-{
+OptionDescriptions::FindOption(const std::string& OptName) const {
   const_iterator I = Descriptions.find(OptName);
   if (I != Descriptions.end())
     return I->second;
@@ -390,16 +410,39 @@
 }
 
 const OptionDescription&
-OptionDescriptions::FindSwitch(const std::string& OptName) const
-{
+OptionDescriptions::FindSwitch(const std::string& OptName) const {
   const OptionDescription& OptDesc = this->FindOption(OptName);
   if (!OptDesc.isSwitch())
     throw OptName + ": incorrect option type - should be a switch!";
   return OptDesc;
 }
 
-void OptionDescriptions::InsertDescription (const OptionDescription& o)
-{
+const OptionDescription&
+OptionDescriptions::FindList(const std::string& OptName) const {
+  const OptionDescription& OptDesc = this->FindOption(OptName);
+  if (!OptDesc.isList())
+    throw OptName + ": incorrect option type - should be a list!";
+  return OptDesc;
+}
+
+const OptionDescription&
+OptionDescriptions::FindParameter(const std::string& OptName) const {
+  const OptionDescription& OptDesc = this->FindOption(OptName);
+  if (!OptDesc.isParameter())
+    throw OptName + ": incorrect option type - should be a parameter!";
+  return OptDesc;
+}
+
+const OptionDescription&
+OptionDescriptions::FindListOrParameter(const std::string& OptName) const {
+  const OptionDescription& OptDesc = this->FindOption(OptName);
+  if (!OptDesc.isList() && !OptDesc.isParameter())
+    throw OptName
+      + ": incorrect option type - should be a list or parameter!";
+  return OptDesc;
+}
+
+void OptionDescriptions::InsertDescription (const OptionDescription& o) {
   container_type::iterator I = Descriptions.find(o.Name);
   if (I != Descriptions.end()) {
     OptionDescription& D = I->second;
@@ -1067,49 +1110,57 @@
   return false;
 }
 
-/// EmitCaseTest1ArgStr - Helper function used by EmitCaseTest1Arg();
-bool EmitCaseTest1ArgStr(const std::string& TestName,
-                         const DagInit& d,
-                         const OptionDescriptions& OptDescs,
-                         raw_ostream& O) {
-  const std::string& OptName = InitPtrToString(d.getArg(0));
+/// EmitListTest - Helper function used by EmitCaseTest1ArgList().
+template <typename F>
+void EmitListTest(const ListInit& L, const char* LogicOp,
+                  F Callback, raw_ostream& O)
+{
+  // This is a lot like EmitLogicalOperationTest, but works on ListInits instead
+  // of Dags...
+  bool isFirst = true;
+  for (ListInit::const_iterator B = L.begin(), E = L.end(); B != E; ++B) {
+    if (isFirst)
+      isFirst = false;
+    else
+      O << " || ";
+    Callback(InitPtrToString(*B), O);
+  }
+}
 
-  if (TestName == "switch_on") {
-    const OptionDescription& OptDesc = OptDescs.FindSwitch(OptName);
+// Callbacks for use with EmitListTest.
+
+class EmitSwitchOn {
+  const OptionDescriptions& OptDescs_;
+public:
+  EmitSwitchOn(const OptionDescriptions& OptDescs) : OptDescs_(OptDescs)
+  {}
+
+  void operator()(const std::string& OptName, raw_ostream& O) const {
+    const OptionDescription& OptDesc = OptDescs_.FindSwitch(OptName);
     O << OptDesc.GenVariableName();
-    return true;
-  }
-  else if (TestName == "input_languages_contain") {
-    O << "InLangs.count(\"" << OptName << "\") != 0";
-    return true;
   }
-  else if (TestName == "in_language") {
-    // This works only for single-argument Tool::GenerateAction. Join
-    // tools can process several files in different languages simultaneously.
+};
 
-    // TODO: make this work with Edge::Weight (if possible).
-    O << "LangMap.GetLanguage(inFile) == \"" << OptName << '\"';
-    return true;
-  }
-  else if (TestName == "not_empty" || TestName == "empty") {
-    const char* Test = (TestName == "empty") ? "" : "!";
+class EmitEmptyTest {
+  bool EmitNegate_;
+  const OptionDescriptions& OptDescs_;
+public:
+  EmitEmptyTest(bool EmitNegate, const OptionDescriptions& OptDescs)
+    : EmitNegate_(EmitNegate), OptDescs_(OptDescs)
+  {}
 
+  void operator()(const std::string& OptName, raw_ostream& O) const {
+    const char* Neg = (EmitNegate_ ? "!" : "");
     if (OptName == "o") {
-      O << Test << "OutputFilename.empty()";
-      return true;
+      O << Neg << "OutputFilename.empty()";
     }
     else {
-      const OptionDescription& OptDesc = OptDescs.FindOption(OptName);
-      if (OptDesc.isSwitch())
-        throw OptName
-          + ": incorrect option type - should be a list or parameter!";
-      O << Test << OptDesc.GenVariableName() << ".empty()";
-      return true;
+      const OptionDescription& OptDesc = OptDescs_.FindListOrParameter(OptName);
+      O << Neg << OptDesc.GenVariableName() << ".empty()";
     }
   }
+};
 
-  return false;
-}
 
 /// EmitCaseTest1ArgList - Helper function used by EmitCaseTest1Arg();
 bool EmitCaseTest1ArgList(const std::string& TestName,
@@ -1119,23 +1170,61 @@
   const ListInit& L = *static_cast<ListInit*>(d.getArg(0));
 
   if (TestName == "any_switch_on") {
-    bool isFirst = true;
+    EmitListTest(L, "||", EmitSwitchOn(OptDescs), O);
+    return true;
+  }
+  else if (TestName == "switch_on") {
+    EmitListTest(L, "&&", EmitSwitchOn(OptDescs), O);
+    return true;
+  }
+  else if (TestName == "any_not_empty") {
+    EmitListTest(L, "||", EmitEmptyTest(true, OptDescs), O);
+    return true;
+  }
+  else if (TestName == "any_empty") {
+    EmitListTest(L, "||", EmitEmptyTest(false, OptDescs), O);
+    return true;
+  }
+  else if (TestName == "not_empty") {
+    EmitListTest(L, "&&", EmitEmptyTest(true, OptDescs), O);
+    return true;
+  }
+  else if (TestName == "empty") {
+    EmitListTest(L, "&&", EmitEmptyTest(false, OptDescs), O);
+    return true;
+  }
 
-    for (ListInit::const_iterator B = L.begin(), E = L.end(); B != E; ++B) {
-      const std::string& OptName = InitPtrToString(*B);
-      const OptionDescription& OptDesc = OptDescs.FindSwitch(OptName);
+  return false;
+}
 
-      if (isFirst)
-        isFirst = false;
-      else
-        O << " || ";
-      O << OptDesc.GenVariableName();
-    }
+/// EmitCaseTest1ArgStr - Helper function used by EmitCaseTest1Arg();
+bool EmitCaseTest1ArgStr(const std::string& TestName,
+                         const DagInit& d,
+                         const OptionDescriptions& OptDescs,
+                         raw_ostream& O) {
+  const std::string& OptName = InitPtrToString(d.getArg(0));
 
+  if (TestName == "switch_on") {
+    apply(EmitSwitchOn(OptDescs), OptName, O);
+    return true;
+  }
+  else if (TestName == "input_languages_contain") {
+    O << "InLangs.count(\"" << OptName << "\") != 0";
     return true;
   }
+  else if (TestName == "in_language") {
+    // This works only for single-argument Tool::GenerateAction. Join
+    // tools can process several files in different languages simultaneously.
 
-  // TODO: implement any_not_empty, any_empty, switch_on [..], empty [..]
+    // TODO: make this work with Edge::Weight (if possible).
+    O << "LangMap.GetLanguage(inFile) == \"" << OptName << '\"';
+    return true;
+  }
+  else if (TestName == "not_empty" || TestName == "empty") {
+    bool EmitNegate = (TestName == "not_empty");
+    apply(EmitEmptyTest(EmitNegate, OptDescs), OptName, O);
+    return true;
+  }
 
   return false;
 }
@@ -1161,17 +1250,14 @@
   checkNumberOfArguments(&d, 2);
   const std::string& OptName = InitPtrToString(d.getArg(0));
   const std::string& OptArg = InitPtrToString(d.getArg(1));
-  const OptionDescription& OptDesc = OptDescs.FindOption(OptName);
 
   if (TestName == "parameter_equals") {
-    if (!OptDesc.isParameter())
-      throw OptName + ": incorrect option type - should be a parameter!";
+    const OptionDescription& OptDesc = OptDescs.FindParameter(OptName);
     O << OptDesc.GenVariableName() << " == \"" << OptArg << "\"";
     return true;
   }
   else if (TestName == "element_in_list") {
-    if (!OptDesc.isList())
-      throw OptName + ": incorrect option type - should be a list!";
+    const OptionDescription& OptDesc = OptDescs.FindList(OptName);
     const std::string& VarName = OptDesc.GenVariableName();
     O << "std::find(" << VarName << ".begin(),\n";
     O.indent(IndentLevel + Indent1)




From foldr at codedgers.com  Tue Oct 20 21:13:53 2009
From: foldr at codedgers.com (Mikhail Glushenkov)
Date: Wed, 21 Oct 2009 02:13:53 -0000
Subject: [llvm-commits] [llvm] r84729 -
	/llvm/trunk/tools/llvmc/doc/LLVMC-Reference.rst
Message-ID: <200910210213.n9L2Drmn002002@zion.cs.uiuc.edu>

Author: foldr
Date: Tue Oct 20 21:13:52 2009
New Revision: 84729

URL: http://llvm.org/viewvc/llvm-project?rev=84729&view=rev
Log:
Clarify documentation on multi_val options.

Modified:
    llvm/trunk/tools/llvmc/doc/LLVMC-Reference.rst

Modified: llvm/trunk/tools/llvmc/doc/LLVMC-Reference.rst
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvmc/doc/LLVMC-Reference.rst?rev=84729&r1=84728&r2=84729&view=diff

==============================================================================
--- llvm/trunk/tools/llvmc/doc/LLVMC-Reference.rst (original)
+++ llvm/trunk/tools/llvmc/doc/LLVMC-Reference.rst Tue Oct 20 21:13:52 2009
@@ -349,8 +349,9 @@
 
    - ``multi_val n`` - this option takes *n* arguments (can be useful in some
      special cases). Usage example: ``(parameter_list_option "foo", (multi_val
-     3))``. Only list options can have this attribute; you can, however, use
-     the ``one_or_more`` and ``zero_or_one`` properties.
+     3))``; the command-line syntax is '-foo a b c'. Only list options can have
+     this attribute; you can, however, use the ``one_or_more``, ``zero_or_one``
+     and ``required`` properties.
 
    - ``init`` - this option has a default value, either a string (if it is a
      parameter), or a boolean (if it is a switch; boolean constants are called




From bob.wilson at apple.com  Tue Oct 20 21:15:46 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Wed, 21 Oct 2009 02:15:46 -0000
Subject: [llvm-commits] [llvm] r84730 - in /llvm/trunk/lib/Target/ARM:
 ARMInstrFormats.td ARMInstrNEON.td
Message-ID: <200910210215.n9L2FlZD002117@zion.cs.uiuc.edu>

Author: bwilson
Date: Tue Oct 20 21:15:46 2009
New Revision: 84730

URL: http://llvm.org/viewvc/llvm-project?rev=84730&view=rev
Log:
Leave some NEON instruction encoding bits unspecified instead of setting
a default value of zero.  This is important for decoding the instructions.
Patch by Johnny Chen, with some changes from me, too.

Modified:
    llvm/trunk/lib/Target/ARM/ARMInstrFormats.td
    llvm/trunk/lib/Target/ARM/ARMInstrNEON.td

Modified: llvm/trunk/lib/Target/ARM/ARMInstrFormats.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrFormats.td?rev=84730&r1=84729&r2=84730&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrFormats.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrFormats.td Tue Oct 20 21:15:46 2009
@@ -1262,15 +1262,26 @@
   let Inst{4} = op4;
 }
 
+// NEON Vector Duplicate (scalar).
+// Inst{19-16} is specified by subclasses.
+class N2VDup<bits<2> op24_23, bits<2> op21_20, bits<5> op11_7, bit op6, bit op4,
+             dag oops, dag iops, InstrItinClass itin,
+             string asm, string cstr, list<dag> pattern>
+  : NDataI<oops, iops, itin, asm, cstr, pattern> {
+  let Inst{24-23} = op24_23;
+  let Inst{21-20} = op21_20;
+  let Inst{11-7} = op11_7;
+  let Inst{6} = op6;
+  let Inst{4} = op4;
+}
+
 // NEON 2 vector register with immediate.
-class N2VImm<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-             bit op6, bit op4,
+class N2VImm<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
              dag oops, dag iops, InstrItinClass itin,
              string asm, string cstr, list<dag> pattern>
   : NDataI<oops, iops, itin, asm, cstr, pattern> {
   let Inst{24} = op24;
   let Inst{23} = op23;
-  let Inst{21-16} = op21_16;
   let Inst{11-8} = op11_8;
   let Inst{7} = op7;
   let Inst{6} = op6;

Modified: llvm/trunk/lib/Target/ARM/ARMInstrNEON.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrNEON.td?rev=84730&r1=84729&r2=84730&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrNEON.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrNEON.td Tue Oct 20 21:15:46 2009
@@ -623,12 +623,12 @@
         (ins QPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
         [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
 
-// Long 2-register intrinsics.  (This is currently only used for VMOVL and is
-// derived from N2VImm instead of N2V because of the way the size is encoded.)
-class N2VLInt<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-              bit op6, bit op4, InstrItinClass itin, string OpcodeStr,
+// Long 2-register intrinsics (currently only used for VMOVL).
+class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
+              bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
+              InstrItinClass itin, string OpcodeStr,
               ValueType TyQ, ValueType TyD, Intrinsic IntOp>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4, (outs QPR:$dst),
+  : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
         (ins DPR:$src), itin, !strconcat(OpcodeStr, "\t$dst, $src"), "",
         [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
 
@@ -1016,36 +1016,33 @@
 
 // Shift by immediate,
 // both double- and quad-register.
-class N2VDSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-             bit op4, InstrItinClass itin, string OpcodeStr,
-             ValueType Ty, SDNode OpNode>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
+class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
+             InstrItinClass itin, string OpcodeStr, ValueType Ty, SDNode OpNode>
+  : N2VImm<op24, op23, op11_8, op7, 0, op4,
            (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), itin,
            !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
            [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
-class N2VQSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-             bit op4, InstrItinClass itin, string OpcodeStr,
-             ValueType Ty, SDNode OpNode>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
+class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
+             InstrItinClass itin, string OpcodeStr, ValueType Ty, SDNode OpNode>
+  : N2VImm<op24, op23, op11_8, op7, 1, op4,
            (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
            !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
            [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
 
 // Long shift by immediate.
-class N2VLSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-             bit op6, bit op4, string OpcodeStr, ValueType ResTy,
-             ValueType OpTy, SDNode OpNode>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4,
+class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
+             string OpcodeStr, ValueType ResTy, ValueType OpTy, SDNode OpNode>
+  : N2VImm<op24, op23, op11_8, op7, op6, op4,
            (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VSHLiD,
            !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
            [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
                                           (i32 imm:$SIMM))))]>;
 
 // Narrow shift by immediate.
-class N2VNSh<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-             bit op6, bit op4, InstrItinClass itin, string OpcodeStr,
+class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
+             InstrItinClass itin, string OpcodeStr,
              ValueType ResTy, ValueType OpTy, SDNode OpNode>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, op6, op4,
+  : N2VImm<op24, op23, op11_8, op7, op6, op4,
            (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), itin,
            !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
            [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
@@ -1053,53 +1050,49 @@
 
 // Shift right by immediate and accumulate,
 // both double- and quad-register.
-class N2VDShAdd<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-                bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
-           (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, i32imm:$SIMM),
-           IIC_VPALiD, 
+class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
+                string OpcodeStr, ValueType Ty, SDNode ShOp>
+  : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
+           (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VPALiD, 
            !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
            [(set DPR:$dst, (Ty (add DPR:$src1,
                                 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
-class N2VQShAdd<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-                bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
-           (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, i32imm:$SIMM),
-           IIC_VPALiD, 
+class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
+                string OpcodeStr, ValueType Ty, SDNode ShOp>
+  : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
+           (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VPALiD, 
            !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
            [(set QPR:$dst, (Ty (add QPR:$src1,
                                 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
 
 // Shift by immediate and insert,
 // both double- and quad-register.
-class N2VDShIns<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-                bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
-           (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, i32imm:$SIMM),
-           IIC_VSHLiD, 
+class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
+                string OpcodeStr, ValueType Ty, SDNode ShOp>
+  : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
+           (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), IIC_VSHLiD, 
            !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
            [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
-class N2VQShIns<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-                bit op4, string OpcodeStr, ValueType Ty, SDNode ShOp>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
-           (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, i32imm:$SIMM),
-           IIC_VSHLiQ, 
+class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
+                string OpcodeStr, ValueType Ty, SDNode ShOp>
+  : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
+           (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), IIC_VSHLiQ, 
            !strconcat(OpcodeStr, "\t$dst, $src2, $SIMM"), "$src1 = $dst",
            [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
 
 // Convert, with fractional bits immediate,
 // both double- and quad-register.
-class N2VCvtD<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-              bit op4, string OpcodeStr, ValueType ResTy, ValueType OpTy,
+class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
+              string OpcodeStr, ValueType ResTy, ValueType OpTy,
               Intrinsic IntOp>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, 0, op4,
+  : N2VImm<op24, op23, op11_8, op7, 0, op4,
            (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), IIC_VUNAD, 
            !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
            [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
-class N2VCvtQ<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
-              bit op4, string OpcodeStr, ValueType ResTy, ValueType OpTy,
+class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
+              string OpcodeStr, ValueType ResTy, ValueType OpTy,
               Intrinsic IntOp>
-  : N2VImm<op24, op23, op21_16, op11_8, op7, 1, op4,
+  : N2VImm<op24, op23, op11_8, op7, 1, op4,
            (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), IIC_VUNAQ, 
            !strconcat(OpcodeStr, "\t$dst, $src, $SIMM"), "",
            [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
@@ -1175,14 +1168,14 @@
 
 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
 //   source operand element sizes of 16, 32 and 64 bits:
-multiclass N2VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
-                       bit op4, string OpcodeStr, Intrinsic IntOp> {
-  def v8i16 : N2VLInt<op24, op23, 0b001000, op11_8, op7, op6, op4,
-                      IIC_VQUNAiD, !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp>;
-  def v4i32 : N2VLInt<op24, op23, 0b010000, op11_8, op7, op6, op4,
-                      IIC_VQUNAiD, !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
-  def v2i64 : N2VLInt<op24, op23, 0b100000, op11_8, op7, op6, op4,
-                      IIC_VQUNAiD, !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
+multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
+                       string OpcodeStr, Intrinsic IntOp> {
+  def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
+                      !strconcat(OpcodeStr, "8"), v8i16, v8i8, IntOp>;
+  def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
+                      !strconcat(OpcodeStr, "16"), v4i32, v4i16, IntOp>;
+  def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
+                      !strconcat(OpcodeStr, "32"), v2i64, v2i32, IntOp>;
 }
 
 
@@ -1461,24 +1454,38 @@
 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
                       InstrItinClass itin, string OpcodeStr, SDNode OpNode> {
   // 64-bit vector types.
-  def v8i8  : N2VDSh<op24, op23, 0b001000, op11_8, 0, op4, itin,
-                     !strconcat(OpcodeStr, "8"), v8i8, OpNode>;
-  def v4i16 : N2VDSh<op24, op23, 0b010000, op11_8, 0, op4, itin,
-                     !strconcat(OpcodeStr, "16"), v4i16, OpNode>;
-  def v2i32 : N2VDSh<op24, op23, 0b100000, op11_8, 0, op4, itin,
-                     !strconcat(OpcodeStr, "32"), v2i32, OpNode>;
-  def v1i64 : N2VDSh<op24, op23, 0b000000, op11_8, 1, op4, itin,
+  def v8i8  : N2VDSh<op24, op23, op11_8, 0, op4, itin,
+                     !strconcat(OpcodeStr, "8"), v8i8, OpNode> {
+    let Inst{21-19} = 0b001; // imm6 = 001xxx
+  }
+  def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
+                     !strconcat(OpcodeStr, "16"), v4i16, OpNode> {
+    let Inst{21-20} = 0b01;  // imm6 = 01xxxx
+  }
+  def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, itin,
+                     !strconcat(OpcodeStr, "32"), v2i32, OpNode> {
+    let Inst{21} = 0b1;      // imm6 = 1xxxxx
+  }
+  def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, itin,
                      !strconcat(OpcodeStr, "64"), v1i64, OpNode>;
+                             // imm6 = xxxxxx
 
   // 128-bit vector types.
-  def v16i8 : N2VQSh<op24, op23, 0b001000, op11_8, 0, op4, itin,
-                     !strconcat(OpcodeStr, "8"), v16i8, OpNode>;
-  def v8i16 : N2VQSh<op24, op23, 0b010000, op11_8, 0, op4, itin,
-                     !strconcat(OpcodeStr, "16"), v8i16, OpNode>;
-  def v4i32 : N2VQSh<op24, op23, 0b100000, op11_8, 0, op4, itin,
-                     !strconcat(OpcodeStr, "32"), v4i32, OpNode>;
-  def v2i64 : N2VQSh<op24, op23, 0b000000, op11_8, 1, op4, itin,
+  def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
+                     !strconcat(OpcodeStr, "8"), v16i8, OpNode> {
+    let Inst{21-19} = 0b001; // imm6 = 001xxx
+  }
+  def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
+                     !strconcat(OpcodeStr, "16"), v8i16, OpNode> {
+    let Inst{21-20} = 0b01;  // imm6 = 01xxxx
+  }
+  def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, itin,
+                     !strconcat(OpcodeStr, "32"), v4i32, OpNode> {
+    let Inst{21} = 0b1;      // imm6 = 1xxxxx
+  }
+  def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, itin,
                      !strconcat(OpcodeStr, "64"), v2i64, OpNode>;
+                             // imm6 = xxxxxx
 }
 
 
@@ -1487,24 +1494,38 @@
 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
                          string OpcodeStr, SDNode ShOp> {
   // 64-bit vector types.
-  def v8i8  : N2VDShAdd<op24, op23, 0b001000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "8"), v8i8, ShOp>;
-  def v4i16 : N2VDShAdd<op24, op23, 0b010000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
-  def v2i32 : N2VDShAdd<op24, op23, 0b100000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
-  def v1i64 : N2VDShAdd<op24, op23, 0b000000, op11_8, 1, op4,
+  def v8i8  : N2VDShAdd<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "8"), v8i8, ShOp> {
+    let Inst{21-19} = 0b001; // imm6 = 001xxx
+  }
+  def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "16"), v4i16, ShOp> {
+    let Inst{21-20} = 0b01;  // imm6 = 01xxxx
+  }
+  def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "32"), v2i32, ShOp> {
+    let Inst{21} = 0b1;      // imm6 = 1xxxxx
+  }
+  def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
                         !strconcat(OpcodeStr, "64"), v1i64, ShOp>;
+                             // imm6 = xxxxxx
 
   // 128-bit vector types.
-  def v16i8 : N2VQShAdd<op24, op23, 0b001000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "8"), v16i8, ShOp>;
-  def v8i16 : N2VQShAdd<op24, op23, 0b010000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "16"), v8i16, ShOp>;
-  def v4i32 : N2VQShAdd<op24, op23, 0b100000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "32"), v4i32, ShOp>;
-  def v2i64 : N2VQShAdd<op24, op23, 0b000000, op11_8, 1, op4,
+  def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "8"), v16i8, ShOp> {
+    let Inst{21-19} = 0b001; // imm6 = 001xxx
+  }
+  def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "16"), v8i16, ShOp> {
+    let Inst{21-20} = 0b01;  // imm6 = 01xxxx
+  }
+  def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "32"), v4i32, ShOp> {
+    let Inst{21} = 0b1;      // imm6 = 1xxxxx
+  }
+  def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
                         !strconcat(OpcodeStr, "64"), v2i64, ShOp>;
+                             // imm6 = xxxxxx
 }
 
 
@@ -1513,24 +1534,75 @@
 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
                          string OpcodeStr, SDNode ShOp> {
   // 64-bit vector types.
-  def v8i8  : N2VDShIns<op24, op23, 0b001000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "8"), v8i8, ShOp>;
-  def v4i16 : N2VDShIns<op24, op23, 0b010000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "16"), v4i16, ShOp>;
-  def v2i32 : N2VDShIns<op24, op23, 0b100000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "32"), v2i32, ShOp>;
-  def v1i64 : N2VDShIns<op24, op23, 0b000000, op11_8, 1, op4,
+  def v8i8  : N2VDShIns<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "8"), v8i8, ShOp> {
+    let Inst{21-19} = 0b001; // imm6 = 001xxx
+  }
+  def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "16"), v4i16, ShOp> {
+    let Inst{21-20} = 0b01;  // imm6 = 01xxxx
+  }
+  def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "32"), v2i32, ShOp> {
+    let Inst{21} = 0b1;      // imm6 = 1xxxxx
+  }
+  def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
                         !strconcat(OpcodeStr, "64"), v1i64, ShOp>;
+                             // imm6 = xxxxxx
 
   // 128-bit vector types.
-  def v16i8 : N2VQShIns<op24, op23, 0b001000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "8"), v16i8, ShOp>;
-  def v8i16 : N2VQShIns<op24, op23, 0b010000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "16"), v8i16, ShOp>;
-  def v4i32 : N2VQShIns<op24, op23, 0b100000, op11_8, 0, op4,
-                        !strconcat(OpcodeStr, "32"), v4i32, ShOp>;
-  def v2i64 : N2VQShIns<op24, op23, 0b000000, op11_8, 1, op4,
+  def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "8"), v16i8, ShOp> {
+    let Inst{21-19} = 0b001; // imm6 = 001xxx
+  }
+  def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "16"), v8i16, ShOp> {
+    let Inst{21-20} = 0b01;  // imm6 = 01xxxx
+  }
+  def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
+                        !strconcat(OpcodeStr, "32"), v4i32, ShOp> {
+    let Inst{21} = 0b1;      // imm6 = 1xxxxx
+  }
+  def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
                         !strconcat(OpcodeStr, "64"), v2i64, ShOp>;
+                             // imm6 = xxxxxx
+}
+
+// Neon Shift Long operations,
+//   element sizes of 8, 16, 32 bits:
+multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
+                      bit op4, string OpcodeStr, SDNode OpNode> {
+  def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
+                 !strconcat(OpcodeStr, "8"), v8i16, v8i8, OpNode> {
+    let Inst{21-19} = 0b001; // imm6 = 001xxx
+  }
+  def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
+                  !strconcat(OpcodeStr, "16"), v4i32, v4i16, OpNode> {
+    let Inst{21-20} = 0b01;  // imm6 = 01xxxx
+  }
+  def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
+                  !strconcat(OpcodeStr, "32"), v2i64, v2i32, OpNode> {
+    let Inst{21} = 0b1;      // imm6 = 1xxxxx
+  }
+}
+
+// Neon Shift Narrow operations,
+//   element sizes of 16, 32, 64 bits:
+multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
+                      bit op4, InstrItinClass itin, string OpcodeStr,
+                      SDNode OpNode> {
+  def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
+                    !strconcat(OpcodeStr, "16"), v8i8, v8i16, OpNode> {
+    let Inst{21-19} = 0b001; // imm6 = 001xxx
+  }
+  def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
+                     !strconcat(OpcodeStr, "32"), v4i16, v4i32, OpNode> {
+    let Inst{21-20} = 0b01;  // imm6 = 01xxxx
+  }
+  def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
+                     !strconcat(OpcodeStr, "64"), v2i32, v2i64, OpNode> {
+    let Inst{21} = 0b1;      // imm6 = 1xxxxx
+  }
 }
 
 //===----------------------------------------------------------------------===//
@@ -2044,34 +2116,25 @@
 defm VSHRu    : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr.u", NEONvshru>;
 
 //   VSHLL    : Vector Shift Left Long
-def  VSHLLs8  : N2VLSh<0, 1, 0b001000, 0b1010, 0, 0, 1, "vshll.s8",
-                       v8i16, v8i8, NEONvshlls>;
-def  VSHLLs16 : N2VLSh<0, 1, 0b010000, 0b1010, 0, 0, 1, "vshll.s16",
-                       v4i32, v4i16, NEONvshlls>;
-def  VSHLLs32 : N2VLSh<0, 1, 0b100000, 0b1010, 0, 0, 1, "vshll.s32",
-                       v2i64, v2i32, NEONvshlls>;
-def  VSHLLu8  : N2VLSh<1, 1, 0b001000, 0b1010, 0, 0, 1, "vshll.u8",
-                       v8i16, v8i8, NEONvshllu>;
-def  VSHLLu16 : N2VLSh<1, 1, 0b010000, 0b1010, 0, 0, 1, "vshll.u16",
-                       v4i32, v4i16, NEONvshllu>;
-def  VSHLLu32 : N2VLSh<1, 1, 0b100000, 0b1010, 0, 0, 1, "vshll.u32",
-                       v2i64, v2i32, NEONvshllu>;
+defm VSHLLs   : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll.s", NEONvshlls>;
+defm VSHLLu   : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll.u", NEONvshllu>;
 
 //   VSHLL    : Vector Shift Left Long (with maximum shift count)
-def  VSHLLi8  : N2VLSh<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll.i8",
-                       v8i16, v8i8, NEONvshlli>;
-def  VSHLLi16 : N2VLSh<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll.i16",
-                       v4i32, v4i16, NEONvshlli>;
-def  VSHLLi32 : N2VLSh<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll.i32",
-                       v2i64, v2i32, NEONvshlli>;
+class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
+                bit op6, bit op4, string OpcodeStr, ValueType ResTy,
+                ValueType OpTy, SDNode OpNode>
+  : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, ResTy, OpTy, OpNode> {
+  let Inst{21-16} = op21_16;
+}
+def  VSHLLi8  : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll.i8",
+                          v8i16, v8i8, NEONvshlli>;
+def  VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll.i16",
+                          v4i32, v4i16, NEONvshlli>;
+def  VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll.i32",
+                          v2i64, v2i32, NEONvshlli>;
 
 //   VSHRN    : Vector Shift Right and Narrow
-def  VSHRN16  : N2VNSh<0, 1, 0b001000, 0b1000, 0, 0, 1, 
-                       IIC_VSHLiD, "vshrn.i16", v8i8, v8i16, NEONvshrn>;
-def  VSHRN32  : N2VNSh<0, 1, 0b010000, 0b1000, 0, 0, 1,
-                       IIC_VSHLiD, "vshrn.i32", v4i16, v4i32, NEONvshrn>;
-def  VSHRN64  : N2VNSh<0, 1, 0b100000, 0b1000, 0, 0, 1,
-                       IIC_VSHLiD, "vshrn.i64", v2i32, v2i64, NEONvshrn>;
+defm VSHRN    : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn.i", NEONvshrn>;
 
 //   VRSHL    : Vector Rounding Shift
 defm VRSHLs   : N3VInt_QHSD<0,0,0b0101,0, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
@@ -2083,12 +2146,8 @@
 defm VRSHRu   : N2VSh_QHSD<1, 1, 0b0010, 1, IIC_VSHLi4D, "vrshr.u", NEONvrshru>;
 
 //   VRSHRN   : Vector Rounding Shift Right and Narrow
-def  VRSHRN16 : N2VNSh<0, 1, 0b001000, 0b1000, 0, 1, 1,
-                       IIC_VSHLi4D, "vrshrn.i16", v8i8, v8i16, NEONvrshrn>;
-def  VRSHRN32 : N2VNSh<0, 1, 0b010000, 0b1000, 0, 1, 1, 
-                       IIC_VSHLi4D, "vrshrn.i32", v4i16, v4i32, NEONvrshrn>;
-def  VRSHRN64 : N2VNSh<0, 1, 0b100000, 0b1000, 0, 1, 1,
-                       IIC_VSHLi4D, "vrshrn.i64", v2i32, v2i64, NEONvrshrn>;
+defm VRSHRN   : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn.i",
+                           NEONvrshrn>;
 
 //   VQSHL    : Vector Saturating Shift
 defm VQSHLs   : N3VInt_QHSD<0,0,0b0100,1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
@@ -2102,26 +2161,14 @@
 defm VQSHLsu  : N2VSh_QHSD<1, 1, 0b0110, 1, IIC_VSHLi4D, "vqshlu.s", NEONvqshlsu>;
 
 //   VQSHRN   : Vector Saturating Shift Right and Narrow
-def VQSHRNs16 : N2VNSh<0, 1, 0b001000, 0b1001, 0, 0, 1, 
-                       IIC_VSHLi4D, "vqshrn.s16", v8i8, v8i16, NEONvqshrns>;
-def VQSHRNs32 : N2VNSh<0, 1, 0b010000, 0b1001, 0, 0, 1,
-                       IIC_VSHLi4D, "vqshrn.s32", v4i16, v4i32, NEONvqshrns>;
-def VQSHRNs64 : N2VNSh<0, 1, 0b100000, 0b1001, 0, 0, 1, 
-                       IIC_VSHLi4D, "vqshrn.s64", v2i32, v2i64, NEONvqshrns>;
-def VQSHRNu16 : N2VNSh<1, 1, 0b001000, 0b1001, 0, 0, 1,
-                       IIC_VSHLi4D, "vqshrn.u16", v8i8, v8i16, NEONvqshrnu>;
-def VQSHRNu32 : N2VNSh<1, 1, 0b010000, 0b1001, 0, 0, 1,
-                       IIC_VSHLi4D, "vqshrn.u32", v4i16, v4i32, NEONvqshrnu>;
-def VQSHRNu64 : N2VNSh<1, 1, 0b100000, 0b1001, 0, 0, 1,
-                       IIC_VSHLi4D, "vqshrn.u64", v2i32, v2i64, NEONvqshrnu>;
+defm VQSHRNs  : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn.s",
+                           NEONvqshrns>;
+defm VQSHRNu  : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn.u",
+                           NEONvqshrnu>;
 
 //   VQSHRUN  : Vector Saturating Shift Right and Narrow (Unsigned)
-def VQSHRUN16 : N2VNSh<1, 1, 0b001000, 0b1000, 0, 0, 1,
-                       IIC_VSHLi4D, "vqshrun.s16", v8i8, v8i16, NEONvqshrnsu>;
-def VQSHRUN32 : N2VNSh<1, 1, 0b010000, 0b1000, 0, 0, 1,
-                       IIC_VSHLi4D, "vqshrun.s32", v4i16, v4i32, NEONvqshrnsu>;
-def VQSHRUN64 : N2VNSh<1, 1, 0b100000, 0b1000, 0, 0, 1,
-                       IIC_VSHLi4D, "vqshrun.s64", v2i32, v2i64, NEONvqshrnsu>;
+defm VQSHRUN  : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun.s",
+                           NEONvqshrnsu>;
 
 //   VQRSHL   : Vector Saturating Rounding Shift
 defm VQRSHLs  : N3VInt_QHSD<0, 0, 0b0101, 1, IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q,
@@ -2130,26 +2177,14 @@
                             IIC_VSHLi4Q, "vqrshl.u", int_arm_neon_vqrshiftu, 0>;
 
 //   VQRSHRN  : Vector Saturating Rounding Shift Right and Narrow
-def VQRSHRNs16: N2VNSh<0, 1, 0b001000, 0b1001, 0, 1, 1,
-                       IIC_VSHLi4D, "vqrshrn.s16", v8i8, v8i16, NEONvqrshrns>;
-def VQRSHRNs32: N2VNSh<0, 1, 0b010000, 0b1001, 0, 1, 1,
-                       IIC_VSHLi4D, "vqrshrn.s32", v4i16, v4i32, NEONvqrshrns>;
-def VQRSHRNs64: N2VNSh<0, 1, 0b100000, 0b1001, 0, 1, 1,
-                       IIC_VSHLi4D, "vqrshrn.s64", v2i32, v2i64, NEONvqrshrns>;
-def VQRSHRNu16: N2VNSh<1, 1, 0b001000, 0b1001, 0, 1, 1,
-                       IIC_VSHLi4D, "vqrshrn.u16", v8i8, v8i16, NEONvqrshrnu>;
-def VQRSHRNu32: N2VNSh<1, 1, 0b010000, 0b1001, 0, 1, 1,
-                       IIC_VSHLi4D, "vqrshrn.u32", v4i16, v4i32, NEONvqrshrnu>;
-def VQRSHRNu64: N2VNSh<1, 1, 0b100000, 0b1001, 0, 1, 1, 
-                       IIC_VSHLi4D, "vqrshrn.u64", v2i32, v2i64, NEONvqrshrnu>;
+defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn.s",
+                           NEONvqrshrns>;
+defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn.u",
+                           NEONvqrshrnu>;
 
 //   VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
-def VQRSHRUN16: N2VNSh<1, 1, 0b001000, 0b1000, 0, 1, 1,
-                       IIC_VSHLi4D, "vqrshrun.s16", v8i8, v8i16, NEONvqrshrnsu>;
-def VQRSHRUN32: N2VNSh<1, 1, 0b010000, 0b1000, 0, 1, 1, 
-                       IIC_VSHLi4D, "vqrshrun.s32", v4i16, v4i32, NEONvqrshrnsu>;
-def VQRSHRUN64: N2VNSh<1, 1, 0b100000, 0b1000, 0, 1, 1,
-                       IIC_VSHLi4D, "vqrshrun.s64", v2i32, v2i64, NEONvqrshrnsu>;
+defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun.s",
+                           NEONvqrshrnsu>;
 
 //   VSRA     : Vector Shift Right and Accumulate
 defm VSRAs    : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra.s", NEONvshrs>;
@@ -2491,27 +2526,28 @@
 
 //   VDUP     : Vector Duplicate Lane (from scalar to all elements)
 
-class VDUPLND<bits<2> op19_18, bits<2> op17_16, string OpcodeStr, ValueType Ty>
-  : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 0, 0,
+class VDUPLND<string OpcodeStr, ValueType Ty>
+  : N2VDup<0b11, 0b11, 0b11000, 0, 0,
         (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
         !strconcat(OpcodeStr, "\t$dst, $src[$lane]"), "",
         [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
 
-class VDUPLNQ<bits<2> op19_18, bits<2> op17_16, string OpcodeStr,
-              ValueType ResTy, ValueType OpTy>
-  : N2V<0b11, 0b11, op19_18, op17_16, 0b11000, 1, 0,
+class VDUPLNQ<string OpcodeStr, ValueType ResTy, ValueType OpTy>
+  : N2VDup<0b11, 0b11, 0b11000, 1, 0,
         (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane), IIC_VMOVD,
         !strconcat(OpcodeStr, "\t$dst, $src[$lane]"), "",
         [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src), imm:$lane)))]>;
 
-def VDUPLN8d  : VDUPLND<0b00, 0b01, "vdup.8", v8i8>;
-def VDUPLN16d : VDUPLND<0b00, 0b10, "vdup.16", v4i16>;
-def VDUPLN32d : VDUPLND<0b01, 0b00, "vdup.32", v2i32>;
-def VDUPLNfd  : VDUPLND<0b01, 0b00, "vdup.32", v2f32>;
-def VDUPLN8q  : VDUPLNQ<0b00, 0b01, "vdup.8", v16i8, v8i8>;
-def VDUPLN16q : VDUPLNQ<0b00, 0b10, "vdup.16", v8i16, v4i16>;
-def VDUPLN32q : VDUPLNQ<0b01, 0b00, "vdup.32", v4i32, v2i32>;
-def VDUPLNfq  : VDUPLNQ<0b01, 0b00, "vdup.32", v4f32, v2f32>;
+// Inst{19-16} is partially specified depending on the element size.
+
+def VDUPLN8d  : VDUPLND<"vdup.8", v8i8> { let Inst{16} = 1; }
+def VDUPLN16d : VDUPLND<"vdup.16", v4i16> { let Inst{17-16} = 0b10; }
+def VDUPLN32d : VDUPLND<"vdup.32", v2i32> { let Inst{18-16} = 0b100; }
+def VDUPLNfd  : VDUPLND<"vdup.32", v2f32> { let Inst{18-16} = 0b100; }
+def VDUPLN8q  : VDUPLNQ<"vdup.8", v16i8, v8i8> { let Inst{16} = 1; }
+def VDUPLN16q : VDUPLNQ<"vdup.16", v8i16, v4i16> { let Inst{17-16} = 0b10; }
+def VDUPLN32q : VDUPLNQ<"vdup.32", v4i32, v2i32> { let Inst{18-16} = 0b100; }
+def VDUPLNfq  : VDUPLNQ<"vdup.32", v4f32, v2f32> { let Inst{18-16} = 0b100; }
 
 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
           (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
@@ -2530,15 +2566,19 @@
                                    (DSubReg_i32_reg imm:$lane))),
                            (SubReg_i32_lane imm:$lane)))>;
 
-def VDUPfdf   : N2V<0b11, 0b11, 0b01, 0b00, 0b11000, 0, 0,
-                    (outs DPR:$dst), (ins SPR:$src),
-                    IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
-                    [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
-
-def VDUPfqf   : N2V<0b11, 0b11, 0b01, 0b00, 0b11000, 1, 0,
-                    (outs QPR:$dst), (ins SPR:$src),
-                    IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
-                    [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
+def  VDUPfdf  : N2VDup<0b11, 0b11, 0b11000, 0, 0,
+                       (outs DPR:$dst), (ins SPR:$src),
+                       IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
+                       [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]> {
+  let Inst{18-16} = 0b100;
+}
+
+def  VDUPfqf  : N2VDup<0b11, 0b11, 0b11000, 1, 0,
+                       (outs QPR:$dst), (ins SPR:$src),
+                       IIC_VMOVD, "vdup.32\t$dst, ${src:lane}", "",
+                       [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]> {
+  let Inst{18-16} = 0b100;
+}
 
 def : Pat<(v2i64 (NEONvduplane (v2i64 QPR:$src), imm:$lane)),
           (INSERT_SUBREG QPR:$src, 
@@ -2560,8 +2600,8 @@
 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD, "vqmovun.s",
                             int_arm_neon_vqmovnsu>;
 //   VMOVL    : Vector Lengthening Move
-defm VMOVLs   : N2VLInt_QHS<0,1,0b1010,0,0,1, "vmovl.s", int_arm_neon_vmovls>;
-defm VMOVLu   : N2VLInt_QHS<1,1,0b1010,0,0,1, "vmovl.u", int_arm_neon_vmovlu>;
+defm VMOVLs   : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl.s", int_arm_neon_vmovls>;
+defm VMOVLu   : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl.u", int_arm_neon_vmovlu>;
 
 // Vector Conversions.
 
@@ -2585,24 +2625,22 @@
                      v4f32, v4i32, uint_to_fp>;
 
 //   VCVT     : Vector Convert Between Floating-Point and Fixed-Point.
-// Note: Some of the opcode bits in the following VCVT instructions need to
-// be encoded based on the immed values.
-def VCVTf2xsd : N2VCvtD<0, 1, 0b000000, 0b1111, 0, 1, "vcvt.s32.f32",
+def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt.s32.f32",
                         v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
-def VCVTf2xud : N2VCvtD<1, 1, 0b000000, 0b1111, 0, 1, "vcvt.u32.f32",
+def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt.u32.f32",
                         v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
-def VCVTxs2fd : N2VCvtD<0, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.s32",
+def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt.f32.s32",
                         v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
-def VCVTxu2fd : N2VCvtD<1, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.u32",
+def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt.f32.u32",
                         v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
 
-def VCVTf2xsq : N2VCvtQ<0, 1, 0b000000, 0b1111, 0, 1, "vcvt.s32.f32",
+def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt.s32.f32",
                         v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
-def VCVTf2xuq : N2VCvtQ<1, 1, 0b000000, 0b1111, 0, 1, "vcvt.u32.f32",
+def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt.u32.f32",
                         v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
-def VCVTxs2fq : N2VCvtQ<0, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.s32",
+def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt.f32.s32",
                         v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
-def VCVTxu2fq : N2VCvtQ<1, 1, 0b000000, 0b1110, 0, 1, "vcvt.f32.u32",
+def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt.f32.u32",
                         v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
 
 // Vector Reverse.




From bob.wilson at apple.com  Tue Oct 20 21:19:32 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Tue, 20 Oct 2009 19:19:32 -0700
Subject: [llvm-commits] ARM .td patch to specify only the	well-known
	encoding bits for imm6 and imm4 fields
In-Reply-To: <FA34BFD6-78D3-45A0-8D9B-B77613EC44B5@apple.com>
References: <E0C31726-F049-46CB-B443-8C01B7FF9AE0@apple.com>
	<FA34BFD6-78D3-45A0-8D9B-B77613EC44B5@apple.com>
Message-ID: <0D809F55-1781-454C-93DD-9389CD11B645@apple.com>


On Oct 20, 2009, at 12:01 PM, Bob Wilson wrote:

> 
> On Oct 20, 2009, at 10:49 AM, Johnny Chen wrote:
> 
>> Hi,
>> 
>> Refs: A8-754 (VSHLL and friends), A8-580 (VCVT between floating- 
>> point and fixed-point, Advanced SIMD), A8-592 (VDUP scalar)
>> 
>> This patch attempts to specify just the encoding bits described in  
>> the manual for several vector shift, vector convert, and vector dupe
>> operations.
>> 
>> One question I have is whether we should explicitly specify that  
>> VCVT's imm6 is of the pattern '1xxxxx'.
>> In the submitted patch, I left the imm6 (Inst{21-16}) of VCVT  
>> unspecified.
>> 
>> Thanks.
> 
> This is exactly the problem I was hoping to avoid.  Please see my  
> response to your VEXT patch before we decide whether to use this  
> approach.

I've committed a revised version of your patch.  There was only one instruction left using the old N2VImm class, and that was really just a minor convenience.  I changed that one class to use N2V instead so that all the other cases can use the N2VImm name.  I also did some refactoring of the new multiclasses that you had added and made some whitespace changes, etc.


From dpatel at apple.com  Tue Oct 20 21:21:34 2009
From: dpatel at apple.com (Devang Patel)
Date: Wed, 21 Oct 2009 02:21:34 -0000
Subject: [llvm-commits] [llvm] r84731 -
	/llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp
Message-ID: <200910210221.n9L2LYAE002400@zion.cs.uiuc.edu>

Author: dpatel
Date: Tue Oct 20 21:21:34 2009
New Revision: 84731

URL: http://llvm.org/viewvc/llvm-project?rev=84731&view=rev
Log:
Do not remove dead metadata for now.

Modified:
    llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp

Modified: llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp?rev=84731&r1=84730&r2=84731&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp (original)
+++ llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp Tue Oct 20 21:21:34 2009
@@ -150,7 +150,8 @@
   AliveGlobals.clear();
 
   // Remove dead metadata.
-  Changed |= M.getContext().RemoveDeadMetadata();
+  // FIXME - Enable this.
+  // Changed |= M.getContext().RemoveDeadMetadata();
   return Changed;
 }
 




From bob.wilson at apple.com  Tue Oct 20 21:27:20 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Wed, 21 Oct 2009 02:27:20 -0000
Subject: [llvm-commits] [llvm] r84732 -
	/llvm/trunk/lib/Target/ARM/ARMInstrNEON.td
Message-ID: <200910210227.n9L2RLQE002761@zion.cs.uiuc.edu>

Author: bwilson
Date: Tue Oct 20 21:27:20 2009
New Revision: 84732

URL: http://llvm.org/viewvc/llvm-project?rev=84732&view=rev
Log:
Fix some more NEON instruction encoding problems.
Thanks to Johnny Chen for discovering the problem.

Modified:
    llvm/trunk/lib/Target/ARM/ARMInstrNEON.td

Modified: llvm/trunk/lib/Target/ARM/ARMInstrNEON.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrNEON.td?rev=84732&r1=84731&r2=84732&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrNEON.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrNEON.td Tue Oct 20 21:27:20 2009
@@ -267,7 +267,7 @@
 
 //   VLD2LN   : Vector Load (single 2-element structure to one lane)
 class VLD2LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b10,op11_8,0b0000, (outs DPR:$dst1, DPR:$dst2),
+  : NLdSt<1,0b10,op11_8,0b0001, (outs DPR:$dst1, DPR:$dst2),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
           IIC_VLD2,
           !strconcat(OpcodeStr, "\t\\{$dst1[$lane],$dst2[$lane]\\}, $addr"),
@@ -287,7 +287,7 @@
 
 //   VLD3LN   : Vector Load (single 3-element structure to one lane)
 class VLD3LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b10,op11_8,0b0000, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
+  : NLdSt<1,0b10,op11_8,0b0010, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
           nohash_imm:$lane), IIC_VLD3,
           !strconcat(OpcodeStr,
@@ -308,7 +308,7 @@
 
 //   VLD4LN   : Vector Load (single 4-element structure to one lane)
 class VLD4LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b10,op11_8,0b0000,
+  : NLdSt<1,0b10,op11_8,0b0011,
           (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
           nohash_imm:$lane), IIC_VLD4,
@@ -450,7 +450,7 @@
 
 //   VST2LN   : Vector Store (single 2-element structure from one lane)
 class VST2LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b00,op11_8,0b0000, (outs),
+  : NLdSt<1,0b00,op11_8,0b0001, (outs),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
           IIC_VST,
           !strconcat(OpcodeStr, "\t\\{$src1[$lane],$src2[$lane]\\}, $addr"),
@@ -470,7 +470,7 @@
 
 //   VST3LN   : Vector Store (single 3-element structure from one lane)
 class VST3LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b00,op11_8,0b0000, (outs),
+  : NLdSt<1,0b00,op11_8,0b0010, (outs),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
            nohash_imm:$lane), IIC_VST,
           !strconcat(OpcodeStr,
@@ -490,7 +490,7 @@
 
 //   VST4LN   : Vector Store (single 4-element structure from one lane)
 class VST4LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b00,op11_8,0b0000, (outs),
+  : NLdSt<1,0b00,op11_8,0b0011, (outs),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
            nohash_imm:$lane), IIC_VST,
           !strconcat(OpcodeStr,




From clattner at apple.com  Tue Oct 20 23:10:03 2009
From: clattner at apple.com (Chris Lattner)
Date: Tue, 20 Oct 2009 21:10:03 -0700
Subject: [llvm-commits] metadata.h/cpp review stuff
In-Reply-To: <F85930CE-3A4B-478B-B0C6-E20EE110B47E@apple.com>
References: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>
	<352a1fb20910200708n4336cdaax503fce6480996830@mail.gmail.com>
	<DD3781CA-F185-4E23-BB49-4D464DA11B22@apple.com>
	<F85930CE-3A4B-478B-B0C6-E20EE110B47E@apple.com>
Message-ID: <8B96849E-A181-404B-831D-2640F85F27A0@apple.com>


On Oct 20, 2009, at 3:10 PM, Devang Patel wrote:

>>>
>>> Metadata used by another metadata is counted as regular uses.  
>>> Metadata
>>> used by non-metadata values is not a normal use (it is invisible to
>>> the non-metadata value). This is why the hierarchy is based on User
>>> and operand list is only counting metadata uses.
>>
>> Ok, I see now that you're doing this.  But why? :)  What capability  
>> does this provide?  Is there another way to get it?  This  
>> capability costs a lot of complexity, memory and compile time.
>
>
> This may be useful to reduce memory pressure by allowing metadata  
> user to delete MDNode and its MDNode users. After stumbling couple  
> of times to get this, I am also wondering whether this is worth it  
> or not or is this the right approach.
>
> MDNodes are uniqued, just like Constant. However Constants become  
> zoombie if its use count is reduced to zero. This may be too  
> expensive for metadata because of possibility of high volume of  
> metadata in the IR due to various reasons.

I'm not convinced :) , what "various reasons" would cause this?   
Again, we have a lot of complexity to support a use case that isn't  
clear.

-Chris
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091020/5e25549e/attachment.html 

From sabre at nondot.org  Tue Oct 20 23:10:31 2009
From: sabre at nondot.org (Chris Lattner)
Date: Wed, 21 Oct 2009 04:10:31 -0000
Subject: [llvm-commits] [llvm] r84738 -
	/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
Message-ID: <200910210410.n9L4AXHu009076@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 23:10:24 2009
New Revision: 84738

URL: http://llvm.org/viewvc/llvm-project?rev=84738&view=rev
Log:
tidy

Modified:
    llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h

Modified: llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h?rev=84738&r1=84737&r2=84738&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h (original)
+++ llvm/trunk/lib/Target/ARM/AsmPrinter/ARMInstPrinter.h Tue Oct 20 23:10:24 2009
@@ -51,7 +51,7 @@
   void printAddrModePCOperand(const MCInst *MI, unsigned OpNum,
                               const char *Modifier = 0);
     
-  void printBitfieldInvMaskImmOperand (const MCInst *MI, unsigned OpNum);
+  void printBitfieldInvMaskImmOperand(const MCInst *MI, unsigned OpNum);
   
   void printThumbITMask(const MCInst *MI, unsigned OpNum) {}
   void printThumbAddrModeRROperand(const MCInst *MI, unsigned OpNum) {}




From sabre at nondot.org  Tue Oct 20 23:11:19 2009
From: sabre at nondot.org (Chris Lattner)
Date: Wed, 21 Oct 2009 04:11:19 -0000
Subject: [llvm-commits] [llvm] r84739 -
	/llvm/trunk/lib/Transforms/Scalar/GVN.cpp
Message-ID: <200910210411.n9L4BKmh009169@zion.cs.uiuc.edu>

Author: lattner
Date: Tue Oct 20 23:11:19 2009
New Revision: 84739

URL: http://llvm.org/viewvc/llvm-project?rev=84739&view=rev
Log:
make GVN work better when TD is not around:

"In the existing code, if the load and the value to replace it with are
of different types *and* target data is available, it tries to use the
target data to coerce the replacement value to the type of the load.
Otherwise, it skips all effort to handle the type mismatch and just
feeds the wrongly-typed replacement value to replaceAllUsesWith, which
triggers an assertion.

The patch replaces it with an outer if checking for type mismatch, and
an inner if-else that checks whether target data is available and, if
not, returns false rather than trying to replace the load."

Patch by Kenneth Uildriks!


Modified:
    llvm/trunk/lib/Transforms/Scalar/GVN.cpp

Modified: llvm/trunk/lib/Transforms/Scalar/GVN.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/GVN.cpp?rev=84739&r1=84738&r2=84739&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/GVN.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/GVN.cpp Tue Oct 20 23:11:19 2009
@@ -1530,15 +1530,18 @@
     // actually have the same type.  See if we know how to reuse the stored
     // value (depending on its type).
     const TargetData *TD = 0;
-    if (StoredVal->getType() != L->getType() &&
-        (TD = getAnalysisIfAvailable<TargetData>())) {
-      StoredVal = CoerceAvailableValueToLoadType(StoredVal, L->getType(),
-                                                 L, *TD);
-      if (StoredVal == 0)
+    if (StoredVal->getType() != L->getType()) {
+      if ((TD = getAnalysisIfAvailable<TargetData>())) {
+        StoredVal = CoerceAvailableValueToLoadType(StoredVal, L->getType(),
+                                                   L, *TD);
+        if (StoredVal == 0)
+          return false;
+        
+        DEBUG(errs() << "GVN COERCED STORE:\n" << *DepSI << '\n' << *StoredVal
+                     << '\n' << *L << "\n\n\n");
+      }
+      else 
         return false;
-      
-      DEBUG(errs() << "GVN COERCED STORE:\n" << *DepSI << '\n' << *StoredVal
-                   << '\n' << *L << "\n\n\n");
     }
 
     // Remove it!
@@ -1557,14 +1560,17 @@
     // the same type.  See if we know how to reuse the previously loaded value
     // (depending on its type).
     const TargetData *TD = 0;
-    if (DepLI->getType() != L->getType() &&
-        (TD = getAnalysisIfAvailable<TargetData>())) {
-      AvailableVal = CoerceAvailableValueToLoadType(DepLI, L->getType(), L,*TD);
-      if (AvailableVal == 0)
-        return false;
+    if (DepLI->getType() != L->getType()) {
+      if ((TD = getAnalysisIfAvailable<TargetData>())) {
+        AvailableVal = CoerceAvailableValueToLoadType(DepLI, L->getType(), L,*TD);
+        if (AvailableVal == 0)
+          return false;
       
-      DEBUG(errs() << "GVN COERCED LOAD:\n" << *DepLI << "\n" << *AvailableVal
-                   << "\n" << *L << "\n\n\n");
+        DEBUG(errs() << "GVN COERCED LOAD:\n" << *DepLI << "\n" << *AvailableVal
+                     << "\n" << *L << "\n\n\n");
+      }
+      else 
+        return false;
     }
     
     // Remove it!




From clattner at apple.com  Tue Oct 20 23:11:31 2009
From: clattner at apple.com (Chris Lattner)
Date: Tue, 20 Oct 2009 21:11:31 -0700
Subject: [llvm-commits] [PATCH] Fix for GVN::processLoad when no
	TargetData available
In-Reply-To: <400d33ea0910201827n39fe1517y6ccb789b42e423f3@mail.gmail.com>
References: <400d33ea0910201827n39fe1517y6ccb789b42e423f3@mail.gmail.com>
Message-ID: <D2C5DB54-016C-4E5C-ABFC-34F375F86841@apple.com>


On Oct 20, 2009, at 6:27 PM, Kenneth Uildriks wrote:

> In the existing code, if the load and the value to replace it with are
> of different types *and* target data is available, it tries to use the
> target data to coerce the replacement value to the type of the load.
> Otherwise, it skips all effort to handle the type mismatch and just
> feeds the wrongly-typed replacement value to replaceAllUsesWith, which
> triggers an assertion.
>
> The patch replaces it with an outer if checking for type mismatch, and
> an inner if-else that checks whether target data is available and, if
> not, returns false rather than trying to replace the load.

Thanks, applied in r84739

-Chris


From sabre at nondot.org  Wed Oct 21 00:07:57 2009
From: sabre at nondot.org (Chris Lattner)
Date: Wed, 21 Oct 2009 05:07:57 -0000
Subject: [llvm-commits] [llvm] r84743 -
	/llvm/trunk/utils/emacs/tablegen-mode.el
Message-ID: <200910210507.n9L57vjp012424@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 00:07:57 2009
New Revision: 84743

URL: http://llvm.org/viewvc/llvm-project?rev=84743&view=rev
Log:
Set comment string, patch by Johnny Chen!

Modified:
    llvm/trunk/utils/emacs/tablegen-mode.el

Modified: llvm/trunk/utils/emacs/tablegen-mode.el
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/emacs/tablegen-mode.el?rev=84743&r1=84742&r2=84743&view=diff

==============================================================================
--- llvm/trunk/utils/emacs/tablegen-mode.el (original)
+++ llvm/trunk/utils/emacs/tablegen-mode.el Wed Oct 21 00:07:57 2009
@@ -112,6 +112,8 @@
         )
 
   (set-syntax-table tablegen-mode-syntax-table)
+  (make-local-variable 'comment-start)
+  (setq comment-start "//")
   (run-hooks 'tablegen-mode-hook))       ; Finally, this permits the user to
                                          ;   customize the mode with a hook.
 




From clattner at apple.com  Wed Oct 21 00:08:09 2009
From: clattner at apple.com (Chris Lattner)
Date: Tue, 20 Oct 2009 22:08:09 -0700
Subject: [llvm-commits] small patch for utils/emacs/tablegen-mode.el
In-Reply-To: <B6735BF7-0D29-4527-92BA-DE8954B800E8@apple.com>
References: <B6735BF7-0D29-4527-92BA-DE8954B800E8@apple.com>
Message-ID: <0EBD16EC-CBE4-4EF6-8F44-A5B8540E93E0@apple.com>

On Oct 20, 2009, at 3:23 PM, Johnny Chen wrote:

> For setting the comment-start string.

Committed as r84743, thanks!

-Chris

>
> <tablegen-mode.patch>_______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From clattner at apple.com  Wed Oct 21 00:51:14 2009
From: clattner at apple.com (Chris Lattner)
Date: Tue, 20 Oct 2009 22:51:14 -0700
Subject: [llvm-commits] patch: teach dse about intrinsics
In-Reply-To: <4ADAA446.9000306@mxc.ca>
References: <4ADAA2D5.2050803@mxc.ca> <4ADAA446.9000306@mxc.ca>
Message-ID: <DEE525BB-BDA0-4149-B362-87E3641906D8@apple.com>


On Oct 17, 2009, at 10:14 PM, Nick Lewycky wrote:

> Nick Lewycky wrote:
>> This patch creates a utility class like an Operator (GEPOperator  
>> and friends) which represents instructions that clobber a pointer  
>> they're given. This groups StoreInst together with certain  
>> intrinsics and helps DSE handle them as a group.
>> This also marks our first optimization based on @llvm.lifetime.end.
>> Please review!
>
> Arr... minor bug. This had code:
>
>      // Because a memcpy or memmove is also a load, we can't skip it  
> if we
>      // didn't remove it.
>      if (isa<MemTransferInst>(C))
>        continue;
>
> the change to use isa<MemTransferInst>(C) was a late change, after I  
> had run the nightly. (Previously I had a C->read() method but that's  
> the same as testing for MemTransferInst.) The correct code negates  
> the test:
>
>      // Because a memcpy or memmove is also a load, we can't skip it  
> if we
>      // didn't remove it.
>      if (!isa<MemTransferInst>(C))
>        continue;
>
> and since I know what you really want is a full patch, the updated  
> patch is attached.

Hi Nick,

This is two changes: 1) to add ClobberingInst, and 2) use it to do new  
stuff.  Because the two are conflated together into two patches, it  
makes it harder to review either.

Because #2 is hidden in noise, I'll address #1 first: why is a new  
pseudo instruction class like this the right way to go?  Why not add a  
simple static helper function instead?

-Chris


From dgregor at apple.com  Wed Oct 21 01:01:55 2009
From: dgregor at apple.com (Douglas Gregor)
Date: Wed, 21 Oct 2009 06:01:55 -0000
Subject: [llvm-commits] [llvm] r84744 -
	/llvm/trunk/lib/ExecutionEngine/JIT/CMakeLists.txt
Message-ID: <200910210601.n9L61t51015591@zion.cs.uiuc.edu>

Author: dgregor
Date: Wed Oct 21 01:01:54 2009
New Revision: 84744

URL: http://llvm.org/viewvc/llvm-project?rev=84744&view=rev
Log:
De-bork CMake build

Modified:
    llvm/trunk/lib/ExecutionEngine/JIT/CMakeLists.txt

Modified: llvm/trunk/lib/ExecutionEngine/JIT/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/CMakeLists.txt?rev=84744&r1=84743&r2=84744&view=diff

==============================================================================
--- llvm/trunk/lib/ExecutionEngine/JIT/CMakeLists.txt (original)
+++ llvm/trunk/lib/ExecutionEngine/JIT/CMakeLists.txt Wed Oct 21 01:01:54 2009
@@ -8,7 +8,6 @@
   JITDwarfEmitter.cpp
   JITEmitter.cpp
   JITMemoryManager.cpp
-  MacOSJITEventListener.cpp
   OProfileJITEventListener.cpp
   TargetSelect.cpp
   )




From evan.cheng at apple.com  Wed Oct 21 02:56:02 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Wed, 21 Oct 2009 07:56:02 -0000
Subject: [llvm-commits] [llvm] r84749 -
	/llvm/trunk/lib/CodeGen/MachineInstr.cpp
Message-ID: <200910210756.n9L7u2F4030076@zion.cs.uiuc.edu>

Author: evancheng
Date: Wed Oct 21 02:56:02 2009
New Revision: 84749

URL: http://llvm.org/viewvc/llvm-project?rev=84749&view=rev
Log:
Need a comma after imp-use.

Modified:
    llvm/trunk/lib/CodeGen/MachineInstr.cpp

Modified: llvm/trunk/lib/CodeGen/MachineInstr.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/MachineInstr.cpp?rev=84749&r1=84748&r2=84749&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/MachineInstr.cpp (original)
+++ llvm/trunk/lib/CodeGen/MachineInstr.cpp Wed Oct 21 02:56:02 2009
@@ -220,8 +220,10 @@
           OS << "imp-";
         OS << "def";
         NeedComma = true;
-      } else if (isImplicit())
+      } else if (isImplicit()) {
           OS << "imp-use";
+          NeedComma = true;
+      }
 
       if (isKill() || isDead() || isUndef()) {
         if (NeedComma) OS << ',';




From evan.cheng at apple.com  Wed Oct 21 03:15:53 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Wed, 21 Oct 2009 08:15:53 -0000
Subject: [llvm-commits] [llvm] r84751 - in /llvm/trunk:
 lib/Target/ARM/ARMISelDAGToDAG.cpp lib/Target/ARM/ARMISelLowering.cpp
 lib/Target/ARM/ARMInstrInfo.td lib/Target/ARM/ARMInstrThumb2.td
 test/CodeGen/ARM/movt.ll test/CodeGen/Thumb2/thumb2-mov2.ll
Message-ID: <200910210815.n9L8FrOd003719@zion.cs.uiuc.edu>

Author: evancheng
Date: Wed Oct 21 03:15:52 2009
New Revision: 84751

URL: http://llvm.org/viewvc/llvm-project?rev=84751&view=rev
Log:
Match more patterns to movt.

Added:
    llvm/trunk/test/CodeGen/ARM/movt.ll
Modified:
    llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp
    llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
    llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
    llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td
    llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll

Modified: llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp?rev=84751&r1=84750&r2=84751&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp Wed Oct 21 03:15:52 2009
@@ -1427,6 +1427,43 @@
       }
     }
     break;
+  case ISD::AND: {
+    // (and (or x, c2), c1) and top 16-bits of c1 and c2 match, lower 16-bits
+    // of c1 are 0xffff, and lower 16-bit of c2 are 0. That is, the top 16-bits
+    // are entirely contributed by c2 and lower 16-bits are entirely contributed
+    // by x. That's equal to (or (and x, 0xffff), (and c1, 0xffff0000)).
+    // Select it to: "movt x, ((c1 & 0xffff) >> 16)
+    EVT VT = Op.getValueType();
+    if (VT != MVT::i32)
+      break;
+    unsigned Opc = (Subtarget->isThumb() && Subtarget->hasThumb2())
+      ? ARM::t2MOVTi16
+      : (Subtarget->hasV6T2Ops() ? ARM::MOVTi16 : 0);
+    if (!Opc)
+      break;
+    SDValue N0 = Op.getOperand(0), N1 = Op.getOperand(1);
+    ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1);
+    if (!N1C)
+      break;
+    if (N0.getOpcode() == ISD::OR && N0.getNode()->hasOneUse()) {
+      SDValue N2 = N0.getOperand(1);
+      ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2);
+      if (!N2C)
+        break;
+      unsigned N1CVal = N1C->getZExtValue();
+      unsigned N2CVal = N2C->getZExtValue();
+      if ((N1CVal & 0xffff0000U) == (N2CVal & 0xffff0000U) &&
+          (N1CVal & 0xffffU) == 0xffffU &&
+          (N2CVal & 0xffffU) == 0x0U) {
+        SDValue Imm16 = CurDAG->getTargetConstant((N2CVal & 0xFFFF0000U) >> 16,
+                                                  MVT::i32);
+        SDValue Ops[] = { N0.getOperand(0), Imm16,
+                          getAL(CurDAG), CurDAG->getRegister(0, MVT::i32) };
+        return CurDAG->getMachineNode(Opc, dl, VT, Ops, 4);
+      }
+    }
+    break;
+  }
   case ARMISD::FMRRD:
     return CurDAG->getMachineNode(ARM::FMRRD, dl, MVT::i32, MVT::i32,
                                   Op.getOperand(0), getAL(CurDAG),

Modified: llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp?rev=84751&r1=84750&r2=84751&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp Wed Oct 21 03:15:52 2009
@@ -3022,7 +3022,6 @@
   return SDValue();
 }
 
-
 /// PerformFMRRDCombine - Target-specific dag combine xforms for ARMISD::FMRRD.
 static SDValue PerformFMRRDCombine(SDNode *N,
                                    TargetLowering::DAGCombinerInfo &DCI) {

Modified: llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrInfo.td?rev=84751&r1=84750&r2=84751&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrInfo.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrInfo.td Wed Oct 21 03:15:52 2009
@@ -980,6 +980,9 @@
   let Inst{25} = 1;
 }
 
+def : ARMPat<(or GPR:$src, 0xffff0000), (MOVTi16 GPR:$src, 0xffff)>,
+      Requires<[IsARM, HasV6T2]>;
+
 let Uses = [CPSR] in
 def MOVrx : AsI1<0b1101, (outs GPR:$dst), (ins GPR:$src), Pseudo, IIC_iMOVsi,
                  "mov", " $dst, $src, rrx",

Modified: llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td?rev=84751&r1=84750&r2=84751&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td Wed Oct 21 03:15:52 2009
@@ -666,6 +666,8 @@
                     [(set GPR:$dst,
                           (or (and GPR:$src, 0xffff), lo16AllZero:$imm))]>;
 
+def : T2Pat<(or GPR:$src, 0xffff0000), (t2MOVTi16 GPR:$src, 0xffff)>;
+
 //===----------------------------------------------------------------------===//
 //  Extend Instructions.
 //

Added: llvm/trunk/test/CodeGen/ARM/movt.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/movt.ll?rev=84751&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/ARM/movt.ll (added)
+++ llvm/trunk/test/CodeGen/ARM/movt.ll Wed Oct 21 03:15:52 2009
@@ -0,0 +1,19 @@
+; RUN: llc < %s -march=arm -mattr=+thumb2 | FileCheck %s
+; rdar://7317664
+
+define i32 @t(i32 %X) nounwind {
+; CHECK: t:
+; CHECK: movt r0, #65535
+entry:
+	%0 = or i32 %X, -65536
+	ret i32 %0
+}
+
+define i32 @t2(i32 %X) nounwind {
+; CHECK: t2:
+; CHECK: movt r0, #65534
+entry:
+	%0 = or i32 %X, -131072
+	%1 = and i32 %0, -65537
+	ret i32 %1
+}

Modified: llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll?rev=84751&r1=84750&r2=84751&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll (original)
+++ llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll Wed Oct 21 03:15:52 2009
@@ -2,10 +2,7 @@
 
 define i32 @t2MOVTi16_ok_1(i32 %a) {
 ; CHECK: t2MOVTi16_ok_1:
-; CHECK:      movs r1, #0
-; CHECK-NEXT: movt r1, #1234
-; CHECK:      movw r1, #65535
-; CHECK-NEXT: movt r1, #1234
+; CHECK: movt r0, #1234
     %1 = and i32 %a, 65535
     %2 = shl i32 1234, 16
     %3 = or  i32 %1, %2
@@ -15,10 +12,7 @@
 
 define i32 @t2MOVTi16_test_1(i32 %a) {
 ; CHECK: t2MOVTi16_test_1:
-; CHECK:      movs r1, #0
-; CHECK-NEXT: movt r1, #1234
-; CHECK:      movw r1, #65535
-; CHECK-NEXT: movt r1, #1234
+; CHECK: movt r0, #1234
     %1 = shl i32  255,   8
     %2 = shl i32 1234,   8
     %3 = or  i32   %1, 255  ; This give us 0xFFFF in %3
@@ -31,10 +25,7 @@
 
 define i32 @t2MOVTi16_test_2(i32 %a) {
 ; CHECK: t2MOVTi16_test_2:
-; CHECK:      movs r1, #0
-; CHECK-NEXT: movt r1, #1234
-; CHECK:      movw r1, #65535
-; CHECK-NEXT: movt r1, #1234
+; CHECK: movt r0, #1234
     %1 = shl i32  255,   8
     %2 = shl i32 1234,   8
     %3 = or  i32   %1, 255  ; This give us 0xFFFF in %3
@@ -48,10 +39,7 @@
 
 define i32 @t2MOVTi16_test_3(i32 %a) {
 ; CHECK: t2MOVTi16_test_3:
-; CHECK:      movs r1, #0
-; CHECK-NEXT: movt r1, #1234
-; CHECK:      movw r1, #65535
-; CHECK-NEXT: movt r1, #1234
+; CHECK: movt r0, #1234
     %1 = shl i32  255,   8
     %2 = shl i32 1234,   8
     %3 = or  i32   %1, 255  ; This give us 0xFFFF in %3




From sanjiv.gupta at microchip.com  Wed Oct 21 05:39:00 2009
From: sanjiv.gupta at microchip.com (Sanjiv Gupta)
Date: Wed, 21 Oct 2009 10:39:00 -0000
Subject: [llvm-commits] [llvm] r84752 - in
 /llvm/trunk/tools/llvmc/example/mcc16: driver/Main.cpp
 plugins/PIC16Base/PIC16Base.td plugins/PIC16Base/PluginMain.cpp
Message-ID: <200910211039.n9LAd1Yv010364@zion.cs.uiuc.edu>

Author: sgupta
Date: Wed Oct 21 05:38:59 2009
New Revision: 84752

URL: http://llvm.org/viewvc/llvm-project?rev=84752&view=rev
Log:
Added more options to mcc16 driver.

Modified:
    llvm/trunk/tools/llvmc/example/mcc16/driver/Main.cpp
    llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PIC16Base.td
    llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PluginMain.cpp

Modified: llvm/trunk/tools/llvmc/example/mcc16/driver/Main.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvmc/example/mcc16/driver/Main.cpp?rev=84752&r1=84751&r2=84752&view=diff

==============================================================================
--- llvm/trunk/tools/llvmc/example/mcc16/driver/Main.cpp (original)
+++ llvm/trunk/tools/llvmc/example/mcc16/driver/Main.cpp Wed Oct 21 05:38:59 2009
@@ -13,18 +13,31 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "llvm/Config/config.h"
 #include "llvm/CompilerDriver/BuiltinOptions.h"
 #include "llvm/CompilerDriver/ForceLinkage.h"
 #include "llvm/System/Path.h"
+#include <iostream>
 
 namespace llvmc {
   int Main(int argc, char** argv);
 }
 
+// Modify the PACKAGE_VERSION to use build number in top level configure file.
+void PIC16VersionPrinter(void) {
+  std::cout << "MPLAB C16 1.0 " << PACKAGE_VERSION << "\n";
+}
+
 int main(int argc, char** argv) {
 
   // HACK
   SaveTemps.setHiddenFlag(llvm::cl::Hidden);
+  TempDirname.setHiddenFlag(llvm::cl::Hidden);
+  Languages.setHiddenFlag(llvm::cl::Hidden);
+  DryRun.setHiddenFlag(llvm::cl::Hidden);
+
+  llvm::cl::SetVersionPrinter(PIC16VersionPrinter); 
+  
   TempDirname = "tmp-objs";
 
   // Remove the temp dir if already exists.

Modified: llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PIC16Base.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PIC16Base.td?rev=84752&r1=84751&r2=84752&view=diff

==============================================================================
--- llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PIC16Base.td (original)
+++ llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PIC16Base.td Wed Oct 21 05:38:59 2009
@@ -11,85 +11,149 @@
 def OptionList : OptionList<[
  (switch_option "g",
     (help "Enable Debugging")),
+ (switch_option "E",
+    (help "Stop after preprocessing, do not compile")),
  (switch_option "S",
     (help "Stop after compilation, do not assemble")),
+ (switch_option "bc",
+    (help "Stop after b-code generation, do not compile")),
  (switch_option "c",
     (help "Stop after assemble, do not link")),
- (parameter_option "I",
+ (prefix_list_option "I",
     (help "Add a directory to include path")),
- (parameter_option "pre-RA-sched",
-    (help "Example of an option that is passed to llc")),
+ (prefix_list_option "L",
+    (help "Add a directory to library path")),
+ (prefix_list_option "K",
+    (help "Add a directory to linker script search path")),
+ (parameter_option "l",
+    (help "Specify a library to link")),
+ (parameter_option "k",
+    (help "Specify a linker script")),
+ (parameter_option "m",
+    (help "Generate linker map file with the given name")),
+ (prefix_list_option "D",
+    (help "Define a macro")),
+ (switch_option "O0",
+    (help "Do not optimize")),
+// (switch_option "O1",
+//    (help "Optimization level 1")),
+// (switch_option "O2",
+//    (help "Optimization level 2. (Default)")),
+// (parameter_option "pre-RA-sched",
+//    (help "Example of an option that is passed to llc")),
  (prefix_list_option "Wa,",
     (help "Pass options to native assembler")),
  (prefix_list_option "Wl,",
-    (help "Pass options to native linker")),
- (prefix_list_option "Wllc,",
-    (help "Pass options to llc")),
- (prefix_list_option "Wo,",
-    (help "Pass options to llvm-ld"))
+    (help "Pass options to native linker"))
+// (prefix_list_option "Wllc,",
+//    (help "Pass options to llc")),
+// (prefix_list_option "Wo,",
+//    (help "Pass options to llvm-ld"))
 ]>;
 
 // Tools
-
-def clang_cc : Tool<[
- (in_language "c"),
+class clang_based<string language, string cmd, string ext_E> : Tool<
+[(in_language language),
  (out_language "llvm-bitcode"),
  (output_suffix "bc"),
- (cmd_line "$CALL(GetBinDir)clang-cc -I $CALL(GetStdHeadersDir) -triple=pic16- -emit-llvm-bc $INFILE -o $OUTFILE"),
- (actions (case
-          (not_empty "I"), (forward "I"))),
+ (cmd_line (case
+           (switch_on "E"),
+           (case 
+              (not_empty "o"), !strconcat(cmd, " -E $INFILE -o $OUTFILE"),
+              (default), !strconcat(cmd, " -E $INFILE")),
+           (default), !strconcat(cmd, " $INFILE -o $OUTFILE"))),
+ (actions (case 
+                (and (multiple_input_files), (or (switch_on "S"), (switch_on "c"))),
+              (error "cannot specify -o with -c or -S with multiple files"),
+                (switch_on "E"), [(stop_compilation), (output_suffix ext_E)],
+                (switch_on "bc"),[(stop_compilation), (output_suffix "bc")],
+                (switch_on "g"), (append_cmd "-g"),
+                (not_empty "D"), (forward "D"),
+                (not_empty "I"), (forward "I"))),
  (sink)
 ]>;
 
+def clang_cc : clang_based<"c", "$CALL(GetBinDir)clang-cc                                                        -I $CALL(GetStdHeadersDir) -triple=pic16-                                       -emit-llvm-bc ", "i">;
+
+//def clang_cc : Tool<[
+// (in_language "c"),
+// (out_language "llvm-bitcode"),
+// (output_suffix "bc"),
+// (cmd_line "$CALL(GetBinDir)clang-cc -I $CALL(GetStdHeadersDir) -triple=pic16- -emit-llvm-bc "),
+// (cmd_line kkkkk
+// (actions (case
+//          (switch_on "g"), (append_cmd "g"),
+//          (not_empty "I"), (forward "I"))),
+// (sink)
+//]>;
+
+
+// pre-link-and-lto step.
 def llvm_ld : Tool<[
  (in_language "llvm-bitcode"),
  (out_language "llvm-bitcode"),
  (output_suffix "bc"),
- (cmd_line "$CALL(GetBinDir)llvm-ld -link-as-library $INFILE -o $OUTFILE"),
+ (cmd_line "$CALL(GetBinDir)llvm-ld -L $CALL(GetStdLibsDir) -disable-gvn -instcombine -disable-inlining                   $INFILE -b $OUTFILE -l std"),
  (actions (case
-          (switch_on "g"), (append_cmd "-disable-opt"),
-          (not_empty "Wo,"), (unpack_values "Wo,")))
+          (switch_on "O0"), (append_cmd "-disable-opt"))),
+ (join)
 ]>;
 
-def llvm_ld_lto : Tool<[
+// optimize single file
+def llvm_ld_optimizer : Tool<[
  (in_language "llvm-bitcode"),
  (out_language "llvm-bitcode"),
  (output_suffix "bc"),
- (cmd_line "$CALL(GetBinDir)llvm-ld -L $CALL(GetStdLibsDir) -l std $INFILE -b $OUTFILE"),
+ (cmd_line "$CALL(GetBinDir)llvm-ld -disable-gvn -instcombine -disable-inlining                   $INFILE -b $OUTFILE"),
  (actions (case
-          (switch_on "g"), (append_cmd "-disable-opt"),
-          (not_empty "Wo,"), (unpack_values "Wo,"))),
- (join)
+          (switch_on "O0"), (append_cmd "-disable-opt")))
+]>;
+
+// optimizer step.
+def pic16passes : Tool<[
+ (in_language "llvm-bitcode"),
+ (out_language "llvm-bitcode"),
+ (output_suffix "obc"),
+ (cmd_line "$CALL(GetBinDir)opt -pic16cg -pic16overlay $INFILE -f -o $OUTFILE"),
+ (actions (case
+          (switch_on "O0"), (append_cmd "-disable-opt")))
 ]>;
 
 def llc : Tool<[
  (in_language "llvm-bitcode"),
  (out_language "assembler"),
  (output_suffix "s"),
- (cmd_line "$CALL(GetBinDir)llc -march=pic16 -disable-jump-tables -f $INFILE -o $OUTFILE"),
+ (cmd_line "$CALL(GetBinDir)llc -march=pic16 -disable-jump-tables -pre-RA-sched=list-burr -regalloc=pbqp -f $INFILE -o $OUTFILE"),
  (actions (case
-          (switch_on "S"), (stop_compilation),
-          (not_empty "Wllc,"), (unpack_values "Wllc,"),
-          (not_empty "pre-RA-sched"), (forward "pre-RA-sched")))
+          (switch_on "S"), (stop_compilation)))
+//          (not_empty "Wllc,"), (unpack_values "Wllc,"),
+//         (not_empty "pre-RA-sched"), (forward "pre-RA-sched")))
 ]>;
 
 def gpasm : Tool<[
  (in_language "assembler"),
  (out_language "object-code"),
  (output_suffix "o"),
- (cmd_line "$CALL(GetBinDir)gpasm -r decimal -p p16F1937 -I $CALL(GetStdAsmHeadersDir) -C -c $INFILE -o $OUTFILE"),
+ (cmd_line "$CALL(GetBinDir)gpasm -r decimal -p p16F1937 -I $CALL(GetStdAsmHeadersDir) -C -c -q $INFILE -o $OUTFILE"),
  (actions (case
           (switch_on "c"), (stop_compilation),
+          (switch_on "g"), (append_cmd "-g"),
           (not_empty "Wa,"), (unpack_values "Wa,")))
 ]>;
 
 def mplink : Tool<[
  (in_language "object-code"),
  (out_language "executable"),
- (output_suffix "out"),
- (cmd_line "$CALL(GetBinDir)mplink.exe -k $CALL(GetStdLinkerScriptsDir) -l $CALL(GetStdLibsDir) 16f1937_g.lkr intrinsics.lib devices.lib $INFILE -o $OUTFILE"),
+ (output_suffix "cof"),
+ (cmd_line "$CALL(GetBinDir)mplink.exe -k $CALL(GetStdLinkerScriptsDir) -l $CALL(GetStdLibsDir) -p 16f1937  intrinsics.lib devices.lib $INFILE -o $OUTFILE"),
  (actions (case
-          (not_empty "Wl,"), (unpack_values "Wl,"))),
+          (not_empty "Wl,"), (unpack_values "Wl,"),
+          (not_empty "L"), (forward_as "L", "-l"),
+          (not_empty "K"), (forward_as "K", "-k"),
+          (not_empty "m"), (forward "m"),
+//          (not_empty "l"), [(unpack_values "l"),(append_cmd ".lib")])),
+          (not_empty "k"), (unpack_values "k"),
+          (not_empty "l"), (unpack_values "l"))),
  (join)
 ]>;
 
@@ -103,19 +167,26 @@
     LangToSuffixes<"llvm-assembler", ["ll"]>,
     LangToSuffixes<"llvm-bitcode", ["bc"]>,
     LangToSuffixes<"object-code", ["o"]>,
-    LangToSuffixes<"executable", ["out"]>
+    LangToSuffixes<"executable", ["cof"]>
 ]>;
 
 // Compilation graph
 
 def CompilationGraph : CompilationGraph<[
     Edge<"root", "clang_cc">,
-    Edge<"clang_cc", "llvm_ld_lto">,
-    Edge<"llvm_ld_lto", "llc">,
-    OptionalEdge<"clang_cc", "llvm_ld", (case 
+    Edge<"root", "llvm_ld">,
+    OptionalEdge<"root", "llvm_ld_optimizer", (case 
+                                         (switch_on "S"), (inc_weight),
+                                         (switch_on "c"), (inc_weight))>,
+    Edge<"root", "gpasm">,
+    Edge<"root", "mplink">,
+    Edge<"clang_cc", "llvm_ld">,
+    OptionalEdge<"clang_cc", "llvm_ld_optimizer", (case 
                                          (switch_on "S"), (inc_weight),
                                          (switch_on "c"), (inc_weight))>,
-    Edge<"llvm_ld", "llc">,
+    Edge<"llvm_ld", "pic16passes">,
+    Edge<"llvm_ld_optimizer", "pic16passes">,
+    Edge<"pic16passes", "llc">,
     Edge<"llc", "gpasm">,
     Edge<"gpasm", "mplink">
 ]>;

Modified: llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PluginMain.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PluginMain.cpp?rev=84752&r1=84751&r2=84752&view=diff

==============================================================================
--- llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PluginMain.cpp (original)
+++ llvm/trunk/tools/llvmc/example/mcc16/plugins/PIC16Base/PluginMain.cpp Wed Oct 21 05:38:59 2009
@@ -10,11 +10,13 @@
 }
 
 // Returns the platform specific directory separator via #ifdefs.
+// FIXME: This currently work on linux and windows only. It does not 
+// work on other unices. 
 static std::string GetDirSeparator() {
-#ifdef _WIN32
-  return "\\";
-#else
+#if __linux__ || __APPLE__
   return "/";
+#else
+  return "\\";
 #endif
 }
 




From sanjiv.gupta at microchip.com  Wed Oct 21 05:42:44 2009
From: sanjiv.gupta at microchip.com (Sanjiv Gupta)
Date: Wed, 21 Oct 2009 10:42:44 -0000
Subject: [llvm-commits] [llvm] r84753 - in /llvm/trunk/lib/Target/PIC16:
 AsmPrinter/PIC16AsmPrinter.cpp AsmPrinter/PIC16AsmPrinter.h Makefile
 PIC16ABINames.h PIC16Passes/ PIC16Passes/Makefile
 PIC16Passes/PIC16Overlay.cpp PIC16Passes/PIC16Overlay.h PIC16Section.cpp
 PIC16Section.h PIC16TargetObjectFile.cpp PIC16TargetObjectFile.h
Message-ID: <200910211042.n9LAgjTm010493@zion.cs.uiuc.edu>

Author: sgupta
Date: Wed Oct 21 05:42:44 2009
New Revision: 84753

URL: http://llvm.org/viewvc/llvm-project?rev=84753&view=rev
Log:
Add a pass to overlay pic16 data sections for function frame and automatic
variables. This pass can be invoked by llvm-ld or opt to traverse over the call graph 
to detect what function frames and their automatic variables can be overlaid.
Currently this builds an archive , but needs to be changed to a loadable module.


Added:
    llvm/trunk/lib/Target/PIC16/PIC16Passes/
    llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
    llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.cpp
    llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.h
Modified:
    llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp
    llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.h
    llvm/trunk/lib/Target/PIC16/Makefile
    llvm/trunk/lib/Target/PIC16/PIC16ABINames.h
    llvm/trunk/lib/Target/PIC16/PIC16Section.cpp
    llvm/trunk/lib/Target/PIC16/PIC16Section.h
    llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.cpp
    llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.h

Modified: llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp?rev=84753&r1=84752&r2=84753&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp Wed Oct 21 05:42:44 2009
@@ -53,6 +53,45 @@
   return true;
 }
 
+static int getFunctionColor(const Function *F) {
+  if (F->hasSection()) {
+    std::string Sectn = F->getSection();
+    std::string StrToFind = "Overlay=";
+    unsigned Pos = Sectn.find(StrToFind);
+
+    // Retreive the color number if the key is found.
+    if (Pos != std::string::npos) {
+      Pos += StrToFind.length();
+      std::string Color = "";
+      char c = Sectn.at(Pos);
+      // A Color can only consist of digits.
+      while (c >= '0' && c<= '9') {
+        Color.append(1,c);
+        Pos++;
+        if (Pos >= Sectn.length())
+          break;
+        c = Sectn.at(Pos);
+      }
+      return atoi(Color.c_str());
+    }
+  }
+
+  // Color was not set for function, so return -1.
+  return -1;
+}
+
+// Color the Auto section of the given function. 
+void PIC16AsmPrinter::ColorAutoSection(const Function *F) {
+  std::string SectionName = PAN::getAutosSectionName(CurrentFnName);
+  PIC16Section* Section = PTOF->findPIC16Section(SectionName);
+  if (Section != NULL) {
+    int Color = getFunctionColor(F);
+    if (Color >= 0)
+      Section->setColor(Color);
+  }
+}
+
+
 /// runOnMachineFunction - This emits the frame section, autos section and 
 /// assembly for each instruction. Also takes care of function begin debug
 /// directive and file begin debug directive (if required) for the function.
@@ -68,6 +107,9 @@
   const Function *F = MF.getFunction();
   CurrentFnName = Mang->getMangledName(F);
 
+  // Put the color information from function to its auto section.
+  ColorAutoSection(F);
+
   // Emit the function frame (args and temps).
   EmitFunctionFrame(MF);
 
@@ -255,7 +297,6 @@
   EmitDefinedVars(M);
   EmitIData(M);
   EmitUData(M);
-  EmitAllAutos(M);
   EmitRomData(M);
   EmitUserSections(M);
   return Result;
@@ -330,6 +371,7 @@
 }
 
 bool PIC16AsmPrinter::doFinalization(Module &M) {
+  EmitAllAutos(M);
   printLibcallDecls();
   DbgInfo.EndModule(M);
   O << "\n\t" << "END\n";
@@ -343,8 +385,10 @@
   // Emit the data section name.
   O << "\n"; 
   
-  const MCSection *fPDataSection =
-    getObjFileLowering().SectionForFrame(CurrentFnName);
+  PIC16Section *fPDataSection = const_cast<PIC16Section *>(getObjFileLowering().
+                                SectionForFrame(CurrentFnName));
+ 
+  fPDataSection->setColor(getFunctionColor(F)); 
   OutStreamer.SwitchSection(fPDataSection);
   
   // Emit function frame label

Modified: llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.h?rev=84753&r1=84752&r2=84753&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.h (original)
+++ llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.h Wed Oct 21 05:42:44 2009
@@ -63,6 +63,7 @@
     void EmitSingleSection(const PIC16Section *S);
     void EmitSectionList(Module &M, 
                          const std::vector< PIC16Section *> &SList);
+    void ColorAutoSection(const Function *F);
   protected:
     bool doInitialization(Module &M);
     bool doFinalization(Module &M);

Modified: llvm/trunk/lib/Target/PIC16/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/Makefile?rev=84753&r1=84752&r2=84753&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/Makefile (original)
+++ llvm/trunk/lib/Target/PIC16/Makefile Wed Oct 21 05:42:44 2009
@@ -17,7 +17,7 @@
 		PIC16GenDAGISel.inc PIC16GenCallingConv.inc \
 		PIC16GenSubtarget.inc
 
-DIRS = AsmPrinter TargetInfo
+DIRS = AsmPrinter TargetInfo PIC16Passes
 
 include $(LEVEL)/Makefile.common
 

Modified: llvm/trunk/lib/Target/PIC16/PIC16ABINames.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16ABINames.h?rev=84753&r1=84752&r2=84753&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16ABINames.h (original)
+++ llvm/trunk/lib/Target/PIC16/PIC16ABINames.h Wed Oct 21 05:42:44 2009
@@ -304,6 +304,21 @@
         }
       }
     }
+
+    /// Return Overlay Name for the section.
+    /// The ABI Convention is: @<Color>.##.<section_tag>
+    /// The section_tag is retrieved from the SectName parameter and
+    /// and Color is passed in parameter.
+    static inline std::string  getOverlayName(std::string SectName, int Color) {
+      // FIXME: Only autos_section and frame_section are colored.
+      // So check and assert if the passed SectName does not have AUTOS_SECTION
+      // or FRAME_SECTION tag in it.
+      std::ostringstream o;
+      o << getTagName(PREFIX_SYMBOL) << Color << ".##" 
+        << SectName.substr(SectName.find("."));
+
+      return o.str();
+    } 
   }; // class PAN.
 } // end namespace llvm;
 

Added: llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile?rev=84753&view=auto

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile (added)
+++ llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile Wed Oct 21 05:42:44 2009
@@ -0,0 +1,17 @@
+##===- lib/Target/PIC16/PIC16Passes/Makefile -----------*- Makefile -*-===##
+# 
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source 
+# License. See LICENSE.TXT for details.
+# 
+##===----------------------------------------------------------------------===##
+LEVEL = ../../../..
+TARGET = PIC16
+LIBRARYNAME = LLVMpic16passes
+BUILD_ARCHIVE = 1
+
+
+
+include $(LEVEL)/Makefile.common
+

Added: llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.cpp?rev=84753&view=auto

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.cpp (added)
+++ llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.cpp Wed Oct 21 05:42:44 2009
@@ -0,0 +1,181 @@
+//===-- PIC16Overlay.cpp - Implementation for PIC16 Frame Overlay===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source 
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the PIC16 Frame Overlay implementation.
+//
+//===----------------------------------------------------------------------===//
+
+
+#include "llvm/Analysis/CallGraph.h"
+#include "llvm/Pass.h"
+#include "llvm/Module.h"
+#include "llvm/Instructions.h"
+#include "llvm/Value.h"
+#include "PIC16Overlay.h"
+#include "llvm/Function.h"
+#include <cstdlib>
+#include <sstream>
+using namespace llvm;
+
+namespace llvm {
+  char PIC16FrameOverlay::ID = 0;
+  ModulePass *createPIC16OverlayPass() { return new PIC16FrameOverlay(); }
+}
+
+void PIC16FrameOverlay::getAnalysisUsage(AnalysisUsage &AU) const {
+  AU.setPreservesAll();
+  AU.addRequired<CallGraph>();
+}
+
+void PIC16FrameOverlay::DFSTraverse(CallGraphNode *CGN, unsigned Depth) {
+  // Do not set any color for external calling node.
+  if (Depth != 0 && CGN->getFunction()) {
+    unsigned Color = getColor(CGN->getFunction());
+
+    // Handle indirectly called functions
+    if (Color >= PIC16Overlay::StartIndirectCallColor || 
+        Depth >= PIC16Overlay::StartIndirectCallColor) {
+      // All functions called from an indirectly called function are given
+      // an unique color.
+      if (Color < PIC16Overlay::StartIndirectCallColor &&
+          Depth >= PIC16Overlay::StartIndirectCallColor)
+        setColor(CGN->getFunction(), Depth);
+
+      for (unsigned int i = 0; i < CGN->size(); i++)
+        DFSTraverse((*CGN)[i], ++IndirectCallColor);
+      return;
+    }
+    // Just return if the node already has a color greater than the current 
+    // depth. A node must be colored with the maximum depth that it has.
+    if (Color >= Depth)
+      return;
+    
+    Depth = ModifyDepthForInterrupt(CGN, Depth);  
+    setColor(CGN->getFunction(), Depth);
+  }
+  
+  // Color all children of this node with color depth+1.
+  for (unsigned int i = 0; i < CGN->size(); i++)
+    DFSTraverse((*CGN)[i], Depth+1);
+}
+
+unsigned PIC16FrameOverlay::ModifyDepthForInterrupt(CallGraphNode *CGN,
+                                                    unsigned Depth) {
+  Function *Fn = CGN->getFunction();
+
+  // Return original Depth if function or section for function do not exist.
+  if (!Fn || !Fn->hasSection())
+    return Depth;
+
+  // Return original Depth if this function is not marked as interrupt.
+  if (Fn->getSection().find("interrupt") == string::npos)
+    return Depth;
+
+  Depth = Depth + InterruptDepth;
+  return Depth;
+}
+
+void PIC16FrameOverlay::setColor(Function *Fn, unsigned Color) {
+  std::string Section = "";
+  if (Fn->hasSection())
+    Section = Fn->getSection();
+
+  size_t Pos = Section.find(OverlayStr);
+
+  // Convert Color to string.
+  std::stringstream ss;
+  ss << Color;
+  std::string ColorString = ss.str();
+
+  // If color is already set then reset it with the new value. Else append 
+  // the Color string to section.
+  if (Pos != std::string::npos) {
+    Pos += OverlayStr.length();
+    char c = Section.at(Pos);
+    unsigned OldColorLength = 0;  
+    while (c >= '0' && c<= '9') {
+      OldColorLength++;    
+      if (Pos < Section.length() - 1)
+        Pos++;
+      else
+        break;
+      c = Section.at(Pos);
+    }
+    // Replace old color with new one.
+    Section.replace(Pos-OldColorLength +1, OldColorLength, ColorString); 
+  }
+  else {
+    // Append Color information to section string.
+    if (Fn->hasSection())
+      Section.append(" ");
+    Section.append(OverlayStr + ColorString);
+  }
+  Fn->setSection(Section);
+}
+
+unsigned PIC16FrameOverlay::getColor(Function *Fn) {
+  int Color = 0;
+  if (!Fn->hasSection())
+    return 0;
+
+  std::string Section = Fn->getSection();
+  size_t Pos = Section.find(OverlayStr);
+  
+  // Return 0 if Color is not set.
+  if (Pos == std::string::npos)
+    return 0;
+
+  // Set Pos to after "Overlay=".
+  Pos += OverlayStr.length();
+  char c = Section.at(Pos);
+  std::string ColorString = "";
+
+  // Find the string representing Color. A Color can only consist of digits.
+  while (c >= '0' && c<= '9') { 
+    ColorString.append(1,c);
+    if (Pos < Section.length() - 1)
+      Pos++;
+    else
+      break;
+    c = Section.at(Pos);
+  }
+  Color = atoi(ColorString.c_str());
+  
+  return Color;    
+}
+
+bool PIC16FrameOverlay::runOnModule(Module &M) {
+  CallGraph &CG = getAnalysis<CallGraph>();
+  CallGraphNode *ECN = CG.getExternalCallingNode();
+
+  MarkIndirectlyCalledFunctions(M); 
+  // Since External Calling Node is the base function, do a depth first 
+  // traversal of CallGraph with ECN as root. Each node with be marked with 
+  // a color that is max(color(callers)) + 1.
+  if(ECN) {
+    DFSTraverse(ECN, 0);
+  }
+  return false;
+}
+
+void PIC16FrameOverlay::MarkIndirectlyCalledFunctions(Module &M) {
+  // If the use of a function is not a call instruction then this
+  // function might be called indirectly. In that case give it
+  // an unique color.
+  for (Module::iterator MI = M.begin(), E = M.end(); MI != E; ++MI) {
+    for (Value::use_iterator I = MI->use_begin(), E = MI->use_end(); I != E;
+         ++I) {
+      if ((!isa<CallInst>(I) && !isa<InvokeInst>(I))
+          || !CallSite(cast<Instruction>(I)).isCallee(I)) {
+        setColor(MI, ++IndirectCallColor);
+        break;
+      }
+    }
+  }
+}

Added: llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.h?rev=84753&view=auto

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.h (added)
+++ llvm/trunk/lib/Target/PIC16/PIC16Passes/PIC16Overlay.h Wed Oct 21 05:42:44 2009
@@ -0,0 +1,55 @@
+//===-- PIC16FrameOverlay.h - Interface for PIC16 Frame Overlay -*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source 
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the PIC16 Overlay infrastructure.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef PIC16FRAMEOVERLAY_H
+#define PIC16FRAMEOVERLAY_H
+ 
+#include "llvm/Analysis/CallGraph.h"
+#include "llvm/Pass.h"
+#include "llvm/CallGraphSCCPass.h"
+
+using std::string;
+using namespace llvm;
+
+namespace  llvm {
+  namespace PIC16Overlay {
+    enum OverlayConsts {
+      StartInterruptColor = 200,
+      StartIndirectCallColor = 300
+    }; 
+  }
+  class PIC16FrameOverlay : public ModulePass {
+    std::string OverlayStr;
+    unsigned InterruptDepth;
+    unsigned IndirectCallColor;
+  public:
+    static char ID; // Class identification 
+    PIC16FrameOverlay() : ModulePass(&ID) {
+      OverlayStr = "Overlay=";
+      InterruptDepth = PIC16Overlay::StartInterruptColor;
+      IndirectCallColor = PIC16Overlay::StartIndirectCallColor;
+    }
+
+    virtual void getAnalysisUsage(AnalysisUsage &AU) const; 
+    virtual bool runOnModule(Module &M);
+
+  private: 
+    unsigned getColor(Function *Fn);
+    void setColor(Function *Fn, unsigned Color);
+    unsigned ModifyDepthForInterrupt(CallGraphNode *CGN, unsigned Depth);
+    void MarkIndirectlyCalledFunctions(Module &M);
+    void DFSTraverse(CallGraphNode *CGN, unsigned Depth);
+  };
+}  // End of  namespace
+
+#endif

Modified: llvm/trunk/lib/Target/PIC16/PIC16Section.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16Section.cpp?rev=84753&r1=84752&r2=84753&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16Section.cpp (original)
+++ llvm/trunk/lib/Target/PIC16/PIC16Section.cpp Wed Oct 21 05:42:44 2009
@@ -68,8 +68,13 @@
 // A generic way to print all types of sections.
 void PIC16Section::PrintSwitchToSection(const MCAsmInfo &MAI,
                                           raw_ostream &OS) const {
-  // Print name.
-  OS << getName() << '\t';
+ 
+  // If the section is overlaid(i.e. it has a color), print overlay name for 
+  // it. Otherwise print its normal name.
+  if (Color != -1)
+    OS << PAN::getOverlayName(getName(), Color) << '\t';
+  else
+    OS << getName() << '\t';
 
   // Print type.
   switch (getType()) {

Modified: llvm/trunk/lib/Target/PIC16/PIC16Section.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16Section.h?rev=84753&r1=84752&r2=84753&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16Section.h (original)
+++ llvm/trunk/lib/Target/PIC16/PIC16Section.h Wed Oct 21 05:42:44 2009
@@ -57,6 +57,7 @@
 
     /// Return the Color of the section.
     int getColor() const { return Color; }
+    void setColor(int color) { Color = color; }
 
     /// Return the size of the section.
     unsigned getSize() const { return Size; }

Modified: llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.cpp?rev=84753&r1=84752&r2=84753&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.cpp (original)
+++ llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.cpp Wed Oct 21 05:42:44 2009
@@ -25,6 +25,18 @@
 PIC16TargetObjectFile::~PIC16TargetObjectFile() {
 }
 
+/// Find a pic16 section. Return null if not found. Do not create one.
+PIC16Section *PIC16TargetObjectFile::
+findPIC16Section(const std::string &Name) {
+  /// Return if we have an already existing one.
+  PIC16Section *Entry = SectionsByName[Name];
+  if (Entry)
+    return Entry;
+
+  return NULL;
+}
+
+
 /// Find a pic16 section. If not found, create one.
 PIC16Section *PIC16TargetObjectFile::
 getPIC16Section(const std::string &Name, PIC16SectionType Ty, 

Modified: llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.h?rev=84753&r1=84752&r2=84753&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.h (original)
+++ llvm/trunk/lib/Target/PIC16/PIC16TargetObjectFile.h Wed Oct 21 05:42:44 2009
@@ -116,6 +116,9 @@
     ~PIC16TargetObjectFile();
     void Initialize(MCContext &Ctx, const TargetMachine &TM);
 
+    /// Return the section with the given Name. Null if not found.
+    PIC16Section *findPIC16Section(const std::string &Name);
+
     /// Override section allocations for user specified sections.
     virtual const MCSection *
     getExplicitSectionGlobal(const GlobalValue *GV, SectionKind Kind, 
@@ -127,6 +130,7 @@
                                                     Mangler *Mang,
                                                     const TargetMachine&) const;
 
+
     /// Return a code section for a function.
     const PIC16Section *SectionForCode (const std::string &FnName) const;
 




From edwintorok at gmail.com  Wed Oct 21 05:49:01 2009
From: edwintorok at gmail.com (Torok Edwin)
Date: Wed, 21 Oct 2009 10:49:01 -0000
Subject: [llvm-commits] [llvm] r84754 - in /llvm/trunk:
 lib/Transforms/Scalar/InstructionCombining.cpp
 test/Transforms/InstCombine/crash.ll
 test/Transforms/InstCombine/fold-intophi.ll
Message-ID: <200910211049.n9LAn18B010770@zion.cs.uiuc.edu>

Author: edwin
Date: Wed Oct 21 05:49:00 2009
New Revision: 84754

URL: http://llvm.org/viewvc/llvm-project?rev=84754&view=rev
Log:
Fix PR5262: when folding select into PHI, make sure all operands are available
in the PHI's Basic Block. This uses a conservative approach, because we don't
have dominator info in instcombine.

Added:
    llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll
Modified:
    llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
    llvm/trunk/test/Transforms/InstCombine/crash.ll

Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=84754&r1=84753&r2=84754&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Wed Oct 21 05:49:00 2009
@@ -1985,6 +1985,42 @@
   return 0;
 }
 
+// Check whether all the operands of the PHI dominate the PHI node,
+// knowing that the PHI's operands either dominate BB, or are defined in the BB.
+static bool checkPHI(PHINode *PN, BasicBlock *BB)
+{
+  BasicBlock *phiBB = PN->getParent();
+  for (unsigned i=0;i<PN->getNumIncomingValues();i++) {
+    Instruction *I = dyn_cast<Instruction>(PN->getIncomingValue(i));
+    if (!I)
+      continue;
+    BasicBlock *pBB = I->getParent();
+    if (pBB == BB) {
+      // another PHI in same BB is always before PN (PN is last phi).
+      if (isa<PHINode>(I))
+        continue;
+      // An instruction in the same BB, and not a phi, this is after PN.
+      return false;
+    }
+    if (phiBB == BB)
+      continue;
+    // We don't have dominator info, so just check that the instruction
+    // is not defined in on of the BB on the unique path between BB and phiBB.
+    // If there is no such unique path, or pBB equals to one of the BBs on that
+    // path we know that this operand doesn't dominate the PHI node.
+    BasicBlock *B = PN->getIncomingBlock(i);
+    while ((B = B->getUniquePredecessor())) {
+      if (pBB == B)
+        return false;
+      if (B == phiBB)
+        break;
+    }
+    // No unique path -> doesn't dominate
+    if (!B)
+      return false;
+  }
+  return true;
+}
 
 /// FoldOpIntoPhi - Given a binary operator, cast instruction, or select which
 /// has a PHI node as operand #0, see if we can fold the instruction into the
@@ -2036,9 +2072,13 @@
   // Okay, we can do the transformation: create the new PHI node.
   PHINode *NewPN = PHINode::Create(I.getType(), "");
   NewPN->reserveOperandSpace(PN->getNumOperands()/2);
-  InsertNewInstBefore(NewPN, *PN);
-  NewPN->takeName(PN);
+  // We must the PHI node as the last PHI, because it may use one of the other
+  // PHIs.
+  BasicBlock::iterator BBIt = PN;
+  while (isa<PHINode>(BBIt)) ++BBIt;
+  InsertNewInstBefore(NewPN, *BBIt);
 
+  SmallVector<Instruction*, 2> tmpWorklist;
   // Next, add all of the operands to the PHI.
   if (SelectInst *SI = dyn_cast<SelectInst>(&I)) {
     // We only currently try to fold the condition of a select when it is a phi,
@@ -2058,7 +2098,7 @@
         InV = SelectInst::Create(PN->getIncomingValue(i), TrueVInPred,
                                  FalseVInPred,
                                  "phitmp", NonConstBB->getTerminator());
-        Worklist.Add(cast<Instruction>(InV));
+        tmpWorklist.push_back(cast<Instruction>(InV));
       }
       NewPN->addIncoming(InV, ThisBB);
     }
@@ -2084,8 +2124,7 @@
                                 NonConstBB->getTerminator());
         else
           llvm_unreachable("Unknown binop!");
-        
-        Worklist.Add(cast<Instruction>(InV));
+        tmpWorklist.push_back(cast<Instruction>(InV));
       }
       NewPN->addIncoming(InV, PN->getIncomingBlock(i));
     }
@@ -2101,11 +2140,26 @@
         InV = CastInst::Create(CI->getOpcode(), PN->getIncomingValue(i), 
                                I.getType(), "phitmp", 
                                NonConstBB->getTerminator());
-        Worklist.Add(cast<Instruction>(InV));
+        tmpWorklist.push_back(cast<Instruction>(InV));
       }
       NewPN->addIncoming(InV, PN->getIncomingBlock(i));
     }
   }
+  // The PHI's operands must dominate the PHI, if we can't prove that
+  // undo this transformation.
+  if (!checkPHI(NewPN, I.getParent())) {
+    Worklist.Remove(NewPN);
+    NewPN->eraseFromParent();
+    while (!tmpWorklist.empty()) {
+      tmpWorklist.pop_back_val()->eraseFromParent();
+    }
+    return 0;
+  }
+  while (!tmpWorklist.empty()) {
+    Worklist.Add(tmpWorklist.pop_back_val());
+  }
+  NewPN->takeName(PN);
+  Worklist.Add(PN);
   return ReplaceInstUsesWith(I, NewPN);
 }
 

Modified: llvm/trunk/test/Transforms/InstCombine/crash.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/crash.ll?rev=84754&r1=84753&r2=84754&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/crash.ll (original)
+++ llvm/trunk/test/Transforms/InstCombine/crash.ll Wed Oct 21 05:49:00 2009
@@ -44,3 +44,58 @@
   store i32 %ins, i32* %arrayidx20
   ret void
 }
+
+; PR5262
+ at tmp2 = global i64 0                              ; <i64*> [#uses=1]
+
+declare void @use(i64) nounwind
+
+define void @foo(i1) nounwind align 2 {
+; <label>:1
+  br i1 %0, label %2, label %3
+
+; <label>:2                                       ; preds = %1
+  br label %3
+
+; <label>:3                                       ; preds = %2, %1
+  %4 = phi i8 [ 1, %2 ], [ 0, %1 ]                ; <i8> [#uses=1]
+  %5 = icmp eq i8 %4, 0                           ; <i1> [#uses=1]
+  %6 = load i64* @tmp2, align 8                   ; <i64> [#uses=1]
+  %7 = select i1 %5, i64 0, i64 %6                ; <i64> [#uses=1]
+  br label %8
+
+; <label>:8                                       ; preds = %3
+  call void @use(i64 %7)
+  ret void
+}
+
+%t0 = type { i32, i32 }
+%t1 = type { i32, i32, i32, i32, i32* }
+
+declare %t0* @bar2(i64)
+
+define void @bar3(i1, i1) nounwind align 2 {
+; <label>:2
+  br i1 %1, label %10, label %3
+
+; <label>:3                                       ; preds = %2
+  %4 = getelementptr inbounds %t0* null, i64 0, i32 1 ; <i32*> [#uses=0]
+  %5 = getelementptr inbounds %t1* null, i64 0, i32 4 ; <i32**> [#uses=1]
+  %6 = load i32** %5, align 8                     ; <i32*> [#uses=1]
+  %7 = icmp ne i32* %6, null                      ; <i1> [#uses=1]
+  %8 = zext i1 %7 to i32                          ; <i32> [#uses=1]
+  %9 = add i32 %8, 0                              ; <i32> [#uses=1]
+  br label %10
+
+; <label>:10                                      ; preds = %3, %2
+  %11 = phi i32 [ %9, %3 ], [ 0, %2 ]             ; <i32> [#uses=1]
+  br i1 %1, label %12, label %13
+
+; <label>:12                                      ; preds = %10
+  br label %13
+
+; <label>:13                                      ; preds = %12, %10
+  %14 = zext i32 %11 to i64                       ; <i64> [#uses=1]
+  %15 = tail call %t0* @bar2(i64 %14) nounwind      ; <%0*> [#uses=0]
+  ret void
+}

Added: llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll?rev=84754&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll (added)
+++ llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll Wed Oct 21 05:49:00 2009
@@ -0,0 +1,54 @@
+; RUN: opt -instcombine -S <%s | FileCheck %s
+; PR5262
+; Make sure the PHI node gets put in a place where all of its operands dominate it
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
+target triple = "x86_64-unknown-linux-gnu"
+
+ at tmp2 = global i64 0                              ; <i64*> [#uses=1]
+
+declare void @use(i64) nounwind
+
+define void @foo(i1) nounwind align 2 {
+; <label>:1
+  br i1 %0, label %2, label %3
+
+; <label>:2                                       ; preds = %1
+  br label %3
+
+; <label>:3                                       ; preds = %2, %1
+; CHECK: <label>:3
+; CHECK-NEXT: %v4 = phi i1 [ false, %2 ], [ true, %1 ]
+; CHECK-NEXT: %v4_ =  phi i1 [ true, %2 ], [ false, %1 ]
+; CHECK-NEXT: br label %l8
+  %v4 = phi i8 [ 1, %2 ], [ 0, %1 ]                ; <i8> [#uses=1]
+  %v4_ = phi i8 [ 0, %2 ], [ 1, %1 ]                ; <i8> [#uses=1]
+  %v5 = icmp eq i8 %v4, 0                           ; <i1> [#uses=1]
+  %v5_ = icmp eq i8 %v4_, 0                           ; <i1> [#uses=1]
+  %v6 = load i64* @tmp2, align 8                   ; <i64> [#uses=1]
+  %v7 = select i1 %v5, i64 0, i64 %v6                ; <i64> [#uses=1]
+  br label %l8
+
+l8:
+  %v9 = load i64* @tmp2, align 8
+  call void @use(i64 %v7)
+  br label %l10
+l10:
+  %v11 = select i1 %v5_, i64 0, i64 %v9                ; <i64> [#uses=1]
+  call void @use(i64 %v11)
+  br label %l11
+l11:
+  %v12 = load i64* @tmp2, align 8
+  %v13 = select i1 %v5_, i64 0, i64 %v12                ; <i64> [#uses=1]
+  call void @use(i64 %v13)
+  br i1 %0, label %l12, label %l13
+l12:
+  br label %l13
+l13:
+;CHECK: l13:
+;CHECK-NEXT: %v14 = phi i64 [ %v12, %l12 ], [ 0, %l11 ]
+;CHECK-NEXT: call void @use(i64 %v14)
+  %v14 = phi i1 [0, %l12], [1, %l11]
+  %v16 = select i1 %v14, i64 0, i64 %v12                ; <i64> [#uses=1]
+  call void @use(i64 %v16)
+  ret void
+}




From ofv at wanadoo.es  Wed Oct 21 08:22:22 2009
From: ofv at wanadoo.es (Oscar Fuentes)
Date: Wed, 21 Oct 2009 13:22:22 -0000
Subject: [llvm-commits] [llvm] r84759 - /llvm/trunk/docs/CMake.html
Message-ID: <200910211322.n9LDMMYx019015@zion.cs.uiuc.edu>

Author: ofv
Date: Wed Oct 21 08:22:20 2009
New Revision: 84759

URL: http://llvm.org/viewvc/llvm-project?rev=84759&view=rev
Log:
Two corrections for docs/CMake.html.

Patch by Victor Zverovich!

Modified:
    llvm/trunk/docs/CMake.html

Modified: llvm/trunk/docs/CMake.html
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/CMake.html?rev=84759&r1=84758&r2=84759&view=diff

==============================================================================
--- llvm/trunk/docs/CMake.html (original)
+++ llvm/trunk/docs/CMake.html Wed Oct 21 08:22:20 2009
@@ -209,7 +209,7 @@
   <dt><b>CMAKE_BUILD_TYPE</b>:STRING</dt>
 
   <dd>Sets the build type for <i>make</i> based generators. Possible
-    values are Release, Debug, RelWithDebInfo and MiniSizeRel. On
+    values are Release, Debug, RelWithDebInfo and MinSizeRel. On
     systems like Visual Studio the user sets the build type with the IDE
     settings.</dd>
 
@@ -260,7 +260,7 @@
   <dd>Build with threads support, if available. Defaults to ON.</dd>
 
   <dt><b>LLVM_ENABLE_ASSERTIONS</b>:BOOL</dt>
-  <dd>Enables code assertions. Defaults to ON if and only if
+  <dd>Enables code assertions. Defaults to OFF if and only if
     CMAKE_BUILD_TYPE is <i>Release</i>.</dd>
 
   <dt><b>LLVM_ENABLE_PIC</b>:BOOL</dt>




From nicolas.geoffray at lip6.fr  Wed Oct 21 10:06:23 2009
From: nicolas.geoffray at lip6.fr (Nicolas Geoffray)
Date: Wed, 21 Oct 2009 17:06:23 +0200
Subject: [llvm-commits] [llvm] r84651 - in /llvm/trunk:
 include/llvm/ExecutionEngine/JITMemoryManager.h
 lib/ExecutionEngine/JIT/JITEmitter.cpp
 lib/ExecutionEngine/JIT/JITMemoryManager.cpp
 unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
In-Reply-To: <200910201813.n9KIDLbV011114@zion.cs.uiuc.edu>
References: <200910201813.n9KIDLbV011114@zion.cs.uiuc.edu>
Message-ID: <4ADF236F.1070905@lip6.fr>

Hi Jeffrey,

There is a problem with your patch when the JIT runs out of memory, 
tries to deallocate the current buffer allocated for the function, and 
SIGSEGV because there is no buffer allocated for the exception table of 
the function.

The following patch fixes the issue. OK to commit?

Thanks,
Nicolas

Jeffrey Yasskin wrote:
> Author: jyasskin
> Date: Tue Oct 20 13:13:21 2009
> New Revision: 84651
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84651&view=rev
> Log:
> Move the Function*->allocated blocks map from the JITMemoryManager to the
> JITEmitter.
>
> I'm gradually making Functions auto-remove themselves from the JIT when they're
> destroyed. In this case, the Function needs to be removed from the JITEmitter,
> but the map recording which Functions need to be removed lived behind the
> JITMemoryManager interface, which made things difficult.
>
> This patch replaces the deallocateMemForFunction(Function*) method with a pair
> of methods deallocateFunctionBody(void *) and deallocateExceptionTable(void *)
> corresponding to the two startFoo/endFoo pairs.
>
>
> Modified:
>     llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h
>     llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp
>     llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
>     llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
>
> Modified: llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h?rev=84651&r1=84650&r2=84651&view=diff
>
> ==============================================================================
> --- llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h (original)
> +++ llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h Tue Oct 20 13:13:21 2009
> @@ -132,9 +132,11 @@
>    ///
>    virtual uint8_t *allocateGlobal(uintptr_t Size, unsigned Alignment) = 0;
>  
> -  /// deallocateMemForFunction - Free JIT memory for the specified function.
> -  /// This is never called when the JIT is currently emitting a function.
> -  virtual void deallocateMemForFunction(const Function *F) = 0;
> +  /// deallocateFunctionBody - Free the specified function body.  The argument
> +  /// must be the return value from a call to startFunctionBody() that hasn't
> +  /// been deallocated yet.  This is never called when the JIT is currently
> +  /// emitting a function.
> +  virtual void deallocateFunctionBody(void *Body) = 0;
>    
>    /// startExceptionTable - When we finished JITing the function, if exception
>    /// handling is set, we emit the exception table.
> @@ -146,6 +148,12 @@
>    virtual void endExceptionTable(const Function *F, uint8_t *TableStart,
>                                   uint8_t *TableEnd, uint8_t* FrameRegister) = 0;
>  
> +  /// deallocateExceptionTable - Free the specified exception table's memory.
> +  /// The argument must be the return value from a call to startExceptionTable()
> +  /// that hasn't been deallocated yet.  This is never called when the JIT is
> +  /// currently emitting an exception table.
> +  virtual void deallocateExceptionTable(void *ET) = 0;
> +
>    /// CheckInvariants - For testing only.  Return true if all internal
>    /// invariants are preserved, or return false and set ErrorStr to a helpful
>    /// error message.
>
> Modified: llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp?rev=84651&r1=84650&r2=84651&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp (original)
> +++ llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp Tue Oct 20 13:13:21 2009
> @@ -42,6 +42,7 @@
>  #include "llvm/System/Disassembler.h"
>  #include "llvm/System/Memory.h"
>  #include "llvm/Target/TargetInstrInfo.h"
> +#include "llvm/ADT/DenseMap.h"
>  #include "llvm/ADT/SmallPtrSet.h"
>  #include "llvm/ADT/SmallVector.h"
>  #include "llvm/ADT/Statistic.h"
> @@ -548,6 +549,13 @@
>      /// finishFunction.
>      JITEvent_EmittedFunctionDetails EmissionDetails;
>  
> +    struct EmittedCode {
> +      void *FunctionBody;
> +      void *ExceptionTable;
> +      EmittedCode() : FunctionBody(0), ExceptionTable(0) {}
> +    };
> +    DenseMap<const Function *, EmittedCode> EmittedFunctions;
> +
>      // CurFnStubUses - For a given Function, a vector of stubs that it
>      // references.  This facilitates the JIT detecting that a stub is no
>      // longer used, so that it may be deallocated.
> @@ -1011,7 +1019,8 @@
>    BufferBegin = CurBufferPtr = MemMgr->startFunctionBody(F.getFunction(),
>                                                           ActualSize);
>    BufferEnd = BufferBegin+ActualSize;
> -  
> +  EmittedFunctions[F.getFunction()].FunctionBody = BufferBegin;
> +
>    // Ensure the constant pool/jump table info is at least 4-byte aligned.
>    emitAlignment(16);
>  
> @@ -1201,6 +1210,7 @@
>      BufferBegin = CurBufferPtr = MemMgr->startExceptionTable(F.getFunction(),
>                                                               ActualSize);
>      BufferEnd = BufferBegin+ActualSize;
> +    EmittedFunctions[F.getFunction()].ExceptionTable = BufferBegin;
>      uint8_t *EhStart;
>      uint8_t *FrameRegister = DE->EmitDwarfTable(F, *this, FnStart, FnEnd,
>                                                  EhStart);
> @@ -1244,7 +1254,13 @@
>  /// deallocateMemForFunction - Deallocate all memory for the specified
>  /// function body.  Also drop any references the function has to stubs.
>  void JITEmitter::deallocateMemForFunction(const Function *F) {
> -  MemMgr->deallocateMemForFunction(F);
> +  DenseMap<const Function *, EmittedCode>::iterator Emitted =
> +    EmittedFunctions.find(F);
> +  if (Emitted != EmittedFunctions.end()) {
> +    MemMgr->deallocateFunctionBody(Emitted->second.FunctionBody);
> +    MemMgr->deallocateExceptionTable(Emitted->second.ExceptionTable);
> +    EmittedFunctions.erase(Emitted);
> +  }
>  
>    // TODO: Do we need to unregister exception handling information from libgcc
>    // here?
>
> Modified: llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp?rev=84651&r1=84650&r2=84651&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp (original)
> +++ llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp Tue Oct 20 13:13:21 2009
> @@ -297,9 +297,6 @@
>  
>      uint8_t *GOTBase;     // Target Specific reserved memory
>      void *DlsymTable;     // Stub external symbol information
> -
> -    std::map<const Function*, MemoryRangeHeader*> FunctionBlocks;
> -    std::map<const Function*, MemoryRangeHeader*> TableBlocks;
>    public:
>      DefaultJITMemoryManager();
>      ~DefaultJITMemoryManager();
> @@ -414,7 +411,6 @@
>               "Mismatched function start/end!");
>  
>        uintptr_t BlockSize = FunctionEnd - (uint8_t *)CurBlock;
> -      FunctionBlocks[F] = CurBlock;
>  
>        // Release the memory at the end of this block that isn't needed.
>        FreeMemoryList =CurBlock->TrimAllocationToSize(FreeMemoryList, BlockSize);
> @@ -464,7 +460,6 @@
>               "Mismatched table start/end!");
>        
>        uintptr_t BlockSize = TableEnd - (uint8_t *)CurBlock;
> -      TableBlocks[F] = CurBlock;
>  
>        // Release the memory at the end of this block that isn't needed.
>        FreeMemoryList =CurBlock->TrimAllocationToSize(FreeMemoryList, BlockSize);
> @@ -478,15 +473,9 @@
>        return DlsymTable;
>      }
>      
> -    /// deallocateMemForFunction - Deallocate all memory for the specified
> -    /// function body.
> -    void deallocateMemForFunction(const Function *F) {
> -      std::map<const Function*, MemoryRangeHeader*>::iterator
> -        I = FunctionBlocks.find(F);
> -      if (I == FunctionBlocks.end()) return;
> -      
> +    void deallocateBlock(void *Block) {
>        // Find the block that is allocated for this function.
> -      MemoryRangeHeader *MemRange = I->second;
> +      MemoryRangeHeader *MemRange = static_cast<MemoryRangeHeader*>(Block) - 1;
>        assert(MemRange->ThisAllocated && "Block isn't allocated!");
>  
>        // Fill the buffer with garbage!
> @@ -496,27 +485,18 @@
>  
>        // Free the memory.
>        FreeMemoryList = MemRange->FreeBlock(FreeMemoryList);
> -      
> -      // Finally, remove this entry from FunctionBlocks.
> -      FunctionBlocks.erase(I);
> -      
> -      I = TableBlocks.find(F);
> -      if (I == TableBlocks.end()) return;
> -      
> -      // Find the block that is allocated for this function.
> -      MemRange = I->second;
> -      assert(MemRange->ThisAllocated && "Block isn't allocated!");
> +    }
>  
> -      // Fill the buffer with garbage!
> -      if (PoisonMemory) {
> -        memset(MemRange+1, 0xCD, MemRange->BlockSize-sizeof(*MemRange));
> -      }
> +    /// deallocateFunctionBody - Deallocate all memory for the specified
> +    /// function body.
> +    void deallocateFunctionBody(void *Body) {
> +      deallocateBlock(Body);
> +    }
>  
> -      // Free the memory.
> -      FreeMemoryList = MemRange->FreeBlock(FreeMemoryList);
> -      
> -      // Finally, remove this entry from TableBlocks.
> -      TableBlocks.erase(I);
> +    /// deallocateExceptionTable - Deallocate memory for the specified
> +    /// exception table.
> +    void deallocateExceptionTable(void *ET) {
> +      deallocateBlock(ET);
>      }
>  
>      /// setMemoryWritable - When code generation is in progress,
>
> Modified: llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp?rev=84651&r1=84650&r2=84651&view=diff
>
> ==============================================================================
> --- llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp (original)
> +++ llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp Tue Oct 20 13:13:21 2009
> @@ -32,37 +32,36 @@
>    OwningPtr<JITMemoryManager> MemMgr(
>        JITMemoryManager::CreateDefaultMemManager());
>    uintptr_t size;
> -  uint8_t *start;
>    std::string Error;
>  
>    // Allocate the functions.
>    OwningPtr<Function> F1(makeFakeFunction());
>    size = 1024;
> -  start = MemMgr->startFunctionBody(F1.get(), size);
> -  memset(start, 0xFF, 1024);
> -  MemMgr->endFunctionBody(F1.get(), start, start + 1024);
> +  uint8_t *FunctionBody1 = MemMgr->startFunctionBody(F1.get(), size);
> +  memset(FunctionBody1, 0xFF, 1024);
> +  MemMgr->endFunctionBody(F1.get(), FunctionBody1, FunctionBody1 + 1024);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  
>    OwningPtr<Function> F2(makeFakeFunction());
>    size = 1024;
> -  start = MemMgr->startFunctionBody(F2.get(), size);
> -  memset(start, 0xFF, 1024);
> -  MemMgr->endFunctionBody(F2.get(), start, start + 1024);
> +  uint8_t *FunctionBody2 = MemMgr->startFunctionBody(F2.get(), size);
> +  memset(FunctionBody2, 0xFF, 1024);
> +  MemMgr->endFunctionBody(F2.get(), FunctionBody2, FunctionBody2 + 1024);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  
>    OwningPtr<Function> F3(makeFakeFunction());
>    size = 1024;
> -  start = MemMgr->startFunctionBody(F3.get(), size);
> -  memset(start, 0xFF, 1024);
> -  MemMgr->endFunctionBody(F3.get(), start, start + 1024);
> +  uint8_t *FunctionBody3 = MemMgr->startFunctionBody(F3.get(), size);
> +  memset(FunctionBody3, 0xFF, 1024);
> +  MemMgr->endFunctionBody(F3.get(), FunctionBody3, FunctionBody3 + 1024);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  
>    // Deallocate them out of order, in case that matters.
> -  MemMgr->deallocateMemForFunction(F2.get());
> +  MemMgr->deallocateFunctionBody(FunctionBody2);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
> -  MemMgr->deallocateMemForFunction(F1.get());
> +  MemMgr->deallocateFunctionBody(FunctionBody1);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
> -  MemMgr->deallocateMemForFunction(F3.get());
> +  MemMgr->deallocateFunctionBody(FunctionBody3);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  }
>  
> @@ -72,7 +71,6 @@
>    OwningPtr<JITMemoryManager> MemMgr(
>        JITMemoryManager::CreateDefaultMemManager());
>    uintptr_t size;
> -  uint8_t *start;
>    std::string Error;
>  
>    // Big functions are a little less than the largest block size.
> @@ -83,26 +81,26 @@
>    // Allocate big functions
>    OwningPtr<Function> F1(makeFakeFunction());
>    size = bigFuncSize;
> -  start = MemMgr->startFunctionBody(F1.get(), size);
> +  uint8_t *FunctionBody1 = MemMgr->startFunctionBody(F1.get(), size);
>    ASSERT_LE(bigFuncSize, size);
> -  memset(start, 0xFF, bigFuncSize);
> -  MemMgr->endFunctionBody(F1.get(), start, start + bigFuncSize);
> +  memset(FunctionBody1, 0xFF, bigFuncSize);
> +  MemMgr->endFunctionBody(F1.get(), FunctionBody1, FunctionBody1 + bigFuncSize);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  
>    OwningPtr<Function> F2(makeFakeFunction());
>    size = bigFuncSize;
> -  start = MemMgr->startFunctionBody(F2.get(), size);
> +  uint8_t *FunctionBody2 = MemMgr->startFunctionBody(F2.get(), size);
>    ASSERT_LE(bigFuncSize, size);
> -  memset(start, 0xFF, bigFuncSize);
> -  MemMgr->endFunctionBody(F2.get(), start, start + bigFuncSize);
> +  memset(FunctionBody2, 0xFF, bigFuncSize);
> +  MemMgr->endFunctionBody(F2.get(), FunctionBody2, FunctionBody2 + bigFuncSize);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  
>    OwningPtr<Function> F3(makeFakeFunction());
>    size = bigFuncSize;
> -  start = MemMgr->startFunctionBody(F3.get(), size);
> +  uint8_t *FunctionBody3 = MemMgr->startFunctionBody(F3.get(), size);
>    ASSERT_LE(bigFuncSize, size);
> -  memset(start, 0xFF, bigFuncSize);
> -  MemMgr->endFunctionBody(F3.get(), start, start + bigFuncSize);
> +  memset(FunctionBody3, 0xFF, bigFuncSize);
> +  MemMgr->endFunctionBody(F3.get(), FunctionBody3, FunctionBody3 + bigFuncSize);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  
>    // Check that each large function took it's own slab.
> @@ -111,43 +109,46 @@
>    // Allocate small functions
>    OwningPtr<Function> F4(makeFakeFunction());
>    size = smallFuncSize;
> -  start = MemMgr->startFunctionBody(F4.get(), size);
> +  uint8_t *FunctionBody4 = MemMgr->startFunctionBody(F4.get(), size);
>    ASSERT_LE(smallFuncSize, size);
> -  memset(start, 0xFF, smallFuncSize);
> -  MemMgr->endFunctionBody(F4.get(), start, start + smallFuncSize);
> +  memset(FunctionBody4, 0xFF, smallFuncSize);
> +  MemMgr->endFunctionBody(F4.get(), FunctionBody4,
> +                          FunctionBody4 + smallFuncSize);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  
>    OwningPtr<Function> F5(makeFakeFunction());
>    size = smallFuncSize;
> -  start = MemMgr->startFunctionBody(F5.get(), size);
> +  uint8_t *FunctionBody5 = MemMgr->startFunctionBody(F5.get(), size);
>    ASSERT_LE(smallFuncSize, size);
> -  memset(start, 0xFF, smallFuncSize);
> -  MemMgr->endFunctionBody(F5.get(), start, start + smallFuncSize);
> +  memset(FunctionBody5, 0xFF, smallFuncSize);
> +  MemMgr->endFunctionBody(F5.get(), FunctionBody5,
> +                          FunctionBody5 + smallFuncSize);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  
>    OwningPtr<Function> F6(makeFakeFunction());
>    size = smallFuncSize;
> -  start = MemMgr->startFunctionBody(F6.get(), size);
> +  uint8_t *FunctionBody6 = MemMgr->startFunctionBody(F6.get(), size);
>    ASSERT_LE(smallFuncSize, size);
> -  memset(start, 0xFF, smallFuncSize);
> -  MemMgr->endFunctionBody(F6.get(), start, start + smallFuncSize);
> +  memset(FunctionBody6, 0xFF, smallFuncSize);
> +  MemMgr->endFunctionBody(F6.get(), FunctionBody6,
> +                          FunctionBody6 + smallFuncSize);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  
>    // Check that the small functions didn't allocate any new slabs.
>    EXPECT_EQ(3U, MemMgr->GetNumCodeSlabs());
>  
>    // Deallocate them out of order, in case that matters.
> -  MemMgr->deallocateMemForFunction(F2.get());
> +  MemMgr->deallocateFunctionBody(FunctionBody2);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
> -  MemMgr->deallocateMemForFunction(F1.get());
> +  MemMgr->deallocateFunctionBody(FunctionBody1);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
> -  MemMgr->deallocateMemForFunction(F4.get());
> +  MemMgr->deallocateFunctionBody(FunctionBody4);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
> -  MemMgr->deallocateMemForFunction(F3.get());
> +  MemMgr->deallocateFunctionBody(FunctionBody3);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
> -  MemMgr->deallocateMemForFunction(F5.get());
> +  MemMgr->deallocateFunctionBody(FunctionBody5);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
> -  MemMgr->deallocateMemForFunction(F6.get());
> +  MemMgr->deallocateFunctionBody(FunctionBody6);
>    EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>  }
>  
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>   

-------------- next part --------------
A non-text attachment was scrubbed...
Name: deallocate.patch
Type: text/x-patch
Size: 692 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091021/b3f23f72/attachment.bin 

From grosbach at apple.com  Wed Oct 21 10:26:25 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Wed, 21 Oct 2009 15:26:25 -0000
Subject: [llvm-commits] [llvm] r84761 -
	/llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp
Message-ID: <200910211526.n9LFQQ79025230@zion.cs.uiuc.edu>

Author: grosbach
Date: Wed Oct 21 10:26:21 2009
New Revision: 84761

URL: http://llvm.org/viewvc/llvm-project?rev=84761&view=rev
Log:
Cleanup of frame index scavenging. Better code flow and more accurately
handles T2 and ARM use cases.

Modified:
    llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp

Modified: llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp?rev=84761&r1=84760&r2=84761&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp (original)
+++ llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp Wed Oct 21 10:26:21 2009
@@ -729,10 +729,12 @@
 /// the instruciton range. Return the operand number of the kill in Operand.
 static MachineBasicBlock::iterator
 findLastUseReg(MachineBasicBlock::iterator I, MachineBasicBlock::iterator ME,
-               unsigned Reg, unsigned *Operand) {
+               unsigned Reg) {
   // Scan forward to find the last use of this virtual register
   for (++I; I != ME; ++I) {
     MachineInstr *MI = I;
+    bool isDefInsn = false;
+    bool isKillInsn = false;
     for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i)
       if (MI->getOperand(i).isReg()) {
         unsigned OpReg = MI->getOperand(i).getReg();
@@ -740,13 +742,14 @@
           continue;
         assert (OpReg == Reg
                 && "overlapping use of scavenged index register!");
-        // If this is the killing use, we're done
-        if (MI->getOperand(i).isKill()) {
-          if (Operand)
-            *Operand = i;
-          return I;
-        }
+        // If this is the killing use, we have a candidate.
+        if (MI->getOperand(i).isKill())
+          isKillInsn = true;
+        else if (MI->getOperand(i).isDef())
+          isDefInsn = true;
       }
+    if (isKillInsn && !isDefInsn)
+      return I;
   }
   // If we hit the end of the basic block, there was no kill of
   // the virtual register, which is wrong.
@@ -763,10 +766,13 @@
        E = Fn.end(); BB != E; ++BB) {
     RS->enterBasicBlock(BB);
 
+    // FIXME: The logic flow in this function is still too convoluted.
+    // It needs a cleanup refactoring. Do that in preparation for tracking
+    // more than one scratch register value and using ranges to find
+    // available scratch registers.
     unsigned CurrentVirtReg = 0;
     unsigned CurrentScratchReg = 0;
     bool havePrevValue = false;
-    unsigned PrevScratchReg = 0;
     int PrevValue = 0;
     MachineInstr *PrevLastUseMI = NULL;
     unsigned PrevLastUseOp = 0;
@@ -776,11 +782,13 @@
 
     // The instruction stream may change in the loop, so check BB->end()
     // directly.
-    for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) {
+    for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ) {
       MachineInstr *MI = I;
       bool isDefInsn = false;
       bool isKillInsn = false;
-      for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i)
+      bool clobbersScratchReg = false;
+      bool DoIncr = true;
+      for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
         if (MI->getOperand(i).isReg()) {
           MachineOperand &MO = MI->getOperand(i);
           unsigned Reg = MO.getReg();
@@ -789,17 +797,14 @@
           if (!TargetRegisterInfo::isVirtualRegister(Reg)) {
             // If we have a previous scratch reg, check and see if anything
             // here kills whatever value is in there.
-            if (Reg == PrevScratchReg) {
+            if (Reg == CurrentScratchReg) {
               if (MO.isUse()) {
                 // Two-address operands implicitly kill
-                if (MO.isKill() || MI->isRegTiedToDefOperand(i)) {
-                  havePrevValue = false;
-                  PrevScratchReg = 0;
-                }
+                if (MO.isKill() || MI->isRegTiedToDefOperand(i))
+                  clobbersScratchReg = true;
               } else {
                 assert (MO.isDef());
-                havePrevValue = false;
-                PrevScratchReg = 0;
+                clobbersScratchReg = true;
               }
             }
             continue;
@@ -840,37 +845,46 @@
               // for the virtual register are exclusively for the purpose
               // of populating the value in the register. That's reasonable
               // for these frame index registers, but it's still a very, very
-              // strong assumption. Perhaps this implies that the frame index
-              // elimination should be before register allocation, with
-              // conservative heuristics since we'll know less then, and
-              // the reuse calculations done directly when doing the code-gen?
+              // strong assumption. rdar://7322732. Better would be to
+              // explicitly check each instruction in the range for references
+              // to the virtual register. Only delete those insns that
+              // touch the virtual register.
 
               // Find the last use of the new virtual register. Remove all
               // instruction between here and there, and update the current
               // instruction to reference the last use insn instead.
               MachineBasicBlock::iterator LastUseMI =
-                findLastUseReg(I, BB->end(), Reg, &i);
+                findLastUseReg(I, BB->end(), Reg);
+
               // Remove all instructions up 'til the last use, since they're
               // just calculating the value we already have.
               BB->erase(I, LastUseMI);
               MI = I = LastUseMI;
-              e = MI->getNumOperands();
 
-              CurrentScratchReg = PrevScratchReg;
-              // Extend the live range of the register
+              // Extend the live range of the scratch register
               PrevLastUseMI->getOperand(PrevLastUseOp).setIsKill(false);
               RS->setUsed(CurrentScratchReg);
-            } else {
               CurrentVirtReg = Reg;
-              const TargetRegisterClass *RC = Fn.getRegInfo().getRegClass(Reg);
-              CurrentScratchReg = RS->FindUnusedReg(RC);
-              if (CurrentScratchReg == 0)
-                // No register is "free". Scavenge a register.
-                CurrentScratchReg = RS->scavengeRegister(RC, I, SPAdj);
 
-              PrevValue = Value;
+              // We deleted the instruction we were scanning the operands of.
+              // Jump back to the instruction iterator loop. Don't increment
+              // past this instruction since we updated the iterator already.
+              DoIncr = false;
+              break;
             }
+
+            // Scavenge a new scratch register
+            CurrentVirtReg = Reg;
+            const TargetRegisterClass *RC = Fn.getRegInfo().getRegClass(Reg);
+            CurrentScratchReg = RS->FindUnusedReg(RC);
+            if (CurrentScratchReg == 0)
+              // No register is "free". Scavenge a register.
+              CurrentScratchReg = RS->scavengeRegister(RC, I, SPAdj);
+
+            PrevValue = Value;
           }
+          // replace this reference to the virtual register with the
+          // scratch register.
           assert (CurrentScratchReg && "Missing scratch register!");
           MI->getOperand(i).setReg(CurrentScratchReg);
 
@@ -880,15 +894,26 @@
             PrevLastUseMI = MI;
           }
         }
+      }
       // If this is the last use of the scratch, stop tracking it. The
       // last use will be a kill operand in an instruction that does
       // not also define the scratch register.
       if (isKillInsn && !isDefInsn) {
-        PrevScratchReg = CurrentScratchReg;
-        CurrentScratchReg = CurrentVirtReg = 0;
+        CurrentVirtReg = 0;
         havePrevValue = trackingCurrentValue;
       }
-      RS->forward(MI);
+      // Similarly, notice if instruction clobbered the value in the
+      // register we're tracking for possible later reuse. This is noted
+      // above, but enforced here since the value is still live while we
+      // process the rest of the operands of the instruction.
+      if (clobbersScratchReg) {
+        havePrevValue = false;
+        CurrentScratchReg = 0;
+      }
+      if (DoIncr) {
+        RS->forward(I);
+        ++I;
+      }
     }
   }
 }




From edwintorok at gmail.com  Wed Oct 21 12:08:21 2009
From: edwintorok at gmail.com (=?ISO-8859-1?Q?T=F6r=F6k_Edwin?=)
Date: Wed, 21 Oct 2009 20:08:21 +0300
Subject: [llvm-commits] [PATCH] Introduce -dbg-quality test
In-Reply-To: <352a1fb20910131016g41abc9e0v3ff54f001324075c@mail.gmail.com>
References: <4ACDAF90.1040202@gmail.com>	
	<B59C3DBC-4810-425A-917A-4E857AEC4CEA@apple.com>	
	<4AD22FBF.3060607@gmail.com>
	<352a1fb20910131016g41abc9e0v3ff54f001324075c@mail.gmail.com>
Message-ID: <4ADF4005.20201@gmail.com>

Hi Devang,

I've written the -dbg-quality pass (attached), and the
TEST.dbgquality.Makefile that tests the optimized (-std-compile-opts)
bitcode file. Should I also add a -std-link-opts tests? I think some
passes (such as -ipsccp) are only run there?
I've attached the results, the maximum score is 10000 (meaning all
non-alloca/non-dbgintrinsic instructions have debug info).
We have 79 tests at 10000 quality, the rest are between 5331 - 9999.

I've also found a bug: static constructors/destructors have out-of-file
line numbers often (PR5270). I've added a hack in -dbg-quality for now
that ignores those functions (denoted by a FIXME).

OK to commit, or should I wait till PR5270 is fixed so I can drop the
ugly hack?

Best regards,
--Edwin
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: sorted.txt
Url: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091021/09fea276/attachment.txt 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: llvm.patch
Type: text/x-diff
Size: 10122 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091021/09fea276/attachment.bin 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: testsuite.patch
Type: text/x-diff
Size: 2411 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091021/09fea276/attachment-0001.bin 

From devang.patel at gmail.com  Wed Oct 21 12:23:21 2009
From: devang.patel at gmail.com (Devang Patel)
Date: Wed, 21 Oct 2009 10:23:21 -0700
Subject: [llvm-commits] [PATCH] Introduce -dbg-quality test
In-Reply-To: <4ADF4005.20201@gmail.com>
References: <4ACDAF90.1040202@gmail.com>
	<B59C3DBC-4810-425A-917A-4E857AEC4CEA@apple.com>
	<4AD22FBF.3060607@gmail.com>
	<352a1fb20910131016g41abc9e0v3ff54f001324075c@mail.gmail.com>
	<4ADF4005.20201@gmail.com>
Message-ID: <352a1fb20910211023n4351ff5ate009c4297ea454df@mail.gmail.com>

Hi

2009/10/21 T?r?k Edwin <edwintorok at gmail.com>:
> Hi Devang,
>
> I've written the -dbg-quality pass (attached), and the
> TEST.dbgquality.Makefile that tests the optimized (-std-compile-opts)
> bitcode file.

This is great. Could you explain how you qualify debug info to get the
following score ?

> Should I also add a -std-link-opts tests?

Yes.

> I think some
> passes (such as -ipsccp) are only run there?
> I've attached the results, the maximum score is 10000 (meaning all
> non-alloca/non-dbgintrinsic instructions have debug info).
> We have 79 tests at 10000 quality, the rest are between 5331 - 9999.
>
> I've also found a bug: static constructors/destructors have out-of-file
> line numbers often (PR5270). I've added a hack in -dbg-quality for now
> that ignores those functions (denoted by a FIXME).

I'll take a look at this PR.

> OK to commit, or should I wait till PR5270 is fixed so I can drop the
> ugly hack?

I do not think Analysis is a good place for this and DebugInfo as
well. I think, DebugInfo is in Analysis, because we do not want it in
VMCore. IMO, debug info related code should be in lib/Support. Chris,
what do you think ?

Thanks for working on this!
-
Devang



From sanjiv.gupta at microchip.com  Wed Oct 21 12:27:23 2009
From: sanjiv.gupta at microchip.com (Sanjiv Gupta)
Date: Wed, 21 Oct 2009 17:27:23 -0000
Subject: [llvm-commits] [llvm] r84764 -
	/llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
Message-ID: <200910211727.n9LHRO3h031223@zion.cs.uiuc.edu>

Author: sgupta
Date: Wed Oct 21 12:27:23 2009
New Revision: 84764

URL: http://llvm.org/viewvc/llvm-project?rev=84764&view=rev
Log:
Build shared lib instead of an archive.

Modified:
    llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile

Modified: llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile?rev=84764&r1=84763&r2=84764&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile (original)
+++ llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile Wed Oct 21 12:27:23 2009
@@ -9,7 +9,7 @@
 LEVEL = ../../../..
 TARGET = PIC16
 LIBRARYNAME = LLVMpic16passes
-BUILD_ARCHIVE = 1
+LOADABLE_MODULE = 1
 
 
 




From dpatel at apple.com  Wed Oct 21 12:33:41 2009
From: dpatel at apple.com (Devang Patel)
Date: Wed, 21 Oct 2009 17:33:41 -0000
Subject: [llvm-commits] [llvm] r84765 - in /llvm/trunk:
 include/llvm/Metadata.h lib/VMCore/Metadata.cpp
Message-ID: <200910211733.n9LHXfVw031547@zion.cs.uiuc.edu>

Author: dpatel
Date: Wed Oct 21 12:33:41 2009
New Revision: 84765

URL: http://llvm.org/viewvc/llvm-project?rev=84765&view=rev
Log:
Incorporate various suggestions Chris gave during metadata review.

- i < getNumElements()  instead of getNumElements() > i
- Make setParent() private
- Fix use of resizeOperands
- Reset HasMetadata bit after removing all metadata attached to an instruction
- Efficient use of iterators


Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84765&r1=84764&r2=84765&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Wed Oct 21 12:33:41 2009
@@ -145,7 +145,7 @@
   
   /// getElement - Return specified element.
   Value *getElement(unsigned i) const {
-    assert(getNumElements() > i && "Invalid element number!");
+    assert(i < getNumElements() && "Invalid element number!");
     return Node[i];
   }
 
@@ -211,6 +211,7 @@
   SmallVector<WeakMetadataVH, 4> Node;
   typedef SmallVectorImpl<WeakMetadataVH>::iterator elem_iterator;
 
+  void setParent(Module *M) { Parent = M; }
 protected:
   explicit NamedMDNode(LLVMContext &C, const Twine &N, MetadataBase*const *Vals, 
                        unsigned NumVals, Module *M = 0);
@@ -240,11 +241,10 @@
   /// getParent - Get the module that holds this named metadata collection.
   inline Module *getParent() { return Parent; }
   inline const Module *getParent() const { return Parent; }
-  void setParent(Module *M) { Parent = M; }
 
   /// getElement - Return specified element.
   MetadataBase *getElement(unsigned i) const {
-    assert(getNumElements() > i && "Invalid element number!");
+    assert(i < getNumElements() && "Invalid element number!");
     return Node[i];
   }
 
@@ -255,7 +255,7 @@
 
   /// addElement - Add metadata element.
   void addElement(MetadataBase *M) {
-    resizeOperands(0);
+    resizeOperands(NumOperands + 1);
     OperandList[NumOperands++] = M;
     Node.push_back(WeakMetadataVH(M));
   }
@@ -319,8 +319,8 @@
   /// removeMD - Remove metadata of given kind attached with an instuction.
   void removeMD(unsigned Kind, Instruction *Inst);
   
-  /// removeMDs - Remove all metadata attached with an instruction.
-  void removeMDs(const Instruction *Inst);
+  /// removeAllMetadata - Remove all metadata attached with an instruction.
+  void removeAllMetadata(Instruction *Inst);
 
   /// copyMD - If metadata is attached with Instruction In1 then attach
   /// the same metadata to In2.
@@ -333,8 +333,8 @@
   /// ValueIsDeleted - This handler is used to update metadata store
   /// when a value is deleted.
   void ValueIsDeleted(const Value *) {}
-  void ValueIsDeleted(const Instruction *Inst) {
-    removeMDs(Inst);
+  void ValueIsDeleted(Instruction *Inst) {
+    removeAllMetadata(Inst);
   }
   void ValueIsRAUWd(Value *V1, Value *V2);
 

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84765&r1=84764&r2=84765&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 12:33:41 2009
@@ -250,10 +250,8 @@
 unsigned MetadataContext::registerMDKind(const char *Name) {
   assert(isValidName(Name) && "Invalid custome metadata name!");
   unsigned Count = MDHandlerNames.size();
-  assert(MDHandlerNames.find(Name) == MDHandlerNames.end() 
-         && "Already registered MDKind!");
-  MDHandlerNames[Name] = Count + 1;
-  return Count + 1;
+  assert(MDHandlerNames.count(Name) == 0 && "Already registered MDKind!");
+  return MDHandlerNames[Name] = Count + 1;
 }
 
 /// isValidName - Return true if Name is a valid custom metadata handler name.
@@ -280,10 +278,11 @@
 /// getMDKind - Return metadata kind. If the requested metadata kind
 /// is not registered then return 0.
 unsigned MetadataContext::getMDKind(const char *Name) {
-  assert(isValidName(Name) && "Invalid custome metadata name!");
   StringMap<unsigned>::iterator I = MDHandlerNames.find(Name);
-  if (I == MDHandlerNames.end())
+  if (I == MDHandlerNames.end()) {
+    assert(isValidName(Name) && "Invalid custome metadata name!");
     return 0;
+  }
 
   return I->getValue();
 }
@@ -292,15 +291,13 @@
 void MetadataContext::addMD(unsigned MDKind, MDNode *Node, Instruction *Inst) {
   assert(Node && "Invalid null MDNode");
   Inst->HasMetadata = true;
-  MDStoreTy::iterator I = MetadataStore.find(Inst);
-  if (I == MetadataStore.end()) {
-    MDMapTy Info;
+  MDMapTy &Info = MetadataStore[Inst];
+  if (Info.empty()) {
     Info.push_back(std::make_pair(MDKind, Node));
     MetadataStore.insert(std::make_pair(Inst, Info));
     return;
   }
 
-  MDMapTy &Info = I->second;
   // If there is an entry for this MDKind then replace it.
   for (unsigned i = 0, e = Info.size(); i != e; ++i) {
     MDPairTy &P = Info[i];
@@ -333,30 +330,20 @@
   return;
 }
   
-/// removeMDs - Remove all metadata attached with an instruction.
-void MetadataContext::removeMDs(const Instruction *Inst) {
-  // Find Metadata handles for this instruction.
-  MDStoreTy::iterator I = MetadataStore.find(Inst);
-  assert(I != MetadataStore.end() && "Invalid custom metadata info!");
-  MDMapTy &Info = I->second;
-  
-  // FIXME : Give all metadata handlers a chance to adjust.
-  
-  // Remove the entries for this instruction.
-  Info.clear();
-  MetadataStore.erase(I);
+/// removeAllMetadata - Remove all metadata attached with an instruction.
+void MetadataContext::removeAllMetadata(Instruction *Inst) {
+  MetadataStore.erase(Inst);
+  Inst->HasMetadata = false;
 }
 
 /// copyMD - If metadata is attached with Instruction In1 then attach
 /// the same metadata to In2.
 void MetadataContext::copyMD(Instruction *In1, Instruction *In2) {
   assert(In1 && In2 && "Invalid instruction!");
-   MDStoreTy::iterator I = MetadataStore.find(In1);
-  if (I == MetadataStore.end())
+  MDMapTy &In1Info = MetadataStore[In1];
+  if (In1Info.empty())
     return;
 
-  MDMapTy &In1Info = I->second;
-  MDMapTy In2Info;
   for (MDMapTy::iterator I = In1Info.begin(), E = In1Info.end(); I != E; ++I)
     if (MDNode *MD = dyn_cast_or_null<MDNode>(I->second))
       addMD(I->first, MD, In2);
@@ -365,11 +352,10 @@
 /// getMD - Get the metadata of given kind attached to an Instruction.
 /// If the metadata is not found then return 0.
 MDNode *MetadataContext::getMD(unsigned MDKind, const Instruction *Inst) {
-  MDStoreTy::iterator I = MetadataStore.find(Inst);
-  if (I == MetadataStore.end())
+  MDMapTy &Info = MetadataStore[Inst];
+  if (Info.empty())
     return NULL;
-  
-  MDMapTy &Info = I->second;
+
   for (MDMapTy::iterator I = Info.begin(), E = Info.end(); I != E; ++I)
     if (I->first == MDKind)
       return dyn_cast_or_null<MDNode>(I->second);




From johnny.chen at apple.com  Wed Oct 21 12:36:42 2009
From: johnny.chen at apple.com (Johnny Chen)
Date: Wed, 21 Oct 2009 10:36:42 -0700
Subject: [llvm-commits] ARM .td patch to specify only the	well-known
	encoding bits for imm6 and imm4 fields
In-Reply-To: <0D809F55-1781-454C-93DD-9389CD11B645@apple.com>
References: <E0C31726-F049-46CB-B443-8C01B7FF9AE0@apple.com>
	<FA34BFD6-78D3-45A0-8D9B-B77613EC44B5@apple.com>
	<0D809F55-1781-454C-93DD-9389CD11B645@apple.com>
Message-ID: <C35AE8B9-C9C6-4E26-A522-64A71663A525@apple.com>

Hi Bob,

Thanks a lot!
Also, there is a typo for the first argument passed to VST2LN<...>.
The Inst{9-8} should be 0b01, not 0b00.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: ARM-NEON.patch
Type: application/octet-stream
Size: 1154 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091021/8384b6ff/attachment.obj 
-------------- next part --------------


On Oct 20, 2009, at 7:19 PM, Bob Wilson wrote:

> 
> On Oct 20, 2009, at 12:01 PM, Bob Wilson wrote:
> 
>> 
>> On Oct 20, 2009, at 10:49 AM, Johnny Chen wrote:
>> 
>>> Hi,
>>> 
>>> Refs: A8-754 (VSHLL and friends), A8-580 (VCVT between floating- 
>>> point and fixed-point, Advanced SIMD), A8-592 (VDUP scalar)
>>> 
>>> This patch attempts to specify just the encoding bits described in  
>>> the manual for several vector shift, vector convert, and vector dupe
>>> operations.
>>> 
>>> One question I have is whether we should explicitly specify that  
>>> VCVT's imm6 is of the pattern '1xxxxx'.
>>> In the submitted patch, I left the imm6 (Inst{21-16}) of VCVT  
>>> unspecified.
>>> 
>>> Thanks.
>> 
>> This is exactly the problem I was hoping to avoid.  Please see my  
>> response to your VEXT patch before we decide whether to use this  
>> approach.
> 
> I've committed a revised version of your patch.  There was only one instruction left using the old N2VImm class, and that was really just a minor convenience.  I changed that one class to use N2V instead so that all the other cases can use the N2VImm name.  I also did some refactoring of the new multiclasses that you had added and made some whitespace changes, etc.


From bob.wilson at apple.com  Wed Oct 21 12:47:08 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Wed, 21 Oct 2009 10:47:08 -0700
Subject: [llvm-commits] ARM .td patch to specify only the	well-known
	encoding bits for imm6 and imm4 fields
In-Reply-To: <C35AE8B9-C9C6-4E26-A522-64A71663A525@apple.com>
References: <E0C31726-F049-46CB-B443-8C01B7FF9AE0@apple.com>
	<FA34BFD6-78D3-45A0-8D9B-B77613EC44B5@apple.com>
	<0D809F55-1781-454C-93DD-9389CD11B645@apple.com>
	<C35AE8B9-C9C6-4E26-A522-64A71663A525@apple.com>
Message-ID: <11EA687C-6483-4D81-8607-B330239E1809@apple.com>

Doh!  You pointed this out yesterday, and I thought I had fixed it in  
r84732, but instead of changing Inst{9-8}, I changed Inst{5-4}.  I'll  
revert that and apply your patch.

On Oct 21, 2009, at 10:36 AM, Johnny Chen wrote:

> Hi Bob,
>
> Thanks a lot!
> Also, there is a typo for the first argument passed to VST2LN<...>.
> The Inst{9-8} should be 0b01, not 0b00.
>
> <ARM-NEON.patch>
>
> On Oct 20, 2009, at 7:19 PM, Bob Wilson wrote:
>
>>
>> On Oct 20, 2009, at 12:01 PM, Bob Wilson wrote:
>>
>>>
>>> On Oct 20, 2009, at 10:49 AM, Johnny Chen wrote:
>>>
>>>> Hi,
>>>>
>>>> Refs: A8-754 (VSHLL and friends), A8-580 (VCVT between floating-
>>>> point and fixed-point, Advanced SIMD), A8-592 (VDUP scalar)
>>>>
>>>> This patch attempts to specify just the encoding bits described in
>>>> the manual for several vector shift, vector convert, and vector  
>>>> dupe
>>>> operations.
>>>>
>>>> One question I have is whether we should explicitly specify that
>>>> VCVT's imm6 is of the pattern '1xxxxx'.
>>>> In the submitted patch, I left the imm6 (Inst{21-16}) of VCVT
>>>> unspecified.
>>>>
>>>> Thanks.
>>>
>>> This is exactly the problem I was hoping to avoid.  Please see my
>>> response to your VEXT patch before we decide whether to use this
>>> approach.
>>
>> I've committed a revised version of your patch.  There was only one  
>> instruction left using the old N2VImm class, and that was really  
>> just a minor convenience.  I changed that one class to use N2V  
>> instead so that all the other cases can use the N2VImm name.  I  
>> also did some refactoring of the new multiclasses that you had  
>> added and made some whitespace changes, etc.
>



From bob.wilson at apple.com  Wed Oct 21 12:52:34 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Wed, 21 Oct 2009 17:52:34 -0000
Subject: [llvm-commits] [llvm] r84766 -
	/llvm/trunk/lib/Target/ARM/ARMInstrNEON.td
Message-ID: <200910211752.n9LHqYXr032123@zion.cs.uiuc.edu>

Author: bwilson
Date: Wed Oct 21 12:52:34 2009
New Revision: 84766

URL: http://llvm.org/viewvc/llvm-project?rev=84766&view=rev
Log:
Revert 84732.  It was the wrong fix.

Modified:
    llvm/trunk/lib/Target/ARM/ARMInstrNEON.td

Modified: llvm/trunk/lib/Target/ARM/ARMInstrNEON.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrNEON.td?rev=84766&r1=84765&r2=84766&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrNEON.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrNEON.td Wed Oct 21 12:52:34 2009
@@ -267,7 +267,7 @@
 
 //   VLD2LN   : Vector Load (single 2-element structure to one lane)
 class VLD2LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b10,op11_8,0b0001, (outs DPR:$dst1, DPR:$dst2),
+  : NLdSt<1,0b10,op11_8,0b0000, (outs DPR:$dst1, DPR:$dst2),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
           IIC_VLD2,
           !strconcat(OpcodeStr, "\t\\{$dst1[$lane],$dst2[$lane]\\}, $addr"),
@@ -287,7 +287,7 @@
 
 //   VLD3LN   : Vector Load (single 3-element structure to one lane)
 class VLD3LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b10,op11_8,0b0010, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
+  : NLdSt<1,0b10,op11_8,0b0000, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
           nohash_imm:$lane), IIC_VLD3,
           !strconcat(OpcodeStr,
@@ -308,7 +308,7 @@
 
 //   VLD4LN   : Vector Load (single 4-element structure to one lane)
 class VLD4LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b10,op11_8,0b0011,
+  : NLdSt<1,0b10,op11_8,0b0000,
           (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
           nohash_imm:$lane), IIC_VLD4,
@@ -450,7 +450,7 @@
 
 //   VST2LN   : Vector Store (single 2-element structure from one lane)
 class VST2LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b00,op11_8,0b0001, (outs),
+  : NLdSt<1,0b00,op11_8,0b0000, (outs),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
           IIC_VST,
           !strconcat(OpcodeStr, "\t\\{$src1[$lane],$src2[$lane]\\}, $addr"),
@@ -470,7 +470,7 @@
 
 //   VST3LN   : Vector Store (single 3-element structure from one lane)
 class VST3LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b00,op11_8,0b0010, (outs),
+  : NLdSt<1,0b00,op11_8,0b0000, (outs),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
            nohash_imm:$lane), IIC_VST,
           !strconcat(OpcodeStr,
@@ -490,7 +490,7 @@
 
 //   VST4LN   : Vector Store (single 4-element structure from one lane)
 class VST4LN<bits<4> op11_8, string OpcodeStr>
-  : NLdSt<1,0b00,op11_8,0b0011, (outs),
+  : NLdSt<1,0b00,op11_8,0b0000, (outs),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
            nohash_imm:$lane), IIC_VST,
           !strconcat(OpcodeStr,




From bob.wilson at apple.com  Wed Oct 21 12:54:01 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Wed, 21 Oct 2009 17:54:01 -0000
Subject: [llvm-commits] [llvm] r84767 -
	/llvm/trunk/lib/Target/ARM/ARMInstrNEON.td
Message-ID: <200910211754.n9LHs1Tg032188@zion.cs.uiuc.edu>

Author: bwilson
Date: Wed Oct 21 12:54:01 2009
New Revision: 84767

URL: http://llvm.org/viewvc/llvm-project?rev=84767&view=rev
Log:
Fix NEON VST2LN instruction encoding.
Patch by Johnny Chen.

Modified:
    llvm/trunk/lib/Target/ARM/ARMInstrNEON.td

Modified: llvm/trunk/lib/Target/ARM/ARMInstrNEON.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrNEON.td?rev=84767&r1=84766&r2=84767&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrNEON.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrNEON.td Wed Oct 21 12:54:01 2009
@@ -456,17 +456,17 @@
           !strconcat(OpcodeStr, "\t\\{$src1[$lane],$src2[$lane]\\}, $addr"),
           "", []>;
 
-def VST2LNd8  : VST2LN<0b0000, "vst2.8">;
-def VST2LNd16 : VST2LN<0b0100, "vst2.16">;
-def VST2LNd32 : VST2LN<0b1000, "vst2.32">;
+def VST2LNd8  : VST2LN<0b0001, "vst2.8">;
+def VST2LNd16 : VST2LN<0b0101, "vst2.16">;
+def VST2LNd32 : VST2LN<0b1001, "vst2.32">;
 
 // vst2 to double-spaced even registers.
-def VST2LNq16a: VST2LN<0b0100, "vst2.16">;
-def VST2LNq32a: VST2LN<0b1000, "vst2.32">;
+def VST2LNq16a: VST2LN<0b0101, "vst2.16">;
+def VST2LNq32a: VST2LN<0b1001, "vst2.32">;
 
 // vst2 to double-spaced odd registers.
-def VST2LNq16b: VST2LN<0b0100, "vst2.16">;
-def VST2LNq32b: VST2LN<0b1000, "vst2.32">;
+def VST2LNq16b: VST2LN<0b0101, "vst2.16">;
+def VST2LNq32b: VST2LN<0b1001, "vst2.32">;
 
 //   VST3LN   : Vector Store (single 3-element structure from one lane)
 class VST3LN<bits<4> op11_8, string OpcodeStr>




From edwintorok at gmail.com  Wed Oct 21 13:19:12 2009
From: edwintorok at gmail.com (=?ISO-8859-1?Q?T=F6r=F6k_Edwin?=)
Date: Wed, 21 Oct 2009 21:19:12 +0300
Subject: [llvm-commits] [PATCH] Introduce -dbg-quality test
In-Reply-To: <352a1fb20910211023n4351ff5ate009c4297ea454df@mail.gmail.com>
References: <4ACDAF90.1040202@gmail.com>	
	<B59C3DBC-4810-425A-917A-4E857AEC4CEA@apple.com>	
	<4AD22FBF.3060607@gmail.com>	
	<352a1fb20910131016g41abc9e0v3ff54f001324075c@mail.gmail.com>	
	<4ADF4005.20201@gmail.com>
	<352a1fb20910211023n4351ff5ate009c4297ea454df@mail.gmail.com>
Message-ID: <4ADF50A0.5080903@gmail.com>

On 2009-10-21 20:23, Devang Patel wrote:
> Hi
>
> 2009/10/21 T?r?k Edwin <edwintorok at gmail.com>:
>   
>> Hi Devang,
>>
>> I've written the -dbg-quality pass (attached), and the
>> TEST.dbgquality.Makefile that tests the optimized (-std-compile-opts)
>> bitcode file.
>>     
>
> This is great. Could you explain how you qualify debug info to get the
> following score ?
>   

It is based on a ratio, so it is not affected that much by optimizers
introducing/deleting instructions.
First a score is calculated like (all is all instructions except alloca
and dbginfo intrinsics):
100000.0*debug_info_in_same_BB/all +
10000.0*debug_info_in_previous_BB/all +
100.0*debug_info_on_instruction/all

This score is then normalized so that the maximum score for a module is
always 10000, so that the score doesn't depend on the size of the module.

I think I should also make a final adjustment, to make sure that the
score is 10000 only if original score == maxscore,
currently due to rounding you can get a 10000 score if score is very
close to maxscore (but not equal) too.

The score numbers are arbitrary, I chose 10000, because it would be
equivalent to 100.00%, but since we can't represent fractions in
statistics I multiplied by 100.

I handle both debug metadata and dbg intrinsics. If the intrinsics are
going away then I can remove that part.

>   
>> Should I also add a -std-link-opts tests?
>>     
>
> Yes.
>   

Ok, I'll do that tomorrow.

I thought to also update TEST.dbgopt and TEST.ipodbgopt to show a
summary, but those seem obsolete now that the dbginfo is attached to the
instructions themselves,
or are they still useful for dbg.variable intrinsics? If those are
switched to metadata by llvm 2.7, then probably not.

>   
>> I think some
>> passes (such as -ipsccp) are only run there?
>> I've attached the results, the maximum score is 10000 (meaning all
>> non-alloca/non-dbgintrinsic instructions have debug info).
>> We have 79 tests at 10000 quality, the rest are between 5331 - 9999.
>>
>> I've also found a bug: static constructors/destructors have out-of-file
>> line numbers often (PR5270). I've added a hack in -dbg-quality for now
>> that ignores those functions (denoted by a FIXME).
>>     
>
> I'll take a look at this PR.
>   

Thanks, there is also PR5246 which prevents debug info from being used:
llc crashes when present.
Unfortunately bugpoint is not able to further reduce the testcase, it
may somehow depend on the order of functions in the bitcode.

>   
>> OK to commit, or should I wait till PR5270 is fixed so I can drop the
>> ugly hack?
>>     
>
> I do not think Analysis is a good place for this and DebugInfo as
> well. 

Sure I can move it if needed, but can we have passes in Support/?

> I think, DebugInfo is in Analysis, because we do not want it in
> VMCore. IMO, debug info related code should be in lib/Support. Chris,
> what do you think ?
>   



Best regards,
--Edwin


From vhernandez at apple.com  Wed Oct 21 14:11:41 2009
From: vhernandez at apple.com (Victor Hernandez)
Date: Wed, 21 Oct 2009 19:11:41 -0000
Subject: [llvm-commits] [llvm] r84772 - in /llvm/trunk:
 lib/AsmParser/LLParser.cpp lib/Bitcode/Reader/BitcodeReader.cpp
 lib/Transforms/Scalar/InstructionCombining.cpp
 lib/Transforms/Scalar/Reassociate.cpp
 test/Transforms/InstCombine/malloc2.ll
Message-ID: <200910211911.n9LJBgn6003044@zion.cs.uiuc.edu>

Author: hernande
Date: Wed Oct 21 14:11:40 2009
New Revision: 84772

URL: http://llvm.org/viewvc/llvm-project?rev=84772&view=rev
Log:
Make changes to rev 84292 as requested by Chris Lattner.

Most changes are cleanup, but there is 1 correctness fix:
I fixed InstCombine so that the icmp is removed only if the malloc call is removed (which requires explicit removal because the Worklist won't DCE any calls since they can have side-effects).

Modified:
    llvm/trunk/lib/AsmParser/LLParser.cpp
    llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp
    llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
    llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp
    llvm/trunk/test/Transforms/InstCombine/malloc2.ll

Modified: llvm/trunk/lib/AsmParser/LLParser.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/AsmParser/LLParser.cpp?rev=84772&r1=84771&r2=84772&view=diff

==============================================================================
--- llvm/trunk/lib/AsmParser/LLParser.cpp (original)
+++ llvm/trunk/lib/AsmParser/LLParser.cpp Wed Oct 21 14:11:40 2009
@@ -69,7 +69,7 @@
 /// ValidateEndOfModule - Do final validity and sanity checks at the end of the
 /// module.
 bool LLParser::ValidateEndOfModule() {
-  // Update auto-upgraded malloc calls from "autoupgrade_malloc" to "malloc".
+  // Update auto-upgraded malloc calls to "malloc".
   // FIXME: Remove in LLVM 3.0.
   if (MallocF) {
     MallocF->setName("malloc");
@@ -77,15 +77,10 @@
     // declaration of "malloc".  In that case, iterate over all calls to MallocF
     // and get them to call the declared "malloc" instead.
     if (MallocF->getName() != "malloc") {
-      Function* realMallocF = M->getFunction("malloc");
-      for (User::use_iterator UI = MallocF->use_begin(), UE= MallocF->use_end();
-           UI != UE; ) {
-        User* user = *UI;
-        UI++;
-        if (CallInst *Call = dyn_cast<CallInst>(user))
-          Call->setCalledFunction(realMallocF);
-      }
-      if (!realMallocF->doesNotAlias(0)) realMallocF->setDoesNotAlias(0);
+      Constant* RealMallocF = M->getFunction("malloc");
+      if (RealMallocF->getType() != MallocF->getType())
+        RealMallocF = ConstantExpr::getBitCast(RealMallocF, MallocF->getType());
+      MallocF->replaceAllUsesWith(RealMallocF);
       MallocF->eraseFromParent();
       MallocF = NULL;
     }
@@ -3482,21 +3477,21 @@
   if (Size && Size->getType() != Type::getInt32Ty(Context))
     return Error(SizeLoc, "element count must be i32");
 
-  if (isAlloca)
+  if (isAlloca) {
     Inst = new AllocaInst(Ty, Size, Alignment);
-  else {
-    // Autoupgrade old malloc instruction to malloc call.
-    const Type* IntPtrTy = Type::getInt32Ty(Context);
-    const Type* Int8PtrTy = PointerType::getUnqual(Type::getInt8Ty(Context));
-    if (!MallocF)
-      // Prototype malloc as "void *autoupgrade_malloc(int32)".
-      MallocF = cast<Function>(M->getOrInsertFunction("autoupgrade_malloc",
-                               Int8PtrTy, IntPtrTy, NULL));
-      // "autoupgrade_malloc" updated to "malloc" in ValidateEndOfModule().
-
-    Inst = cast<Instruction>(CallInst::CreateMalloc(BB, IntPtrTy, Ty,
-                                                    Size, MallocF));
+    return false;
   }
+
+  // Autoupgrade old malloc instruction to malloc call.
+  // FIXME: Remove in LLVM 3.0.
+  const Type *IntPtrTy = Type::getInt32Ty(Context);
+  const Type *Int8PtrTy = Type::getInt8PtrTy(Context);
+  if (!MallocF)
+    // Prototype malloc as "void *(int32)".
+    // This function is renamed as "malloc" in ValidateEndOfModule().
+    MallocF = cast<Function>(M->getOrInsertFunction(NULL, Int8PtrTy, 
+                                                    IntPtrTy, NULL));
+  Inst = CallInst::CreateMalloc(BB, IntPtrTy, Ty, Size, MallocF);
   return false;
 }
 

Modified: llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp?rev=84772&r1=84771&r2=84772&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp (original)
+++ llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp Wed Oct 21 14:11:40 2009
@@ -2045,6 +2045,7 @@
 
     case bitc::FUNC_CODE_INST_MALLOC: { // MALLOC: [instty, op, align]
       // Autoupgrade malloc instruction to malloc call.
+      // FIXME: Remove in LLVM 3.0.
       if (Record.size() < 3)
         return Error("Invalid MALLOC record");
       const PointerType *Ty =
@@ -2052,13 +2053,9 @@
       Value *Size = getFnValueByID(Record[1], Type::getInt32Ty(Context));
       if (!Ty || !Size) return Error("Invalid MALLOC record");
       if (!CurBB) return Error("Invalid malloc instruction with no BB");
-      const Type* Int32Ty = IntegerType::getInt32Ty(CurBB->getContext());
-      if (Size->getType() != Int32Ty)
-        Size = CastInst::CreateIntegerCast(Size, Int32Ty, false /*ZExt*/,
-                                           "", CurBB);
-      Value* Malloc = CallInst::CreateMalloc(CurBB, Int32Ty,
-                                             Ty->getElementType(), Size, NULL);
-      I = cast<Instruction>(Malloc);
+      const Type *Int32Ty = IntegerType::getInt32Ty(CurBB->getContext());
+      I = CallInst::CreateMalloc(CurBB, Int32Ty, Ty->getElementType(),
+                                 Size, NULL);
       InstructionList.push_back(I);
       break;
     }

Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=84772&r1=84771&r2=84772&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Wed Oct 21 14:11:40 2009
@@ -6359,10 +6359,28 @@
         // can assume it is successful and remove the malloc.
         if (isMalloc(LHSI) && LHSI->hasOneUse() &&
             isa<ConstantPointerNull>(RHSC)) {
-          Worklist.Add(LHSI);
-          return ReplaceInstUsesWith(I,
+          // Need to explicitly erase malloc call here, instead of adding it to
+          // Worklist, because it won't get DCE'd from the Worklist since
+          // isInstructionTriviallyDead() returns false for function calls.
+          // It is OK to replace LHSI/MallocCall with Undef because the 
+          // instruction that uses it will be erased via Worklist.
+          if (extractMallocCall(LHSI)) {
+            LHSI->replaceAllUsesWith(UndefValue::get(LHSI->getType()));
+            EraseInstFromFunction(*LHSI);
+            return ReplaceInstUsesWith(I,
                                      ConstantInt::get(Type::getInt1Ty(*Context),
                                                       !I.isTrueWhenEqual()));
+          }
+          if (CallInst* MallocCall = extractMallocCallFromBitCast(LHSI))
+            if (MallocCall->hasOneUse()) {
+              MallocCall->replaceAllUsesWith(
+                                        UndefValue::get(MallocCall->getType()));
+              EraseInstFromFunction(*MallocCall);
+              Worklist.Add(LHSI); // The malloc's bitcast use.
+              return ReplaceInstUsesWith(I,
+                                     ConstantInt::get(Type::getInt1Ty(*Context),
+                                                      !I.isTrueWhenEqual()));
+            }
         }
         break;
       }

Modified: llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp?rev=84772&r1=84771&r2=84772&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp Wed Oct 21 14:11:40 2009
@@ -29,7 +29,6 @@
 #include "llvm/IntrinsicInst.h"
 #include "llvm/LLVMContext.h"
 #include "llvm/Pass.h"
-#include "llvm/Analysis/MallocHelper.h"
 #include "llvm/Assembly/Writer.h"
 #include "llvm/Support/CFG.h"
 #include "llvm/Support/Debug.h"
@@ -122,7 +121,6 @@
   if (I->getOpcode() == Instruction::PHI ||
       I->getOpcode() == Instruction::Alloca ||
       I->getOpcode() == Instruction::Load ||
-      isMalloc(I) ||
       I->getOpcode() == Instruction::Invoke ||
       (I->getOpcode() == Instruction::Call &&
        !isa<DbgInfoIntrinsic>(I)) ||

Modified: llvm/trunk/test/Transforms/InstCombine/malloc2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/malloc2.ll?rev=84772&r1=84771&r2=84772&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/malloc2.ll (original)
+++ llvm/trunk/test/Transforms/InstCombine/malloc2.ll Wed Oct 21 14:11:40 2009
@@ -1,18 +1,22 @@
-; RUN: opt < %s -instcombine -S | grep {ret i32 0}
+; RUN: opt < %s -instcombine -S | FileCheck %s
 ; PR1313
 
 define i32 @test1(i32 %argc, i8* %argv, i8* %envp) {
         %tmp15.i.i.i23 = malloc [2564 x i32]            ; <[2564 x i32]*> [#uses=1]
+; CHECK-NOT: call i8* @malloc
         %c = icmp eq [2564 x i32]* %tmp15.i.i.i23, null              ; <i1>:0 [#uses=1]
         %retval = zext i1 %c to i32             ; <i32> [#uses=1]
         ret i32 %retval
+; CHECK: ret i32 0
 }
 
 define i32 @test2(i32 %argc, i8* %argv, i8* %envp) {
         %tmp15.i.i.i23 = malloc [2564 x i32]            ; <[2564 x i32]*> [#uses=1]
+; CHECK-NOT: call i8* @malloc
         %X = bitcast [2564 x i32]* %tmp15.i.i.i23 to i32*
         %c = icmp ne i32* %X, null
         %retval = zext i1 %c to i32             ; <i32> [#uses=1]
         ret i32 %retval
+; CHECK: ret i32 1
 }
 




From vhernandez at apple.com  Wed Oct 21 14:12:03 2009
From: vhernandez at apple.com (Victor Hernandez)
Date: Wed, 21 Oct 2009 12:12:03 -0700
Subject: [llvm-commits] [llvm] r84292 - in /llvm/trunk: examples/BrainF/
	include/llvm/ lib/AsmParser/ lib/Bitcode/Reader/
	lib/Transforms/IPO/ lib/Transforms/Scalar/ lib/VMCore/
	test/Transforms/GlobalOpt/ test/Transforms/IndMemRem/
	test/Transforms/InstCombine/
In-Reply-To: <797711D6-5B12-4DA3-BFF4-97D38BDC6E58@apple.com>
References: <200910170000.n9H00RWN030901@zion.cs.uiuc.edu>
	<797711D6-5B12-4DA3-BFF4-97D38BDC6E58@apple.com>
Message-ID: <5F7FC9EE-DD30-438F-833D-9F9338021A1A@apple.com>

Chris,

I fixed all your suggestions in rev 84772.

On Oct 17, 2009, at 10:24 PM, Chris Lattner wrote:

> On Oct 16, 2009, at 5:00 PM, Victor Hernandez wrote:
>> URL: http://llvm.org/viewvc/llvm-project?rev=84292&view=rev
>> Log:
>> Autoupgrade malloc insts to malloc calls.
>> Update testcases that rely on malloc insts being present.
>>
>> Also prematurely remove MallocInst handling from IndMemRemoval and  
>> RaiseAllocations to help pass tests in this incremental step.
>
> Thanks for working on this Victor.  I committed a few patches to do  
> random cleanups, here are some more things for you to look at:
>
>> +++ llvm/trunk/lib/AsmParser/LLParser.cpp Fri Oct 16 19:00:19 2009
>> @@ -69,6 +69,27 @@
>> /// ValidateEndOfModule - Do final validity and sanity checks at  
>> the end of the
>> /// module.
>> bool LLParser::ValidateEndOfModule() {
>> +  // Update auto-upgraded malloc calls from "autoupgrade_malloc"  
>> to "malloc".
>> +  if (MallocF) {
>> +    MallocF->setName("malloc");
>> +    // If setName() does not set the name to "malloc", then there  
>> is already a
>> +    // declaration of "malloc".  In that case, iterate over all  
>> calls to MallocF
>> +    // and get them to call the declared "malloc" instead.
>> +    if (MallocF->getName() != "malloc") {
>> +      Function* realMallocF = M->getFunction("malloc");
>> +      for (User::use_iterator UI = MallocF->use_begin(), UE=  
>> MallocF->use_end();
>> +           UI != UE; ) {
>> +        User* user = *UI;
>> +        UI++;
>> +        if (CallInst *Call = dyn_cast<CallInst>(user))
>> +          Call->setCalledFunction(realMallocF);
>> +      }
>
> This should use V->replaceAllUsesWith() instead of manually walking  
> the use list.  Doing this will handle non-call users as well as call  
> users.  What happens if the prototypes don't agree for these two  
> functions?  I think you'll get an abort.  Please fix by using a  
> constantexpr bitcast to the expected type.

Fixed:

   // Update auto-upgraded malloc calls to "malloc".
   // FIXME: Remove in LLVM 3.0.
   if (MallocF) {
     MallocF->setName("malloc");
     // If setName() does not set the name to "malloc", then there is  
already a
     // declaration of "malloc".  In that case, iterate over all calls  
to MallocF
     // and get them to call the declared "malloc" instead.
     if (MallocF->getName() != "malloc") {
       Constant* RealMallocF = M->getFunction("malloc");
       if (RealMallocF->getType() != MallocF->getType())
         RealMallocF = ConstantExpr::getBitCast(RealMallocF, MallocF- 
 >getType());
       MallocF->replaceAllUsesWith(RealMallocF);
       MallocF->eraseFromParent();
       MallocF = NULL;
     }
   }


>
>> +      if (!realMallocF->doesNotAlias(0)) realMallocF- 
>> >setDoesNotAlias(0);
>
> I don't think you have to care about this.  We don't care that old  
> code be performant, just that it work.

See above.
>
>> @@ -3466,10 +3487,21 @@
>>  if (Size && Size->getType() != Type::getInt32Ty(Context))
>>    return Error(SizeLoc, "element count must be i32");
>>
>> -  if (Opc == Instruction::Malloc)
>> -    Inst = new MallocInst(Ty, Size, Alignment);
>> -  else
>> +  if (isAlloca)
>>    Inst = new AllocaInst(Ty, Size, Alignment);
>
> Please use an early exit here to unnest the 'else'.

Fixed.

>
>> +  else {
>> +    // Autoupgrade old malloc instruction to malloc call.
>
> Please add a FIXME to remove this in LLVM 3.0 like I added to other  
> places.

Fixed.

>
>> +    const Type* IntPtrTy = Type::getInt32Ty(Context);
>> +    const Type* Int8PtrTy = PointerType::getUnqual(Type::getInt8Ty 
>> (Context));
>
> Please use 'const Type *IntPtrTy', watch the * placement.  There is  
> now a helper function that you can use instead of  
> 'PointerType::getUnqual(Type::getInt8Ty(Context));', please use it.

Fixed:
   const Type *IntPtrTy = Type::getInt32Ty(Context);
   const Type *Int8PtrTy = Type::getInt8PtrTy(Context);

>
>> +    if (!MallocF)
>> +      // Prototype malloc as "void *autoupgrade_malloc(int32)".
>> +      MallocF = cast<Function>(M->getOrInsertFunction 
>> ("autoupgrade_malloc",
>> +                               Int8PtrTy, IntPtrTy, NULL));
>
> Why do you bother naming autoupgrade_malloc?  This is in the user's  
> namespace, so it can conflict with a function named  
> autoupgrade_malloc in the user's code.  Please just leave the name  
> empty, making it impossible for it to conflict.

Fixed.

>
>> +      // "autoupgrade_malloc" updated to "malloc" in  
>> ValidateEndOfModule().
>> +
>> +    Inst = cast<Instruction>(CallInst::CreateMalloc(BB, IntPtrTy,  
>> Ty,
>> +                                                    Size, MallocF));

Fixed.

>
> CreateMalloc doesn't return something convertible to Instruction?   
> Why do you need the cast<>?

Fixed.

>
>> +++ llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp Fri Oct 16  
>> 19:00:19 2009
>> @@ -2044,14 +2044,21 @@
>>    }
>>
>>    case bitc::FUNC_CODE_INST_MALLOC: { // MALLOC: [instty, op, align]
>> +      // Autoupgrade malloc instruction to malloc call.
>
> FIXME remove in 3.0.

Fixed.

>
>>      if (Record.size() < 3)
>>        return Error("Invalid MALLOC record");
>>      const PointerType *Ty =
>>        dyn_cast_or_null<PointerType>(getTypeByID(Record[0]));
>>      Value *Size = getFnValueByID(Record[1], Type::getInt32Ty 
>> (Context));
>> -      unsigned Align = Record[2];
>>      if (!Ty || !Size) return Error("Invalid MALLOC record");
>> -      I = new MallocInst(Ty->getElementType(), Size, (1 << Align)  
>> >> 1);
>> +      if (!CurBB) return Error("Invalid malloc instruction with no  
>> BB");
>> +      const Type* Int32Ty = IntegerType::getInt32Ty(CurBB- 
>> >getContext());
>
> * placement.

Fixed.

>
>> +      if (Size->getType() != Int32Ty)
>> +        Size = CastInst::CreateIntegerCast(Size, Int32Ty, false / 
>> *ZExt*/,
>> +                                           "", CurBB);
>
> How is this possible?  Size has to be int32 per the code above.

It is not possible, copy-and-past error from other calls to  
CallInst::CreateMalloc().  Deleted the IntegerCast.
>
>
>
>> +      Value* Malloc = CallInst::CreateMalloc(CurBB, Int32Ty,
>> +                                             Ty->getElementType(),  
>> Size, NULL);
>> +      I = cast<Instruction>(Malloc);
>
> again, why the cast?

Fixed.

>
>> +++ llvm/trunk/lib/Transforms/IPO/IndMemRemoval.cpp Fri Oct 16  
>> 19:00:19 2009
>> @@ -24,6 +24,7 @@
>
> I removed this pass, it was specifically their to support malloc/ 
> free instructions, it is not needed anymore (and not used in any  
> case).

Thanks.

>
>> +++ llvm/trunk/lib/Transforms/IPO/RaiseAllocations.cpp Fri Oct 16  
>> 19:00:19 2009
>> @@ -1,4 +1,4 @@
>> -//===- RaiseAllocations.cpp - Convert @malloc & @free calls to  
>> insts ------===//
>> +//===- RaiseAllocations.cpp - Convert @free calls to insts ------ 
>> ===//
>
> Hopefully your future changes to remove FreeInst will completely zap  
> this pass.

Yup, that is next.

>
>> +++ llvm/trunk/lib/Transforms/Scalar/Reassociate.cpp Fri Oct 16  
>> 19:00:19 2009
>> @@ -29,6 +29,7 @@
>> #include "llvm/IntrinsicInst.h"
>> #include "llvm/LLVMContext.h"
>> #include "llvm/Pass.h"
>> +#include "llvm/Analysis/MallocHelper.h"
>> #include "llvm/Assembly/Writer.h"
>> #include "llvm/Support/CFG.h"
>> #include "llvm/Support/Debug.h"
>> @@ -121,7 +122,7 @@
>>  if (I->getOpcode() == Instruction::PHI ||
>>      I->getOpcode() == Instruction::Alloca ||
>>      I->getOpcode() == Instruction::Load ||
>> -      I->getOpcode() == Instruction::Malloc ||
>> +      I->getOpcode() == Instruction::Malloc || isMalloc(I) ||
>>      I->getOpcode() == Instruction::Invoke ||
>>      (I->getOpcode() == Instruction::Call &&
>>       !isa<DbgInfoIntrinsic>(I)) ||
>
> A check for isMalloc isn't needed here at all, because call is  
> handled.  Please remove all the malloc logic and the #include.

Fixed.

>
>> +++ llvm/trunk/test/Transforms/InstCombine/cast-malloc.ll Fri Oct  
>> 16 19:00:19 2009
>> @@ -1,6 +1,6 @@
>
> I removed this, it is now pointless.  You convert it to test that  
> instcombine isn't doing anything.

OK.

>
>>
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> = 
>> =====================================================================
>> --- llvm/trunk/test/Transforms/InstCombine/malloc2.ll (original)
>> +++ llvm/trunk/test/Transforms/InstCombine/malloc2.ll Fri Oct 16  
>> 19:00:19 2009
>> @@ -1,5 +1,4 @@
>> ; RUN: opt < %s -instcombine -S | grep {ret i32 0}
>> -; RUN: opt < %s -instcombine -S | not grep malloc
>> ; PR1313
>>
>> define i32 @test1(i32 %argc, i8* %argv, i8* %envp) {
>
>
> This test is being miscompiled now.  This is a serious bug that one  
> of your patches has introduced.  We're now compiling (e.g.):
>
> define i32 @test1(i32 %argc, i8* %argv, i8* %envp) {
>        %tmp15.i.i.i23 = malloc [2564 x i32]            ; <[2564 x  
> i32]*> [#uses=1]
>        %c = icmp eq [2564 x i32]* %tmp15.i.i.i23,  
> null              ; <i1>:0 [#uses=1]
>        %retval = zext i1 %c to i32             ; <i32> [#uses=1]
>        ret i32 %retval
> }
>
> into:
>
> define i32 @test1(i32 %argc, i8* %argv, i8* %envp) {
>  %malloccall = tail call i8* @malloc(i32 ptrtoint ([2564 x i32]*  
> getelementptr ([2564 x i32]* null, i32 1) to i32)) ; <i8*> [#uses=0]
>  ret i32 0
> }
>
> The old code was correct because it would delete the call to malloc,  
> so it is safe to assume that the malloc (which never got run) never  
> ran out of space.  The transformation that does this should just be  
> removed.

I fixed InstCombine so that the icmp is removed only if the malloc  
call is removed (which requires explicit removal because the Worklist  
won't DCE any calls since they can have side-effects).  I also updated  
the testcase to use FileCheck.

Victor



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091021/2b980146/attachment.html 

From asl at math.spbu.ru  Wed Oct 21 14:16:49 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 19:16:49 -0000
Subject: [llvm-commits] [llvm] r84773 - in /llvm/trunk/lib/Target/MSP430:
 AsmPrinter/MSP430AsmPrinter.cpp AsmPrinter/MSP430InstPrinter.cpp MSP430.h
 MSP430ISelLowering.cpp MSP430InstrInfo.h
Message-ID: <200910211916.n9LJGnag003236@zion.cs.uiuc.edu>

Author: asl
Date: Wed Oct 21 14:16:49 2009
New Revision: 84773

URL: http://llvm.org/viewvc/llvm-project?rev=84773&view=rev
Log:
Cosmetic changes, no functionality changes

Modified:
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
    llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
    llvm/trunk/lib/Target/MSP430/MSP430.h
    llvm/trunk/lib/Target/MSP430/MSP430ISelLowering.cpp
    llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp?rev=84773&r1=84772&r2=84773&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430AsmPrinter.cpp Wed Oct 21 14:16:49 2009
@@ -250,22 +250,22 @@
   default:
    llvm_unreachable("Unsupported CC code");
    break;
-  case MSP430::COND_E:
+  case MSP430CC::COND_E:
    O << "eq";
    break;
-  case MSP430::COND_NE:
+  case MSP430CC::COND_NE:
    O << "ne";
    break;
-  case MSP430::COND_HS:
+  case MSP430CC::COND_HS:
    O << "hs";
    break;
-  case MSP430::COND_LO:
+  case MSP430CC::COND_LO:
    O << "lo";
    break;
-  case MSP430::COND_GE:
+  case MSP430CC::COND_GE:
    O << "ge";
    break;
-  case MSP430::COND_L:
+  case MSP430CC::COND_L:
    O << 'l';
    break;
   }

Modified: llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp?rev=84773&r1=84772&r2=84773&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/AsmPrinter/MSP430InstPrinter.cpp Wed Oct 21 14:16:49 2009
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "asm-printer"
+#include "MSP430.h"
 #include "MSP430InstrInfo.h"
 #include "MSP430InstPrinter.h"
 #include "llvm/MC/MCInst.h"
@@ -19,7 +20,6 @@
 #include "llvm/MC/MCExpr.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
-#include "MSP430GenInstrNames.inc"
 using namespace llvm;
 
 
@@ -94,22 +94,22 @@
   default:
    llvm_unreachable("Unsupported CC code");
    break;
-  case MSP430::COND_E:
+  case MSP430CC::COND_E:
    O << "eq";
    break;
-  case MSP430::COND_NE:
+  case MSP430CC::COND_NE:
    O << "ne";
    break;
-  case MSP430::COND_HS:
+  case MSP430CC::COND_HS:
    O << "hs";
    break;
-  case MSP430::COND_LO:
+  case MSP430CC::COND_LO:
    O << "lo";
    break;
-  case MSP430::COND_GE:
+  case MSP430CC::COND_GE:
    O << "ge";
    break;
-  case MSP430::COND_L:
+  case MSP430CC::COND_L:
    O << 'l';
    break;
   }

Modified: llvm/trunk/lib/Target/MSP430/MSP430.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430.h?rev=84773&r1=84772&r2=84773&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430.h (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430.h Wed Oct 21 14:16:49 2009
@@ -17,6 +17,20 @@
 
 #include "llvm/Target/TargetMachine.h"
 
+namespace MSP430CC {
+  // MSP430 specific condition code.
+  enum CondCodes {
+    COND_E  = 0,  // aka COND_Z
+    COND_NE = 1,  // aka COND_NZ
+    COND_HS = 2,  // aka COND_C
+    COND_LO = 3,  // aka COND_NC
+    COND_GE = 4,
+    COND_L  = 5,
+
+    COND_INVALID = -1
+  };
+}
+
 namespace llvm {
   class MSP430TargetMachine;
   class FunctionPass;

Modified: llvm/trunk/lib/Target/MSP430/MSP430ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430ISelLowering.cpp?rev=84773&r1=84772&r2=84773&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430ISelLowering.cpp Wed Oct 21 14:16:49 2009
@@ -567,44 +567,45 @@
   return DAG.getNode(MSP430ISD::Wrapper, dl, getPointerTy(), Result);;
 }
 
-static SDValue EmitCMP(SDValue &LHS, SDValue &RHS, unsigned &TargetCC,
+static SDValue EmitCMP(SDValue &LHS, SDValue &RHS, SDValue &TargetCC,
                        ISD::CondCode CC,
                        DebugLoc dl, SelectionDAG &DAG) {
   // FIXME: Handle bittests someday
   assert(!LHS.getValueType().isFloatingPoint() && "We don't handle FP yet");
 
   // FIXME: Handle jump negative someday
-  TargetCC = MSP430::COND_INVALID;
+  MSP430CC::CondCodes TCC = MSP430CC::COND_INVALID;
   switch (CC) {
   default: llvm_unreachable("Invalid integer condition!");
   case ISD::SETEQ:
-    TargetCC = MSP430::COND_E;  // aka COND_Z
+    TCC = MSP430CC::COND_E;     // aka COND_Z
     break;
   case ISD::SETNE:
-    TargetCC = MSP430::COND_NE; // aka COND_NZ
+    TCC = MSP430CC::COND_NE;    // aka COND_NZ
     break;
   case ISD::SETULE:
     std::swap(LHS, RHS);        // FALLTHROUGH
   case ISD::SETUGE:
-    TargetCC = MSP430::COND_HS; // aka COND_C
+    TCC = MSP430CC::COND_HS;    // aka COND_C
     break;
   case ISD::SETUGT:
     std::swap(LHS, RHS);        // FALLTHROUGH
   case ISD::SETULT:
-    TargetCC = MSP430::COND_LO; // aka COND_NC
+    TCC = MSP430CC::COND_LO;    // aka COND_NC
     break;
   case ISD::SETLE:
     std::swap(LHS, RHS);        // FALLTHROUGH
   case ISD::SETGE:
-    TargetCC = MSP430::COND_GE;
+    TCC = MSP430CC::COND_GE;
     break;
   case ISD::SETGT:
     std::swap(LHS, RHS);        // FALLTHROUGH
   case ISD::SETLT:
-    TargetCC = MSP430::COND_L;
+    TCC = MSP430CC::COND_L;
     break;
   }
 
+  TargetCC = DAG.getConstant(TCC, MVT::i8);
   return DAG.getNode(MSP430ISD::CMP, dl, MVT::Flag, LHS, RHS);
 }
 
@@ -617,13 +618,11 @@
   SDValue Dest  = Op.getOperand(4);
   DebugLoc dl   = Op.getDebugLoc();
 
-  unsigned TargetCC = MSP430::COND_INVALID;
+  SDValue TargetCC;
   SDValue Flag = EmitCMP(LHS, RHS, TargetCC, CC, dl, DAG);
 
   return DAG.getNode(MSP430ISD::BR_CC, dl, Op.getValueType(),
-                     Chain,
-                     Dest, DAG.getConstant(TargetCC, MVT::i8),
-                     Flag);
+                     Chain, Dest, TargetCC, Flag);
 }
 
 SDValue MSP430TargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) {
@@ -634,14 +633,14 @@
   ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
   DebugLoc dl    = Op.getDebugLoc();
 
-  unsigned TargetCC = MSP430::COND_INVALID;
+  SDValue TargetCC;
   SDValue Flag = EmitCMP(LHS, RHS, TargetCC, CC, dl, DAG);
 
   SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::Flag);
   SmallVector<SDValue, 4> Ops;
   Ops.push_back(TrueV);
   Ops.push_back(FalseV);
-  Ops.push_back(DAG.getConstant(TargetCC, MVT::i8));
+  Ops.push_back(TargetCC);
   Ops.push_back(Flag);
 
   return DAG.getNode(MSP430ISD::SELECT_CC, dl, VTs, &Ops[0], Ops.size());

Modified: llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h?rev=84773&r1=84772&r2=84773&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h Wed Oct 21 14:16:49 2009
@@ -21,20 +21,6 @@
 
 class MSP430TargetMachine;
 
-namespace MSP430 {
-  // MSP430 specific condition code.
-  enum CondCode {
-    COND_E  = 0,  // aka COND_Z
-    COND_NE = 1,  // aka COND_NZ
-    COND_HS = 2,  // aka COND_C
-    COND_LO = 3,  // aka COND_NC
-    COND_GE = 4,
-    COND_L  = 5,
-
-    COND_INVALID
-  };
-}
-
 class MSP430InstrInfo : public TargetInstrInfoImpl {
   const MSP430RegisterInfo RI;
   MSP430TargetMachine &TM;




From asl at math.spbu.ru  Wed Oct 21 14:17:18 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 19:17:18 -0000
Subject: [llvm-commits] [llvm] r84774 - in /llvm/trunk/lib/Target/MSP430:
 MSP430InstrInfo.cpp MSP430InstrInfo.h README.txt
Message-ID: <200910211917.n9LJHIsn003260@zion.cs.uiuc.edu>

Author: asl
Date: Wed Oct 21 14:17:18 2009
New Revision: 84774

URL: http://llvm.org/viewvc/llvm-project?rev=84774&view=rev
Log:
Implement branch folding

Modified:
    llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp
    llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h
    llvm/trunk/lib/Target/MSP430/README.txt

Modified: llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp?rev=84774&r1=84773&r2=84774&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp Wed Oct 21 14:17:18 2009
@@ -151,6 +151,164 @@
   return true;
 }
 
+unsigned MSP430InstrInfo::RemoveBranch(MachineBasicBlock &MBB) const {
+  MachineBasicBlock::iterator I = MBB.end();
+  unsigned Count = 0;
+
+  while (I != MBB.begin()) {
+    --I;
+    if (I->getOpcode() != MSP430::JMP &&
+        I->getOpcode() != MSP430::JCC)
+      break;
+    // Remove the branch.
+    I->eraseFromParent();
+    I = MBB.end();
+    ++Count;
+  }
+
+  return Count;
+}
+
+
+bool MSP430InstrInfo::
+ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const {
+  assert(Cond.size() == 1 && "Invalid Xbranch condition!");
+
+  MSP430CC::CondCodes CC = static_cast<MSP430CC::CondCodes>(Cond[0].getImm());
+
+  switch (CC) {
+  default:
+    assert(0 && "Invalid branch condition!");
+    break;
+  case MSP430CC::COND_E:
+    CC = MSP430CC::COND_NE;
+    break;
+  case MSP430CC::COND_NE:
+    CC = MSP430CC::COND_E;
+    break;
+  case MSP430CC::COND_L:
+    CC = MSP430CC::COND_GE;
+    break;
+  case MSP430CC::COND_GE:
+    CC = MSP430CC::COND_L;
+    break;
+  case MSP430CC::COND_HS:
+    CC = MSP430CC::COND_LO;
+    break;
+  case MSP430CC::COND_LO:
+    CC = MSP430CC::COND_HS;
+    break;
+  }
+
+  Cond[0].setImm(CC);
+  return false;
+}
+
+bool MSP430InstrInfo::BlockHasNoFallThrough(const MachineBasicBlock &MBB)const{
+  if (MBB.empty()) return false;
+
+  switch (MBB.back().getOpcode()) {
+  case MSP430::RET:   // Return.
+  case MSP430::JMP:   // Uncond branch.
+    return true;
+  default: return false;
+  }
+}
+
+bool MSP430InstrInfo::isUnpredicatedTerminator(const MachineInstr *MI) const {
+  const TargetInstrDesc &TID = MI->getDesc();
+  if (!TID.isTerminator()) return false;
+
+  // Conditional branch is a special case.
+  if (TID.isBranch() && !TID.isBarrier())
+    return true;
+  if (!TID.isPredicable())
+    return true;
+  return !isPredicated(MI);
+}
+
+bool MSP430InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,
+                                    MachineBasicBlock *&TBB,
+                                    MachineBasicBlock *&FBB,
+                                    SmallVectorImpl<MachineOperand> &Cond,
+                                    bool AllowModify) const {
+  // Start from the bottom of the block and work up, examining the
+  // terminator instructions.
+  MachineBasicBlock::iterator I = MBB.end();
+  while (I != MBB.begin()) {
+    --I;
+    // Working from the bottom, when we see a non-terminator
+    // instruction, we're done.
+    if (!isUnpredicatedTerminator(I))
+      break;
+
+    // A terminator that isn't a branch can't easily be handled
+    // by this analysis.
+    if (!I->getDesc().isBranch())
+      return true;
+
+    // Handle unconditional branches.
+    if (I->getOpcode() == MSP430::JMP) {
+      if (!AllowModify) {
+        TBB = I->getOperand(0).getMBB();
+        continue;
+      }
+
+      // If the block has any instructions after a JMP, delete them.
+      while (next(I) != MBB.end())
+        next(I)->eraseFromParent();
+      Cond.clear();
+      FBB = 0;
+
+      // Delete the JMP if it's equivalent to a fall-through.
+      if (MBB.isLayoutSuccessor(I->getOperand(0).getMBB())) {
+        TBB = 0;
+        I->eraseFromParent();
+        I = MBB.end();
+        continue;
+      }
+
+      // TBB is used to indicate the unconditinal destination.
+      TBB = I->getOperand(0).getMBB();
+      continue;
+    }
+
+    // Handle conditional branches.
+    assert(I->getOpcode() == MSP430::JCC && "Invalid conditional branch");
+    MSP430CC::CondCodes BranchCode =
+      static_cast<MSP430CC::CondCodes>(I->getOperand(1).getImm());
+    if (BranchCode == MSP430CC::COND_INVALID)
+      return true;  // Can't handle weird stuff.
+
+    // Working from the bottom, handle the first conditional branch.
+    if (Cond.empty()) {
+      FBB = TBB;
+      TBB = I->getOperand(0).getMBB();
+      Cond.push_back(MachineOperand::CreateImm(BranchCode));
+      continue;
+    }
+
+    // Handle subsequent conditional branches. Only handle the case where all
+    // conditional branches branch to the same destination.
+    assert(Cond.size() == 1);
+    assert(TBB);
+
+    // Only handle the case where all conditional branches branch to
+    // the same destination.
+    if (TBB != I->getOperand(0).getMBB())
+      return true;
+
+    MSP430CC::CondCodes OldBranchCode = (MSP430CC::CondCodes)Cond[0].getImm();
+    // If the conditions are the same, we can leave them alone.
+    if (OldBranchCode == BranchCode)
+      continue;
+
+    return true;
+  }
+
+  return false;
+}
+
 unsigned
 MSP430InstrInfo::InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
                               MachineBasicBlock *FBB,
@@ -172,7 +330,13 @@
 
   // Conditional branch.
   unsigned Count = 0;
-  llvm_unreachable("Implement conditional branches!");
+  BuildMI(&MBB, dl, get(MSP430::JCC)).addMBB(TBB).addImm(Cond[0].getImm());
+  ++Count;
 
+  if (FBB) {
+    // Two-way Conditional branch. Insert the second branch.
+    BuildMI(&MBB, dl, get(MSP430::JMP)).addMBB(FBB);
+    ++Count;
+  }
   return Count;
 }

Modified: llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h?rev=84774&r1=84773&r2=84774&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.h Wed Oct 21 14:17:18 2009
@@ -59,9 +59,19 @@
                                            MachineBasicBlock::iterator MI,
                                  const std::vector<CalleeSavedInfo> &CSI) const;
 
-  virtual unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
-                                MachineBasicBlock *FBB,
-                             const SmallVectorImpl<MachineOperand> &Cond) const;
+  // Branch folding goodness
+  bool ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const;
+  bool BlockHasNoFallThrough(const MachineBasicBlock &MBB) const;
+  bool isUnpredicatedTerminator(const MachineInstr *MI) const;
+  bool AnalyzeBranch(MachineBasicBlock &MBB,
+                     MachineBasicBlock *&TBB, MachineBasicBlock *&FBB,
+                     SmallVectorImpl<MachineOperand> &Cond,
+                     bool AllowModify) const;
+
+  unsigned RemoveBranch(MachineBasicBlock &MBB) const;
+  unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
+                        MachineBasicBlock *FBB,
+                        const SmallVectorImpl<MachineOperand> &Cond) const;
 
 };
 

Modified: llvm/trunk/lib/Target/MSP430/README.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/README.txt?rev=84774&r1=84773&r2=84774&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/README.txt (original)
+++ llvm/trunk/lib/Target/MSP430/README.txt Wed Oct 21 14:17:18 2009
@@ -40,5 +40,3 @@
 10. Handle imm in comparisons in better way (see comment in MSP430InstrInfo.td)
 
 11. Implement hooks for better memory op folding, etc.
-
-12. Implement branch folding.




From asl at math.spbu.ru  Wed Oct 21 14:17:55 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 19:17:55 -0000
Subject: [llvm-commits] [llvm] r84775 - in /llvm/trunk:
 lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
 lib/Target/MSP430/MSP430InstrInfo.cpp test/CodeGen/MSP430/Inst16mm.ll
 test/CodeGen/MSP430/Inst8mm.ll
Message-ID: <200910211917.n9LJHtE9003302@zion.cs.uiuc.edu>

Author: asl
Date: Wed Oct 21 14:17:55 2009
New Revision: 84775

URL: http://llvm.org/viewvc/llvm-project?rev=84775&view=rev
Log:
RMW preprocessing stuff was incorrect. Grab the stuff from x86 backend and disable some tests until it will be clever enough to handle them.

Modified:
    llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
    llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp
    llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll
    llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll

Modified: llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp?rev=84775&r1=84774&r2=84775&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp Wed Oct 21 14:17:55 2009
@@ -151,28 +151,15 @@
 /// MoveBelowTokenFactor - Replace TokenFactor operand with load's chain operand
 /// and move load below the TokenFactor. Replace store's chain operand with
 /// load's chain result.
-/// Shamelessly stolen from X86.
 static void MoveBelowTokenFactor(SelectionDAG *CurDAG, SDValue Load,
                                  SDValue Store, SDValue TF) {
   SmallVector<SDValue, 4> Ops;
-  bool isRMW = false;
-  SDValue TF0, TF1, NewTF;
   for (unsigned i = 0, e = TF.getNode()->getNumOperands(); i != e; ++i)
-    if (Load.getNode() == TF.getOperand(i).getNode()) {
-      TF0 = Load.getOperand(0);
-      Ops.push_back(TF0);
-    } else {
-      TF1 = TF.getOperand(i);
-      Ops.push_back(TF1);
-      if (LoadSDNode* LD = dyn_cast<LoadSDNode>(TF1))
-        isRMW = !LD->isVolatile();
-    }
-
-  if (isRMW && TF1.getOperand(0).getNode() == TF0.getNode())
-    NewTF = TF0;
-  else
-    NewTF = CurDAG->UpdateNodeOperands(TF, &Ops[0], Ops.size());
-
+    if (Load.getNode() == TF.getOperand(i).getNode())
+      Ops.push_back(Load.getOperand(0));
+    else
+      Ops.push_back(TF.getOperand(i));
+  SDValue NewTF = CurDAG->UpdateNodeOperands(TF, &Ops[0], Ops.size());
   SDValue NewLoad = CurDAG->UpdateNodeOperands(Load, NewTF,
                                                Load.getOperand(1),
                                                Load.getOperand(2));
@@ -180,10 +167,9 @@
                              Store.getOperand(2), Store.getOperand(3));
 }
 
-/// isRMWLoad - Return true if N is a load that's part of RMW sub-DAG. The chain
-/// produced by the load must only be used by the store's chain operand,
-/// otherwise this may produce a cycle in the DAG.
-/// Shamelessly stolen from X86. FIXME: Should we make this function common?
+/// isRMWLoad - Return true if N is a load that's part of RMW sub-DAG.
+/// The chain produced by the load must only be used by the store's chain
+/// operand, otherwise this may produce a cycle in the DAG.
 static bool isRMWLoad(SDValue N, SDValue Chain, SDValue Address,
                       SDValue &Load) {
   if (N.getOpcode() == ISD::BIT_CONVERT)
@@ -210,52 +196,86 @@
 }
 
 /// PreprocessForRMW - Preprocess the DAG to make instruction selection better.
-/// Shamelessly stolen from X86.
+/// This is only run if not in -O0 mode.
+/// This allows the instruction selector to pick more read-modify-write
+/// instructions. This is a common case:
+///
+///     [Load chain]
+///         ^
+///         |
+///       [Load]
+///       ^    ^
+///       |    |
+///      /      \-
+///     /         |
+/// [TokenFactor] [Op]
+///     ^          ^
+///     |          |
+///      \        /
+///       \      /
+///       [Store]
+///
+/// The fact the store's chain operand != load's chain will prevent the
+/// (store (op (load))) instruction from being selected. We can transform it to:
+///
+///     [Load chain]
+///         ^
+///         |
+///    [TokenFactor]
+///         ^
+///         |
+///       [Load]
+///       ^    ^
+///       |    |
+///       |     \-
+///       |       |
+///       |     [Op]
+///       |       ^
+///       |       |
+///       \      /
+///        \    /
+///       [Store]
 void MSP430DAGToDAGISel::PreprocessForRMW() {
   for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(),
          E = CurDAG->allnodes_end(); I != E; ++I) {
     if (!ISD::isNON_TRUNCStore(I))
       continue;
-
     SDValue Chain = I->getOperand(0);
+
     if (Chain.getNode()->getOpcode() != ISD::TokenFactor)
       continue;
 
-    SDValue N1 = I->getOperand(1); // Value to store
-    SDValue N2 = I->getOperand(2); // Address of store
-
-    if (!N1.hasOneUse())
+    SDValue N1 = I->getOperand(1);
+    SDValue N2 = I->getOperand(2);
+    if ((N1.getValueType().isFloatingPoint() &&
+         !N1.getValueType().isVector()) ||
+        !N1.hasOneUse())
       continue;
 
     bool RModW = false;
     SDValue Load;
     unsigned Opcode = N1.getNode()->getOpcode();
     switch (Opcode) {
-      case ISD::ADD:
-      case ISD::AND:
-      case ISD::OR:
-      case ISD::XOR:
-      case ISD::ADDC:
-      case ISD::ADDE: {
-        SDValue N10 = N1.getOperand(0);
-        SDValue N11 = N1.getOperand(1);
-        RModW = isRMWLoad(N10, Chain, N2, Load);
-
-        if (!RModW && isRMWLoad(N11, Chain, N2, Load)) {
-          // Swap the operands, making the RMW load the first operand seems
-          // to help selection and prevent token chain loops.
-          N1 = CurDAG->UpdateNodeOperands(N1, N11, N10);
-          RModW = true;
-        }
-       break;
-      }
-      case ISD::SUB:
-      case ISD::SUBC:
-      case ISD::SUBE: {
-        SDValue N10 = N1.getOperand(0);
-        RModW = isRMWLoad(N10, Chain, N2, Load);
-        break;
-      }
+    case ISD::ADD:
+    case ISD::AND:
+    case ISD::OR:
+    case ISD::XOR:
+    case ISD::ADDC:
+    case ISD::ADDE: {
+      SDValue N10 = N1.getOperand(0);
+      SDValue N11 = N1.getOperand(1);
+      RModW = isRMWLoad(N10, Chain, N2, Load);
+      if (!RModW)
+        RModW = isRMWLoad(N11, Chain, N2, Load);
+      break;
+    }
+    case ISD::SUB:
+    case ISD::SUBC:
+    case ISD::SUBE: {
+      SDValue N10 = N1.getOperand(0);
+      RModW = isRMWLoad(N10, Chain, N2, Load);
+      break;
+    }
     }
 
     if (RModW) {

Modified: llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp?rev=84775&r1=84774&r2=84775&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430InstrInfo.cpp Wed Oct 21 14:17:55 2009
@@ -169,7 +169,6 @@
   return Count;
 }
 
-
 bool MSP430InstrInfo::
 ReverseBranchCondition(SmallVectorImpl<MachineOperand> &Cond) const {
   assert(Cond.size() == 1 && "Invalid Xbranch condition!");

Modified: llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll?rev=84775&r1=84774&r2=84775&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll (original)
+++ llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll Wed Oct 21 14:17:55 2009
@@ -1,4 +1,5 @@
 ; RUN: llc -march=msp430 < %s | FileCheck %s
+; XFAIL
 target datalayout = "e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8"
 target triple = "msp430-generic-generic"
 @foo = common global i16 0, align 2

Modified: llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll?rev=84775&r1=84774&r2=84775&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll (original)
+++ llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll Wed Oct 21 14:17:55 2009
@@ -1,4 +1,5 @@
 ; RUN: llc -march=msp430 < %s | FileCheck %s
+; XFAIL
 target datalayout = "e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8"
 target triple = "msp430-generic-generic"
 




From asl at math.spbu.ru  Wed Oct 21 14:18:28 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 19:18:28 -0000
Subject: [llvm-commits] [llvm] r84776 - in /llvm/trunk:
 lib/Target/MSP430/MSP430ISelDAGToDAG.cpp test/CodeGen/MSP430/Inst16mm.ll
 test/CodeGen/MSP430/Inst8mm.ll
Message-ID: <200910211918.n9LJISD0003347@zion.cs.uiuc.edu>

Author: asl
Date: Wed Oct 21 14:18:28 2009
New Revision: 84776

URL: http://llvm.org/viewvc/llvm-project?rev=84776&view=rev
Log:
Add DAG printing for RMW stuff debugging

Modified:
    llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
    llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll
    llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll

Modified: llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp?rev=84776&r1=84775&r2=84776&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp Wed Oct 21 14:18:28 2009
@@ -26,6 +26,7 @@
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/Target/TargetLowering.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
@@ -34,6 +35,14 @@
 
 using namespace llvm;
 
+#ifndef NDEBUG
+static cl::opt<bool>
+ViewRMWDAGs("view-msp430-rmw-dags", cl::Hidden,
+          cl::desc("Pop up a window to show isel dags after RMW preprocess"));
+#else
+static const bool ViewRMWDAGs = false;
+#endif
+
 STATISTIC(NumLoadMoved, "Number of loads moved below TokenFactor");
 
 /// MSP430DAGToDAGISel - MSP430 specific code to select MSP430 machine
@@ -288,8 +297,15 @@
 /// InstructionSelect - This callback is invoked by
 /// SelectionDAGISel when it has created a SelectionDAG for us to codegen.
 void MSP430DAGToDAGISel::InstructionSelect() {
+  std::string BlockName;
+  if (ViewRMWDAGs)
+    BlockName = MF->getFunction()->getNameStr() + ":" +
+                BB->getBasicBlock()->getNameStr();
+
   PreprocessForRMW();
 
+  if (ViewRMWDAGs) CurDAG->viewGraph("RMW preprocessed:" + BlockName);
+
   DEBUG(errs() << "Selection DAG after RMW preprocessing:\n");
   DEBUG(CurDAG->dump());
 

Modified: llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll?rev=84776&r1=84775&r2=84776&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll (original)
+++ llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll Wed Oct 21 14:18:28 2009
@@ -1,5 +1,5 @@
 ; RUN: llc -march=msp430 < %s | FileCheck %s
-; XFAIL
+; XFAIL: *
 target datalayout = "e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8"
 target triple = "msp430-generic-generic"
 @foo = common global i16 0, align 2

Modified: llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll?rev=84776&r1=84775&r2=84776&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll (original)
+++ llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll Wed Oct 21 14:18:28 2009
@@ -1,5 +1,5 @@
 ; RUN: llc -march=msp430 < %s | FileCheck %s
-; XFAIL
+; XFAIL: *
 target datalayout = "e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8"
 target triple = "msp430-generic-generic"
 




From grosbach at apple.com  Wed Oct 21 15:44:35 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Wed, 21 Oct 2009 20:44:35 -0000
Subject: [llvm-commits] [llvm] r84778 - in /llvm/trunk:
 lib/Target/ARM/ARMAddressingModes.h lib/Target/ARM/ARMInstrInfo.td
 lib/Target/ARM/ARMInstrThumb2.td test/CodeGen/Thumb2/thumb2-mov.ll
 test/CodeGen/Thumb2/thumb2-mov2.ll
Message-ID: <200910212044.n9LKia3w008119@zion.cs.uiuc.edu>

Author: grosbach
Date: Wed Oct 21 15:44:34 2009
New Revision: 84778

URL: http://llvm.org/viewvc/llvm-project?rev=84778&view=rev
Log:
Improve handling of immediates by splitting 32-bit immediates into two 16-bit
immediate operands when they will fit into the using instruction.


Modified:
    llvm/trunk/lib/Target/ARM/ARMAddressingModes.h
    llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
    llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td
    llvm/trunk/test/CodeGen/Thumb2/thumb2-mov.ll
    llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll

Modified: llvm/trunk/lib/Target/ARM/ARMAddressingModes.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMAddressingModes.h?rev=84778&r1=84777&r2=84778&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMAddressingModes.h (original)
+++ llvm/trunk/lib/Target/ARM/ARMAddressingModes.h Wed Oct 21 15:44:34 2009
@@ -341,6 +341,66 @@
     return -1;
   }
 
+  static inline unsigned getT2SOImmValRotate(unsigned V) {
+    if ((V & ~255U) == 0) return 0;
+    // Use CTZ to compute the rotate amount.
+    unsigned RotAmt = CountTrailingZeros_32(V);
+    return (32 - RotAmt) & 31;
+  }
+
+  static inline bool isT2SOImmTwoPartVal (unsigned Imm) {
+    unsigned V = Imm;
+    // Passing values can be any combination of splat values and shifter
+    // values. If this can be handled with a single shifter or splat, bail
+    // out. Those should be handled directly, not with a two-part val.
+    if (getT2SOImmValSplatVal(V) != -1)
+      return false;
+    V = rotr32 (~255U, getT2SOImmValRotate(V)) & V;
+    if (V == 0)
+      return false;
+
+    // If this can be handled as an immediate, accept.
+    if (getT2SOImmVal(V) != -1) return true;
+
+    // Likewise, try masking out a splat value first.
+    V = Imm;
+    if (getT2SOImmValSplatVal(V & 0xff00ff00U) != -1)
+      V &= ~0xff00ff00U;
+    else if (getT2SOImmValSplatVal(V & 0x00ff00ffU) != -1)
+      V &= ~0x00ff00ffU;
+    // If what's left can be handled as an immediate, accept.
+    if (getT2SOImmVal(V) != -1) return true;
+
+    // Otherwise, do not accept.
+    return false;
+  }
+
+  static inline unsigned getT2SOImmTwoPartFirst(unsigned Imm) {
+    assert (isT2SOImmTwoPartVal(Imm) &&
+            "Immedate cannot be encoded as two part immediate!");
+    // Try a shifter operand as one part
+    unsigned V = rotr32 (~255, getT2SOImmValRotate(Imm)) & Imm;
+    // If the rest is encodable as an immediate, then return it.
+    if (getT2SOImmVal(V) != -1) return V;
+
+    // Try masking out a splat value first.
+    if (getT2SOImmValSplatVal(Imm & 0xff00ff00U) != -1)
+      return Imm & 0xff00ff00U;
+
+    // The other splat is all that's left as an option.
+    assert (getT2SOImmValSplatVal(Imm & 0x00ff00ffU) != -1);
+    return Imm & 0x00ff00ffU;
+  }
+
+  static inline unsigned getT2SOImmTwoPartSecond(unsigned Imm) {
+    // Mask out the first hunk
+    Imm ^= getT2SOImmTwoPartFirst(Imm);
+    // Return what's left
+    assert (getT2SOImmVal(Imm) != -1 &&
+            "Unable to encode second part of T2 two part SO immediate");
+    return Imm;
+  }
+
 
   //===--------------------------------------------------------------------===//
   // Addressing Mode #2

Modified: llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrInfo.td?rev=84778&r1=84777&r2=84778&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrInfo.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrInfo.td Wed Oct 21 15:44:34 2009
@@ -1583,6 +1583,12 @@
 def : ARMPat<(xor GPR:$LHS, so_imm2part:$RHS),
              (EORri (EORri GPR:$LHS, (so_imm2part_1 imm:$RHS)),
                     (so_imm2part_2 imm:$RHS))>;
+def : ARMPat<(add GPR:$LHS, so_imm2part:$RHS),
+             (ADDri (ADDri GPR:$LHS, (so_imm2part_1 imm:$RHS)),
+                    (so_imm2part_2 imm:$RHS))>;
+def : ARMPat<(sub GPR:$LHS, so_imm2part:$RHS),
+             (SUBri (SUBri GPR:$LHS, (so_imm2part_1 imm:$RHS)),
+                    (so_imm2part_2 imm:$RHS))>;
 
 // 32-bit immediate using movw + movt.
 // This is a single pseudo instruction, the benefit is that it can be remat'd

Modified: llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td?rev=84778&r1=84777&r2=84778&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrThumb2.td Wed Oct 21 15:44:34 2009
@@ -69,6 +69,25 @@
   return ARM_AM::getT2SOImmVal(-((int)N->getZExtValue())) != -1;
 }], t2_so_imm_neg_XFORM>;
 
+// Break t2_so_imm's up into two pieces.  This handles immediates with up to 16
+// bits set in them.  This uses t2_so_imm2part to match and t2_so_imm2part_[12]
+// to get the first/second pieces.
+def t2_so_imm2part : Operand<i32>,
+                  PatLeaf<(imm), [{
+      return ARM_AM::isT2SOImmTwoPartVal((unsigned)N->getZExtValue());
+    }]> {
+}
+
+def t2_so_imm2part_1 : SDNodeXForm<imm, [{
+  unsigned V = ARM_AM::getT2SOImmTwoPartFirst((unsigned)N->getZExtValue());
+  return CurDAG->getTargetConstant(V, MVT::i32);
+}]>;
+
+def t2_so_imm2part_2 : SDNodeXForm<imm, [{
+  unsigned V = ARM_AM::getT2SOImmTwoPartSecond((unsigned)N->getZExtValue());
+  return CurDAG->getTargetConstant(V, MVT::i32);
+}]>;
+
 /// imm1_31 predicate - True if the 32-bit immediate is in the range [1,31].
 def imm1_31 : PatLeaf<(i32 imm), [{
   return (int32_t)N->getZExtValue() >= 1 && (int32_t)N->getZExtValue() < 32;
@@ -1131,6 +1150,20 @@
 // Non-Instruction Patterns
 //
 
+// Two piece so_imms.
+def : T2Pat<(or GPR:$LHS, t2_so_imm2part:$RHS),
+             (t2ORRri (t2ORRri GPR:$LHS, (t2_so_imm2part_1 imm:$RHS)),
+                    (t2_so_imm2part_2 imm:$RHS))>;
+def : T2Pat<(xor GPR:$LHS, t2_so_imm2part:$RHS),
+             (t2EORri (t2EORri GPR:$LHS, (t2_so_imm2part_1 imm:$RHS)),
+                    (t2_so_imm2part_2 imm:$RHS))>;
+def : T2Pat<(add GPR:$LHS, t2_so_imm2part:$RHS),
+             (t2ADDri (t2ADDri GPR:$LHS, (t2_so_imm2part_1 imm:$RHS)),
+                    (t2_so_imm2part_2 imm:$RHS))>;
+def : T2Pat<(sub GPR:$LHS, t2_so_imm2part:$RHS),
+             (t2SUBri (t2SUBri GPR:$LHS, (t2_so_imm2part_1 imm:$RHS)),
+                    (t2_so_imm2part_2 imm:$RHS))>;
+
 // ConstantPool, GlobalAddress, and JumpTable
 def : T2Pat<(ARMWrapper  tglobaladdr :$dst), (t2LEApcrel tglobaladdr :$dst)>;
 def : T2Pat<(ARMWrapper  tconstpool  :$dst), (t2LEApcrel tconstpool  :$dst)>;

Modified: llvm/trunk/test/CodeGen/Thumb2/thumb2-mov.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Thumb2/thumb2-mov.ll?rev=84778&r1=84777&r2=84778&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/Thumb2/thumb2-mov.ll (original)
+++ llvm/trunk/test/CodeGen/Thumb2/thumb2-mov.ll Wed Oct 21 15:44:34 2009
@@ -10,29 +10,32 @@
     ret i32 %ret
 }
 
-define i32 @t2_const_var2_1_fail_1(i32 %lhs) {
-;CHECK: t2_const_var2_1_fail_1:
-;CHECK: movt
+define i32 @t2_const_var2_1_ok_2(i32 %lhs) {
+;CHECK: t2_const_var2_1_ok_2:
+;CHECK: #11206656
+;CHECK: #187
     %ret = add i32 %lhs, 11206843 ; 0x00ab00bb
     ret i32 %ret
 }
 
-define i32 @t2_const_var2_1_fail_2(i32 %lhs) {
-;CHECK: t2_const_var2_1_fail_2:
-;CHECK: movt
+define i32 @t2_const_var2_1_ok_3(i32 %lhs) {
+;CHECK: t2_const_var2_1_ok_3:
+;CHECK: #11206827
+;CHECK: #16777216
     %ret = add i32 %lhs, 27984043 ; 0x01ab00ab
     ret i32 %ret
 }
 
-define i32 @t2_const_var2_1_fail_3(i32 %lhs) {
-;CHECK: t2_const_var2_1_fail_3:
-;CHECK: movt
+define i32 @t2_const_var2_1_ok_4(i32 %lhs) {
+;CHECK: t2_const_var2_1_ok_4:
+;CHECK: #16777472
+;CHECK: #11206827
     %ret = add i32 %lhs, 27984299 ; 0x01ab01ab
     ret i32 %ret
 }
 
-define i32 @t2_const_var2_1_fail_4(i32 %lhs) {
-;CHECK: t2_const_var2_1_fail_4:
+define i32 @t2_const_var2_1_fail_1(i32 %lhs) {
+;CHECK: t2_const_var2_1_fail_1:
 ;CHECK: movt
     %ret = add i32 %lhs, 28027649 ; 0x01abab01
     ret i32 %ret
@@ -46,29 +49,31 @@
     ret i32 %ret
 }
 
-define i32 @t2_const_var2_2_fail_1(i32 %lhs) {
-;CHECK: t2_const_var2_2_fail_1:
-;CHECK: movt
+define i32 @t2_const_var2_2_ok_2(i32 %lhs) {
+;CHECK: t2_const_var2_2_ok_2:
+;CHECK: #-1426063360
+;CHECK: #47616
     %ret = add i32 %lhs, 2868951552 ; 0xab00ba00
     ret i32 %ret
 }
 
-define i32 @t2_const_var2_2_fail_2(i32 %lhs) {
-;CHECK: t2_const_var2_2_fail_2:
-;CHECK: movt
+define i32 @t2_const_var2_2_ok_3(i32 %lhs) {
+;CHECK: t2_const_var2_2_ok_3:
+;CHECK: #-1426019584
     %ret = add i32 %lhs, 2868947728 ; 0xab00ab10
     ret i32 %ret
 }
 
-define i32 @t2_const_var2_2_fail_3(i32 %lhs) {
-;CHECK: t2_const_var2_2_fail_3:
-;CHECK: movt
+define i32 @t2_const_var2_2_ok_4(i32 %lhs) {
+;CHECK: t2_const_var2_2_ok_4:
+;CHECK: #-1426019584
+;CHECK: #1048592
     %ret = add i32 %lhs, 2869996304 ; 0xab10ab10
     ret i32 %ret
 }
 
-define i32 @t2_const_var2_2_fail_4(i32 %lhs) {
-;CHECK: t2_const_var2_2_fail_4:
+define i32 @t2_const_var2_2_fail_1(i32 %lhs) {
+;CHECK: t2_const_var2_2_fail_1:
 ;CHECK: movt
     %ret = add i32 %lhs, 279685904 ; 0x10abab10
     ret i32 %ret
@@ -125,9 +130,10 @@
     ret i32 %ret
 }
 
-define i32 @t2_const_var3_2_fail_1(i32 %lhs) {
-;CHECK: t2_const_var3_2_fail_1:
-;CHECK: movt
+define i32 @t2_const_var3_2_ok_2(i32 %lhs) {
+;CHECK: t2_const_var3_2_ok_2:
+;CHECK: #2097152
+;CHECK: #1843200
     %ret = add i32 %lhs, 3940352 ; 0b00000000001111000010000000000000
     ret i32 %ret
 }

Modified: llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll?rev=84778&r1=84777&r2=84778&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll (original)
+++ llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll Wed Oct 21 15:44:34 2009
@@ -55,10 +55,10 @@
 
 define i32 @t2MOVTi16_test_nomatch_1(i32 %a) {
 ; CHECK: t2MOVTi16_test_nomatch_1:
-; CHECK:      movw r1, #16384
-; CHECK-NEXT: movt r1, #154
+; CHECK:      #8388608
 ; CHECK:      movw r1, #65535
 ; CHECK-NEXT: movt r1, #154
+; CHECK:      #1720320
     %1 = shl i32  255,   8
     %2 = shl i32 1234,   8
     %3 = or  i32   %1, 255  ; This give us 0xFFFF in %3




From isanbard at gmail.com  Wed Oct 21 16:09:48 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Wed, 21 Oct 2009 21:09:48 -0000
Subject: [llvm-commits] [llvm] r84782 -
	/llvm/trunk/test/FrontendC++/member-alignment.cpp
Message-ID: <200910212109.n9LL9mqq009255@zion.cs.uiuc.edu>

Author: void
Date: Wed Oct 21 16:09:48 2009
New Revision: 84782

URL: http://llvm.org/viewvc/llvm-project?rev=84782&view=rev
Log:
XFAIL this test for PPC.

Modified:
    llvm/trunk/test/FrontendC++/member-alignment.cpp

Modified: llvm/trunk/test/FrontendC++/member-alignment.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC%2B%2B/member-alignment.cpp?rev=84782&r1=84781&r2=84782&view=diff

==============================================================================
--- llvm/trunk/test/FrontendC++/member-alignment.cpp (original)
+++ llvm/trunk/test/FrontendC++/member-alignment.cpp Wed Oct 21 16:09:48 2009
@@ -1,5 +1,5 @@
 // RUN: %llvmgxx -S -emit-llvm %s -o - | FileCheck %s
-// XFAIL: arm
+// XFAIL: arm,powerpc
 
 // rdar://7268289
 




From deeppatel1987 at gmail.com  Wed Oct 21 16:11:18 2009
From: deeppatel1987 at gmail.com (Sandeep Patel)
Date: Wed, 21 Oct 2009 21:11:18 +0000
Subject: [llvm-commits] [llvm] r84764 -
	/llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
In-Reply-To: <200910211727.n9LHRO3h031223@zion.cs.uiuc.edu>
References: <200910211727.n9LHRO3h031223@zion.cs.uiuc.edu>
Message-ID: <305d6f60910211411t4ec59710tcb61f6271eb63dcb@mail.gmail.com>

This causes build failures for my MingW canadian cross:

/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.text+0x12):
undefined reference to `llvm::CallGraph::ID'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.text+0x17):
undefined reference to `llvm::Pass::lookupPassInfo(int)'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.text+0x560):
undefined reference to `llvm::CallSite::CallSite(llvm::Instruction*)'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.data+0x3f4):
undefined reference to `llvm::CallGraphLinkVar'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.text$linkonce__ZN4llvm17PIC16FrameOverlayD0Ev+0x28):
undefined reference to `llvm::ModulePass::~ModulePass()'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.text$linkonce__ZN4llvm17PIC16FrameOverlayD1Ev+0x28):
undefined reference to `llvm::ModulePass::~ModulePass()'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.text$linkonce__ZNK4llvm18value_use_iteratorINS_4UserEEdeEv+0x11):
undefined reference to `llvm::Use::getUser() const'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.text$linkonce__ZNK4llvm4Pass11getAnalysisINS_9CallGraphEEERT_v+0x11):
undefined reference to `llvm::CallGraph::ID'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.text$linkonce__ZNK4llvm4Pass11getAnalysisINS_9CallGraphEEERT_v+0x16):
undefined reference to `llvm::Pass::lookupPassInfo(int)'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.rdata$linkonce__ZTVN4llvm17PIC16FrameOverlayE+0x10):
undefined reference to `llvm::Pass::getPassName() const'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.rdata$linkonce__ZTVN4llvm17PIC16FrameOverlayE+0x14):
undefined reference to `llvm::Pass::print(llvm::raw_ostream&,
llvm::Module const*) const'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.rdata$linkonce__ZTVN4llvm17PIC16FrameOverlayE+0x18):
undefined reference to
`llvm::ModulePass::assignPassManager(llvm::PMStack&,
llvm::PassManagerType)'
/Volumes/src/mingwbuild/llvm/lib/Target/PIC16/PIC16Passes/Release/PIC16Overlay.o:fake:(.rdata$linkonce__ZTVN4llvm17PIC16FrameOverlayE+0x30):
undefined reference to `llvm::Pass::dumpPassStructure(unsigned int)'

deep

On Wed, Oct 21, 2009 at 5:27 PM, Sanjiv Gupta
<sanjiv.gupta at microchip.com> wrote:
> Author: sgupta
> Date: Wed Oct 21 12:27:23 2009
> New Revision: 84764
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84764&view=rev
> Log:
> Build shared lib instead of an archive.
>
> Modified:
> ? ?llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
>
> Modified: llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile?rev=84764&r1=84763&r2=84764&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile (original)
> +++ llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile Wed Oct 21 12:27:23 2009
> @@ -9,7 +9,7 @@
> ?LEVEL = ../../../..
> ?TARGET = PIC16
> ?LIBRARYNAME = LLVMpic16passes
> -BUILD_ARCHIVE = 1
> +LOADABLE_MODULE = 1
>
>
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>



From asl at math.spbu.ru  Wed Oct 21 16:15:18 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Wed, 21 Oct 2009 21:15:18 -0000
Subject: [llvm-commits] [llvm] r84783 -
	/llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
Message-ID: <200910212115.n9LLFIOe009491@zion.cs.uiuc.edu>

Author: asl
Date: Wed Oct 21 16:15:18 2009
New Revision: 84783

URL: http://llvm.org/viewvc/llvm-project?rev=84783&view=rev
Log:
Revert r84764, it breaks mingw build

Modified:
    llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile

Modified: llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile?rev=84783&r1=84782&r2=84783&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile (original)
+++ llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile Wed Oct 21 16:15:18 2009
@@ -9,7 +9,7 @@
 LEVEL = ../../../..
 TARGET = PIC16
 LIBRARYNAME = LLVMpic16passes
-LOADABLE_MODULE = 1
+BUILD_ARCHIVE = 1
 
 
 




From anton at korobeynikov.info  Wed Oct 21 16:15:58 2009
From: anton at korobeynikov.info (Anton Korobeynikov)
Date: Thu, 22 Oct 2009 01:15:58 +0400
Subject: [llvm-commits] [llvm] r84764 -
	/llvm/trunk/lib/Target/PIC16/PIC16Passes/Makefile
In-Reply-To: <305d6f60910211411t4ec59710tcb61f6271eb63dcb@mail.gmail.com>
References: <200910211727.n9LHRO3h031223@zion.cs.uiuc.edu>
	<305d6f60910211411t4ec59710tcb61f6271eb63dcb@mail.gmail.com>
Message-ID: <b087fca00910211415y5191e334ne66f5fcb9c8c6749@mail.gmail.com>

> This causes build failures for my MingW canadian cross:
I've reverted the commit

-- 
With best regards, Anton Korobeynikov
Faculty of Mathematics and Mechanics, Saint Petersburg State University


From dpatel at apple.com  Wed Oct 21 16:25:10 2009
From: dpatel at apple.com (Devang Patel)
Date: Wed, 21 Oct 2009 21:25:10 -0000
Subject: [llvm-commits] [llvm] r84784 - in /llvm/trunk:
 include/llvm/Metadata.h lib/Bitcode/Writer/ValueEnumerator.cpp
 lib/VMCore/AsmWriter.cpp lib/VMCore/Metadata.cpp
Message-ID: <200910212125.n9LLPAEL009873@zion.cs.uiuc.edu>

Author: dpatel
Date: Wed Oct 21 16:25:09 2009
New Revision: 84784

URL: http://llvm.org/viewvc/llvm-project?rev=84784&view=rev
Log:
Do not use SmallVector to store MDNode elements.

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp
    llvm/trunk/lib/VMCore/AsmWriter.cpp
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84784&r1=84783&r2=84784&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Wed Oct 21 16:25:09 2009
@@ -110,6 +110,7 @@
   // Use CallbackVH to hold MDNOde elements.
   struct ElementVH : public CallbackVH {
     MDNode *Parent;
+    ElementVH() {}
     ElementVH(Value *V, MDNode *P) : CallbackVH(V), Parent(P) {}
     ~ElementVH() {}
 
@@ -124,7 +125,8 @@
   // Replace each instance of F from the element list of this node with T.
   void replaceElement(Value *F, Value *T);
 
-  SmallVector<ElementVH, 4> Node;
+  ElementVH *Node;
+  unsigned NodeSize;
 
 protected:
   explicit MDNode(LLVMContext &C, Value *const *Vals, unsigned NumVals);
@@ -150,19 +152,7 @@
   }
 
   /// getNumElements - Return number of MDNode elements.
-  unsigned getNumElements() const {
-    return Node.size();
-  }
-
-  // Element access
-  typedef SmallVectorImpl<ElementVH>::const_iterator const_elem_iterator;
-  typedef SmallVectorImpl<ElementVH>::iterator elem_iterator;
-  /// elem_empty - Return true if MDNode is empty.
-  bool elem_empty() const                { return Node.empty(); }
-  const_elem_iterator elem_begin() const { return Node.begin(); }
-  const_elem_iterator elem_end() const   { return Node.end();   }
-  elem_iterator elem_begin()             { return Node.begin(); }
-  elem_iterator elem_end()               { return Node.end();   }
+  unsigned getNumElements() const { return NodeSize; }
 
   /// Profile - calculate a unique identifier for this MDNode to collapse
   /// duplicates

Modified: llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp?rev=84784&r1=84783&r2=84784&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp (original)
+++ llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp Wed Oct 21 16:25:09 2009
@@ -214,10 +214,9 @@
     MDValues.push_back(std::make_pair(MD, 1U));
     MDValueMap[MD] = MDValues.size();
     MDValueID = MDValues.size();
-    for (MDNode::const_elem_iterator I = N->elem_begin(), E = N->elem_end();
-         I != E; ++I) {
-      if (*I)
-        EnumerateValue(*I);
+    for (unsigned i = 0, e = N->getNumElements(); i != e; ++i) {    
+      if (Value *V = N->getElement(i))
+        EnumerateValue(V);
       else
         EnumerateType(Type::getVoidTy(MD->getContext()));
     }

Modified: llvm/trunk/lib/VMCore/AsmWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/AsmWriter.cpp?rev=84784&r1=84783&r2=84784&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/AsmWriter.cpp (original)
+++ llvm/trunk/lib/VMCore/AsmWriter.cpp Wed Oct 21 16:25:09 2009
@@ -818,9 +818,8 @@
   unsigned DestSlot = mdnNext++;
   mdnMap[N] = DestSlot;
 
-  for (MDNode::const_elem_iterator MDI = N->elem_begin(),
-         MDE = N->elem_end(); MDI != MDE; ++MDI) {
-    const Value *TV = *MDI;
+  for (unsigned i = 0, e = N->getNumElements(); i != e; ++i) {
+    const Value *TV = N->getElement(i);
     if (TV)
       if (const MDNode *N2 = dyn_cast<MDNode>(TV))
         CreateMetadataSlot(N2);
@@ -906,9 +905,8 @@
     Out << '!' << i << " = metadata ";
     const MDNode *Node = Nodes[i];
     Out << "!{";
-    for (MDNode::const_elem_iterator NI = Node->elem_begin(),
-           NE = Node->elem_end(); NI != NE;) {
-      const Value *V = *NI;
+    for (unsigned mi = 0, me = Node->getNumElements(); mi != me; ++mi) {
+      const Value *V = Node->getElement(mi);
       if (!V)
         Out << "null";
       else if (const MDNode *N = dyn_cast<MDNode>(V)) {
@@ -916,11 +914,12 @@
         Out << '!' << Machine.getMetadataSlot(N);
       }
       else {
-        TypePrinter.print((*NI)->getType(), Out);
+        TypePrinter.print(V->getType(), Out);
         Out << ' ';
-        WriteAsOperandInternal(Out, *NI, &TypePrinter, &Machine);
+        WriteAsOperandInternal(Out, Node->getElement(mi), 
+                               &TypePrinter, &Machine);
       }
-      if (++NI != NE)
+      if (mi + 1 != me)
         Out << ", ";
     }
 

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84784&r1=84783&r2=84784&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 16:25:09 2009
@@ -67,6 +67,9 @@
   : MetadataBase(Type::getMetadataTy(C), Value::MDNodeVal) {
   NumOperands = 0;
   resizeOperands(NumVals);
+  NodeSize = NumVals;
+  Node = new ElementVH[NodeSize];
+  ElementVH *Ptr = Node;
   for (unsigned i = 0; i != NumVals; ++i) {
     // Only record metadata uses.
     if (MetadataBase *MB = dyn_cast_or_null<MetadataBase>(Vals[i]))
@@ -74,13 +77,13 @@
     else if(Vals[i] && 
             Vals[i]->getType()->getTypeID() == Type::MetadataTyID)
       OperandList[NumOperands++] = Vals[i];
-    Node.push_back(ElementVH(Vals[i], this));
+    *Ptr++ = ElementVH(Vals[i], this);
   }
 }
 
 void MDNode::Profile(FoldingSetNodeID &ID) const {
-  for (const_elem_iterator I = elem_begin(), E = elem_end(); I != E; ++I)
-    ID.AddPointer(*I);
+  for (unsigned i = 0, e = getNumElements(); i != e; ++i)
+    ID.AddPointer(getElement(i));
 }
 
 MDNode *MDNode::get(LLVMContext &Context, Value*const* Vals, unsigned NumVals) {
@@ -109,7 +112,8 @@
 /// dropAllReferences - Remove all uses and clear node vector.
 void MDNode::dropAllReferences() {
   User::dropAllReferences();
-  Node.clear();
+  delete [] Node;
+  Node = NULL;
 }
 
 MDNode::~MDNode() {
@@ -132,9 +136,8 @@
   // From in this MDNode's element list.
   SmallVector<unsigned, 4> Indexes;
   unsigned Index = 0;
-  for (SmallVector<ElementVH, 4>::iterator I = Node.begin(),
-         E = Node.end(); I != E; ++I, ++Index) {
-    Value *V = *I;
+  for (unsigned i = 0, e = getNumElements(); i != e; ++i, ++Index) {
+    Value *V = getElement(i);
     if (V && V == From) 
       Indexes.push_back(Index);
   }




From kennethuil at gmail.com  Wed Oct 21 16:31:27 2009
From: kennethuil at gmail.com (Kenneth Uildriks)
Date: Wed, 21 Oct 2009 16:31:27 -0500
Subject: [llvm-commits] [PATCH] Allow "opt" to run without target data
Message-ID: <400d33ea0910211431r5efc19fan571752db8914bde3@mail.gmail.com>

Currently, there is no way to prevent "opt" from putting in default
target data layout if the Module has none, or to supply an alternate
target data layout.  The default target data layout is bogus on most
platforms.

This patch adds the following command-line options:

-default-data-layout=<layout-string> - data layout string to use if
not specified by module
-no-default-data-layout - no data layout assumptions unless module
specifies data layout.

The -no-default-data-layout prevents a TargetData pass from being
added to opt's PassManager and (theoretically) prevents platform data
layout dependent optimization from being applied.  I think this should
eventually be the default, but setting it to the default might break
existing users and test scripts.  For now, if neither option is
supplied, opt's behavior will remain unchanged from the current
(trunk) version.


From kennethuil at gmail.com  Wed Oct 21 16:33:01 2009
From: kennethuil at gmail.com (Kenneth Uildriks)
Date: Wed, 21 Oct 2009 16:33:01 -0500
Subject: [llvm-commits] [PATCH] Allow "opt" to run without target data
In-Reply-To: <400d33ea0910211431r5efc19fan571752db8914bde3@mail.gmail.com>
References: <400d33ea0910211431r5efc19fan571752db8914bde3@mail.gmail.com>
Message-ID: <400d33ea0910211433w758b2491k50b1c3d4e7ac4370@mail.gmail.com>

On Wed, Oct 21, 2009 at 4:31 PM, Kenneth Uildriks <kennethuil at gmail.com> wrote:
> Currently, there is no way to prevent "opt" from putting in default
> target data layout if the Module has none, or to supply an alternate
> target data layout. ?The default target data layout is bogus on most
> platforms.
>
> This patch adds the following command-line options:
>
> -default-data-layout=<layout-string> - data layout string to use if
> not specified by module
> -no-default-data-layout - no data layout assumptions unless module
> specifies data layout.
>
> The -no-default-data-layout prevents a TargetData pass from being
> added to opt's PassManager and (theoretically) prevents platform data
> layout dependent optimization from being applied. ?I think this should
> eventually be the default, but setting it to the default might break
> existing users and test scripts. ?For now, if neither option is
> supplied, opt's behavior will remain unchanged from the current
> (trunk) version.
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: opt.patch
Type: text/x-patch
Size: 1632 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091021/858d1c25/attachment.bin 

From bob.wilson at apple.com  Wed Oct 21 16:36:27 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Wed, 21 Oct 2009 21:36:27 -0000
Subject: [llvm-commits] [llvm] r84785 -
	/llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
Message-ID: <200910212136.n9LLaRXl010355@zion.cs.uiuc.edu>

Author: bwilson
Date: Wed Oct 21 16:36:27 2009
New Revision: 84785

URL: http://llvm.org/viewvc/llvm-project?rev=84785&view=rev
Log:
Most of the NEON shuffle instructions do not support 64-bit element types.

Modified:
    llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp

Modified: llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp?rev=84785&r1=84784&r2=84785&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp Wed Oct 21 16:36:27 2009
@@ -2361,8 +2361,11 @@
   assert((BlockSize==16 || BlockSize==32 || BlockSize==64) &&
          "Only possible block sizes for VREV are: 16, 32, 64");
 
-  unsigned NumElts = VT.getVectorNumElements();
   unsigned EltSz = VT.getVectorElementType().getSizeInBits();
+  if (EltSz == 64)
+    return false;
+
+  unsigned NumElts = VT.getVectorNumElements();
   unsigned BlockElts = M[0] + 1;
 
   if (BlockSize <= EltSz || BlockSize != BlockElts * EltSz)
@@ -2379,6 +2382,10 @@
 
 static bool isVTRNMask(const SmallVectorImpl<int> &M, EVT VT,
                        unsigned &WhichResult) {
+  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
+  if (EltSz == 64)
+    return false;
+
   unsigned NumElts = VT.getVectorNumElements();
   WhichResult = (M[0] == 0 ? 0 : 1);
   for (unsigned i = 0; i < NumElts; i += 2) {
@@ -2391,6 +2398,10 @@
 
 static bool isVUZPMask(const SmallVectorImpl<int> &M, EVT VT,
                        unsigned &WhichResult) {
+  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
+  if (EltSz == 64)
+    return false;
+
   unsigned NumElts = VT.getVectorNumElements();
   WhichResult = (M[0] == 0 ? 0 : 1);
   for (unsigned i = 0; i != NumElts; ++i) {
@@ -2399,7 +2410,7 @@
   }
 
   // VUZP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32.
-  if (VT.is64BitVector() && VT.getVectorElementType().getSizeInBits() == 32)
+  if (VT.is64BitVector() && EltSz == 32)
     return false;
 
   return true;
@@ -2407,6 +2418,10 @@
 
 static bool isVZIPMask(const SmallVectorImpl<int> &M, EVT VT,
                        unsigned &WhichResult) {
+  unsigned EltSz = VT.getVectorElementType().getSizeInBits();
+  if (EltSz == 64)
+    return false;
+
   unsigned NumElts = VT.getVectorNumElements();
   WhichResult = (M[0] == 0 ? 0 : 1);
   unsigned Idx = WhichResult * NumElts / 2;
@@ -2418,7 +2433,7 @@
   }
 
   // VZIP.32 for 64-bit vectors is a pseudo-instruction alias for VTRN.32.
-  if (VT.is64BitVector() && VT.getVectorElementType().getSizeInBits() == 32)
+  if (VT.is64BitVector() && EltSz == 32)
     return false;
 
   return true;




From benny.kra at googlemail.com  Wed Oct 21 16:44:02 2009
From: benny.kra at googlemail.com (Benjamin Kramer)
Date: Wed, 21 Oct 2009 23:44:02 +0200
Subject: [llvm-commits] [llvm] r84753 - in /llvm/trunk/lib/Target/PIC16:
	AsmPrinter/PIC16AsmPrinter.cpp AsmPrinter/PIC16AsmPrinter.h
	Makefile PIC16ABINames.h PIC16Passes/ PIC16Passes/Makefile
	PIC16Passes/PIC16Overlay.cpp PIC16Passes/PIC16Overlay.h
	PIC16Section.cpp PIC16Section.h PIC16TargetObjectFile.cpp
	PIC16TargetObjectFile.h
In-Reply-To: <200910211042.n9LAgjTm010493@zion.cs.uiuc.edu>
References: <200910211042.n9LAgjTm010493@zion.cs.uiuc.edu>
Message-ID: <2DA24F6E-62F5-4E21-A434-E66B3432C745@gmail.com>


Am 21.10.2009 um 12:42 schrieb Sanjiv Gupta:

> Author: sgupta
> Date: Wed Oct 21 05:42:44 2009
> New Revision: 84753
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84753&view=rev
> Log:
> Add a pass to overlay pic16 data sections for function frame and  
> automatic
> variables. This pass can be invoked by llvm-ld or opt to traverse  
> over the call graph
> to detect what function frames and their automatic variables can be  
> overlaid.
> Currently this builds an archive , but needs to be changed to a  
> loadable module.
>
> Modified: llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp?rev=84753&r1=84752&r2=84753&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp  
> (original)
> +++ llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp Wed  
> Oct 21 05:42:44 2009
> @@ -53,6 +53,45 @@
>   return true;
> }
>
> +static int getFunctionColor(const Function *F) {
> +  if (F->hasSection()) {
> +    std::string Sectn = F->getSection();
> +    std::string StrToFind = "Overlay=";
> +    unsigned Pos = Sectn.find(StrToFind);
> +
> +    // Retreive the color number if the key is found.
> +    if (Pos != std::string::npos) {

GCC warns (64 bit osx):

PIC16AsmPrinter.cpp: In function ?int getFunctionColor(const  
llvm::Function*)?:
PIC16AsmPrinter.cpp:63: warning: comparison is always true due to  
limited range of data type

Please use std::string::size_type instead of unsigned.

> +      Pos += StrToFind.length();
> +      std::string Color = "";
> +      char c = Sectn.at(Pos);
> +      // A Color can only consist of digits.
> +      while (c >= '0' && c<= '9') {
> +        Color.append(1,c);
> +        Pos++;
> +        if (Pos >= Sectn.length())
> +          break;
> +        c = Sectn.at(Pos);
> +      }
> +      return atoi(Color.c_str());
> +    }
> +  }
> +
> +  // Color was not set for function, so return -1.
> +  return -1;
> +}



From dpatel at apple.com  Wed Oct 21 16:57:13 2009
From: dpatel at apple.com (Devang Patel)
Date: Wed, 21 Oct 2009 21:57:13 -0000
Subject: [llvm-commits] [llvm] r84786 - in /llvm/trunk:
 include/llvm/Metadata.h lib/VMCore/Metadata.cpp
Message-ID: <200910212157.n9LLvDQk011333@zion.cs.uiuc.edu>

Author: dpatel
Date: Wed Oct 21 16:57:13 2009
New Revision: 84786

URL: http://llvm.org/viewvc/llvm-project?rev=84786&view=rev
Log:
Use StringRef.

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84786&r1=84785&r2=84786&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Wed Oct 21 16:57:13 2009
@@ -287,14 +287,14 @@
 public:
   /// registerMDKind - Register a new metadata kind and return its ID.
   /// A metadata kind can be registered only once. 
-  unsigned registerMDKind(const char *Name);
+  unsigned registerMDKind(const StringRef Name);
 
   /// getMDKind - Return metadata kind. If the requested metadata kind
   /// is not registered then return 0.
-  unsigned getMDKind(const char *Name);
+  unsigned getMDKind(const StringRef Name) const;
 
   /// isValidName - Return true if Name is a valid custom metadata handler name.
-  bool isValidName(const char *Name);
+  static bool isValidName(const StringRef Name);
 
   /// getMD - Get the metadata of given kind attached to an Instruction.
   /// If the metadata is not found then return 0.

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84786&r1=84785&r2=84786&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 16:57:13 2009
@@ -250,7 +250,7 @@
 
 /// registerMDKind - Register a new metadata kind and return its ID.
 /// A metadata kind can be registered only once. 
-unsigned MetadataContext::registerMDKind(const char *Name) {
+unsigned MetadataContext::registerMDKind(const StringRef Name) {
   assert(isValidName(Name) && "Invalid custome metadata name!");
   unsigned Count = MDHandlerNames.size();
   assert(MDHandlerNames.count(Name) == 0 && "Already registered MDKind!");
@@ -258,7 +258,8 @@
 }
 
 /// isValidName - Return true if Name is a valid custom metadata handler name.
-bool MetadataContext::isValidName(const char *Name) {
+bool MetadataContext::isValidName(const StringRef MDName) {
+  const char *Name = MDName.data();
   if (!Name)
     return false;
 
@@ -280,8 +281,8 @@
 
 /// getMDKind - Return metadata kind. If the requested metadata kind
 /// is not registered then return 0.
-unsigned MetadataContext::getMDKind(const char *Name) {
-  StringMap<unsigned>::iterator I = MDHandlerNames.find(Name);
+unsigned MetadataContext::getMDKind(const StringRef Name) const {
+  StringMap<unsigned>::const_iterator I = MDHandlerNames.find(Name);
   if (I == MDHandlerNames.end()) {
     assert(isValidName(Name) && "Invalid custome metadata name!");
     return 0;




From benny.kra at googlemail.com  Wed Oct 21 17:55:52 2009
From: benny.kra at googlemail.com (Benjamin Kramer)
Date: Wed, 21 Oct 2009 22:55:52 -0000
Subject: [llvm-commits] [llvm] r84790 - /llvm/trunk/lib/VMCore/Metadata.cpp
Message-ID: <200910212255.n9LMtqjr013741@zion.cs.uiuc.edu>

Author: d0k
Date: Wed Oct 21 17:55:51 2009
New Revision: 84790

URL: http://llvm.org/viewvc/llvm-project?rev=84790&view=rev
Log:
Simplify code. No intended functionality change.

Modified:
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84790&r1=84789&r2=84790&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 17:55:51 2009
@@ -259,23 +259,17 @@
 
 /// isValidName - Return true if Name is a valid custom metadata handler name.
 bool MetadataContext::isValidName(const StringRef MDName) {
-  const char *Name = MDName.data();
-  if (!Name)
+  if (MDName.empty())
     return false;
 
-  if (!isalpha(*Name))
+  if (!isalpha(MDName[1]))
     return false;
 
-  unsigned Length = strlen(Name);  
-  unsigned Count = 1;
-  ++Name;
-  while (Name &&
-         (isalnum(*Name) || *Name == '_' || *Name == '-' || *Name == '.')) {
-    ++Name;
-    ++Count;
+  for (StringRef::iterator I = MDName.begin() + 1, E = MDName.end(); I != E;
+       ++I) {
+    if (!isalnum(*I) && *I != '_' && *I != '-' && *I != '.')
+        return false;
   }
-  if (Length != Count)
-    return false;
   return true;
 }
 




From grosbach at apple.com  Wed Oct 21 17:59:24 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Wed, 21 Oct 2009 22:59:24 -0000
Subject: [llvm-commits] [llvm] r84791 -
	/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
Message-ID: <200910212259.n9LMxOfe013902@zion.cs.uiuc.edu>

Author: grosbach
Date: Wed Oct 21 17:59:24 2009
New Revision: 84791

URL: http://llvm.org/viewvc/llvm-project?rev=84791&view=rev
Log:
Conditionalize ARM/T2 frame index post-scavenging while working out fixes
for a few bugs.

Modified:
    llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp

Modified: llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp?rev=84791&r1=84790&r2=84791&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp Wed Oct 21 17:59:24 2009
@@ -40,6 +40,11 @@
 using namespace llvm;
 
 static cl::opt<bool>
+ScavengeFrameIndexVals("arm-virtual-frame-index-vals", cl::Hidden,
+          cl::init(false),
+          cl::desc("Resolve frame index values via scavenging in PEI"));
+
+static cl::opt<bool>
 ReuseFrameIndexVals("arm-reuse-frame-index-vals", cl::Hidden, cl::init(false),
           cl::desc("Reuse repeated frame index values"));
 
@@ -941,7 +946,7 @@
 }
 bool ARMBaseRegisterInfo::
 requiresFrameIndexScavenging(const MachineFunction &MF) const {
-  return true;
+  return ScavengeFrameIndexVals;
 }
 
 // hasReservedCallFrame - Under normal circumstances, when a frame pointer is
@@ -1019,6 +1024,17 @@
   MBB.erase(I);
 }
 
+/// findScratchRegister - Find a 'free' ARM register. If register scavenger
+/// is not being used, R12 is available. Otherwise, try for a call-clobbered
+/// register first and then a spilled callee-saved register if that fails.
+static
+unsigned findScratchRegister(RegScavenger *RS, const TargetRegisterClass *RC,
+                             ARMFunctionInfo *AFI) {
+  unsigned Reg = RS ? RS->FindUnusedReg(RC) : (unsigned) ARM::R12;
+  assert(!AFI->isThumb1OnlyFunction());
+  return Reg;
+}
+
 unsigned
 ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
                                          int SPAdj, int *Value,
@@ -1082,8 +1098,19 @@
     // Must be addrmode4.
     MI.getOperand(i).ChangeToRegister(FrameReg, false, false, false);
   else {
-    ScratchReg = MF.getRegInfo().createVirtualRegister(ARM::GPRRegisterClass);
-    *Value = Offset;
+    if (!ScavengeFrameIndexVals) {
+      // Insert a set of r12 with the full address: r12 = sp + offset
+      // If the offset we have is too large to fit into the instruction, we need
+      // to form it with a series of ADDri's.  Do this by taking 8-bit chunks
+      // out of 'Offset'.
+      ScratchReg = findScratchRegister(RS, ARM::GPRRegisterClass, AFI);
+      if (ScratchReg == 0)
+        // No register is "free". Scavenge a register.
+        ScratchReg = RS->scavengeRegister(ARM::GPRRegisterClass, II, SPAdj);
+    } else {
+      ScratchReg = MF.getRegInfo().createVirtualRegister(ARM::GPRRegisterClass);
+      *Value = Offset;
+    }
     if (!AFI->isThumbFunction())
       emitARMRegPlusImmediate(MBB, II, MI.getDebugLoc(), ScratchReg, FrameReg,
                               Offset, Pred, PredReg, TII);
@@ -1093,7 +1120,7 @@
                              Offset, Pred, PredReg, TII);
     }
     MI.getOperand(i).ChangeToRegister(ScratchReg, false, false, true);
-    if (!ReuseFrameIndexVals)
+    if (!ReuseFrameIndexVals || !ScavengeFrameIndexVals)
       ScratchReg = 0;
   }
   return ScratchReg;




From grosbach at apple.com  Wed Oct 21 17:59:56 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Wed, 21 Oct 2009 22:59:56 -0000
Subject: [llvm-commits] [llvm] r84792 -
	/llvm/trunk/lib/CodeGen/RegisterScavenging.cpp
Message-ID: <200910212259.n9LMxuAO013925@zion.cs.uiuc.edu>

Author: grosbach
Date: Wed Oct 21 17:59:56 2009
New Revision: 84792

URL: http://llvm.org/viewvc/llvm-project?rev=84792&view=rev
Log:
The spill restore needs to be resolved to the SP/FP just like the spill

Modified:
    llvm/trunk/lib/CodeGen/RegisterScavenging.cpp

Modified: llvm/trunk/lib/CodeGen/RegisterScavenging.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/RegisterScavenging.cpp?rev=84792&r1=84791&r2=84792&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/RegisterScavenging.cpp (original)
+++ llvm/trunk/lib/CodeGen/RegisterScavenging.cpp Wed Oct 21 17:59:56 2009
@@ -310,6 +310,8 @@
 
     // Restore the scavenged register before its use (or first terminator).
     TII->loadRegFromStackSlot(*MBB, UseMI, SReg, ScavengingFrameIndex, RC);
+    II = prior(UseMI);
+    TRI->eliminateFrameIndex(II, SPAdj, NULL, this);
   }
 
   ScavengeRestore = prior(UseMI);




From jyasskin at google.com  Wed Oct 21 18:01:29 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Wed, 21 Oct 2009 16:01:29 -0700
Subject: [llvm-commits] [llvm] r84651 - in /llvm/trunk:
	include/llvm/ExecutionEngine/JITMemoryManager.h
	lib/ExecutionEngine/JIT/JITEmitter.cpp
	lib/ExecutionEngine/JIT/JITMemoryManager.cpp
	unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
In-Reply-To: <4ADF236F.1070905@lip6.fr>
References: <200910201813.n9KIDLbV011114@zion.cs.uiuc.edu>
	<4ADF236F.1070905@lip6.fr>
Message-ID: <bb55930e0910211601v7c83d6d7ue9286766dedc37ba@mail.gmail.com>

Oops!

You probably don't need the "if (DwarfExceptionHandling ||
JITEmitDebugInfo)", and it could probably also work to require the
deallocate functions to check for NULL (by analogy to free()).

But this patch is fine with me. Please check it in with or without the
above tweaks.

On Wed, Oct 21, 2009 at 8:06 AM, Nicolas Geoffray
<nicolas.geoffray at lip6.fr> wrote:
> Hi Jeffrey,
>
> There is a problem with your patch when the JIT runs out of memory, tries to
> deallocate the current buffer allocated for the function, and SIGSEGV
> because there is no buffer allocated for the exception table of the
> function.
>
> The following patch fixes the issue. OK to commit?
>
> Thanks,
> Nicolas
>
> Jeffrey Yasskin wrote:
>>
>> Author: jyasskin
>> Date: Tue Oct 20 13:13:21 2009
>> New Revision: 84651
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=84651&view=rev
>> Log:
>> Move the Function*->allocated blocks map from the JITMemoryManager to the
>> JITEmitter.
>>
>> I'm gradually making Functions auto-remove themselves from the JIT when
>> they're
>> destroyed. In this case, the Function needs to be removed from the
>> JITEmitter,
>> but the map recording which Functions need to be removed lived behind the
>> JITMemoryManager interface, which made things difficult.
>>
>> This patch replaces the deallocateMemForFunction(Function*) method with a
>> pair
>> of methods deallocateFunctionBody(void *) and
>> deallocateExceptionTable(void *)
>> corresponding to the two startFoo/endFoo pairs.
>>
>>
>> Modified:
>> ? ?llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h
>> ? ?llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp
>> ? ?llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
>> ? ?llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
>>
>> Modified: llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h?rev=84651&r1=84650&r2=84651&view=diff
>>
>>
>> ==============================================================================
>> --- llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h (original)
>> +++ llvm/trunk/include/llvm/ExecutionEngine/JITMemoryManager.h Tue Oct 20
>> 13:13:21 2009
>> @@ -132,9 +132,11 @@
>> ? ///
>> ? virtual uint8_t *allocateGlobal(uintptr_t Size, unsigned Alignment) = 0;
>> ?- ?/// deallocateMemForFunction - Free JIT memory for the specified
>> function.
>> - ?/// This is never called when the JIT is currently emitting a function.
>> - ?virtual void deallocateMemForFunction(const Function *F) = 0;
>> + ?/// deallocateFunctionBody - Free the specified function body. ?The
>> argument
>> + ?/// must be the return value from a call to startFunctionBody() that
>> hasn't
>> + ?/// been deallocated yet. ?This is never called when the JIT is
>> currently
>> + ?/// emitting a function.
>> + ?virtual void deallocateFunctionBody(void *Body) = 0;
>> ? ? /// startExceptionTable - When we finished JITing the function, if
>> exception
>> ? /// handling is set, we emit the exception table.
>> @@ -146,6 +148,12 @@
>> ? virtual void endExceptionTable(const Function *F, uint8_t *TableStart,
>> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?uint8_t *TableEnd, uint8_t*
>> FrameRegister) = 0;
>> ?+ ?/// deallocateExceptionTable - Free the specified exception table's
>> memory.
>> + ?/// The argument must be the return value from a call to
>> startExceptionTable()
>> + ?/// that hasn't been deallocated yet. ?This is never called when the
>> JIT is
>> + ?/// currently emitting an exception table.
>> + ?virtual void deallocateExceptionTable(void *ET) = 0;
>> +
>> ? /// CheckInvariants - For testing only. ?Return true if all internal
>> ? /// invariants are preserved, or return false and set ErrorStr to a
>> helpful
>> ? /// error message.
>>
>> Modified: llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp?rev=84651&r1=84650&r2=84651&view=diff
>>
>>
>> ==============================================================================
>> --- llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp (original)
>> +++ llvm/trunk/lib/ExecutionEngine/JIT/JITEmitter.cpp Tue Oct 20 13:13:21
>> 2009
>> @@ -42,6 +42,7 @@
>> ?#include "llvm/System/Disassembler.h"
>> ?#include "llvm/System/Memory.h"
>> ?#include "llvm/Target/TargetInstrInfo.h"
>> +#include "llvm/ADT/DenseMap.h"
>> ?#include "llvm/ADT/SmallPtrSet.h"
>> ?#include "llvm/ADT/SmallVector.h"
>> ?#include "llvm/ADT/Statistic.h"
>> @@ -548,6 +549,13 @@
>> ? ? /// finishFunction.
>> ? ? JITEvent_EmittedFunctionDetails EmissionDetails;
>> ?+ ? ?struct EmittedCode {
>> + ? ? ?void *FunctionBody;
>> + ? ? ?void *ExceptionTable;
>> + ? ? ?EmittedCode() : FunctionBody(0), ExceptionTable(0) {}
>> + ? ?};
>> + ? ?DenseMap<const Function *, EmittedCode> EmittedFunctions;
>> +
>> ? ? // CurFnStubUses - For a given Function, a vector of stubs that it
>> ? ? // references. ?This facilitates the JIT detecting that a stub is no
>> ? ? // longer used, so that it may be deallocated.
>> @@ -1011,7 +1019,8 @@
>> ? BufferBegin = CurBufferPtr = MemMgr->startFunctionBody(F.getFunction(),
>> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?ActualSize);
>> ? BufferEnd = BufferBegin+ActualSize;
>> - ?+ ?EmittedFunctions[F.getFunction()].FunctionBody = BufferBegin;
>> +
>> ? // Ensure the constant pool/jump table info is at least 4-byte aligned.
>> ? emitAlignment(16);
>> ?@@ -1201,6 +1210,7 @@
>> ? ? BufferBegin = CurBufferPtr =
>> MemMgr->startExceptionTable(F.getFunction(),
>> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?ActualSize);
>> ? ? BufferEnd = BufferBegin+ActualSize;
>> + ? ?EmittedFunctions[F.getFunction()].ExceptionTable = BufferBegin;
>> ? ? uint8_t *EhStart;
>> ? ? uint8_t *FrameRegister = DE->EmitDwarfTable(F, *this, FnStart, FnEnd,
>> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? EhStart);
>> @@ -1244,7 +1254,13 @@
>> ?/// deallocateMemForFunction - Deallocate all memory for the specified
>> ?/// function body. ?Also drop any references the function has to stubs.
>> ?void JITEmitter::deallocateMemForFunction(const Function *F) {
>> - ?MemMgr->deallocateMemForFunction(F);
>> + ?DenseMap<const Function *, EmittedCode>::iterator Emitted =
>> + ? ?EmittedFunctions.find(F);
>> + ?if (Emitted != EmittedFunctions.end()) {
>> + ? ?MemMgr->deallocateFunctionBody(Emitted->second.FunctionBody);
>> + ? ?MemMgr->deallocateExceptionTable(Emitted->second.ExceptionTable);
>> + ? ?EmittedFunctions.erase(Emitted);
>> + ?}
>> ? ? // TODO: Do we need to unregister exception handling information from
>> libgcc
>> ? // here?
>>
>> Modified: llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp?rev=84651&r1=84650&r2=84651&view=diff
>>
>>
>> ==============================================================================
>> --- llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp (original)
>> +++ llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp Tue Oct 20
>> 13:13:21 2009
>> @@ -297,9 +297,6 @@
>> ? ? ? uint8_t *GOTBase; ? ? // Target Specific reserved memory
>> ? ? void *DlsymTable; ? ? // Stub external symbol information
>> -
>> - ? ?std::map<const Function*, MemoryRangeHeader*> FunctionBlocks;
>> - ? ?std::map<const Function*, MemoryRangeHeader*> TableBlocks;
>> ? public:
>> ? ? DefaultJITMemoryManager();
>> ? ? ~DefaultJITMemoryManager();
>> @@ -414,7 +411,6 @@
>> ? ? ? ? ? ? ?"Mismatched function start/end!");
>> ? ? ? ? uintptr_t BlockSize = FunctionEnd - (uint8_t *)CurBlock;
>> - ? ? ?FunctionBlocks[F] = CurBlock;
>> ? ? ? ? // Release the memory at the end of this block that isn't needed.
>> ? ? ? FreeMemoryList =CurBlock->TrimAllocationToSize(FreeMemoryList,
>> BlockSize);
>> @@ -464,7 +460,6 @@
>> ? ? ? ? ? ? ?"Mismatched table start/end!");
>> ? ? ? ? ? ? uintptr_t BlockSize = TableEnd - (uint8_t *)CurBlock;
>> - ? ? ?TableBlocks[F] = CurBlock;
>> ? ? ? ? // Release the memory at the end of this block that isn't needed.
>> ? ? ? FreeMemoryList =CurBlock->TrimAllocationToSize(FreeMemoryList,
>> BlockSize);
>> @@ -478,15 +473,9 @@
>> ? ? ? return DlsymTable;
>> ? ? }
>> ? ? - ? ?/// deallocateMemForFunction - Deallocate all memory for the
>> specified
>> - ? ?/// function body.
>> - ? ?void deallocateMemForFunction(const Function *F) {
>> - ? ? ?std::map<const Function*, MemoryRangeHeader*>::iterator
>> - ? ? ? ?I = FunctionBlocks.find(F);
>> - ? ? ?if (I == FunctionBlocks.end()) return;
>> - ? ? ?+ ? ?void deallocateBlock(void *Block) {
>> ? ? ? // Find the block that is allocated for this function.
>> - ? ? ?MemoryRangeHeader *MemRange = I->second;
>> + ? ? ?MemoryRangeHeader *MemRange =
>> static_cast<MemoryRangeHeader*>(Block) - 1;
>> ? ? ? assert(MemRange->ThisAllocated && "Block isn't allocated!");
>> ? ? ? ? // Fill the buffer with garbage!
>> @@ -496,27 +485,18 @@
>> ? ? ? ? // Free the memory.
>> ? ? ? FreeMemoryList = MemRange->FreeBlock(FreeMemoryList);
>> - ? ? ?- ? ? ?// Finally, remove this entry from FunctionBlocks.
>> - ? ? ?FunctionBlocks.erase(I);
>> - ? ? ?- ? ? ?I = TableBlocks.find(F);
>> - ? ? ?if (I == TableBlocks.end()) return;
>> - ? ? ?- ? ? ?// Find the block that is allocated for this function.
>> - ? ? ?MemRange = I->second;
>> - ? ? ?assert(MemRange->ThisAllocated && "Block isn't allocated!");
>> + ? ?}
>> ?- ? ? ?// Fill the buffer with garbage!
>> - ? ? ?if (PoisonMemory) {
>> - ? ? ? ?memset(MemRange+1, 0xCD, MemRange->BlockSize-sizeof(*MemRange));
>> - ? ? ?}
>> + ? ?/// deallocateFunctionBody - Deallocate all memory for the specified
>> + ? ?/// function body.
>> + ? ?void deallocateFunctionBody(void *Body) {
>> + ? ? ?deallocateBlock(Body);
>> + ? ?}
>> ?- ? ? ?// Free the memory.
>> - ? ? ?FreeMemoryList = MemRange->FreeBlock(FreeMemoryList);
>> - ? ? ?- ? ? ?// Finally, remove this entry from TableBlocks.
>> - ? ? ?TableBlocks.erase(I);
>> + ? ?/// deallocateExceptionTable - Deallocate memory for the specified
>> + ? ?/// exception table.
>> + ? ?void deallocateExceptionTable(void *ET) {
>> + ? ? ?deallocateBlock(ET);
>> ? ? }
>> ? ? ? /// setMemoryWritable - When code generation is in progress,
>>
>> Modified:
>> llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp?rev=84651&r1=84650&r2=84651&view=diff
>>
>>
>> ==============================================================================
>> --- llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp
>> (original)
>> +++ llvm/trunk/unittests/ExecutionEngine/JIT/JITMemoryManagerTest.cpp Tue
>> Oct 20 13:13:21 2009
>> @@ -32,37 +32,36 @@
>> ? OwningPtr<JITMemoryManager> MemMgr(
>> ? ? ? JITMemoryManager::CreateDefaultMemManager());
>> ? uintptr_t size;
>> - ?uint8_t *start;
>> ? std::string Error;
>> ? ? // Allocate the functions.
>> ? OwningPtr<Function> F1(makeFakeFunction());
>> ? size = 1024;
>> - ?start = MemMgr->startFunctionBody(F1.get(), size);
>> - ?memset(start, 0xFF, 1024);
>> - ?MemMgr->endFunctionBody(F1.get(), start, start + 1024);
>> + ?uint8_t *FunctionBody1 = MemMgr->startFunctionBody(F1.get(), size);
>> + ?memset(FunctionBody1, 0xFF, 1024);
>> + ?MemMgr->endFunctionBody(F1.get(), FunctionBody1, FunctionBody1 + 1024);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ? ? OwningPtr<Function> F2(makeFakeFunction());
>> ? size = 1024;
>> - ?start = MemMgr->startFunctionBody(F2.get(), size);
>> - ?memset(start, 0xFF, 1024);
>> - ?MemMgr->endFunctionBody(F2.get(), start, start + 1024);
>> + ?uint8_t *FunctionBody2 = MemMgr->startFunctionBody(F2.get(), size);
>> + ?memset(FunctionBody2, 0xFF, 1024);
>> + ?MemMgr->endFunctionBody(F2.get(), FunctionBody2, FunctionBody2 + 1024);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ? ? OwningPtr<Function> F3(makeFakeFunction());
>> ? size = 1024;
>> - ?start = MemMgr->startFunctionBody(F3.get(), size);
>> - ?memset(start, 0xFF, 1024);
>> - ?MemMgr->endFunctionBody(F3.get(), start, start + 1024);
>> + ?uint8_t *FunctionBody3 = MemMgr->startFunctionBody(F3.get(), size);
>> + ?memset(FunctionBody3, 0xFF, 1024);
>> + ?MemMgr->endFunctionBody(F3.get(), FunctionBody3, FunctionBody3 + 1024);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ? ? // Deallocate them out of order, in case that matters.
>> - ?MemMgr->deallocateMemForFunction(F2.get());
>> + ?MemMgr->deallocateFunctionBody(FunctionBody2);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> - ?MemMgr->deallocateMemForFunction(F1.get());
>> + ?MemMgr->deallocateFunctionBody(FunctionBody1);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> - ?MemMgr->deallocateMemForFunction(F3.get());
>> + ?MemMgr->deallocateFunctionBody(FunctionBody3);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ?}
>> ?@@ -72,7 +71,6 @@
>> ? OwningPtr<JITMemoryManager> MemMgr(
>> ? ? ? JITMemoryManager::CreateDefaultMemManager());
>> ? uintptr_t size;
>> - ?uint8_t *start;
>> ? std::string Error;
>> ? ? // Big functions are a little less than the largest block size.
>> @@ -83,26 +81,26 @@
>> ? // Allocate big functions
>> ? OwningPtr<Function> F1(makeFakeFunction());
>> ? size = bigFuncSize;
>> - ?start = MemMgr->startFunctionBody(F1.get(), size);
>> + ?uint8_t *FunctionBody1 = MemMgr->startFunctionBody(F1.get(), size);
>> ? ASSERT_LE(bigFuncSize, size);
>> - ?memset(start, 0xFF, bigFuncSize);
>> - ?MemMgr->endFunctionBody(F1.get(), start, start + bigFuncSize);
>> + ?memset(FunctionBody1, 0xFF, bigFuncSize);
>> + ?MemMgr->endFunctionBody(F1.get(), FunctionBody1, FunctionBody1 +
>> bigFuncSize);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ? ? OwningPtr<Function> F2(makeFakeFunction());
>> ? size = bigFuncSize;
>> - ?start = MemMgr->startFunctionBody(F2.get(), size);
>> + ?uint8_t *FunctionBody2 = MemMgr->startFunctionBody(F2.get(), size);
>> ? ASSERT_LE(bigFuncSize, size);
>> - ?memset(start, 0xFF, bigFuncSize);
>> - ?MemMgr->endFunctionBody(F2.get(), start, start + bigFuncSize);
>> + ?memset(FunctionBody2, 0xFF, bigFuncSize);
>> + ?MemMgr->endFunctionBody(F2.get(), FunctionBody2, FunctionBody2 +
>> bigFuncSize);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ? ? OwningPtr<Function> F3(makeFakeFunction());
>> ? size = bigFuncSize;
>> - ?start = MemMgr->startFunctionBody(F3.get(), size);
>> + ?uint8_t *FunctionBody3 = MemMgr->startFunctionBody(F3.get(), size);
>> ? ASSERT_LE(bigFuncSize, size);
>> - ?memset(start, 0xFF, bigFuncSize);
>> - ?MemMgr->endFunctionBody(F3.get(), start, start + bigFuncSize);
>> + ?memset(FunctionBody3, 0xFF, bigFuncSize);
>> + ?MemMgr->endFunctionBody(F3.get(), FunctionBody3, FunctionBody3 +
>> bigFuncSize);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ? ? // Check that each large function took it's own slab.
>> @@ -111,43 +109,46 @@
>> ? // Allocate small functions
>> ? OwningPtr<Function> F4(makeFakeFunction());
>> ? size = smallFuncSize;
>> - ?start = MemMgr->startFunctionBody(F4.get(), size);
>> + ?uint8_t *FunctionBody4 = MemMgr->startFunctionBody(F4.get(), size);
>> ? ASSERT_LE(smallFuncSize, size);
>> - ?memset(start, 0xFF, smallFuncSize);
>> - ?MemMgr->endFunctionBody(F4.get(), start, start + smallFuncSize);
>> + ?memset(FunctionBody4, 0xFF, smallFuncSize);
>> + ?MemMgr->endFunctionBody(F4.get(), FunctionBody4,
>> + ? ? ? ? ? ? ? ? ? ? ? ? ?FunctionBody4 + smallFuncSize);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ? ? OwningPtr<Function> F5(makeFakeFunction());
>> ? size = smallFuncSize;
>> - ?start = MemMgr->startFunctionBody(F5.get(), size);
>> + ?uint8_t *FunctionBody5 = MemMgr->startFunctionBody(F5.get(), size);
>> ? ASSERT_LE(smallFuncSize, size);
>> - ?memset(start, 0xFF, smallFuncSize);
>> - ?MemMgr->endFunctionBody(F5.get(), start, start + smallFuncSize);
>> + ?memset(FunctionBody5, 0xFF, smallFuncSize);
>> + ?MemMgr->endFunctionBody(F5.get(), FunctionBody5,
>> + ? ? ? ? ? ? ? ? ? ? ? ? ?FunctionBody5 + smallFuncSize);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ? ? OwningPtr<Function> F6(makeFakeFunction());
>> ? size = smallFuncSize;
>> - ?start = MemMgr->startFunctionBody(F6.get(), size);
>> + ?uint8_t *FunctionBody6 = MemMgr->startFunctionBody(F6.get(), size);
>> ? ASSERT_LE(smallFuncSize, size);
>> - ?memset(start, 0xFF, smallFuncSize);
>> - ?MemMgr->endFunctionBody(F6.get(), start, start + smallFuncSize);
>> + ?memset(FunctionBody6, 0xFF, smallFuncSize);
>> + ?MemMgr->endFunctionBody(F6.get(), FunctionBody6,
>> + ? ? ? ? ? ? ? ? ? ? ? ? ?FunctionBody6 + smallFuncSize);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ? ? // Check that the small functions didn't allocate any new slabs.
>> ? EXPECT_EQ(3U, MemMgr->GetNumCodeSlabs());
>> ? ? // Deallocate them out of order, in case that matters.
>> - ?MemMgr->deallocateMemForFunction(F2.get());
>> + ?MemMgr->deallocateFunctionBody(FunctionBody2);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> - ?MemMgr->deallocateMemForFunction(F1.get());
>> + ?MemMgr->deallocateFunctionBody(FunctionBody1);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> - ?MemMgr->deallocateMemForFunction(F4.get());
>> + ?MemMgr->deallocateFunctionBody(FunctionBody4);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> - ?MemMgr->deallocateMemForFunction(F3.get());
>> + ?MemMgr->deallocateFunctionBody(FunctionBody3);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> - ?MemMgr->deallocateMemForFunction(F5.get());
>> + ?MemMgr->deallocateFunctionBody(FunctionBody5);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> - ?MemMgr->deallocateMemForFunction(F6.get());
>> + ?MemMgr->deallocateFunctionBody(FunctionBody6);
>> ? EXPECT_TRUE(MemMgr->CheckInvariants(Error)) << Error;
>> ?}
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>
>



From clattner at apple.com  Wed Oct 21 18:17:24 2009
From: clattner at apple.com (Chris Lattner)
Date: Wed, 21 Oct 2009 16:17:24 -0700
Subject: [llvm-commits] [llvm] r84778 - in /llvm/trunk:
	lib/Target/ARM/ARMAddressingModes.h lib/Target/ARM/ARMInstrInfo.td
	lib/Target/ARM/ARMInstrThumb2.td test/CodeGen/Thumb2/thumb2-mov.ll
	test/CodeGen/Thumb2/thumb2-mov2.ll
In-Reply-To: <200910212044.n9LKia3w008119@zion.cs.uiuc.edu>
References: <200910212044.n9LKia3w008119@zion.cs.uiuc.edu>
Message-ID: <18349C55-1D53-4269-89A4-47DEFC9BFDFC@apple.com>


On Oct 21, 2009, at 1:44 PM, Jim Grosbach wrote:

> Author: grosbach
> Date: Wed Oct 21 15:44:34 2009
> New Revision: 84778
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84778&view=rev
> Log:
> Improve handling of immediates by splitting 32-bit immediates into  
> two 16-bit
> immediate operands when they will fit into the using instruction.

Nice

> -define i32 @t2_const_var2_1_fail_1(i32 %lhs) {
> -;CHECK: t2_const_var2_1_fail_1:
> -;CHECK: movt
> +define i32 @t2_const_var2_1_ok_2(i32 %lhs) {
> +;CHECK: t2_const_var2_1_ok_2:
> +;CHECK: #11206656
> +;CHECK: #187

Please add the expected instructions opcodes too, there is no reason  
to be this "fuzzy".

-Chris

>     %ret = add i32 %lhs, 11206843 ; 0x00ab00bb
>     ret i32 %ret
> }
>
> -define i32 @t2_const_var2_1_fail_2(i32 %lhs) {
> -;CHECK: t2_const_var2_1_fail_2:
> -;CHECK: movt
> +define i32 @t2_const_var2_1_ok_3(i32 %lhs) {
> +;CHECK: t2_const_var2_1_ok_3:
> +;CHECK: #11206827
> +;CHECK: #16777216
>     %ret = add i32 %lhs, 27984043 ; 0x01ab00ab
>     ret i32 %ret
> }
>
> -define i32 @t2_const_var2_1_fail_3(i32 %lhs) {
> -;CHECK: t2_const_var2_1_fail_3:
> -;CHECK: movt
> +define i32 @t2_const_var2_1_ok_4(i32 %lhs) {
> +;CHECK: t2_const_var2_1_ok_4:
> +;CHECK: #16777472
> +;CHECK: #11206827
>     %ret = add i32 %lhs, 27984299 ; 0x01ab01ab
>     ret i32 %ret
> }
>
> -define i32 @t2_const_var2_1_fail_4(i32 %lhs) {
> -;CHECK: t2_const_var2_1_fail_4:
> +define i32 @t2_const_var2_1_fail_1(i32 %lhs) {
> +;CHECK: t2_const_var2_1_fail_1:
> ;CHECK: movt
>     %ret = add i32 %lhs, 28027649 ; 0x01abab01
>     ret i32 %ret
> @@ -46,29 +49,31 @@
>     ret i32 %ret
> }
>
> -define i32 @t2_const_var2_2_fail_1(i32 %lhs) {
> -;CHECK: t2_const_var2_2_fail_1:
> -;CHECK: movt
> +define i32 @t2_const_var2_2_ok_2(i32 %lhs) {
> +;CHECK: t2_const_var2_2_ok_2:
> +;CHECK: #-1426063360
> +;CHECK: #47616
>     %ret = add i32 %lhs, 2868951552 ; 0xab00ba00
>     ret i32 %ret
> }
>
> -define i32 @t2_const_var2_2_fail_2(i32 %lhs) {
> -;CHECK: t2_const_var2_2_fail_2:
> -;CHECK: movt
> +define i32 @t2_const_var2_2_ok_3(i32 %lhs) {
> +;CHECK: t2_const_var2_2_ok_3:
> +;CHECK: #-1426019584
>     %ret = add i32 %lhs, 2868947728 ; 0xab00ab10
>     ret i32 %ret
> }
>
> -define i32 @t2_const_var2_2_fail_3(i32 %lhs) {
> -;CHECK: t2_const_var2_2_fail_3:
> -;CHECK: movt
> +define i32 @t2_const_var2_2_ok_4(i32 %lhs) {
> +;CHECK: t2_const_var2_2_ok_4:
> +;CHECK: #-1426019584
> +;CHECK: #1048592
>     %ret = add i32 %lhs, 2869996304 ; 0xab10ab10
>     ret i32 %ret
> }
>
> -define i32 @t2_const_var2_2_fail_4(i32 %lhs) {
> -;CHECK: t2_const_var2_2_fail_4:
> +define i32 @t2_const_var2_2_fail_1(i32 %lhs) {
> +;CHECK: t2_const_var2_2_fail_1:
> ;CHECK: movt
>     %ret = add i32 %lhs, 279685904 ; 0x10abab10
>     ret i32 %ret
> @@ -125,9 +130,10 @@
>     ret i32 %ret
> }
>
> -define i32 @t2_const_var3_2_fail_1(i32 %lhs) {
> -;CHECK: t2_const_var3_2_fail_1:
> -;CHECK: movt
> +define i32 @t2_const_var3_2_ok_2(i32 %lhs) {
> +;CHECK: t2_const_var3_2_ok_2:
> +;CHECK: #2097152
> +;CHECK: #1843200
>     %ret = add i32 %lhs, 3940352 ; 0b00000000001111000010000000000000
>     ret i32 %ret
> }
>
> Modified: llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll?rev=84778&r1=84777&r2=84778&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll (original)
> +++ llvm/trunk/test/CodeGen/Thumb2/thumb2-mov2.ll Wed Oct 21  
> 15:44:34 2009
> @@ -55,10 +55,10 @@
>
> define i32 @t2MOVTi16_test_nomatch_1(i32 %a) {
> ; CHECK: t2MOVTi16_test_nomatch_1:
> -; CHECK:      movw r1, #16384
> -; CHECK-NEXT: movt r1, #154
> +; CHECK:      #8388608
> ; CHECK:      movw r1, #65535
> ; CHECK-NEXT: movt r1, #154
> +; CHECK:      #1720320
>     %1 = shl i32  255,   8
>     %2 = shl i32 1234,   8
>     %3 = or  i32   %1, 255  ; This give us 0xFFFF in %3
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From clattner at apple.com  Wed Oct 21 18:22:01 2009
From: clattner at apple.com (Chris Lattner)
Date: Wed, 21 Oct 2009 16:22:01 -0700
Subject: [llvm-commits] [llvm] r84790 -
	/llvm/trunk/lib/VMCore/Metadata.cpp
In-Reply-To: <200910212255.n9LMtqjr013741@zion.cs.uiuc.edu>
References: <200910212255.n9LMtqjr013741@zion.cs.uiuc.edu>
Message-ID: <A18EE01F-9D83-415E-A753-B035AED15E59@apple.com>


On Oct 21, 2009, at 3:55 PM, Benjamin Kramer wrote:

> Author: d0k
> Date: Wed Oct 21 17:55:51 2009
> New Revision: 84790
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84790&view=rev
> Log:
> Simplify code. No intended functionality change.

Thanks Benjamin!
> +++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 17:55:51 2009
> @@ -259,23 +259,17 @@
>
> /// isValidName - Return true if Name is a valid custom metadata  
> handler name.
> bool MetadataContext::isValidName(const StringRef MDName) {
> -  const char *Name = MDName.data();
> -  if (!Name)
> +  if (MDName.empty())
>     return false;
>
> -  if (!isalpha(*Name))
> +  if (!isalpha(MDName[1]))

I think this should be [0] not [1].

-Chris

>     return false;
>
> -  unsigned Length = strlen(Name);
> -  unsigned Count = 1;
> -  ++Name;
> -  while (Name &&
> -         (isalnum(*Name) || *Name == '_' || *Name == '-' || *Name  
> == '.')) {
> -    ++Name;
> -    ++Count;
> +  for (StringRef::iterator I = MDName.begin() + 1, E =  
> MDName.end(); I != E;
> +       ++I) {
> +    if (!isalnum(*I) && *I != '_' && *I != '-' && *I != '.')
> +        return false;
>   }
> -  if (Length != Count)
> -    return false;
>   return true;
> }
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From benny.kra at googlemail.com  Wed Oct 21 18:27:54 2009
From: benny.kra at googlemail.com (Benjamin Kramer)
Date: Wed, 21 Oct 2009 23:27:54 -0000
Subject: [llvm-commits] [llvm] r84794 - /llvm/trunk/lib/VMCore/Metadata.cpp
Message-ID: <200910212327.n9LNRsq5014899@zion.cs.uiuc.edu>

Author: d0k
Date: Wed Oct 21 18:27:54 2009
New Revision: 84794

URL: http://llvm.org/viewvc/llvm-project?rev=84794&view=rev
Log:
Remove pointless return; at end of function.

Modified:
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84794&r1=84793&r2=84794&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 18:27:54 2009
@@ -307,7 +307,6 @@
 
   // Otherwise add a new entry.
   Info.push_back(std::make_pair(MDKind, Node));
-  return;
 }
 
 /// removeMD - Remove metadata of given kind attached with an instuction.
@@ -324,8 +323,6 @@
       return;
     }
   }
-
-  return;
 }
   
 /// removeAllMetadata - Remove all metadata attached with an instruction.




From dalej at apple.com  Wed Oct 21 18:28:00 2009
From: dalej at apple.com (Dale Johannesen)
Date: Wed, 21 Oct 2009 23:28:00 -0000
Subject: [llvm-commits] [llvm] r84795 - in /llvm/trunk: docs/LangRef.html
 include/llvm/InlineAsm.h lib/AsmParser/LLLexer.cpp
 lib/AsmParser/LLParser.cpp lib/AsmParser/LLToken.h
 lib/Bitcode/Reader/BitcodeReader.cpp lib/Bitcode/Writer/BitcodeWriter.cpp
 lib/VMCore/AsmWriter.cpp lib/VMCore/Core.cpp lib/VMCore/InlineAsm.cpp
Message-ID: <200910212328.n9LNS1ng014924@zion.cs.uiuc.edu>

Author: johannes
Date: Wed Oct 21 18:28:00 2009
New Revision: 84795

URL: http://llvm.org/viewvc/llvm-project?rev=84795&view=rev
Log:
Rename msasm to alignstack per review.


Modified:
    llvm/trunk/docs/LangRef.html
    llvm/trunk/include/llvm/InlineAsm.h
    llvm/trunk/lib/AsmParser/LLLexer.cpp
    llvm/trunk/lib/AsmParser/LLParser.cpp
    llvm/trunk/lib/AsmParser/LLToken.h
    llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp
    llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
    llvm/trunk/lib/VMCore/AsmWriter.cpp
    llvm/trunk/lib/VMCore/Core.cpp
    llvm/trunk/lib/VMCore/InlineAsm.cpp

Modified: llvm/trunk/docs/LangRef.html
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/LangRef.html?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/docs/LangRef.html (original)
+++ llvm/trunk/docs/LangRef.html Wed Oct 21 18:28:00 2009
@@ -2339,9 +2339,9 @@
    a special value.  This value represents the inline assembler as a string
    (containing the instructions to emit), a list of operand constraints (stored
    as a string), a flag that indicates whether or not the inline asm
-   expression has side effects, and a flag indicating whether the asm came
-   originally from an asm block.  An example inline assembler
-   expression is:</p>
+   expression has side effects, and a flag indicating whether the function
+   containing the asm needs to align its stack conservatively.  An example
+   inline assembler expression is:</p>
 
 <div class="doc_code">
 <pre>
@@ -2369,12 +2369,16 @@
 </pre>
 </div>
 
-<p>Inline asms derived from asm blocks are similarly marked with the
-   '<tt>msasm</tt>' keyword:</p>
+<p>In some cases inline asms will contain code that will not work unless the
+   stack is aligned in some way, such as calls or SSE instructions on x86,
+   yet will not contain code that does that alignment within the asm.
+   The compiler should make conservative assumptions about what the asm might
+   contain and should generate its usual stack alignment code in the prologue
+   if the '<tt>alignstack</tt>' keyword is present:</p>
 
 <div class="doc_code">
 <pre>
-call void asm msasm "eieio", ""()
+call void asm alignstack "eieio", ""()
 </pre>
 </div>
 

Modified: llvm/trunk/include/llvm/InlineAsm.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/InlineAsm.h?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/include/llvm/InlineAsm.h (original)
+++ llvm/trunk/include/llvm/InlineAsm.h Wed Oct 21 18:28:00 2009
@@ -31,11 +31,11 @@
 
   std::string AsmString, Constraints;
   bool HasSideEffects;
-  bool IsMsAsm;
+  bool IsAlignStack;
   
   InlineAsm(const FunctionType *Ty, const StringRef &AsmString,
             const StringRef &Constraints, bool hasSideEffects,
-            bool isMsAsm = false);
+            bool isAlignStack = false);
   virtual ~InlineAsm();
 public:
 
@@ -43,10 +43,10 @@
   ///
   static InlineAsm *get(const FunctionType *Ty, const StringRef &AsmString,
                         const StringRef &Constraints, bool hasSideEffects,
-                        bool isMsAsm = false);
+                        bool isAlignStack = false);
   
   bool hasSideEffects() const { return HasSideEffects; }
-  bool isMsAsm() const { return IsMsAsm; }
+  bool isAlignStack() const { return IsAlignStack; }
   
   /// getType - InlineAsm's are always pointers.
   ///

Modified: llvm/trunk/lib/AsmParser/LLLexer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/AsmParser/LLLexer.cpp?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/lib/AsmParser/LLLexer.cpp (original)
+++ llvm/trunk/lib/AsmParser/LLLexer.cpp Wed Oct 21 18:28:00 2009
@@ -529,7 +529,7 @@
   KEYWORD(module);
   KEYWORD(asm);
   KEYWORD(sideeffect);
-  KEYWORD(msasm);
+  KEYWORD(alignstack);
   KEYWORD(gc);
 
   KEYWORD(ccc);

Modified: llvm/trunk/lib/AsmParser/LLParser.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/AsmParser/LLParser.cpp?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/lib/AsmParser/LLParser.cpp (original)
+++ llvm/trunk/lib/AsmParser/LLParser.cpp Wed Oct 21 18:28:00 2009
@@ -1974,17 +1974,17 @@
     return false;
 
   case lltok::kw_asm: {
-    // ValID ::= 'asm' SideEffect? MsAsm? STRINGCONSTANT ',' STRINGCONSTANT
-    bool HasSideEffect, MsAsm;
+    // ValID ::= 'asm' SideEffect? AlignStack? STRINGCONSTANT ',' STRINGCONSTANT
+    bool HasSideEffect, AlignStack;
     Lex.Lex();
     if (ParseOptionalToken(lltok::kw_sideeffect, HasSideEffect) ||
-        ParseOptionalToken(lltok::kw_msasm, MsAsm) ||
+        ParseOptionalToken(lltok::kw_alignstack, AlignStack) ||
         ParseStringConstant(ID.StrVal) ||
         ParseToken(lltok::comma, "expected comma in inline asm expression") ||
         ParseToken(lltok::StringConstant, "expected constraint string"))
       return true;
     ID.StrVal2 = Lex.getStrVal();
-    ID.UIntVal = HasSideEffect | ((unsigned)MsAsm<<1);
+    ID.UIntVal = HasSideEffect | ((unsigned)AlignStack<<1);
     ID.Kind = ValID::t_InlineAsm;
     return false;
   }

Modified: llvm/trunk/lib/AsmParser/LLToken.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/AsmParser/LLToken.h?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/lib/AsmParser/LLToken.h (original)
+++ llvm/trunk/lib/AsmParser/LLToken.h Wed Oct 21 18:28:00 2009
@@ -62,7 +62,7 @@
     kw_module,
     kw_asm,
     kw_sideeffect,
-    kw_msasm,
+    kw_alignstack,
     kw_gc,
     kw_c,
 

Modified: llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp (original)
+++ llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp Wed Oct 21 18:28:00 2009
@@ -1165,7 +1165,7 @@
       if (Record.size() < 2) return Error("Invalid INLINEASM record");
       std::string AsmStr, ConstrStr;
       bool HasSideEffects = Record[0] & 1;
-      bool IsMsAsm = Record[0] >> 1;
+      bool IsAlignStack = Record[0] >> 1;
       unsigned AsmStrSize = Record[1];
       if (2+AsmStrSize >= Record.size())
         return Error("Invalid INLINEASM record");
@@ -1179,7 +1179,7 @@
         ConstrStr += (char)Record[3+AsmStrSize+i];
       const PointerType *PTy = cast<PointerType>(CurTy);
       V = InlineAsm::get(cast<FunctionType>(PTy->getElementType()),
-                         AsmStr, ConstrStr, HasSideEffects, IsMsAsm);
+                         AsmStr, ConstrStr, HasSideEffects, IsAlignStack);
       break;
     }
     }

Modified: llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp (original)
+++ llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp Wed Oct 21 18:28:00 2009
@@ -678,7 +678,7 @@
 
     if (const InlineAsm *IA = dyn_cast<InlineAsm>(V)) {
       Record.push_back(unsigned(IA->hasSideEffects()) |
-                       unsigned(IA->isMsAsm()) << 1);
+                       unsigned(IA->isAlignStack()) << 1);
 
       // Add the asm string.
       const std::string &AsmStr = IA->getAsmString();

Modified: llvm/trunk/lib/VMCore/AsmWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/AsmWriter.cpp?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/AsmWriter.cpp (original)
+++ llvm/trunk/lib/VMCore/AsmWriter.cpp Wed Oct 21 18:28:00 2009
@@ -1205,8 +1205,8 @@
     Out << "asm ";
     if (IA->hasSideEffects())
       Out << "sideeffect ";
-    if (IA->isMsAsm())
-      Out << "msasm ";
+    if (IA->isAlignStack())
+      Out << "alignstack ";
     Out << '"';
     PrintEscapedString(IA->getAsmString(), Out);
     Out << "\", \"";

Modified: llvm/trunk/lib/VMCore/Core.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Core.cpp?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Core.cpp (original)
+++ llvm/trunk/lib/VMCore/Core.cpp Wed Oct 21 18:28:00 2009
@@ -885,9 +885,9 @@
 
 LLVMValueRef LLVMConstInlineAsm(LLVMTypeRef Ty, const char *AsmString, 
                                 const char *Constraints, int HasSideEffects,
-                                int IsMsAsm) {
+                                int IsAlignStack) {
   return wrap(InlineAsm::get(dyn_cast<FunctionType>(unwrap(Ty)), AsmString, 
-                             Constraints, HasSideEffects, IsMsAsm));
+                             Constraints, HasSideEffects, IsAlignStack));
 }
 
 /*--.. Operations on global variables, functions, and aliases (globals) ....--*/

Modified: llvm/trunk/lib/VMCore/InlineAsm.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/InlineAsm.cpp?rev=84795&r1=84794&r2=84795&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/InlineAsm.cpp (original)
+++ llvm/trunk/lib/VMCore/InlineAsm.cpp Wed Oct 21 18:28:00 2009
@@ -28,18 +28,20 @@
 
 InlineAsm *InlineAsm::get(const FunctionType *Ty, const StringRef &AsmString,
                           const StringRef &Constraints, bool hasSideEffects,
-                          bool isMsAsm) {
+                          bool isAlignStack) {
   // FIXME: memoize!
-  return new InlineAsm(Ty, AsmString, Constraints, hasSideEffects, isMsAsm);  
+  return new InlineAsm(Ty, AsmString, Constraints, hasSideEffects, 
+                       isAlignStack);
 }
 
 InlineAsm::InlineAsm(const FunctionType *Ty, const StringRef &asmString,
                      const StringRef &constraints, bool hasSideEffects,
-                     bool isMsAsm)
+                     bool isAlignStack)
   : Value(PointerType::getUnqual(Ty), 
           Value::InlineAsmVal), 
     AsmString(asmString), 
-    Constraints(constraints), HasSideEffects(hasSideEffects), IsMsAsm(isMsAsm) {
+    Constraints(constraints), HasSideEffects(hasSideEffects), 
+    IsAlignStack(isAlignStack) {
 
   // Do various checks on the constraint string and type.
   assert(Verify(Ty, constraints) && "Function type not legal for constraints!");




From dalej at apple.com  Wed Oct 21 18:29:12 2009
From: dalej at apple.com (Dale Johannesen)
Date: Wed, 21 Oct 2009 23:29:12 -0000
Subject: [llvm-commits] [llvm] r84796 - in /llvm/trunk/test:
 Assembler/alignstack.ll Assembler/msasm.ll FrontendC++/alignstack.cpp
 FrontendC++/msasm.cpp FrontendC/alignstack.c FrontendC/msasm.c
Message-ID: <200910212329.n9LNTD8J015007@zion.cs.uiuc.edu>

Author: johannes
Date: Wed Oct 21 18:29:12 2009
New Revision: 84796

URL: http://llvm.org/viewvc/llvm-project?rev=84796&view=rev
Log:
Adjust testcases for msasm -> alignstack.


Added:
    llvm/trunk/test/Assembler/alignstack.ll
    llvm/trunk/test/FrontendC++/alignstack.cpp
    llvm/trunk/test/FrontendC/alignstack.c
Removed:
    llvm/trunk/test/Assembler/msasm.ll
    llvm/trunk/test/FrontendC++/msasm.cpp
    llvm/trunk/test/FrontendC/msasm.c

Added: llvm/trunk/test/Assembler/alignstack.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Assembler/alignstack.ll?rev=84796&view=auto

==============================================================================
--- llvm/trunk/test/Assembler/alignstack.ll (added)
+++ llvm/trunk/test/Assembler/alignstack.ll Wed Oct 21 18:29:12 2009
@@ -0,0 +1,36 @@
+; RUN: llvm-as < %s | llvm-dis | FileCheck %s
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target triple = "i386-apple-darwin10.0"
+
+define void @test1() nounwind {
+; CHECK: test1
+; CHECK: sideeffect
+; CHECK-NOT: alignstack
+	tail call void asm sideeffect "mov", "~{dirflag},~{fpsr},~{flags}"() nounwind
+	ret void
+; CHECK: ret
+}
+define void @test2() nounwind {
+; CHECK: test2
+; CHECK: sideeffect
+; CHECK: alignstack
+	tail call void asm sideeffect alignstack "mov", "~{dirflag},~{fpsr},~{flags}"() nounwind
+	ret void
+; CHECK: ret
+}
+define void @test3() nounwind {
+; CHECK: test3
+; CHECK-NOT: sideeffect
+; CHECK: alignstack
+	tail call void asm alignstack "mov", "~{dirflag},~{fpsr},~{flags}"() nounwind
+	ret void
+; CHECK: ret
+}
+define void @test4() nounwind {
+; CHECK: test4
+; CHECK-NOT: sideeffect
+; CHECK-NOT: alignstack
+	tail call void asm  "mov", "~{dirflag},~{fpsr},~{flags}"() nounwind
+	ret void
+; CHECK: ret
+}

Removed: llvm/trunk/test/Assembler/msasm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Assembler/msasm.ll?rev=84795&view=auto

==============================================================================
--- llvm/trunk/test/Assembler/msasm.ll (original)
+++ llvm/trunk/test/Assembler/msasm.ll (removed)
@@ -1,36 +0,0 @@
-; RUN: llvm-as < %s | llvm-dis | FileCheck %s
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
-target triple = "i386-apple-darwin10.0"
-
-define void @test1() nounwind {
-; CHECK: test1
-; CHECK: sideeffect
-; CHECK-NOT: msasm
-	tail call void asm sideeffect "mov", "~{dirflag},~{fpsr},~{flags}"() nounwind
-	ret void
-; CHECK: ret
-}
-define void @test2() nounwind {
-; CHECK: test2
-; CHECK: sideeffect
-; CHECK: msasm
-	tail call void asm sideeffect msasm "mov", "~{dirflag},~{fpsr},~{flags}"() nounwind
-	ret void
-; CHECK: ret
-}
-define void @test3() nounwind {
-; CHECK: test3
-; CHECK-NOT: sideeffect
-; CHECK: msasm
-	tail call void asm msasm "mov", "~{dirflag},~{fpsr},~{flags}"() nounwind
-	ret void
-; CHECK: ret
-}
-define void @test4() nounwind {
-; CHECK: test4
-; CHECK-NOT: sideeffect
-; CHECK-NOT: msasm
-	tail call void asm  "mov", "~{dirflag},~{fpsr},~{flags}"() nounwind
-	ret void
-; CHECK: ret
-}

Added: llvm/trunk/test/FrontendC++/alignstack.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC%2B%2B/alignstack.cpp?rev=84796&view=auto

==============================================================================
--- llvm/trunk/test/FrontendC++/alignstack.cpp (added)
+++ llvm/trunk/test/FrontendC++/alignstack.cpp Wed Oct 21 18:29:12 2009
@@ -0,0 +1,23 @@
+// RUN: %llvmgxx %s -fasm-blocks -S -o - | FileCheck %s
+// Complicated expression as jump target
+// XFAIL: *
+// XTARGET: x86,i386,i686
+
+void Method3()
+{
+// CHECK: Method3
+// CHECK-NOT: alignstack
+    asm("foo:");
+// CHECK: return
+}
+
+void Method4()
+{
+// CHECK: Method4
+// CHECK: alignstack
+  asm {
+    bar:
+  }
+// CHECK: return
+}
+

Removed: llvm/trunk/test/FrontendC++/msasm.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC%2B%2B/msasm.cpp?rev=84795&view=auto

==============================================================================
--- llvm/trunk/test/FrontendC++/msasm.cpp (original)
+++ llvm/trunk/test/FrontendC++/msasm.cpp (removed)
@@ -1,23 +0,0 @@
-// RUN: %llvmgxx %s -fasm-blocks -S -o - | FileCheck %s
-// Complicated expression as jump target
-// XFAIL: *
-// XTARGET: x86,i386,i686
-
-void Method3()
-{
-// CHECK: Method3
-// CHECK-NOT: msasm
-    asm("foo:");
-// CHECK: return
-}
-
-void Method4()
-{
-// CHECK: Method4
-// CHECK: msasm
-  asm {
-    bar:
-  }
-// CHECK: return
-}
-

Added: llvm/trunk/test/FrontendC/alignstack.c
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC/alignstack.c?rev=84796&view=auto

==============================================================================
--- llvm/trunk/test/FrontendC/alignstack.c (added)
+++ llvm/trunk/test/FrontendC/alignstack.c Wed Oct 21 18:29:12 2009
@@ -0,0 +1,23 @@
+// RUN: %llvmgcc %s -fasm-blocks -S -o - | FileCheck %s
+// Complicated expression as jump target
+// XFAIL: *
+// XTARGET: x86,i386,i686,darwin
+
+void Method3()
+{
+// CHECK: Method3
+// CHECK-NOT: alignstack
+    asm("foo:");
+// CHECK: return
+}
+
+void Method4()
+{
+// CHECK: Method4
+// CHECK: alignstack
+  asm {
+    bar:
+  }
+// CHECK: return
+}
+

Removed: llvm/trunk/test/FrontendC/msasm.c
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC/msasm.c?rev=84795&view=auto

==============================================================================
--- llvm/trunk/test/FrontendC/msasm.c (original)
+++ llvm/trunk/test/FrontendC/msasm.c (removed)
@@ -1,23 +0,0 @@
-// RUN: %llvmgcc %s -fasm-blocks -S -o - | FileCheck %s
-// Complicated expression as jump target
-// XFAIL: *
-// XTARGET: x86,i386,i686,darwin
-
-void Method3()
-{
-// CHECK: Method3
-// CHECK-NOT: msasm
-    asm("foo:");
-// CHECK: return
-}
-
-void Method4()
-{
-// CHECK: Method4
-// CHECK: msasm
-  asm {
-    bar:
-  }
-// CHECK: return
-}
-




From benny.kra at googlemail.com  Wed Oct 21 18:29:32 2009
From: benny.kra at googlemail.com (Benjamin Kramer)
Date: Wed, 21 Oct 2009 23:29:32 -0000
Subject: [llvm-commits] [llvm] r84797 - /llvm/trunk/lib/VMCore/Metadata.cpp
Message-ID: <200910212329.n9LNTWHd015027@zion.cs.uiuc.edu>

Author: d0k
Date: Wed Oct 21 18:29:32 2009
New Revision: 84797

URL: http://llvm.org/viewvc/llvm-project?rev=84797&view=rev
Log:
Fix thinko noticed by Chris.

Modified:
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84797&r1=84796&r2=84797&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 18:29:32 2009
@@ -262,7 +262,7 @@
   if (MDName.empty())
     return false;
 
-  if (!isalpha(MDName[1]))
+  if (!isalpha(MDName[0]))
     return false;
 
   for (StringRef::iterator I = MDName.begin() + 1, E = MDName.end(); I != E;




From benny.kra at googlemail.com  Wed Oct 21 18:33:30 2009
From: benny.kra at googlemail.com (Benjamin Kramer)
Date: Thu, 22 Oct 2009 01:33:30 +0200
Subject: [llvm-commits] [llvm] r84790 -
	/llvm/trunk/lib/VMCore/Metadata.cpp
In-Reply-To: <A18EE01F-9D83-415E-A753-B035AED15E59@apple.com>
References: <200910212255.n9LMtqjr013741@zion.cs.uiuc.edu>
	<A18EE01F-9D83-415E-A753-B035AED15E59@apple.com>
Message-ID: <A35A620A-070C-4B2F-B15B-4846A72B31E9@gmail.com>


Am 22.10.2009 um 01:22 schrieb Chris Lattner:

>
> On Oct 21, 2009, at 3:55 PM, Benjamin Kramer wrote:
>
>> Author: d0k
>> Date: Wed Oct 21 17:55:51 2009
>> New Revision: 84790
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=84790&view=rev
>> Log:
>> Simplify code. No intended functionality change.
>
> Thanks Benjamin!
>> +++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 17:55:51 2009
>> @@ -259,23 +259,17 @@
>>
>> /// isValidName - Return true if Name is a valid custom metadata  
>> handler name.
>> bool MetadataContext::isValidName(const StringRef MDName) {
>> -  const char *Name = MDName.data();
>> -  if (!Name)
>> +  if (MDName.empty())
>>    return false;
>>
>> -  if (!isalpha(*Name))
>> +  if (!isalpha(MDName[1]))
>
> I think this should be [0] not [1].
>
> -Chris

Thanks for noticing that stupid mistake. Fixed in 84797.



From grosbach at apple.com  Wed Oct 21 18:40:57 2009
From: grosbach at apple.com (Jim Grosbach)
Date: Wed, 21 Oct 2009 23:40:57 -0000
Subject: [llvm-commits] [llvm] r84798 - in /llvm/trunk/lib/Target/ARM:
 ARMBaseRegisterInfo.cpp Thumb1RegisterInfo.cpp Thumb1RegisterInfo.h
Message-ID: <200910212340.n9LNevtn015459@zion.cs.uiuc.edu>

Author: grosbach
Date: Wed Oct 21 18:40:56 2009
New Revision: 84798

URL: http://llvm.org/viewvc/llvm-project?rev=84798&view=rev
Log:
Missing piece of the ARM frame index post-scavenging conditionalization

Modified:
    llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
    llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
    llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h

Modified: llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp?rev=84798&r1=84797&r2=84798&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMBaseRegisterInfo.cpp Wed Oct 21 18:40:56 2009
@@ -944,6 +944,7 @@
 requiresRegisterScavenging(const MachineFunction &MF) const {
   return true;
 }
+
 bool ARMBaseRegisterInfo::
 requiresFrameIndexScavenging(const MachineFunction &MF) const {
   return ScavengeFrameIndexVals;

Modified: llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp?rev=84798&r1=84797&r2=84798&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp Wed Oct 21 18:40:56 2009
@@ -77,6 +77,18 @@
   return TargetRegisterInfo::getPhysicalRegisterRegClass(Reg, VT);
 }
 
+bool
+Thumb1RegisterInfo::requiresRegisterScavenging(const MachineFunction &MF) const {
+  return true;
+}
+
+bool
+Thumb1RegisterInfo::requiresFrameIndexScavenging(const MachineFunction &MF)
+  const {
+  return true;
+}
+
+
 bool Thumb1RegisterInfo::hasReservedCallFrame(MachineFunction &MF) const {
   const MachineFrameInfo *FFI = MF.getFrameInfo();
   unsigned CFSize = FFI->getMaxCallFrameSize();

Modified: llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h?rev=84798&r1=84797&r2=84798&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h (original)
+++ llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.h Wed Oct 21 18:40:56 2009
@@ -40,6 +40,9 @@
   const TargetRegisterClass *
     getPhysicalRegisterRegClass(unsigned Reg, EVT VT = MVT::Other) const;
 
+  bool requiresRegisterScavenging(const MachineFunction &MF) const;
+  bool requiresFrameIndexScavenging(const MachineFunction &MF) const;
+
   bool hasReservedCallFrame(MachineFunction &MF) const;
 
   void eliminateCallFramePseudoInstr(MachineFunction &MF,




From sabre at nondot.org  Wed Oct 21 18:41:58 2009
From: sabre at nondot.org (Chris Lattner)
Date: Wed, 21 Oct 2009 23:41:58 -0000
Subject: [llvm-commits] [llvm] r84799 - in /llvm/trunk:
 lib/Transforms/Scalar/InstructionCombining.cpp
 test/Transforms/InstCombine/fold-intophi.ll
Message-ID: <200910212341.n9LNfxIQ015498@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 18:41:58 2009
New Revision: 84799

URL: http://llvm.org/viewvc/llvm-project?rev=84799&view=rev
Log:
revert r84754, it isn't the right approach.  Edwin, please propose 
patches for fixes like this instead of committing them directly.


Removed:
    llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll
Modified:
    llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp

Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=84799&r1=84798&r2=84799&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Wed Oct 21 18:41:58 2009
@@ -1985,42 +1985,6 @@
   return 0;
 }
 
-// Check whether all the operands of the PHI dominate the PHI node,
-// knowing that the PHI's operands either dominate BB, or are defined in the BB.
-static bool checkPHI(PHINode *PN, BasicBlock *BB)
-{
-  BasicBlock *phiBB = PN->getParent();
-  for (unsigned i=0;i<PN->getNumIncomingValues();i++) {
-    Instruction *I = dyn_cast<Instruction>(PN->getIncomingValue(i));
-    if (!I)
-      continue;
-    BasicBlock *pBB = I->getParent();
-    if (pBB == BB) {
-      // another PHI in same BB is always before PN (PN is last phi).
-      if (isa<PHINode>(I))
-        continue;
-      // An instruction in the same BB, and not a phi, this is after PN.
-      return false;
-    }
-    if (phiBB == BB)
-      continue;
-    // We don't have dominator info, so just check that the instruction
-    // is not defined in on of the BB on the unique path between BB and phiBB.
-    // If there is no such unique path, or pBB equals to one of the BBs on that
-    // path we know that this operand doesn't dominate the PHI node.
-    BasicBlock *B = PN->getIncomingBlock(i);
-    while ((B = B->getUniquePredecessor())) {
-      if (pBB == B)
-        return false;
-      if (B == phiBB)
-        break;
-    }
-    // No unique path -> doesn't dominate
-    if (!B)
-      return false;
-  }
-  return true;
-}
 
 /// FoldOpIntoPhi - Given a binary operator, cast instruction, or select which
 /// has a PHI node as operand #0, see if we can fold the instruction into the
@@ -2072,13 +2036,9 @@
   // Okay, we can do the transformation: create the new PHI node.
   PHINode *NewPN = PHINode::Create(I.getType(), "");
   NewPN->reserveOperandSpace(PN->getNumOperands()/2);
-  // We must the PHI node as the last PHI, because it may use one of the other
-  // PHIs.
-  BasicBlock::iterator BBIt = PN;
-  while (isa<PHINode>(BBIt)) ++BBIt;
-  InsertNewInstBefore(NewPN, *BBIt);
+  InsertNewInstBefore(NewPN, *PN);
+  NewPN->takeName(PN);
 
-  SmallVector<Instruction*, 2> tmpWorklist;
   // Next, add all of the operands to the PHI.
   if (SelectInst *SI = dyn_cast<SelectInst>(&I)) {
     // We only currently try to fold the condition of a select when it is a phi,
@@ -2098,7 +2058,7 @@
         InV = SelectInst::Create(PN->getIncomingValue(i), TrueVInPred,
                                  FalseVInPred,
                                  "phitmp", NonConstBB->getTerminator());
-        tmpWorklist.push_back(cast<Instruction>(InV));
+        Worklist.Add(cast<Instruction>(InV));
       }
       NewPN->addIncoming(InV, ThisBB);
     }
@@ -2124,7 +2084,8 @@
                                 NonConstBB->getTerminator());
         else
           llvm_unreachable("Unknown binop!");
-        tmpWorklist.push_back(cast<Instruction>(InV));
+        
+        Worklist.Add(cast<Instruction>(InV));
       }
       NewPN->addIncoming(InV, PN->getIncomingBlock(i));
     }
@@ -2140,26 +2101,11 @@
         InV = CastInst::Create(CI->getOpcode(), PN->getIncomingValue(i), 
                                I.getType(), "phitmp", 
                                NonConstBB->getTerminator());
-        tmpWorklist.push_back(cast<Instruction>(InV));
+        Worklist.Add(cast<Instruction>(InV));
       }
       NewPN->addIncoming(InV, PN->getIncomingBlock(i));
     }
   }
-  // The PHI's operands must dominate the PHI, if we can't prove that
-  // undo this transformation.
-  if (!checkPHI(NewPN, I.getParent())) {
-    Worklist.Remove(NewPN);
-    NewPN->eraseFromParent();
-    while (!tmpWorklist.empty()) {
-      tmpWorklist.pop_back_val()->eraseFromParent();
-    }
-    return 0;
-  }
-  while (!tmpWorklist.empty()) {
-    Worklist.Add(tmpWorklist.pop_back_val());
-  }
-  NewPN->takeName(PN);
-  Worklist.Add(PN);
   return ReplaceInstUsesWith(I, NewPN);
 }
 

Removed: llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll?rev=84798&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll (original)
+++ llvm/trunk/test/Transforms/InstCombine/fold-intophi.ll (removed)
@@ -1,54 +0,0 @@
-; RUN: opt -instcombine -S <%s | FileCheck %s
-; PR5262
-; Make sure the PHI node gets put in a place where all of its operands dominate it
-target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
-target triple = "x86_64-unknown-linux-gnu"
-
- at tmp2 = global i64 0                              ; <i64*> [#uses=1]
-
-declare void @use(i64) nounwind
-
-define void @foo(i1) nounwind align 2 {
-; <label>:1
-  br i1 %0, label %2, label %3
-
-; <label>:2                                       ; preds = %1
-  br label %3
-
-; <label>:3                                       ; preds = %2, %1
-; CHECK: <label>:3
-; CHECK-NEXT: %v4 = phi i1 [ false, %2 ], [ true, %1 ]
-; CHECK-NEXT: %v4_ =  phi i1 [ true, %2 ], [ false, %1 ]
-; CHECK-NEXT: br label %l8
-  %v4 = phi i8 [ 1, %2 ], [ 0, %1 ]                ; <i8> [#uses=1]
-  %v4_ = phi i8 [ 0, %2 ], [ 1, %1 ]                ; <i8> [#uses=1]
-  %v5 = icmp eq i8 %v4, 0                           ; <i1> [#uses=1]
-  %v5_ = icmp eq i8 %v4_, 0                           ; <i1> [#uses=1]
-  %v6 = load i64* @tmp2, align 8                   ; <i64> [#uses=1]
-  %v7 = select i1 %v5, i64 0, i64 %v6                ; <i64> [#uses=1]
-  br label %l8
-
-l8:
-  %v9 = load i64* @tmp2, align 8
-  call void @use(i64 %v7)
-  br label %l10
-l10:
-  %v11 = select i1 %v5_, i64 0, i64 %v9                ; <i64> [#uses=1]
-  call void @use(i64 %v11)
-  br label %l11
-l11:
-  %v12 = load i64* @tmp2, align 8
-  %v13 = select i1 %v5_, i64 0, i64 %v12                ; <i64> [#uses=1]
-  call void @use(i64 %v13)
-  br i1 %0, label %l12, label %l13
-l12:
-  br label %l13
-l13:
-;CHECK: l13:
-;CHECK-NEXT: %v14 = phi i64 [ %v12, %l12 ], [ 0, %l11 ]
-;CHECK-NEXT: call void @use(i64 %v14)
-  %v14 = phi i1 [0, %l12], [1, %l11]
-  %v16 = select i1 %v14, i64 0, i64 %v12                ; <i64> [#uses=1]
-  call void @use(i64 %v16)
-  ret void
-}




From dpatel at apple.com  Wed Oct 21 18:57:35 2009
From: dpatel at apple.com (Devang Patel)
Date: Wed, 21 Oct 2009 23:57:35 -0000
Subject: [llvm-commits] [llvm] r84801 - in /llvm/trunk:
 include/llvm/LLVMContext.h include/llvm/Metadata.h
 lib/Transforms/IPO/GlobalDCE.cpp lib/Transforms/IPO/StripSymbols.cpp
 lib/VMCore/LLVMContext.cpp lib/VMCore/LLVMContextImpl.h
 lib/VMCore/Metadata.cpp
Message-ID: <200910212357.n9LNvZ5A016097@zion.cs.uiuc.edu>

Author: dpatel
Date: Wed Oct 21 18:57:35 2009
New Revision: 84801

URL: http://llvm.org/viewvc/llvm-project?rev=84801&view=rev
Log:
Derive metadata hierarchy from Value instead of User.


Modified:
    llvm/trunk/include/llvm/LLVMContext.h
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp
    llvm/trunk/lib/Transforms/IPO/StripSymbols.cpp
    llvm/trunk/lib/VMCore/LLVMContext.cpp
    llvm/trunk/lib/VMCore/LLVMContextImpl.h
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/LLVMContext.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/LLVMContext.h?rev=84801&r1=84800&r2=84801&view=diff

==============================================================================
--- llvm/trunk/include/llvm/LLVMContext.h (original)
+++ llvm/trunk/include/llvm/LLVMContext.h Wed Oct 21 18:57:35 2009
@@ -33,7 +33,6 @@
 public:
   LLVMContextImpl* const pImpl;
   MetadataContext &getMetadata();
-  bool RemoveDeadMetadata();
   LLVMContext();
   ~LLVMContext();
 };

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84801&r1=84800&r2=84801&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Wed Oct 21 18:57:35 2009
@@ -16,7 +16,7 @@
 #ifndef LLVM_METADATA_H
 #define LLVM_METADATA_H
 
-#include "llvm/User.h"
+#include "llvm/Value.h"
 #include "llvm/Type.h"
 #include "llvm/ADT/FoldingSet.h"
 #include "llvm/ADT/SmallVector.h"
@@ -33,17 +33,11 @@
 
 //===----------------------------------------------------------------------===//
 // MetadataBase  - A base class for MDNode, MDString and NamedMDNode.
-class MetadataBase : public User {
-private:
-  /// ReservedSpace - The number of operands actually allocated.  NumOperands is
-  /// the number actually in use.
-  unsigned ReservedSpace;
-
+class MetadataBase : public Value {
 protected:
   MetadataBase(const Type *Ty, unsigned scid)
-    : User(Ty, scid, NULL, 0), ReservedSpace(0) {}
+    : Value(Ty, scid) {}
 
-  void resizeOperands(unsigned NumOps);
 public:
 
   /// Methods for support type inquiry through isa, cast, and dyn_cast:
@@ -60,8 +54,6 @@
 /// MDString is always unnamd.
 class MDString : public MetadataBase {
   MDString(const MDString &);            // DO NOT IMPLEMENT
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
-  unsigned getNumOperands();             // DO NOT IMPLEMENT
 
   StringRef Str;
 protected:
@@ -69,10 +61,6 @@
     : MetadataBase(Type::getMetadataTy(C), Value::MDStringVal), Str(begin, l) {}
 
 public:
-  // Do not allocate any space for operands.
-  void *operator new(size_t s) {
-    return User::operator new(s, 0);
-  }
   static MDString *get(LLVMContext &Context, const StringRef &Str);
   
   StringRef getString() const { return Str; }
@@ -102,9 +90,6 @@
 /// MDNode is always unnamed.
 class MDNode : public MetadataBase, public FoldingSetNode {
   MDNode(const MDNode &);                // DO NOT IMPLEMENT
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
-  // getNumOperands - Make this only available for private uses.
-  unsigned getNumOperands() { return User::getNumOperands();  }
 
   friend class ElementVH;
   // Use CallbackVH to hold MDNOde elements.
@@ -131,17 +116,10 @@
 protected:
   explicit MDNode(LLVMContext &C, Value *const *Vals, unsigned NumVals);
 public:
-  // Do not allocate any space for operands.
-  void *operator new(size_t s) {
-    return User::operator new(s, 0);
-  }
   // Constructors and destructors.
   static MDNode *get(LLVMContext &Context, 
                      Value *const *Vals, unsigned NumVals);
 
-  /// dropAllReferences - Remove all uses and clear node vector.
-  void dropAllReferences();
-
   /// ~MDNode - Destroy MDNode.
   ~MDNode();
   
@@ -193,9 +171,6 @@
   friend class LLVMContextImpl;
 
   NamedMDNode(const NamedMDNode &);      // DO NOT IMPLEMENT
-  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
-  // getNumOperands - Make this only available for private uses.
-  unsigned getNumOperands() { return User::getNumOperands();  }
 
   Module *Parent;
   SmallVector<WeakMetadataVH, 4> Node;
@@ -206,10 +181,6 @@
   explicit NamedMDNode(LLVMContext &C, const Twine &N, MetadataBase*const *Vals, 
                        unsigned NumVals, Module *M = 0);
 public:
-  // Do not allocate any space for operands.
-  void *operator new(size_t s) {
-    return User::operator new(s, 0);
-  }
   static NamedMDNode *Create(LLVMContext &C, const Twine &N, 
                              MetadataBase *const *MDs, 
                              unsigned NumMDs, Module *M = 0) {
@@ -245,8 +216,6 @@
 
   /// addElement - Add metadata element.
   void addElement(MetadataBase *M) {
-    resizeOperands(NumOperands + 1);
-    OperandList[NumOperands++] = M;
     Node.push_back(WeakMetadataVH(M));
   }
 

Modified: llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp?rev=84801&r1=84800&r2=84801&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp (original)
+++ llvm/trunk/lib/Transforms/IPO/GlobalDCE.cpp Wed Oct 21 18:57:35 2009
@@ -149,9 +149,6 @@
   // Make sure that all memory is released
   AliveGlobals.clear();
 
-  // Remove dead metadata.
-  // FIXME - Enable this.
-  // Changed |= M.getContext().RemoveDeadMetadata();
   return Changed;
 }
 

Modified: llvm/trunk/lib/Transforms/IPO/StripSymbols.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/StripSymbols.cpp?rev=84801&r1=84800&r2=84801&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/IPO/StripSymbols.cpp (original)
+++ llvm/trunk/lib/Transforms/IPO/StripSymbols.cpp Wed Oct 21 18:57:35 2009
@@ -250,8 +250,6 @@
   if (NMD)
     NMD->eraseFromParent();
 
-  // Remove dead metadata.
-  M.getContext().RemoveDeadMetadata();
   return true;
 }
 

Modified: llvm/trunk/lib/VMCore/LLVMContext.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/LLVMContext.cpp?rev=84801&r1=84800&r2=84801&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/LLVMContext.cpp (original)
+++ llvm/trunk/lib/VMCore/LLVMContext.cpp Wed Oct 21 18:57:35 2009
@@ -45,32 +45,6 @@
     OperandList[i+1] = IdxList[i];
 }
 
-bool LLVMContext::RemoveDeadMetadata() {
-  std::vector<WeakVH> DeadMDNodes;
-  bool Changed = false;
-  while (1) {
-
-    for (FoldingSet<MDNode>::iterator 
-           I = pImpl->MDNodeSet.begin(),
-           E = pImpl->MDNodeSet.end(); I != E; ++I) {
-      MDNode *N = &(*I);
-      if (N->use_empty()) 
-        DeadMDNodes.push_back(WeakVH(N));
-    }
-    
-    if (DeadMDNodes.empty())
-      return Changed;
-
-    while (!DeadMDNodes.empty()) {
-      Value *V = DeadMDNodes.back(); DeadMDNodes.pop_back();
-      if (const MDNode *N = dyn_cast_or_null<MDNode>(V))
-        if (N->use_empty())
-          delete N;
-    }
-  }
-  return Changed;
-}
-
 MetadataContext &LLVMContext::getMetadata() {
   return pImpl->TheMetadata;
 }

Modified: llvm/trunk/lib/VMCore/LLVMContextImpl.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/LLVMContextImpl.h?rev=84801&r1=84800&r2=84801&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/LLVMContextImpl.h (original)
+++ llvm/trunk/lib/VMCore/LLVMContextImpl.h Wed Oct 21 18:57:35 2009
@@ -203,9 +203,6 @@
     AggZeroConstants.freeConstants();
     NullPtrConstants.freeConstants();
     UndefValueConstants.freeConstants();
-    for (FoldingSet<MDNode>::iterator I = MDNodeSet.begin(), 
-           E = MDNodeSet.end(); I != E; ++I)
-      I->dropAllReferences();
     for (IntMapTy::iterator I = IntConstants.begin(), E = IntConstants.end(); 
          I != E; ++I) {
       if (I->second->use_empty())
@@ -216,6 +213,7 @@
       if (I->second->use_empty())
         delete I->second;
     }
+    MDNodeSet.clear();
   }
 };
 

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84801&r1=84800&r2=84801&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 18:57:35 2009
@@ -23,30 +23,6 @@
 // MetadataBase implementation.
 //
 
-/// resizeOperands - Metadata keeps track of other metadata uses using 
-/// OperandList. Resize this list to hold anticipated number of metadata
-/// operands.
-void MetadataBase::resizeOperands(unsigned NumOps) {
-  unsigned e = getNumOperands();
-  if (NumOps == 0) {
-    NumOps = e*2;
-    if (NumOps < 2) NumOps = 2;  
-  } else if (NumOps > NumOperands) {
-    // No resize needed.
-    if (ReservedSpace >= NumOps) return;
-  } else if (NumOps == NumOperands) {
-    if (ReservedSpace == NumOps) return;
-  } else {
-    return;
-  }
-
-  ReservedSpace = NumOps;
-  Use *OldOps = OperandList;
-  Use *NewOps = allocHungoffUses(NumOps);
-  std::copy(OldOps, OldOps + e, NewOps);
-  OperandList = NewOps;
-  if (OldOps) Use::zap(OldOps, OldOps + e, true);
-}
 //===----------------------------------------------------------------------===//
 // MDString implementation.
 //
@@ -65,20 +41,11 @@
 //
 MDNode::MDNode(LLVMContext &C, Value *const *Vals, unsigned NumVals)
   : MetadataBase(Type::getMetadataTy(C), Value::MDNodeVal) {
-  NumOperands = 0;
-  resizeOperands(NumVals);
   NodeSize = NumVals;
   Node = new ElementVH[NodeSize];
   ElementVH *Ptr = Node;
-  for (unsigned i = 0; i != NumVals; ++i) {
-    // Only record metadata uses.
-    if (MetadataBase *MB = dyn_cast_or_null<MetadataBase>(Vals[i]))
-      OperandList[NumOperands++] = MB;
-    else if(Vals[i] && 
-            Vals[i]->getType()->getTypeID() == Type::MetadataTyID)
-      OperandList[NumOperands++] = Vals[i];
+  for (unsigned i = 0; i != NumVals; ++i) 
     *Ptr++ = ElementVH(Vals[i], this);
-  }
 }
 
 void MDNode::Profile(FoldingSetNodeID &ID) const {
@@ -109,19 +76,14 @@
   return N;
 }
 
-/// dropAllReferences - Remove all uses and clear node vector.
-void MDNode::dropAllReferences() {
-  User::dropAllReferences();
-  delete [] Node;
-  Node = NULL;
-}
-
+/// ~MDNode - Destroy MDNode.
 MDNode::~MDNode() {
   {
     LLVMContextImpl *pImpl = getType()->getContext().pImpl;
     pImpl->MDNodeSet.RemoveNode(this);
   }
-  dropAllReferences();
+  delete [] Node;
+  Node = NULL;
 }
 
 // Replace value from this node's element list.
@@ -148,27 +110,6 @@
   // Remove "this" from the context map. 
   pImpl->MDNodeSet.RemoveNode(this);
 
-  // MDNode only lists metadata elements in operand list, because MDNode
-  // used by MDNode is considered a valid use. However on the side, MDNode
-  // using a non-metadata value is not considered a "use" of non-metadata
-  // value.
-  SmallVector<unsigned, 4> OpIndexes;
-  unsigned OpIndex = 0;
-  for (User::op_iterator OI = op_begin(), OE = op_end();
-       OI != OE; ++OI, OpIndex++) {
-    if (*OI == From)
-      OpIndexes.push_back(OpIndex);
-  }
-  if (MetadataBase *MDTo = dyn_cast_or_null<MetadataBase>(To)) {
-    for (SmallVector<unsigned, 4>::iterator OI = OpIndexes.begin(),
-           OE = OpIndexes.end(); OI != OE; ++OI)
-      setOperand(*OI, MDTo);
-  } else {
-    for (SmallVector<unsigned, 4>::iterator OI = OpIndexes.begin(),
-           OE = OpIndexes.end(); OI != OE; ++OI)
-      setOperand(*OI, 0);
-  }
-
   // Replace From element(s) in place.
   for (SmallVector<unsigned, 4>::iterator I = Indexes.begin(), E = Indexes.end(); 
        I != E; ++I) {
@@ -207,14 +148,10 @@
                          unsigned NumMDs, Module *ParentModule)
   : MetadataBase(Type::getMetadataTy(C), Value::NamedMDNodeVal), Parent(0) {
   setName(N);
-  NumOperands = 0;
-  resizeOperands(NumMDs);
 
-  for (unsigned i = 0; i != NumMDs; ++i) {
-    if (MDs[i])
-      OperandList[NumOperands++] = MDs[i];
+  for (unsigned i = 0; i != NumMDs; ++i)
     Node.push_back(WeakMetadataVH(MDs[i]));
-  }
+
   if (ParentModule)
     ParentModule->getNamedMDList().push_back(this);
 }
@@ -236,7 +173,6 @@
 
 /// dropAllReferences - Remove all uses and clear node vector.
 void NamedMDNode::dropAllReferences() {
-  User::dropAllReferences();
   Node.clear();
 }
 




From gohman at apple.com  Wed Oct 21 19:03:58 2009
From: gohman at apple.com (Dan Gohman)
Date: Thu, 22 Oct 2009 00:03:58 -0000
Subject: [llvm-commits] [llvm] r84803 - in /llvm/trunk:
 lib/CodeGen/BranchFolding.cpp test/CodeGen/CellSPU/2009-01-01-BrCond.ll
 test/CodeGen/Generic/switch-lower-feature-2.ll
 test/CodeGen/X86/loop-blocks.ll
Message-ID: <200910220003.n9M03x0b016487@zion.cs.uiuc.edu>

Author: djg
Date: Wed Oct 21 19:03:58 2009
New Revision: 84803

URL: http://llvm.org/viewvc/llvm-project?rev=84803&view=rev
Log:
Revert the main portion of r31856. It was causing BranchFolding
to break up CFG diamonds by banishing one of the blocks to the end of
the function, which is bad for code density and branch size.

This does pessimize MultiSource/Benchmarks/Ptrdist/yacr2, the
benchmark cited as the reason for the change, however I've examined
the code and it looks more like a case of gaming a particular
branch than of being generally applicable.

Modified:
    llvm/trunk/lib/CodeGen/BranchFolding.cpp
    llvm/trunk/test/CodeGen/CellSPU/2009-01-01-BrCond.ll
    llvm/trunk/test/CodeGen/Generic/switch-lower-feature-2.ll
    llvm/trunk/test/CodeGen/X86/loop-blocks.ll

Modified: llvm/trunk/lib/CodeGen/BranchFolding.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/BranchFolding.cpp?rev=84803&r1=84802&r2=84803&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/BranchFolding.cpp (original)
+++ llvm/trunk/lib/CodeGen/BranchFolding.cpp Wed Oct 21 19:03:58 2009
@@ -945,15 +945,15 @@
       }
     }
     
-    // If this block doesn't fall through (e.g. it ends with an uncond branch or
-    // has no successors) and if the pred falls through into this block, and if
-    // it would otherwise fall through into the block after this, move this
-    // block to the end of the function.
+    // If this block has no successors (e.g. it is a return block or ends with
+    // a call to a no-return function like abort or __cxa_throw) and if the pred
+    // falls through into this block, and if it would otherwise fall through
+    // into the block after this, move this block to the end of the function.
     //
     // We consider it more likely that execution will stay in the function (e.g.
     // due to loops) than it is to exit it.  This asserts in loops etc, moving
     // the assert condition out of the loop body.
-    if (!PriorCond.empty() && PriorFBB == 0 &&
+    if (MBB->succ_empty() && !PriorCond.empty() && PriorFBB == 0 &&
         MachineFunction::iterator(PriorTBB) == FallThrough &&
         !CanFallThrough(MBB)) {
       bool DoTransform = true;

Modified: llvm/trunk/test/CodeGen/CellSPU/2009-01-01-BrCond.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/CellSPU/2009-01-01-BrCond.ll?rev=84803&r1=84802&r2=84803&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/CellSPU/2009-01-01-BrCond.ll (original)
+++ llvm/trunk/test/CodeGen/CellSPU/2009-01-01-BrCond.ll Wed Oct 21 19:03:58 2009
@@ -1,4 +1,4 @@
-; RUN: llc < %s -march=cellspu -o - | grep brz
+; RUN: llc < %s -march=cellspu -o - | grep brnz
 ; PR3274
 
 target datalayout = "E-p:32:32:128-i1:8:128-i8:8:128-i16:16:128-i32:32:128-i64:32:128-f32:32:128-f64:64:128-v64:64:64-v128:128:128-a0:0:128-s0:128:128"

Modified: llvm/trunk/test/CodeGen/Generic/switch-lower-feature-2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Generic/switch-lower-feature-2.ll?rev=84803&r1=84802&r2=84803&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/Generic/switch-lower-feature-2.ll (original)
+++ llvm/trunk/test/CodeGen/Generic/switch-lower-feature-2.ll Wed Oct 21 19:03:58 2009
@@ -5,9 +5,9 @@
 ; RUN: grep 1023 %t | count 1
 ; RUN: grep 119  %t | count 1
 ; RUN: grep JTI %t | count 2
-; RUN: grep jg %t | count 1
+; RUN: grep jg %t | count 3
 ; RUN: grep ja %t | count 1
-; RUN: grep js %t | count 1
+; RUN: grep jns %t | count 1
 
 target triple = "i686-pc-linux-gnu"
 

Modified: llvm/trunk/test/CodeGen/X86/loop-blocks.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/loop-blocks.ll?rev=84803&r1=84802&r2=84803&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/loop-blocks.ll (original)
+++ llvm/trunk/test/CodeGen/X86/loop-blocks.ll Wed Oct 21 19:03:58 2009
@@ -74,16 +74,16 @@
 ; CHECK: yet_more_involved:
 ;      CHECK:   jmp .LBB3_1
 ; CHECK-NEXT:   align
-; CHECK-NEXT: .LBB3_7:
-; CHECK-NEXT:   call block_a_true_func
-; CHECK-NEXT:   jmp .LBB3_4
-; CHECK-NEXT: .LBB3_2:
+; CHECK-NEXT: .LBB3_3:
 ; CHECK-NEXT:   call bar99
 ; CHECK-NEXT:   call get
 ; CHECK-NEXT:   cmpl $2999, %eax
-; CHECK-NEXT:   jle .LBB3_7
+; CHECK-NEXT:   jg .LBB3_5
+; CHECK-NEXT:   call block_a_true_func
+; CHECK-NEXT:   jmp .LBB3_6
+; CHECK-NEXT: .LBB3_5:
 ; CHECK-NEXT:   call block_a_false_func
-; CHECK-NEXT: .LBB3_4:
+; CHECK-NEXT: .LBB3_6:
 ; CHECK-NEXT:   call block_a_merge_func
 ; CHECK-NEXT: .LBB3_1:
 ; CHECK-NEXT:   call body




From dpatel at apple.com  Wed Oct 21 19:10:16 2009
From: dpatel at apple.com (Devang Patel)
Date: Thu, 22 Oct 2009 00:10:16 -0000
Subject: [llvm-commits] [llvm] r84804 - in /llvm/trunk:
 include/llvm/Metadata.h lib/VMCore/Metadata.cpp
Message-ID: <200910220010.n9M0AGPE016727@zion.cs.uiuc.edu>

Author: dpatel
Date: Wed Oct 21 19:10:15 2009
New Revision: 84804

URL: http://llvm.org/viewvc/llvm-project?rev=84804&view=rev
Log:
Remove meaningless const.
Pass StringRef by value.

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84804&r1=84803&r2=84804&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Wed Oct 21 19:10:15 2009
@@ -61,7 +61,7 @@
     : MetadataBase(Type::getMetadataTy(C), Value::MDStringVal), Str(begin, l) {}
 
 public:
-  static MDString *get(LLVMContext &Context, const StringRef &Str);
+  static MDString *get(LLVMContext &Context, StringRef Str);
   
   StringRef getString() const { return Str; }
 
@@ -256,14 +256,14 @@
 public:
   /// registerMDKind - Register a new metadata kind and return its ID.
   /// A metadata kind can be registered only once. 
-  unsigned registerMDKind(const StringRef Name);
+  unsigned registerMDKind(StringRef Name);
 
   /// getMDKind - Return metadata kind. If the requested metadata kind
   /// is not registered then return 0.
-  unsigned getMDKind(const StringRef Name) const;
+  unsigned getMDKind(StringRef Name) const;
 
   /// isValidName - Return true if Name is a valid custom metadata handler name.
-  static bool isValidName(const StringRef Name);
+  static bool isValidName(StringRef Name);
 
   /// getMD - Get the metadata of given kind attached to an Instruction.
   /// If the metadata is not found then return 0.

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84804&r1=84803&r2=84804&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 19:10:15 2009
@@ -26,7 +26,7 @@
 //===----------------------------------------------------------------------===//
 // MDString implementation.
 //
-MDString *MDString::get(LLVMContext &Context, const StringRef &Str) {
+MDString *MDString::get(LLVMContext &Context, StringRef Str) {
   LLVMContextImpl *pImpl = Context.pImpl;
   StringMapEntry<MDString *> &Entry = 
     pImpl->MDStringCache.GetOrCreateValue(Str);
@@ -186,7 +186,7 @@
 
 /// registerMDKind - Register a new metadata kind and return its ID.
 /// A metadata kind can be registered only once. 
-unsigned MetadataContext::registerMDKind(const StringRef Name) {
+unsigned MetadataContext::registerMDKind(StringRef Name) {
   assert(isValidName(Name) && "Invalid custome metadata name!");
   unsigned Count = MDHandlerNames.size();
   assert(MDHandlerNames.count(Name) == 0 && "Already registered MDKind!");
@@ -194,7 +194,7 @@
 }
 
 /// isValidName - Return true if Name is a valid custom metadata handler name.
-bool MetadataContext::isValidName(const StringRef MDName) {
+bool MetadataContext::isValidName(StringRef MDName) {
   if (MDName.empty())
     return false;
 
@@ -211,7 +211,7 @@
 
 /// getMDKind - Return metadata kind. If the requested metadata kind
 /// is not registered then return 0.
-unsigned MetadataContext::getMDKind(const StringRef Name) const {
+unsigned MetadataContext::getMDKind(StringRef Name) const {
   StringMap<unsigned>::const_iterator I = MDHandlerNames.find(Name);
   if (I == MDHandlerNames.end()) {
     assert(isValidName(Name) && "Invalid custome metadata name!");




From asl at math.spbu.ru  Wed Oct 21 19:15:17 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Thu, 22 Oct 2009 00:15:17 -0000
Subject: [llvm-commits] [llvm] r84806 -
	/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
Message-ID: <200910220015.n9M0FIb6016961@zion.cs.uiuc.edu>

Author: asl
Date: Wed Oct 21 19:15:17 2009
New Revision: 84806

URL: http://llvm.org/viewvc/llvm-project?rev=84806&view=rev
Log:
Fix null pointer dereference.

Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp?rev=84806&r1=84805&r2=84806&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp Wed Oct 21 19:15:17 2009
@@ -4601,7 +4601,7 @@
       N->InitOperands(new SDUse[NumOps], Ops, NumOps);
       N->OperandsNeedDelete = true;
     } else
-      MN->InitOperands(MN->OperandList, Ops, NumOps);
+      N->InitOperands(N->OperandList, Ops, NumOps);
   }
 
   // Delete any nodes that are still dead after adding the uses for the




From asl at math.spbu.ru  Wed Oct 21 19:16:01 2009
From: asl at math.spbu.ru (Anton Korobeynikov)
Date: Thu, 22 Oct 2009 00:16:01 -0000
Subject: [llvm-commits] [llvm] r84808 - in /llvm/trunk:
 lib/Target/MSP430/MSP430ISelDAGToDAG.cpp test/CodeGen/MSP430/Inst16mm.ll
 test/CodeGen/MSP430/Inst8mm.ll
Message-ID: <200910220016.n9M0G1qP017006@zion.cs.uiuc.edu>

Author: asl
Date: Wed Oct 21 19:16:00 2009
New Revision: 84808

URL: http://llvm.org/viewvc/llvm-project?rev=84808&view=rev
Log:
Use special DAG-to-DAG preprocessing to allow mem-mem instructions to be selected.
Yay for ASCII graphics!

Modified:
    llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
    llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll
    llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll

Modified: llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp?rev=84808&r1=84807&r2=84808&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp Wed Oct 21 19:16:00 2009
@@ -65,6 +65,9 @@
       return "MSP430 DAG->DAG Pattern Instruction Selection";
     }
 
+    bool IsLegalAndProfitableToFold(SDNode *N, SDNode *U,
+                                    SDNode *Root) const;
+
     virtual bool
     SelectInlineAsmMemoryOperand(const SDValue &Op, char ConstraintCode,
                                  std::vector<SDValue> &OutOps);
@@ -73,6 +76,7 @@
   #include "MSP430GenDAGISel.inc"
 
   private:
+    DenseMap<SDNode*, SDNode*> RMWStores;
     void PreprocessForRMW();
     SDNode *Select(SDValue Op);
     bool SelectAddr(SDValue Op, SDValue Addr, SDValue &Base, SDValue &Disp);
@@ -139,7 +143,6 @@
   return true;
 }
 
-
 bool MSP430DAGToDAGISel::
 SelectInlineAsmMemoryOperand(const SDValue &Op, char ConstraintCode,
                              std::vector<SDValue> &OutOps) {
@@ -157,6 +160,42 @@
   return false;
 }
 
+bool MSP430DAGToDAGISel::IsLegalAndProfitableToFold(SDNode *N, SDNode *U,
+                                                    SDNode *Root) const {
+  if (OptLevel == CodeGenOpt::None) return false;
+
+  /// RMW preprocessing creates the following code:
+  ///       [Load1]
+  ///       ^     ^
+  ///       |      \
+  ///       |       \
+  ///       [Load2] |
+  ///       ^    ^  |
+  ///       |    |  |
+  ///       |     \-|
+  ///       |       |
+  ///       |     [Op]
+  ///       |       ^
+  ///       |       |
+  ///       \      /
+  ///        \    /
+  ///       [Store]
+  ///
+  /// The path Store => Load2 => Load1 is via chain. Note that in general it is
+  /// not allowed to fold Load1 into Op (and Store) since it will creates a
+  /// cycle. However, this is perfectly legal for the loads moved below the
+  /// TokenFactor by PreprocessForRMW. Query the map Store => Load1 (created
+  /// during preprocessing) to determine whether it's legal to introduce such
+  /// "cycle" for a moment.
+  DenseMap<SDNode*, SDNode*>::iterator I = RMWStores.find(Root);
+  if (I != RMWStores.end() && I->second == N)
+    return true;
+
+  // Proceed to 'generic' cycle finder code
+  return SelectionDAGISel::IsLegalAndProfitableToFold(N, U, Root);
+}
+
+
 /// MoveBelowTokenFactor - Replace TokenFactor operand with load's chain operand
 /// and move load below the TokenFactor. Replace store's chain operand with
 /// load's chain result.
@@ -176,11 +215,43 @@
                              Store.getOperand(2), Store.getOperand(3));
 }
 
-/// isRMWLoad - Return true if N is a load that's part of RMW sub-DAG.
-/// The chain produced by the load must only be used by the store's chain
-/// operand, otherwise this may produce a cycle in the DAG.
-static bool isRMWLoad(SDValue N, SDValue Chain, SDValue Address,
-                      SDValue &Load) {
+/// MoveBelowTokenFactor2 - Replace TokenFactor operand with load's chain operand
+/// and move load below the TokenFactor. Replace store's chain operand with
+/// load's chain result. This a version which sinks two loads below token factor.
+/// Look into PreprocessForRMW comments for explanation of transform.
+static void MoveBelowTokenFactor2(SelectionDAG *CurDAG,
+                                  SDValue Load1, SDValue Load2,
+                                  SDValue Store, SDValue TF) {
+  SmallVector<SDValue, 4> Ops;
+  for (unsigned i = 0, e = TF.getNode()->getNumOperands(); i != e; ++i) {
+    SDNode* N = TF.getOperand(i).getNode();
+    if (Load2.getNode() == N)
+      Ops.push_back(Load2.getOperand(0));
+    else if (Load1.getNode() != N)
+      Ops.push_back(TF.getOperand(i));
+  }
+
+  SDValue NewTF = SDValue(CurDAG->MorphNodeTo(TF.getNode(),
+                                  TF.getOpcode(),
+                                  TF.getNode()->getVTList(),
+                                  &Ops[0], Ops.size()), TF.getResNo());
+  SDValue NewLoad2 = CurDAG->UpdateNodeOperands(Load2, NewTF,
+                                                Load2.getOperand(1),
+                                                Load2.getOperand(2));
+
+  SDValue NewLoad1 = CurDAG->UpdateNodeOperands(Load1, NewLoad2.getValue(1),
+                                                Load1.getOperand(1),
+                                                Load1.getOperand(2));
+
+  CurDAG->UpdateNodeOperands(Store,
+                             NewLoad1.getValue(1),
+                             Store.getOperand(1),
+                             Store.getOperand(2), Store.getOperand(3));
+}
+
+/// isAllowedToSink - return true if N a load which can be moved below token
+/// factor. Basically, the load should be non-volatile and has single use.
+static bool isLoadAllowedToSink(SDValue N, SDValue Chain) {
   if (N.getOpcode() == ISD::BIT_CONVERT)
     N = N.getOperand(0);
 
@@ -194,10 +265,19 @@
   if (ExtType != ISD::NON_EXTLOAD && ExtType != ISD::EXTLOAD)
     return false;
 
-  if (N.hasOneUse() &&
-      LD->hasNUsesOfValue(1, 1) &&
-      N.getOperand(1) == Address &&
-      LD->isOperandOf(Chain.getNode())) {
+  return (N.hasOneUse() &&
+          LD->hasNUsesOfValue(1, 1) &&
+          LD->isOperandOf(Chain.getNode()));
+}
+
+
+/// isRMWLoad - Return true if N is a load that's part of RMW sub-DAG.
+/// The chain produced by the load must only be used by the store's chain
+/// operand, otherwise this may produce a cycle in the DAG.
+static bool isRMWLoad(SDValue N, SDValue Chain, SDValue Address,
+                      SDValue &Load) {
+  if (isLoadAllowedToSink(N, Chain) &&
+      N.getOperand(1) == Address) {
     Load = N;
     return true;
   }
@@ -244,6 +324,34 @@
 ///       \      /
 ///        \    /
 ///       [Store]
+///
+/// We also recognize the case where second operand of Op is load as well and
+/// move it below token factor as well creating DAG as follows:
+///
+///     [Load chain]
+///         ^
+///         |
+///    [TokenFactor]
+///         ^
+///         |
+///       [Load1]
+///       ^     ^
+///       |      \
+///       |       \
+///       [Load2] |
+///       ^    ^  |
+///       |    |  |
+///       |     \-|
+///       |       |
+///       |     [Op]
+///       |       ^
+///       |       |
+///       \      /
+///        \    /
+///       [Store]
+///
+/// This allows selection of mem-mem instructions. Yay!
+
 void MSP430DAGToDAGISel::PreprocessForRMW() {
   for (SelectionDAG::allnodes_iterator I = CurDAG->allnodes_begin(),
          E = CurDAG->allnodes_end(); I != E; ++I) {
@@ -261,8 +369,8 @@
         !N1.hasOneUse())
       continue;
 
-    bool RModW = false;
-    SDValue Load;
+    unsigned RModW = 0;
+    SDValue Load1, Load2;
     unsigned Opcode = N1.getNode()->getOpcode();
     switch (Opcode) {
     case ISD::ADD:
@@ -273,23 +381,44 @@
     case ISD::ADDE: {
       SDValue N10 = N1.getOperand(0);
       SDValue N11 = N1.getOperand(1);
-      RModW = isRMWLoad(N10, Chain, N2, Load);
-      if (!RModW)
-        RModW = isRMWLoad(N11, Chain, N2, Load);
+      if (isRMWLoad(N10, Chain, N2, Load1)) {
+        if (isLoadAllowedToSink(N11, Chain)) {
+          Load2 = N11;
+          RModW = 2;
+        } else
+          RModW = 1;
+      } else if (isRMWLoad(N11, Chain, N2, Load1)) {
+        if (isLoadAllowedToSink(N10, Chain)) {
+          Load2 = N10;
+          RModW = 2;
+        } else
+          RModW = 1;
+      }
       break;
     }
     case ISD::SUB:
     case ISD::SUBC:
     case ISD::SUBE: {
       SDValue N10 = N1.getOperand(0);
-      RModW = isRMWLoad(N10, Chain, N2, Load);
+      SDValue N11 = N1.getOperand(1);
+      if (isRMWLoad(N10, Chain, N2, Load1)) {
+        if (isLoadAllowedToSink(N11, Chain)) {
+          Load2 = N11;
+          RModW = 2;
+        } else
+          RModW = 1;
+      }
       break;
     }
     }
 
-    if (RModW) {
-      MoveBelowTokenFactor(CurDAG, Load, SDValue(I, 0), Chain);
-      ++NumLoadMoved;
+    NumLoadMoved += RModW;
+    if (RModW == 1)
+      MoveBelowTokenFactor(CurDAG, Load1, SDValue(I, 0), Chain);
+    else if (RModW == 2) {
+      MoveBelowTokenFactor2(CurDAG, Load1, Load2, SDValue(I, 0), Chain);
+      SDNode* Store = I;
+      RMWStores[Store] = Load2.getNode();
     }
   }
 }
@@ -318,6 +447,7 @@
   DEBUG(errs() << "===== Instruction selection ends:\n");
 
   CurDAG->RemoveDeadNodes();
+  RMWStores.clear();
 }
 
 SDNode *MSP430DAGToDAGISel::Select(SDValue Op) {

Modified: llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll?rev=84808&r1=84807&r2=84808&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll (original)
+++ llvm/trunk/test/CodeGen/MSP430/Inst16mm.ll Wed Oct 21 19:16:00 2009
@@ -1,5 +1,4 @@
 ; RUN: llc -march=msp430 < %s | FileCheck %s
-; XFAIL: *
 target datalayout = "e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8"
 target triple = "msp430-generic-generic"
 @foo = common global i16 0, align 2

Modified: llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll?rev=84808&r1=84807&r2=84808&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll (original)
+++ llvm/trunk/test/CodeGen/MSP430/Inst8mm.ll Wed Oct 21 19:16:00 2009
@@ -1,5 +1,4 @@
 ; RUN: llc -march=msp430 < %s | FileCheck %s
-; XFAIL: *
 target datalayout = "e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8"
 target triple = "msp430-generic-generic"
 




From sabre at nondot.org  Wed Oct 21 19:17:27 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 00:17:27 -0000
Subject: [llvm-commits] [llvm] r84810 - in /llvm/trunk:
 lib/Transforms/Scalar/InstructionCombining.cpp
 test/Transforms/InstCombine/crash.ll
Message-ID: <200910220017.n9M0HRkA017077@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 19:17:26 2009
New Revision: 84810

URL: http://llvm.org/viewvc/llvm-project?rev=84810&view=rev
Log:
fix PR5262.

Modified:
    llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
    llvm/trunk/test/Transforms/InstCombine/crash.ll

Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=84810&r1=84809&r2=84810&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Wed Oct 21 19:17:26 2009
@@ -9274,14 +9274,44 @@
   return Changed ? &SI : 0;
 }
 
-/// isDefinedInBB - Return true if the value is an instruction defined in the
-/// specified basicblock.
-static bool isDefinedInBB(const Value *V, const BasicBlock *BB) {
+
+/// CanSelectOperandBeMappingIntoPredBlock - SI is a select whose condition is a
+/// PHI node (but the two may be in different blocks).  See if the true/false
+/// values (V) are live in all of the predecessor blocks of the PHI.  For
+/// example, cases like this cannot be mapped:
+///
+///   X = phi [ C1, BB1], [C2, BB2]
+///   Y = add
+///   Z = select X, Y, 0
+///
+/// because Y is not live in BB1/BB2.
+///
+static bool CanSelectOperandBeMappingIntoPredBlock(const Value *V,
+                                                   const SelectInst &SI) {
+  // If the value is a non-instruction value like a constant or argument, it
+  // can always be mapped.
   const Instruction *I = dyn_cast<Instruction>(V);
-  return I != 0 && I->getParent() == BB;
+  if (I == 0) return true;
+  
+  // If V is a PHI node defined in the same block as the condition PHI, we can
+  // map the arguments.
+  const PHINode *CondPHI = cast<PHINode>(SI.getCondition());
+  
+  if (const PHINode *VP = dyn_cast<PHINode>(I))
+    if (VP->getParent() == CondPHI->getParent())
+      return true;
+  
+  // Otherwise, if the PHI and select are defined in the same block and if V is
+  // defined in a different block, then we can transform it.
+  if (SI.getParent() == CondPHI->getParent() &&
+      I->getParent() != CondPHI->getParent())
+    return true;
+  
+  // Otherwise we have a 'hard' case and we can't tell without doing more
+  // detailed dominator based analysis, punt.
+  return false;
 }
 
-
 Instruction *InstCombiner::visitSelectInst(SelectInst &SI) {
   Value *CondVal = SI.getCondition();
   Value *TrueVal = SI.getTrueValue();
@@ -9493,16 +9523,13 @@
       return FoldI;
   }
 
-  // See if we can fold the select into a phi node.  The true/false values have
-  // to be live in the predecessor blocks.  If they are instructions in SI's
-  // block, we can't map to the predecessor.
-  if (isa<PHINode>(SI.getCondition()) &&
-      (!isDefinedInBB(SI.getTrueValue(), SI.getParent()) ||
-       isa<PHINode>(SI.getTrueValue())) &&
-      (!isDefinedInBB(SI.getFalseValue(), SI.getParent()) ||
-       isa<PHINode>(SI.getFalseValue())))
-    if (Instruction *NV = FoldOpIntoPhi(SI))
-      return NV;
+  // See if we can fold the select into a phi node if the condition is a select.
+  if (isa<PHINode>(SI.getCondition())) 
+    // The true/false values have to be live in the PHI predecessor's blocks.
+    if (CanSelectOperandBeMappingIntoPredBlock(TrueVal, SI) &&
+        CanSelectOperandBeMappingIntoPredBlock(FalseVal, SI))
+      if (Instruction *NV = FoldOpIntoPhi(SI))
+        return NV;
 
   if (BinaryOperator::isNot(CondVal)) {
     SI.setOperand(0, BinaryOperator::getNotArgument(CondVal));

Modified: llvm/trunk/test/Transforms/InstCombine/crash.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/crash.ll?rev=84810&r1=84809&r2=84810&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/crash.ll (original)
+++ llvm/trunk/test/Transforms/InstCombine/crash.ll Wed Oct 21 19:17:26 2009
@@ -1,4 +1,4 @@
-; RUN: opt < %s -instcombine | llvm-dis
+; RUN: opt < %s -instcombine -S
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
 target triple = "i386-apple-darwin10.0"
 
@@ -99,3 +99,29 @@
   %15 = tail call %t0* @bar2(i64 %14) nounwind      ; <%0*> [#uses=0]
   ret void
 }
+
+
+
+
+; PR5262
+; Make sure the PHI node gets put in a place where all of its operands dominate
+; it.
+define i64 @test4(i1 %c, i64* %P) nounwind align 2 {
+BB0:
+  br i1 %c, label %BB1, label %BB2
+
+BB1:
+  br label %BB2
+
+BB2:
+  %v5_ = phi i1 [ true, %BB0], [false, %BB1]
+  %v6 = load i64* %P
+  br label %l8
+
+l8:
+  br label %l10
+  
+l10:
+  %v11 = select i1 %v5_, i64 0, i64 %v6
+  ret i64 %v11
+}




From dpatel at apple.com  Wed Oct 21 19:22:06 2009
From: dpatel at apple.com (Devang Patel)
Date: Thu, 22 Oct 2009 00:22:06 -0000
Subject: [llvm-commits] [llvm] r84811 - in /llvm/trunk:
 include/llvm/Metadata.h lib/VMCore/Metadata.cpp
Message-ID: <200910220022.n9M0M6PO017321@zion.cs.uiuc.edu>

Author: dpatel
Date: Wed Oct 21 19:22:05 2009
New Revision: 84811

URL: http://llvm.org/viewvc/llvm-project?rev=84811&view=rev
Log:
Use StringRef to construct MDString.

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84811&r1=84810&r2=84811&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Wed Oct 21 19:22:05 2009
@@ -57,8 +57,8 @@
 
   StringRef Str;
 protected:
-  explicit MDString(LLVMContext &C, const char *begin, unsigned l)
-    : MetadataBase(Type::getMetadataTy(C), Value::MDStringVal), Str(begin, l) {}
+  explicit MDString(LLVMContext &C, StringRef S)
+    : MetadataBase(Type::getMetadataTy(C), Value::MDStringVal), Str(S) {}
 
 public:
   static MDString *get(LLVMContext &Context, StringRef Str);

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84811&r1=84810&r2=84811&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 19:22:05 2009
@@ -33,7 +33,8 @@
   MDString *&S = Entry.getValue();
   if (S) return S;
   
-  return S = new MDString(Context, Entry.getKeyData(), Entry.getKeyLength());
+  return S = 
+    new MDString(Context, StringRef(Entry.getKeyData(), Entry.getKeyLength()));
 }
 
 //===----------------------------------------------------------------------===//




From clattner at apple.com  Wed Oct 21 19:26:44 2009
From: clattner at apple.com (Chris Lattner)
Date: Wed, 21 Oct 2009 17:26:44 -0700
Subject: [llvm-commits] [llvm] r84811 - in /llvm/trunk:
	include/llvm/Metadata.h lib/VMCore/Metadata.cpp
In-Reply-To: <200910220022.n9M0M6PO017321@zion.cs.uiuc.edu>
References: <200910220022.n9M0M6PO017321@zion.cs.uiuc.edu>
Message-ID: <66E1589A-91E0-4E20-9B59-8E02F7FD2E4B@apple.com>

On Oct 21, 2009, at 5:22 PM, Devang Patel wrote:
> Author: dpatel
> Date: Wed Oct 21 19:22:05 2009
> New Revision: 84811
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84811&view=rev
> Log:
> Use StringRef to construct MDString.

Cool:

> +++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 19:22:05 2009
> @@ -33,7 +33,8 @@
>   MDString *&S = Entry.getValue();
>   if (S) return S;
>
> -  return S = new MDString(Context, Entry.getKeyData(),  
> Entry.getKeyLength());
> +  return S =
> +    new MDString(Context, StringRef(Entry.getKeyData(),  
> Entry.getKeyLength()));
> }

Entry.getKey() returns the key as a stringref.

-Chris


>
> // 
> = 
> = 
> = 
> ----------------------------------------------------------------------= 
> ==//
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From clattner at apple.com  Wed Oct 21 19:29:53 2009
From: clattner at apple.com (Chris Lattner)
Date: Wed, 21 Oct 2009 17:29:53 -0700
Subject: [llvm-commits] (half precision) [llvm] r84239 - in /llvm/trunk:
	include/llvm/ADT/APFloat.h lib/Support/APFloat.cpp
In-Reply-To: <3E890328-4BC7-47D4-B3BE-58CE46E21262@apple.com>
References: <200910160213.n9G2Dp0C003020@zion.cs.uiuc.edu>
	<3E890328-4BC7-47D4-B3BE-58CE46E21262@apple.com>
Message-ID: <2FB4E34F-F914-4275-BBB7-1BD226DCDD52@apple.com>


On Oct 20, 2009, at 5:23 PM, Dale Johannesen wrote:

> Some testcases to exercise basic functionality would be nice.  I think
> there is a bug here that should have shown up in testing.

Peter, can you take a look at Dale's feedback?

-Chris

>
> On Oct 15, 2009, at 7:13 PM, Chris Lattner wrote:
>> +APInt
>> +APFloat::convertHalfAPFloatToAPInt() const
>> +{
>> +  assert(semantics == (const llvm::fltSemantics*)&IEEEhalf);
>> +  assert (partCount()==1);
>> +
>> +  uint32_t myexponent, mysignificand;
>> +
>> +  if (category==fcNormal) {
>> +    myexponent = exponent+15; //bias
>> +    mysignificand = (uint32_t)*significandParts();
>> +    if (myexponent == 1 && !(mysignificand & 0x400))
>> +      myexponent = 0;   // denormal
>> +  } else if (category==fcZero) {
>> +    myexponent = 0;
>> +    mysignificand = 0;
>> +  } else if (category==fcInfinity) {
>> +    myexponent = 0xff;
>
> 0x1f (doesn't matter functionally, but cleaner)
>
>> +    mysignificand = 0;
>> +  } else {
>> +    assert(category == fcNaN && "Unknown category!");
>> +    myexponent = 0xff;
>
> 0x1f
>
>> +    mysignificand = (uint32_t)*significandParts();
>> +  }
>> +
>> +  return APInt(16, (((sign&1) << 15) | ((myexponent&0x1f) << 10) |
>> +                    (mysignificand & 0x3ff)));
>> +}
>> +
>
>
>
>> +void
>> +APFloat::initFromHalfAPInt(const APInt & api)
>> +{
>> +  assert(api.getBitWidth()==16);
>> +  uint32_t i = (uint32_t)*api.getRawData();
>> +  uint32_t myexponent = (i >> 15) & 0x1f;
>
> Bug: should be >> 10
>
>> +  uint32_t mysignificand = i & 0x3ff;
>> +
>> +  initialize(&APFloat::IEEEhalf);
>> +  assert(partCount()==1);
>> +
>> +  sign = i >> 15;
>> +  if (myexponent==0 && mysignificand==0) {
>> +    // exponent, significand meaningless
>> +    category = fcZero;
>> +  } else if (myexponent==0x1f && mysignificand==0) {
>> +    // exponent, significand meaningless
>> +    category = fcInfinity;
>> +  } else if (myexponent==0x1f && mysignificand!=0) {
>> +    // sign, exponent, significand meaningless
>> +    category = fcNaN;
>> +    *significandParts() = mysignificand;
>> +  } else {
>> +    category = fcNormal;
>> +    exponent = myexponent - 15;  //bias
>> +    *significandParts() = mysignificand;
>> +    if (myexponent==0)    // denormal
>> +      exponent = -14;
>> +    else
>> +      *significandParts() |= 0x400; // integer bit
>> +  }
>> +}
>> +
>> /// Treat api as containing the bits of a floating point number.
>> Currently
>> /// we infer the floating point type from the size of the APInt.  The
>> /// isIEEE argument distinguishes between PPC128 and IEEE128 (not
>> meaningful
>> @@ -3058,7 +3124,9 @@
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From clattner at apple.com  Wed Oct 21 19:38:49 2009
From: clattner at apple.com (Chris Lattner)
Date: Wed, 21 Oct 2009 17:38:49 -0700
Subject: [llvm-commits] [llvm] r84292 - in /llvm/trunk: examples/BrainF/
	include/llvm/ lib/AsmParser/ lib/Bitcode/Reader/
	lib/Transforms/IPO/ lib/Transforms/Scalar/ lib/VMCore/
	test/Transforms/GlobalOpt/ test/Transforms/IndMemRem/
	test/Transforms/InstCombine/
In-Reply-To: <5F7FC9EE-DD30-438F-833D-9F9338021A1A@apple.com>
References: <200910170000.n9H00RWN030901@zion.cs.uiuc.edu>
	<797711D6-5B12-4DA3-BFF4-97D38BDC6E58@apple.com>
	<5F7FC9EE-DD30-438F-833D-9F9338021A1A@apple.com>
Message-ID: <7B90F86B-6555-42E8-9ED7-C45C1216CF62@apple.com>

On Oct 21, 2009, at 12:12 PM, Victor Hernandez wrote:
>> This should use V->replaceAllUsesWith() instead of manually walking  
>> the use list.  Doing this will handle non-call users as well as  
>> call users.  What happens if the prototypes don't agree for these  
>> two functions?  I think you'll get an abort.  Please fix by using a  
>> constantexpr bitcast to the expected type.
>
> Fixed:

Much nicer, thanks.

>>> +    const Type* IntPtrTy = Type::getInt32Ty(Context);
>>> +    const Type* Int8PtrTy =  
>>> PointerType::getUnqual(Type::getInt8Ty(Context));
>>
>> Please use 'const Type *IntPtrTy', watch the * placement.  There is  
>> now a helper function that you can use instead of  
>> 'PointerType::getUnqual(Type::getInt8Ty(Context));', please use it.
>
> Fixed:
>   const Type *IntPtrTy = Type::getInt32Ty(Context);
>   const Type *Int8PtrTy = Type::getInt8PtrTy(Context);

Ok:

+  // Autoupgrade old malloc instruction to malloc call.
+  // FIXME: Remove in LLVM 3.0.
+  const Type *IntPtrTy = Type::getInt32Ty(Context);
+  const Type *Int8PtrTy = Type::getInt8PtrTy(Context);
+  if (!MallocF)
+    // Prototype malloc as "void *(int32)".
+    // This function is renamed as "malloc" in ValidateEndOfModule().
+    MallocF = cast<Function>(M->getOrInsertFunction(NULL, Int8PtrTy,
+                                                    IntPtrTy, NULL));
+  Inst = CallInst::CreateMalloc(BB, IntPtrTy, Ty, Size, MallocF);

It doesn't matter because this is very cold code, but you might as  
well sink the 'get' of Int8PtrTy into the 'if' for tidiness.

>
>>
>>> +    if (!MallocF)
>>> +      // Prototype malloc as "void *autoupgrade_malloc(int32)".
>>> +      MallocF = cast<Function>(M- 
>>> >getOrInsertFunction("autoupgrade_malloc",
>>> +                               Int8PtrTy, IntPtrTy, NULL));
>>
>> Why do you bother naming autoupgrade_malloc?  This is in the user's  
>> namespace, so it can conflict with a function named  
>> autoupgrade_malloc in the user's code.  Please just leave the name  
>> empty, making it impossible for it to conflict.
>
> Fixed.

Thanks.  While passing in NULL to getOrInsertFunction will probably  
work, please pass in "" instead.  This makes it obvious that it is the  
name you are specifying.

>>
>> The old code was correct because it would delete the call to  
>> malloc, so it is safe to assume that the malloc (which never got  
>> run) never ran out of space.  The transformation that does this  
>> should just be removed.
>
> I fixed InstCombine so that the icmp is removed only if the malloc  
> call is removed (which requires explicit removal because the  
> Worklist won't DCE any calls since they can have side-effects).  I  
> also updated the testcase to use FileCheck.

Looks good.

Thanks Victor,

-Chris
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091021/f178e735/attachment.html 

From evan.cheng at apple.com  Wed Oct 21 19:40:00 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 00:40:00 -0000
Subject: [llvm-commits] [llvm] r84813 - in /llvm/trunk:
 lib/Target/ARM/ARMISelDAGToDAG.cpp test/CodeGen/ARM/sbfx.ll
Message-ID: <200910220040.n9M0e1R3018004@zion.cs.uiuc.edu>

Author: evancheng
Date: Wed Oct 21 19:40:00 2009
New Revision: 84813

URL: http://llvm.org/viewvc/llvm-project?rev=84813&view=rev
Log:
Don't generate sbfx / ubfx with negative lsb field. Patch by David Conrad.

Modified:
    llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp
    llvm/trunk/test/CodeGen/ARM/sbfx.ll

Modified: llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp?rev=84813&r1=84812&r2=84813&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp Wed Oct 21 19:40:00 2009
@@ -1287,7 +1287,7 @@
       assert(Srl_imm > 0 && Srl_imm < 32 && "bad amount in shift node!");
       unsigned Width = 32 - Srl_imm;
       int LSB = Srl_imm - Shl_imm;
-      if ((LSB + Width) > 32)
+      if (LSB < 0)
         return NULL;
       SDValue Reg0 = CurDAG->getRegister(0, MVT::i32);
       SDValue Ops[] = { Op.getOperand(0).getOperand(0),

Modified: llvm/trunk/test/CodeGen/ARM/sbfx.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/sbfx.ll?rev=84813&r1=84812&r2=84813&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/ARM/sbfx.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/sbfx.ll Wed Oct 21 19:40:00 2009
@@ -35,3 +35,13 @@
     %tmp2 = lshr i32 %tmp, 29
     ret i32 %tmp2
 }
+
+define i32 @f5(i32 %a) {
+entry:
+; CHECK: f5:
+; CHECK-NOT: sbfx
+; CHECK: bx
+    %tmp = shl i32 %a, 3
+    %tmp2 = ashr i32 %tmp, 1
+    ret i32 %tmp2
+}




From sabre at nondot.org  Wed Oct 21 19:44:11 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 00:44:11 -0000
Subject: [llvm-commits] [llvm] r84814 - /llvm/trunk/tools/opt/opt.cpp
Message-ID: <200910220044.n9M0iB0X018137@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 19:44:10 2009
New Revision: 84814

URL: http://llvm.org/viewvc/llvm-project?rev=84814&view=rev
Log:
Add some command line options for twiddling the default data layout
used by opt when a module doesn't specify one.  Patch from Kenneth Uildriks!

Modified:
    llvm/trunk/tools/opt/opt.cpp

Modified: llvm/trunk/tools/opt/opt.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/opt/opt.cpp?rev=84814&r1=84813&r2=84814&view=diff

==============================================================================
--- llvm/trunk/tools/opt/opt.cpp (original)
+++ llvm/trunk/tools/opt/opt.cpp Wed Oct 21 19:44:10 2009
@@ -127,6 +127,15 @@
 static cl::opt<bool>
 AnalyzeOnly("analyze", cl::desc("Only perform analysis, no optimization"));
 
+static cl::opt<std::string>
+DefaultDataLayout("default-data-layout", 
+          cl::desc("data layout string to use if not specified by module"),
+          cl::value_desc("layout-string"), cl::init(""));
+
+static cl::opt<bool>
+NoDefaultDataLayout("no-default-data-layout",
+  cl::desc("no data layout assumptions unless module specifies data layout"));
+
 // ---------- Define Printers for module and function passes ------------
 namespace {
 
@@ -388,12 +397,21 @@
     PassManager Passes;
 
     // Add an appropriate TargetData instance for this module...
-    Passes.add(new TargetData(M.get()));
+    TargetData *TD = 0;
+    const std::string &ModuleDataLayout = M.get()->getDataLayout();
+    if (!ModuleDataLayout.empty())
+      TD = new TargetData(ModuleDataLayout);
+    else if (!NoDefaultDataLayout)
+      TD = new TargetData(DefaultDataLayout);
+
+    if (TD)
+      Passes.add(TD);
 
     FunctionPassManager *FPasses = NULL;
     if (OptLevelO1 || OptLevelO2 || OptLevelO3) {
       FPasses = new FunctionPassManager(new ExistingModuleProvider(M.get()));
-      FPasses->add(new TargetData(M.get()));
+      if (TD)
+        FPasses->add(new TargetData(*TD));
     }
 
     // If the -strip-debug command line option was specified, add it.  If




From clattner at apple.com  Wed Oct 21 19:44:25 2009
From: clattner at apple.com (Chris Lattner)
Date: Wed, 21 Oct 2009 17:44:25 -0700
Subject: [llvm-commits] [PATCH] Allow "opt" to run without target data
In-Reply-To: <400d33ea0910211431r5efc19fan571752db8914bde3@mail.gmail.com>
References: <400d33ea0910211431r5efc19fan571752db8914bde3@mail.gmail.com>
Message-ID: <174B3F81-4322-44D6-95C1-F873C0FC8BC7@apple.com>

Thanks, I committed a cleaned up version in r84814.

-Chris

On Oct 21, 2009, at 2:31 PM, Kenneth Uildriks wrote:

> Currently, there is no way to prevent "opt" from putting in default
> target data layout if the Module has none, or to supply an alternate
> target data layout.  The default target data layout is bogus on most
> platforms.
>
> This patch adds the following command-line options:
>
> -default-data-layout=<layout-string> - data layout string to use if
> not specified by module
> -no-default-data-layout - no data layout assumptions unless module
> specifies data layout.
>
> The -no-default-data-layout prevents a TargetData pass from being
> added to opt's PassManager and (theoretically) prevents platform data
> layout dependent optimization from being applied.  I think this should
> eventually be the default, but setting it to the default might break
> existing users and test scripts.  For now, if neither option is
> supplied, opt's behavior will remain unchanged from the current
> (trunk) version.
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From sabre at nondot.org  Wed Oct 21 19:46:42 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 00:46:42 -0000
Subject: [llvm-commits] [llvm] r84816 - in /llvm/trunk/tools/opt:
	CMakeLists.txt Makefile opt.cpp
Message-ID: <200910220046.n9M0kg6s018240@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 19:46:41 2009
New Revision: 84816

URL: http://llvm.org/viewvc/llvm-project?rev=84816&view=rev
Log:
nothing opt uses can throw, remove the try block and -fexceptions when
building opt.

Modified:
    llvm/trunk/tools/opt/CMakeLists.txt
    llvm/trunk/tools/opt/Makefile
    llvm/trunk/tools/opt/opt.cpp

Modified: llvm/trunk/tools/opt/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/opt/CMakeLists.txt?rev=84816&r1=84815&r2=84816&view=diff

==============================================================================
--- llvm/trunk/tools/opt/CMakeLists.txt (original)
+++ llvm/trunk/tools/opt/CMakeLists.txt Wed Oct 21 19:46:41 2009
@@ -1,4 +1,3 @@
-set(LLVM_REQUIRES_EH 1)
 set(LLVM_LINK_COMPONENTS bitreader asmparser bitwriter instrumentation scalaropts ipo)
 
 add_llvm_tool(opt

Modified: llvm/trunk/tools/opt/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/opt/Makefile?rev=84816&r1=84815&r2=84816&view=diff

==============================================================================
--- llvm/trunk/tools/opt/Makefile (original)
+++ llvm/trunk/tools/opt/Makefile Wed Oct 21 19:46:41 2009
@@ -8,7 +8,6 @@
 ##===----------------------------------------------------------------------===##
 LEVEL = ../..
 TOOLNAME = opt
-REQUIRES_EH := 1
 
 LINK_COMPONENTS := bitreader bitwriter asmparser instrumentation scalaropts ipo
 

Modified: llvm/trunk/tools/opt/opt.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/opt/opt.cpp?rev=84816&r1=84815&r2=84816&view=diff

==============================================================================
--- llvm/trunk/tools/opt/opt.cpp (original)
+++ llvm/trunk/tools/opt/opt.cpp Wed Oct 21 19:46:41 2009
@@ -346,198 +346,187 @@
 int main(int argc, char **argv) {
   llvm_shutdown_obj X;  // Call llvm_shutdown() on exit.
   LLVMContext &Context = getGlobalContext();
-  try {
-    cl::ParseCommandLineOptions(argc, argv,
-      "llvm .bc -> .bc modular optimizer and analysis printer\n");
-    sys::PrintStackTraceOnErrorSignal();
-
-    // Allocate a full target machine description only if necessary.
-    // FIXME: The choice of target should be controllable on the command line.
-    std::auto_ptr<TargetMachine> target;
-
-    SMDiagnostic Err;
-
-    // Load the input module...
-    std::auto_ptr<Module> M;
-    M.reset(ParseIRFile(InputFilename, Err, Context));
+  
+  cl::ParseCommandLineOptions(argc, argv,
+    "llvm .bc -> .bc modular optimizer and analysis printer\n");
+  sys::PrintStackTraceOnErrorSignal();
+
+  // Allocate a full target machine description only if necessary.
+  // FIXME: The choice of target should be controllable on the command line.
+  std::auto_ptr<TargetMachine> target;
+
+  SMDiagnostic Err;
+
+  // Load the input module...
+  std::auto_ptr<Module> M;
+  M.reset(ParseIRFile(InputFilename, Err, Context));
+
+  if (M.get() == 0) {
+    Err.Print(argv[0], errs());
+    return 1;
+  }
 
-    if (M.get() == 0) {
-      Err.Print(argv[0], errs());
+  // Figure out what stream we are supposed to write to...
+  // FIXME: outs() is not binary!
+  raw_ostream *Out = &outs();  // Default to printing to stdout...
+  if (OutputFilename != "-") {
+    // Make sure that the Output file gets unlinked from the disk if we get a
+    // SIGINT
+    sys::RemoveFileOnSignal(sys::Path(OutputFilename));
+
+    std::string ErrorInfo;
+    Out = new raw_fd_ostream(OutputFilename.c_str(), ErrorInfo,
+                             raw_fd_ostream::F_Binary);
+    if (!ErrorInfo.empty()) {
+      errs() << ErrorInfo << '\n';
+      delete Out;
       return 1;
     }
+  }
 
-    // Figure out what stream we are supposed to write to...
-    // FIXME: outs() is not binary!
-    raw_ostream *Out = &outs();  // Default to printing to stdout...
-    if (OutputFilename != "-") {
-      // Make sure that the Output file gets unlinked from the disk if we get a
-      // SIGINT
-      sys::RemoveFileOnSignal(sys::Path(OutputFilename));
-
-      std::string ErrorInfo;
-      Out = new raw_fd_ostream(OutputFilename.c_str(), ErrorInfo,
-                               raw_fd_ostream::F_Binary);
-      if (!ErrorInfo.empty()) {
-        errs() << ErrorInfo << '\n';
-        delete Out;
-        return 1;
-      }
-    }
-
-    // If the output is set to be emitted to standard out, and standard out is a
-    // console, print out a warning message and refuse to do it.  We don't
-    // impress anyone by spewing tons of binary goo to a terminal.
-    if (!Force && !NoOutput && !OutputAssembly)
-      if (CheckBitcodeOutputToConsole(*Out, !Quiet))
-        NoOutput = true;
-
-    // Create a PassManager to hold and optimize the collection of passes we are
-    // about to build...
-    //
-    PassManager Passes;
-
-    // Add an appropriate TargetData instance for this module...
-    TargetData *TD = 0;
-    const std::string &ModuleDataLayout = M.get()->getDataLayout();
-    if (!ModuleDataLayout.empty())
-      TD = new TargetData(ModuleDataLayout);
-    else if (!NoDefaultDataLayout)
-      TD = new TargetData(DefaultDataLayout);
-
+  // If the output is set to be emitted to standard out, and standard out is a
+  // console, print out a warning message and refuse to do it.  We don't
+  // impress anyone by spewing tons of binary goo to a terminal.
+  if (!Force && !NoOutput && !OutputAssembly)
+    if (CheckBitcodeOutputToConsole(*Out, !Quiet))
+      NoOutput = true;
+
+  // Create a PassManager to hold and optimize the collection of passes we are
+  // about to build...
+  //
+  PassManager Passes;
+
+  // Add an appropriate TargetData instance for this module...
+  TargetData *TD = 0;
+  const std::string &ModuleDataLayout = M.get()->getDataLayout();
+  if (!ModuleDataLayout.empty())
+    TD = new TargetData(ModuleDataLayout);
+  else if (!NoDefaultDataLayout)
+    TD = new TargetData(DefaultDataLayout);
+
+  if (TD)
+    Passes.add(TD);
+
+  FunctionPassManager *FPasses = NULL;
+  if (OptLevelO1 || OptLevelO2 || OptLevelO3) {
+    FPasses = new FunctionPassManager(new ExistingModuleProvider(M.get()));
     if (TD)
-      Passes.add(TD);
-
-    FunctionPassManager *FPasses = NULL;
-    if (OptLevelO1 || OptLevelO2 || OptLevelO3) {
-      FPasses = new FunctionPassManager(new ExistingModuleProvider(M.get()));
-      if (TD)
-        FPasses->add(new TargetData(*TD));
-    }
-
-    // If the -strip-debug command line option was specified, add it.  If
-    // -std-compile-opts was also specified, it will handle StripDebug.
-    if (StripDebug && !StandardCompileOpts)
-      addPass(Passes, createStripSymbolsPass(true));
-
-    // Create a new optimization pass for each one specified on the command line
-    for (unsigned i = 0; i < PassList.size(); ++i) {
-      // Check to see if -std-compile-opts was specified before this option.  If
-      // so, handle it.
-      if (StandardCompileOpts &&
-          StandardCompileOpts.getPosition() < PassList.getPosition(i)) {
-        AddStandardCompilePasses(Passes);
-        StandardCompileOpts = false;
-      }
-
-      if (StandardLinkOpts &&
-          StandardLinkOpts.getPosition() < PassList.getPosition(i)) {
-        AddStandardLinkPasses(Passes);
-        StandardLinkOpts = false;
-      }
-
-      if (OptLevelO1 && OptLevelO1.getPosition() < PassList.getPosition(i)) {
-        AddOptimizationPasses(Passes, *FPasses, 1);
-        OptLevelO1 = false;
-      }
-
-      if (OptLevelO2 && OptLevelO2.getPosition() < PassList.getPosition(i)) {
-        AddOptimizationPasses(Passes, *FPasses, 2);
-        OptLevelO2 = false;
-      }
-
-      if (OptLevelO3 && OptLevelO3.getPosition() < PassList.getPosition(i)) {
-        AddOptimizationPasses(Passes, *FPasses, 3);
-        OptLevelO3 = false;
-      }
-
-      const PassInfo *PassInf = PassList[i];
-      Pass *P = 0;
-      if (PassInf->getNormalCtor())
-        P = PassInf->getNormalCtor()();
-      else
-        errs() << argv[0] << ": cannot create pass: "
-               << PassInf->getPassName() << "\n";
-      if (P) {
-        bool isBBPass = dynamic_cast<BasicBlockPass*>(P) != 0;
-        bool isLPass = !isBBPass && dynamic_cast<LoopPass*>(P) != 0;
-        bool isFPass = !isLPass && dynamic_cast<FunctionPass*>(P) != 0;
-        bool isCGSCCPass = !isFPass && dynamic_cast<CallGraphSCCPass*>(P) != 0;
-
-        addPass(Passes, P);
-
-        if (AnalyzeOnly) {
-          if (isBBPass)
-            Passes.add(new BasicBlockPassPrinter(PassInf));
-          else if (isLPass)
-            Passes.add(new LoopPassPrinter(PassInf));
-          else if (isFPass)
-            Passes.add(new FunctionPassPrinter(PassInf));
-          else if (isCGSCCPass)
-            Passes.add(new CallGraphSCCPassPrinter(PassInf));
-          else
-            Passes.add(new ModulePassPrinter(PassInf));
-        }
-      }
-
-      if (PrintEachXForm)
-        Passes.add(createPrintModulePass(&errs()));
-    }
+      FPasses->add(new TargetData(*TD));
+  }
 
-    // If -std-compile-opts was specified at the end of the pass list, add them.
-    if (StandardCompileOpts) {
+  // If the -strip-debug command line option was specified, add it.  If
+  // -std-compile-opts was also specified, it will handle StripDebug.
+  if (StripDebug && !StandardCompileOpts)
+    addPass(Passes, createStripSymbolsPass(true));
+
+  // Create a new optimization pass for each one specified on the command line
+  for (unsigned i = 0; i < PassList.size(); ++i) {
+    // Check to see if -std-compile-opts was specified before this option.  If
+    // so, handle it.
+    if (StandardCompileOpts &&
+        StandardCompileOpts.getPosition() < PassList.getPosition(i)) {
       AddStandardCompilePasses(Passes);
       StandardCompileOpts = false;
     }
 
-    if (StandardLinkOpts) {
+    if (StandardLinkOpts &&
+        StandardLinkOpts.getPosition() < PassList.getPosition(i)) {
       AddStandardLinkPasses(Passes);
       StandardLinkOpts = false;
     }
 
-    if (OptLevelO1) {
+    if (OptLevelO1 && OptLevelO1.getPosition() < PassList.getPosition(i)) {
       AddOptimizationPasses(Passes, *FPasses, 1);
+      OptLevelO1 = false;
     }
 
-    if (OptLevelO2) {
+    if (OptLevelO2 && OptLevelO2.getPosition() < PassList.getPosition(i)) {
       AddOptimizationPasses(Passes, *FPasses, 2);
+      OptLevelO2 = false;
     }
 
-    if (OptLevelO3) {
+    if (OptLevelO3 && OptLevelO3.getPosition() < PassList.getPosition(i)) {
       AddOptimizationPasses(Passes, *FPasses, 3);
+      OptLevelO3 = false;
     }
 
-    if (OptLevelO1 || OptLevelO2 || OptLevelO3) {
-      FPasses->doInitialization();
-      for (Module::iterator I = M.get()->begin(), E = M.get()->end();
-           I != E; ++I)
-        FPasses->run(*I);
-    }
-
-    // Check that the module is well formed on completion of optimization
-    if (!NoVerify && !VerifyEach)
-      Passes.add(createVerifierPass());
-
-    // Write bitcode or assembly  out to disk or outs() as the last step...
-    if (!NoOutput && !AnalyzeOnly) {
-      if (OutputAssembly)
-        Passes.add(createPrintModulePass(Out));
-      else
-        Passes.add(createBitcodeWriterPass(*Out));
+    const PassInfo *PassInf = PassList[i];
+    Pass *P = 0;
+    if (PassInf->getNormalCtor())
+      P = PassInf->getNormalCtor()();
+    else
+      errs() << argv[0] << ": cannot create pass: "
+             << PassInf->getPassName() << "\n";
+    if (P) {
+      bool isBBPass = dynamic_cast<BasicBlockPass*>(P) != 0;
+      bool isLPass = !isBBPass && dynamic_cast<LoopPass*>(P) != 0;
+      bool isFPass = !isLPass && dynamic_cast<FunctionPass*>(P) != 0;
+      bool isCGSCCPass = !isFPass && dynamic_cast<CallGraphSCCPass*>(P) != 0;
+
+      addPass(Passes, P);
+
+      if (AnalyzeOnly) {
+        if (isBBPass)
+          Passes.add(new BasicBlockPassPrinter(PassInf));
+        else if (isLPass)
+          Passes.add(new LoopPassPrinter(PassInf));
+        else if (isFPass)
+          Passes.add(new FunctionPassPrinter(PassInf));
+        else if (isCGSCCPass)
+          Passes.add(new CallGraphSCCPassPrinter(PassInf));
+        else
+          Passes.add(new ModulePassPrinter(PassInf));
+      }
     }
 
-    // Now that we have all of the passes ready, run them.
-    Passes.run(*M.get());
+    if (PrintEachXForm)
+      Passes.add(createPrintModulePass(&errs()));
+  }
 
-    // Delete the raw_fd_ostream.
-    if (Out != &outs())
-      delete Out;
-    return 0;
+  // If -std-compile-opts was specified at the end of the pass list, add them.
+  if (StandardCompileOpts) {
+    AddStandardCompilePasses(Passes);
+    StandardCompileOpts = false;
+  }
 
-  } catch (const std::string& msg) {
-    errs() << argv[0] << ": " << msg << "\n";
-  } catch (...) {
-    errs() << argv[0] << ": Unexpected unknown exception occurred.\n";
+  if (StandardLinkOpts) {
+    AddStandardLinkPasses(Passes);
+    StandardLinkOpts = false;
   }
-  llvm_shutdown();
-  return 1;
+
+  if (OptLevelO1)
+    AddOptimizationPasses(Passes, *FPasses, 1);
+
+  if (OptLevelO2)
+    AddOptimizationPasses(Passes, *FPasses, 2);
+
+  if (OptLevelO3)
+    AddOptimizationPasses(Passes, *FPasses, 3);
+
+  if (OptLevelO1 || OptLevelO2 || OptLevelO3) {
+    FPasses->doInitialization();
+    for (Module::iterator I = M.get()->begin(), E = M.get()->end();
+         I != E; ++I)
+      FPasses->run(*I);
+  }
+
+  // Check that the module is well formed on completion of optimization
+  if (!NoVerify && !VerifyEach)
+    Passes.add(createVerifierPass());
+
+  // Write bitcode or assembly  out to disk or outs() as the last step...
+  if (!NoOutput && !AnalyzeOnly) {
+    if (OutputAssembly)
+      Passes.add(createPrintModulePass(Out));
+    else
+      Passes.add(createBitcodeWriterPass(*Out));
+  }
+
+  // Now that we have all of the passes ready, run them.
+  Passes.run(*M.get());
+
+  // Delete the raw_fd_ostream.
+  if (Out != &outs())
+    delete Out;
+  return 0;
 }




From sabre at nondot.org  Wed Oct 21 19:50:25 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 00:50:25 -0000
Subject: [llvm-commits] [llvm] r84818 - in /llvm/trunk/tools/llvm-prof:
 CMakeLists.txt Makefile llvm-prof.cpp
Message-ID: <200910220050.n9M0oPYg018434@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 19:50:24 2009
New Revision: 84818

URL: http://llvm.org/viewvc/llvm-project?rev=84818&view=rev
Log:
this doesn't use EH either.

Modified:
    llvm/trunk/tools/llvm-prof/CMakeLists.txt
    llvm/trunk/tools/llvm-prof/Makefile
    llvm/trunk/tools/llvm-prof/llvm-prof.cpp

Modified: llvm/trunk/tools/llvm-prof/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-prof/CMakeLists.txt?rev=84818&r1=84817&r2=84818&view=diff

==============================================================================
--- llvm/trunk/tools/llvm-prof/CMakeLists.txt (original)
+++ llvm/trunk/tools/llvm-prof/CMakeLists.txt Wed Oct 21 19:50:24 2009
@@ -1,5 +1,4 @@
 set(LLVM_LINK_COMPONENTS bitreader analysis)
-set(LLVM_REQUIRES_EH 1)
 
 add_llvm_tool(llvm-prof
   llvm-prof.cpp

Modified: llvm/trunk/tools/llvm-prof/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-prof/Makefile?rev=84818&r1=84817&r2=84818&view=diff

==============================================================================
--- llvm/trunk/tools/llvm-prof/Makefile (original)
+++ llvm/trunk/tools/llvm-prof/Makefile Wed Oct 21 19:50:24 2009
@@ -10,7 +10,6 @@
 
 TOOLNAME = llvm-prof
 LINK_COMPONENTS = bitreader analysis
-REQUIRES_EH := 1
 
 # This tool has no plugins, optimize startup time.
 TOOL_NO_EXPORTS = 1

Modified: llvm/trunk/tools/llvm-prof/llvm-prof.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-prof/llvm-prof.cpp?rev=84818&r1=84817&r2=84818&view=diff

==============================================================================
--- llvm/trunk/tools/llvm-prof/llvm-prof.cpp (original)
+++ llvm/trunk/tools/llvm-prof/llvm-prof.cpp Wed Oct 21 19:50:24 2009
@@ -255,41 +255,34 @@
 
   LLVMContext &Context = getGlobalContext();
   llvm_shutdown_obj Y;  // Call llvm_shutdown() on exit.
-  try {
-    cl::ParseCommandLineOptions(argc, argv, "llvm profile dump decoder\n");
+  
+  cl::ParseCommandLineOptions(argc, argv, "llvm profile dump decoder\n");
 
-    // Read in the bitcode file...
-    std::string ErrorMessage;
-    Module *M = 0;
-    if (MemoryBuffer *Buffer = MemoryBuffer::getFileOrSTDIN(BitcodeFile,
-                                                            &ErrorMessage)) {
-      M = ParseBitcodeFile(Buffer, Context, &ErrorMessage);
-      delete Buffer;
-    }
-    if (M == 0) {
-      errs() << argv[0] << ": " << BitcodeFile << ": "
-        << ErrorMessage << "\n";
-      return 1;
-    }
+  // Read in the bitcode file...
+  std::string ErrorMessage;
+  Module *M = 0;
+  if (MemoryBuffer *Buffer = MemoryBuffer::getFileOrSTDIN(BitcodeFile,
+                                                          &ErrorMessage)) {
+    M = ParseBitcodeFile(Buffer, Context, &ErrorMessage);
+    delete Buffer;
+  }
+  if (M == 0) {
+    errs() << argv[0] << ": " << BitcodeFile << ": "
+      << ErrorMessage << "\n";
+    return 1;
+  }
 
-    // Read the profiling information. This is redundant since we load it again
-    // using the standard profile info provider pass, but for now this gives us
-    // access to additional information not exposed via the ProfileInfo
-    // interface.
-    ProfileInfoLoader PIL(argv[0], ProfileDataFile, *M);
+  // Read the profiling information. This is redundant since we load it again
+  // using the standard profile info provider pass, but for now this gives us
+  // access to additional information not exposed via the ProfileInfo
+  // interface.
+  ProfileInfoLoader PIL(argv[0], ProfileDataFile, *M);
 
-    // Run the printer pass.
-    PassManager PassMgr;
-    PassMgr.add(createProfileLoaderPass(ProfileDataFile));
-    PassMgr.add(new ProfileInfoPrinterPass(PIL));
-    PassMgr.run(*M);
+  // Run the printer pass.
+  PassManager PassMgr;
+  PassMgr.add(createProfileLoaderPass(ProfileDataFile));
+  PassMgr.add(new ProfileInfoPrinterPass(PIL));
+  PassMgr.run(*M);
 
-    return 0;
-  } catch (const std::string& msg) {
-    errs() << argv[0] << ": " << msg << "\n";
-  } catch (...) {
-    errs() << argv[0] << ": Unexpected unknown exception occurred.\n";
-  }
-  
-  return 1;
+  return 0;
 }




From sabre at nondot.org  Wed Oct 21 19:52:28 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 00:52:28 -0000
Subject: [llvm-commits] [llvm] r84819 - in /llvm/trunk/tools/llvm-ld:
 CMakeLists.txt Makefile llvm-ld.cpp
Message-ID: <200910220052.n9M0qTDj018525@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 19:52:28 2009
New Revision: 84819

URL: http://llvm.org/viewvc/llvm-project?rev=84819&view=rev
Log:
llvm-ld doesn't throw.

Modified:
    llvm/trunk/tools/llvm-ld/CMakeLists.txt
    llvm/trunk/tools/llvm-ld/Makefile
    llvm/trunk/tools/llvm-ld/llvm-ld.cpp

Modified: llvm/trunk/tools/llvm-ld/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-ld/CMakeLists.txt?rev=84819&r1=84818&r2=84819&view=diff

==============================================================================
--- llvm/trunk/tools/llvm-ld/CMakeLists.txt (original)
+++ llvm/trunk/tools/llvm-ld/CMakeLists.txt Wed Oct 21 19:52:28 2009
@@ -1,5 +1,4 @@
 set(LLVM_LINK_COMPONENTS ipo scalaropts linker archive bitwriter)
-set(LLVM_REQUIRES_EH 1)
 
 add_llvm_tool(llvm-ld
   Optimize.cpp

Modified: llvm/trunk/tools/llvm-ld/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-ld/Makefile?rev=84819&r1=84818&r2=84819&view=diff

==============================================================================
--- llvm/trunk/tools/llvm-ld/Makefile (original)
+++ llvm/trunk/tools/llvm-ld/Makefile Wed Oct 21 19:52:28 2009
@@ -11,6 +11,5 @@
 
 TOOLNAME = llvm-ld
 LINK_COMPONENTS = ipo scalaropts linker archive bitwriter
-REQUIRES_EH := 1
 
 include $(LEVEL)/Makefile.common

Modified: llvm/trunk/tools/llvm-ld/llvm-ld.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/llvm-ld/llvm-ld.cpp?rev=84819&r1=84818&r2=84819&view=diff

==============================================================================
--- llvm/trunk/tools/llvm-ld/llvm-ld.cpp (original)
+++ llvm/trunk/tools/llvm-ld/llvm-ld.cpp Wed Oct 21 19:52:28 2009
@@ -513,205 +513,200 @@
 
   LLVMContext &Context = getGlobalContext();
   llvm_shutdown_obj Y;  // Call llvm_shutdown() on exit.
-  try {
-    // Initial global variable above for convenience printing of program name.
-    progname = sys::Path(argv[0]).getBasename();
-
-    // Parse the command line options
-    cl::ParseCommandLineOptions(argc, argv, "llvm linker\n");
-
-    // Construct a Linker (now that Verbose is set)
-    Linker TheLinker(progname, OutputFilename, Context, Verbose);
-
-    // Keep track of the native link items (versus the bitcode items)
-    Linker::ItemList NativeLinkItems;
-
-    // Add library paths to the linker
-    TheLinker.addPaths(LibPaths);
-    TheLinker.addSystemPaths();
-
-    // Remove any consecutive duplicates of the same library...
-    Libraries.erase(std::unique(Libraries.begin(), Libraries.end()),
-                    Libraries.end());
-
-    if (LinkAsLibrary) {
-      std::vector<sys::Path> Files;
-      for (unsigned i = 0; i < InputFilenames.size(); ++i )
-        Files.push_back(sys::Path(InputFilenames[i]));
-      if (TheLinker.LinkInFiles(Files))
-        return 1; // Error already printed
-
-      // The libraries aren't linked in but are noted as "dependent" in the
-      // module.
-      for (cl::list<std::string>::const_iterator I = Libraries.begin(),
-           E = Libraries.end(); I != E ; ++I) {
-        TheLinker.getModule()->addLibrary(*I);
-      }
-    } else {
-      // Build a list of the items from our command line
-      Linker::ItemList Items;
-      BuildLinkItems(Items, InputFilenames, Libraries);
-
-      // Link all the items together
-      if (TheLinker.LinkInItems(Items, NativeLinkItems) )
-        return 1; // Error already printed
+  
+  // Initial global variable above for convenience printing of program name.
+  progname = sys::Path(argv[0]).getBasename();
+
+  // Parse the command line options
+  cl::ParseCommandLineOptions(argc, argv, "llvm linker\n");
+
+  // Construct a Linker (now that Verbose is set)
+  Linker TheLinker(progname, OutputFilename, Context, Verbose);
+
+  // Keep track of the native link items (versus the bitcode items)
+  Linker::ItemList NativeLinkItems;
+
+  // Add library paths to the linker
+  TheLinker.addPaths(LibPaths);
+  TheLinker.addSystemPaths();
+
+  // Remove any consecutive duplicates of the same library...
+  Libraries.erase(std::unique(Libraries.begin(), Libraries.end()),
+                  Libraries.end());
+
+  if (LinkAsLibrary) {
+    std::vector<sys::Path> Files;
+    for (unsigned i = 0; i < InputFilenames.size(); ++i )
+      Files.push_back(sys::Path(InputFilenames[i]));
+    if (TheLinker.LinkInFiles(Files))
+      return 1; // Error already printed
+
+    // The libraries aren't linked in but are noted as "dependent" in the
+    // module.
+    for (cl::list<std::string>::const_iterator I = Libraries.begin(),
+         E = Libraries.end(); I != E ; ++I) {
+      TheLinker.getModule()->addLibrary(*I);
     }
+  } else {
+    // Build a list of the items from our command line
+    Linker::ItemList Items;
+    BuildLinkItems(Items, InputFilenames, Libraries);
+
+    // Link all the items together
+    if (TheLinker.LinkInItems(Items, NativeLinkItems) )
+      return 1; // Error already printed
+  }
 
-    std::auto_ptr<Module> Composite(TheLinker.releaseModule());
+  std::auto_ptr<Module> Composite(TheLinker.releaseModule());
 
-    // Optimize the module
-    Optimize(Composite.get());
+  // Optimize the module
+  Optimize(Composite.get());
 
 #if defined(_WIN32) || defined(__CYGWIN__)
-    if (!LinkAsLibrary) {
-      // Default to "a.exe" instead of "a.out".
-      if (OutputFilename.getNumOccurrences() == 0)
-        OutputFilename = "a.exe";
-
-      // If there is no suffix add an "exe" one.
-      sys::Path ExeFile( OutputFilename );
-      if (ExeFile.getSuffix() == "") {
-        ExeFile.appendSuffix("exe");
-        OutputFilename = ExeFile.str();
-      }
+  if (!LinkAsLibrary) {
+    // Default to "a.exe" instead of "a.out".
+    if (OutputFilename.getNumOccurrences() == 0)
+      OutputFilename = "a.exe";
+
+    // If there is no suffix add an "exe" one.
+    sys::Path ExeFile( OutputFilename );
+    if (ExeFile.getSuffix() == "") {
+      ExeFile.appendSuffix("exe");
+      OutputFilename = ExeFile.str();
     }
+  }
 #endif
 
-    // Generate the bitcode for the optimized module.
-    // If -b wasn't specified, use the name specified
-    // with -o to construct BitcodeOutputFilename.
-    if (BitcodeOutputFilename.empty()) {
-      BitcodeOutputFilename = OutputFilename;
-      if (!LinkAsLibrary) BitcodeOutputFilename += ".bc";
-    }
+  // Generate the bitcode for the optimized module.
+  // If -b wasn't specified, use the name specified
+  // with -o to construct BitcodeOutputFilename.
+  if (BitcodeOutputFilename.empty()) {
+    BitcodeOutputFilename = OutputFilename;
+    if (!LinkAsLibrary) BitcodeOutputFilename += ".bc";
+  }
 
-    GenerateBitcode(Composite.get(), BitcodeOutputFilename);
+  GenerateBitcode(Composite.get(), BitcodeOutputFilename);
 
-    // If we are not linking a library, generate either a native executable
-    // or a JIT shell script, depending upon what the user wants.
-    if (!LinkAsLibrary) {
-      // If the user wants to run a post-link optimization, run it now.
-      if (!PostLinkOpts.empty()) {
-        std::vector<std::string> opts = PostLinkOpts;
-        for (std::vector<std::string>::iterator I = opts.begin(),
-             E = opts.end(); I != E; ++I) {
-          sys::Path prog(*I);
-          if (!prog.canExecute()) {
-            prog = sys::Program::FindProgramByName(*I);
-            if (prog.isEmpty())
-              PrintAndExit(std::string("Optimization program '") + *I +
-                "' is not found or not executable.");
-          }
-          // Get the program arguments
-          sys::Path tmp_output("opt_result");
-          std::string ErrMsg;
-          if (tmp_output.createTemporaryFileOnDisk(true, &ErrMsg))
-            PrintAndExit(ErrMsg);
-
-          const char* args[4];
-          args[0] = I->c_str();
-          args[1] = BitcodeOutputFilename.c_str();
-          args[2] = tmp_output.c_str();
-          args[3] = 0;
-          if (0 == sys::Program::ExecuteAndWait(prog, args, 0,0,0,0, &ErrMsg)) {
-            if (tmp_output.isBitcodeFile() || tmp_output.isBitcodeFile()) {
-              sys::Path target(BitcodeOutputFilename);
-              target.eraseFromDisk();
-              if (tmp_output.renamePathOnDisk(target, &ErrMsg))
-                PrintAndExit(ErrMsg, 2);
-            } else
-              PrintAndExit("Post-link optimization output is not bitcode");
-          } else {
-            PrintAndExit(ErrMsg);
-          }
+  // If we are not linking a library, generate either a native executable
+  // or a JIT shell script, depending upon what the user wants.
+  if (!LinkAsLibrary) {
+    // If the user wants to run a post-link optimization, run it now.
+    if (!PostLinkOpts.empty()) {
+      std::vector<std::string> opts = PostLinkOpts;
+      for (std::vector<std::string>::iterator I = opts.begin(),
+           E = opts.end(); I != E; ++I) {
+        sys::Path prog(*I);
+        if (!prog.canExecute()) {
+          prog = sys::Program::FindProgramByName(*I);
+          if (prog.isEmpty())
+            PrintAndExit(std::string("Optimization program '") + *I +
+              "' is not found or not executable.");
         }
-      }
-
-      // If the user wants to generate a native executable, compile it from the
-      // bitcode file.
-      //
-      // Otherwise, create a script that will run the bitcode through the JIT.
-      if (Native) {
-        // Name of the Assembly Language output file
-        sys::Path AssemblyFile ( OutputFilename);
-        AssemblyFile.appendSuffix("s");
-
-        // Mark the output files for removal if we get an interrupt.
-        sys::RemoveFileOnSignal(AssemblyFile);
-        sys::RemoveFileOnSignal(sys::Path(OutputFilename));
-
-        // Determine the locations of the llc and gcc programs.
-        sys::Path llc = FindExecutable("llc", argv[0],
-                                       (void *)(intptr_t)&Optimize);
-        if (llc.isEmpty())
-          PrintAndExit("Failed to find llc");
-
-        sys::Path gcc = sys::Program::FindProgramByName("gcc");
-        if (gcc.isEmpty())
-          PrintAndExit("Failed to find gcc");
-
-        // Generate an assembly language file for the bitcode.
+        // Get the program arguments
+        sys::Path tmp_output("opt_result");
         std::string ErrMsg;
-        if (0 != GenerateAssembly(AssemblyFile.str(), BitcodeOutputFilename,
-            llc, ErrMsg))
+        if (tmp_output.createTemporaryFileOnDisk(true, &ErrMsg))
           PrintAndExit(ErrMsg);
 
-        if (0 != GenerateNative(OutputFilename, AssemblyFile.str(),
-                                NativeLinkItems, gcc, envp, ErrMsg))
+        const char* args[4];
+        args[0] = I->c_str();
+        args[1] = BitcodeOutputFilename.c_str();
+        args[2] = tmp_output.c_str();
+        args[3] = 0;
+        if (0 == sys::Program::ExecuteAndWait(prog, args, 0,0,0,0, &ErrMsg)) {
+          if (tmp_output.isBitcodeFile() || tmp_output.isBitcodeFile()) {
+            sys::Path target(BitcodeOutputFilename);
+            target.eraseFromDisk();
+            if (tmp_output.renamePathOnDisk(target, &ErrMsg))
+              PrintAndExit(ErrMsg, 2);
+          } else
+            PrintAndExit("Post-link optimization output is not bitcode");
+        } else {
           PrintAndExit(ErrMsg);
+        }
+      }
+    }
 
-        // Remove the assembly language file.
-        AssemblyFile.eraseFromDisk();
-      } else if (NativeCBE) {
-        sys::Path CFile (OutputFilename);
-        CFile.appendSuffix("cbe.c");
-
-        // Mark the output files for removal if we get an interrupt.
-        sys::RemoveFileOnSignal(CFile);
-        sys::RemoveFileOnSignal(sys::Path(OutputFilename));
-
-        // Determine the locations of the llc and gcc programs.
-        sys::Path llc = FindExecutable("llc", argv[0],
-                                       (void *)(intptr_t)&Optimize);
-        if (llc.isEmpty())
-          PrintAndExit("Failed to find llc");
-
-        sys::Path gcc = sys::Program::FindProgramByName("gcc");
-        if (gcc.isEmpty())
-          PrintAndExit("Failed to find gcc");
-
-        // Generate an assembly language file for the bitcode.
-        std::string ErrMsg;
-        if (GenerateCFile(CFile.str(), BitcodeOutputFilename, llc, ErrMsg))
-          PrintAndExit(ErrMsg);
+    // If the user wants to generate a native executable, compile it from the
+    // bitcode file.
+    //
+    // Otherwise, create a script that will run the bitcode through the JIT.
+    if (Native) {
+      // Name of the Assembly Language output file
+      sys::Path AssemblyFile ( OutputFilename);
+      AssemblyFile.appendSuffix("s");
+
+      // Mark the output files for removal if we get an interrupt.
+      sys::RemoveFileOnSignal(AssemblyFile);
+      sys::RemoveFileOnSignal(sys::Path(OutputFilename));
+
+      // Determine the locations of the llc and gcc programs.
+      sys::Path llc = FindExecutable("llc", argv[0],
+                                     (void *)(intptr_t)&Optimize);
+      if (llc.isEmpty())
+        PrintAndExit("Failed to find llc");
+
+      sys::Path gcc = sys::Program::FindProgramByName("gcc");
+      if (gcc.isEmpty())
+        PrintAndExit("Failed to find gcc");
 
-        if (GenerateNative(OutputFilename, CFile.str(), 
-                           NativeLinkItems, gcc, envp, ErrMsg))
-          PrintAndExit(ErrMsg);
+      // Generate an assembly language file for the bitcode.
+      std::string ErrMsg;
+      if (0 != GenerateAssembly(AssemblyFile.str(), BitcodeOutputFilename,
+          llc, ErrMsg))
+        PrintAndExit(ErrMsg);
 
-        // Remove the assembly language file.
-        CFile.eraseFromDisk();
+      if (0 != GenerateNative(OutputFilename, AssemblyFile.str(),
+                              NativeLinkItems, gcc, envp, ErrMsg))
+        PrintAndExit(ErrMsg);
 
-      } else {
-        EmitShellScript(argv);
-      }
+      // Remove the assembly language file.
+      AssemblyFile.eraseFromDisk();
+    } else if (NativeCBE) {
+      sys::Path CFile (OutputFilename);
+      CFile.appendSuffix("cbe.c");
+
+      // Mark the output files for removal if we get an interrupt.
+      sys::RemoveFileOnSignal(CFile);
+      sys::RemoveFileOnSignal(sys::Path(OutputFilename));
+
+      // Determine the locations of the llc and gcc programs.
+      sys::Path llc = FindExecutable("llc", argv[0],
+                                     (void *)(intptr_t)&Optimize);
+      if (llc.isEmpty())
+        PrintAndExit("Failed to find llc");
+
+      sys::Path gcc = sys::Program::FindProgramByName("gcc");
+      if (gcc.isEmpty())
+        PrintAndExit("Failed to find gcc");
 
-      // Make the script executable...
+      // Generate an assembly language file for the bitcode.
       std::string ErrMsg;
-      if (sys::Path(OutputFilename).makeExecutableOnDisk(&ErrMsg))
+      if (GenerateCFile(CFile.str(), BitcodeOutputFilename, llc, ErrMsg))
         PrintAndExit(ErrMsg);
 
-      // Make the bitcode file readable and directly executable in LLEE as well
-      if (sys::Path(BitcodeOutputFilename).makeExecutableOnDisk(&ErrMsg))
+      if (GenerateNative(OutputFilename, CFile.str(), 
+                         NativeLinkItems, gcc, envp, ErrMsg))
         PrintAndExit(ErrMsg);
 
-      if (sys::Path(BitcodeOutputFilename).makeReadableOnDisk(&ErrMsg))
-        PrintAndExit(ErrMsg);
+      // Remove the assembly language file.
+      CFile.eraseFromDisk();
+
+    } else {
+      EmitShellScript(argv);
     }
-  } catch (const std::string& msg) {
-    PrintAndExit(msg,2);
-  } catch (...) {
-    PrintAndExit("Unexpected unknown exception occurred.", 2);
+
+    // Make the script executable...
+    std::string ErrMsg;
+    if (sys::Path(OutputFilename).makeExecutableOnDisk(&ErrMsg))
+      PrintAndExit(ErrMsg);
+
+    // Make the bitcode file readable and directly executable in LLEE as well
+    if (sys::Path(BitcodeOutputFilename).makeExecutableOnDisk(&ErrMsg))
+      PrintAndExit(ErrMsg);
+
+    if (sys::Path(BitcodeOutputFilename).makeReadableOnDisk(&ErrMsg))
+      PrintAndExit(ErrMsg);
   }
 
   // Graceful exit




From clattner at apple.com  Wed Oct 21 19:54:02 2009
From: clattner at apple.com (Chris Lattner)
Date: Wed, 21 Oct 2009 17:54:02 -0700
Subject: [llvm-commits] [llvm] r84795 - in /llvm/trunk:
	docs/LangRef.html include/llvm/InlineAsm.h
	lib/AsmParser/LLLexer.cpp lib/AsmParser/LLParser.cpp
	lib/AsmParser/LLToken.h lib/Bitcode/Reader/BitcodeReader.cpp
	lib/Bitcode/Writer/BitcodeWriter.cpp lib/VMCore/AsmWriter.cpp
	lib/VMCore/Core.cpp lib/VMCore/InlineAsm.cpp
In-Reply-To: <200910212328.n9LNS1ng014924@zion.cs.uiuc.edu>
References: <200910212328.n9LNS1ng014924@zion.cs.uiuc.edu>
Message-ID: <49662842-C925-4EF3-9F64-9E5A0CDEE8BC@apple.com>


On Oct 21, 2009, at 4:28 PM, Dale Johannesen wrote:

> Author: johannes
> Date: Wed Oct 21 18:28:00 2009
> New Revision: 84795
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84795&view=rev
> Log:
> Rename msasm to alignstack per review.

Nice, thanks Dale!

-Chris



From dpatel at apple.com  Wed Oct 21 20:01:24 2009
From: dpatel at apple.com (Devang Patel)
Date: Thu, 22 Oct 2009 01:01:24 -0000
Subject: [llvm-commits] [llvm] r84820 - in /llvm/trunk:
 include/llvm/Metadata.h lib/Bitcode/Writer/BitcodeWriter.cpp
 lib/VMCore/AsmWriter.cpp lib/VMCore/Metadata.cpp
Message-ID: <200910220101.n9M11PUF018859@zion.cs.uiuc.edu>

Author: dpatel
Date: Wed Oct 21 20:01:24 2009
New Revision: 84820

URL: http://llvm.org/viewvc/llvm-project?rev=84820&view=rev
Log:
Fix getHandlerNames() interface. Now it populate clinet supplied small vector with handler names.

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
    llvm/trunk/lib/VMCore/AsmWriter.cpp
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84820&r1=84819&r2=84820&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Wed Oct 21 20:01:24 2009
@@ -287,7 +287,7 @@
 
   /// getHandlerNames - Get handler names. This is used by bitcode
   /// writer.
-  const StringMap<unsigned> *getHandlerNames();
+  void getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef> >&N) const;
 
   /// ValueIsDeleted - This handler is used to update metadata store
   /// when a value is deleted.

Modified: llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp?rev=84820&r1=84819&r2=84820&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp (original)
+++ llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp Wed Oct 21 20:01:24 2009
@@ -602,11 +602,13 @@
   // Write metadata kinds
   // METADATA_KIND - [n x [id, name]]
   MetadataContext &TheMetadata = M->getContext().getMetadata();
-  const StringMap<unsigned> *Kinds = TheMetadata.getHandlerNames();
-  for (StringMap<unsigned>::const_iterator
-         I = Kinds->begin(), E = Kinds->end(); I != E; ++I) {
-    Record.push_back(I->second);
-    StringRef KName = I->first();
+  SmallVector<std::pair<unsigned, StringRef>, 4> Names;
+  TheMetadata.getHandlerNames(Names);
+  for (SmallVector<std::pair<unsigned, StringRef>, 4>::iterator 
+         I = Names.begin(),
+         E = Names.end(); I != E; ++I) {
+    Record.push_back(I->first);
+    StringRef KName = I->second;
     for (unsigned i = 0, e = KName.size(); i != e; ++i)
       Record.push_back(KName[i]);
     if (!StartedMetadataBlock)  {

Modified: llvm/trunk/lib/VMCore/AsmWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/AsmWriter.cpp?rev=84820&r1=84819&r2=84820&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/AsmWriter.cpp (original)
+++ llvm/trunk/lib/VMCore/AsmWriter.cpp Wed Oct 21 20:01:24 2009
@@ -1295,7 +1295,7 @@
   TypePrinting TypePrinter;
   AssemblyAnnotationWriter *AnnotationWriter;
   std::vector<const Type*> NumberedTypes;
-  DenseMap<unsigned, const char *> MDNames;
+  DenseMap<unsigned, StringRef> MDNames;
 
 public:
   inline AssemblyWriter(formatted_raw_ostream &o, SlotTracker &Mac,
@@ -1306,11 +1306,12 @@
     // FIXME: Provide MDPrinter
     if (M) {
       MetadataContext &TheMetadata = M->getContext().getMetadata();
-      const StringMap<unsigned> *Names = TheMetadata.getHandlerNames();
-      for (StringMapConstIterator<unsigned> I = Names->begin(),
-             E = Names->end(); I != E; ++I) {
-        const StringMapEntry<unsigned> &Entry = *I;
-        MDNames[I->second] = Entry.getKeyData();
+      SmallVector<std::pair<unsigned, StringRef>, 4> Names;
+      TheMetadata.getHandlerNames(Names);
+      for (SmallVector<std::pair<unsigned, StringRef>, 4>::iterator 
+             I = Names.begin(),
+             E = Names.end(); I != E; ++I) {
+      MDNames[I->first] = I->second;
       }
     }
   }

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84820&r1=84819&r2=84820&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Wed Oct 21 20:01:24 2009
@@ -305,9 +305,12 @@
 }
 
 /// getHandlerNames - Get handler names. This is used by bitcode
-/// writer.
-const StringMap<unsigned> *MetadataContext::getHandlerNames() {
-  return &MDHandlerNames;
+/// writer and aswm writer.
+void MetadataContext::
+getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef> >&Names) const {
+  for (StringMap<unsigned>::const_iterator I = MDHandlerNames.begin(),
+         E = MDHandlerNames.end(); I != E; ++I) 
+    Names.push_back(std::make_pair(I->second, I->first()));
 }
 
 /// ValueIsCloned - This handler is used to update metadata store




From sabre at nondot.org  Wed Oct 21 22:42:32 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 03:42:32 -0000
Subject: [llvm-commits] [llvm] r84826 -
	/llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp
Message-ID: <200910220342.n9M3gWYL024714@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 22:42:27 2009
New Revision: 84826

URL: http://llvm.org/viewvc/llvm-project?rev=84826&view=rev
Log:
fix warning.

Modified:
    llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp

Modified: llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp?rev=84826&r1=84825&r2=84826&view=diff

==============================================================================
--- llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp (original)
+++ llvm/trunk/lib/Target/PIC16/AsmPrinter/PIC16AsmPrinter.cpp Wed Oct 21 22:42:27 2009
@@ -57,7 +57,7 @@
   if (F->hasSection()) {
     std::string Sectn = F->getSection();
     std::string StrToFind = "Overlay=";
-    unsigned Pos = Sectn.find(StrToFind);
+    std::string::size_type Pos = Sectn.find(StrToFind);
 
     // Retreive the color number if the key is found.
     if (Pos != std::string::npos) {




From foldr at codedgers.com  Wed Oct 21 23:15:08 2009
From: foldr at codedgers.com (Mikhail Glushenkov)
Date: Thu, 22 Oct 2009 04:15:08 -0000
Subject: [llvm-commits] [llvm] r84827 -
	/llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp
Message-ID: <200910220415.n9M4F8K7025980@zion.cs.uiuc.edu>

Author: foldr
Date: Wed Oct 21 23:15:07 2009
New Revision: 84827

URL: http://llvm.org/viewvc/llvm-project?rev=84827&view=rev
Log:
Make 'unset_option' work on list options.

Modified:
    llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp

Modified: llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp?rev=84827&r1=84826&r2=84827&view=diff

==============================================================================
--- llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp (original)
+++ llvm/trunk/utils/TableGen/LLVMCConfigurationEmitter.cpp Wed Oct 21 23:15:07 2009
@@ -2127,18 +2127,18 @@
   void onUnsetOption(Init* i, unsigned IndentLevel, raw_ostream& O) {
     const std::string& OptName = InitPtrToString(i);
     const OptionDescription& OptDesc = OptDescs_.FindOption(OptName);
-    const OptionType::OptionType OptType = OptDesc.Type;
 
-    if (OptType == OptionType::Switch) {
+    if (OptDesc.isSwitch()) {
       O.indent(IndentLevel) << OptDesc.GenVariableName() << " = false;\n";
     }
-    else if (OptType == OptionType::Parameter
-             || OptType == OptionType::Prefix) {
+    else if (OptDesc.isParameter()) {
       O.indent(IndentLevel) << OptDesc.GenVariableName() << " = \"\";\n";
     }
+    else if (OptDesc.isList()) {
+      O.indent(IndentLevel) << OptDesc.GenVariableName() << ".clear();\n";
+    }
     else {
-      throw std::string("'unset_option' can only be applied to "
-                        "switches or parameter/prefix options.");
+      throw "Can't apply 'unset_option' to alias option '" + OptName + "'";
     }
   }
 




From foldr at codedgers.com  Wed Oct 21 23:15:24 2009
From: foldr at codedgers.com (Mikhail Glushenkov)
Date: Thu, 22 Oct 2009 04:15:24 -0000
Subject: [llvm-commits] [llvm] r84828 -
	/llvm/trunk/include/llvm/CompilerDriver/Common.td
Message-ID: <200910220415.n9M4FO1u026004@zion.cs.uiuc.edu>

Author: foldr
Date: Wed Oct 21 23:15:24 2009
New Revision: 84828

URL: http://llvm.org/viewvc/llvm-project?rev=84828&view=rev
Log:
Forgot a declaration.

Modified:
    llvm/trunk/include/llvm/CompilerDriver/Common.td

Modified: llvm/trunk/include/llvm/CompilerDriver/Common.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/CompilerDriver/Common.td?rev=84828&r1=84827&r2=84828&view=diff

==============================================================================
--- llvm/trunk/include/llvm/CompilerDriver/Common.td (original)
+++ llvm/trunk/include/llvm/CompilerDriver/Common.td Wed Oct 21 23:15:24 2009
@@ -70,6 +70,7 @@
 def multiple_input_files;
 def any_switch_on;
 def any_not_empty;
+def any_empty;
 
 // Possible actions.
 




From sabre at nondot.org  Wed Oct 21 23:47:09 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 04:47:09 -0000
Subject: [llvm-commits] [llvm] r84830 - in /llvm/trunk/test/FrontendC++:
 2009-08-03-Varargs.cpp varargs.cpp
Message-ID: <200910220447.n9M4l9Hw027146@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 23:47:09 2009
New Revision: 84830

URL: http://llvm.org/viewvc/llvm-project?rev=84830&view=rev
Log:
testcase for PR4678 & rdar://7309675


Added:
    llvm/trunk/test/FrontendC++/varargs.cpp
Removed:
    llvm/trunk/test/FrontendC++/2009-08-03-Varargs.cpp

Removed: llvm/trunk/test/FrontendC++/2009-08-03-Varargs.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC%2B%2B/2009-08-03-Varargs.cpp?rev=84829&view=auto

==============================================================================
--- llvm/trunk/test/FrontendC++/2009-08-03-Varargs.cpp (original)
+++ llvm/trunk/test/FrontendC++/2009-08-03-Varargs.cpp (removed)
@@ -1,5 +0,0 @@
-// RUN: %llvmgxx %s -S -emit-llvm -o - | grep _Z1az\(\.\.\.\)
-// XFAIL: *
-// PR4678
-void a(...) {
-}

Added: llvm/trunk/test/FrontendC++/varargs.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC%2B%2B/varargs.cpp?rev=84830&view=auto

==============================================================================
--- llvm/trunk/test/FrontendC++/varargs.cpp (added)
+++ llvm/trunk/test/FrontendC++/varargs.cpp Wed Oct 21 23:47:09 2009
@@ -0,0 +1,19 @@
+// RUN: %llvmgxx -S -emit-llvm %s -o - | FileCheck %s
+// rdar://7309675
+// PR4678
+
+// test1 should be compmiled to be a varargs function in the IR even 
+// though there is no way to do a va_begin.  Otherwise, the optimizer
+// will warn about 'dropped arguments' at the call site.
+
+// CHECK: define i32 @_Z5test1z(...)
+int test1(...) {
+  return -1;
+}
+
+// CHECK: call i32 (...)* @_Z5test1z(i32 0)
+void test() {
+  test1(0);
+}
+
+




From sabre at nondot.org  Wed Oct 21 23:47:07 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 04:47:07 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84829 -
	/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
Message-ID: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>

Author: lattner
Date: Wed Oct 21 23:47:07 2009
New Revision: 84829

URL: http://llvm.org/viewvc/llvm-project?rev=84829&view=rev
Log:
add a somewhat despicable (but precendented) hack to fix PR4678 & rdar://7309675.

Basically, we want to compile C/ObjC functions like:

void foo() {

into nullary LLVM IR functions.  This is important so we don't get varargs ABI
stuff, and also important because we want to get "arguments are dropped" warnings
from instcombine if you have calls like "foo(42)".

OTOH, we don't want to do this for C++ functions like:

void foo(...) {

While this foo can't do a va_begin, we do want to compile it to a varargs LLVM IR
function.  Not doing so will cause 'arguments are dropped' warning for "foo(42)",
which (unlike the C case) makes perfect sense.

This is a hack because it is predicated on the language kind.  This is because GCC's
tree structure apparently represents these things exactly the same way, so there is
no way to distinguish them.  However, this is a *precedented* hack, because GCC 
already does this in mudflap, fold-const.c etc.

If someone has a better way to get this information, I'm all ears.  Until then, I'm 
just happy to have this old issue fixed. Testcase here: test/FrontendC++/varargs.cpp.



Modified:
    llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp

Modified: llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp?rev=84829&r1=84828&r2=84829&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp (original)
+++ llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp Wed Oct 21 23:47:07 2009
@@ -404,6 +404,29 @@
   return false;
 }
 
+
+/// isCompilingCCode - Return true if we are compiling C or Objective-C code.
+static bool isCompilingCCode() {
+  // If we've already determined this, return it.
+  static unsigned Val = 2;
+  if (Val != 2) return (bool)Val;
+  
+  StringRef LanguageName = lang_hooks.name;
+  
+  if (LanguageName == "GNU C" || LanguageName == "GNU Objective-C")
+    return (Val = true);
+  
+  if (LanguageName == "GNU C++" ||
+      LanguageName == "GNU Ada" ||
+      LanguageName == "GNU F77" ||
+      LanguageName == "GNU Pascal" ||
+      LanguageName == "GNU Java" ||
+      LanguageName == "GNU Objective-C++")
+    return (Val = false);
+
+  return (Val = true);
+}
+
 void TreeToLLVM::StartFunctionBody() {
   const char *Name = "";
   // Get the name of the function.
@@ -421,7 +444,10 @@
   // allows C functions declared as "T foo() {}" to be treated like
   // "T foo(void) {}" and allows us to handle functions with K&R-style
   // definitions correctly.
-  if (TYPE_ARG_TYPES(TREE_TYPE(FnDecl)) == 0) {
+  //
+  // Note that we only do this in C/Objective-C.  Doing this in C++ for
+  // functions explicitly declared as taking (...) is bad.
+  if (TYPE_ARG_TYPES(TREE_TYPE(FnDecl)) == 0 && isCompilingCCode()) {
     FTy = TheTypeConverter->ConvertArgListToFnType(TREE_TYPE(FnDecl),
                                                    DECL_ARGUMENTS(FnDecl),
                                                    static_chain,




From johnson.peter at gmail.com  Wed Oct 21 23:41:00 2009
From: johnson.peter at gmail.com (Peter Johnson)
Date: Wed, 21 Oct 2009 21:41:00 -0700
Subject: [llvm-commits] (half precision) [llvm] r84239 - in /llvm/trunk:
	include/llvm/ADT/APFloat.h lib/Support/APFloat.cpp
In-Reply-To: <2FB4E34F-F914-4275-BBB7-1BD226DCDD52@apple.com>
References: <200910160213.n9G2Dp0C003020@zion.cs.uiuc.edu>
	<3E890328-4BC7-47D4-B3BE-58CE46E21262@apple.com>
	<2FB4E34F-F914-4275-BBB7-1BD226DCDD52@apple.com>
Message-ID: <1233f1b60910212141p4b5b1d19q8c0ec3f41786a887@mail.gmail.com>

On Wed, Oct 21, 2009 at 5:29 PM, Chris Lattner <clattner at apple.com> wrote:
>
> On Oct 20, 2009, at 5:23 PM, Dale Johannesen wrote:
>
>> Some testcases to exercise basic functionality would be nice. ?I think
>> there is a bug here that should have shown up in testing.
>
> Peter, can you take a look at Dale's feedback?
>
> -Chris

Dale is quite correct.  I had caught the 0x1f cleanup (see my bug 5195
comment) but had missed the >>10 bug.

Agreed that unit tests would be good.  There don't seem to be any
APFloat unit tests for APInt conversions.  I'll work on writing some
for the half precision case but probably won't try to tackle the full
complement.  Should I just add them to the unittest/ADT/APFloat.cpp
file?

-Peter

>
>>
>> On Oct 15, 2009, at 7:13 PM, Chris Lattner wrote:
>>>
>>> +APInt
>>> +APFloat::convertHalfAPFloatToAPInt() const
>>> +{
>>> + ?assert(semantics == (const llvm::fltSemantics*)&IEEEhalf);
>>> + ?assert (partCount()==1);
>>> +
>>> + ?uint32_t myexponent, mysignificand;
>>> +
>>> + ?if (category==fcNormal) {
>>> + ? ?myexponent = exponent+15; //bias
>>> + ? ?mysignificand = (uint32_t)*significandParts();
>>> + ? ?if (myexponent == 1 && !(mysignificand & 0x400))
>>> + ? ? ?myexponent = 0; ? // denormal
>>> + ?} else if (category==fcZero) {
>>> + ? ?myexponent = 0;
>>> + ? ?mysignificand = 0;
>>> + ?} else if (category==fcInfinity) {
>>> + ? ?myexponent = 0xff;
>>
>> 0x1f (doesn't matter functionally, but cleaner)
>>
>>> + ? ?mysignificand = 0;
>>> + ?} else {
>>> + ? ?assert(category == fcNaN && "Unknown category!");
>>> + ? ?myexponent = 0xff;
>>
>> 0x1f
>>
>>> + ? ?mysignificand = (uint32_t)*significandParts();
>>> + ?}
>>> +
>>> + ?return APInt(16, (((sign&1) << 15) | ((myexponent&0x1f) << 10) |
>>> + ? ? ? ? ? ? ? ? ? ?(mysignificand & 0x3ff)));
>>> +}
>>> +
>>
>>
>>
>>> +void
>>> +APFloat::initFromHalfAPInt(const APInt & api)
>>> +{
>>> + ?assert(api.getBitWidth()==16);
>>> + ?uint32_t i = (uint32_t)*api.getRawData();
>>> + ?uint32_t myexponent = (i >> 15) & 0x1f;
>>
>> Bug: should be >> 10
>>
>>> + ?uint32_t mysignificand = i & 0x3ff;
>>> +
>>> + ?initialize(&APFloat::IEEEhalf);
>>> + ?assert(partCount()==1);
>>> +
>>> + ?sign = i >> 15;
>>> + ?if (myexponent==0 && mysignificand==0) {
>>> + ? ?// exponent, significand meaningless
>>> + ? ?category = fcZero;
>>> + ?} else if (myexponent==0x1f && mysignificand==0) {
>>> + ? ?// exponent, significand meaningless
>>> + ? ?category = fcInfinity;
>>> + ?} else if (myexponent==0x1f && mysignificand!=0) {
>>> + ? ?// sign, exponent, significand meaningless
>>> + ? ?category = fcNaN;
>>> + ? ?*significandParts() = mysignificand;
>>> + ?} else {
>>> + ? ?category = fcNormal;
>>> + ? ?exponent = myexponent - 15; ?//bias
>>> + ? ?*significandParts() = mysignificand;
>>> + ? ?if (myexponent==0) ? ?// denormal
>>> + ? ? ?exponent = -14;
>>> + ? ?else
>>> + ? ? ?*significandParts() |= 0x400; // integer bit
>>> + ?}
>>> +}
>>> +
>>> /// Treat api as containing the bits of a floating point number.
>>> Currently
>>> /// we infer the floating point type from the size of the APInt. ?The
>>> /// isIEEE argument distinguishes between PPC128 and IEEE128 (not
>>> meaningful
>>> @@ -3058,7 +3124,9 @@
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>



From clattner at apple.com  Wed Oct 21 23:51:39 2009
From: clattner at apple.com (Chris Lattner)
Date: Wed, 21 Oct 2009 21:51:39 -0700
Subject: [llvm-commits] (half precision) [llvm] r84239 - in /llvm/trunk:
	include/llvm/ADT/APFloat.h lib/Support/APFloat.cpp
In-Reply-To: <1233f1b60910212141p4b5b1d19q8c0ec3f41786a887@mail.gmail.com>
References: <200910160213.n9G2Dp0C003020@zion.cs.uiuc.edu>
	<3E890328-4BC7-47D4-B3BE-58CE46E21262@apple.com>
	<2FB4E34F-F914-4275-BBB7-1BD226DCDD52@apple.com>
	<1233f1b60910212141p4b5b1d19q8c0ec3f41786a887@mail.gmail.com>
Message-ID: <CB63A93F-75F3-41D1-814D-0BD299069CF0@apple.com>


On Oct 21, 2009, at 9:41 PM, Peter Johnson wrote:

> On Wed, Oct 21, 2009 at 5:29 PM, Chris Lattner <clattner at apple.com>  
> wrote:
>>
>> On Oct 20, 2009, at 5:23 PM, Dale Johannesen wrote:
>>
>>> Some testcases to exercise basic functionality would be nice.  I  
>>> think
>>> there is a bug here that should have shown up in testing.
>>
>> Peter, can you take a look at Dale's feedback?
>>
>> -Chris
>
> Dale is quite correct.  I had caught the 0x1f cleanup (see my bug 5195
> comment) but had missed the >>10 bug.
>
> Agreed that unit tests would be good.  There don't seem to be any
> APFloat unit tests for APInt conversions.  I'll work on writing some
> for the half precision case but probably won't try to tackle the full
> complement.  Should I just add them to the unittest/ADT/APFloat.cpp
> file?

Sounds great to me, thanks Peter!

-Chris

>
> -Peter
>
>>
>>>
>>> On Oct 15, 2009, at 7:13 PM, Chris Lattner wrote:
>>>>
>>>> +APInt
>>>> +APFloat::convertHalfAPFloatToAPInt() const
>>>> +{
>>>> +  assert(semantics == (const llvm::fltSemantics*)&IEEEhalf);
>>>> +  assert (partCount()==1);
>>>> +
>>>> +  uint32_t myexponent, mysignificand;
>>>> +
>>>> +  if (category==fcNormal) {
>>>> +    myexponent = exponent+15; //bias
>>>> +    mysignificand = (uint32_t)*significandParts();
>>>> +    if (myexponent == 1 && !(mysignificand & 0x400))
>>>> +      myexponent = 0;   // denormal
>>>> +  } else if (category==fcZero) {
>>>> +    myexponent = 0;
>>>> +    mysignificand = 0;
>>>> +  } else if (category==fcInfinity) {
>>>> +    myexponent = 0xff;
>>>
>>> 0x1f (doesn't matter functionally, but cleaner)
>>>
>>>> +    mysignificand = 0;
>>>> +  } else {
>>>> +    assert(category == fcNaN && "Unknown category!");
>>>> +    myexponent = 0xff;
>>>
>>> 0x1f
>>>
>>>> +    mysignificand = (uint32_t)*significandParts();
>>>> +  }
>>>> +
>>>> +  return APInt(16, (((sign&1) << 15) | ((myexponent&0x1f) << 10) |
>>>> +                    (mysignificand & 0x3ff)));
>>>> +}
>>>> +
>>>
>>>
>>>
>>>> +void
>>>> +APFloat::initFromHalfAPInt(const APInt & api)
>>>> +{
>>>> +  assert(api.getBitWidth()==16);
>>>> +  uint32_t i = (uint32_t)*api.getRawData();
>>>> +  uint32_t myexponent = (i >> 15) & 0x1f;
>>>
>>> Bug: should be >> 10
>>>
>>>> +  uint32_t mysignificand = i & 0x3ff;
>>>> +
>>>> +  initialize(&APFloat::IEEEhalf);
>>>> +  assert(partCount()==1);
>>>> +
>>>> +  sign = i >> 15;
>>>> +  if (myexponent==0 && mysignificand==0) {
>>>> +    // exponent, significand meaningless
>>>> +    category = fcZero;
>>>> +  } else if (myexponent==0x1f && mysignificand==0) {
>>>> +    // exponent, significand meaningless
>>>> +    category = fcInfinity;
>>>> +  } else if (myexponent==0x1f && mysignificand!=0) {
>>>> +    // sign, exponent, significand meaningless
>>>> +    category = fcNaN;
>>>> +    *significandParts() = mysignificand;
>>>> +  } else {
>>>> +    category = fcNormal;
>>>> +    exponent = myexponent - 15;  //bias
>>>> +    *significandParts() = mysignificand;
>>>> +    if (myexponent==0)    // denormal
>>>> +      exponent = -14;
>>>> +    else
>>>> +      *significandParts() |= 0x400; // integer bit
>>>> +  }
>>>> +}
>>>> +
>>>> /// Treat api as containing the bits of a floating point number.
>>>> Currently
>>>> /// we infer the floating point type from the size of the APInt.   
>>>> The
>>>> /// isIEEE argument distinguishes between PPC128 and IEEE128 (not
>>>> meaningful
>>>> @@ -3058,7 +3124,9 @@
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>>



From evan.cheng at apple.com  Thu Oct 22 00:08:50 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 05:08:50 -0000
Subject: [llvm-commits] [llvm] r84831 -
	/llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp
Message-ID: <200910220508.n9M58o0I027859@zion.cs.uiuc.edu>

Author: evancheng
Date: Thu Oct 22 00:08:49 2009
New Revision: 84831

URL: http://llvm.org/viewvc/llvm-project?rev=84831&view=rev
Log:
Trim include.

Modified:
    llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp

Modified: llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp?rev=84831&r1=84830&r2=84831&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp Thu Oct 22 00:08:49 2009
@@ -16,7 +16,6 @@
 #include "ARM.h"
 #include "llvm/PassManager.h"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/FormattedStream.h"
 #include "llvm/Target/TargetOptions.h"
 #include "llvm/Target/TargetRegistry.h"




From evan.cheng at apple.com  Thu Oct 22 00:11:01 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 05:11:01 -0000
Subject: [llvm-commits] [llvm] r84832 - in /llvm/trunk/lib/Target/ARM:
 ARMInstrInfo.cpp Thumb1RegisterInfo.cpp Thumb2RegisterInfo.cpp
Message-ID: <200910220511.n9M5B17q027976@zion.cs.uiuc.edu>

Author: evancheng
Date: Thu Oct 22 00:11:00 2009
New Revision: 84832

URL: http://llvm.org/viewvc/llvm-project?rev=84832&view=rev
Log:
Trim more includes.

Modified:
    llvm/trunk/lib/Target/ARM/ARMInstrInfo.cpp
    llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
    llvm/trunk/lib/Target/ARM/Thumb2RegisterInfo.cpp

Modified: llvm/trunk/lib/Target/ARM/ARMInstrInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrInfo.cpp?rev=84832&r1=84831&r2=84832&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrInfo.cpp Thu Oct 22 00:11:00 2009
@@ -22,7 +22,6 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
 #include "llvm/MC/MCAsmInfo.h"
-#include "llvm/Support/CommandLine.h"
 using namespace llvm;
 
 ARMInstrInfo::ARMInstrInfo(const ARMSubtarget &STI)

Modified: llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp?rev=84832&r1=84831&r2=84832&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/Thumb1RegisterInfo.cpp Thu Oct 22 00:11:00 2009
@@ -32,7 +32,6 @@
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/SmallVector.h"
-#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
 using namespace llvm;

Modified: llvm/trunk/lib/Target/ARM/Thumb2RegisterInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/Thumb2RegisterInfo.cpp?rev=84832&r1=84831&r2=84832&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/Thumb2RegisterInfo.cpp (original)
+++ llvm/trunk/lib/Target/ARM/Thumb2RegisterInfo.cpp Thu Oct 22 00:11:00 2009
@@ -32,7 +32,6 @@
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/SmallVector.h"
-#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/ErrorHandling.h"
 using namespace llvm;
 




From clattner at apple.com  Thu Oct 22 00:22:09 2009
From: clattner at apple.com (Chris Lattner)
Date: Wed, 21 Oct 2009 22:22:09 -0700
Subject: [llvm-commits] [llvm] r84820 - in /llvm/trunk:
	include/llvm/Metadata.h lib/Bitcode/Writer/BitcodeWriter.cpp
	lib/VMCore/AsmWriter.cpp lib/VMCore/Metadata.cpp
In-Reply-To: <200910220101.n9M11PUF018859@zion.cs.uiuc.edu>
References: <200910220101.n9M11PUF018859@zion.cs.uiuc.edu>
Message-ID: <EE7833CB-FCEB-426C-8962-24ABCBD7B571@apple.com>


On Oct 21, 2009, at 6:01 PM, Devang Patel wrote:

> Author: dpatel
> Date: Wed Oct 21 20:01:24 2009
> New Revision: 84820
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84820&view=rev
> Log:
> Fix getHandlerNames() interface. Now it populate clinet supplied  
> small vector with handler names.

Thanks Devang,

> +++ llvm/trunk/include/llvm/Metadata.h Wed Oct 21 20:01:24 2009
> @@ -287,7 +287,7 @@
>
>   /// getHandlerNames - Get handler names. This is used by bitcode
>   /// writer.
> -  const StringMap<unsigned> *getHandlerNames();
> +  void getHandlerNames(SmallVectorImpl<std::pair<unsigned,  
> StringRef> >&N) const;

Please improve this comment: the bitcode writer shouldn't be mentioned.

> +/// writer and aswm writer.
> +void MetadataContext::
> +getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef>  
> >&Names) const {
> +  for (StringMap<unsigned>::const_iterator I = MDHandlerNames.begin 
> (),
> +         E = MDHandlerNames.end(); I != E; ++I)
> +    Names.push_back(std::make_pair(I->second, I->first()));

This implementation is returning the names in a nondeterminstic  
order.  Please return them according to their index.  Since the  
indices are dense, you can do something like:

   Names.resize(MDHandlerNames.size());
   for (StringMap<unsigned>::const_iterator I = MDHandlerNames.begin(),
          E = MDHandlerNames.end(); I != E; ++I)
     Names[I->second] = I->first;

Being careful for fencepost errors.

Thanks!

-Chris


From sabre at nondot.org  Thu Oct 22 01:25:13 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 06:25:13 -0000
Subject: [llvm-commits] [llvm] r84836 - in /llvm/trunk:
 include/llvm/Analysis/ConstantFolding.h lib/Analysis/ConstantFolding.cpp
 lib/Transforms/Scalar/InstructionCombining.cpp
Message-ID: <200910220625.n9M6PDsH030808@zion.cs.uiuc.edu>

Author: lattner
Date: Thu Oct 22 01:25:11 2009
New Revision: 84836

URL: http://llvm.org/viewvc/llvm-project?rev=84836&view=rev
Log:
Move some constant folding logic for loads out of instcombine into
Analysis/ConstantFolding.cpp.  This doesn't change the behavior of
instcombine but makes other clients of ConstantFoldInstruction
able to handle loads.  This was partially extracted from Eli's patch
in PR3152.


Modified:
    llvm/trunk/include/llvm/Analysis/ConstantFolding.h
    llvm/trunk/lib/Analysis/ConstantFolding.cpp
    llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp

Modified: llvm/trunk/include/llvm/Analysis/ConstantFolding.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Analysis/ConstantFolding.h?rev=84836&r1=84835&r2=84836&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Analysis/ConstantFolding.h (original)
+++ llvm/trunk/include/llvm/Analysis/ConstantFolding.h Thu Oct 22 01:25:11 2009
@@ -62,6 +62,10 @@
                                           LLVMContext &Context,
                                           const TargetData *TD = 0);
 
+/// ConstantFoldLoadFromConstPtr - Return the value that a load from C would
+/// produce if it is constant and determinable.  If this is not determinable,
+/// return null.
+Constant *ConstantFoldLoadFromConstPtr(Constant *C, const TargetData *TD = 0);
 
 /// ConstantFoldLoadThroughGEPConstantExpr - Given a constant and a
 /// getelementptr constantexpr, return the constant value being addressed by the

Modified: llvm/trunk/lib/Analysis/ConstantFolding.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ConstantFolding.cpp?rev=84836&r1=84835&r2=84836&view=diff

==============================================================================
--- llvm/trunk/lib/Analysis/ConstantFolding.cpp (original)
+++ llvm/trunk/lib/Analysis/ConstantFolding.cpp Thu Oct 22 01:25:11 2009
@@ -30,6 +30,7 @@
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Support/MathExtras.h"
+#include "llvm/GlobalVariable.h"
 #include <cerrno>
 #include <cmath>
 using namespace llvm;
@@ -92,6 +93,37 @@
   return false;
 }
 
+/// ConstantFoldLoadFromConstPtr - Return the value that a load from C would
+/// produce if it is constant and determinable.  If this is not determinable,
+/// return null.
+Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C,
+                                             const TargetData *TD) {
+  // First, try the easy cases:
+  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(C))
+    if (GV->isConstant() && GV->hasDefinitiveInitializer())
+      return GV->getInitializer();
+
+  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
+    if (CE->getOpcode() == Instruction::GetElementPtr) {
+      if (GlobalVariable *GV = dyn_cast<GlobalVariable>(CE->getOperand(0)))
+        if (GV->isConstant() && GV->hasDefinitiveInitializer())
+          if (Constant *V = 
+               ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE))
+            return V;
+    }
+  }
+
+  return 0;
+}
+
+static Constant *ConstantFoldLoadInst(const LoadInst *LI, const TargetData *TD){
+  if (LI->isVolatile()) return 0;
+  
+  if (Constant *C = dyn_cast<Constant>(LI->getOperand(0)))
+    return ConstantFoldLoadFromConstPtr(C, TD);
+    
+  return 0;
+}
 
 /// SymbolicallyEvaluateBinop - One of Op0/Op1 is a constant expression.
 /// Attempt to symbolically evaluate the result of a binary operator merging
@@ -380,6 +412,9 @@
                                            Ops.data(), Ops.size(), 
                                            Context, TD);
   
+  if (const LoadInst *LI = dyn_cast<LoadInst>(I))
+    return ConstantFoldLoadInst(LI, TD);
+  
   return ConstantFoldInstOperands(I->getOpcode(), I->getType(),
                                   Ops.data(), Ops.size(), Context, TD);
 }

Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=84836&r1=84835&r2=84836&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Thu Oct 22 01:25:11 2009
@@ -11454,6 +11454,7 @@
   if (Value *AvailableVal = FindAvailableLoadedValue(Op, LI.getParent(), BBI,6))
     return ReplaceInstUsesWith(LI, AvailableVal);
 
+  // load(gep null, ...) -> unreachable
   if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(Op)) {
     const Value *GEPI0 = GEPI->getOperand(0);
     // TODO: Consider a target hook for valid address spaces for this xform.
@@ -11468,49 +11469,24 @@
     }
   } 
 
-  if (Constant *C = dyn_cast<Constant>(Op)) {
-    // load null/undef -> undef
-    // TODO: Consider a target hook for valid address spaces for this xform.
-    if (isa<UndefValue>(C) ||
-        (C->isNullValue() && LI.getPointerAddressSpace() == 0)) {
-      // Insert a new store to null instruction before the load to indicate that
-      // this code is not reachable.  We do this instead of inserting an
-      // unreachable instruction directly because we cannot modify the CFG.
-      new StoreInst(UndefValue::get(LI.getType()),
-                    Constant::getNullValue(Op->getType()), &LI);
-      return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
-    }
-
-    // Instcombine load (constant global) into the value loaded.
-    if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Op))
-      if (GV->isConstant() && GV->hasDefinitiveInitializer())
-        return ReplaceInstUsesWith(LI, GV->getInitializer());
-
-    // Instcombine load (constantexpr_GEP global, 0, ...) into the value loaded.
-    if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Op)) {
-      if (CE->getOpcode() == Instruction::GetElementPtr) {
-        if (GlobalVariable *GV = dyn_cast<GlobalVariable>(CE->getOperand(0)))
-          if (GV->isConstant() && GV->hasDefinitiveInitializer())
-            if (Constant *V = 
-               ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE))
-              return ReplaceInstUsesWith(LI, V);
-        if (CE->getOperand(0)->isNullValue()) {
-          // Insert a new store to null instruction before the load to indicate
-          // that this code is not reachable.  We do this instead of inserting
-          // an unreachable instruction directly because we cannot modify the
-          // CFG.
-          new StoreInst(UndefValue::get(LI.getType()),
-                        Constant::getNullValue(Op->getType()), &LI);
-          return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
-        }
-
-      } else if (CE->isCast()) {
-        if (Instruction *Res = InstCombineLoadCast(*this, LI, TD))
-          return Res;
-      }
-    }
+  // load null/undef -> unreachable
+  // TODO: Consider a target hook for valid address spaces for this xform.
+  if (isa<UndefValue>(Op) ||
+      (isa<ConstantPointerNull>(Op) && LI.getPointerAddressSpace() == 0)) {
+    // Insert a new store to null instruction before the load to indicate that
+    // this code is not reachable.  We do this instead of inserting an
+    // unreachable instruction directly because we cannot modify the CFG.
+    new StoreInst(UndefValue::get(LI.getType()),
+                  Constant::getNullValue(Op->getType()), &LI);
+    return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
   }
-    
+
+  // Instcombine load (constantexpr_cast global) -> cast (load global)
+  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Op))
+    if (CE->isCast())
+      if (Instruction *Res = InstCombineLoadCast(*this, LI, TD))
+        return Res;
+  
   // If this load comes from anywhere in a constant global, and if the global
   // is all undef or zero, we know what it loads.
   if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Op->getUnderlyingObject())){




From evan.cheng at apple.com  Thu Oct 22 01:33:10 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 06:33:10 -0000
Subject: [llvm-commits] [test-suite] r84837 -
	/test-suite/trunk/Makefile.programs
Message-ID: <200910220633.n9M6XAhj031169@zion.cs.uiuc.edu>

Author: evancheng
Date: Thu Oct 22 01:33:10 2009
New Revision: 84837

URL: http://llvm.org/viewvc/llvm-project?rev=84837&view=rev
Log:
Thumb llcbeta is -combiner-alias-analysis.

Modified:
    test-suite/trunk/Makefile.programs

Modified: test-suite/trunk/Makefile.programs
URL: http://llvm.org/viewvc/llvm-project/test-suite/trunk/Makefile.programs?rev=84837&r1=84836&r2=84837&view=diff

==============================================================================
--- test-suite/trunk/Makefile.programs (original)
+++ test-suite/trunk/Makefile.programs Thu Oct 22 01:33:10 2009
@@ -248,8 +248,7 @@
 #-schedule-livein-copies
 endif
 ifeq ($(ARCH),THUMB)
-LLCBETAOPTION :=
-#-combiner-alias-analysis
+LLCBETAOPTION := -combiner-alias-analysis
 #--enable-thumb-reg-scavenging
 endif
 




From evan.cheng at apple.com  Thu Oct 22 01:34:53 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 06:34:53 -0000
Subject: [llvm-commits] [test-suite] r84838 -
	/test-suite/trunk/Makefile.programs
Message-ID: <200910220634.n9M6YrIE031261@zion.cs.uiuc.edu>

Author: evancheng
Date: Thu Oct 22 01:34:52 2009
New Revision: 84838

URL: http://llvm.org/viewvc/llvm-project?rev=84838&view=rev
Log:
x86 llcbeta is -hoist-constants.

Modified:
    test-suite/trunk/Makefile.programs

Modified: test-suite/trunk/Makefile.programs
URL: http://llvm.org/viewvc/llvm-project/test-suite/trunk/Makefile.programs?rev=84838&r1=84837&r2=84838&view=diff

==============================================================================
--- test-suite/trunk/Makefile.programs (original)
+++ test-suite/trunk/Makefile.programs Thu Oct 22 01:34:52 2009
@@ -223,12 +223,14 @@
 LLCBETAOPTION := -sched=simple 
 endif
 ifeq ($(ARCH),x86_64)
-LLCBETAOPTION := -combiner-alias-analysis
+LLCBETAOPTION := -host-constants
+#-combiner-alias-analysis
 #-pre-alloc-split
 #-remat-pic-stub-load
 endif
 ifeq ($(ARCH),x86)
-LLCBETAOPTION := -combiner-alias-analysis
+LLCBETAOPTION := -host-constants
+#-combiner-alias-analysis
 #-pre-alloc-split
 #-remat-pic-stub-load
 #-combiner-global-alias-analysis




From evan.cheng at apple.com  Thu Oct 22 01:35:28 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 06:35:28 -0000
Subject: [llvm-commits] [test-suite] r84839 -
	/test-suite/trunk/Makefile.programs
Message-ID: <200910220635.n9M6ZSuY031299@zion.cs.uiuc.edu>

Author: evancheng
Date: Thu Oct 22 01:35:28 2009
New Revision: 84839

URL: http://llvm.org/viewvc/llvm-project?rev=84839&view=rev
Log:
-hoist-constants, not -host-constants.

Modified:
    test-suite/trunk/Makefile.programs

Modified: test-suite/trunk/Makefile.programs
URL: http://llvm.org/viewvc/llvm-project/test-suite/trunk/Makefile.programs?rev=84839&r1=84838&r2=84839&view=diff

==============================================================================
--- test-suite/trunk/Makefile.programs (original)
+++ test-suite/trunk/Makefile.programs Thu Oct 22 01:35:28 2009
@@ -223,13 +223,13 @@
 LLCBETAOPTION := -sched=simple 
 endif
 ifeq ($(ARCH),x86_64)
-LLCBETAOPTION := -host-constants
+LLCBETAOPTION := -hoist-constants
 #-combiner-alias-analysis
 #-pre-alloc-split
 #-remat-pic-stub-load
 endif
 ifeq ($(ARCH),x86)
-LLCBETAOPTION := -host-constants
+LLCBETAOPTION := -hoist-constants
 #-combiner-alias-analysis
 #-pre-alloc-split
 #-remat-pic-stub-load




From sabre at nondot.org  Thu Oct 22 01:38:35 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 06:38:35 -0000
Subject: [llvm-commits] [llvm] r84840 - in /llvm/trunk/lib:
 Analysis/ConstantFolding.cpp Transforms/Scalar/InstructionCombining.cpp
Message-ID: <200910220638.n9M6cZm6031443@zion.cs.uiuc.edu>

Author: lattner
Date: Thu Oct 22 01:38:35 2009
New Revision: 84840

URL: http://llvm.org/viewvc/llvm-project?rev=84840&view=rev
Log:
move 'loading i32 from string' optimization from instcombine
to libanalysis.  Instcombine shrinking... does this even 
make sense???

Modified:
    llvm/trunk/lib/Analysis/ConstantFolding.cpp
    llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp

Modified: llvm/trunk/lib/Analysis/ConstantFolding.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ConstantFolding.cpp?rev=84840&r1=84839&r2=84840&view=diff

==============================================================================
--- llvm/trunk/lib/Analysis/ConstantFolding.cpp (original)
+++ llvm/trunk/lib/Analysis/ConstantFolding.cpp Thu Oct 22 01:38:35 2009
@@ -24,13 +24,13 @@
 #include "llvm/Instructions.h"
 #include "llvm/Intrinsics.h"
 #include "llvm/LLVMContext.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Target/TargetData.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringMap.h"
-#include "llvm/Target/TargetData.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Support/MathExtras.h"
-#include "llvm/GlobalVariable.h"
 #include <cerrno>
 #include <cmath>
 using namespace llvm;
@@ -111,6 +111,35 @@
                ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE))
             return V;
     }
+    
+    // Instead of loading constant c string, use corresponding integer value
+    // directly if string length is small enough.
+    std::string Str;
+    if (TD && GetConstantStringInfo(CE->getOperand(0), Str) && !Str.empty()) {
+      unsigned len = Str.length();
+      const Type *Ty = cast<PointerType>(CE->getType())->getElementType();
+      unsigned numBits = Ty->getPrimitiveSizeInBits();
+      // Replace LI with immediate integer store.
+      if ((numBits >> 3) == len + 1) {
+        APInt StrVal(numBits, 0);
+        APInt SingleChar(numBits, 0);
+        if (TD->isLittleEndian()) {
+          for (signed i = len-1; i >= 0; i--) {
+            SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
+            StrVal = (StrVal << 8) | SingleChar;
+          }
+        } else {
+          for (unsigned i = 0; i < len; i++) {
+            SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
+            StrVal = (StrVal << 8) | SingleChar;
+          }
+          // Append NULL at the end.
+          SingleChar = 0;
+          StrVal = (StrVal << 8) | SingleChar;
+        }
+        return ConstantInt::get(CE->getContext(), StrVal);
+      }
+    }
   }
 
   return 0;
@@ -675,15 +704,15 @@
           C = UndefValue::get(ATy->getElementType());
         else
           return 0;
-      } else if (const VectorType *PTy = dyn_cast<VectorType>(*I)) {
-        if (CI->getZExtValue() >= PTy->getNumElements())
+      } else if (const VectorType *VTy = dyn_cast<VectorType>(*I)) {
+        if (CI->getZExtValue() >= VTy->getNumElements())
           return 0;
         if (ConstantVector *CP = dyn_cast<ConstantVector>(C))
           C = CP->getOperand(CI->getZExtValue());
         else if (isa<ConstantAggregateZero>(C))
-          C = Constant::getNullValue(PTy->getElementType());
+          C = Constant::getNullValue(VTy->getElementType());
         else if (isa<UndefValue>(C))
-          C = UndefValue::get(PTy->getElementType());
+          C = UndefValue::get(VTy->getElementType());
         else
           return 0;
       } else {

Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=84840&r1=84839&r2=84840&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Thu Oct 22 01:38:35 2009
@@ -11345,40 +11345,6 @@
   Value *CastOp = CI->getOperand(0);
   LLVMContext *Context = IC.getContext();
 
-  if (TD) {
-    if (ConstantExpr *CE = dyn_cast<ConstantExpr>(CI)) {
-      // Instead of loading constant c string, use corresponding integer value
-      // directly if string length is small enough.
-      std::string Str;
-      if (GetConstantStringInfo(CE->getOperand(0), Str) && !Str.empty()) {
-        unsigned len = Str.length();
-        const Type *Ty = cast<PointerType>(CE->getType())->getElementType();
-        unsigned numBits = Ty->getPrimitiveSizeInBits();
-        // Replace LI with immediate integer store.
-        if ((numBits >> 3) == len + 1) {
-          APInt StrVal(numBits, 0);
-          APInt SingleChar(numBits, 0);
-          if (TD->isLittleEndian()) {
-            for (signed i = len-1; i >= 0; i--) {
-              SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
-              StrVal = (StrVal << 8) | SingleChar;
-            }
-          } else {
-            for (unsigned i = 0; i < len; i++) {
-              SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
-              StrVal = (StrVal << 8) | SingleChar;
-            }
-            // Append NULL at the end.
-            SingleChar = 0;
-            StrVal = (StrVal << 8) | SingleChar;
-          }
-          Value *NL = ConstantInt::get(*Context, StrVal);
-          return IC.ReplaceInstUsesWith(LI, NL);
-        }
-      }
-    }
-  }
-
   const PointerType *DestTy = cast<PointerType>(CI->getType());
   const Type *DestPTy = DestTy->getElementType();
   if (const PointerType *SrcTy = dyn_cast<PointerType>(CastOp->getType())) {




From sabre at nondot.org  Thu Oct 22 01:44:08 2009
From: sabre at nondot.org (Chris Lattner)
Date: Thu, 22 Oct 2009 06:44:08 -0000
Subject: [llvm-commits] [llvm] r84841 - in /llvm/trunk/lib:
 Analysis/ConstantFolding.cpp Transforms/Scalar/InstructionCombining.cpp
Message-ID: <200910220644.n9M6i8G4031666@zion.cs.uiuc.edu>

Author: lattner
Date: Thu Oct 22 01:44:07 2009
New Revision: 84841

URL: http://llvm.org/viewvc/llvm-project?rev=84841&view=rev
Log:
move another load optimization from instcombine -> libanalysis.
 

Modified:
    llvm/trunk/lib/Analysis/ConstantFolding.cpp
    llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp

Modified: llvm/trunk/lib/Analysis/ConstantFolding.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ConstantFolding.cpp?rev=84841&r1=84840&r2=84841&view=diff

==============================================================================
--- llvm/trunk/lib/Analysis/ConstantFolding.cpp (original)
+++ llvm/trunk/lib/Analysis/ConstantFolding.cpp Thu Oct 22 01:44:07 2009
@@ -103,45 +103,59 @@
     if (GV->isConstant() && GV->hasDefinitiveInitializer())
       return GV->getInitializer();
 
-  if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
-    if (CE->getOpcode() == Instruction::GetElementPtr) {
-      if (GlobalVariable *GV = dyn_cast<GlobalVariable>(CE->getOperand(0)))
-        if (GV->isConstant() && GV->hasDefinitiveInitializer())
-          if (Constant *V = 
-               ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE))
-            return V;
-    }
-    
-    // Instead of loading constant c string, use corresponding integer value
-    // directly if string length is small enough.
-    std::string Str;
-    if (TD && GetConstantStringInfo(CE->getOperand(0), Str) && !Str.empty()) {
-      unsigned len = Str.length();
-      const Type *Ty = cast<PointerType>(CE->getType())->getElementType();
-      unsigned numBits = Ty->getPrimitiveSizeInBits();
-      // Replace LI with immediate integer store.
-      if ((numBits >> 3) == len + 1) {
-        APInt StrVal(numBits, 0);
-        APInt SingleChar(numBits, 0);
-        if (TD->isLittleEndian()) {
-          for (signed i = len-1; i >= 0; i--) {
-            SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
-            StrVal = (StrVal << 8) | SingleChar;
-          }
-        } else {
-          for (unsigned i = 0; i < len; i++) {
-            SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
-            StrVal = (StrVal << 8) | SingleChar;
-          }
-          // Append NULL at the end.
-          SingleChar = 0;
+  // If the loaded value isn't a constant expr, we can't handle it.
+  ConstantExpr *CE = dyn_cast<ConstantExpr>(C);
+  if (!CE) return 0;
+  
+  if (CE->getOpcode() == Instruction::GetElementPtr) {
+    if (GlobalVariable *GV = dyn_cast<GlobalVariable>(CE->getOperand(0)))
+      if (GV->isConstant() && GV->hasDefinitiveInitializer())
+        if (Constant *V = 
+             ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE))
+          return V;
+  }
+  
+  // Instead of loading constant c string, use corresponding integer value
+  // directly if string length is small enough.
+  std::string Str;
+  if (TD && GetConstantStringInfo(CE->getOperand(0), Str) && !Str.empty()) {
+    unsigned len = Str.length();
+    const Type *Ty = cast<PointerType>(CE->getType())->getElementType();
+    unsigned numBits = Ty->getPrimitiveSizeInBits();
+    // Replace LI with immediate integer store.
+    if ((numBits >> 3) == len + 1) {
+      APInt StrVal(numBits, 0);
+      APInt SingleChar(numBits, 0);
+      if (TD->isLittleEndian()) {
+        for (signed i = len-1; i >= 0; i--) {
+          SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
           StrVal = (StrVal << 8) | SingleChar;
         }
-        return ConstantInt::get(CE->getContext(), StrVal);
-      }
+      } else {
+        for (unsigned i = 0; i < len; i++) {
+          SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
+          StrVal = (StrVal << 8) | SingleChar;
+        }
+        // Append NULL at the end.
+        SingleChar = 0;
+        StrVal = (StrVal << 8) | SingleChar;
+      }
+      return ConstantInt::get(CE->getContext(), StrVal);
     }
   }
-
+  
+  // If this load comes from anywhere in a constant global, and if the global
+  // is all undef or zero, we know what it loads.
+  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(CE->getUnderlyingObject())){
+    if (GV->isConstant() && GV->hasDefinitiveInitializer()) {
+      const Type *ResTy = cast<PointerType>(C->getType())->getElementType();
+      if (GV->getInitializer()->isNullValue())
+        return Constant::getNullValue(ResTy);
+      if (isa<UndefValue>(GV->getInitializer()))
+        return UndefValue::get(ResTy);
+    }
+  }
+  
   return 0;
 }
 

Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=84841&r1=84840&r2=84841&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Thu Oct 22 01:44:07 2009
@@ -11364,7 +11364,8 @@
         if (Constant *CSrc = dyn_cast<Constant>(CastOp))
           if (ASrcTy->getNumElements() != 0) {
             Value *Idxs[2];
-            Idxs[0] = Idxs[1] = Constant::getNullValue(Type::getInt32Ty(*Context));
+            Idxs[0] = Constant::getNullValue(Type::getInt32Ty(*Context));
+            Idxs[1] = Idxs[0];
             CastOp = ConstantExpr::getGetElementPtr(CSrc, Idxs, 2);
             SrcTy = cast<PointerType>(CastOp->getType());
             SrcPTy = SrcTy->getElementType();
@@ -11453,17 +11454,6 @@
       if (Instruction *Res = InstCombineLoadCast(*this, LI, TD))
         return Res;
   
-  // If this load comes from anywhere in a constant global, and if the global
-  // is all undef or zero, we know what it loads.
-  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Op->getUnderlyingObject())){
-    if (GV->isConstant() && GV->hasDefinitiveInitializer()) {
-      if (GV->getInitializer()->isNullValue())
-        return ReplaceInstUsesWith(LI, Constant::getNullValue(LI.getType()));
-      else if (isa<UndefValue>(GV->getInitializer()))
-        return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
-    }
-  }
-
   if (Op->hasOneUse()) {
     // Change select and PHI nodes to select values instead of addresses: this
     // helps alias analysis out a lot, allows many others simplifications, and




From evan.cheng at apple.com  Thu Oct 22 01:47:35 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 06:47:35 -0000
Subject: [llvm-commits] [llvm] r84842 -
	/llvm/trunk/lib/Target/ARM/ARMLoadStoreOptimizer.cpp
Message-ID: <200910220647.n9M6lZHO031873@zion.cs.uiuc.edu>

Author: evancheng
Date: Thu Oct 22 01:47:35 2009
New Revision: 84842

URL: http://llvm.org/viewvc/llvm-project?rev=84842&view=rev
Log:
Load / store multiple was missing opportunites when the load / store bundles are at the end of the bb. Test case is already in, the bug is exposed by subsequent commit.

Modified:
    llvm/trunk/lib/Target/ARM/ARMLoadStoreOptimizer.cpp

Modified: llvm/trunk/lib/Target/ARM/ARMLoadStoreOptimizer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMLoadStoreOptimizer.cpp?rev=84842&r1=84841&r2=84842&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMLoadStoreOptimizer.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMLoadStoreOptimizer.cpp Thu Oct 22 01:47:35 2009
@@ -974,6 +974,9 @@
     if (Advance) {
       ++Position;
       ++MBBI;
+      if (MBBI == E)
+        // Reach the end of the block, try merging the memory instructions.
+        TryMerge = true;
     } else
       TryMerge = true;
 




From evan.cheng at apple.com  Thu Oct 22 01:48:32 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 06:48:32 -0000
Subject: [llvm-commits] [llvm] r84843 - in /llvm/trunk:
 lib/Target/ARM/ARMTargetMachine.cpp test/CodeGen/ARM/ifcvt5.ll
Message-ID: <200910220648.n9M6mWIn031918@zion.cs.uiuc.edu>

Author: evancheng
Date: Thu Oct 22 01:48:32 2009
New Revision: 84843

URL: http://llvm.org/viewvc/llvm-project?rev=84843&view=rev
Log:
Move if-conversion before post-regalloc scheduling so the predicated instruction get scheduled properly.

Modified:
    llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp
    llvm/trunk/test/CodeGen/ARM/ifcvt5.ll

Modified: llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp?rev=84843&r1=84842&r2=84843&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp Thu Oct 22 01:48:32 2009
@@ -103,18 +103,16 @@
 bool ARMBaseTargetMachine::addPreSched2(PassManagerBase &PM,
                                         CodeGenOpt::Level OptLevel) {
   // FIXME: temporarily disabling load / store optimization pass for Thumb1.
-  if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only())
+  if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only()) {
     PM.add(createARMLoadStoreOptimizationPass());
+    PM.add(createIfConverterPass());
+  }
 
   return true;
 }
 
 bool ARMBaseTargetMachine::addPreEmitPass(PassManagerBase &PM,
                                           CodeGenOpt::Level OptLevel) {
-  // FIXME: temporarily disabling load / store optimization pass for Thumb1.
-  if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only())
-    PM.add(createIfConverterPass());
-
   if (Subtarget.isThumb2()) {
     PM.add(createThumb2ITBlockPass());
     PM.add(createThumb2SizeReductionPass());

Modified: llvm/trunk/test/CodeGen/ARM/ifcvt5.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/ifcvt5.ll?rev=84843&r1=84842&r2=84843&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/ARM/ifcvt5.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/ifcvt5.ll Thu Oct 22 01:48:32 2009
@@ -11,7 +11,8 @@
 
 define void @t1(i32 %a, i32 %b) {
 ; CHECK: t1:
-; CHECK: ldmltfd sp!, {r7, pc}
+; CHECK: movge
+; CHECK: blge _foo
 entry:
 	%tmp1 = icmp sgt i32 %a, 10		; <i1> [#uses=1]
 	br i1 %tmp1, label %cond_true, label %UnifiedReturnBlock




From edwintorok at gmail.com  Thu Oct 22 02:18:00 2009
From: edwintorok at gmail.com (=?ISO-8859-1?Q?T=F6r=F6k_Edwin?=)
Date: Thu, 22 Oct 2009 10:18:00 +0300
Subject: [llvm-commits] [llvm] r84799 - in /llvm/trunk:
 lib/Transforms/Scalar/InstructionCombining.cpp
 test/Transforms/InstCombine/fold-intophi.ll
In-Reply-To: <200910212341.n9LNfxIQ015498@zion.cs.uiuc.edu>
References: <200910212341.n9LNfxIQ015498@zion.cs.uiuc.edu>
Message-ID: <4AE00728.1090707@gmail.com>

On 2009-10-22 02:41, Chris Lattner wrote:
> Author: lattner
> Date: Wed Oct 21 18:41:58 2009
> New Revision: 84799
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84799&view=rev
> Log:
> revert r84754, it isn't the right approach.  Edwin, please propose 
> patches for fixes like this instead of committing them directly.
>   

Sorry about that, will do that in the future.

Best regards,
--Edwin



From benny.kra at googlemail.com  Thu Oct 22 04:28:49 2009
From: benny.kra at googlemail.com (Benjamin Kramer)
Date: Thu, 22 Oct 2009 09:28:49 -0000
Subject: [llvm-commits] [llvm] r84849 -
	/llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
Message-ID: <200910220928.n9M9SokC020091@zion.cs.uiuc.edu>

Author: d0k
Date: Thu Oct 22 04:28:49 2009
New Revision: 84849

URL: http://llvm.org/viewvc/llvm-project?rev=84849&view=rev
Log:
Shift art to the right to keep GCC from complaining about multi-line comments.

Modified:
    llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp

Modified: llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp?rev=84849&r1=84848&r2=84849&view=diff

==============================================================================
--- llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/MSP430/MSP430ISelDAGToDAG.cpp Thu Oct 22 04:28:49 2009
@@ -165,10 +165,10 @@
   if (OptLevel == CodeGenOpt::None) return false;
 
   /// RMW preprocessing creates the following code:
-  ///       [Load1]
-  ///       ^     ^
-  ///       |      \
-  ///       |       \
+  ///         [Load1]
+  ///         ^     ^
+  ///        /      |
+  ///       /       |
   ///       [Load2] |
   ///       ^    ^  |
   ///       |    |  |
@@ -328,16 +328,16 @@
 /// We also recognize the case where second operand of Op is load as well and
 /// move it below token factor as well creating DAG as follows:
 ///
-///     [Load chain]
-///         ^
-///         |
-///    [TokenFactor]
-///         ^
-///         |
-///       [Load1]
-///       ^     ^
-///       |      \
-///       |       \
+///       [Load chain]
+///            ^
+///            |
+///      [TokenFactor]
+///            ^
+///            |
+///         [Load1]
+///         ^     ^
+///        /      |
+///       /       |
 ///       [Load2] |
 ///       ^    ^  |
 ///       |    |  |




From baldrick at free.fr  Thu Oct 22 04:38:50 2009
From: baldrick at free.fr (Duncan Sands)
Date: Thu, 22 Oct 2009 11:38:50 +0200
Subject: [llvm-commits] [llvm-gcc-4.2] r84829
	-	/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
In-Reply-To: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>
References: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>
Message-ID: <4AE0282A.5080300@free.fr>

Hi Chris,

> add a somewhat despicable (but precendented) hack to fix PR4678 & rdar://7309675.

this is awful.  Was there something wrong with Rafael's proposed patch from the
PR?

Ciao,

Duncan.


From baldrick at free.fr  Thu Oct 22 05:02:10 2009
From: baldrick at free.fr (Duncan Sands)
Date: Thu, 22 Oct 2009 10:02:10 -0000
Subject: [llvm-commits] [llvm] r84851 -
	/llvm/trunk/test/FrontendAda/negative_field_offset.adb
Message-ID: <200910221002.n9MA2APe021445@zion.cs.uiuc.edu>

Author: baldrick
Date: Thu Oct 22 05:02:10 2009
New Revision: 84851

URL: http://llvm.org/viewvc/llvm-project?rev=84851&view=rev
Log:
Test handling of record fields with negative offsets.

Added:
    llvm/trunk/test/FrontendAda/negative_field_offset.adb

Added: llvm/trunk/test/FrontendAda/negative_field_offset.adb
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendAda/negative_field_offset.adb?rev=84851&view=auto

==============================================================================
--- llvm/trunk/test/FrontendAda/negative_field_offset.adb (added)
+++ llvm/trunk/test/FrontendAda/negative_field_offset.adb Thu Oct 22 05:02:10 2009
@@ -0,0 +1,10 @@
+-- RUN: %llvmgcc -c %s
+with System;
+procedure Negative_Field_Offset (N : Integer) is
+   type String_Pointer is access String;
+   --  Force use of a thin pointer.
+   for String_Pointer'Size use System.Word_Size;
+   P : String_Pointer;
+begin
+   P := new String (1 .. N);
+end;




From baldrick at free.fr  Thu Oct 22 07:53:26 2009
From: baldrick at free.fr (Duncan Sands)
Date: Thu, 22 Oct 2009 12:53:26 -0000
Subject: [llvm-commits] [llvm] r84853 -
	/llvm/trunk/test/FrontendAda/negative_field_offset.adb
Message-ID: <200910221253.n9MCrQDW027414@zion.cs.uiuc.edu>

Author: baldrick
Date: Thu Oct 22 07:53:25 2009
New Revision: 84853

URL: http://llvm.org/viewvc/llvm-project?rev=84853&view=rev
Log:
Check that accessing a struct field that occurs before the start
of the struct (!) works correctly.

Modified:
    llvm/trunk/test/FrontendAda/negative_field_offset.adb

Modified: llvm/trunk/test/FrontendAda/negative_field_offset.adb
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendAda/negative_field_offset.adb?rev=84853&r1=84852&r2=84853&view=diff

==============================================================================
--- llvm/trunk/test/FrontendAda/negative_field_offset.adb (original)
+++ llvm/trunk/test/FrontendAda/negative_field_offset.adb Thu Oct 22 07:53:25 2009
@@ -5,6 +5,12 @@
    --  Force use of a thin pointer.
    for String_Pointer'Size use System.Word_Size;
    P : String_Pointer;
+
+   procedure Q (P : String_Pointer) is
+   begin
+      P (1) := 'Z';
+   end;
 begin
    P := new String (1 .. N);
+   Q (P);
 end;




From espindola at google.com  Thu Oct 22 08:44:53 2009
From: espindola at google.com (Rafael Espindola)
Date: Thu, 22 Oct 2009 09:44:53 -0400
Subject: [llvm-commits] [llvm-gcc-4.2] r84829 -
	/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
In-Reply-To: <4AE0282A.5080300@free.fr>
References: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>
	<4AE0282A.5080300@free.fr>
Message-ID: <38a0d8450910220644o5ef90796w6e0b97c881305f31@mail.gmail.com>

> this is awful. ?Was there something wrong with Rafael's proposed patch from the
> PR?

It looks a bit like my first attempt at fixing it.

The following patches tried to change both the function definition and
the function call. The problem with them is that we would produce
different outputs at -O0 and -O1. The warning would show up in strange
combinations like -O0 + opt and probably -O0 + lto.

I haven't looked at the bug for some time, but I am OK with both solutions.

> Ciao,
>
> Duncan.

Cheers,
-- 
Rafael ?vila de Esp?ndola



From nicolas.geoffray at lip6.fr  Thu Oct 22 09:35:57 2009
From: nicolas.geoffray at lip6.fr (Nicolas Geoffray)
Date: Thu, 22 Oct 2009 14:35:57 -0000
Subject: [llvm-commits] [llvm] r84859 -
	/llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
Message-ID: <200910221435.n9MEZw4i031232@zion.cs.uiuc.edu>

Author: geoffray
Date: Thu Oct 22 09:35:57 2009
New Revision: 84859

URL: http://llvm.org/viewvc/llvm-project?rev=84859&view=rev
Log:
Verify that the function and exception table have been allocated
before freeing them.


Modified:
    llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp

Modified: llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp?rev=84859&r1=84858&r2=84859&view=diff

==============================================================================
--- llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp (original)
+++ llvm/trunk/lib/ExecutionEngine/JIT/JITMemoryManager.cpp Thu Oct 22 09:35:57 2009
@@ -490,13 +490,13 @@
     /// deallocateFunctionBody - Deallocate all memory for the specified
     /// function body.
     void deallocateFunctionBody(void *Body) {
-      deallocateBlock(Body);
+      if (Body) deallocateBlock(Body);
     }
 
     /// deallocateExceptionTable - Deallocate memory for the specified
     /// exception table.
     void deallocateExceptionTable(void *ET) {
-      deallocateBlock(ET);
+      if (ET) deallocateBlock(ET);
     }
 
     /// setMemoryWritable - When code generation is in progress,




From baldrick at free.fr  Thu Oct 22 11:03:32 2009
From: baldrick at free.fr (Duncan Sands)
Date: Thu, 22 Oct 2009 16:03:32 -0000
Subject: [llvm-commits] [llvm] r84861 -
	/llvm/trunk/tools/gold/gold-plugin.cpp
Message-ID: <200910221603.n9MG3W0R002154@zion.cs.uiuc.edu>

Author: baldrick
Date: Thu Oct 22 11:03:32 2009
New Revision: 84861

URL: http://llvm.org/viewvc/llvm-project?rev=84861&view=rev
Log:
Include config.h in order to have HAVE_STDINT_H be defined.
In the latest binutils the plugin-api.h needs this - without
it the LLVM gold plugin fails to compile.

Modified:
    llvm/trunk/tools/gold/gold-plugin.cpp

Modified: llvm/trunk/tools/gold/gold-plugin.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/tools/gold/gold-plugin.cpp?rev=84861&r1=84860&r2=84861&view=diff

==============================================================================
--- llvm/trunk/tools/gold/gold-plugin.cpp (original)
+++ llvm/trunk/tools/gold/gold-plugin.cpp Thu Oct 22 11:03:32 2009
@@ -12,6 +12,7 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "llvm/Config/config.h"
 #include "plugin-api.h"
 
 #include "llvm-c/lto.h"




From clattner at apple.com  Thu Oct 22 11:22:18 2009
From: clattner at apple.com (Chris Lattner)
Date: Thu, 22 Oct 2009 09:22:18 -0700
Subject: [llvm-commits] [llvm-gcc-4.2] r84829	-
	/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
In-Reply-To: <4AE0282A.5080300@free.fr>
References: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>
	<4AE0282A.5080300@free.fr>
Message-ID: <FE66EE67-889E-47F2-ABA9-FC1C429E542A@apple.com>


On Oct 22, 2009, at 2:38 AM, Duncan Sands wrote:

> Hi Chris,
>
>> add a somewhat despicable (but precendented) hack to fix PR4678 & rdar://7309675 
>> .
>
> this is awful.  Was there something wrong with Rafael's proposed  
> patch from the
> PR?


The two fixes take a different approach (as far as I understand it).   
The major difference is that I want the warning from instcombine when  
you do this in C:

void foo() {.}
  ... foo(4);  ...

but not when you do this in C++:

void foo(...) {.}
  ... foo(4);  ...

My understanding is that Rafael's patch won't produce a warning in the  
first case.

-Chris


From baldrick at free.fr  Thu Oct 22 11:33:34 2009
From: baldrick at free.fr (Duncan Sands)
Date: Thu, 22 Oct 2009 18:33:34 +0200
Subject: [llvm-commits] [llvm-gcc-4.2] r84829	-
	/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
In-Reply-To: <FE66EE67-889E-47F2-ABA9-FC1C429E542A@apple.com>
References: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>
	<4AE0282A.5080300@free.fr>
	<FE66EE67-889E-47F2-ABA9-FC1C429E542A@apple.com>
Message-ID: <4AE0895E.8000900@free.fr>

Hi Chris,

> The two fixes take a different approach (as far as I understand it).  
> The major difference is that I want the warning from instcombine when 
> you do this in C:
> 
> void foo() {.}
>  ... foo(4);  ...

gcc doesn't warn about this, even gcc-4.4, so why should llvm-gcc?  And if you
really want llvm-gcc to warn about it, why not add the logic for that to the
C frontend?

> but not when you do this in C++:
> 
> void foo(...) {.}
>  ... foo(4);  ...

> 
> My understanding is that Rafael's patch won't produce a warning in the 
> first case.

I don't remember, hopefully Rafael will comment.

Ciao,

Duncan.


From espindola at google.com  Thu Oct 22 11:47:13 2009
From: espindola at google.com (Rafael Espindola)
Date: Thu, 22 Oct 2009 12:47:13 -0400
Subject: [llvm-commits] [llvm-gcc-4.2] r84829 -
	/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
In-Reply-To: <FE66EE67-889E-47F2-ABA9-FC1C429E542A@apple.com>
References: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>
	<4AE0282A.5080300@free.fr>
	<FE66EE67-889E-47F2-ABA9-FC1C429E542A@apple.com>
Message-ID: <38a0d8450910220947l29488bc7y7938044932f59b7b@mail.gmail.com>

> My understanding is that Rafael's patch won't produce a warning in the
> first case.

My last patch will not. That is correct.

> -Chris

Cheers,
-- 
Rafael ?vila de Esp?ndola



From bob.wilson at apple.com  Thu Oct 22 11:52:22 2009
From: bob.wilson at apple.com (Bob Wilson)
Date: Thu, 22 Oct 2009 16:52:22 -0000
Subject: [llvm-commits] [llvm] r84868 - in /llvm/trunk:
 lib/Target/ARM/ARMTargetMachine.cpp test/CodeGen/ARM/ifcvt5.ll
Message-ID: <200910221652.n9MGqMCt006879@zion.cs.uiuc.edu>

Author: bwilson
Date: Thu Oct 22 11:52:21 2009
New Revision: 84868

URL: http://llvm.org/viewvc/llvm-project?rev=84868&view=rev
Log:
Revert 84843.  Evan, this was breaking some of the if-conversion tests.

Modified:
    llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp
    llvm/trunk/test/CodeGen/ARM/ifcvt5.ll

Modified: llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp?rev=84868&r1=84867&r2=84868&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp Thu Oct 22 11:52:21 2009
@@ -103,16 +103,18 @@
 bool ARMBaseTargetMachine::addPreSched2(PassManagerBase &PM,
                                         CodeGenOpt::Level OptLevel) {
   // FIXME: temporarily disabling load / store optimization pass for Thumb1.
-  if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only()) {
+  if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only())
     PM.add(createARMLoadStoreOptimizationPass());
-    PM.add(createIfConverterPass());
-  }
 
   return true;
 }
 
 bool ARMBaseTargetMachine::addPreEmitPass(PassManagerBase &PM,
                                           CodeGenOpt::Level OptLevel) {
+  // FIXME: temporarily disabling load / store optimization pass for Thumb1.
+  if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only())
+    PM.add(createIfConverterPass());
+
   if (Subtarget.isThumb2()) {
     PM.add(createThumb2ITBlockPass());
     PM.add(createThumb2SizeReductionPass());

Modified: llvm/trunk/test/CodeGen/ARM/ifcvt5.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/ifcvt5.ll?rev=84868&r1=84867&r2=84868&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/ARM/ifcvt5.ll (original)
+++ llvm/trunk/test/CodeGen/ARM/ifcvt5.ll Thu Oct 22 11:52:21 2009
@@ -11,8 +11,7 @@
 
 define void @t1(i32 %a, i32 %b) {
 ; CHECK: t1:
-; CHECK: movge
-; CHECK: blge _foo
+; CHECK: ldmltfd sp!, {r7, pc}
 entry:
 	%tmp1 = icmp sgt i32 %a, 10		; <i1> [#uses=1]
 	br i1 %tmp1, label %cond_true, label %UnifiedReturnBlock




From stuart at apple.com  Thu Oct 22 12:22:37 2009
From: stuart at apple.com (Stuart Hastings)
Date: Thu, 22 Oct 2009 17:22:37 -0000
Subject: [llvm-commits] [llvm] r84872 - in /llvm/trunk: Makefile
 utils/buildit/GNUmakefile utils/buildit/build_llvm
Message-ID: <200910221722.n9MHMbDq012872@zion.cs.uiuc.edu>

Author: stuart
Date: Thu Oct 22 12:22:37 2009
New Revision: 84872

URL: http://llvm.org/viewvc/llvm-project?rev=84872&view=rev
Log:
Trying again to tweak the top-level Makefile to facilitate an Apple-style build.
Now with Clang-compatibility.

Modified:
    llvm/trunk/Makefile
    llvm/trunk/utils/buildit/GNUmakefile
    llvm/trunk/utils/buildit/build_llvm

Modified: llvm/trunk/Makefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/Makefile?rev=84872&r1=84871&r2=84872&view=diff

==============================================================================
--- llvm/trunk/Makefile (original)
+++ llvm/trunk/Makefile Thu Oct 22 12:22:37 2009
@@ -19,6 +19,12 @@
 #
 # When cross-compiling, there are some things (tablegen) that need to
 # be build for the build system first.
+
+# If "RC_ProjectName" exists in the environment, and its value is
+# "llvmCore", then this is an "Apple-style" build; search for
+# "Apple-style" in the comments for more info.  Anything else is a
+# normal build.
+ifneq ($(RC_ProjectName),llvmCore)  # Normal build (not "Apple-style").
 ifeq ($(BUILD_DIRS_ONLY),1)
   DIRS := lib/System lib/Support utils
   OPTIONAL_DIRS :=
@@ -211,3 +217,9 @@
 
 .NOTPARALLEL:
 
+else # Building "Apple-style."
+# In an Apple-style build, once configuration is done, lines marked
+# "Apple-style" are removed with sed!  Please don't remove these!
+# Look for the string "Apple-style" in utils/buildit/build_llvm.
+include $(shell find . -name GNUmakefile) # Building "Apple-style."
+endif # Building "Apple-style."

Modified: llvm/trunk/utils/buildit/GNUmakefile
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/GNUmakefile?rev=84872&r1=84871&r2=84872&view=diff

==============================================================================
--- llvm/trunk/utils/buildit/GNUmakefile (original)
+++ llvm/trunk/utils/buildit/GNUmakefile Thu Oct 22 12:22:37 2009
@@ -59,7 +59,7 @@
 # NOTE : Always put version numbers at the end because they are optional.
 install: $(OBJROOT) $(SYMROOT) $(DSTROOT)
 	cd $(OBJROOT) && \
-	  $(SRC)/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
+	  $(SRC)/utils/buildit/build_llvm "$(RC_ARCHS)" "$(TARGETS)" \
 	    $(SRC) $(PREFIX) $(DSTROOT) $(SYMROOT) \
 	    $(LLVM_ASSERTIONS) $(LLVM_OPTIMIZED) \
 	    $(RC_ProjectSourceVersion) $(RC_ProjectSourceSubversion) 

Modified: llvm/trunk/utils/buildit/build_llvm
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/buildit/build_llvm?rev=84872&r1=84871&r2=84872&view=diff

==============================================================================
--- llvm/trunk/utils/buildit/build_llvm (original)
+++ llvm/trunk/utils/buildit/build_llvm Thu Oct 22 12:22:37 2009
@@ -82,6 +82,10 @@
 rm -rf $SRC_DIR || exit 1
 mkdir $SRC_DIR || exit 1
 ln -s $ORIG_SRC_DIR/* $SRC_DIR/ || exit 1
+# We can't use the top-level Makefile as-is.  Remove the soft link:
+rm $SRC_DIR/Makefile || exit 1
+# Now create our own by editing the top-level Makefile, deleting every line marked "Apple-style":
+sed -e '/[Aa]pple-style/d' -e '/include.*GNUmakefile/d' $ORIG_SRC_DIR/Makefile > $SRC_DIR/Makefile || exit 1
 
 # Build the LLVM tree universal.
 mkdir -p $DIR/obj-llvm || exit 1




From evan.cheng at apple.com  Thu Oct 22 12:30:22 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 10:30:22 -0700
Subject: [llvm-commits] [llvm] r84868 - in /llvm/trunk:
	lib/Target/ARM/ARMTargetMachine.cpp test/CodeGen/ARM/ifcvt5.ll
In-Reply-To: <200910221652.n9MGqMCt006879@zion.cs.uiuc.edu>
References: <200910221652.n9MGqMCt006879@zion.cs.uiuc.edu>
Message-ID: <017AE698-C17E-4593-911F-AC1E40AA5533@apple.com>

Yes sorry. I was going to revert it myself. The problem is post-regalloc scheduling can change register allocation. That doesn't work for diamond shape if-conversion when two instructions with mutually exclusive predicates must target the same register.

Evan

On Oct 22, 2009, at 9:52 AM, Bob Wilson wrote:

> Author: bwilson
> Date: Thu Oct 22 11:52:21 2009
> New Revision: 84868
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=84868&view=rev
> Log:
> Revert 84843.  Evan, this was breaking some of the if-conversion tests.
> 
> Modified:
>    llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp
>    llvm/trunk/test/CodeGen/ARM/ifcvt5.ll
> 
> Modified: llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp?rev=84868&r1=84867&r2=84868&view=diff
> 
> ==============================================================================
> --- llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp (original)
> +++ llvm/trunk/lib/Target/ARM/ARMTargetMachine.cpp Thu Oct 22 11:52:21 2009
> @@ -103,16 +103,18 @@
> bool ARMBaseTargetMachine::addPreSched2(PassManagerBase &PM,
>                                         CodeGenOpt::Level OptLevel) {
>   // FIXME: temporarily disabling load / store optimization pass for Thumb1.
> -  if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only()) {
> +  if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only())
>     PM.add(createARMLoadStoreOptimizationPass());
> -    PM.add(createIfConverterPass());
> -  }
> 
>   return true;
> }
> 
> bool ARMBaseTargetMachine::addPreEmitPass(PassManagerBase &PM,
>                                           CodeGenOpt::Level OptLevel) {
> +  // FIXME: temporarily disabling load / store optimization pass for Thumb1.
> +  if (OptLevel != CodeGenOpt::None && !Subtarget.isThumb1Only())
> +    PM.add(createIfConverterPass());
> +
>   if (Subtarget.isThumb2()) {
>     PM.add(createThumb2ITBlockPass());
>     PM.add(createThumb2SizeReductionPass());
> 
> Modified: llvm/trunk/test/CodeGen/ARM/ifcvt5.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/ifcvt5.ll?rev=84868&r1=84867&r2=84868&view=diff
> 
> ==============================================================================
> --- llvm/trunk/test/CodeGen/ARM/ifcvt5.ll (original)
> +++ llvm/trunk/test/CodeGen/ARM/ifcvt5.ll Thu Oct 22 11:52:21 2009
> @@ -11,8 +11,7 @@
> 
> define void @t1(i32 %a, i32 %b) {
> ; CHECK: t1:
> -; CHECK: movge
> -; CHECK: blge _foo
> +; CHECK: ldmltfd sp!, {r7, pc}
> entry:
> 	%tmp1 = icmp sgt i32 %a, 10		; <i1> [#uses=1]
> 	br i1 %tmp1, label %cond_true, label %UnifiedReturnBlock
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits




From dpatel at apple.com  Thu Oct 22 12:40:38 2009
From: dpatel at apple.com (Devang Patel)
Date: Thu, 22 Oct 2009 17:40:38 -0000
Subject: [llvm-commits] [llvm] r84878 - in /llvm/trunk:
 include/llvm/Metadata.h lib/VMCore/Metadata.cpp
Message-ID: <200910221740.n9MHecFP016341@zion.cs.uiuc.edu>

Author: dpatel
Date: Thu Oct 22 12:40:37 2009
New Revision: 84878

URL: http://llvm.org/viewvc/llvm-project?rev=84878&view=rev
Log:
Sort handler names to ensure deterministic behavior.

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84878&r1=84877&r2=84878&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Thu Oct 22 12:40:37 2009
@@ -285,9 +285,9 @@
   /// the same metadata to In2.
   void copyMD(Instruction *In1, Instruction *In2);
 
-  /// getHandlerNames - Get handler names. This is used by bitcode
-  /// writer.
-  void getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef> >&N) const;
+  /// getHandlerNames - Populate client supplied smallvector using custome
+  /// metadata name and ID.
+  void getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef> >&) const;
 
   /// ValueIsDeleted - This handler is used to update metadata store
   /// when a value is deleted.

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84878&r1=84877&r2=84878&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Thu Oct 22 12:40:37 2009
@@ -304,13 +304,14 @@
   return &I->second;
 }
 
-/// getHandlerNames - Get handler names. This is used by bitcode
-/// writer and aswm writer.
+/// getHandlerNames - Populate client supplied smallvector using custome
+/// metadata name and ID.
 void MetadataContext::
 getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef> >&Names) const {
   for (StringMap<unsigned>::const_iterator I = MDHandlerNames.begin(),
          E = MDHandlerNames.end(); I != E; ++I) 
     Names.push_back(std::make_pair(I->second, I->first()));
+  std::sort(Names.begin(), Names.end());
 }
 
 /// ValueIsCloned - This handler is used to update metadata store




From wendling at apple.com  Thu Oct 22 13:00:58 2009
From: wendling at apple.com (Bill Wendling)
Date: Thu, 22 Oct 2009 11:00:58 -0700
Subject: [llvm-commits] [llvm] r84130 - in /llvm/trunk/include/llvm:
	ADT/DenseMapInfo.h
	CodeGen/ScheduleDAG.h	ExecutionEngine/JITMemoryManager.h
	Metadata.h Operator.h	Pass.h Support/CommandLine.h
	Support/raw_ostream.h
In-Reply-To: <A4B4E402-D1E7-4387-A8ED-CF32C2739117@apple.com>
References: <200910142028.n9EKSXp1021686@zion.cs.uiuc.edu>
	<A4B4E402-D1E7-4387-A8ED-CF32C2739117@apple.com>
Message-ID: <5A367888-05B2-4B7E-B558-55C5B6111B1A@apple.com>


On Oct 14, 2009, at 2:00 PM, Eric Christopher wrote:

>
> On Oct 14, 2009, at 1:28 PM, Eric Christopher wrote:
>
>> -  virtual raw_ostream &changeColor(enum Colors colors, bool
>> bold=false,
>> -                                   bool  bg=false) { return *this; }
>> +  virtual raw_ostream &changeColor(enum Colors, bool, bool)
>> { return *this; }
>
> Yep. Clang build failure from this.  Fixing.  That'll teach me to
> forget to copy clang over when I checkout a fresh llvm.
>
It's allowed that you can do this:

   virtual raw_ostream &changeColor(enum Colors, bool = false, bool =  
false)
    { return *this; }

-bw



From dpatel at apple.com  Thu Oct 22 13:25:29 2009
From: dpatel at apple.com (Devang Patel)
Date: Thu, 22 Oct 2009 18:25:29 -0000
Subject: [llvm-commits] [llvm] r84884 - in /llvm/trunk:
 include/llvm/CodeGen/MachineModuleInfo.h include/llvm/Metadata.h
 lib/Bitcode/Writer/BitcodeWriter.cpp lib/Bitcode/Writer/ValueEnumerator.cpp
 lib/VMCore/AsmWriter.cpp lib/VMCore/Metadata.cpp
Message-ID: <200910221825.n9MIPUR9025036@zion.cs.uiuc.edu>

Author: dpatel
Date: Thu Oct 22 13:25:28 2009
New Revision: 84884

URL: http://llvm.org/viewvc/llvm-project?rev=84884&view=rev
Log:
Using TrackingVH instead of WeakVH or WeakMetadataVH.

Modified:
    llvm/trunk/include/llvm/CodeGen/MachineModuleInfo.h
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
    llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp
    llvm/trunk/lib/VMCore/AsmWriter.cpp
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/CodeGen/MachineModuleInfo.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/CodeGen/MachineModuleInfo.h?rev=84884&r1=84883&r2=84884&view=diff

==============================================================================
--- llvm/trunk/include/llvm/CodeGen/MachineModuleInfo.h (original)
+++ llvm/trunk/include/llvm/CodeGen/MachineModuleInfo.h Thu Oct 22 13:25:28 2009
@@ -150,7 +150,8 @@
 public:
   static char ID; // Pass identification, replacement for typeid
 
-  typedef SmallVector< std::pair< WeakMetadataVH, unsigned>, 4 > VariableDbgInfoMapTy;
+  typedef SmallVector< std::pair<TrackingVH<MDNode>, unsigned>, 4 > 
+    VariableDbgInfoMapTy;
   VariableDbgInfoMapTy VariableDbgInfo;
 
   MachineModuleInfo();

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84884&r1=84883&r2=84884&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Thu Oct 22 13:25:28 2009
@@ -144,23 +144,6 @@
 };
 
 //===----------------------------------------------------------------------===//
-/// WeakMetadataVH - a weak value handle for metadata.
-class WeakMetadataVH : public WeakVH {
-public:
-  WeakMetadataVH() : WeakVH() {}
-  WeakMetadataVH(MetadataBase *M) : WeakVH(M) {}
-  WeakMetadataVH(const WeakMetadataVH &RHS) : WeakVH(RHS) {}
-  
-  operator Value*() const {
-    llvm_unreachable("WeakMetadataVH only handles Metadata");
-  }
-
-  operator MetadataBase*() const {
-   return dyn_cast_or_null<MetadataBase>(getValPtr());
-  }
-};
-
-//===----------------------------------------------------------------------===//
 /// NamedMDNode - a tuple of other metadata. 
 /// NamedMDNode is always named. All NamedMDNode element has a type of metadata.
 template<typename ValueSubClass, typename ItemParentClass>
@@ -173,8 +156,7 @@
   NamedMDNode(const NamedMDNode &);      // DO NOT IMPLEMENT
 
   Module *Parent;
-  SmallVector<WeakMetadataVH, 4> Node;
-  typedef SmallVectorImpl<WeakMetadataVH>::iterator elem_iterator;
+  SmallVector<TrackingVH<MetadataBase>, 4> Node;
 
   void setParent(Module *M) { Parent = M; }
 protected:
@@ -216,10 +198,12 @@
 
   /// addElement - Add metadata element.
   void addElement(MetadataBase *M) {
-    Node.push_back(WeakMetadataVH(M));
+    Node.push_back(TrackingVH<MetadataBase>(M));
   }
 
-  typedef SmallVectorImpl<WeakMetadataVH>::const_iterator const_elem_iterator;
+  typedef SmallVectorImpl<TrackingVH<MetadataBase> >::iterator elem_iterator;
+  typedef SmallVectorImpl<TrackingVH<MetadataBase> >::const_iterator 
+    const_elem_iterator;
   bool elem_empty() const                { return Node.empty(); }
   const_elem_iterator elem_begin() const { return Node.begin(); }
   const_elem_iterator elem_end() const   { return Node.end();   }
@@ -241,7 +225,7 @@
 /// is [a-zA-Z$._][a-zA-Z$._0-9]*
 class MetadataContext {
 public:
-  typedef std::pair<unsigned, WeakVH> MDPairTy;
+  typedef std::pair<unsigned, TrackingVH<MDNode> > MDPairTy;
   typedef SmallVector<MDPairTy, 2> MDMapTy;
   typedef DenseMap<const Instruction *, MDMapTy> MDStoreTy;
   friend class BitcodeReader;

Modified: llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp?rev=84884&r1=84883&r2=84884&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp (original)
+++ llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp Thu Oct 22 13:25:28 2009
@@ -569,14 +569,12 @@
       bool RecordedInstruction = false;
       for (MetadataContext::MDMapTy::const_iterator PI = P->begin(), 
              PE = P->end(); PI != PE; ++PI) {
-        if (MDNode *ND = dyn_cast_or_null<MDNode>(PI->second)) {
-          if (RecordedInstruction == false) {
-            Record.push_back(VE.getInstructionID(I));
-            RecordedInstruction = true;
-          }
-          Record.push_back(PI->first);
-          Record.push_back(VE.getValueID(ND));
+        if (RecordedInstruction == false) {
+          Record.push_back(VE.getInstructionID(I));
+          RecordedInstruction = true;
         }
+        Record.push_back(PI->first);
+        Record.push_back(VE.getValueID(PI->second));
       }
       if (!Record.empty()) {
         if (!StartedMetadataBlock)  {

Modified: llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp?rev=84884&r1=84883&r2=84884&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp (original)
+++ llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp Thu Oct 22 13:25:28 2009
@@ -103,8 +103,7 @@
         if (MDs)
           for (MetadataContext::MDMapTy::const_iterator MI = MDs->begin(),
                  ME = MDs->end(); MI != ME; ++MI)
-            if (MDNode *MDN = dyn_cast_or_null<MDNode>(MI->second))
-              EnumerateMetadata(MDN);
+            EnumerateMetadata(MI->second);
       }
   }
 

Modified: llvm/trunk/lib/VMCore/AsmWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/AsmWriter.cpp?rev=84884&r1=84883&r2=84884&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/AsmWriter.cpp (original)
+++ llvm/trunk/lib/VMCore/AsmWriter.cpp Thu Oct 22 13:25:28 2009
@@ -700,8 +700,7 @@
       if (MDs)
         for (MetadataContext::MDMapTy::const_iterator MI = MDs->begin(),
                ME = MDs->end(); MI != ME; ++MI)
-          if (MDNode *MDN = dyn_cast_or_null<MDNode>(MI->second))
-            CreateMetadataSlot(MDN);
+          CreateMetadataSlot(MI->second);
     }
   }
 
@@ -2039,9 +2038,8 @@
     if (MDMap)
       for (MetadataContext::MDMapTy::const_iterator MI = MDMap->begin(),
              ME = MDMap->end(); MI != ME; ++MI)
-        if (const MDNode *MD = dyn_cast_or_null<MDNode>(MI->second))
-          Out << ", !" << MDNames[MI->first]
-              << " !" << Machine.getMetadataSlot(MD);
+        Out << ", !" << MDNames[MI->first]
+            << " !" << Machine.getMetadataSlot(MI->second);
   }
   printInfoComment(I);
 }

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84884&r1=84883&r2=84884&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Thu Oct 22 13:25:28 2009
@@ -34,7 +34,7 @@
   if (S) return S;
   
   return S = 
-    new MDString(Context, StringRef(Entry.getKeyData(), Entry.getKeyLength()));
+    new MDString(Context, Entry.getKey());
 }
 
 //===----------------------------------------------------------------------===//
@@ -151,7 +151,7 @@
   setName(N);
 
   for (unsigned i = 0; i != NumMDs; ++i)
-    Node.push_back(WeakMetadataVH(MDs[i]));
+    Node.push_back(TrackingVH<MetadataBase>(MDs[i]));
 
   if (ParentModule)
     ParentModule->getNamedMDList().push_back(this);
@@ -277,8 +277,7 @@
     return;
 
   for (MDMapTy::iterator I = In1Info.begin(), E = In1Info.end(); I != E; ++I)
-    if (MDNode *MD = dyn_cast_or_null<MDNode>(I->second))
-      addMD(I->first, MD, In2);
+    addMD(I->first, I->second, In2);
 }
 
 /// getMD - Get the metadata of given kind attached to an Instruction.
@@ -290,7 +289,7 @@
 
   for (MDMapTy::iterator I = Info.begin(), E = Info.end(); I != E; ++I)
     if (I->first == MDKind)
-      return dyn_cast_or_null<MDNode>(I->second);
+      return I->second;
   return NULL;
 }
 
@@ -326,8 +325,7 @@
   MDMapTy &In1Info = I->second;
   MDMapTy In2Info;
   for (MDMapTy::iterator I = In1Info.begin(), E = In1Info.end(); I != E; ++I)
-    if (MDNode *MD = dyn_cast_or_null<MDNode>(I->second))
-      addMD(I->first, MD, In2);
+    addMD(I->first, I->second, In2);
 }
 
 /// ValueIsRAUWd - This handler is used when V1's all uses are replaced by




From dpatel at apple.com  Thu Oct 22 13:55:16 2009
From: dpatel at apple.com (Devang Patel)
Date: Thu, 22 Oct 2009 18:55:16 -0000
Subject: [llvm-commits] [llvm] r84885 - in /llvm/trunk:
 include/llvm/Metadata.h lib/Bitcode/Writer/BitcodeWriter.cpp
 lib/Bitcode/Writer/ValueEnumerator.cpp lib/VMCore/AsmWriter.cpp
 lib/VMCore/Metadata.cpp
Message-ID: <200910221855.n9MItGM7031353@zion.cs.uiuc.edu>

Author: dpatel
Date: Thu Oct 22 13:55:16 2009
New Revision: 84885

URL: http://llvm.org/viewvc/llvm-project?rev=84885&view=rev
Log:
Fix getMDs() interface such that it does not expose implementation details.

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
    llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp
    llvm/trunk/lib/VMCore/AsmWriter.cpp
    llvm/trunk/lib/VMCore/Metadata.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84885&r1=84884&r2=84885&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Thu Oct 22 13:55:16 2009
@@ -254,7 +254,7 @@
   MDNode *getMD(unsigned Kind, const Instruction *Inst);
 
   /// getMDs - Get the metadata attached to an Instruction.
-  const MDMapTy *getMDs(const Instruction *Inst);
+  void getMDs(const Instruction *Inst, SmallVectorImpl<MDPairTy> &MDs) const;
 
   /// addMD - Attach the metadata of given kind to an Instruction.
   void addMD(unsigned Kind, MDNode *Node, Instruction *Inst);

Modified: llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp?rev=84885&r1=84884&r2=84885&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp (original)
+++ llvm/trunk/lib/Bitcode/Writer/BitcodeWriter.cpp Thu Oct 22 13:55:16 2009
@@ -561,14 +561,16 @@
   // Write metadata attachments
   // METADATA_ATTACHMENT - [m x [value, [n x [id, mdnode]]]
   MetadataContext &TheMetadata = F.getContext().getMetadata();
+  typedef SmallVector<std::pair<unsigned, TrackingVH<MDNode> >, 2> MDMapTy;
+  MDMapTy MDs;
   for (Function::const_iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
     for (BasicBlock::const_iterator I = BB->begin(), E = BB->end();
          I != E; ++I) {
-      const MetadataContext::MDMapTy *P = TheMetadata.getMDs(I);
-      if (!P) continue;
+      MDs.clear();
+      TheMetadata.getMDs(I, MDs);
       bool RecordedInstruction = false;
-      for (MetadataContext::MDMapTy::const_iterator PI = P->begin(), 
-             PE = P->end(); PI != PE; ++PI) {
+      for (MDMapTy::const_iterator PI = MDs.begin(), PE = MDs.end();
+             PI != PE; ++PI) {
         if (RecordedInstruction == false) {
           Record.push_back(VE.getInstructionID(I));
           RecordedInstruction = true;

Modified: llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp?rev=84885&r1=84884&r2=84885&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp (original)
+++ llvm/trunk/lib/Bitcode/Writer/ValueEnumerator.cpp Thu Oct 22 13:55:16 2009
@@ -87,6 +87,8 @@
       EnumerateType(I->getType());
 
     MetadataContext &TheMetadata = F->getContext().getMetadata();
+    typedef SmallVector<std::pair<unsigned, TrackingVH<MDNode> >, 2> MDMapTy;
+    MDMapTy MDs;
     for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
       for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E;++I){
         for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
@@ -99,11 +101,11 @@
           EnumerateAttributes(II->getAttributes());
 
         // Enumerate metadata attached with this instruction.
-        const MetadataContext::MDMapTy *MDs = TheMetadata.getMDs(I);
-        if (MDs)
-          for (MetadataContext::MDMapTy::const_iterator MI = MDs->begin(),
-                 ME = MDs->end(); MI != ME; ++MI)
-            EnumerateMetadata(MI->second);
+        MDs.clear();
+        TheMetadata.getMDs(I, MDs);
+        for (MDMapTy::const_iterator MI = MDs.begin(), ME = MDs.end(); MI != ME;
+             ++MI)
+          EnumerateMetadata(MI->second);
       }
   }
 

Modified: llvm/trunk/lib/VMCore/AsmWriter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/AsmWriter.cpp?rev=84885&r1=84884&r2=84885&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/AsmWriter.cpp (original)
+++ llvm/trunk/lib/VMCore/AsmWriter.cpp Thu Oct 22 13:55:16 2009
@@ -680,6 +680,8 @@
   ST_DEBUG("Inserting Instructions:\n");
 
   MetadataContext &TheMetadata = TheFunction->getContext().getMetadata();
+  typedef SmallVector<std::pair<unsigned, TrackingVH<MDNode> >, 2> MDMapTy;
+  MDMapTy MDs;
 
   // Add all of the basic blocks and instructions with no names.
   for (Function::const_iterator BB = TheFunction->begin(),
@@ -696,11 +698,11 @@
           CreateMetadataSlot(N);
 
       // Process metadata attached with this instruction.
-      const MetadataContext::MDMapTy *MDs = TheMetadata.getMDs(I);
-      if (MDs)
-        for (MetadataContext::MDMapTy::const_iterator MI = MDs->begin(),
-               ME = MDs->end(); MI != ME; ++MI)
-          CreateMetadataSlot(MI->second);
+      MDs.clear();
+      TheMetadata.getMDs(I, MDs);
+      for (MDMapTy::const_iterator MI = MDs.begin(), ME = MDs.end(); MI != ME; 
+           ++MI)
+        CreateMetadataSlot(MI->second);
     }
   }
 
@@ -2034,12 +2036,13 @@
   // Print Metadata info
   if (!MDNames.empty()) {
     MetadataContext &TheMetadata = I.getContext().getMetadata();
-    const MetadataContext::MDMapTy *MDMap = TheMetadata.getMDs(&I);
-    if (MDMap)
-      for (MetadataContext::MDMapTy::const_iterator MI = MDMap->begin(),
-             ME = MDMap->end(); MI != ME; ++MI)
-        Out << ", !" << MDNames[MI->first]
-            << " !" << Machine.getMetadataSlot(MI->second);
+    typedef SmallVector<std::pair<unsigned, TrackingVH<MDNode> >, 2> MDMapTy;
+    MDMapTy MDs;
+    TheMetadata.getMDs(&I, MDs);
+    for (MDMapTy::const_iterator MI = MDs.begin(), ME = MDs.end(); MI != ME; 
+         ++MI)
+      Out << ", !" << MDNames[MI->first]
+          << " !" << Machine.getMetadataSlot(MI->second);
   }
   printInfoComment(I);
 }

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84885&r1=84884&r2=84885&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Thu Oct 22 13:55:16 2009
@@ -294,13 +294,15 @@
 }
 
 /// getMDs - Get the metadata attached to an Instruction.
-const MetadataContext::MDMapTy *
-MetadataContext::getMDs(const Instruction *Inst) {
+void MetadataContext::
+getMDs(const Instruction *Inst, SmallVectorImpl<MDPairTy> &MDs) const {
   MDStoreTy::iterator I = MetadataStore.find(Inst);
   if (I == MetadataStore.end())
-    return NULL;
-  
-  return &I->second;
+    return;
+  for (MDMapTy::iterator MI = I->second.begin(), ME = I->second.end();
+       MI != ME; ++MI)
+    MDs.push_back(std::make_pair(MI->first, MI->second));
+  std::sort(MDs.begin(), MDs.end());
 }
 
 /// getHandlerNames - Populate client supplied smallvector using custome




From dpatel at apple.com  Thu Oct 22 14:37:01 2009
From: dpatel at apple.com (Devang Patel)
Date: Thu, 22 Oct 2009 19:37:01 -0000
Subject: [llvm-commits] [llvm] r84886 - in /llvm/trunk:
 include/llvm/Metadata.h include/llvm/Value.h
 lib/Bitcode/Reader/BitcodeReader.cpp lib/VMCore/Metadata.cpp
 lib/VMCore/Pass.cpp
Message-ID: <200910221937.n9MJb4a1006964@zion.cs.uiuc.edu>

Author: dpatel
Date: Thu Oct 22 14:36:54 2009
New Revision: 84886

URL: http://llvm.org/viewvc/llvm-project?rev=84886&view=rev
Log:
Hide MetadataContext implementation details.

Modified:
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/include/llvm/Value.h
    llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp
    llvm/trunk/lib/VMCore/Metadata.cpp
    llvm/trunk/lib/VMCore/Pass.cpp

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84886&r1=84885&r2=84886&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Thu Oct 22 14:36:54 2009
@@ -20,8 +20,6 @@
 #include "llvm/Type.h"
 #include "llvm/ADT/FoldingSet.h"
 #include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/StringMap.h"
 #include "llvm/ADT/ilist_node.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/ValueHandle.h"
@@ -30,6 +28,7 @@
 class Constant;
 class Instruction;
 class LLVMContext;
+class MetadataContextImpl;
 
 //===----------------------------------------------------------------------===//
 // MetadataBase  - A base class for MDNode, MDString and NamedMDNode.
@@ -224,20 +223,15 @@
 /// must start with an alphabet. The regular expression used to check name
 /// is [a-zA-Z$._][a-zA-Z$._0-9]*
 class MetadataContext {
-public:
-  typedef std::pair<unsigned, TrackingVH<MDNode> > MDPairTy;
-  typedef SmallVector<MDPairTy, 2> MDMapTy;
-  typedef DenseMap<const Instruction *, MDMapTy> MDStoreTy;
-  friend class BitcodeReader;
-private:
-
-  /// MetadataStore - Collection of metadata used in this context.
-  MDStoreTy MetadataStore;
-
-  /// MDHandlerNames - Map to hold metadata handler names.
-  StringMap<unsigned> MDHandlerNames;
+  // DO NOT IMPLEMENT
+  MetadataContext(MetadataContext&);
+  void operator=(MetadataContext&);
 
+  MetadataContextImpl *const pImpl;
 public:
+  MetadataContext();
+  ~MetadataContext();
+
   /// registerMDKind - Register a new metadata kind and return its ID.
   /// A metadata kind can be registered only once. 
   unsigned registerMDKind(StringRef Name);
@@ -254,7 +248,8 @@
   MDNode *getMD(unsigned Kind, const Instruction *Inst);
 
   /// getMDs - Get the metadata attached to an Instruction.
-  void getMDs(const Instruction *Inst, SmallVectorImpl<MDPairTy> &MDs) const;
+  void getMDs(const Instruction *Inst, 
+        SmallVectorImpl<std::pair<unsigned, TrackingVH<MDNode> > > &MDs) const;
 
   /// addMD - Attach the metadata of given kind to an Instruction.
   void addMD(unsigned Kind, MDNode *Node, Instruction *Inst);
@@ -276,9 +271,7 @@
   /// ValueIsDeleted - This handler is used to update metadata store
   /// when a value is deleted.
   void ValueIsDeleted(const Value *) {}
-  void ValueIsDeleted(Instruction *Inst) {
-    removeAllMetadata(Inst);
-  }
+  void ValueIsDeleted(Instruction *Inst);
   void ValueIsRAUWd(Value *V1, Value *V2);
 
   /// ValueIsCloned - This handler is used to update metadata store

Modified: llvm/trunk/include/llvm/Value.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Value.h?rev=84886&r1=84885&r2=84886&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Value.h (original)
+++ llvm/trunk/include/llvm/Value.h Thu Oct 22 14:36:54 2009
@@ -42,7 +42,7 @@
 class AssemblyAnnotationWriter;
 class ValueHandleBase;
 class LLVMContext;
-class MetadataContext;
+class MetadataContextImpl;
 
 //===----------------------------------------------------------------------===//
 //                                 Value Class
@@ -83,7 +83,7 @@
   friend class ValueSymbolTable; // Allow ValueSymbolTable to directly mod Name.
   friend class SymbolTable;      // Allow SymbolTable to directly poke Name.
   friend class ValueHandleBase;
-  friend class MetadataContext;
+  friend class MetadataContextImpl;
   friend class AbstractTypeUser;
   ValueName *Name;
 

Modified: llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp?rev=84886&r1=84885&r2=84886&view=diff

==============================================================================
--- llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp (original)
+++ llvm/trunk/lib/Bitcode/Reader/BitcodeReader.cpp Thu Oct 22 14:36:54 2009
@@ -840,7 +840,15 @@
       for (unsigned i = 1; i != RecordLength; ++i)
         Name[i-1] = Record[i];
       MetadataContext &TheMetadata = Context.getMetadata();
-      TheMetadata.MDHandlerNames[Name.str()] = Kind;
+      unsigned ExistingKind = TheMetadata.getMDKind(Name.str());
+      if (ExistingKind == 0) {
+        unsigned NewKind = TheMetadata.registerMDKind(Name.str());
+        assert (Kind == NewKind 
+                && "Unable to handle custom metadata mismatch!");
+      } else {
+        assert (ExistingKind == Kind 
+                && "Unable to handle custom metadata mismatch!");
+      }
       break;
     }
     }

Modified: llvm/trunk/lib/VMCore/Metadata.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84886&r1=84885&r2=84886&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Metadata.cpp (original)
+++ llvm/trunk/lib/VMCore/Metadata.cpp Thu Oct 22 14:36:54 2009
@@ -16,6 +16,8 @@
 #include "llvm/LLVMContext.h"
 #include "llvm/Module.h"
 #include "llvm/Instruction.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/StringMap.h"
 #include "SymbolTableListTraitsImpl.h"
 using namespace llvm;
 
@@ -182,48 +184,91 @@
 }
 
 //===----------------------------------------------------------------------===//
-// MetadataContext implementation.
+// MetadataContextImpl implementation.
 //
+namespace llvm {
+class MetadataContextImpl {
+public:
+  typedef std::pair<unsigned, TrackingVH<MDNode> > MDPairTy;
+  typedef SmallVector<MDPairTy, 2> MDMapTy;
+  typedef DenseMap<const Instruction *, MDMapTy> MDStoreTy;
+  friend class BitcodeReader;
+private:
+
+  /// MetadataStore - Collection of metadata used in this context.
+  MDStoreTy MetadataStore;
+
+  /// MDHandlerNames - Map to hold metadata handler names.
+  StringMap<unsigned> MDHandlerNames;
+
+public:
+  /// registerMDKind - Register a new metadata kind and return its ID.
+  /// A metadata kind can be registered only once. 
+  unsigned registerMDKind(StringRef Name);
+
+  /// getMDKind - Return metadata kind. If the requested metadata kind
+  /// is not registered then return 0.
+  unsigned getMDKind(StringRef Name) const;
+
+  /// getMD - Get the metadata of given kind attached to an Instruction.
+  /// If the metadata is not found then return 0.
+  MDNode *getMD(unsigned Kind, const Instruction *Inst);
+
+  /// getMDs - Get the metadata attached to an Instruction.
+  void getMDs(const Instruction *Inst, SmallVectorImpl<MDPairTy> &MDs) const;
+
+  /// addMD - Attach the metadata of given kind to an Instruction.
+  void addMD(unsigned Kind, MDNode *Node, Instruction *Inst);
+  
+  /// removeMD - Remove metadata of given kind attached with an instuction.
+  void removeMD(unsigned Kind, Instruction *Inst);
+  
+  /// removeAllMetadata - Remove all metadata attached with an instruction.
+  void removeAllMetadata(Instruction *Inst);
+
+  /// copyMD - If metadata is attached with Instruction In1 then attach
+  /// the same metadata to In2.
+  void copyMD(Instruction *In1, Instruction *In2);
+
+  /// getHandlerNames - Populate client supplied smallvector using custome
+  /// metadata name and ID.
+  void getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef> >&) const;
+
+  /// ValueIsDeleted - This handler is used to update metadata store
+  /// when a value is deleted.
+  void ValueIsDeleted(const Value *) {}
+  void ValueIsDeleted(Instruction *Inst) {
+    removeAllMetadata(Inst);
+  }
+  void ValueIsRAUWd(Value *V1, Value *V2);
+
+  /// ValueIsCloned - This handler is used to update metadata store
+  /// when In1 is cloned to create In2.
+  void ValueIsCloned(const Instruction *In1, Instruction *In2);
+};
+}
 
 /// registerMDKind - Register a new metadata kind and return its ID.
 /// A metadata kind can be registered only once. 
-unsigned MetadataContext::registerMDKind(StringRef Name) {
-  assert(isValidName(Name) && "Invalid custome metadata name!");
+unsigned MetadataContextImpl::registerMDKind(StringRef Name) {
   unsigned Count = MDHandlerNames.size();
   assert(MDHandlerNames.count(Name) == 0 && "Already registered MDKind!");
   return MDHandlerNames[Name] = Count + 1;
 }
 
-/// isValidName - Return true if Name is a valid custom metadata handler name.
-bool MetadataContext::isValidName(StringRef MDName) {
-  if (MDName.empty())
-    return false;
-
-  if (!isalpha(MDName[0]))
-    return false;
-
-  for (StringRef::iterator I = MDName.begin() + 1, E = MDName.end(); I != E;
-       ++I) {
-    if (!isalnum(*I) && *I != '_' && *I != '-' && *I != '.')
-        return false;
-  }
-  return true;
-}
-
 /// getMDKind - Return metadata kind. If the requested metadata kind
 /// is not registered then return 0.
-unsigned MetadataContext::getMDKind(StringRef Name) const {
+unsigned MetadataContextImpl::getMDKind(StringRef Name) const {
   StringMap<unsigned>::const_iterator I = MDHandlerNames.find(Name);
-  if (I == MDHandlerNames.end()) {
-    assert(isValidName(Name) && "Invalid custome metadata name!");
+  if (I == MDHandlerNames.end())
     return 0;
-  }
 
   return I->getValue();
 }
 
 /// addMD - Attach the metadata of given kind to an Instruction.
-void MetadataContext::addMD(unsigned MDKind, MDNode *Node, Instruction *Inst) {
+void MetadataContextImpl::addMD(unsigned MDKind, MDNode *Node, 
+                                Instruction *Inst) {
   assert(Node && "Invalid null MDNode");
   Inst->HasMetadata = true;
   MDMapTy &Info = MetadataStore[Inst];
@@ -247,7 +292,7 @@
 }
 
 /// removeMD - Remove metadata of given kind attached with an instuction.
-void MetadataContext::removeMD(unsigned Kind, Instruction *Inst) {
+void MetadataContextImpl::removeMD(unsigned Kind, Instruction *Inst) {
   MDStoreTy::iterator I = MetadataStore.find(Inst);
   if (I == MetadataStore.end())
     return;
@@ -263,14 +308,14 @@
 }
   
 /// removeAllMetadata - Remove all metadata attached with an instruction.
-void MetadataContext::removeAllMetadata(Instruction *Inst) {
+void MetadataContextImpl::removeAllMetadata(Instruction *Inst) {
   MetadataStore.erase(Inst);
   Inst->HasMetadata = false;
 }
 
 /// copyMD - If metadata is attached with Instruction In1 then attach
 /// the same metadata to In2.
-void MetadataContext::copyMD(Instruction *In1, Instruction *In2) {
+void MetadataContextImpl::copyMD(Instruction *In1, Instruction *In2) {
   assert(In1 && In2 && "Invalid instruction!");
   MDMapTy &In1Info = MetadataStore[In1];
   if (In1Info.empty())
@@ -282,7 +327,7 @@
 
 /// getMD - Get the metadata of given kind attached to an Instruction.
 /// If the metadata is not found then return 0.
-MDNode *MetadataContext::getMD(unsigned MDKind, const Instruction *Inst) {
+MDNode *MetadataContextImpl::getMD(unsigned MDKind, const Instruction *Inst) {
   MDMapTy &Info = MetadataStore[Inst];
   if (Info.empty())
     return NULL;
@@ -294,7 +339,7 @@
 }
 
 /// getMDs - Get the metadata attached to an Instruction.
-void MetadataContext::
+void MetadataContextImpl::
 getMDs(const Instruction *Inst, SmallVectorImpl<MDPairTy> &MDs) const {
   MDStoreTy::iterator I = MetadataStore.find(Inst);
   if (I == MetadataStore.end())
@@ -307,7 +352,7 @@
 
 /// getHandlerNames - Populate client supplied smallvector using custome
 /// metadata name and ID.
-void MetadataContext::
+void MetadataContextImpl::
 getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef> >&Names) const {
   for (StringMap<unsigned>::const_iterator I = MDHandlerNames.begin(),
          E = MDHandlerNames.end(); I != E; ++I) 
@@ -317,7 +362,8 @@
 
 /// ValueIsCloned - This handler is used to update metadata store
 /// when In1 is cloned to create In2.
-void MetadataContext::ValueIsCloned(const Instruction *In1, Instruction *In2) {
+void MetadataContextImpl::ValueIsCloned(const Instruction *In1, 
+                                        Instruction *In2) {
   // Find Metadata handles for In1.
   MDStoreTy::iterator I = MetadataStore.find(In1);
   assert(I != MetadataStore.end() && "Invalid custom metadata info!");
@@ -332,7 +378,7 @@
 
 /// ValueIsRAUWd - This handler is used when V1's all uses are replaced by
 /// V2.
-void MetadataContext::ValueIsRAUWd(Value *V1, Value *V2) {
+void MetadataContextImpl::ValueIsRAUWd(Value *V1, Value *V2) {
   Instruction *I1 = dyn_cast<Instruction>(V1);
   Instruction *I2 = dyn_cast<Instruction>(V2);
   if (!I1 || !I2)
@@ -342,3 +388,94 @@
   ValueIsCloned(I1, I2);
 }
 
+//===----------------------------------------------------------------------===//
+// MetadataContext implementation.
+//
+MetadataContext::MetadataContext() 
+  : pImpl(new MetadataContextImpl()) { }
+MetadataContext::~MetadataContext() { delete pImpl; }
+
+/// isValidName - Return true if Name is a valid custom metadata handler name.
+bool MetadataContext::isValidName(StringRef MDName) {
+  if (MDName.empty())
+    return false;
+
+  if (!isalpha(MDName[0]))
+    return false;
+
+  for (StringRef::iterator I = MDName.begin() + 1, E = MDName.end(); I != E;
+       ++I) {
+    if (!isalnum(*I) && *I != '_' && *I != '-' && *I != '.')
+        return false;
+  }
+  return true;
+}
+
+/// registerMDKind - Register a new metadata kind and return its ID.
+/// A metadata kind can be registered only once. 
+unsigned MetadataContext::registerMDKind(StringRef Name) {
+  assert(isValidName(Name) && "Invalid custome metadata name!");
+  return pImpl->registerMDKind(Name);
+}
+
+/// getMDKind - Return metadata kind. If the requested metadata kind
+/// is not registered then return 0.
+unsigned MetadataContext::getMDKind(StringRef Name) const {
+  return pImpl->getMDKind(Name);
+}
+
+/// getMD - Get the metadata of given kind attached to an Instruction.
+/// If the metadata is not found then return 0.
+MDNode *MetadataContext::getMD(unsigned Kind, const Instruction *Inst) {
+  return pImpl->getMD(Kind, Inst);
+}
+
+/// getMDs - Get the metadata attached to an Instruction.
+void MetadataContext::
+getMDs(const Instruction *Inst, 
+       SmallVectorImpl<std::pair<unsigned, TrackingVH<MDNode> > > &MDs) const {
+  return pImpl->getMDs(Inst, MDs);
+}
+
+/// addMD - Attach the metadata of given kind to an Instruction.
+void MetadataContext::addMD(unsigned Kind, MDNode *Node, Instruction *Inst) {
+  pImpl->addMD(Kind, Node, Inst);
+}
+  
+/// removeMD - Remove metadata of given kind attached with an instuction.
+void MetadataContext::removeMD(unsigned Kind, Instruction *Inst) {
+  pImpl->removeMD(Kind, Inst);
+}
+  
+/// removeAllMetadata - Remove all metadata attached with an instruction.
+void MetadataContext::removeAllMetadata(Instruction *Inst) {
+  pImpl->removeAllMetadata(Inst);
+}
+
+/// copyMD - If metadata is attached with Instruction In1 then attach
+/// the same metadata to In2.
+void MetadataContext::copyMD(Instruction *In1, Instruction *In2) {
+  pImpl->copyMD(In1, In2);
+}
+
+/// getHandlerNames - Populate client supplied smallvector using custome
+/// metadata name and ID.
+void MetadataContext::
+getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef> >&N) const {
+  pImpl->getHandlerNames(N);
+}
+
+/// ValueIsDeleted - This handler is used to update metadata store
+/// when a value is deleted.
+void MetadataContext::ValueIsDeleted(Instruction *Inst) {
+  pImpl->ValueIsDeleted(Inst);
+}
+void MetadataContext::ValueIsRAUWd(Value *V1, Value *V2) {
+  pImpl->ValueIsRAUWd(V1, V2);
+}
+
+/// ValueIsCloned - This handler is used to update metadata store
+/// when In1 is cloned to create In2.
+void MetadataContext::ValueIsCloned(const Instruction *In1, Instruction *In2) {
+  pImpl->ValueIsCloned(In1, In2);
+}

Modified: llvm/trunk/lib/VMCore/Pass.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Pass.cpp?rev=84886&r1=84885&r2=84886&view=diff

==============================================================================
--- llvm/trunk/lib/VMCore/Pass.cpp (original)
+++ llvm/trunk/lib/VMCore/Pass.cpp Thu Oct 22 14:36:54 2009
@@ -18,6 +18,7 @@
 #include "llvm/Module.h"
 #include "llvm/ModuleProvider.h"
 #include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/StringMap.h"
 #include "llvm/Support/ManagedStatic.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/System/Atomic.h"




From devang.patel at gmail.com  Thu Oct 22 14:53:07 2009
From: devang.patel at gmail.com (Devang Patel)
Date: Thu, 22 Oct 2009 12:53:07 -0700
Subject: [llvm-commits] metadata.h/cpp review stuff
In-Reply-To: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>
References: <95965B98-8FE7-4E5E-89F3-211A932FD5C5@apple.com>
Message-ID: <352a1fb20910221253m29a2093dra229a2e741cdd2d5@mail.gmail.com>

Chris,

Thanks for detailed review. I have covered most of the points you
brought up, except few.

On Mon, Oct 19, 2009 at 12:06 AM, Chris Lattner <clattner at apple.com> wrote:
> Hi Devang,
>
> I'm just now getting caught up on your work. ?Here are some thoughts/
> questions on Metadata.h/cpp:
>
>
> The metadata hierarchy is:
>
> User
> ? -> MetadataBase
> ? ? ?-> MDString
> ? ? ?-> MDNode
> ? ? ?-> NamedMDNode
>
> Why not:
>
> Value
> ? -> MetadataBase
> ? ? ?-> MDString
> ? ? ?-> MDNode
> ? ? ?-> NamedMDNode
>
> This would shrink them significantly and reduce the games you have to
> play with hiding getNumOperands, overloading operator new, etc. ?Also,
> MetadataBase doesn't have much common state, it isn't clear it needs
> to exist at all.

Done.

> There is some seriously confused code circling around
> MetadataBase::resizeOperands. ?It looks like half the code thinks that
> operands can happen, but half doesn't. ?Please remove all this stuff.

This code disappeared once I fixed class hierarchy.

>
> In Metadata.h:
>
>
> #include "llvm/User.h"
> #include "llvm/Type.h"
> #include "llvm/OperandTraits.h"
> #include "llvm/ADT/FoldingSet.h"
> #include "llvm/ADT/SmallVector.h"
> #include "llvm/ADT/SmallPtrSet.h"
> #include "llvm/ADT/DenseMap.h"
> #include "llvm/ADT/StringMap.h"
> #include "llvm/ADT/ilist_node.h"
> #include "llvm/Support/ErrorHandling.h"
> #include "llvm/Support/ValueHandle.h"
>
> This is including way too much. ?Please reduce this as aggressively as
> possible.

Done.

> class MDString : public MetadataBase {
> ...
> ? explicit MDString(LLVMContext &C, const char *begin, unsigned l)
>
> This ctor should just take a StringRef.

Done.

>
> ...
> ? static MDString *get(LLVMContext &Context, const StringRef &Str);
>
> Please take StringRef's by value, not by const&.

Done.

> class MDNode : public MetadataBase, public FoldingSetNode {
> ...
> ? friend class ElementVH;
> ? // Use CallbackVH to hold MDNOde elements.
> ? struct ElementVH : public CallbackVH {
> ? ? MDNode *Parent;
> ? ? ElementVH(Value *V, MDNode *P) : CallbackVH(V), Parent(P) {}
> ? ? ~ElementVH() {}
>
> ? ? virtual void deleted() {
> ? ? ? Parent->replaceElement(this->operator Value*(), 0);
> ? ? }
>
> ? ? virtual void allUsesReplacedWith(Value *NV) {
> ? ? ? Parent->replaceElement(this->operator Value*(), NV);
> ? ? }
> ? };
> ? // Replace each instance of F from the element list of this node
> with T.
> ? void replaceElement(Value *F, Value *T);
>
> ? SmallVector<ElementVH, 4> Node;
>
>
> Please move all of this stuff out of line to Metadata.cpp. ?If needbe,
> Node should be declared as a 'void*' in the header. ?Also, there is no
> reason to use a SmallVector here: the size is fixed on node
> construction. ?Just use a pointer to array and new[]/delete[] the
> array. ?Keeping a SmallVector in NamedMDNode (but not MDNode) is
> completely fine.

Now, node is a pointer instead of SmallVector.

>
> Why does MDNode::dropAllReferences need to exist? ?It seems *highly*
> unsafe because it clears 'Node' but doesn't update the FoldingSet.

Now dropAllReferences() does not exist. Earlier it was invoked after
FoldingSet was updated.

>
> ? Value *getElement(unsigned i) const {
> ? ? assert(getNumElements() > i && "Invalid element number!");
>
> This is fine, but please use "i < getNumElements()" because that is
> more idiomatic. ?This happens multiple times in the file.

Done.

>
> /// WeakMetadataVH - a weak value handle for metadata.
> class WeakMetadataVH : public WeakVH {
> public:
>
> I think this should be removed and the use replaced with TrackingVH??

Done.
>
> NamedMDNode::setParent shouldn't be public. ?Does
> NamedMDNode::dropAllReferences need to exist?
>
> I don't understand NamedMDNode::Create(const NamedMDNode *NMD, Module
> *M). ?Is this a weird copy ctor? ?Why does it need to exist?

It is used while linking modules.

>
> ? void addElement(MetadataBase *M) {
> ? ? resizeOperands(0);
> ? ? OperandList[NumOperands++] = M;
> ? ? Node.push_back(WeakMetadataVH(M));
> ? }
>
> How can this work? ?Isn't this an invalid access into OperandList?

resizeOperands() did some magic to make this work:) Now it is gone.

>
> /// MetadataContext -
> /// MetadataContext handles uniquing and assignment of IDs for custom
> metadata
> /// types. Custom metadata handler names do not contain spaces. And
> the name
> /// must start with an alphabet. The regular expression used to check
> name
>
> "start with an alphabet" is missing something.
>
> class MetadataContext {
> public:
> ? typedef std::pair<unsigned, WeakVH> MDPairTy;
> ? typedef SmallVector<MDPairTy, 2> MDMapTy;
> ? typedef DenseMap<const Instruction *, MDMapTy> MDStoreTy;
> ? friend class BitcodeReader;
> private:
>
> ? /// MetadataStore - Collection of metadata used in this context.
> ? MDStoreTy MetadataStore;
>
> ? /// MDHandlerNames - Map to hold metadata handler names.
> ? StringMap<unsigned> MDHandlerNames;
>
>
> These datastructures are all implementation details. ?They should be
> hidden behind a void*, allowing you to avoid #include of DenseMap.h,
> StringMap.h etc. ?MDPairTy should be replaced with TrackingVH<MDNode>.

These data structures are now hiding inside Metadata.cpp

>
> ? /// RegisterMDKind - Register a new metadata kind and return its ID.
> ? /// A metadata kind can be registered only once.
> ? unsigned RegisterMDKind(const char *Name);
>
> ? /// getMDKind - Return metadata kind. If the requested metadata kind
> ? /// is not registered then return 0.
> ? unsigned getMDKind(const char *Name);
>
> ? /// validName - Return true if Name is a valid custom metadata
> handler name.
> ? bool validName(const char *Name);
>
> These methods should all take StringRef for consistency. ?The second
> should be const, the third should be static. ?validName ->
> isValidName. ?RegisterMDKind -> registerMDKind.

Done.

> ? /// getMD - Get the metadata of given kind attached with an
> Instruction.
> ? /// If the metadata is not found then return 0.
> ? MDNode *getMD(unsigned Kind, const Instruction *Inst);
>
> "with an instruction" -> "to an instruction". ? This occurs several
> times in both files.

Fixed.

> ? /// removeMDs - Remove all metadata attached with an instruction.
> ? void removeMDs(const Instruction *Inst);
>
> removeAllMetadata()?

ok, renamed.

>
> ? /// getHandlerNames - Get handler names. This is used by bitcode
> ? /// writer.
> ? const StringMap<unsigned> *getHandlerNames();
>
> This is gross for several reasons: at the least, it should be const.
> It also exposes implementation details. ?Finally, the bitcode writer
> is currently writing these out in iterator order which is not at all
> deterministic. ?It would be better to have something like:
>
> ? void getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef>
> ?> &) const;
>
> and have the implementation of getHandlerNames do the sort (with
> array_pod_sort).

Fixed.

>
> ? /// ValueIsDeleted - This handler is used to update metadata store
> ? /// when a value is deleted.
> ? void ValueIsDeleted(const Value *) {}
> ? void ValueIsDeleted(const Instruction *Inst) {
> ? ? removeMDs(Inst);
> ? }
> ? void ValueIsRAUWd(Value *V1, Value *V2);
>
> ? /// ValueIsCloned - This handler is used to update metadata store
> ? /// when In1 is cloned to create In2.
> ? void ValueIsCloned(const Instruction *In1, Instruction *In2);
>
> These should not be public and/or should not be in the MetadataContext
> class.

I am not sure why ?

>
> Metadata.cpp:
>
> void MDNode::replaceElement(Value *From, Value *To) {
> ...
>
> You don't know that MDNode::replaceElement will be called before or
> after the VH operands uses are updated.

I do. This is a private method used by VH handlers only.

> ?I don't think this method in
> general is safe because a bunch of stuff depends on pointer equality
> with the "from" value.
>
>
> ? // MDNode only lists metadata elements in operand list, because
> MDNode
> ? // used by MDNode is considered a valid use. However on the side,
> MDNode
> ? // using a non-metadata value is not considered a "use" of non-
> metadata
> ? // value.
>
> Why do MD uses try to show up in the operand list? ?This seems really
> complex, what benefit does it provide?

Now MD uses do not show up in the operand list :)

>
> ? if (MetadataBase *MDTo = dyn_cast_or_null<MetadataBase>(To)) {
> ? ? for (SmallVector<unsigned, 4>::iterator OI = OpIndexes.begin(),
> ? ? ? ? ? ?OE = OpIndexes.end(); OI != OE; ++OI)
> ? ? ? setOperand(*OI, MDTo);
> ? } else {
> ? ? for (SmallVector<unsigned, 4>::iterator OI = OpIndexes.begin(),
> ? ? ? ? ? ?OE = OpIndexes.end(); OI != OE; ++OI)
> ? ? ? setOperand(*OI, 0);
> ? }
>
> This doesn't make sense to me. ?If one MD operand is replaced, it sets
> them all to the new thing?

This code is gone now.

>
> NamedMDNode::NamedMDNode(LLVMContext &C, const Twine &N,
> ...
> ? resizeOperands(NumMDs);
>
> ? for (unsigned i = 0; i != NumMDs; ++i) {
> ? ? if (MDs[i])
> ? ? ? OperandList[NumOperands++] = MDs[i];
> ? ? Node.push_back(WeakMetadataVH(MDs[i]));
> ? }
>
> This is wrong unless resizeOperands does something really strange
> (likely).

It did :)

> ?Why does NamedMDNode put anything in the 'User' operands
> list at all? ?I think this should just be removed.

This code is gone, because now metadata is not derived from User.

>
> unsigned MetadataContext::RegisterMDKind(const char *Name) {
> ..
> ? assert(MDHandlerNames.find(Name) == MDHandlerNames.end()
> ? ? ? ? ?&& "Already registered MDKind!");
>
> Just use: assert(MDHandlerNames.count(Name) == 0 && ...

Done.

>
>
> ? MDHandlerNames[Name] = Count + 1;
> ? return Count + 1;
>
> How about: return MDHandlerNames[Name] = Count + 1;

Done.

>
> unsigned MetadataContext::getMDKind(const char *Name) {
> ? assert(validName(Name) && "Invalid custome metadata name!");
> ? StringMap<unsigned>::iterator I = MDHandlerNames.find(Name);
>
> Random microoptimization for assert builds: the check for a valid name
> can be sunk into the case where the query fails. ?If the query
> succeeds (common case) the name was already validated.
>

Done.

> void MetadataContext::addMD(unsigned MDKind, MDNode *Node, Instruction
> *Inst) {
> ...
> ? MDStoreTy::iterator I = MetadataStore.find(Inst);
> ? if (I == MetadataStore.end()) {
> ? ? MDMapTy Info;
> ? ? Info.push_back(std::make_pair(MDKind, Node));
> ? ? MetadataStore.insert(std::make_pair(Inst, Info));
> ? ? return;
> ? }
>
> It is actually better to just rely on default construction in the
> map. ?Just use:
> ? MDMapTy &Info = MetadataStore[Inst];
> and delete the code above.

Done.

>
> /// removeMD - Remove metadata of given kind attached with an
> instuction.
> void MetadataContext::removeMD(unsigned Kind, Instruction *Inst) {
> ? MDStoreTy::iterator I = MetadataStore.find(Inst);
> ? if (I == MetadataStore.end())
> ? ? return;
> ? MDMapTy &Info = I->second;
>
> This seems wrong. ?The method should only be called if the
> 'HasMetadata' bit is set on the instruction. ?If it is set and there
> is no entry in MetaDataStore, then it is a logic bug (so it should
> assert). ?Just do:

This is intentional, can be used by a pass to strip all debug info
where each instruction may not have debug info attached to itself.

> MDMapTy &Info = MetadataStore[Inst];
> assert(!Info.empty() && ..
>
> or whatever.
>
>
>
> /// removeMDs - Remove all metadata attached with an instruction.
> void MetadataContext::removeMDs(const Instruction *Inst) {
> ? // Find Metadata handles for this instruction.
> ? MDStoreTy::iterator I = MetadataStore.find(Inst);
> ? assert(I != MetadataStore.end() && "Invalid custom metadata info!");
> ? MDMapTy &Info = I->second;
>
> ? // FIXME : Give all metadata handlers a chance to adjust.
>
> ? // Remove the entries for this instruction.
> ? Info.clear();
> ? MetadataStore.erase(I);
> }
>
>
> This doesn't look fully implemented. ? In addition to the FIXME (What
> does it mean?) this should clear the ?HasMetadata bit on the
> instruction. ?Also, don't use find here. ?This entire method can
> currently be replaced with this, which is identical in behavior:
>
> void MetadataContext::removeMDs(const Instruction *Inst) {
> ? MetadataStore.erase(Inst);
> }

Done.

>
> void MetadataContext::copyMD(Instruction *In1, Instruction *In2) {
> ? assert(In1 && In2 && "Invalid instruction!");
> ? ?MDStoreTy::iterator I = MetadataStore.find(In1);
> ? if (I == MetadataStore.end())
> ? ? return;
>
> ? MDMapTy &In1Info = I->second;
> ? MDMapTy In2Info;
> ? for (MDMapTy::iterator I = In1Info.begin(), E = In1Info.end(); I !=
> E; ++I)
> ? ? if (MDNode *MD = dyn_cast_or_null<MDNode>(I->second))
> ? ? ? addMD(I->first, MD, In2);
>
>
> Similar comments to above. ?Also, In2Info is dead here.
> dyn_cast_or_null should go away when you start using trackingvh
> instead of weakvh (not just here, check the rest of the file).

Done.

>
>
> MDNode *MetadataContext::getMD(unsigned MDKind, const Instruction
> *Inst) {
> ? MDStoreTy::iterator I = MetadataStore.find(Inst);
> ? if (I == MetadataStore.end())
> ? ? return NULL;
>
> Should be an assert as above.

It is used by clients to see if metadata of given kind is attached to
this inst or not.

>
> const MetadataContext::MDMapTy *
> MetadataContext::getMDs(const Instruction *Inst) {
>
> I really don't like this method. ?Exposing implementation details like
> this is bad. ?IF this is for the bc writer, please follow something
> similar to the getHandlerNames approach above.

Fixed.

> ?void MetadataContext::ValueIsCloned(const Instruction *In1,
> Instruction *In2) {
> ? // Find Metadata handles for In1.
> ? MDStoreTy::iterator I = MetadataStore.find(In1);
> ? assert(I != MetadataStore.end() && "Invalid custom metadata info!");
>
> ? // FIXME : Give all metadata handlers a chance to adjust.
>
> ? MDMapTy &In1Info = I->second;
> ? MDMapTy In2Info;
> ? for (MDMapTy::iterator I = In1Info.begin(), E = In1Info.end(); I !=
> E; ++I)
> ? ? if (MDNode *MD = dyn_cast_or_null<MDNode>(I->second))
> ? ? ? addMD(I->first, MD, In2);
> }
>
> How is this different than copyMD?

ValueIsCloned() is used by VH handlers and it has assertion checks.
copyMD() is available to clients to copy metadata, if available, from
one instruction to another instruction.

> /// ValueIsRAUWd - This handler is used when V1's all uses are
> replaced by
> /// V2.
> void MetadataContext::ValueIsRAUWd(Value *V1, Value *V2) {
> ? Instruction *I1 = dyn_cast<Instruction>(V1);
> ? Instruction *I2 = dyn_cast<Instruction>(V2);
> ? if (!I1 || !I2)
>
> This seems very very wrong. ?If I do:
>
> X->RAUW(Y);
>
> Then the properties on X are not necessarily true in Y's scope. ?The
> debug info (for example) should just be discarded.

I am not sure. Even if an instruction is replaced by another
instruction the new instruction still represents original source code
location. Otherwise, IMO, this will put huge burden on optimizer to
keep dbg info available through all RAUW.

>?Value range info
> is another example: X may be defined lower in the dom tree where some
> property holds that doesn't hold on Y.

That is true for this use. That's why the comment
"FIXME: Give custom handlers a chance to override this".

Please let me know if I missed anything or misinterpreted your comments.
Thanks!
-
Devang



From echristo at apple.com  Thu Oct 22 15:04:07 2009
From: echristo at apple.com (Eric Christopher)
Date: Thu, 22 Oct 2009 13:04:07 -0700
Subject: [llvm-commits] [llvm] r84130 - in /llvm/trunk/include/llvm:
	ADT/DenseMapInfo.h
	CodeGen/ScheduleDAG.h	ExecutionEngine/JITMemoryManager.h
	Metadata.h Operator.h	Pass.h Support/CommandLine.h
	Support/raw_ostream.h
In-Reply-To: <5A367888-05B2-4B7E-B558-55C5B6111B1A@apple.com>
References: <200910142028.n9EKSXp1021686@zion.cs.uiuc.edu>
	<A4B4E402-D1E7-4387-A8ED-CF32C2739117@apple.com>
	<5A367888-05B2-4B7E-B558-55C5B6111B1A@apple.com>
Message-ID: <F58CE6FB-436A-4E4F-88D5-0A9E2FCD1224@apple.com>


On Oct 22, 2009, at 11:00 AM, Bill Wendling wrote:

> 
> On Oct 14, 2009, at 2:00 PM, Eric Christopher wrote:
> 
>> 
>> On Oct 14, 2009, at 1:28 PM, Eric Christopher wrote:
>> 
>>> -  virtual raw_ostream &changeColor(enum Colors colors, bool
>>> bold=false,
>>> -                                   bool  bg=false) { return *this; }
>>> +  virtual raw_ostream &changeColor(enum Colors, bool, bool)
>>> { return *this; }
>> 
>> Yep. Clang build failure from this.  Fixing.  That'll teach me to
>> forget to copy clang over when I checkout a fresh llvm.
>> 
> It's allowed that you can do this:
> 
>  virtual raw_ostream &changeColor(enum Colors, bool = false, bool = false)
>   { return *this; }

Yup. Dan mentioned that one to me :)

-eric


From jyasskin at google.com  Thu Oct 22 15:10:21 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Thu, 22 Oct 2009 20:10:21 -0000
Subject: [llvm-commits] [llvm] r84890 - in /llvm/trunk:
 docs/ProgrammersManual.html include/llvm/ADT/ValueMap.h
 include/llvm/Support/type_traits.h unittests/ADT/ValueMapTest.cpp
Message-ID: <200910222010.n9MKALHs012740@zion.cs.uiuc.edu>

Author: jyasskin
Date: Thu Oct 22 15:10:20 2009
New Revision: 84890

URL: http://llvm.org/viewvc/llvm-project?rev=84890&view=rev
Log:
Add a ValueMap<ValueOrSubclass*, T> type.  ValueMap<Value*, T> is safe to use
even when keys get RAUWed and deleted during its lifetime. By default the keys
act like WeakVHs, but users can pass a third template parameter to configure
how updates work and whether to do anything beyond updating the map on each
action.

It's also possible to automatically acquire a lock around ValueMap updates
triggered by RAUWs and deletes, to support the ExecutionEngine.

Added:
    llvm/trunk/include/llvm/ADT/ValueMap.h
    llvm/trunk/unittests/ADT/ValueMapTest.cpp
Modified:
    llvm/trunk/docs/ProgrammersManual.html
    llvm/trunk/include/llvm/Support/type_traits.h

Modified: llvm/trunk/docs/ProgrammersManual.html
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/ProgrammersManual.html?rev=84890&r1=84889&r2=84890&view=diff

==============================================================================
--- llvm/trunk/docs/ProgrammersManual.html (original)
+++ llvm/trunk/docs/ProgrammersManual.html Thu Oct 22 15:10:20 2009
@@ -83,6 +83,7 @@
       <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
       <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
       <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
+      <li><a href="#dss_valuemap">"llvm/ADT/ValueMap.h"</a></li>
       <li><a href="#dss_map">&lt;map&gt;</a></li>
       <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
     </ul></li>
@@ -1492,6 +1493,23 @@
 
 <!-- _______________________________________________________________________ -->
 <div class="doc_subsubsection">
+  <a name="dss_valuemap">"llvm/ADT/ValueMap.h"</a>
+</div>
+
+<div class="doc_text">
+
+<p>
+ValueMap is a wrapper around a <a href="#dss_densemap">DenseMap</a> mapping
+Value*s (or subclasses) to another type.  When a Value is deleted or RAUW'ed,
+ValueMap will update itself so the new version of the key is mapped to the same
+value, just as if the key were a WeakVH.  You can configure exactly how this
+happens, and what else happens on these two events, by passing
+a <code>Config</code> parameter to the ValueMap template.</p>
+
+</div>
+
+<!-- _______________________________________________________________________ -->
+<div class="doc_subsubsection">
   <a name="dss_map">&lt;map&gt;</a>
 </div>
 

Added: llvm/trunk/include/llvm/ADT/ValueMap.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ADT/ValueMap.h?rev=84890&view=auto

==============================================================================
--- llvm/trunk/include/llvm/ADT/ValueMap.h (added)
+++ llvm/trunk/include/llvm/ADT/ValueMap.h Thu Oct 22 15:10:20 2009
@@ -0,0 +1,365 @@
+//===- llvm/ADT/ValueMap.h - Safe map from Values to data -------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the ValueMap class.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_ADT_VALUEMAP_H
+#define LLVM_ADT_VALUEMAP_H
+
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/Support/ValueHandle.h"
+#include "llvm/Support/type_traits.h"
+#include "llvm/System/Mutex.h"
+
+#include <iterator>
+
+namespace llvm {
+
+template<typename KeyT, typename ValueT, typename Config, typename ValueInfoT>
+class ValueMapCallbackVH;
+
+template<typename DenseMapT, typename KeyT>
+class ValueMapIterator;
+template<typename DenseMapT, typename KeyT>
+class ValueMapConstIterator;
+
+template<typename KeyT>
+struct ValueMapConfig {
+  /// If FollowRAUW is true, the ValueMap will update mappings on RAUW. If it's
+  /// false, the ValueMap will leave the original mapping in place.
+  enum { FollowRAUW = true };
+
+  // All methods will be called with a first argument of type ExtraData.  The
+  // default implementations in this class take a templated first argument so
+  // that users' subclasses can use any type they want without having to
+  // override all the defaults.
+  struct ExtraData {};
+
+  template<typename ExtraDataT>
+  static void onRAUW(const ExtraDataT &Data, KeyT Old, KeyT New) {}
+  template<typename ExtraDataT>
+  static void onDeleted(const ExtraDataT &Data, KeyT Old) {}
+
+  /// Returns a mutex that should be acquired around any changes to the map.
+  /// This is only acquired from the CallbackVH (and held around calls to onRAUW
+  /// and onDeleted) and not inside other ValueMap methods.  NULL means that no
+  /// mutex is necessary.
+  template<typename ExtraDataT>
+  static sys::Mutex *getMutex(const ExtraDataT &Data) { return NULL; }
+};
+
+/// ValueMap maps Value* or any subclass to an arbitrary other
+/// type. It provides the DenseMap interface.  When the key values are
+/// deleted or RAUWed, ValueMap relies on the Config to decide what to
+/// do.  Config parameters should inherit from ValueMapConfig<KeyT> to
+/// get default implementations of all the methods ValueMap uses.
+///
+/// By default, when a key is RAUWed from V1 to V2, the old mapping
+/// V1->target is removed, and a new mapping V2->target is added.  If
+/// V2 already existed, its old target is overwritten.  When a key is
+/// deleted, its mapping is removed.  You can override Config to get
+/// called back on each event.
+template<typename KeyT, typename ValueT, typename Config = ValueMapConfig<KeyT>,
+         typename ValueInfoT = DenseMapInfo<ValueT> >
+class ValueMap {
+  friend class ValueMapCallbackVH<KeyT, ValueT, Config, ValueInfoT>;
+  typedef ValueMapCallbackVH<KeyT, ValueT, Config, ValueInfoT> ValueMapCVH;
+  typedef DenseMap<ValueMapCVH, ValueT, DenseMapInfo<ValueMapCVH>,
+                   ValueInfoT> MapT;
+  typedef typename Config::ExtraData ExtraData;
+  MapT Map;
+  ExtraData Data;
+public:
+  typedef KeyT key_type;
+  typedef ValueT mapped_type;
+  typedef std::pair<KeyT, ValueT> value_type;
+
+  ValueMap(const ValueMap& Other) : Map(Other.Map), Data(Other.Data) {}
+
+  explicit ValueMap(unsigned NumInitBuckets = 64)
+    : Map(NumInitBuckets), Data() {}
+  explicit ValueMap(const ExtraData &Data, unsigned NumInitBuckets = 64)
+    : Map(NumInitBuckets), Data(Data) {}
+
+  ~ValueMap() {}
+
+  typedef ValueMapIterator<MapT, KeyT> iterator;
+  typedef ValueMapConstIterator<MapT, KeyT> const_iterator;
+  inline iterator begin() { return iterator(Map.begin()); }
+  inline iterator end() { return iterator(Map.end()); }
+  inline const_iterator begin() const { return const_iterator(Map.begin()); }
+  inline const_iterator end() const { return const_iterator(Map.end()); }
+
+  bool empty() const { return Map.empty(); }
+  unsigned size() const { return Map.size(); }
+
+  /// Grow the map so that it has at least Size buckets. Does not shrink
+  void resize(size_t Size) { Map.resize(Size); }
+
+  void clear() { Map.clear(); }
+
+  /// count - Return true if the specified key is in the map.
+  bool count(const KeyT &Val) const {
+    return Map.count(Wrap(Val));
+  }
+
+  iterator find(const KeyT &Val) {
+    return iterator(Map.find(Wrap(Val)));
+  }
+  const_iterator find(const KeyT &Val) const {
+    return const_iterator(Map.find(Wrap(Val)));
+  }
+
+  /// lookup - Return the entry for the specified key, or a default
+  /// constructed value if no such entry exists.
+  ValueT lookup(const KeyT &Val) const {
+    return Map.lookup(Wrap(Val));
+  }
+
+  // Inserts key,value pair into the map if the key isn't already in the map.
+  // If the key is already in the map, it returns false and doesn't update the
+  // value.
+  std::pair<iterator, bool> insert(const std::pair<KeyT, ValueT> &KV) {
+    std::pair<typename MapT::iterator, bool> map_result=
+      Map.insert(std::make_pair(Wrap(KV.first), KV.second));
+    return std::make_pair(iterator(map_result.first), map_result.second);
+  }
+
+  /// insert - Range insertion of pairs.
+  template<typename InputIt>
+  void insert(InputIt I, InputIt E) {
+    for (; I != E; ++I)
+      insert(*I);
+  }
+
+
+  bool erase(const KeyT &Val) {
+    return Map.erase(Wrap(Val));
+  }
+  bool erase(iterator I) {
+    return Map.erase(I.base());
+  }
+
+  value_type& FindAndConstruct(const KeyT &Key) {
+    return Map.FindAndConstruct(Wrap(Key));
+  }
+
+  ValueT &operator[](const KeyT &Key) {
+    return Map[Wrap(Key)];
+  }
+
+  ValueMap& operator=(const ValueMap& Other) {
+    Map = Other.Map;
+    Data = Other.Data;
+    return *this;
+  }
+
+  /// isPointerIntoBucketsArray - Return true if the specified pointer points
+  /// somewhere into the ValueMap's array of buckets (i.e. either to a key or
+  /// value in the ValueMap).
+  bool isPointerIntoBucketsArray(const void *Ptr) const {
+    return Map.isPointerIntoBucketsArray(Ptr);
+  }
+
+  /// getPointerIntoBucketsArray() - Return an opaque pointer into the buckets
+  /// array.  In conjunction with the previous method, this can be used to
+  /// determine whether an insertion caused the ValueMap to reallocate.
+  const void *getPointerIntoBucketsArray() const {
+    return Map.getPointerIntoBucketsArray();
+  }
+
+private:
+  ValueMapCVH Wrap(KeyT key) const {
+    // The only way the resulting CallbackVH could try to modify *this (making
+    // the const_cast incorrect) is if it gets inserted into the map.  But then
+    // this function must have been called from a non-const method, making the
+    // const_cast ok.
+    return ValueMapCVH(key, const_cast<ValueMap*>(this));
+  }
+};
+
+template<typename KeyT, typename ValueT, typename Config, typename ValueInfoT>
+class ValueMapCallbackVH : public CallbackVH {
+  friend class ValueMap<KeyT, ValueT, Config, ValueInfoT>;
+  friend class DenseMapInfo<ValueMapCallbackVH>;
+  typedef ValueMap<KeyT, ValueT, Config, ValueInfoT> ValueMap;
+  typedef typename llvm::remove_pointer<KeyT>::type KeySansPointerT;
+
+  ValueMap *Map;
+
+  ValueMapCallbackVH(KeyT Key, ValueMap *Map)
+      : CallbackVH(const_cast<Value*>(static_cast<const Value*>(Key))),
+        Map(Map) {}
+
+public:
+  KeyT Unwrap() const { return cast_or_null<KeySansPointerT>(getValPtr()); }
+
+  virtual void deleted() {
+    // Make a copy that won't get changed even when *this is destroyed.
+    ValueMapCallbackVH Copy(*this);
+    sys::Mutex *M = Config::getMutex(Copy.Map->Data);
+    if (M)
+      M->acquire();
+    Config::onDeleted(Copy.Map->Data, Copy.Unwrap());  // May destroy *this.
+    Copy.Map->Map.erase(Copy);  // Definitely destroys *this.
+    if (M)
+      M->release();
+  }
+  virtual void allUsesReplacedWith(Value *new_key) {
+    assert(isa<KeySansPointerT>(new_key) &&
+           "Invalid RAUW on key of ValueMap<>");
+    // Make a copy that won't get changed even when *this is destroyed.
+    ValueMapCallbackVH Copy(*this);
+    sys::Mutex *M = Config::getMutex(Copy.Map->Data);
+    if (M)
+      M->acquire();
+
+    KeyT typed_new_key = cast<KeySansPointerT>(new_key);
+    // Can destroy *this:
+    Config::onRAUW(Copy.Map->Data, Copy.Unwrap(), typed_new_key);
+    if (Config::FollowRAUW) {
+      typename ValueMap::MapT::iterator I = Copy.Map->Map.find(Copy);
+      // I could == Copy.Map->Map.end() if the onRAUW callback already
+      // removed the old mapping.
+      if (I != Copy.Map->Map.end()) {
+        ValueT Target(I->second);
+        Copy.Map->Map.erase(I);  // Definitely destroys *this.
+        Copy.Map->insert(std::make_pair(typed_new_key, Target));
+      }
+    }
+    if (M)
+      M->release();
+  }
+};
+
+template<typename KeyT, typename ValueT, typename Config, typename ValueInfoT>
+struct DenseMapInfo<ValueMapCallbackVH<KeyT, ValueT, Config, ValueInfoT> > {
+  typedef ValueMapCallbackVH<KeyT, ValueT, Config, ValueInfoT> VH;
+  typedef DenseMapInfo<KeyT> PointerInfo;
+
+  static inline VH getEmptyKey() {
+    return VH(PointerInfo::getEmptyKey(), NULL);
+  }
+  static inline VH getTombstoneKey() {
+    return VH(PointerInfo::getTombstoneKey(), NULL);
+  }
+  static unsigned getHashValue(const VH &Val) {
+    return PointerInfo::getHashValue(Val.Unwrap());
+  }
+  static bool isEqual(const VH &LHS, const VH &RHS) {
+    return LHS == RHS;
+  }
+  static bool isPod() { return false; }
+};
+
+
+template<typename DenseMapT, typename KeyT>
+class ValueMapIterator :
+    public std::iterator<std::forward_iterator_tag,
+                         std::pair<KeyT, typename DenseMapT::mapped_type>,
+                         ptrdiff_t> {
+  typedef typename DenseMapT::iterator BaseT;
+  typedef typename DenseMapT::mapped_type ValueT;
+  BaseT I;
+public:
+  ValueMapIterator() : I() {}
+
+  ValueMapIterator(BaseT I) : I(I) {}
+
+  BaseT base() const { return I; }
+
+  struct ValueTypeProxy {
+    const KeyT first;
+    ValueT& second;
+    ValueTypeProxy *operator->()  { return this; }
+    operator std::pair<KeyT, ValueT>() const {
+      return std::make_pair(first, second);
+    }
+  };
+
+  ValueTypeProxy operator*() const {
+    ValueTypeProxy Result = {I->first.Unwrap(), I->second};
+    return Result;
+  }
+
+  ValueTypeProxy operator->() const {
+    return operator*();
+  }
+
+  bool operator==(const ValueMapIterator &RHS) const {
+    return I == RHS.I;
+  }
+  bool operator!=(const ValueMapIterator &RHS) const {
+    return I != RHS.I;
+  }
+
+  inline ValueMapIterator& operator++() {          // Preincrement
+    ++I;
+    return *this;
+  }
+  ValueMapIterator operator++(int) {        // Postincrement
+    ValueMapIterator tmp = *this; ++*this; return tmp;
+  }
+};
+
+template<typename DenseMapT, typename KeyT>
+class ValueMapConstIterator :
+    public std::iterator<std::forward_iterator_tag,
+                         std::pair<KeyT, typename DenseMapT::mapped_type>,
+                         ptrdiff_t> {
+  typedef typename DenseMapT::const_iterator BaseT;
+  typedef typename DenseMapT::mapped_type ValueT;
+  BaseT I;
+public:
+  ValueMapConstIterator() : I() {}
+  ValueMapConstIterator(BaseT I) : I(I) {}
+  ValueMapConstIterator(ValueMapIterator<DenseMapT, KeyT> Other)
+    : I(Other.base()) {}
+
+  BaseT base() const { return I; }
+
+  struct ValueTypeProxy {
+    const KeyT first;
+    const ValueT& second;
+    ValueTypeProxy *operator->()  { return this; }
+    operator std::pair<KeyT, ValueT>() const {
+      return std::make_pair(first, second);
+    }
+  };
+
+  ValueTypeProxy operator*() const {
+    ValueTypeProxy Result = {I->first.Unwrap(), I->second};
+    return Result;
+  }
+
+  ValueTypeProxy operator->() const {
+    return operator*();
+  }
+
+  bool operator==(const ValueMapConstIterator &RHS) const {
+    return I == RHS.I;
+  }
+  bool operator!=(const ValueMapConstIterator &RHS) const {
+    return I != RHS.I;
+  }
+
+  inline ValueMapConstIterator& operator++() {          // Preincrement
+    ++I;
+    return *this;
+  }
+  ValueMapConstIterator operator++(int) {        // Postincrement
+    ValueMapConstIterator tmp = *this; ++*this; return tmp;
+  }
+};
+
+} // end namespace llvm
+
+#endif

Modified: llvm/trunk/include/llvm/Support/type_traits.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/type_traits.h?rev=84890&r1=84889&r2=84890&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Support/type_traits.h (original)
+++ llvm/trunk/include/llvm/Support/type_traits.h Thu Oct 22 15:10:20 2009
@@ -87,6 +87,15 @@
       sizeof(char) == sizeof(dont_use::base_of_helper<Base>((Derived*)0));
 };
 
+// remove_pointer - Metafunction to turn Foo* into Foo.  Defined in
+// C++0x [meta.trans.ptr].
+template <typename T> struct remove_pointer { typedef T type; };
+template <typename T> struct remove_pointer<T*> { typedef T type; };
+template <typename T> struct remove_pointer<T*const> { typedef T type; };
+template <typename T> struct remove_pointer<T*volatile> { typedef T type; };
+template <typename T> struct remove_pointer<T*const volatile> {
+    typedef T type; };
+
 }
 
 #endif

Added: llvm/trunk/unittests/ADT/ValueMapTest.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ADT/ValueMapTest.cpp?rev=84890&view=auto

==============================================================================
--- llvm/trunk/unittests/ADT/ValueMapTest.cpp (added)
+++ llvm/trunk/unittests/ADT/ValueMapTest.cpp Thu Oct 22 15:10:20 2009
@@ -0,0 +1,291 @@
+//===- llvm/unittest/ADT/ValueMapTest.cpp - ValueMap unit tests -*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ADT/ValueMap.h"
+
+#include "llvm/Instructions.h"
+#include "llvm/ADT/OwningPtr.h"
+
+#include "gtest/gtest.h"
+
+using namespace llvm;
+
+namespace {
+
+// Test fixture
+template<typename T>
+class ValueMapTest : public testing::Test {
+protected:
+  Constant *ConstantV;
+  OwningPtr<BitCastInst> BitcastV;
+  OwningPtr<BinaryOperator> AddV;
+
+  ValueMapTest() :
+    ConstantV(ConstantInt::get(Type::getInt32Ty(getGlobalContext()), 0)),
+    BitcastV(new BitCastInst(ConstantV, Type::getInt32Ty(getGlobalContext()))),
+    AddV(BinaryOperator::CreateAdd(ConstantV, ConstantV)) {
+  }
+};
+
+// Run everything on Value*, a subtype to make sure that casting works as
+// expected, and a const subtype to make sure we cast const correctly.
+typedef ::testing::Types<Value, Instruction, const Instruction> KeyTypes;
+TYPED_TEST_CASE(ValueMapTest, KeyTypes);
+
+TYPED_TEST(ValueMapTest, Null) {
+  ValueMap<TypeParam*, int> VM1;
+  VM1[NULL] = 7;
+  EXPECT_EQ(7, VM1.lookup(NULL));
+}
+
+TYPED_TEST(ValueMapTest, FollowsValue) {
+  ValueMap<TypeParam*, int> VM;
+  VM[this->BitcastV.get()] = 7;
+  EXPECT_EQ(7, VM.lookup(this->BitcastV.get()));
+  EXPECT_EQ(0, VM.count(this->AddV.get()));
+  this->BitcastV->replaceAllUsesWith(this->AddV.get());
+  EXPECT_EQ(7, VM.lookup(this->AddV.get()));
+  EXPECT_EQ(0, VM.count(this->BitcastV.get()));
+  this->AddV.reset();
+  EXPECT_EQ(0, VM.count(this->AddV.get()));
+  EXPECT_EQ(0, VM.count(this->BitcastV.get()));
+  EXPECT_EQ(0U, VM.size());
+}
+
+TYPED_TEST(ValueMapTest, OperationsWork) {
+  ValueMap<TypeParam*, int> VM;
+  ValueMap<TypeParam*, int> VM2(16);
+  typename ValueMapConfig<TypeParam*>::ExtraData Data;
+  ValueMap<TypeParam*, int> VM3(Data, 16);
+  EXPECT_TRUE(VM.empty());
+
+  VM[this->BitcastV.get()] = 7;
+
+  // Find:
+  typename ValueMap<TypeParam*, int>::iterator I =
+    VM.find(this->BitcastV.get());
+  ASSERT_TRUE(I != VM.end());
+  EXPECT_EQ(this->BitcastV.get(), I->first);
+  EXPECT_EQ(7, I->second);
+  EXPECT_TRUE(VM.find(this->AddV.get()) == VM.end());
+
+  // Const find:
+  const ValueMap<TypeParam*, int> &CVM = VM;
+  typename ValueMap<TypeParam*, int>::const_iterator CI =
+    CVM.find(this->BitcastV.get());
+  ASSERT_TRUE(CI != CVM.end());
+  EXPECT_EQ(this->BitcastV.get(), CI->first);
+  EXPECT_EQ(7, CI->second);
+  EXPECT_TRUE(CVM.find(this->AddV.get()) == CVM.end());
+
+  // Insert:
+  std::pair<typename ValueMap<TypeParam*, int>::iterator, bool> InsertResult1 =
+    VM.insert(std::make_pair(this->AddV.get(), 3));
+  EXPECT_EQ(this->AddV.get(), InsertResult1.first->first);
+  EXPECT_EQ(3, InsertResult1.first->second);
+  EXPECT_TRUE(InsertResult1.second);
+  EXPECT_EQ(true, VM.count(this->AddV.get()));
+  std::pair<typename ValueMap<TypeParam*, int>::iterator, bool> InsertResult2 =
+    VM.insert(std::make_pair(this->AddV.get(), 5));
+  EXPECT_EQ(this->AddV.get(), InsertResult2.first->first);
+  EXPECT_EQ(3, InsertResult2.first->second);
+  EXPECT_FALSE(InsertResult2.second);
+
+  // Erase:
+  VM.erase(InsertResult2.first);
+  EXPECT_EQ(false, VM.count(this->AddV.get()));
+  EXPECT_EQ(true, VM.count(this->BitcastV.get()));
+  VM.erase(this->BitcastV.get());
+  EXPECT_EQ(false, VM.count(this->BitcastV.get()));
+  EXPECT_EQ(0U, VM.size());
+
+  // Range insert:
+  SmallVector<std::pair<Instruction*, int>, 2> Elems;
+  Elems.push_back(std::make_pair(this->AddV.get(), 1));
+  Elems.push_back(std::make_pair(this->BitcastV.get(), 2));
+  VM.insert(Elems.begin(), Elems.end());
+  EXPECT_EQ(1, VM.lookup(this->AddV.get()));
+  EXPECT_EQ(2, VM.lookup(this->BitcastV.get()));
+}
+
+template<typename ExpectedType, typename VarType>
+void CompileAssertHasType(VarType) {
+  typedef char assert[is_same<ExpectedType, VarType>::value ? 1 : -1];
+}
+
+TYPED_TEST(ValueMapTest, Iteration) {
+  ValueMap<TypeParam*, int> VM;
+  VM[this->BitcastV.get()] = 2;
+  VM[this->AddV.get()] = 3;
+  size_t size = 0;
+  for (typename ValueMap<TypeParam*, int>::iterator I = VM.begin(), E = VM.end();
+       I != E; ++I) {
+    ++size;
+    std::pair<TypeParam*, int> value = *I;
+    CompileAssertHasType<TypeParam*>(I->first);
+    if (I->second == 2) {
+      EXPECT_EQ(this->BitcastV.get(), I->first);
+      I->second = 5;
+    } else if (I->second == 3) {
+      EXPECT_EQ(this->AddV.get(), I->first);
+      I->second = 6;
+    } else {
+      ADD_FAILURE() << "Iterated through an extra value.";
+    }
+  }
+  EXPECT_EQ(2U, size);
+  EXPECT_EQ(5, VM[this->BitcastV.get()]);
+  EXPECT_EQ(6, VM[this->AddV.get()]);
+
+  size = 0;
+  // Cast to const ValueMap to avoid a bug in DenseMap's iterators.
+  const ValueMap<TypeParam*, int>& CVM = VM;
+  for (typename ValueMap<TypeParam*, int>::const_iterator I = CVM.begin(),
+         E = CVM.end(); I != E; ++I) {
+    ++size;
+    std::pair<TypeParam*, int> value = *I;
+    CompileAssertHasType<TypeParam*>(I->first);
+    if (I->second == 5) {
+      EXPECT_EQ(this->BitcastV.get(), I->first);
+    } else if (I->second == 6) {
+      EXPECT_EQ(this->AddV.get(), I->first);
+    } else {
+      ADD_FAILURE() << "Iterated through an extra value.";
+    }
+  }
+  EXPECT_EQ(2U, size);
+}
+
+TYPED_TEST(ValueMapTest, DefaultCollisionBehavior) {
+  // By default, we overwrite the old value with the replaced value.
+  ValueMap<TypeParam*, int> VM;
+  VM[this->BitcastV.get()] = 7;
+  VM[this->AddV.get()] = 9;
+  this->BitcastV->replaceAllUsesWith(this->AddV.get());
+  EXPECT_EQ(0, VM.count(this->BitcastV.get()));
+  EXPECT_EQ(9, VM.lookup(this->AddV.get()));
+}
+
+TYPED_TEST(ValueMapTest, ConfiguredCollisionBehavior) {
+  // TODO: Implement this when someone needs it.
+}
+
+template<typename KeyT>
+struct LockMutex : ValueMapConfig<KeyT> {
+  struct ExtraData {
+    sys::Mutex *M;
+    bool *CalledRAUW;
+    bool *CalledDeleted;
+  };
+  static void onRAUW(const ExtraData &Data, KeyT Old, KeyT New) {
+    *Data.CalledRAUW = true;
+    EXPECT_FALSE(Data.M->tryacquire()) << "Mutex should already be locked.";
+  }
+  static void onDeleted(const ExtraData &Data, KeyT Old) {
+    *Data.CalledDeleted = true;
+    EXPECT_FALSE(Data.M->tryacquire()) << "Mutex should already be locked.";
+  }
+  static sys::Mutex *getMutex(const ExtraData &Data) { return Data.M; }
+};
+TYPED_TEST(ValueMapTest, LocksMutex) {
+  sys::Mutex M(false);  // Not recursive.
+  bool CalledRAUW = false, CalledDeleted = false;
+  typename LockMutex<TypeParam*>::ExtraData Data =
+    {&M, &CalledRAUW, &CalledDeleted};
+  ValueMap<TypeParam*, int, LockMutex<TypeParam*> > VM(Data);
+  VM[this->BitcastV.get()] = 7;
+  this->BitcastV->replaceAllUsesWith(this->AddV.get());
+  this->AddV.reset();
+  EXPECT_TRUE(CalledRAUW);
+  EXPECT_TRUE(CalledDeleted);
+}
+
+template<typename KeyT>
+struct NoFollow : ValueMapConfig<KeyT> {
+  enum { FollowRAUW = false };
+};
+
+TYPED_TEST(ValueMapTest, NoFollowRAUW) {
+  ValueMap<TypeParam*, int, NoFollow<TypeParam*> > VM;
+  VM[this->BitcastV.get()] = 7;
+  EXPECT_EQ(7, VM.lookup(this->BitcastV.get()));
+  EXPECT_EQ(0, VM.count(this->AddV.get()));
+  this->BitcastV->replaceAllUsesWith(this->AddV.get());
+  EXPECT_EQ(7, VM.lookup(this->BitcastV.get()));
+  EXPECT_EQ(0, VM.lookup(this->AddV.get()));
+  this->AddV.reset();
+  EXPECT_EQ(7, VM.lookup(this->BitcastV.get()));
+  EXPECT_EQ(0, VM.lookup(this->AddV.get()));
+  this->BitcastV.reset();
+  EXPECT_EQ(0, VM.lookup(this->BitcastV.get()));
+  EXPECT_EQ(0, VM.lookup(this->AddV.get()));
+  EXPECT_EQ(0U, VM.size());
+}
+
+template<typename KeyT>
+struct CountOps : ValueMapConfig<KeyT> {
+  struct ExtraData {
+    int *Deletions;
+    int *RAUWs;
+  };
+
+  static void onRAUW(const ExtraData &Data, KeyT Old, KeyT New) {
+    ++*Data.RAUWs;
+  }
+  static void onDeleted(const ExtraData &Data, KeyT Old) {
+    ++*Data.Deletions;
+  }
+};
+
+TYPED_TEST(ValueMapTest, CallsConfig) {
+  int Deletions = 0, RAUWs = 0;
+  typename CountOps<TypeParam*>::ExtraData Data = {&Deletions, &RAUWs};
+  ValueMap<TypeParam*, int, CountOps<TypeParam*> > VM(Data);
+  VM[this->BitcastV.get()] = 7;
+  this->BitcastV->replaceAllUsesWith(this->AddV.get());
+  EXPECT_EQ(0, Deletions);
+  EXPECT_EQ(1, RAUWs);
+  this->AddV.reset();
+  EXPECT_EQ(1, Deletions);
+  EXPECT_EQ(1, RAUWs);
+  this->BitcastV.reset();
+  EXPECT_EQ(1, Deletions);
+  EXPECT_EQ(1, RAUWs);
+}
+
+template<typename KeyT>
+struct ModifyingConfig : ValueMapConfig<KeyT> {
+  // We'll put a pointer here back to the ValueMap this key is in, so
+  // that we can modify it (and clobber *this) before the ValueMap
+  // tries to do the same modification.  In previous versions of
+  // ValueMap, that exploded.
+  typedef ValueMap<KeyT, int, ModifyingConfig<KeyT> > **ExtraData;
+
+  static void onRAUW(ExtraData Map, KeyT Old, KeyT New) {
+    (*Map)->erase(Old);
+  }
+  static void onDeleted(ExtraData Map, KeyT Old) {
+    (*Map)->erase(Old);
+  }
+};
+TYPED_TEST(ValueMapTest, SurvivesModificationByConfig) {
+  ValueMap<TypeParam*, int, ModifyingConfig<TypeParam*> > *MapAddress;
+  ValueMap<TypeParam*, int, ModifyingConfig<TypeParam*> > VM(&MapAddress);
+  MapAddress = &VM;
+  // Now the ModifyingConfig can modify the Map inside a callback.
+  VM[this->BitcastV.get()] = 7;
+  this->BitcastV->replaceAllUsesWith(this->AddV.get());
+  EXPECT_FALSE(VM.count(this->BitcastV.get()));
+  EXPECT_FALSE(VM.count(this->AddV.get()));
+  VM[this->AddV.get()] = 7;
+  this->AddV.reset();
+  EXPECT_FALSE(VM.count(this->AddV.get()));
+}
+
+}




From jyasskin at google.com  Thu Oct 22 15:23:43 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Thu, 22 Oct 2009 20:23:43 -0000
Subject: [llvm-commits] [llvm] r84892 - in /llvm/trunk:
 docs/ProgrammersManual.html include/llvm/ADT/ValueMap.h
 include/llvm/Support/type_traits.h unittests/ADT/ValueMapTest.cpp
Message-ID: <200910222023.n9MKNhDt014903@zion.cs.uiuc.edu>

Author: jyasskin
Date: Thu Oct 22 15:23:43 2009
New Revision: 84892

URL: http://llvm.org/viewvc/llvm-project?rev=84892&view=rev
Log:
Revert r84890, which broke the linux build.

Removed:
    llvm/trunk/include/llvm/ADT/ValueMap.h
    llvm/trunk/unittests/ADT/ValueMapTest.cpp
Modified:
    llvm/trunk/docs/ProgrammersManual.html
    llvm/trunk/include/llvm/Support/type_traits.h

Modified: llvm/trunk/docs/ProgrammersManual.html
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/ProgrammersManual.html?rev=84892&r1=84891&r2=84892&view=diff

==============================================================================
--- llvm/trunk/docs/ProgrammersManual.html (original)
+++ llvm/trunk/docs/ProgrammersManual.html Thu Oct 22 15:23:43 2009
@@ -83,7 +83,6 @@
       <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
       <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
       <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
-      <li><a href="#dss_valuemap">"llvm/ADT/ValueMap.h"</a></li>
       <li><a href="#dss_map">&lt;map&gt;</a></li>
       <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
     </ul></li>
@@ -1493,23 +1492,6 @@
 
 <!-- _______________________________________________________________________ -->
 <div class="doc_subsubsection">
-  <a name="dss_valuemap">"llvm/ADT/ValueMap.h"</a>
-</div>
-
-<div class="doc_text">
-
-<p>
-ValueMap is a wrapper around a <a href="#dss_densemap">DenseMap</a> mapping
-Value*s (or subclasses) to another type.  When a Value is deleted or RAUW'ed,
-ValueMap will update itself so the new version of the key is mapped to the same
-value, just as if the key were a WeakVH.  You can configure exactly how this
-happens, and what else happens on these two events, by passing
-a <code>Config</code> parameter to the ValueMap template.</p>
-
-</div>
-
-<!-- _______________________________________________________________________ -->
-<div class="doc_subsubsection">
   <a name="dss_map">&lt;map&gt;</a>
 </div>
 

Removed: llvm/trunk/include/llvm/ADT/ValueMap.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ADT/ValueMap.h?rev=84891&view=auto

==============================================================================
--- llvm/trunk/include/llvm/ADT/ValueMap.h (original)
+++ llvm/trunk/include/llvm/ADT/ValueMap.h (removed)
@@ -1,365 +0,0 @@
-//===- llvm/ADT/ValueMap.h - Safe map from Values to data -------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This file defines the ValueMap class.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef LLVM_ADT_VALUEMAP_H
-#define LLVM_ADT_VALUEMAP_H
-
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/Support/ValueHandle.h"
-#include "llvm/Support/type_traits.h"
-#include "llvm/System/Mutex.h"
-
-#include <iterator>
-
-namespace llvm {
-
-template<typename KeyT, typename ValueT, typename Config, typename ValueInfoT>
-class ValueMapCallbackVH;
-
-template<typename DenseMapT, typename KeyT>
-class ValueMapIterator;
-template<typename DenseMapT, typename KeyT>
-class ValueMapConstIterator;
-
-template<typename KeyT>
-struct ValueMapConfig {
-  /// If FollowRAUW is true, the ValueMap will update mappings on RAUW. If it's
-  /// false, the ValueMap will leave the original mapping in place.
-  enum { FollowRAUW = true };
-
-  // All methods will be called with a first argument of type ExtraData.  The
-  // default implementations in this class take a templated first argument so
-  // that users' subclasses can use any type they want without having to
-  // override all the defaults.
-  struct ExtraData {};
-
-  template<typename ExtraDataT>
-  static void onRAUW(const ExtraDataT &Data, KeyT Old, KeyT New) {}
-  template<typename ExtraDataT>
-  static void onDeleted(const ExtraDataT &Data, KeyT Old) {}
-
-  /// Returns a mutex that should be acquired around any changes to the map.
-  /// This is only acquired from the CallbackVH (and held around calls to onRAUW
-  /// and onDeleted) and not inside other ValueMap methods.  NULL means that no
-  /// mutex is necessary.
-  template<typename ExtraDataT>
-  static sys::Mutex *getMutex(const ExtraDataT &Data) { return NULL; }
-};
-
-/// ValueMap maps Value* or any subclass to an arbitrary other
-/// type. It provides the DenseMap interface.  When the key values are
-/// deleted or RAUWed, ValueMap relies on the Config to decide what to
-/// do.  Config parameters should inherit from ValueMapConfig<KeyT> to
-/// get default implementations of all the methods ValueMap uses.
-///
-/// By default, when a key is RAUWed from V1 to V2, the old mapping
-/// V1->target is removed, and a new mapping V2->target is added.  If
-/// V2 already existed, its old target is overwritten.  When a key is
-/// deleted, its mapping is removed.  You can override Config to get
-/// called back on each event.
-template<typename KeyT, typename ValueT, typename Config = ValueMapConfig<KeyT>,
-         typename ValueInfoT = DenseMapInfo<ValueT> >
-class ValueMap {
-  friend class ValueMapCallbackVH<KeyT, ValueT, Config, ValueInfoT>;
-  typedef ValueMapCallbackVH<KeyT, ValueT, Config, ValueInfoT> ValueMapCVH;
-  typedef DenseMap<ValueMapCVH, ValueT, DenseMapInfo<ValueMapCVH>,
-                   ValueInfoT> MapT;
-  typedef typename Config::ExtraData ExtraData;
-  MapT Map;
-  ExtraData Data;
-public:
-  typedef KeyT key_type;
-  typedef ValueT mapped_type;
-  typedef std::pair<KeyT, ValueT> value_type;
-
-  ValueMap(const ValueMap& Other) : Map(Other.Map), Data(Other.Data) {}
-
-  explicit ValueMap(unsigned NumInitBuckets = 64)
-    : Map(NumInitBuckets), Data() {}
-  explicit ValueMap(const ExtraData &Data, unsigned NumInitBuckets = 64)
-    : Map(NumInitBuckets), Data(Data) {}
-
-  ~ValueMap() {}
-
-  typedef ValueMapIterator<MapT, KeyT> iterator;
-  typedef ValueMapConstIterator<MapT, KeyT> const_iterator;
-  inline iterator begin() { return iterator(Map.begin()); }
-  inline iterator end() { return iterator(Map.end()); }
-  inline const_iterator begin() const { return const_iterator(Map.begin()); }
-  inline const_iterator end() const { return const_iterator(Map.end()); }
-
-  bool empty() const { return Map.empty(); }
-  unsigned size() const { return Map.size(); }
-
-  /// Grow the map so that it has at least Size buckets. Does not shrink
-  void resize(size_t Size) { Map.resize(Size); }
-
-  void clear() { Map.clear(); }
-
-  /// count - Return true if the specified key is in the map.
-  bool count(const KeyT &Val) const {
-    return Map.count(Wrap(Val));
-  }
-
-  iterator find(const KeyT &Val) {
-    return iterator(Map.find(Wrap(Val)));
-  }
-  const_iterator find(const KeyT &Val) const {
-    return const_iterator(Map.find(Wrap(Val)));
-  }
-
-  /// lookup - Return the entry for the specified key, or a default
-  /// constructed value if no such entry exists.
-  ValueT lookup(const KeyT &Val) const {
-    return Map.lookup(Wrap(Val));
-  }
-
-  // Inserts key,value pair into the map if the key isn't already in the map.
-  // If the key is already in the map, it returns false and doesn't update the
-  // value.
-  std::pair<iterator, bool> insert(const std::pair<KeyT, ValueT> &KV) {
-    std::pair<typename MapT::iterator, bool> map_result=
-      Map.insert(std::make_pair(Wrap(KV.first), KV.second));
-    return std::make_pair(iterator(map_result.first), map_result.second);
-  }
-
-  /// insert - Range insertion of pairs.
-  template<typename InputIt>
-  void insert(InputIt I, InputIt E) {
-    for (; I != E; ++I)
-      insert(*I);
-  }
-
-
-  bool erase(const KeyT &Val) {
-    return Map.erase(Wrap(Val));
-  }
-  bool erase(iterator I) {
-    return Map.erase(I.base());
-  }
-
-  value_type& FindAndConstruct(const KeyT &Key) {
-    return Map.FindAndConstruct(Wrap(Key));
-  }
-
-  ValueT &operator[](const KeyT &Key) {
-    return Map[Wrap(Key)];
-  }
-
-  ValueMap& operator=(const ValueMap& Other) {
-    Map = Other.Map;
-    Data = Other.Data;
-    return *this;
-  }
-
-  /// isPointerIntoBucketsArray - Return true if the specified pointer points
-  /// somewhere into the ValueMap's array of buckets (i.e. either to a key or
-  /// value in the ValueMap).
-  bool isPointerIntoBucketsArray(const void *Ptr) const {
-    return Map.isPointerIntoBucketsArray(Ptr);
-  }
-
-  /// getPointerIntoBucketsArray() - Return an opaque pointer into the buckets
-  /// array.  In conjunction with the previous method, this can be used to
-  /// determine whether an insertion caused the ValueMap to reallocate.
-  const void *getPointerIntoBucketsArray() const {
-    return Map.getPointerIntoBucketsArray();
-  }
-
-private:
-  ValueMapCVH Wrap(KeyT key) const {
-    // The only way the resulting CallbackVH could try to modify *this (making
-    // the const_cast incorrect) is if it gets inserted into the map.  But then
-    // this function must have been called from a non-const method, making the
-    // const_cast ok.
-    return ValueMapCVH(key, const_cast<ValueMap*>(this));
-  }
-};
-
-template<typename KeyT, typename ValueT, typename Config, typename ValueInfoT>
-class ValueMapCallbackVH : public CallbackVH {
-  friend class ValueMap<KeyT, ValueT, Config, ValueInfoT>;
-  friend class DenseMapInfo<ValueMapCallbackVH>;
-  typedef ValueMap<KeyT, ValueT, Config, ValueInfoT> ValueMap;
-  typedef typename llvm::remove_pointer<KeyT>::type KeySansPointerT;
-
-  ValueMap *Map;
-
-  ValueMapCallbackVH(KeyT Key, ValueMap *Map)
-      : CallbackVH(const_cast<Value*>(static_cast<const Value*>(Key))),
-        Map(Map) {}
-
-public:
-  KeyT Unwrap() const { return cast_or_null<KeySansPointerT>(getValPtr()); }
-
-  virtual void deleted() {
-    // Make a copy that won't get changed even when *this is destroyed.
-    ValueMapCallbackVH Copy(*this);
-    sys::Mutex *M = Config::getMutex(Copy.Map->Data);
-    if (M)
-      M->acquire();
-    Config::onDeleted(Copy.Map->Data, Copy.Unwrap());  // May destroy *this.
-    Copy.Map->Map.erase(Copy);  // Definitely destroys *this.
-    if (M)
-      M->release();
-  }
-  virtual void allUsesReplacedWith(Value *new_key) {
-    assert(isa<KeySansPointerT>(new_key) &&
-           "Invalid RAUW on key of ValueMap<>");
-    // Make a copy that won't get changed even when *this is destroyed.
-    ValueMapCallbackVH Copy(*this);
-    sys::Mutex *M = Config::getMutex(Copy.Map->Data);
-    if (M)
-      M->acquire();
-
-    KeyT typed_new_key = cast<KeySansPointerT>(new_key);
-    // Can destroy *this:
-    Config::onRAUW(Copy.Map->Data, Copy.Unwrap(), typed_new_key);
-    if (Config::FollowRAUW) {
-      typename ValueMap::MapT::iterator I = Copy.Map->Map.find(Copy);
-      // I could == Copy.Map->Map.end() if the onRAUW callback already
-      // removed the old mapping.
-      if (I != Copy.Map->Map.end()) {
-        ValueT Target(I->second);
-        Copy.Map->Map.erase(I);  // Definitely destroys *this.
-        Copy.Map->insert(std::make_pair(typed_new_key, Target));
-      }
-    }
-    if (M)
-      M->release();
-  }
-};
-
-template<typename KeyT, typename ValueT, typename Config, typename ValueInfoT>
-struct DenseMapInfo<ValueMapCallbackVH<KeyT, ValueT, Config, ValueInfoT> > {
-  typedef ValueMapCallbackVH<KeyT, ValueT, Config, ValueInfoT> VH;
-  typedef DenseMapInfo<KeyT> PointerInfo;
-
-  static inline VH getEmptyKey() {
-    return VH(PointerInfo::getEmptyKey(), NULL);
-  }
-  static inline VH getTombstoneKey() {
-    return VH(PointerInfo::getTombstoneKey(), NULL);
-  }
-  static unsigned getHashValue(const VH &Val) {
-    return PointerInfo::getHashValue(Val.Unwrap());
-  }
-  static bool isEqual(const VH &LHS, const VH &RHS) {
-    return LHS == RHS;
-  }
-  static bool isPod() { return false; }
-};
-
-
-template<typename DenseMapT, typename KeyT>
-class ValueMapIterator :
-    public std::iterator<std::forward_iterator_tag,
-                         std::pair<KeyT, typename DenseMapT::mapped_type>,
-                         ptrdiff_t> {
-  typedef typename DenseMapT::iterator BaseT;
-  typedef typename DenseMapT::mapped_type ValueT;
-  BaseT I;
-public:
-  ValueMapIterator() : I() {}
-
-  ValueMapIterator(BaseT I) : I(I) {}
-
-  BaseT base() const { return I; }
-
-  struct ValueTypeProxy {
-    const KeyT first;
-    ValueT& second;
-    ValueTypeProxy *operator->()  { return this; }
-    operator std::pair<KeyT, ValueT>() const {
-      return std::make_pair(first, second);
-    }
-  };
-
-  ValueTypeProxy operator*() const {
-    ValueTypeProxy Result = {I->first.Unwrap(), I->second};
-    return Result;
-  }
-
-  ValueTypeProxy operator->() const {
-    return operator*();
-  }
-
-  bool operator==(const ValueMapIterator &RHS) const {
-    return I == RHS.I;
-  }
-  bool operator!=(const ValueMapIterator &RHS) const {
-    return I != RHS.I;
-  }
-
-  inline ValueMapIterator& operator++() {          // Preincrement
-    ++I;
-    return *this;
-  }
-  ValueMapIterator operator++(int) {        // Postincrement
-    ValueMapIterator tmp = *this; ++*this; return tmp;
-  }
-};
-
-template<typename DenseMapT, typename KeyT>
-class ValueMapConstIterator :
-    public std::iterator<std::forward_iterator_tag,
-                         std::pair<KeyT, typename DenseMapT::mapped_type>,
-                         ptrdiff_t> {
-  typedef typename DenseMapT::const_iterator BaseT;
-  typedef typename DenseMapT::mapped_type ValueT;
-  BaseT I;
-public:
-  ValueMapConstIterator() : I() {}
-  ValueMapConstIterator(BaseT I) : I(I) {}
-  ValueMapConstIterator(ValueMapIterator<DenseMapT, KeyT> Other)
-    : I(Other.base()) {}
-
-  BaseT base() const { return I; }
-
-  struct ValueTypeProxy {
-    const KeyT first;
-    const ValueT& second;
-    ValueTypeProxy *operator->()  { return this; }
-    operator std::pair<KeyT, ValueT>() const {
-      return std::make_pair(first, second);
-    }
-  };
-
-  ValueTypeProxy operator*() const {
-    ValueTypeProxy Result = {I->first.Unwrap(), I->second};
-    return Result;
-  }
-
-  ValueTypeProxy operator->() const {
-    return operator*();
-  }
-
-  bool operator==(const ValueMapConstIterator &RHS) const {
-    return I == RHS.I;
-  }
-  bool operator!=(const ValueMapConstIterator &RHS) const {
-    return I != RHS.I;
-  }
-
-  inline ValueMapConstIterator& operator++() {          // Preincrement
-    ++I;
-    return *this;
-  }
-  ValueMapConstIterator operator++(int) {        // Postincrement
-    ValueMapConstIterator tmp = *this; ++*this; return tmp;
-  }
-};
-
-} // end namespace llvm
-
-#endif

Modified: llvm/trunk/include/llvm/Support/type_traits.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/type_traits.h?rev=84892&r1=84891&r2=84892&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Support/type_traits.h (original)
+++ llvm/trunk/include/llvm/Support/type_traits.h Thu Oct 22 15:23:43 2009
@@ -87,15 +87,6 @@
       sizeof(char) == sizeof(dont_use::base_of_helper<Base>((Derived*)0));
 };
 
-// remove_pointer - Metafunction to turn Foo* into Foo.  Defined in
-// C++0x [meta.trans.ptr].
-template <typename T> struct remove_pointer { typedef T type; };
-template <typename T> struct remove_pointer<T*> { typedef T type; };
-template <typename T> struct remove_pointer<T*const> { typedef T type; };
-template <typename T> struct remove_pointer<T*volatile> { typedef T type; };
-template <typename T> struct remove_pointer<T*const volatile> {
-    typedef T type; };
-
 }
 
 #endif

Removed: llvm/trunk/unittests/ADT/ValueMapTest.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/ADT/ValueMapTest.cpp?rev=84891&view=auto

==============================================================================
--- llvm/trunk/unittests/ADT/ValueMapTest.cpp (original)
+++ llvm/trunk/unittests/ADT/ValueMapTest.cpp (removed)
@@ -1,291 +0,0 @@
-//===- llvm/unittest/ADT/ValueMapTest.cpp - ValueMap unit tests -*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#include "llvm/ADT/ValueMap.h"
-
-#include "llvm/Instructions.h"
-#include "llvm/ADT/OwningPtr.h"
-
-#include "gtest/gtest.h"
-
-using namespace llvm;
-
-namespace {
-
-// Test fixture
-template<typename T>
-class ValueMapTest : public testing::Test {
-protected:
-  Constant *ConstantV;
-  OwningPtr<BitCastInst> BitcastV;
-  OwningPtr<BinaryOperator> AddV;
-
-  ValueMapTest() :
-    ConstantV(ConstantInt::get(Type::getInt32Ty(getGlobalContext()), 0)),
-    BitcastV(new BitCastInst(ConstantV, Type::getInt32Ty(getGlobalContext()))),
-    AddV(BinaryOperator::CreateAdd(ConstantV, ConstantV)) {
-  }
-};
-
-// Run everything on Value*, a subtype to make sure that casting works as
-// expected, and a const subtype to make sure we cast const correctly.
-typedef ::testing::Types<Value, Instruction, const Instruction> KeyTypes;
-TYPED_TEST_CASE(ValueMapTest, KeyTypes);
-
-TYPED_TEST(ValueMapTest, Null) {
-  ValueMap<TypeParam*, int> VM1;
-  VM1[NULL] = 7;
-  EXPECT_EQ(7, VM1.lookup(NULL));
-}
-
-TYPED_TEST(ValueMapTest, FollowsValue) {
-  ValueMap<TypeParam*, int> VM;
-  VM[this->BitcastV.get()] = 7;
-  EXPECT_EQ(7, VM.lookup(this->BitcastV.get()));
-  EXPECT_EQ(0, VM.count(this->AddV.get()));
-  this->BitcastV->replaceAllUsesWith(this->AddV.get());
-  EXPECT_EQ(7, VM.lookup(this->AddV.get()));
-  EXPECT_EQ(0, VM.count(this->BitcastV.get()));
-  this->AddV.reset();
-  EXPECT_EQ(0, VM.count(this->AddV.get()));
-  EXPECT_EQ(0, VM.count(this->BitcastV.get()));
-  EXPECT_EQ(0U, VM.size());
-}
-
-TYPED_TEST(ValueMapTest, OperationsWork) {
-  ValueMap<TypeParam*, int> VM;
-  ValueMap<TypeParam*, int> VM2(16);
-  typename ValueMapConfig<TypeParam*>::ExtraData Data;
-  ValueMap<TypeParam*, int> VM3(Data, 16);
-  EXPECT_TRUE(VM.empty());
-
-  VM[this->BitcastV.get()] = 7;
-
-  // Find:
-  typename ValueMap<TypeParam*, int>::iterator I =
-    VM.find(this->BitcastV.get());
-  ASSERT_TRUE(I != VM.end());
-  EXPECT_EQ(this->BitcastV.get(), I->first);
-  EXPECT_EQ(7, I->second);
-  EXPECT_TRUE(VM.find(this->AddV.get()) == VM.end());
-
-  // Const find:
-  const ValueMap<TypeParam*, int> &CVM = VM;
-  typename ValueMap<TypeParam*, int>::const_iterator CI =
-    CVM.find(this->BitcastV.get());
-  ASSERT_TRUE(CI != CVM.end());
-  EXPECT_EQ(this->BitcastV.get(), CI->first);
-  EXPECT_EQ(7, CI->second);
-  EXPECT_TRUE(CVM.find(this->AddV.get()) == CVM.end());
-
-  // Insert:
-  std::pair<typename ValueMap<TypeParam*, int>::iterator, bool> InsertResult1 =
-    VM.insert(std::make_pair(this->AddV.get(), 3));
-  EXPECT_EQ(this->AddV.get(), InsertResult1.first->first);
-  EXPECT_EQ(3, InsertResult1.first->second);
-  EXPECT_TRUE(InsertResult1.second);
-  EXPECT_EQ(true, VM.count(this->AddV.get()));
-  std::pair<typename ValueMap<TypeParam*, int>::iterator, bool> InsertResult2 =
-    VM.insert(std::make_pair(this->AddV.get(), 5));
-  EXPECT_EQ(this->AddV.get(), InsertResult2.first->first);
-  EXPECT_EQ(3, InsertResult2.first->second);
-  EXPECT_FALSE(InsertResult2.second);
-
-  // Erase:
-  VM.erase(InsertResult2.first);
-  EXPECT_EQ(false, VM.count(this->AddV.get()));
-  EXPECT_EQ(true, VM.count(this->BitcastV.get()));
-  VM.erase(this->BitcastV.get());
-  EXPECT_EQ(false, VM.count(this->BitcastV.get()));
-  EXPECT_EQ(0U, VM.size());
-
-  // Range insert:
-  SmallVector<std::pair<Instruction*, int>, 2> Elems;
-  Elems.push_back(std::make_pair(this->AddV.get(), 1));
-  Elems.push_back(std::make_pair(this->BitcastV.get(), 2));
-  VM.insert(Elems.begin(), Elems.end());
-  EXPECT_EQ(1, VM.lookup(this->AddV.get()));
-  EXPECT_EQ(2, VM.lookup(this->BitcastV.get()));
-}
-
-template<typename ExpectedType, typename VarType>
-void CompileAssertHasType(VarType) {
-  typedef char assert[is_same<ExpectedType, VarType>::value ? 1 : -1];
-}
-
-TYPED_TEST(ValueMapTest, Iteration) {
-  ValueMap<TypeParam*, int> VM;
-  VM[this->BitcastV.get()] = 2;
-  VM[this->AddV.get()] = 3;
-  size_t size = 0;
-  for (typename ValueMap<TypeParam*, int>::iterator I = VM.begin(), E = VM.end();
-       I != E; ++I) {
-    ++size;
-    std::pair<TypeParam*, int> value = *I;
-    CompileAssertHasType<TypeParam*>(I->first);
-    if (I->second == 2) {
-      EXPECT_EQ(this->BitcastV.get(), I->first);
-      I->second = 5;
-    } else if (I->second == 3) {
-      EXPECT_EQ(this->AddV.get(), I->first);
-      I->second = 6;
-    } else {
-      ADD_FAILURE() << "Iterated through an extra value.";
-    }
-  }
-  EXPECT_EQ(2U, size);
-  EXPECT_EQ(5, VM[this->BitcastV.get()]);
-  EXPECT_EQ(6, VM[this->AddV.get()]);
-
-  size = 0;
-  // Cast to const ValueMap to avoid a bug in DenseMap's iterators.
-  const ValueMap<TypeParam*, int>& CVM = VM;
-  for (typename ValueMap<TypeParam*, int>::const_iterator I = CVM.begin(),
-         E = CVM.end(); I != E; ++I) {
-    ++size;
-    std::pair<TypeParam*, int> value = *I;
-    CompileAssertHasType<TypeParam*>(I->first);
-    if (I->second == 5) {
-      EXPECT_EQ(this->BitcastV.get(), I->first);
-    } else if (I->second == 6) {
-      EXPECT_EQ(this->AddV.get(), I->first);
-    } else {
-      ADD_FAILURE() << "Iterated through an extra value.";
-    }
-  }
-  EXPECT_EQ(2U, size);
-}
-
-TYPED_TEST(ValueMapTest, DefaultCollisionBehavior) {
-  // By default, we overwrite the old value with the replaced value.
-  ValueMap<TypeParam*, int> VM;
-  VM[this->BitcastV.get()] = 7;
-  VM[this->AddV.get()] = 9;
-  this->BitcastV->replaceAllUsesWith(this->AddV.get());
-  EXPECT_EQ(0, VM.count(this->BitcastV.get()));
-  EXPECT_EQ(9, VM.lookup(this->AddV.get()));
-}
-
-TYPED_TEST(ValueMapTest, ConfiguredCollisionBehavior) {
-  // TODO: Implement this when someone needs it.
-}
-
-template<typename KeyT>
-struct LockMutex : ValueMapConfig<KeyT> {
-  struct ExtraData {
-    sys::Mutex *M;
-    bool *CalledRAUW;
-    bool *CalledDeleted;
-  };
-  static void onRAUW(const ExtraData &Data, KeyT Old, KeyT New) {
-    *Data.CalledRAUW = true;
-    EXPECT_FALSE(Data.M->tryacquire()) << "Mutex should already be locked.";
-  }
-  static void onDeleted(const ExtraData &Data, KeyT Old) {
-    *Data.CalledDeleted = true;
-    EXPECT_FALSE(Data.M->tryacquire()) << "Mutex should already be locked.";
-  }
-  static sys::Mutex *getMutex(const ExtraData &Data) { return Data.M; }
-};
-TYPED_TEST(ValueMapTest, LocksMutex) {
-  sys::Mutex M(false);  // Not recursive.
-  bool CalledRAUW = false, CalledDeleted = false;
-  typename LockMutex<TypeParam*>::ExtraData Data =
-    {&M, &CalledRAUW, &CalledDeleted};
-  ValueMap<TypeParam*, int, LockMutex<TypeParam*> > VM(Data);
-  VM[this->BitcastV.get()] = 7;
-  this->BitcastV->replaceAllUsesWith(this->AddV.get());
-  this->AddV.reset();
-  EXPECT_TRUE(CalledRAUW);
-  EXPECT_TRUE(CalledDeleted);
-}
-
-template<typename KeyT>
-struct NoFollow : ValueMapConfig<KeyT> {
-  enum { FollowRAUW = false };
-};
-
-TYPED_TEST(ValueMapTest, NoFollowRAUW) {
-  ValueMap<TypeParam*, int, NoFollow<TypeParam*> > VM;
-  VM[this->BitcastV.get()] = 7;
-  EXPECT_EQ(7, VM.lookup(this->BitcastV.get()));
-  EXPECT_EQ(0, VM.count(this->AddV.get()));
-  this->BitcastV->replaceAllUsesWith(this->AddV.get());
-  EXPECT_EQ(7, VM.lookup(this->BitcastV.get()));
-  EXPECT_EQ(0, VM.lookup(this->AddV.get()));
-  this->AddV.reset();
-  EXPECT_EQ(7, VM.lookup(this->BitcastV.get()));
-  EXPECT_EQ(0, VM.lookup(this->AddV.get()));
-  this->BitcastV.reset();
-  EXPECT_EQ(0, VM.lookup(this->BitcastV.get()));
-  EXPECT_EQ(0, VM.lookup(this->AddV.get()));
-  EXPECT_EQ(0U, VM.size());
-}
-
-template<typename KeyT>
-struct CountOps : ValueMapConfig<KeyT> {
-  struct ExtraData {
-    int *Deletions;
-    int *RAUWs;
-  };
-
-  static void onRAUW(const ExtraData &Data, KeyT Old, KeyT New) {
-    ++*Data.RAUWs;
-  }
-  static void onDeleted(const ExtraData &Data, KeyT Old) {
-    ++*Data.Deletions;
-  }
-};
-
-TYPED_TEST(ValueMapTest, CallsConfig) {
-  int Deletions = 0, RAUWs = 0;
-  typename CountOps<TypeParam*>::ExtraData Data = {&Deletions, &RAUWs};
-  ValueMap<TypeParam*, int, CountOps<TypeParam*> > VM(Data);
-  VM[this->BitcastV.get()] = 7;
-  this->BitcastV->replaceAllUsesWith(this->AddV.get());
-  EXPECT_EQ(0, Deletions);
-  EXPECT_EQ(1, RAUWs);
-  this->AddV.reset();
-  EXPECT_EQ(1, Deletions);
-  EXPECT_EQ(1, RAUWs);
-  this->BitcastV.reset();
-  EXPECT_EQ(1, Deletions);
-  EXPECT_EQ(1, RAUWs);
-}
-
-template<typename KeyT>
-struct ModifyingConfig : ValueMapConfig<KeyT> {
-  // We'll put a pointer here back to the ValueMap this key is in, so
-  // that we can modify it (and clobber *this) before the ValueMap
-  // tries to do the same modification.  In previous versions of
-  // ValueMap, that exploded.
-  typedef ValueMap<KeyT, int, ModifyingConfig<KeyT> > **ExtraData;
-
-  static void onRAUW(ExtraData Map, KeyT Old, KeyT New) {
-    (*Map)->erase(Old);
-  }
-  static void onDeleted(ExtraData Map, KeyT Old) {
-    (*Map)->erase(Old);
-  }
-};
-TYPED_TEST(ValueMapTest, SurvivesModificationByConfig) {
-  ValueMap<TypeParam*, int, ModifyingConfig<TypeParam*> > *MapAddress;
-  ValueMap<TypeParam*, int, ModifyingConfig<TypeParam*> > VM(&MapAddress);
-  MapAddress = &VM;
-  // Now the ModifyingConfig can modify the Map inside a callback.
-  VM[this->BitcastV.get()] = 7;
-  this->BitcastV->replaceAllUsesWith(this->AddV.get());
-  EXPECT_FALSE(VM.count(this->BitcastV.get()));
-  EXPECT_FALSE(VM.count(this->AddV.get()));
-  VM[this->AddV.get()] = 7;
-  this->AddV.reset();
-  EXPECT_FALSE(VM.count(this->AddV.get()));
-}
-
-}




From dpatel at apple.com  Thu Oct 22 15:47:06 2009
From: dpatel at apple.com (Devang Patel)
Date: Thu, 22 Oct 2009 13:47:06 -0700
Subject: [llvm-commits] PATCH: Additional DebugFactory methods
In-Reply-To: <d1b71f830910182212t4257ab63g5d860b65197b8950@mail.gmail.com>
References: <d1b71f830910052248naeeba24pe4822e09598119c3@mail.gmail.com>
	<7083196B-95FE-4F1D-89AE-2D38B2D0D72F@apple.com>
	<d1b71f830910182212t4257ab63g5d860b65197b8950@mail.gmail.com>
Message-ID: <A9943CCF-A2AF-4370-AB9B-857AB38E4BE2@apple.com>


On Oct 18, 2009, at 10:12 PM, Talin wrote:

> Here's an updated version of the patch which uses StringRef as  
> requested.
>
> Also, I have changed the method names to avoid the ambiguity  
> problems encountered in the last patch. The new methods end with  
> "Ex" which can stand either for "Expression" or "Extension"  
> depending on your preference. :)


Looks good. Please apply.
-
Devang


From isanbard at gmail.com  Thu Oct 22 15:48:59 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Thu, 22 Oct 2009 20:48:59 -0000
Subject: [llvm-commits] [llvm] r84894 -
	/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp
Message-ID: <200910222048.n9MKmxCi019077@zion.cs.uiuc.edu>

Author: void
Date: Thu Oct 22 15:48:59 2009
New Revision: 84894

URL: http://llvm.org/viewvc/llvm-project?rev=84894&view=rev
Log:
Tidying up some code and comments. No functionality change.

Modified:
    llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp

Modified: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp?rev=84894&r1=84893&r2=84894&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp (original)
+++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp Thu Oct 22 15:48:59 2009
@@ -871,28 +871,7 @@
     Asm->EOL("Next action");
   }
 
-  // Emit the Catch Clauses. The code for the catch clauses following the same
-  // try is similar to a switch statement. The catch clause action record
-  // informs the runtime about the type of a catch clause and about the
-  // associated switch value.
-  //
-  //  Action Record Fields:
-  //
-  //   * Filter Value
-  //     Positive value, starting at 1. Index in the types table of the
-  //     __typeinfo for the catch-clause type. 1 is the first word preceding
-  //     TTBase, 2 is the second word, and so on. Used by the runtime to check
-  //     if the thrown exception type matches the catch-clause type. Back-end
-  //     generated switch statements check against this value.
-  //
-  //   * Next
-  //     Signed offset, in bytes from the start of this field, to the next
-  //     chained action record, or zero if none.
-  //
-  // The order of the action records determined by the next field is the order
-  // of the catch clauses as they appear in the source code, and must be kept in
-  // the same order. As a result, changing the order of the catch clause would
-  // change the semantics of the program.
+  // Emit the Catch TypeInfos.
   for (std::vector<GlobalVariable *>::const_reverse_iterator
          I = TypeInfos.rbegin(), E = TypeInfos.rend(); I != E; ++I) {
     const GlobalVariable *GV = *I;
@@ -907,12 +886,15 @@
     Asm->EOL("TypeInfo");
   }
 
-  // Emit the Type Table.
+  // Emit the Exception Specifications.
   for (std::vector<unsigned>::const_iterator
          I = FilterIds.begin(), E = FilterIds.end(); I < E; ++I) {
     unsigned TypeID = *I;
     Asm->EmitULEB128Bytes(TypeID);
-    Asm->EOL("Filter TypeInfo index");
+    if (TypeID != 0)
+      Asm->EOL("Exception specification");
+    else
+      Asm->EOL();
   }
 
   Asm->EmitAlignment(2, 0, 0, false);




From jyasskin at google.com  Thu Oct 22 15:57:35 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Thu, 22 Oct 2009 20:57:35 -0000
Subject: [llvm-commits] [llvm] r84895 -
 /llvm/trunk/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp
Message-ID: <200910222057.n9MKvZIJ020275@zion.cs.uiuc.edu>

Author: jyasskin
Date: Thu Oct 22 15:57:35 2009
New Revision: 84895

URL: http://llvm.org/viewvc/llvm-project?rev=84895&view=rev
Log:
Fix OProfileJITEventListener after r84054 renamed CompileUnit to Scope.

Modified:
    llvm/trunk/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp

Modified: llvm/trunk/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp?rev=84895&r1=84894&r2=84895&view=diff

==============================================================================
--- llvm/trunk/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp (original)
+++ llvm/trunk/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp Thu Oct 22 15:57:35 2009
@@ -69,16 +69,16 @@
 }
 
 class FilenameCache {
-  // Holds the filename of each CompileUnit, so that we can pass the
+  // Holds the filename of each Scope, so that we can pass the
   // pointer into oprofile.  These char*s are freed in the destructor.
   DenseMap<MDNode*, char*> Filenames;
 
  public:
-  const char *getFilename(MDNode *CompileUnit) {
-    char *&Filename = Filenames[CompileUnit];
+  const char *getFilename(MDNode *Scope) {
+    char *&Filename = Filenames[Scope];
     if (Filename == NULL) {
-      DICompileUnit CU(CompileUnit);
-      Filename = strdup(CU.getFilename());
+      DIScope S(Scope);
+      Filename = strdup(S.getFilename());
     }
     return Filename;
   }
@@ -97,7 +97,7 @@
   Result.vma = Address;
   const DebugLocTuple &tuple = MF.getDebugLocTuple(Loc);
   Result.lineno = tuple.Line;
-  Result.filename = Filenames.getFilename(tuple.CompileUnit);
+  Result.filename = Filenames.getFilename(tuple.Scope);
   DEBUG(errs() << "Mapping " << reinterpret_cast<void*>(Result.vma) << " to "
                << Result.filename << ":" << Result.lineno << "\n");
   return Result;




From clattner at apple.com  Thu Oct 22 16:28:14 2009
From: clattner at apple.com (Chris Lattner)
Date: Thu, 22 Oct 2009 14:28:14 -0700
Subject: [llvm-commits] [llvm-gcc-4.2] r84829	-
	/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
In-Reply-To: <4AE0895E.8000900@free.fr>
References: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>
	<4AE0282A.5080300@free.fr>
	<FE66EE67-889E-47F2-ABA9-FC1C429E542A@apple.com>
	<4AE0895E.8000900@free.fr>
Message-ID: <AB20C91E-8E2F-45C8-8A89-1B4E4ED8D165@apple.com>


On Oct 22, 2009, at 9:33 AM, Duncan Sands wrote:

> Hi Chris,
>
>> The two fixes take a different approach (as far as I understand  
>> it).  The major difference is that I want the warning from  
>> instcombine when you do this in C:
>> void foo() {.}
>> ... foo(4);  ...
>
> gcc doesn't warn about this, even gcc-4.4, so why should llvm-gcc?   
> And if you
> really want llvm-gcc to warn about it, why not add the logic for  
> that to the
> C frontend?

This warning is most useful at LTO time.  I agree that the front-end  
should also catch this in reasonable situations within a translation  
unit, even though GCC doesn't.

-Chris



From clattner at apple.com  Thu Oct 22 16:28:55 2009
From: clattner at apple.com (Chris Lattner)
Date: Thu, 22 Oct 2009 14:28:55 -0700
Subject: [llvm-commits] [llvm] r84878 - in /llvm/trunk:
	include/llvm/Metadata.h lib/VMCore/Metadata.cpp
In-Reply-To: <200910221740.n9MHecFP016341@zion.cs.uiuc.edu>
References: <200910221740.n9MHecFP016341@zion.cs.uiuc.edu>
Message-ID: <25868FA3-DF9C-4F03-8682-9CD71214C58D@apple.com>


On Oct 22, 2009, at 10:40 AM, Devang Patel wrote:

> Author: dpatel
> Date: Thu Oct 22 12:40:37 2009
> New Revision: 84878
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84878&view=rev
> Log:
> Sort handler names to ensure deterministic behavior.

Please don't use std::sort.  It is both much slower and bigger than  
using the simple loop I suggested.

-Chris

>
> Modified:
>    llvm/trunk/include/llvm/Metadata.h
>    llvm/trunk/lib/VMCore/Metadata.cpp
>
> Modified: llvm/trunk/include/llvm/Metadata.h
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84878&r1=84877&r2=84878&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/include/llvm/Metadata.h (original)
> +++ llvm/trunk/include/llvm/Metadata.h Thu Oct 22 12:40:37 2009
> @@ -285,9 +285,9 @@
>   /// the same metadata to In2.
>   void copyMD(Instruction *In1, Instruction *In2);
>
> -  /// getHandlerNames - Get handler names. This is used by bitcode
> -  /// writer.
> -  void getHandlerNames(SmallVectorImpl<std::pair<unsigned,  
> StringRef> >&N) const;
> +  /// getHandlerNames - Populate client supplied smallvector using  
> custome
> +  /// metadata name and ID.
> +  void getHandlerNames(SmallVectorImpl<std::pair<unsigned,  
> StringRef> >&) const;
>
>   /// ValueIsDeleted - This handler is used to update metadata store
>   /// when a value is deleted.
>
> Modified: llvm/trunk/lib/VMCore/Metadata.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Metadata.cpp?rev=84878&r1=84877&r2=84878&view=diff
>
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- llvm/trunk/lib/VMCore/Metadata.cpp (original)
> +++ llvm/trunk/lib/VMCore/Metadata.cpp Thu Oct 22 12:40:37 2009
> @@ -304,13 +304,14 @@
>   return &I->second;
> }
>
> -/// getHandlerNames - Get handler names. This is used by bitcode
> -/// writer and aswm writer.
> +/// getHandlerNames - Populate client supplied smallvector using  
> custome
> +/// metadata name and ID.
> void MetadataContext::
> getHandlerNames(SmallVectorImpl<std::pair<unsigned, StringRef>  
> >&Names) const {
>   for (StringMap<unsigned>::const_iterator I = MDHandlerNames.begin(),
>          E = MDHandlerNames.end(); I != E; ++I)
>     Names.push_back(std::make_pair(I->second, I->first()));
> +  std::sort(Names.begin(), Names.end());
> }
>
> /// ValueIsCloned - This handler is used to update metadata store
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



From benny.kra at googlemail.com  Thu Oct 22 16:49:42 2009
From: benny.kra at googlemail.com (Benjamin Kramer)
Date: Thu, 22 Oct 2009 21:49:42 -0000
Subject: [llvm-commits] [llvm] r84898 - in /llvm/trunk:
 include/llvm/ADT/FoldingSet.h include/llvm/Metadata.h
 lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp
Message-ID: <200910222149.n9MLngqM028060@zion.cs.uiuc.edu>

Author: d0k
Date: Thu Oct 22 16:49:41 2009
New Revision: 84898

URL: http://llvm.org/viewvc/llvm-project?rev=84898&view=rev
Log:
Random include cleanup.

Modified:
    llvm/trunk/include/llvm/ADT/FoldingSet.h
    llvm/trunk/include/llvm/Metadata.h
    llvm/trunk/lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp

Modified: llvm/trunk/include/llvm/ADT/FoldingSet.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ADT/FoldingSet.h?rev=84898&r1=84897&r2=84898&view=diff

==============================================================================
--- llvm/trunk/include/llvm/ADT/FoldingSet.h (original)
+++ llvm/trunk/include/llvm/ADT/FoldingSet.h Thu Oct 22 16:49:41 2009
@@ -19,7 +19,6 @@
 #include "llvm/Support/DataTypes.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringRef.h"
-#include <iterator>
 
 namespace llvm {
   class APFloat;

Modified: llvm/trunk/include/llvm/Metadata.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Metadata.h?rev=84898&r1=84897&r2=84898&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Metadata.h (original)
+++ llvm/trunk/include/llvm/Metadata.h Thu Oct 22 16:49:41 2009
@@ -21,7 +21,6 @@
 #include "llvm/ADT/FoldingSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/ilist_node.h"
-#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/ValueHandle.h"
 
 namespace llvm {

Modified: llvm/trunk/lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp?rev=84898&r1=84897&r2=84898&view=diff

==============================================================================
--- llvm/trunk/lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp (original)
+++ llvm/trunk/lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp Thu Oct 22 16:49:41 2009
@@ -22,6 +22,7 @@
 #include "llvm/Target/TargetOptions.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/OwningPtr.h"
+#include "llvm/Support/Compiler.h"
 #include "llvm/Support/MutexGuard.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/System/Mutex.h"




From echristo at apple.com  Thu Oct 22 17:06:50 2009
From: echristo at apple.com (Eric Christopher)
Date: Thu, 22 Oct 2009 22:06:50 -0000
Subject: [llvm-commits] [llvm] r84900 -
	/llvm/trunk/include/llvm/ADT/StringRef.h
Message-ID: <200910222206.n9MM6oVI030595@zion.cs.uiuc.edu>

Author: echristo
Date: Thu Oct 22 17:06:50 2009
New Revision: 84900

URL: http://llvm.org/viewvc/llvm-project?rev=84900&view=rev
Log:
size_t, not unsigned here to silence a warning.

Modified:
    llvm/trunk/include/llvm/ADT/StringRef.h

Modified: llvm/trunk/include/llvm/ADT/StringRef.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ADT/StringRef.h?rev=84900&r1=84899&r2=84900&view=diff

==============================================================================
--- llvm/trunk/include/llvm/ADT/StringRef.h (original)
+++ llvm/trunk/include/llvm/ADT/StringRef.h Thu Oct 22 17:06:50 2009
@@ -29,7 +29,7 @@
     typedef const char *iterator;
     static const size_t npos = ~size_t(0);
     typedef size_t size_type;
-    
+
   private:
     /// The start of the string, in an external buffer.
     const char *Data;
@@ -45,15 +45,15 @@
     /*implicit*/ StringRef() : Data(0), Length(0) {}
 
     /// Construct a string ref from a cstring.
-    /*implicit*/ StringRef(const char *Str) 
+    /*implicit*/ StringRef(const char *Str)
       : Data(Str) { if (Str) Length = ::strlen(Str); else Length = 0; }
- 
+
     /// Construct a string ref from a pointer and length.
-    /*implicit*/ StringRef(const char *data, unsigned length)
+    /*implicit*/ StringRef(const char *data, size_t length)
       : Data(data), Length(length) {}
 
     /// Construct a string ref from an std::string.
-    /*implicit*/ StringRef(const std::string &Str) 
+    /*implicit*/ StringRef(const std::string &Str)
       : Data(Str.c_str()), Length(Str.length()) {}
 
     /// @}
@@ -83,7 +83,7 @@
       assert(!empty());
       return Data[0];
     }
-    
+
     /// back - Get the last character in the string.
     char back() const {
       assert(!empty());
@@ -93,7 +93,7 @@
     /// equals - Check for string equality, this is more efficient than
     /// compare() when the relative ordering of inequal strings isn't needed.
     bool equals(const StringRef &RHS) const {
-      return (Length == RHS.Length && 
+      return (Length == RHS.Length &&
               memcmp(Data, RHS.Data, RHS.Length) == 0);
     }
 
@@ -117,9 +117,9 @@
     /// @name Operator Overloads
     /// @{
 
-    char operator[](size_t Index) const { 
+    char operator[](size_t Index) const {
       assert(Index < Length && "Invalid index!");
-      return Data[Index]; 
+      return Data[Index];
     }
 
     /// @}
@@ -135,7 +135,7 @@
     /// @{
 
     /// startswith - Check if this string starts with the given \arg Prefix.
-    bool startswith(const StringRef &Prefix) const { 
+    bool startswith(const StringRef &Prefix) const {
       return substr(0, Prefix.Length).equals(Prefix);
     }
 
@@ -164,7 +164,7 @@
     /// \return - The index of the first occurence of \arg Str, or npos if not
     /// found.
     size_t find(const StringRef &Str) const;
-    
+
     /// rfind - Search for the last character \arg C in the string.
     ///
     /// \return - The index of the last occurence of \arg C, or npos if not
@@ -179,29 +179,29 @@
       }
       return npos;
     }
-    
+
     /// rfind - Search for the last string \arg Str in the string.
     ///
     /// \return - The index of the last occurence of \arg Str, or npos if not
     /// found.
     size_t rfind(const StringRef &Str) const;
-    
+
     /// find_first_of - Find the first instance of the specified character or
     /// return npos if not in string.  Same as find.
     size_type find_first_of(char C) const { return find(C); }
-    
+
     /// find_first_of - Find the first character from the string 'Chars' in the
     /// current string or return npos if not in string.
     size_type find_first_of(StringRef Chars) const;
-    
+
     /// find_first_not_of - Find the first character in the string that is not
     /// in the string 'Chars' or return npos if all are in string. Same as find.
     size_type find_first_not_of(StringRef Chars) const;
-    
+
     /// @}
     /// @name Helpful Algorithms
     /// @{
-    
+
     /// count - Return the number of occurrences of \arg C in the string.
     size_t count(char C) const {
       size_t Count = 0;
@@ -210,11 +210,11 @@
           ++Count;
       return Count;
     }
-    
+
     /// count - Return the number of non-overlapped occurrences of \arg Str in
     /// the string.
     size_t count(const StringRef &Str) const;
-    
+
     /// getAsInteger - Parse the current string as an integer of the specified
     /// radix.  If Radix is specified as zero, this does radix autosensing using
     /// extended C rules: 0 is octal, 0x is hex, 0b is binary.
@@ -229,7 +229,7 @@
     bool getAsInteger(unsigned Radix, unsigned &Result) const;
 
     // TODO: Provide overloads for int/unsigned that check for overflow.
-    
+
     /// @}
     /// @name Substring Operations
     /// @{
@@ -308,24 +308,24 @@
     return LHS.equals(RHS);
   }
 
-  inline bool operator!=(const StringRef &LHS, const StringRef &RHS) { 
+  inline bool operator!=(const StringRef &LHS, const StringRef &RHS) {
     return !(LHS == RHS);
   }
-  
+
   inline bool operator<(const StringRef &LHS, const StringRef &RHS) {
-    return LHS.compare(RHS) == -1; 
+    return LHS.compare(RHS) == -1;
   }
 
   inline bool operator<=(const StringRef &LHS, const StringRef &RHS) {
-    return LHS.compare(RHS) != 1; 
+    return LHS.compare(RHS) != 1;
   }
 
   inline bool operator>(const StringRef &LHS, const StringRef &RHS) {
-    return LHS.compare(RHS) == 1; 
+    return LHS.compare(RHS) == 1;
   }
 
   inline bool operator>=(const StringRef &LHS, const StringRef &RHS) {
-    return LHS.compare(RHS) != -1; 
+    return LHS.compare(RHS) != -1;
   }
 
   /// @}




From jyasskin at google.com  Thu Oct 22 17:11:22 2009
From: jyasskin at google.com (Jeffrey Yasskin)
Date: Thu, 22 Oct 2009 22:11:22 -0000
Subject: [llvm-commits] [llvm] r84902 - in /llvm/trunk:
 docs/ProgrammersManual.html include/llvm/ADT/ValueMap.h
 include/llvm/Support/type_traits.h unittests/ADT/ValueMapTest.cpp
Message-ID: <200910222211.n9MMBM93031051@zion.cs.uiuc.edu>

Author: jyasskin
Date: Thu Oct 22 17:11:22 2009
New Revision: 84902

URL: http://llvm.org/viewvc/llvm-project?rev=84902&view=rev
Log:
Try r84890 again (adding ValueMap<>), now that I've tested the compile on
gcc-4.4.

Added:
    llvm/trunk/include/llvm/ADT/ValueMap.h
      - copied, changed from r84891, llvm/trunk/include/llvm/ADT/ValueMap.h
    llvm/trunk/unittests/ADT/ValueMapTest.cpp
      - copied unchanged from r84891, llvm/trunk/unittests/ADT/ValueMapTest.cpp
Modified:
    llvm/trunk/docs/ProgrammersManual.html
    llvm/trunk/include/llvm/Support/type_traits.h

Modified: llvm/trunk/docs/ProgrammersManual.html
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/ProgrammersManual.html?rev=84902&r1=84901&r2=84902&view=diff

==============================================================================
--- llvm/trunk/docs/ProgrammersManual.html (original)
+++ llvm/trunk/docs/ProgrammersManual.html Thu Oct 22 17:11:22 2009
@@ -83,6 +83,7 @@
       <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
       <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
       <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
+      <li><a href="#dss_valuemap">"llvm/ADT/ValueMap.h"</a></li>
       <li><a href="#dss_map">&lt;map&gt;</a></li>
       <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
     </ul></li>
@@ -1492,6 +1493,23 @@
 
 <!-- _______________________________________________________________________ -->
 <div class="doc_subsubsection">
+  <a name="dss_valuemap">"llvm/ADT/ValueMap.h"</a>
+</div>
+
+<div class="doc_text">
+
+<p>
+ValueMap is a wrapper around a <a href="#dss_densemap">DenseMap</a> mapping
+Value*s (or subclasses) to another type.  When a Value is deleted or RAUW'ed,
+ValueMap will update itself so the new version of the key is mapped to the same
+value, just as if the key were a WeakVH.  You can configure exactly how this
+happens, and what else happens on these two events, by passing
+a <code>Config</code> parameter to the ValueMap template.</p>
+
+</div>
+
+<!-- _______________________________________________________________________ -->
+<div class="doc_subsubsection">
   <a name="dss_map">&lt;map&gt;</a>
 </div>
 

Copied: llvm/trunk/include/llvm/ADT/ValueMap.h (from r84891, llvm/trunk/include/llvm/ADT/ValueMap.h)
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/ADT/ValueMap.h?p2=llvm/trunk/include/llvm/ADT/ValueMap.h&p1=llvm/trunk/include/llvm/ADT/ValueMap.h&r1=84891&r2=84902&rev=84902&view=diff

==============================================================================
--- llvm/trunk/include/llvm/ADT/ValueMap.h (original)
+++ llvm/trunk/include/llvm/ADT/ValueMap.h Thu Oct 22 17:11:22 2009
@@ -190,12 +190,12 @@
 class ValueMapCallbackVH : public CallbackVH {
   friend class ValueMap<KeyT, ValueT, Config, ValueInfoT>;
   friend class DenseMapInfo<ValueMapCallbackVH>;
-  typedef ValueMap<KeyT, ValueT, Config, ValueInfoT> ValueMap;
+  typedef ValueMap<KeyT, ValueT, Config, ValueInfoT> ValueMapT;
   typedef typename llvm::remove_pointer<KeyT>::type KeySansPointerT;
 
-  ValueMap *Map;
+  ValueMapT *Map;
 
-  ValueMapCallbackVH(KeyT Key, ValueMap *Map)
+  ValueMapCallbackVH(KeyT Key, ValueMapT *Map)
       : CallbackVH(const_cast<Value*>(static_cast<const Value*>(Key))),
         Map(Map) {}
 
@@ -226,7 +226,7 @@
     // Can destroy *this:
     Config::onRAUW(Copy.Map->Data, Copy.Unwrap(), typed_new_key);
     if (Config::FollowRAUW) {
-      typename ValueMap::MapT::iterator I = Copy.Map->Map.find(Copy);
+      typename ValueMapT::MapT::iterator I = Copy.Map->Map.find(Copy);
       // I could == Copy.Map->Map.end() if the onRAUW callback already
       // removed the old mapping.
       if (I != Copy.Map->Map.end()) {

Modified: llvm/trunk/include/llvm/Support/type_traits.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/type_traits.h?rev=84902&r1=84901&r2=84902&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Support/type_traits.h (original)
+++ llvm/trunk/include/llvm/Support/type_traits.h Thu Oct 22 17:11:22 2009
@@ -87,6 +87,15 @@
       sizeof(char) == sizeof(dont_use::base_of_helper<Base>((Derived*)0));
 };
 
+// remove_pointer - Metafunction to turn Foo* into Foo.  Defined in
+// C++0x [meta.trans.ptr].
+template <typename T> struct remove_pointer { typedef T type; };
+template <typename T> struct remove_pointer<T*> { typedef T type; };
+template <typename T> struct remove_pointer<T*const> { typedef T type; };
+template <typename T> struct remove_pointer<T*volatile> { typedef T type; };
+template <typename T> struct remove_pointer<T*const volatile> {
+    typedef T type; };
+
 }
 
 #endif




From kremenek at apple.com  Thu Oct 22 17:16:17 2009
From: kremenek at apple.com (Ted Kremenek)
Date: Thu, 22 Oct 2009 22:16:17 -0000
Subject: [llvm-commits] [llvm] r84903 -
	/llvm/trunk/lib/System/Unix/Program.inc
Message-ID: <200910222216.n9MMGHfQ031553@zion.cs.uiuc.edu>

Author: kremenek
Date: Thu Oct 22 17:16:17 2009
New Revision: 84903

URL: http://llvm.org/viewvc/llvm-project?rev=84903&view=rev
Log:
Use 'waitpid' instead of 'wait'.  Basing Program::Wait() on 'wait()' prevents it being correct within a multithreaded context.

This address: PR 5277 (Program::Wait is unsafe to call from multiple threads).

Note: If waitpid() turns out to be non-portable, we can add more autoconf magic, or look into
another solution.

Modified:
    llvm/trunk/lib/System/Unix/Program.inc

Modified: llvm/trunk/lib/System/Unix/Program.inc
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/System/Unix/Program.inc?rev=84903&r1=84902&r2=84903&view=diff

==============================================================================
--- llvm/trunk/lib/System/Unix/Program.inc (original)
+++ llvm/trunk/lib/System/Unix/Program.inc Thu Oct 22 17:16:17 2009
@@ -244,7 +244,7 @@
   int status;
   uint64_t pid = reinterpret_cast<uint64_t>(Data_);
   pid_t child = static_cast<pid_t>(pid);
-  while (wait(&status) != child)
+  while (waitpid(pid, &status, 0) != child)
     if (secondsToWait && errno == EINTR) {
       // Kill the child.
       kill(child, SIGKILL);




From jyasskin at gmail.com  Thu Oct 22 17:43:55 2009
From: jyasskin at gmail.com (jyasskin at gmail.com)
Date: Thu, 22 Oct 2009 22:43:55 +0000
Subject: [llvm-commits] [PATCH] Allow Module deletion after function
	compiled.
Message-ID: <00163630ffd99145b604768dd492@google.com>

Evan, could you take a look?

New patch at
http://codereview.appspot.com/download/issue133043_3001.diff

http://codereview.appspot.com/133043


From david_goodwin at apple.com  Thu Oct 22 18:19:18 2009
From: david_goodwin at apple.com (David Goodwin)
Date: Thu, 22 Oct 2009 23:19:18 -0000
Subject: [llvm-commits] [llvm] r84911 - in /llvm/trunk:
 include/llvm/Target/TargetSubtarget.h lib/CodeGen/PostRASchedulerList.cpp
 lib/Target/ARM/ARMSubtarget.h lib/Target/X86/X86Subtarget.h
 test/CodeGen/X86/2007-01-08-InstrSched.ll test/CodeGen/X86/sse2.ll
 test/CodeGen/X86/sse3.ll test/CodeGen/X86/vshift-1.ll
 test/CodeGen/X86/vshift-2.ll test/CodeGen/X86/vshift-3.ll
 test/CodeGen/X86/vshift-5.ll
Message-ID: <200910222319.n9MNJIrY005681@zion.cs.uiuc.edu>

Author: david_goodwin
Date: Thu Oct 22 18:19:17 2009
New Revision: 84911

URL: http://llvm.org/viewvc/llvm-project?rev=84911&view=rev
Log:
Allow the target to select the level of anti-dependence breaking that should be performed by the post-RA scheduler. The default is none.

Modified:
    llvm/trunk/include/llvm/Target/TargetSubtarget.h
    llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
    llvm/trunk/lib/Target/ARM/ARMSubtarget.h
    llvm/trunk/lib/Target/X86/X86Subtarget.h
    llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
    llvm/trunk/test/CodeGen/X86/sse2.ll
    llvm/trunk/test/CodeGen/X86/sse3.ll
    llvm/trunk/test/CodeGen/X86/vshift-1.ll
    llvm/trunk/test/CodeGen/X86/vshift-2.ll
    llvm/trunk/test/CodeGen/X86/vshift-3.ll
    llvm/trunk/test/CodeGen/X86/vshift-5.ll

Modified: llvm/trunk/include/llvm/Target/TargetSubtarget.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Target/TargetSubtarget.h?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/include/llvm/Target/TargetSubtarget.h (original)
+++ llvm/trunk/include/llvm/Target/TargetSubtarget.h Thu Oct 22 18:19:17 2009
@@ -33,6 +33,10 @@
 protected: // Can only create subclasses...
   TargetSubtarget();
 public:
+  // AntiDepBreakMode - Type of anti-dependence breaking that should
+  // be performed before post-RA scheduling.
+  typedef enum { ANTIDEP_NONE, ANTIDEP_CRITICAL, ANTIDEP_ALL } AntiDepBreakMode;
+
   virtual ~TargetSubtarget();
 
   /// getSpecialAddressLatency - For targets where it is beneficial to
@@ -43,8 +47,10 @@
 
   // enablePostRAScheduler - If the target can benefit from post-regalloc
   // scheduling and the specified optimization level meets the requirement
-  // return true to enable post-register-allocation scheduling.
-  virtual bool enablePostRAScheduler(CodeGenOpt::Level OptLevel) const {
+  // return true to enable post-register-allocation scheduling. 
+  virtual bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,
+                                     AntiDepBreakMode& mode) const {
+    mode = ANTIDEP_NONE;
     return false;
   }
 

Modified: llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp (original)
+++ llvm/trunk/lib/CodeGen/PostRASchedulerList.cpp Thu Oct 22 18:19:17 2009
@@ -128,6 +128,9 @@
     /// AA - AliasAnalysis for making memory reference queries.
     AliasAnalysis *AA;
 
+    /// AntiDepMode - Anti-dependence breaking mode
+    TargetSubtarget::AntiDepBreakMode AntiDepMode;
+
     /// Classes - For live regs that are only used in one register class in a
     /// live range, the register class. If the register is not live, the
     /// corresponding value is null. If the register is live but used in
@@ -156,10 +159,11 @@
                          const MachineLoopInfo &MLI,
                          const MachineDominatorTree &MDT,
                          ScheduleHazardRecognizer *HR,
-                         AliasAnalysis *aa)
+                         AliasAnalysis *aa,
+                         TargetSubtarget::AntiDepBreakMode adm)
       : ScheduleDAGInstrs(MF, MLI, MDT), Topo(SUnits),
         AllocatableSet(TRI->getAllocatableSet(MF)),
-        HazardRec(HR), AA(aa) {}
+      HazardRec(HR), AA(aa), AntiDepMode(adm) {}
 
     ~SchedulePostRATDList() {
       delete HazardRec;
@@ -234,16 +238,23 @@
   AA = &getAnalysis<AliasAnalysis>();
 
   // Check for explicit enable/disable of post-ra scheduling.
+  TargetSubtarget::AntiDepBreakMode AntiDepMode = TargetSubtarget::ANTIDEP_NONE;
   if (EnablePostRAScheduler.getPosition() > 0) {
     if (!EnablePostRAScheduler)
       return false;
   } else {
     // Check that post-RA scheduling is enabled for this target.
     const TargetSubtarget &ST = Fn.getTarget().getSubtarget<TargetSubtarget>();
-    if (!ST.enablePostRAScheduler(OptLevel))
+    if (!ST.enablePostRAScheduler(OptLevel, AntiDepMode))
       return false;
   }
 
+  // Check for antidep breaking override...
+  if (EnableAntiDepBreaking.getPosition() > 0) {
+    AntiDepMode = (EnableAntiDepBreaking) ? 
+      TargetSubtarget::ANTIDEP_CRITICAL : TargetSubtarget::ANTIDEP_NONE;
+  }
+
   DEBUG(errs() << "PostRAScheduler\n");
 
   const MachineLoopInfo &MLI = getAnalysis<MachineLoopInfo>();
@@ -253,7 +264,7 @@
     (ScheduleHazardRecognizer *)new ExactHazardRecognizer(InstrItins) :
     (ScheduleHazardRecognizer *)new SimpleHazardRecognizer();
 
-  SchedulePostRATDList Scheduler(Fn, MLI, MDT, HR, AA);
+  SchedulePostRATDList Scheduler(Fn, MLI, MDT, HR, AA, AntiDepMode);
 
   // Loop over all of the basic blocks
   for (MachineFunction::iterator MBB = Fn.begin(), MBBe = Fn.end();
@@ -393,7 +404,7 @@
   // Build the scheduling graph.
   BuildSchedGraph(AA);
 
-  if (EnableAntiDepBreaking) {
+  if (AntiDepMode != TargetSubtarget::ANTIDEP_NONE) {
     if (BreakAntiDependencies()) {
       // We made changes. Update the dependency graph.
       // Theoretically we could update the graph in place:

Modified: llvm/trunk/lib/Target/ARM/ARMSubtarget.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMSubtarget.h?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMSubtarget.h (original)
+++ llvm/trunk/lib/Target/ARM/ARMSubtarget.h Thu Oct 22 18:19:17 2009
@@ -128,7 +128,9 @@
   
   /// enablePostRAScheduler - True at 'More' optimization except
   /// for Thumb1.
-  bool enablePostRAScheduler(CodeGenOpt::Level OptLevel) const {
+  bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,
+                             TargetSubtarget::AntiDepBreakMode& mode) const {
+    mode = TargetSubtarget::ANTIDEP_NONE;
     return PostRAScheduler && OptLevel >= CodeGenOpt::Default;
   }
 

Modified: llvm/trunk/lib/Target/X86/X86Subtarget.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Subtarget.h?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86Subtarget.h (original)
+++ llvm/trunk/lib/Target/X86/X86Subtarget.h Thu Oct 22 18:19:17 2009
@@ -218,7 +218,9 @@
 
   /// enablePostRAScheduler - X86 target is enabling post-alloc scheduling
   /// at 'More' optimization level.
-  bool enablePostRAScheduler(CodeGenOpt::Level OptLevel) const {
+  bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,
+                             TargetSubtarget::AntiDepBreakMode& mode) const {
+    mode = TargetSubtarget::ANTIDEP_NONE;
     return OptLevel >= CodeGenOpt::Default;
   }
 };

Modified: llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll (original)
+++ llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll Thu Oct 22 18:19:17 2009
@@ -11,11 +11,11 @@
     %tmp14 = fadd float %tmp12, %tmp7
     ret float %tmp14
 
-; CHECK:      mulss	LCPI1_3(%rip)
-; CHECK-NEXT: mulss	LCPI1_0(%rip)
+; CHECK:      mulss	LCPI1_0(%rip)
 ; CHECK-NEXT: mulss	LCPI1_1(%rip)
-; CHECK-NEXT: mulss	LCPI1_2(%rip)
 ; CHECK-NEXT: addss
+; CHECK:      mulss	LCPI1_3(%rip)
+; CHECK-NEXT: mulss	LCPI1_2(%rip)
 ; CHECK-NEXT: addss
 ; CHECK-NEXT: addss
 ; CHECK-NEXT: ret

Modified: llvm/trunk/test/CodeGen/X86/sse2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse2.ll?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/sse2.ll (original)
+++ llvm/trunk/test/CodeGen/X86/sse2.ll Thu Oct 22 18:19:17 2009
@@ -10,10 +10,10 @@
         
 ; CHECK: t1:
 ; CHECK: 	movl	8(%esp), %eax
-; CHECK-NEXT: 	movl	4(%esp), %ecx
 ; CHECK-NEXT: 	movapd	(%eax), %xmm0
+; CHECK-NEXT: 	movl	4(%esp), %eax
 ; CHECK-NEXT: 	movlpd	12(%esp), %xmm0
-; CHECK-NEXT: 	movapd	%xmm0, (%ecx)
+; CHECK-NEXT: 	movapd	%xmm0, (%eax)
 ; CHECK-NEXT: 	ret
 }
 
@@ -26,9 +26,9 @@
         
 ; CHECK: t2:
 ; CHECK: 	movl	8(%esp), %eax
-; CHECK-NEXT: 	movl	4(%esp), %ecx
 ; CHECK-NEXT: 	movapd	(%eax), %xmm0
+; CHECK-NEXT: 	movl	4(%esp), %eax
 ; CHECK-NEXT: 	movhpd	12(%esp), %xmm0
-; CHECK-NEXT: 	movapd	%xmm0, (%ecx)
+; CHECK-NEXT: 	movapd	%xmm0, (%eax)
 ; CHECK-NEXT: 	ret
 }

Modified: llvm/trunk/test/CodeGen/X86/sse3.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse3.ll?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/sse3.ll (original)
+++ llvm/trunk/test/CodeGen/X86/sse3.ll Thu Oct 22 18:19:17 2009
@@ -168,11 +168,11 @@
         ret void
 ; X64: 	t10:
 ; X64: 		pextrw	$4, %xmm0, %eax
-; X64: 		pextrw	$6, %xmm0, %edx
 ; X64: 		movlhps	%xmm1, %xmm1
 ; X64: 		pshuflw	$8, %xmm1, %xmm1
 ; X64: 		pinsrw	$2, %eax, %xmm1
-; X64: 		pinsrw	$3, %edx, %xmm1
+; X64: 		pextrw	$6, %xmm0, %eax
+; X64: 		pinsrw	$3, %eax, %xmm1
 }
 
 

Modified: llvm/trunk/test/CodeGen/X86/vshift-1.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vshift-1.ll?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/vshift-1.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vshift-1.ll Thu Oct 22 18:19:17 2009
@@ -63,7 +63,7 @@
 ; CHECK: shift3b:
 ; CHECK: movzwl
 ; CHECK: movd
-; CHECK-NEXT: psllw
+; CHECK: psllw
   %0 = insertelement <8 x i16> undef, i16 %amt, i32 0
   %1 = insertelement <8 x i16> %0, i16 %amt, i32 1
   %2 = insertelement <8 x i16> %0, i16 %amt, i32 2

Modified: llvm/trunk/test/CodeGen/X86/vshift-2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vshift-2.ll?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/vshift-2.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vshift-2.ll Thu Oct 22 18:19:17 2009
@@ -63,7 +63,7 @@
 ; CHECK: shift3b:
 ; CHECK: movzwl
 ; CHECK: movd
-; CHECK-NEXT: psrlw
+; CHECK: psrlw
   %0 = insertelement <8 x i16> undef, i16 %amt, i32 0
   %1 = insertelement <8 x i16> %0, i16 %amt, i32 1
   %2 = insertelement <8 x i16> %0, i16 %amt, i32 2

Modified: llvm/trunk/test/CodeGen/X86/vshift-3.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vshift-3.ll?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/vshift-3.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vshift-3.ll Thu Oct 22 18:19:17 2009
@@ -52,7 +52,7 @@
 ; CHECK: shift3b:
 ; CHECK: movzwl
 ; CHECK: movd
-; CHECK-NEXT: psraw
+; CHECK: psraw
   %0 = insertelement <8 x i16> undef, i16 %amt, i32 0
   %1 = insertelement <8 x i16> %0, i16 %amt, i32 1
   %2 = insertelement <8 x i16> %0, i16 %amt, i32 2

Modified: llvm/trunk/test/CodeGen/X86/vshift-5.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vshift-5.ll?rev=84911&r1=84910&r2=84911&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/vshift-5.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vshift-5.ll Thu Oct 22 18:19:17 2009
@@ -6,7 +6,7 @@
 entry:
 ; CHECK: shift5a:
 ; CHECK: movd
-; CHECK-NEXT: pslld
+; CHECK: pslld
   %amt = load i32* %pamt 
   %tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
   %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer 
@@ -20,7 +20,7 @@
 entry:
 ; CHECK: shift5b:
 ; CHECK: movd
-; CHECK-NEXT: psrad
+; CHECK: psrad
   %amt = load i32* %pamt 
   %tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
   %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer 




From jyasskin at gmail.com  Thu Oct 22 18:29:45 2009
From: jyasskin at gmail.com (jyasskin at gmail.com)
Date: Thu, 22 Oct 2009 23:29:45 +0000
Subject: [llvm-commits] [PATCH] Free machine code when its Function is
	destroyed
Message-ID: <0016367b65c678169a04768e7871@google.com>

Reviewers: ,

Message:
Please take a look.

Description:
Patch at http://codereview.appspot.com/download/issue140041_1.diff

Automatically do the equivalent of freeMachineCodeForFunction(F) when F
is being destroyed. This allows users to run global optimizations like
globaldce even after some functions have been jitted.

This patch also removes the Function* parameter to
JITEventListener::NotifyFreeingMachineCode() since it can cause that to
be called when the Function is partially destroyed. This change will be
even more helpful later when I think we'll want to allow machine code to
actually outlive its Function.

This patch depends slightly on the test in
http://codereview.appspot.com/133043, but I can make them independent if
this gets reviewed first.

Please review this at http://codereview.appspot.com/140041

Affected files:
   M include/llvm/ExecutionEngine/ExecutionEngine.h
   M include/llvm/ExecutionEngine/JITEventListener.h
   M lib/ExecutionEngine/JIT/JIT.cpp
   M lib/ExecutionEngine/JIT/JIT.h
   M lib/ExecutionEngine/JIT/JITEmitter.cpp
   M lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp
   M unittests/ExecutionEngine/JIT/JITEventListenerTest.cpp
   M unittests/ExecutionEngine/JIT/JITTest.cpp




From stuart at apple.com  Thu Oct 22 18:52:41 2009
From: stuart at apple.com (Stuart Hastings)
Date: Thu, 22 Oct 2009 23:52:41 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84915 - /llvm-gcc-4.2/trunk/build_gcc
Message-ID: <200910222352.n9MNqfeA009083@zion.cs.uiuc.edu>

Author: stuart
Date: Thu Oct 22 18:52:41 2009
New Revision: 84915

URL: http://llvm.org/viewvc/llvm-project?rev=84915&view=rev
Log:
Temporary Barolo compatibility hack; look for C++ headers under i686-apple-darwin10, not darwin11.

Modified:
    llvm-gcc-4.2/trunk/build_gcc

Modified: llvm-gcc-4.2/trunk/build_gcc
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/build_gcc?rev=84915&r1=84914&r2=84915&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/build_gcc (original)
+++ llvm-gcc-4.2/trunk/build_gcc Thu Oct 22 18:52:41 2009
@@ -116,7 +116,11 @@
 NON_ARM_CONFIGFLAGS="--with-gxx-include-dir=/usr/include/c++/$LIBSTDCXX_VERSION"
 # LLVM LOCAL end
 
-DARWIN_VERS=`uname -r | sed 's/\..*//'`
+#DARWIN_VERS=`uname -r | sed 's/\..*//'`
+# KLUDGE: Barolo compatbility hack!
+# Remove when /usr/include/c++/4.2.1/i686-apple-darwin11 is established!
+DARWIN_VERS=`sw_vers -productVersion | sed 's/\..*//'`
+# end KLUDGE
 echo DARWIN_VERS = $DARWIN_VERS
 
 # APPLE LOCAL begin ARM




From isanbard at gmail.com  Thu Oct 22 19:01:05 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Fri, 23 Oct 2009 00:01:05 -0000
Subject: [llvm-commits] [llvm] r84916 -
	/llvm/trunk/lib/CodeGen/StackProtector.cpp
Message-ID: <200910230001.n9N015nr009989@zion.cs.uiuc.edu>

Author: void
Date: Thu Oct 22 19:01:05 2009
New Revision: 84916

URL: http://llvm.org/viewvc/llvm-project?rev=84916&view=rev
Log:
Neuter stack protectors by only checking character arrays. This is what GCC
does.

Modified:
    llvm/trunk/lib/CodeGen/StackProtector.cpp

Modified: llvm/trunk/lib/CodeGen/StackProtector.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/StackProtector.cpp?rev=84916&r1=84915&r2=84916&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/StackProtector.cpp (original)
+++ llvm/trunk/lib/CodeGen/StackProtector.cpp Thu Oct 22 19:01:05 2009
@@ -111,11 +111,16 @@
           // protectors.
           return true;
 
-        if (const ArrayType *AT = dyn_cast<ArrayType>(AI->getAllocatedType()))
+        if (const ArrayType *AT = dyn_cast<ArrayType>(AI->getAllocatedType())) {
+          // We apparently only care about character arrays.
+          if (AT->getElementType() != Type::getInt8Ty(AT->getContext()))
+            continue;
+
           // If an array has more than SSPBufferSize bytes of allocated space,
           // then we emit stack protectors.
           if (SSPBufferSize <= TD->getTypeAllocSize(AT))
             return true;
+        }
       }
   }
 




From stuart at apple.com  Thu Oct 22 19:06:00 2009
From: stuart at apple.com (Stuart Hastings)
Date: Fri, 23 Oct 2009 00:06:00 -0000
Subject: [llvm-commits] [llvm-gcc-4.2] r84918 - /llvm-gcc-4.2/trunk/build_gcc
Message-ID: <200910230006.n9N060tL010397@zion.cs.uiuc.edu>

Author: stuart
Date: Thu Oct 22 19:06:00 2009
New Revision: 84918

URL: http://llvm.org/viewvc/llvm-project?rev=84918&view=rev
Log:
Revert.

Modified:
    llvm-gcc-4.2/trunk/build_gcc

Modified: llvm-gcc-4.2/trunk/build_gcc
URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/build_gcc?rev=84918&r1=84917&r2=84918&view=diff

==============================================================================
--- llvm-gcc-4.2/trunk/build_gcc (original)
+++ llvm-gcc-4.2/trunk/build_gcc Thu Oct 22 19:06:00 2009
@@ -116,11 +116,7 @@
 NON_ARM_CONFIGFLAGS="--with-gxx-include-dir=/usr/include/c++/$LIBSTDCXX_VERSION"
 # LLVM LOCAL end
 
-#DARWIN_VERS=`uname -r | sed 's/\..*//'`
-# KLUDGE: Barolo compatbility hack!
-# Remove when /usr/include/c++/4.2.1/i686-apple-darwin11 is established!
-DARWIN_VERS=`sw_vers -productVersion | sed 's/\..*//'`
-# end KLUDGE
+DARWIN_VERS=`uname -r | sed 's/\..*//'`
 echo DARWIN_VERS = $DARWIN_VERS
 
 # APPLE LOCAL begin ARM




From vhernandez at apple.com  Thu Oct 22 19:59:10 2009
From: vhernandez at apple.com (Victor Hernandez)
Date: Fri, 23 Oct 2009 00:59:10 -0000
Subject: [llvm-commits] [llvm] r84919 -
	/llvm/trunk/lib/AsmParser/LLParser.cpp
Message-ID: <200910230059.n9N0xAmr014825@zion.cs.uiuc.edu>

Author: hernande
Date: Thu Oct 22 19:59:10 2009
New Revision: 84919

URL: http://llvm.org/viewvc/llvm-project?rev=84919&view=rev
Log:
Minor code cleanup.


Modified:
    llvm/trunk/lib/AsmParser/LLParser.cpp

Modified: llvm/trunk/lib/AsmParser/LLParser.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/AsmParser/LLParser.cpp?rev=84919&r1=84918&r2=84919&view=diff

==============================================================================
--- llvm/trunk/lib/AsmParser/LLParser.cpp (original)
+++ llvm/trunk/lib/AsmParser/LLParser.cpp Thu Oct 22 19:59:10 2009
@@ -3485,12 +3485,11 @@
   // Autoupgrade old malloc instruction to malloc call.
   // FIXME: Remove in LLVM 3.0.
   const Type *IntPtrTy = Type::getInt32Ty(Context);
-  const Type *Int8PtrTy = Type::getInt8PtrTy(Context);
   if (!MallocF)
     // Prototype malloc as "void *(int32)".
     // This function is renamed as "malloc" in ValidateEndOfModule().
-    MallocF = cast<Function>(M->getOrInsertFunction(NULL, Int8PtrTy, 
-                                                    IntPtrTy, NULL));
+    MallocF = cast<Function>(
+       M->getOrInsertFunction("", Type::getInt8PtrTy(Context), IntPtrTy, NULL));
   Inst = CallInst::CreateMalloc(BB, IntPtrTy, Ty, Size, MallocF);
   return false;
 }




From vhernandez at apple.com  Thu Oct 22 19:59:40 2009
From: vhernandez at apple.com (Victor Hernandez)
Date: Thu, 22 Oct 2009 17:59:40 -0700
Subject: [llvm-commits] [llvm] r84292 - in /llvm/trunk: examples/BrainF/
	include/llvm/ lib/AsmParser/ lib/Bitcode/Reader/
	lib/Transforms/IPO/ lib/Transforms/Scalar/ lib/VMCore/
	test/Transforms/GlobalOpt/ test/Transforms/IndMemRem/
	test/Transforms/InstCombine/
In-Reply-To: <7B90F86B-6555-42E8-9ED7-C45C1216CF62@apple.com>
References: <200910170000.n9H00RWN030901@zion.cs.uiuc.edu>
	<797711D6-5B12-4DA3-BFF4-97D38BDC6E58@apple.com>
	<5F7FC9EE-DD30-438F-833D-9F9338021A1A@apple.com>
	<7B90F86B-6555-42E8-9ED7-C45C1216CF62@apple.com>
Message-ID: <49AC3589-633A-4AAE-8BC5-40E0DCC59A35@apple.com>

I just made your suggested changes in rev 84919.

Victor

On Oct 21, 2009, at 5:38 PM, Chris Lattner wrote:

> On Oct 21, 2009, at 12:12 PM, Victor Hernandez wrote:
>>> This should use V->replaceAllUsesWith() instead of manually  
>>> walking the use list.  Doing this will handle non-call users as  
>>> well as call users.  What happens if the prototypes don't agree  
>>> for these two functions?  I think you'll get an abort.  Please fix  
>>> by using a constantexpr bitcast to the expected type.
>>
>> Fixed:
>
> Much nicer, thanks.
>
>>>> +    const Type* IntPtrTy = Type::getInt32Ty(Context);
>>>> +    const Type* Int8PtrTy = PointerType::getUnqual 
>>>> (Type::getInt8Ty(Context));
>>>
>>> Please use 'const Type *IntPtrTy', watch the * placement.  There  
>>> is now a helper function that you can use instead of  
>>> 'PointerType::getUnqual(Type::getInt8Ty(Context));', please use it.
>>
>> Fixed:
>>   const Type *IntPtrTy = Type::getInt32Ty(Context);
>>   const Type *Int8PtrTy = Type::getInt8PtrTy(Context);
>
> Ok:
>
> +  // Autoupgrade old malloc instruction to malloc call.
> +  // FIXME: Remove in LLVM 3.0.
> +  const Type *IntPtrTy = Type::getInt32Ty(Context);
> +  const Type *Int8PtrTy = Type::getInt8PtrTy(Context);
> +  if (!MallocF)
> +    // Prototype malloc as "void *(int32)".
> +    // This function is renamed as "malloc" in ValidateEndOfModule().
> +    MallocF = cast<Function>(M->getOrInsertFunction(NULL, Int8PtrTy,
> +                                                    IntPtrTy, NULL));
> +  Inst = CallInst::CreateMalloc(BB, IntPtrTy, Ty, Size, MallocF);
>
> It doesn't matter because this is very cold code, but you might as  
> well sink the 'get' of Int8PtrTy into the 'if' for tidiness.
>
>>
>>>
>>>> +    if (!MallocF)
>>>> +      // Prototype malloc as "void *autoupgrade_malloc(int32)".
>>>> +      MallocF = cast<Function>(M->getOrInsertFunction 
>>>> ("autoupgrade_malloc",
>>>> +                               Int8PtrTy, IntPtrTy, NULL));
>>>
>>> Why do you bother naming autoupgrade_malloc?  This is in the  
>>> user's namespace, so it can conflict with a function named  
>>> autoupgrade_malloc in the user's code.  Please just leave the name  
>>> empty, making it impossible for it to conflict.
>>
>> Fixed.
>
> Thanks.  While passing in NULL to getOrInsertFunction will probably  
> work, please pass in "" instead.  This makes it obvious that it is  
> the name you are specifying.
>
>>>
>>> The old code was correct because it would delete the call to  
>>> malloc, so it is safe to assume that the malloc (which never got  
>>> run) never ran out of space.  The transformation that does this  
>>> should just be removed.
>>
>> I fixed InstCombine so that the icmp is removed only if the malloc  
>> call is removed (which requires explicit removal because the  
>> Worklist won't DCE any calls since they can have side-effects).  I  
>> also updated the testcase to use FileCheck.
>
> Looks good.
>
> Thanks Victor,
>
> -Chris

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091022/78712fe5/attachment.html 

From isanbard at gmail.com  Thu Oct 22 20:37:01 2009
From: isanbard at gmail.com (Bill Wendling)
Date: Fri, 23 Oct 2009 01:37:01 -0000
Subject: [llvm-commits] [llvm] r84921 -
	/llvm/trunk/test/FrontendC++/alignstack.cpp
Message-ID: <200910230137.n9N1b26a017746@zion.cs.uiuc.edu>

Author: void
Date: Thu Oct 22 20:37:01 2009
New Revision: 84921

URL: http://llvm.org/viewvc/llvm-project?rev=84921&view=rev
Log:
This is passing on Darwin PPC.

Modified:
    llvm/trunk/test/FrontendC++/alignstack.cpp

Modified: llvm/trunk/test/FrontendC++/alignstack.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC%2B%2B/alignstack.cpp?rev=84921&r1=84920&r2=84921&view=diff

==============================================================================
--- llvm/trunk/test/FrontendC++/alignstack.cpp (original)
+++ llvm/trunk/test/FrontendC++/alignstack.cpp Thu Oct 22 20:37:01 2009
@@ -1,7 +1,7 @@
 // RUN: %llvmgxx %s -fasm-blocks -S -o - | FileCheck %s
 // Complicated expression as jump target
 // XFAIL: *
-// XTARGET: x86,i386,i686
+// XTARGET: x86,i386,i686,darwin
 
 void Method3()
 {




From vkutuzov at accesssoftek.com  Thu Oct 22 20:28:25 2009
From: vkutuzov at accesssoftek.com (Viktor Kutuzov)
Date: Thu, 22 Oct 2009 18:28:25 -0700
Subject: [llvm-commits] [PATCH] Fix to pass options from Gold plugin to LTO
	codegen
References: <3FE2DECEF38F9247AE7C8BC1FD2B015F0817022C@ALLISON>
Message-ID: <7F9FE9028F544389BE51F80E181E6B0C@andreic6e7fe55>

Hello everyone,

Please find the patch attached.
This should fix the issue when gold plugin doesn't pass the  -plugin-opt options down to the LTO codegen.
See the thread "[LLVMdev] getting gold plugin to work?" for more details.

Best regards,
Viktor
-------------- next part --------------
A non-text attachment was scrubbed...
Name: llvm-gold-plugin-process_command_line.diff
Type: application/octet-stream
Size: 7031 bytes
Desc: not available
Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20091022/f38be070/attachment.obj 

From deeppatel1987 at gmail.com  Thu Oct 22 21:04:49 2009
From: deeppatel1987 at gmail.com (Sandeep Patel)
Date: Fri, 23 Oct 2009 02:04:49 +0000
Subject: [llvm-commits] [llvm] r84431 - in /llvm/trunk:
	lib/Target/X86/X86Subtarget.h
	test/CodeGen/X86/2007-01-08-InstrSched.ll
	test/CodeGen/X86/2008-07-11-SpillerBug.ll
	test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
	test/CodeGen/X86/2009-09-19-SchedCustom
Message-ID: <305d6f60910221904p64a00aacxe873f7f891d889d2@mail.gmail.com>

This is breaking our canadian cross that runs on MinGW.

An example of the problem from gcc.o:record_temp_file (the first
function in gcc.o to exhibit the problem, but not the cause of the
crash):

bad code:
5d: a3 14 00 00 00  mov    %eax,0x14	5e: dir32	.bss
62: 89 48 04        mov    %ecx,0x4(%eax)
65: 89 38           mov    %edi,(%eax)

good code:
5d: 89 48 04        mov    %ecx,0x4(%eax)
60: 89 38           mov    %edi,(%eax)
62: a3 14 00 00 00  mov    %eax,0x14	63: dir32	.bss

deep

On Sun, Oct 18, 2009 at 7:57 PM, Evan Cheng <evan.cheng at apple.com> wrote:
> Author: evancheng
> Date: Sun Oct 18 14:57:27 2009
> New Revision: 84431
>
> URL: http://llvm.org/viewvc/llvm-project?rev=84431&view=rev
> Log:
> Turn on post-alloc scheduling for x86.
>
> Modified:
> ? ?llvm/trunk/lib/Target/X86/X86Subtarget.h
> ? ?llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
> ? ?llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll
> ? ?llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
> ? ?llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll
> ? ?llvm/trunk/test/CodeGen/X86/abi-isel.ll
> ? ?llvm/trunk/test/CodeGen/X86/fastcc.ll
> ? ?llvm/trunk/test/CodeGen/X86/peep-test-3.ll
> ? ?llvm/trunk/test/CodeGen/X86/pic.ll
> ? ?llvm/trunk/test/CodeGen/X86/sink-hoist.ll
> ? ?llvm/trunk/test/CodeGen/X86/sse2.ll
> ? ?llvm/trunk/test/CodeGen/X86/sse3.ll
> ? ?llvm/trunk/test/CodeGen/X86/tailcallstack64.ll
> ? ?llvm/trunk/test/CodeGen/X86/widen_arith-3.ll
>
> Modified: llvm/trunk/lib/Target/X86/X86Subtarget.h
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Subtarget.h?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/X86Subtarget.h (original)
> +++ llvm/trunk/lib/Target/X86/X86Subtarget.h Sun Oct 18 14:57:27 2009
> @@ -219,8 +219,7 @@
> ? /// enablePostRAScheduler - X86 target is enabling post-alloc scheduling
> ? /// at 'More' optimization level.
> ? bool enablePostRAScheduler(CodeGenOpt::Level OptLevel) const {
> - ? ?// FIXME: This causes llvm to miscompile itself on i386. :-(
> - ? ?return false /*OptLevel >= CodeGenOpt::Default*/;
> + ? ?return OptLevel >= CodeGenOpt::Default;
> ? }
> ?};
>
>
> Modified: llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll Sun Oct 18 14:57:27 2009
> @@ -11,9 +11,12 @@
> ? ? %tmp14 = fadd float %tmp12, %tmp7
> ? ? ret float %tmp14
>
> -; CHECK: ? ? ?mulss ? ?LCPI1_2(%rip)
> +; CHECK: ? ? ?mulss ? ?LCPI1_3(%rip)
> +; CHECK-NEXT: mulss ? ?LCPI1_0(%rip)
> +; CHECK-NEXT: mulss ? ?LCPI1_1(%rip)
> +; CHECK-NEXT: mulss ? ?LCPI1_2(%rip)
> +; CHECK-NEXT: addss
> ?; CHECK-NEXT: addss
> -; CHECK-NEXT: mulss ? ?LCPI1_3(%rip)
> ?; CHECK-NEXT: addss
> ?; CHECK-NEXT: ret
> ?}
>
> Modified: llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll Sun Oct 18 14:57:27 2009
> @@ -1,4 +1,4 @@
> -; RUN: llc < %s -march=x86 -relocation-model=static -disable-fp-elim | FileCheck %s
> +; RUN: llc < %s -march=x86 -relocation-model=static -disable-fp-elim -post-RA-scheduler=false | FileCheck %s
> ?; PR2536
>
>
>
> Modified: llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll Sun Oct 18 14:57:27 2009
> @@ -1,5 +1,5 @@
> ?; RUN: llc -mtriple=i386-apple-darwin10.0 -relocation-model=pic \
> -; RUN: ? ? -disable-fp-elim -mattr=-sse41,-sse3,+sse2 < %s | \
> +; RUN: ? ? -disable-fp-elim -mattr=-sse41,-sse3,+sse2 -post-RA-scheduler=false < %s | \
> ?; RUN: ? FileCheck %s
> ?; rdar://6808032
>
>
> Modified: llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll Sun Oct 18 14:57:27 2009
> @@ -9,9 +9,7 @@
> ? br label %bb
>
> ?bb: ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ; preds = %bb1, %entry
> -; CHECK: ? ? ?movl %e
> -; CHECK-NEXT: addl $1
> -; CHECK-NEXT: movl %e
> +; CHECK: ? ? ?addl $1
> ?; CHECK-NEXT: adcl $0
> ? %i.0 = phi i64 [ 0, %entry ], [ %0, %bb1 ] ? ? ?; <i64> [#uses=1]
> ? %0 = add nsw i64 %i.0, 1 ? ? ? ? ? ? ? ? ? ? ? ?; <i64> [#uses=2]
>
> Modified: llvm/trunk/test/CodeGen/X86/abi-isel.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/abi-isel.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/abi-isel.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/abi-isel.ll Sun Oct 18 14:57:27 2009
> @@ -1,16 +1,16 @@
> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-32-STATIC
> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-32-PIC
> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-32-STATIC
> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-32-PIC
>
> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-64-STATIC
> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=LINUX-64-PIC
> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-64-STATIC
> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-64-PIC
>
> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=DARWIN-32-STATIC
> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=dynamic-no-pic -code-model=small | FileCheck %s -check-prefix=DARWIN-32-DYNAMIC
> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=DARWIN-32-PIC
> -
> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=DARWIN-64-STATIC
> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=dynamic-no-pic -code-model=small | FileCheck %s -check-prefix=DARWIN-64-DYNAMIC
> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=DARWIN-64-PIC
> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-STATIC
> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=dynamic-no-pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-DYNAMIC
> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-PIC
> +
> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-STATIC
> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=dynamic-no-pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-DYNAMIC
> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-PIC
>
> ?@src = external global [131072 x i32]
> ?@dst = external global [131072 x i32]
>
> Modified: llvm/trunk/test/CodeGen/X86/fastcc.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/fastcc.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/fastcc.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/fastcc.ll Sun Oct 18 14:57:27 2009
> @@ -1,4 +1,4 @@
> -; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+sse2 | FileCheck %s
> +; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+sse2 -post-RA-scheduler=false | FileCheck %s
> ?; CHECK: movsd %xmm0, 8(%esp)
> ?; CHECK: xorl %ecx, %ecx
>
>
> Modified: llvm/trunk/test/CodeGen/X86/peep-test-3.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/peep-test-3.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/peep-test-3.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/peep-test-3.ll Sun Oct 18 14:57:27 2009
> @@ -1,4 +1,4 @@
> -; RUN: llc < %s -march=x86 | FileCheck %s
> +; RUN: llc < %s -march=x86 -post-RA-scheduler=false | FileCheck %s
> ?; rdar://7226797
>
> ?; LLVM should omit the testl and use the flags result from the orl.
>
> Modified: llvm/trunk/test/CodeGen/X86/pic.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pic.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/pic.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/pic.ll Sun Oct 18 14:57:27 2009
> @@ -1,4 +1,4 @@
> -; RUN: llc < %s -mtriple=i686-pc-linux-gnu -relocation-model=pic -asm-verbose=false | FileCheck %s -check-prefix=LINUX
> +; RUN: llc < %s -mtriple=i686-pc-linux-gnu -relocation-model=pic -asm-verbose=false -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX
>
> ?@ptr = external global i32*
> ?@dst = external global i32
>
> Modified: llvm/trunk/test/CodeGen/X86/sink-hoist.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sink-hoist.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/sink-hoist.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/sink-hoist.ll Sun Oct 18 14:57:27 2009
> @@ -6,10 +6,10 @@
> ?; that it's conditionally evaluated.
>
> ?; CHECK: foo:
> -; CHECK-NEXT: divsd
> -; CHECK: ? ? ?testb $1, %dil
> -; CHECK-NEXT: jne
> ?; CHECK: ? ? ?divsd
> +; CHECK-NEXT: testb $1, %dil
> +; CHECK-NEXT: jne
> +; CHECK-NEXT: divsd
>
> ?define double @foo(double %x, double %y, i1 %c) nounwind {
> ? %a = fdiv double %x, 3.2
>
> Modified: llvm/trunk/test/CodeGen/X86/sse2.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse2.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/sse2.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/sse2.ll Sun Oct 18 14:57:27 2009
> @@ -10,10 +10,10 @@
>
> ?; CHECK: t1:
> ?; CHECK: ? ? ? movl ? ?8(%esp), %eax
> +; CHECK-NEXT: ?movl ? ?4(%esp), %ecx
> ?; CHECK-NEXT: ?movapd ?(%eax), %xmm0
> ?; CHECK-NEXT: ?movlpd ?12(%esp), %xmm0
> -; CHECK-NEXT: ?movl ? ?4(%esp), %eax
> -; CHECK-NEXT: ?movapd ?%xmm0, (%eax)
> +; CHECK-NEXT: ?movapd ?%xmm0, (%ecx)
> ?; CHECK-NEXT: ?ret
> ?}
>
> @@ -26,9 +26,9 @@
>
> ?; CHECK: t2:
> ?; CHECK: ? ? ? movl ? ?8(%esp), %eax
> +; CHECK-NEXT: ?movl ? ?4(%esp), %ecx
> ?; CHECK-NEXT: ?movapd ?(%eax), %xmm0
> ?; CHECK-NEXT: ?movhpd ?12(%esp), %xmm0
> -; CHECK-NEXT: ?movl ? ?4(%esp), %eax
> -; CHECK-NEXT: ?movapd ?%xmm0, (%eax)
> +; CHECK-NEXT: ?movapd ?%xmm0, (%ecx)
> ?; CHECK-NEXT: ?ret
> ?}
>
> Modified: llvm/trunk/test/CodeGen/X86/sse3.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse3.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/sse3.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/sse3.ll Sun Oct 18 14:57:27 2009
> @@ -17,8 +17,8 @@
>
> ?; X64: t0:
> ?; X64: ? ? ? ? movddup (%rsi), %xmm0
> -; X64: ?pshuflw ? ? ? ?$0, %xmm0, %xmm0
> ?; X64: xorl ? ?%eax, %eax
> +; X64: ?pshuflw ? ? ? ?$0, %xmm0, %xmm0
> ?; X64: pinsrw ?$0, %eax, %xmm0
> ?; X64: movaps ?%xmm0, (%rdi)
> ?; X64: ret
> @@ -167,18 +167,12 @@
> ? ? ? ? store <4 x i16> %6, <4 x i16>* @g2, align 8
> ? ? ? ? ret void
> ?; X64: ? ? ? ? t10:
> -; X64: ? ? ? ? ? ? ? ? movq ? ?_g1 at GOTPCREL(%rip), %rax
> -; X64: ? ? ? ? ? ? ? ? movaps ?(%rax), %xmm0
> ?; X64: ? ? ? ? ? ? ? ? pextrw ?$4, %xmm0, %eax
> -; X64: ? ? ? ? ? ? ? ? movaps ?%xmm0, %xmm1
> +; X64: ? ? ? ? ? ? ? ? pextrw ?$6, %xmm0, %edx
> ?; X64: ? ? ? ? ? ? ? ? movlhps %xmm1, %xmm1
> ?; X64: ? ? ? ? ? ? ? ? pshuflw $8, %xmm1, %xmm1
> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$2, %eax, %xmm1
> -; X64: ? ? ? ? ? ? ? ? pextrw ?$6, %xmm0, %eax
> -; X64: ? ? ? ? ? ? ? ? pinsrw ?$3, %eax, %xmm1
> -; X64: ? ? ? ? ? ? ? ? movq ? ?_g2 at GOTPCREL(%rip), %rax
> -; X64: ? ? ? ? ? ? ? ? movq ? ?%xmm1, (%rax)
> -; X64: ? ? ? ? ? ? ? ? ret
> +; X64: ? ? ? ? ? ? ? ? pinsrw ?$3, %edx, %xmm1
> ?}
>
>
> @@ -189,8 +183,8 @@
> ? ? ? ?ret <8 x i16> %tmp7
>
> ?; X64: t11:
> -; X64: movd ? ?%xmm1, %eax
> ?; X64: movlhps %xmm0, %xmm0
> +; X64: movd ? ?%xmm1, %eax
> ?; X64: pshuflw $1, %xmm0, %xmm0
> ?; X64: pinsrw ?$1, %eax, %xmm0
> ?; X64: ret
> @@ -203,8 +197,8 @@
> ? ? ? ?ret <8 x i16> %tmp9
>
> ?; X64: t12:
> -; X64: ? ? ? ? pextrw ?$3, %xmm1, %eax
> ?; X64: ? ? ? ? movlhps %xmm0, %xmm0
> +; X64: ? ? ? ? pextrw ?$3, %xmm1, %eax
> ?; X64: ? ? ? ? pshufhw $3, %xmm0, %xmm0
> ?; X64: ? ? ? ? pinsrw ?$5, %eax, %xmm0
> ?; X64: ? ? ? ? ret
> @@ -256,18 +250,12 @@
> ? ? ? ? %tmp9 = shufflevector <16 x i8> %tmp8, <16 x i8> %T0, ?<16 x i32> < i32 0, i32 1, i32 2, i32 17, ?i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef , i32 undef >
> ? ? ? ? ret <16 x i8> %tmp9
> ?; X64: ? ? ? ? t16:
> -; X64: ? ? ? ? ? ? ? ? movaps ?LCPI17_0(%rip), %xmm1
> -; X64: ? ? ? ? ? ? ? ? movd ? ?%xmm1, %eax
> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$0, %eax, %xmm1
> ?; X64: ? ? ? ? ? ? ? ? pextrw ?$8, %xmm0, %eax
> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$1, %eax, %xmm1
> ?; X64: ? ? ? ? ? ? ? ? pextrw ?$1, %xmm1, %ecx
> ?; X64: ? ? ? ? ? ? ? ? movd ? ?%xmm1, %edx
> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$0, %edx, %xmm1
> -; X64: ? ? ? ? ? ? ? ? movzbl ?%cl, %ecx
> -; X64: ? ? ? ? ? ? ? ? andw ? ?$-256, %ax
> -; X64: ? ? ? ? ? ? ? ? orw ? ? %cx, %ax
> -; X64: ? ? ? ? ? ? ? ? movaps ?%xmm1, %xmm0
> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$1, %eax, %xmm0
> ?; X64: ? ? ? ? ? ? ? ? ret
> ?}
>
> Modified: llvm/trunk/test/CodeGen/X86/tailcallstack64.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/tailcallstack64.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/tailcallstack64.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/tailcallstack64.ll Sun Oct 18 14:57:27 2009
> @@ -3,19 +3,18 @@
> ?; Check that lowered arguments on the stack do not overwrite each other.
> ?; Add %in1 %p1 to a different temporary register (%eax).
> ?; CHECK: movl ?%edi, %eax
> -; CHECK: addl ?32(%rsp), %eax
> ?; Move param %in1 to temp register (%r10d).
> ?; CHECK: movl ?40(%rsp), %r10d
> -; Move result of addition to stack.
> -; CHECK: movl ?%eax, 40(%rsp)
> ?; Move param %in2 to stack.
> ?; CHECK: movl ?%r10d, 32(%rsp)
> +; Move result of addition to stack.
> +; CHECK: movl ?%eax, 40(%rsp)
> ?; Eventually, do a TAILCALL
> ?; CHECK: TAILCALL
>
> -declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %a, i32 %b)
> +declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %a, i32 %b) nounwind
>
> -define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in1, i32 %in2) {
> +define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in1, i32 %in2) nounwind {
> ?entry:
> ? ? ? ? %tmp = add i32 %in1, %p1
> ? ? ? ? %retval = tail call fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in2,i32 %tmp)
>
> Modified: llvm/trunk/test/CodeGen/X86/widen_arith-3.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/widen_arith-3.ll?rev=84431&r1=84430&r2=84431&view=diff
>
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/widen_arith-3.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/widen_arith-3.ll Sun Oct 18 14:57:27 2009
> @@ -1,7 +1,7 @@
> ?; RUN: llc < %s -march=x86 -mattr=+sse42 -disable-mmx | FileCheck %s
> ?; CHECK: paddw
> -; CHECK: movd
> ?; CHECK: pextrw
> +; CHECK: movd
>
> ?; Widen a v3i16 to v8i16 to do a vector add
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>



From evan.cheng at apple.com  Thu Oct 22 21:10:50 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 19:10:50 -0700
Subject: [llvm-commits] [llvm] r84431 - in /llvm/trunk:
	lib/Target/X86/X86Subtarget.h
	test/CodeGen/X86/2007-01-08-InstrSched.ll
	test/CodeGen/X86/2008-07-11-SpillerBug.ll
	test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
	test/CodeGen/X86/2009-09-19-SchedCustom
In-Reply-To: <305d6f60910221904p64a00aacxe873f7f891d889d2@mail.gmail.com>
References: <305d6f60910221904p64a00aacxe873f7f891d889d2@mail.gmail.com>
Message-ID: <C97226D4-B60A-4E1A-876B-14CC8F5E833E@apple.com>


On Oct 22, 2009, at 7:04 PM, Sandeep Patel wrote:

> This is breaking our canadian cross that runs on MinGW.
> 
> An example of the problem from gcc.o:record_temp_file (the first
> function in gcc.o to exhibit the problem, but not the cause of the
> crash):
> 
> bad code:
> 5d: a3 14 00 00 00  mov    %eax,0x14	5e: dir32	.bss
> 62: 89 48 04        mov    %ecx,0x4(%eax)
> 65: 89 38           mov    %edi,(%eax)
> 
> good code:
> 5d: 89 48 04        mov    %ecx,0x4(%eax)
> 60: 89 38           mov    %edi,(%eax)
> 62: a3 14 00 00 00  mov    %eax,0x14	63: dir32	.bss

What's wrong with the "bad code"?

You can disable it for MinGW to workaround the problem for now.

Evan

> 
> deep
> 
> On Sun, Oct 18, 2009 at 7:57 PM, Evan Cheng <evan.cheng at apple.com> wrote:
>> Author: evancheng
>> Date: Sun Oct 18 14:57:27 2009
>> New Revision: 84431
>> 
>> URL: http://llvm.org/viewvc/llvm-project?rev=84431&view=rev
>> Log:
>> Turn on post-alloc scheduling for x86.
>> 
>> Modified:
>>    llvm/trunk/lib/Target/X86/X86Subtarget.h
>>    llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
>>    llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll
>>    llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
>>    llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll
>>    llvm/trunk/test/CodeGen/X86/abi-isel.ll
>>    llvm/trunk/test/CodeGen/X86/fastcc.ll
>>    llvm/trunk/test/CodeGen/X86/peep-test-3.ll
>>    llvm/trunk/test/CodeGen/X86/pic.ll
>>    llvm/trunk/test/CodeGen/X86/sink-hoist.ll
>>    llvm/trunk/test/CodeGen/X86/sse2.ll
>>    llvm/trunk/test/CodeGen/X86/sse3.ll
>>    llvm/trunk/test/CodeGen/X86/tailcallstack64.ll
>>    llvm/trunk/test/CodeGen/X86/widen_arith-3.ll
>> 
>> Modified: llvm/trunk/lib/Target/X86/X86Subtarget.h
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Subtarget.h?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/lib/Target/X86/X86Subtarget.h (original)
>> +++ llvm/trunk/lib/Target/X86/X86Subtarget.h Sun Oct 18 14:57:27 2009
>> @@ -219,8 +219,7 @@
>>   /// enablePostRAScheduler - X86 target is enabling post-alloc scheduling
>>   /// at 'More' optimization level.
>>   bool enablePostRAScheduler(CodeGenOpt::Level OptLevel) const {
>> -    // FIXME: This causes llvm to miscompile itself on i386. :-(
>> -    return false /*OptLevel >= CodeGenOpt::Default*/;
>> +    return OptLevel >= CodeGenOpt::Default;
>>   }
>>  };
>> 
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll Sun Oct 18 14:57:27 2009
>> @@ -11,9 +11,12 @@
>>     %tmp14 = fadd float %tmp12, %tmp7
>>     ret float %tmp14
>> 
>> -; CHECK:      mulss    LCPI1_2(%rip)
>> +; CHECK:      mulss    LCPI1_3(%rip)
>> +; CHECK-NEXT: mulss    LCPI1_0(%rip)
>> +; CHECK-NEXT: mulss    LCPI1_1(%rip)
>> +; CHECK-NEXT: mulss    LCPI1_2(%rip)
>> +; CHECK-NEXT: addss
>>  ; CHECK-NEXT: addss
>> -; CHECK-NEXT: mulss    LCPI1_3(%rip)
>>  ; CHECK-NEXT: addss
>>  ; CHECK-NEXT: ret
>>  }
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll Sun Oct 18 14:57:27 2009
>> @@ -1,4 +1,4 @@
>> -; RUN: llc < %s -march=x86 -relocation-model=static -disable-fp-elim | FileCheck %s
>> +; RUN: llc < %s -march=x86 -relocation-model=static -disable-fp-elim -post-RA-scheduler=false | FileCheck %s
>>  ; PR2536
>> 
>> 
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll Sun Oct 18 14:57:27 2009
>> @@ -1,5 +1,5 @@
>>  ; RUN: llc -mtriple=i386-apple-darwin10.0 -relocation-model=pic \
>> -; RUN:     -disable-fp-elim -mattr=-sse41,-sse3,+sse2 < %s | \
>> +; RUN:     -disable-fp-elim -mattr=-sse41,-sse3,+sse2 -post-RA-scheduler=false < %s | \
>>  ; RUN:   FileCheck %s
>>  ; rdar://6808032
>> 
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll Sun Oct 18 14:57:27 2009
>> @@ -9,9 +9,7 @@
>>   br label %bb
>> 
>>  bb:                                               ; preds = %bb1, %entry
>> -; CHECK:      movl %e
>> -; CHECK-NEXT: addl $1
>> -; CHECK-NEXT: movl %e
>> +; CHECK:      addl $1
>>  ; CHECK-NEXT: adcl $0
>>   %i.0 = phi i64 [ 0, %entry ], [ %0, %bb1 ]      ; <i64> [#uses=1]
>>   %0 = add nsw i64 %i.0, 1                        ; <i64> [#uses=2]
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/abi-isel.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/abi-isel.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/abi-isel.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/abi-isel.ll Sun Oct 18 14:57:27 2009
>> @@ -1,16 +1,16 @@
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-32-STATIC
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-32-PIC
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-32-STATIC
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-32-PIC
>> 
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-64-STATIC
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=LINUX-64-PIC
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-64-STATIC
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-64-PIC
>> 
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=DARWIN-32-STATIC
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=dynamic-no-pic -code-model=small | FileCheck %s -check-prefix=DARWIN-32-DYNAMIC
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=DARWIN-32-PIC
>> -
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=DARWIN-64-STATIC
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=dynamic-no-pic -code-model=small | FileCheck %s -check-prefix=DARWIN-64-DYNAMIC
>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=DARWIN-64-PIC
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-STATIC
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=dynamic-no-pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-DYNAMIC
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-PIC
>> +
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-STATIC
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=dynamic-no-pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-DYNAMIC
>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-PIC
>> 
>>  @src = external global [131072 x i32]
>>  @dst = external global [131072 x i32]
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/fastcc.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/fastcc.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/fastcc.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/fastcc.ll Sun Oct 18 14:57:27 2009
>> @@ -1,4 +1,4 @@
>> -; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+sse2 | FileCheck %s
>> +; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+sse2 -post-RA-scheduler=false | FileCheck %s
>>  ; CHECK: movsd %xmm0, 8(%esp)
>>  ; CHECK: xorl %ecx, %ecx
>> 
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/peep-test-3.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/peep-test-3.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/peep-test-3.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/peep-test-3.ll Sun Oct 18 14:57:27 2009
>> @@ -1,4 +1,4 @@
>> -; RUN: llc < %s -march=x86 | FileCheck %s
>> +; RUN: llc < %s -march=x86 -post-RA-scheduler=false | FileCheck %s
>>  ; rdar://7226797
>> 
>>  ; LLVM should omit the testl and use the flags result from the orl.
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/pic.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pic.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/pic.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/pic.ll Sun Oct 18 14:57:27 2009
>> @@ -1,4 +1,4 @@
>> -; RUN: llc < %s -mtriple=i686-pc-linux-gnu -relocation-model=pic -asm-verbose=false | FileCheck %s -check-prefix=LINUX
>> +; RUN: llc < %s -mtriple=i686-pc-linux-gnu -relocation-model=pic -asm-verbose=false -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX
>> 
>>  @ptr = external global i32*
>>  @dst = external global i32
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/sink-hoist.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sink-hoist.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/sink-hoist.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/sink-hoist.ll Sun Oct 18 14:57:27 2009
>> @@ -6,10 +6,10 @@
>>  ; that it's conditionally evaluated.
>> 
>>  ; CHECK: foo:
>> -; CHECK-NEXT: divsd
>> -; CHECK:      testb $1, %dil
>> -; CHECK-NEXT: jne
>>  ; CHECK:      divsd
>> +; CHECK-NEXT: testb $1, %dil
>> +; CHECK-NEXT: jne
>> +; CHECK-NEXT: divsd
>> 
>>  define double @foo(double %x, double %y, i1 %c) nounwind {
>>   %a = fdiv double %x, 3.2
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/sse2.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse2.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/sse2.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/sse2.ll Sun Oct 18 14:57:27 2009
>> @@ -10,10 +10,10 @@
>> 
>>  ; CHECK: t1:
>>  ; CHECK:       movl    8(%esp), %eax
>> +; CHECK-NEXT:  movl    4(%esp), %ecx
>>  ; CHECK-NEXT:  movapd  (%eax), %xmm0
>>  ; CHECK-NEXT:  movlpd  12(%esp), %xmm0
>> -; CHECK-NEXT:  movl    4(%esp), %eax
>> -; CHECK-NEXT:  movapd  %xmm0, (%eax)
>> +; CHECK-NEXT:  movapd  %xmm0, (%ecx)
>>  ; CHECK-NEXT:  ret
>>  }
>> 
>> @@ -26,9 +26,9 @@
>> 
>>  ; CHECK: t2:
>>  ; CHECK:       movl    8(%esp), %eax
>> +; CHECK-NEXT:  movl    4(%esp), %ecx
>>  ; CHECK-NEXT:  movapd  (%eax), %xmm0
>>  ; CHECK-NEXT:  movhpd  12(%esp), %xmm0
>> -; CHECK-NEXT:  movl    4(%esp), %eax
>> -; CHECK-NEXT:  movapd  %xmm0, (%eax)
>> +; CHECK-NEXT:  movapd  %xmm0, (%ecx)
>>  ; CHECK-NEXT:  ret
>>  }
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/sse3.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse3.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/sse3.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/sse3.ll Sun Oct 18 14:57:27 2009
>> @@ -17,8 +17,8 @@
>> 
>>  ; X64: t0:
>>  ; X64:         movddup (%rsi), %xmm0
>> -; X64:  pshuflw        $0, %xmm0, %xmm0
>>  ; X64: xorl    %eax, %eax
>> +; X64:  pshuflw        $0, %xmm0, %xmm0
>>  ; X64: pinsrw  $0, %eax, %xmm0
>>  ; X64: movaps  %xmm0, (%rdi)
>>  ; X64: ret
>> @@ -167,18 +167,12 @@
>>         store <4 x i16> %6, <4 x i16>* @g2, align 8
>>         ret void
>>  ; X64:         t10:
>> -; X64:                 movq    _g1 at GOTPCREL(%rip), %rax
>> -; X64:                 movaps  (%rax), %xmm0
>>  ; X64:                 pextrw  $4, %xmm0, %eax
>> -; X64:                 movaps  %xmm0, %xmm1
>> +; X64:                 pextrw  $6, %xmm0, %edx
>>  ; X64:                 movlhps %xmm1, %xmm1
>>  ; X64:                 pshuflw $8, %xmm1, %xmm1
>>  ; X64:                 pinsrw  $2, %eax, %xmm1
>> -; X64:                 pextrw  $6, %xmm0, %eax
>> -; X64:                 pinsrw  $3, %eax, %xmm1
>> -; X64:                 movq    _g2 at GOTPCREL(%rip), %rax
>> -; X64:                 movq    %xmm1, (%rax)
>> -; X64:                 ret
>> +; X64:                 pinsrw  $3, %edx, %xmm1
>>  }
>> 
>> 
>> @@ -189,8 +183,8 @@
>>        ret <8 x i16> %tmp7
>> 
>>  ; X64: t11:
>> -; X64: movd    %xmm1, %eax
>>  ; X64: movlhps %xmm0, %xmm0
>> +; X64: movd    %xmm1, %eax
>>  ; X64: pshuflw $1, %xmm0, %xmm0
>>  ; X64: pinsrw  $1, %eax, %xmm0
>>  ; X64: ret
>> @@ -203,8 +197,8 @@
>>        ret <8 x i16> %tmp9
>> 
>>  ; X64: t12:
>> -; X64:         pextrw  $3, %xmm1, %eax
>>  ; X64:         movlhps %xmm0, %xmm0
>> +; X64:         pextrw  $3, %xmm1, %eax
>>  ; X64:         pshufhw $3, %xmm0, %xmm0
>>  ; X64:         pinsrw  $5, %eax, %xmm0
>>  ; X64:         ret
>> @@ -256,18 +250,12 @@
>>         %tmp9 = shufflevector <16 x i8> %tmp8, <16 x i8> %T0,  <16 x i32> < i32 0, i32 1, i32 2, i32 17,  i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef , i32 undef >
>>         ret <16 x i8> %tmp9
>>  ; X64:         t16:
>> -; X64:                 movaps  LCPI17_0(%rip), %xmm1
>> -; X64:                 movd    %xmm1, %eax
>>  ; X64:                 pinsrw  $0, %eax, %xmm1
>>  ; X64:                 pextrw  $8, %xmm0, %eax
>>  ; X64:                 pinsrw  $1, %eax, %xmm1
>>  ; X64:                 pextrw  $1, %xmm1, %ecx
>>  ; X64:                 movd    %xmm1, %edx
>>  ; X64:                 pinsrw  $0, %edx, %xmm1
>> -; X64:                 movzbl  %cl, %ecx
>> -; X64:                 andw    $-256, %ax
>> -; X64:                 orw     %cx, %ax
>> -; X64:                 movaps  %xmm1, %xmm0
>>  ; X64:                 pinsrw  $1, %eax, %xmm0
>>  ; X64:                 ret
>>  }
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/tailcallstack64.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/tailcallstack64.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/tailcallstack64.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/tailcallstack64.ll Sun Oct 18 14:57:27 2009
>> @@ -3,19 +3,18 @@
>>  ; Check that lowered arguments on the stack do not overwrite each other.
>>  ; Add %in1 %p1 to a different temporary register (%eax).
>>  ; CHECK: movl  %edi, %eax
>> -; CHECK: addl  32(%rsp), %eax
>>  ; Move param %in1 to temp register (%r10d).
>>  ; CHECK: movl  40(%rsp), %r10d
>> -; Move result of addition to stack.
>> -; CHECK: movl  %eax, 40(%rsp)
>>  ; Move param %in2 to stack.
>>  ; CHECK: movl  %r10d, 32(%rsp)
>> +; Move result of addition to stack.
>> +; CHECK: movl  %eax, 40(%rsp)
>>  ; Eventually, do a TAILCALL
>>  ; CHECK: TAILCALL
>> 
>> -declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %a, i32 %b)
>> +declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %a, i32 %b) nounwind
>> 
>> -define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in1, i32 %in2) {
>> +define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in1, i32 %in2) nounwind {
>>  entry:
>>         %tmp = add i32 %in1, %p1
>>         %retval = tail call fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in2,i32 %tmp)
>> 
>> Modified: llvm/trunk/test/CodeGen/X86/widen_arith-3.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/widen_arith-3.ll?rev=84431&r1=84430&r2=84431&view=diff
>> 
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/widen_arith-3.ll (original)
>> +++ llvm/trunk/test/CodeGen/X86/widen_arith-3.ll Sun Oct 18 14:57:27 2009
>> @@ -1,7 +1,7 @@
>>  ; RUN: llc < %s -march=x86 -mattr=+sse42 -disable-mmx | FileCheck %s
>>  ; CHECK: paddw
>> -; CHECK: movd
>>  ; CHECK: pextrw
>> +; CHECK: movd
>> 
>>  ; Widen a v3i16 to v8i16 to do a vector add
>> 
>> 
>> 
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> 




From deeppatel1987 at gmail.com  Thu Oct 22 21:38:21 2009
From: deeppatel1987 at gmail.com (Sandeep Patel)
Date: Fri, 23 Oct 2009 02:38:21 +0000
Subject: [llvm-commits] [llvm] r84431 - in /llvm/trunk:
	lib/Target/X86/X86Subtarget.h
	test/CodeGen/X86/2007-01-08-InstrSched.ll
	test/CodeGen/X86/2008-07-11-SpillerBug.ll
	test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
	test/CodeGen/X86/2009-09-19-SchedCustom
In-Reply-To: <C97226D4-B60A-4E1A-876B-14CC8F5E833E@apple.com>
References: <305d6f60910221904p64a00aacxe873f7f891d889d2@mail.gmail.com>
	<C97226D4-B60A-4E1A-876B-14CC8F5E833E@apple.com>
Message-ID: <305d6f60910221938w1af4b535l26ebb955805d26de@mail.gmail.com>

On Fri, Oct 23, 2009 at 2:10 AM, Evan Cheng <evan.cheng at apple.com> wrote:
>
> On Oct 22, 2009, at 7:04 PM, Sandeep Patel wrote:
>
>> This is breaking our canadian cross that runs on MinGW.
>>
>> An example of the problem from gcc.o:record_temp_file (the first
>> function in gcc.o to exhibit the problem, but not the cause of the
>> crash):
>>
>> bad code:
>> 5d: a3 14 00 00 00 ?mov ? ?%eax,0x14 ?5e: dir32 ? ? ? .bss
>> 62: 89 48 04 ? ? ? ?mov ? ?%ecx,0x4(%eax)
>> 65: 89 38 ? ? ? ? ? mov ? ?%edi,(%eax)
>>
>> good code:
>> 5d: 89 48 04 ? ? ? ?mov ? ?%ecx,0x4(%eax)
>> 60: 89 38 ? ? ? ? ? mov ? ?%edi,(%eax)
>> 62: a3 14 00 00 00 ?mov ? ?%eax,0x14 ?63: dir32 ? ? ? .bss
>
> What's wrong with the "bad code"?

Sorry, AT&T vs Intel syntax parse problem. :-)

Going to the actual function where the crash is (used_arg):

bad code:
2678: 89 e6            mov    %esp,%esi
267a: c1 e0 04         shl    $0x4,%eax
267d: e8 00 00 00 00   call   2682 <_used_arg+0x62>	267e: DISP32	__alloca

good code:
2668: c1 e0 04         shl    $0x4,%eax
266b: e8 00 00 00 00   call   2670 <_used_arg+0x60>	266c: DISP32	__alloca
2670: 89 e6            mov    %esp,%esi

deep

> You can disable it for MinGW to workaround the problem for now.
>
> Evan
>
>>
>> deep
>>
>> On Sun, Oct 18, 2009 at 7:57 PM, Evan Cheng <evan.cheng at apple.com> wrote:
>>> Author: evancheng
>>> Date: Sun Oct 18 14:57:27 2009
>>> New Revision: 84431
>>>
>>> URL: http://llvm.org/viewvc/llvm-project?rev=84431&view=rev
>>> Log:
>>> Turn on post-alloc scheduling for x86.
>>>
>>> Modified:
>>> ? ?llvm/trunk/lib/Target/X86/X86Subtarget.h
>>> ? ?llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/abi-isel.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/fastcc.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/peep-test-3.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/pic.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/sink-hoist.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/sse2.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/sse3.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/tailcallstack64.ll
>>> ? ?llvm/trunk/test/CodeGen/X86/widen_arith-3.ll
>>>
>>> Modified: llvm/trunk/lib/Target/X86/X86Subtarget.h
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Subtarget.h?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/lib/Target/X86/X86Subtarget.h (original)
>>> +++ llvm/trunk/lib/Target/X86/X86Subtarget.h Sun Oct 18 14:57:27 2009
>>> @@ -219,8 +219,7 @@
>>> ? /// enablePostRAScheduler - X86 target is enabling post-alloc scheduling
>>> ? /// at 'More' optimization level.
>>> ? bool enablePostRAScheduler(CodeGenOpt::Level OptLevel) const {
>>> - ? ?// FIXME: This causes llvm to miscompile itself on i386. :-(
>>> - ? ?return false /*OptLevel >= CodeGenOpt::Default*/;
>>> + ? ?return OptLevel >= CodeGenOpt::Default;
>>> ? }
>>> ?};
>>>
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll Sun Oct 18 14:57:27 2009
>>> @@ -11,9 +11,12 @@
>>> ? ? %tmp14 = fadd float %tmp12, %tmp7
>>> ? ? ret float %tmp14
>>>
>>> -; CHECK: ? ? ?mulss ? ?LCPI1_2(%rip)
>>> +; CHECK: ? ? ?mulss ? ?LCPI1_3(%rip)
>>> +; CHECK-NEXT: mulss ? ?LCPI1_0(%rip)
>>> +; CHECK-NEXT: mulss ? ?LCPI1_1(%rip)
>>> +; CHECK-NEXT: mulss ? ?LCPI1_2(%rip)
>>> +; CHECK-NEXT: addss
>>> ?; CHECK-NEXT: addss
>>> -; CHECK-NEXT: mulss ? ?LCPI1_3(%rip)
>>> ?; CHECK-NEXT: addss
>>> ?; CHECK-NEXT: ret
>>> ?}
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll Sun Oct 18 14:57:27 2009
>>> @@ -1,4 +1,4 @@
>>> -; RUN: llc < %s -march=x86 -relocation-model=static -disable-fp-elim | FileCheck %s
>>> +; RUN: llc < %s -march=x86 -relocation-model=static -disable-fp-elim -post-RA-scheduler=false | FileCheck %s
>>> ?; PR2536
>>>
>>>
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll Sun Oct 18 14:57:27 2009
>>> @@ -1,5 +1,5 @@
>>> ?; RUN: llc -mtriple=i386-apple-darwin10.0 -relocation-model=pic \
>>> -; RUN: ? ? -disable-fp-elim -mattr=-sse41,-sse3,+sse2 < %s | \
>>> +; RUN: ? ? -disable-fp-elim -mattr=-sse41,-sse3,+sse2 -post-RA-scheduler=false < %s | \
>>> ?; RUN: ? FileCheck %s
>>> ?; rdar://6808032
>>>
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll Sun Oct 18 14:57:27 2009
>>> @@ -9,9 +9,7 @@
>>> ? br label %bb
>>>
>>> ?bb: ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ; preds = %bb1, %entry
>>> -; CHECK: ? ? ?movl %e
>>> -; CHECK-NEXT: addl $1
>>> -; CHECK-NEXT: movl %e
>>> +; CHECK: ? ? ?addl $1
>>> ?; CHECK-NEXT: adcl $0
>>> ? %i.0 = phi i64 [ 0, %entry ], [ %0, %bb1 ] ? ? ?; <i64> [#uses=1]
>>> ? %0 = add nsw i64 %i.0, 1 ? ? ? ? ? ? ? ? ? ? ? ?; <i64> [#uses=2]
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/abi-isel.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/abi-isel.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/abi-isel.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/abi-isel.ll Sun Oct 18 14:57:27 2009
>>> @@ -1,16 +1,16 @@
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-32-STATIC
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-32-PIC
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-32-STATIC
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-32-PIC
>>>
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-64-STATIC
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=LINUX-64-PIC
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-64-STATIC
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-64-PIC
>>>
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=DARWIN-32-STATIC
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=dynamic-no-pic -code-model=small | FileCheck %s -check-prefix=DARWIN-32-DYNAMIC
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=DARWIN-32-PIC
>>> -
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=DARWIN-64-STATIC
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=dynamic-no-pic -code-model=small | FileCheck %s -check-prefix=DARWIN-64-DYNAMIC
>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=DARWIN-64-PIC
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-STATIC
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=dynamic-no-pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-DYNAMIC
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-PIC
>>> +
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-STATIC
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=dynamic-no-pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-DYNAMIC
>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-PIC
>>>
>>> ?@src = external global [131072 x i32]
>>> ?@dst = external global [131072 x i32]
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/fastcc.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/fastcc.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/fastcc.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/fastcc.ll Sun Oct 18 14:57:27 2009
>>> @@ -1,4 +1,4 @@
>>> -; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+sse2 | FileCheck %s
>>> +; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+sse2 -post-RA-scheduler=false | FileCheck %s
>>> ?; CHECK: movsd %xmm0, 8(%esp)
>>> ?; CHECK: xorl %ecx, %ecx
>>>
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/peep-test-3.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/peep-test-3.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/peep-test-3.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/peep-test-3.ll Sun Oct 18 14:57:27 2009
>>> @@ -1,4 +1,4 @@
>>> -; RUN: llc < %s -march=x86 | FileCheck %s
>>> +; RUN: llc < %s -march=x86 -post-RA-scheduler=false | FileCheck %s
>>> ?; rdar://7226797
>>>
>>> ?; LLVM should omit the testl and use the flags result from the orl.
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/pic.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pic.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/pic.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/pic.ll Sun Oct 18 14:57:27 2009
>>> @@ -1,4 +1,4 @@
>>> -; RUN: llc < %s -mtriple=i686-pc-linux-gnu -relocation-model=pic -asm-verbose=false | FileCheck %s -check-prefix=LINUX
>>> +; RUN: llc < %s -mtriple=i686-pc-linux-gnu -relocation-model=pic -asm-verbose=false -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX
>>>
>>> ?@ptr = external global i32*
>>> ?@dst = external global i32
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/sink-hoist.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sink-hoist.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/sink-hoist.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/sink-hoist.ll Sun Oct 18 14:57:27 2009
>>> @@ -6,10 +6,10 @@
>>> ?; that it's conditionally evaluated.
>>>
>>> ?; CHECK: foo:
>>> -; CHECK-NEXT: divsd
>>> -; CHECK: ? ? ?testb $1, %dil
>>> -; CHECK-NEXT: jne
>>> ?; CHECK: ? ? ?divsd
>>> +; CHECK-NEXT: testb $1, %dil
>>> +; CHECK-NEXT: jne
>>> +; CHECK-NEXT: divsd
>>>
>>> ?define double @foo(double %x, double %y, i1 %c) nounwind {
>>> ? %a = fdiv double %x, 3.2
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/sse2.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse2.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/sse2.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/sse2.ll Sun Oct 18 14:57:27 2009
>>> @@ -10,10 +10,10 @@
>>>
>>> ?; CHECK: t1:
>>> ?; CHECK: ? ? ? movl ? ?8(%esp), %eax
>>> +; CHECK-NEXT: ?movl ? ?4(%esp), %ecx
>>> ?; CHECK-NEXT: ?movapd ?(%eax), %xmm0
>>> ?; CHECK-NEXT: ?movlpd ?12(%esp), %xmm0
>>> -; CHECK-NEXT: ?movl ? ?4(%esp), %eax
>>> -; CHECK-NEXT: ?movapd ?%xmm0, (%eax)
>>> +; CHECK-NEXT: ?movapd ?%xmm0, (%ecx)
>>> ?; CHECK-NEXT: ?ret
>>> ?}
>>>
>>> @@ -26,9 +26,9 @@
>>>
>>> ?; CHECK: t2:
>>> ?; CHECK: ? ? ? movl ? ?8(%esp), %eax
>>> +; CHECK-NEXT: ?movl ? ?4(%esp), %ecx
>>> ?; CHECK-NEXT: ?movapd ?(%eax), %xmm0
>>> ?; CHECK-NEXT: ?movhpd ?12(%esp), %xmm0
>>> -; CHECK-NEXT: ?movl ? ?4(%esp), %eax
>>> -; CHECK-NEXT: ?movapd ?%xmm0, (%eax)
>>> +; CHECK-NEXT: ?movapd ?%xmm0, (%ecx)
>>> ?; CHECK-NEXT: ?ret
>>> ?}
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/sse3.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse3.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/sse3.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/sse3.ll Sun Oct 18 14:57:27 2009
>>> @@ -17,8 +17,8 @@
>>>
>>> ?; X64: t0:
>>> ?; X64: ? ? ? ? movddup (%rsi), %xmm0
>>> -; X64: ?pshuflw ? ? ? ?$0, %xmm0, %xmm0
>>> ?; X64: xorl ? ?%eax, %eax
>>> +; X64: ?pshuflw ? ? ? ?$0, %xmm0, %xmm0
>>> ?; X64: pinsrw ?$0, %eax, %xmm0
>>> ?; X64: movaps ?%xmm0, (%rdi)
>>> ?; X64: ret
>>> @@ -167,18 +167,12 @@
>>> ? ? ? ? store <4 x i16> %6, <4 x i16>* @g2, align 8
>>> ? ? ? ? ret void
>>> ?; X64: ? ? ? ? t10:
>>> -; X64: ? ? ? ? ? ? ? ? movq ? ?_g1 at GOTPCREL(%rip), %rax
>>> -; X64: ? ? ? ? ? ? ? ? movaps ?(%rax), %xmm0
>>> ?; X64: ? ? ? ? ? ? ? ? pextrw ?$4, %xmm0, %eax
>>> -; X64: ? ? ? ? ? ? ? ? movaps ?%xmm0, %xmm1
>>> +; X64: ? ? ? ? ? ? ? ? pextrw ?$6, %xmm0, %edx
>>> ?; X64: ? ? ? ? ? ? ? ? movlhps %xmm1, %xmm1
>>> ?; X64: ? ? ? ? ? ? ? ? pshuflw $8, %xmm1, %xmm1
>>> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$2, %eax, %xmm1
>>> -; X64: ? ? ? ? ? ? ? ? pextrw ?$6, %xmm0, %eax
>>> -; X64: ? ? ? ? ? ? ? ? pinsrw ?$3, %eax, %xmm1
>>> -; X64: ? ? ? ? ? ? ? ? movq ? ?_g2 at GOTPCREL(%rip), %rax
>>> -; X64: ? ? ? ? ? ? ? ? movq ? ?%xmm1, (%rax)
>>> -; X64: ? ? ? ? ? ? ? ? ret
>>> +; X64: ? ? ? ? ? ? ? ? pinsrw ?$3, %edx, %xmm1
>>> ?}
>>>
>>>
>>> @@ -189,8 +183,8 @@
>>> ? ? ? ?ret <8 x i16> %tmp7
>>>
>>> ?; X64: t11:
>>> -; X64: movd ? ?%xmm1, %eax
>>> ?; X64: movlhps %xmm0, %xmm0
>>> +; X64: movd ? ?%xmm1, %eax
>>> ?; X64: pshuflw $1, %xmm0, %xmm0
>>> ?; X64: pinsrw ?$1, %eax, %xmm0
>>> ?; X64: ret
>>> @@ -203,8 +197,8 @@
>>> ? ? ? ?ret <8 x i16> %tmp9
>>>
>>> ?; X64: t12:
>>> -; X64: ? ? ? ? pextrw ?$3, %xmm1, %eax
>>> ?; X64: ? ? ? ? movlhps %xmm0, %xmm0
>>> +; X64: ? ? ? ? pextrw ?$3, %xmm1, %eax
>>> ?; X64: ? ? ? ? pshufhw $3, %xmm0, %xmm0
>>> ?; X64: ? ? ? ? pinsrw ?$5, %eax, %xmm0
>>> ?; X64: ? ? ? ? ret
>>> @@ -256,18 +250,12 @@
>>> ? ? ? ? %tmp9 = shufflevector <16 x i8> %tmp8, <16 x i8> %T0, ?<16 x i32> < i32 0, i32 1, i32 2, i32 17, ?i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef , i32 undef >
>>> ? ? ? ? ret <16 x i8> %tmp9
>>> ?; X64: ? ? ? ? t16:
>>> -; X64: ? ? ? ? ? ? ? ? movaps ?LCPI17_0(%rip), %xmm1
>>> -; X64: ? ? ? ? ? ? ? ? movd ? ?%xmm1, %eax
>>> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$0, %eax, %xmm1
>>> ?; X64: ? ? ? ? ? ? ? ? pextrw ?$8, %xmm0, %eax
>>> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$1, %eax, %xmm1
>>> ?; X64: ? ? ? ? ? ? ? ? pextrw ?$1, %xmm1, %ecx
>>> ?; X64: ? ? ? ? ? ? ? ? movd ? ?%xmm1, %edx
>>> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$0, %edx, %xmm1
>>> -; X64: ? ? ? ? ? ? ? ? movzbl ?%cl, %ecx
>>> -; X64: ? ? ? ? ? ? ? ? andw ? ?$-256, %ax
>>> -; X64: ? ? ? ? ? ? ? ? orw ? ? %cx, %ax
>>> -; X64: ? ? ? ? ? ? ? ? movaps ?%xmm1, %xmm0
>>> ?; X64: ? ? ? ? ? ? ? ? pinsrw ?$1, %eax, %xmm0
>>> ?; X64: ? ? ? ? ? ? ? ? ret
>>> ?}
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/tailcallstack64.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/tailcallstack64.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/tailcallstack64.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/tailcallstack64.ll Sun Oct 18 14:57:27 2009
>>> @@ -3,19 +3,18 @@
>>> ?; Check that lowered arguments on the stack do not overwrite each other.
>>> ?; Add %in1 %p1 to a different temporary register (%eax).
>>> ?; CHECK: movl ?%edi, %eax
>>> -; CHECK: addl ?32(%rsp), %eax
>>> ?; Move param %in1 to temp register (%r10d).
>>> ?; CHECK: movl ?40(%rsp), %r10d
>>> -; Move result of addition to stack.
>>> -; CHECK: movl ?%eax, 40(%rsp)
>>> ?; Move param %in2 to stack.
>>> ?; CHECK: movl ?%r10d, 32(%rsp)
>>> +; Move result of addition to stack.
>>> +; CHECK: movl ?%eax, 40(%rsp)
>>> ?; Eventually, do a TAILCALL
>>> ?; CHECK: TAILCALL
>>>
>>> -declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %a, i32 %b)
>>> +declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %a, i32 %b) nounwind
>>>
>>> -define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in1, i32 %in2) {
>>> +define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in1, i32 %in2) nounwind {
>>> ?entry:
>>> ? ? ? ? %tmp = add i32 %in1, %p1
>>> ? ? ? ? %retval = tail call fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in2,i32 %tmp)
>>>
>>> Modified: llvm/trunk/test/CodeGen/X86/widen_arith-3.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/widen_arith-3.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>
>>> ==============================================================================
>>> --- llvm/trunk/test/CodeGen/X86/widen_arith-3.ll (original)
>>> +++ llvm/trunk/test/CodeGen/X86/widen_arith-3.ll Sun Oct 18 14:57:27 2009
>>> @@ -1,7 +1,7 @@
>>> ?; RUN: llc < %s -march=x86 -mattr=+sse42 -disable-mmx | FileCheck %s
>>> ?; CHECK: paddw
>>> -; CHECK: movd
>>> ?; CHECK: pextrw
>>> +; CHECK: movd
>>>
>>> ?; Widen a v3i16 to v8i16 to do a vector add
>>>
>>>
>>>
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>
>
>



From nicholas at mxc.ca  Thu Oct 22 22:33:25 2009
From: nicholas at mxc.ca (Nick Lewycky)
Date: Thu, 22 Oct 2009 20:33:25 -0700
Subject: [llvm-commits] [PATCH] Fix to pass options from Gold plugin to
 LTO	codegen
In-Reply-To: <7F9FE9028F544389BE51F80E181E6B0C@andreic6e7fe55>
References: <3FE2DECEF38F9247AE7C8BC1FD2B015F0817022C@ALLISON>
	<7F9FE9028F544389BE51F80E181E6B0C@andreic6e7fe55>
Message-ID: <4AE12405.9010502@mxc.ca>

Viktor Kutuzov wrote:
> Hello everyone,
> 
> Please find the patch attached.
> This should fix the issue when gold plugin doesn't pass the  -plugin-opt 
> options down to the LTO codegen.
> See the thread "[LLVMdev] getting gold plugin to work?" for more details.

What's the difference between setCodeGenDebugOptions and 
setCodeGenOption? It looks like they should be merged into one then 
exposed by lto_codegen_set_option. And that would be even better if it 
took an argc+argv pair, even if that's a little harder to construct from 
the gold plugin, many option parsing packages work by removing the 
options they recognize from argc+argv and leaving the rest in place.

Nick


From dalej at apple.com  Thu Oct 22 23:02:51 2009
From: dalej at apple.com (Dale Johannesen)
Date: Fri, 23 Oct 2009 04:02:51 -0000
Subject: [llvm-commits] [llvm] r84928 - /llvm/trunk/lib/Support/APFloat.cpp
Message-ID: <200910230402.n9N42psH029166@zion.cs.uiuc.edu>

Author: johannes
Date: Thu Oct 22 23:02:51 2009
New Revision: 84928

URL: http://llvm.org/viewvc/llvm-project?rev=84928&view=rev
Log:
Commit fixes for half precision I noted in review, so
they don't get lost; I don't think the originator has
write access.


Modified:
    llvm/trunk/lib/Support/APFloat.cpp

Modified: llvm/trunk/lib/Support/APFloat.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Support/APFloat.cpp?rev=84928&r1=84927&r2=84928&view=diff

==============================================================================
--- llvm/trunk/lib/Support/APFloat.cpp (original)
+++ llvm/trunk/lib/Support/APFloat.cpp Thu Oct 22 23:02:51 2009
@@ -2830,11 +2830,11 @@
     myexponent = 0;
     mysignificand = 0;
   } else if (category==fcInfinity) {
-    myexponent = 0xff;
+    myexponent = 0x1f;
     mysignificand = 0;
   } else {
     assert(category == fcNaN && "Unknown category!");
-    myexponent = 0xff;
+    myexponent = 0x1f;
     mysignificand = (uint32_t)*significandParts();
   }
 
@@ -3089,7 +3089,7 @@
 {
   assert(api.getBitWidth()==16);
   uint32_t i = (uint32_t)*api.getRawData();
-  uint32_t myexponent = (i >> 15) & 0x1f;
+  uint32_t myexponent = (i >> 10) & 0x1f;
   uint32_t mysignificand = i & 0x3ff;
 
   initialize(&APFloat::IEEEhalf);




From baldrick at free.fr  Thu Oct 22 23:05:54 2009
From: baldrick at free.fr (Duncan Sands)
Date: Fri, 23 Oct 2009 06:05:54 +0200
Subject: [llvm-commits] [llvm] r84900
	-	/llvm/trunk/include/llvm/ADT/StringRef.h
In-Reply-To: <200910222206.n9MM6oVI030595@zion.cs.uiuc.edu>
References: <200910222206.n9MM6oVI030595@zion.cs.uiuc.edu>
Message-ID: <4AE12BA2.8020503@free.fr>

Hi Eric,

> size_t, not unsigned here to silence a warning.

while I'm all for white space cleanups, can you please not mix them
in with real changes since it makes it harder to see what the real
changes were.

Thanks,

Duncan.


From baldrick at free.fr  Thu Oct 22 23:49:29 2009
From: baldrick at free.fr (Duncan Sands)
Date: Fri, 23 Oct 2009 06:49:29 +0200
Subject: [llvm-commits] [llvm-gcc-4.2] r84829	-
	/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
In-Reply-To: <AB20C91E-8E2F-45C8-8A89-1B4E4ED8D165@apple.com>
References: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>
	<4AE0282A.5080300@free.fr>
	<FE66EE67-889E-47F2-ABA9-FC1C429E542A@apple.com>
	<4AE0895E.8000900@free.fr>
	<AB20C91E-8E2F-45C8-8A89-1B4E4ED8D165@apple.com>
Message-ID: <4AE135D9.7050304@free.fr>

Hi Chris,

>>> The two fixes take a different approach (as far as I understand it).  
>>> The major difference is that I want the warning from instcombine when 
>>> you do this in C:
>>> void foo() {.}
>>> ... foo(4);  ...
>>
>> gcc doesn't warn about this, even gcc-4.4, so why should llvm-gcc?  
>> And if you
>> really want llvm-gcc to warn about it, why not add the logic for that 
>> to the
>> C frontend?
> 
> This warning is most useful at LTO time.  I agree that the front-end 
> should also catch this in reasonable situations within a translation 
> unit, even though GCC doesn't.

so would you be happy with Rafael's fix (assuming it works correctly)
plus a new PR asking for llvm-gcc to emit a warning in the C case (to
be implemented in the C frontend)?

Ciao,

Duncan.


From clattner at apple.com  Fri Oct 23 00:01:23 2009
From: clattner at apple.com (Chris Lattner)
Date: Thu, 22 Oct 2009 22:01:23 -0700
Subject: [llvm-commits] [llvm-gcc-4.2] r84829	-
	/llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp
In-Reply-To: <4AE135D9.7050304@free.fr>
References: <200910220447.n9M4l7QJ027132@zion.cs.uiuc.edu>
	<4AE0282A.5080300@free.fr>
	<FE66EE67-889E-47F2-ABA9-FC1C429E542A@apple.com>
	<4AE0895E.8000900@free.fr>
	<AB20C91E-8E2F-45C8-8A89-1B4E4ED8D165@apple.com>
	<4AE135D9.7050304@free.fr>
Message-ID: <CD2774F5-FE3D-41A4-B101-3B30F1EDA055@apple.com>


On Oct 22, 2009, at 9:49 PM, Duncan Sands wrote:

> Hi Chris,
>
>>>> The two fixes take a different approach (as far as I understand  
>>>> it).  The major difference is that I want the warning from  
>>>> instcombine when you do this in C:
>>>> void foo() {.}
>>>> ... foo(4);  ...
>>>
>>> gcc doesn't warn about this, even gcc-4.4, so why should llvm- 
>>> gcc?  And if you
>>> really want llvm-gcc to warn about it, why not add the logic for  
>>> that to the
>>> C frontend?
>> This warning is most useful at LTO time.  I agree that the front- 
>> end should also catch this in reasonable situations within a  
>> translation unit, even though GCC doesn't.
>
> so would you be happy with Rafael's fix (assuming it works correctly)
> plus a new PR asking for llvm-gcc to emit a warning in the C case (to
> be implemented in the C frontend)?

Until someone actually adds the warning, I like my fix better.  A PR  
is less useful than something that works today :)

-Chris


From evan.cheng at apple.com  Fri Oct 23 00:35:18 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Thu, 22 Oct 2009 22:35:18 -0700
Subject: [llvm-commits] [llvm] r84431 - in /llvm/trunk:
	lib/Target/X86/X86Subtarget.h
	test/CodeGen/X86/2007-01-08-InstrSched.ll
	test/CodeGen/X86/2008-07-11-SpillerBug.ll
	test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
	test/CodeGen/X86/2009-09-19-SchedCustom
In-Reply-To: <305d6f60910221938w1af4b535l26ebb955805d26de@mail.gmail.com>
References: <305d6f60910221904p64a00aacxe873f7f891d889d2@mail.gmail.com>
	<C97226D4-B60A-4E1A-876B-14CC8F5E833E@apple.com>
	<305d6f60910221938w1af4b535l26ebb955805d26de@mail.gmail.com>
Message-ID: <8C108567-FD07-4F66-986E-CD9FC70092C7@apple.com>


On Oct 22, 2009, at 7:38 PM, Sandeep Patel wrote:

> On Fri, Oct 23, 2009 at 2:10 AM, Evan Cheng <evan.cheng at apple.com> wrote:
>> 
>> On Oct 22, 2009, at 7:04 PM, Sandeep Patel wrote:
>> 
>>> This is breaking our canadian cross that runs on MinGW.
>>> 
>>> An example of the problem from gcc.o:record_temp_file (the first
>>> function in gcc.o to exhibit the problem, but not the cause of the
>>> crash):
>>> 
>>> bad code:
>>> 5d: a3 14 00 00 00  mov    %eax,0x14  5e: dir32       .bss
>>> 62: 89 48 04        mov    %ecx,0x4(%eax)
>>> 65: 89 38           mov    %edi,(%eax)
>>> 
>>> good code:
>>> 5d: 89 48 04        mov    %ecx,0x4(%eax)
>>> 60: 89 38           mov    %edi,(%eax)
>>> 62: a3 14 00 00 00  mov    %eax,0x14  63: dir32       .bss
>> 
>> What's wrong with the "bad code"?
> 
> Sorry, AT&T vs Intel syntax parse problem. :-)
> 
> Going to the actual function where the crash is (used_arg):
> 
> bad code:
> 2678: 89 e6            mov    %esp,%esi
> 267a: c1 e0 04         shl    $0x4,%eax
> 267d: e8 00 00 00 00   call   2682 <_used_arg+0x62>	267e: DISP32	__alloca
> 
> good code:
> 2668: c1 e0 04         shl    $0x4,%eax
> 266b: e8 00 00 00 00   call   2670 <_used_arg+0x60>	266c: DISP32	__alloca
> 2670: 89 e6            mov    %esp,%esi
> 
> deep

Please file a bug with bitcode test cases.

Evan

> 
>> You can disable it for MinGW to workaround the problem for now.
>> 
>> Evan
>> 
>>> 
>>> deep
>>> 
>>> On Sun, Oct 18, 2009 at 7:57 PM, Evan Cheng <evan.cheng at apple.com> wrote:
>>>> Author: evancheng
>>>> Date: Sun Oct 18 14:57:27 2009
>>>> New Revision: 84431
>>>> 
>>>> URL: http://llvm.org/viewvc/llvm-project?rev=84431&view=rev
>>>> Log:
>>>> Turn on post-alloc scheduling for x86.
>>>> 
>>>> Modified:
>>>>    llvm/trunk/lib/Target/X86/X86Subtarget.h
>>>>    llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
>>>>    llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll
>>>>    llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
>>>>    llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll
>>>>    llvm/trunk/test/CodeGen/X86/abi-isel.ll
>>>>    llvm/trunk/test/CodeGen/X86/fastcc.ll
>>>>    llvm/trunk/test/CodeGen/X86/peep-test-3.ll
>>>>    llvm/trunk/test/CodeGen/X86/pic.ll
>>>>    llvm/trunk/test/CodeGen/X86/sink-hoist.ll
>>>>    llvm/trunk/test/CodeGen/X86/sse2.ll
>>>>    llvm/trunk/test/CodeGen/X86/sse3.ll
>>>>    llvm/trunk/test/CodeGen/X86/tailcallstack64.ll
>>>>    llvm/trunk/test/CodeGen/X86/widen_arith-3.ll
>>>> 
>>>> Modified: llvm/trunk/lib/Target/X86/X86Subtarget.h
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Subtarget.h?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/lib/Target/X86/X86Subtarget.h (original)
>>>> +++ llvm/trunk/lib/Target/X86/X86Subtarget.h Sun Oct 18 14:57:27 2009
>>>> @@ -219,8 +219,7 @@
>>>>   /// enablePostRAScheduler - X86 target is enabling post-alloc scheduling
>>>>   /// at 'More' optimization level.
>>>>   bool enablePostRAScheduler(CodeGenOpt::Level OptLevel) const {
>>>> -    // FIXME: This causes llvm to miscompile itself on i386. :-(
>>>> -    return false /*OptLevel >= CodeGenOpt::Default*/;
>>>> +    return OptLevel >= CodeGenOpt::Default;
>>>>   }
>>>>  };
>>>> 
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll Sun Oct 18 14:57:27 2009
>>>> @@ -11,9 +11,12 @@
>>>>     %tmp14 = fadd float %tmp12, %tmp7
>>>>     ret float %tmp14
>>>> 
>>>> -; CHECK:      mulss    LCPI1_2(%rip)
>>>> +; CHECK:      mulss    LCPI1_3(%rip)
>>>> +; CHECK-NEXT: mulss    LCPI1_0(%rip)
>>>> +; CHECK-NEXT: mulss    LCPI1_1(%rip)
>>>> +; CHECK-NEXT: mulss    LCPI1_2(%rip)
>>>> +; CHECK-NEXT: addss
>>>>  ; CHECK-NEXT: addss
>>>> -; CHECK-NEXT: mulss    LCPI1_3(%rip)
>>>>  ; CHECK-NEXT: addss
>>>>  ; CHECK-NEXT: ret
>>>>  }
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/2008-07-11-SpillerBug.ll Sun Oct 18 14:57:27 2009
>>>> @@ -1,4 +1,4 @@
>>>> -; RUN: llc < %s -march=x86 -relocation-model=static -disable-fp-elim | FileCheck %s
>>>> +; RUN: llc < %s -march=x86 -relocation-model=static -disable-fp-elim -post-RA-scheduler=false | FileCheck %s
>>>>  ; PR2536
>>>> 
>>>> 
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/2009-04-21-NoReloadImpDef.ll Sun Oct 18 14:57:27 2009
>>>> @@ -1,5 +1,5 @@
>>>>  ; RUN: llc -mtriple=i386-apple-darwin10.0 -relocation-model=pic \
>>>> -; RUN:     -disable-fp-elim -mattr=-sse41,-sse3,+sse2 < %s | \
>>>> +; RUN:     -disable-fp-elim -mattr=-sse41,-sse3,+sse2 -post-RA-scheduler=false < %s | \
>>>>  ; RUN:   FileCheck %s
>>>>  ; rdar://6808032
>>>> 
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll Sun Oct 18 14:57:27 2009
>>>> @@ -9,9 +9,7 @@
>>>>   br label %bb
>>>> 
>>>>  bb:                                               ; preds = %bb1, %entry
>>>> -; CHECK:      movl %e
>>>> -; CHECK-NEXT: addl $1
>>>> -; CHECK-NEXT: movl %e
>>>> +; CHECK:      addl $1
>>>>  ; CHECK-NEXT: adcl $0
>>>>   %i.0 = phi i64 [ 0, %entry ], [ %0, %bb1 ]      ; <i64> [#uses=1]
>>>>   %0 = add nsw i64 %i.0, 1                        ; <i64> [#uses=2]
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/abi-isel.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/abi-isel.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/abi-isel.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/abi-isel.ll Sun Oct 18 14:57:27 2009
>>>> @@ -1,16 +1,16 @@
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-32-STATIC
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-32-PIC
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-32-STATIC
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-unknown-linux-gnu -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-32-PIC
>>>> 
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=LINUX-64-STATIC
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=LINUX-64-PIC
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-64-STATIC
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-unknown-linux-gnu -march=x86-64 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX-64-PIC
>>>> 
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=DARWIN-32-STATIC
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=dynamic-no-pic -code-model=small | FileCheck %s -check-prefix=DARWIN-32-DYNAMIC
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=DARWIN-32-PIC
>>>> -
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=static -code-model=small | FileCheck %s -check-prefix=DARWIN-64-STATIC
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=dynamic-no-pic -code-model=small | FileCheck %s -check-prefix=DARWIN-64-DYNAMIC
>>>> -; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=pic -code-model=small | FileCheck %s -check-prefix=DARWIN-64-PIC
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-STATIC
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=dynamic-no-pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-DYNAMIC
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=i686-apple-darwin -march=x86 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-32-PIC
>>>> +
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=static -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-STATIC
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=dynamic-no-pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-DYNAMIC
>>>> +; RUN: llc < %s -asm-verbose=0 -mtriple=x86_64-apple-darwin -march=x86-64 -relocation-model=pic -code-model=small -post-RA-scheduler=false | FileCheck %s -check-prefix=DARWIN-64-PIC
>>>> 
>>>>  @src = external global [131072 x i32]
>>>>  @dst = external global [131072 x i32]
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/fastcc.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/fastcc.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/fastcc.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/fastcc.ll Sun Oct 18 14:57:27 2009
>>>> @@ -1,4 +1,4 @@
>>>> -; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+sse2 | FileCheck %s
>>>> +; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+sse2 -post-RA-scheduler=false | FileCheck %s
>>>>  ; CHECK: movsd %xmm0, 8(%esp)
>>>>  ; CHECK: xorl %ecx, %ecx
>>>> 
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/peep-test-3.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/peep-test-3.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/peep-test-3.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/peep-test-3.ll Sun Oct 18 14:57:27 2009
>>>> @@ -1,4 +1,4 @@
>>>> -; RUN: llc < %s -march=x86 | FileCheck %s
>>>> +; RUN: llc < %s -march=x86 -post-RA-scheduler=false | FileCheck %s
>>>>  ; rdar://7226797
>>>> 
>>>>  ; LLVM should omit the testl and use the flags result from the orl.
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/pic.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pic.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/pic.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/pic.ll Sun Oct 18 14:57:27 2009
>>>> @@ -1,4 +1,4 @@
>>>> -; RUN: llc < %s -mtriple=i686-pc-linux-gnu -relocation-model=pic -asm-verbose=false | FileCheck %s -check-prefix=LINUX
>>>> +; RUN: llc < %s -mtriple=i686-pc-linux-gnu -relocation-model=pic -asm-verbose=false -post-RA-scheduler=false | FileCheck %s -check-prefix=LINUX
>>>> 
>>>>  @ptr = external global i32*
>>>>  @dst = external global i32
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/sink-hoist.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sink-hoist.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/sink-hoist.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/sink-hoist.ll Sun Oct 18 14:57:27 2009
>>>> @@ -6,10 +6,10 @@
>>>>  ; that it's conditionally evaluated.
>>>> 
>>>>  ; CHECK: foo:
>>>> -; CHECK-NEXT: divsd
>>>> -; CHECK:      testb $1, %dil
>>>> -; CHECK-NEXT: jne
>>>>  ; CHECK:      divsd
>>>> +; CHECK-NEXT: testb $1, %dil
>>>> +; CHECK-NEXT: jne
>>>> +; CHECK-NEXT: divsd
>>>> 
>>>>  define double @foo(double %x, double %y, i1 %c) nounwind {
>>>>   %a = fdiv double %x, 3.2
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/sse2.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse2.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/sse2.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/sse2.ll Sun Oct 18 14:57:27 2009
>>>> @@ -10,10 +10,10 @@
>>>> 
>>>>  ; CHECK: t1:
>>>>  ; CHECK:       movl    8(%esp), %eax
>>>> +; CHECK-NEXT:  movl    4(%esp), %ecx
>>>>  ; CHECK-NEXT:  movapd  (%eax), %xmm0
>>>>  ; CHECK-NEXT:  movlpd  12(%esp), %xmm0
>>>> -; CHECK-NEXT:  movl    4(%esp), %eax
>>>> -; CHECK-NEXT:  movapd  %xmm0, (%eax)
>>>> +; CHECK-NEXT:  movapd  %xmm0, (%ecx)
>>>>  ; CHECK-NEXT:  ret
>>>>  }
>>>> 
>>>> @@ -26,9 +26,9 @@
>>>> 
>>>>  ; CHECK: t2:
>>>>  ; CHECK:       movl    8(%esp), %eax
>>>> +; CHECK-NEXT:  movl    4(%esp), %ecx
>>>>  ; CHECK-NEXT:  movapd  (%eax), %xmm0
>>>>  ; CHECK-NEXT:  movhpd  12(%esp), %xmm0
>>>> -; CHECK-NEXT:  movl    4(%esp), %eax
>>>> -; CHECK-NEXT:  movapd  %xmm0, (%eax)
>>>> +; CHECK-NEXT:  movapd  %xmm0, (%ecx)
>>>>  ; CHECK-NEXT:  ret
>>>>  }
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/sse3.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse3.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/sse3.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/sse3.ll Sun Oct 18 14:57:27 2009
>>>> @@ -17,8 +17,8 @@
>>>> 
>>>>  ; X64: t0:
>>>>  ; X64:         movddup (%rsi), %xmm0
>>>> -; X64:  pshuflw        $0, %xmm0, %xmm0
>>>>  ; X64: xorl    %eax, %eax
>>>> +; X64:  pshuflw        $0, %xmm0, %xmm0
>>>>  ; X64: pinsrw  $0, %eax, %xmm0
>>>>  ; X64: movaps  %xmm0, (%rdi)
>>>>  ; X64: ret
>>>> @@ -167,18 +167,12 @@
>>>>         store <4 x i16> %6, <4 x i16>* @g2, align 8
>>>>         ret void
>>>>  ; X64:         t10:
>>>> -; X64:                 movq    _g1 at GOTPCREL(%rip), %rax
>>>> -; X64:                 movaps  (%rax), %xmm0
>>>>  ; X64:                 pextrw  $4, %xmm0, %eax
>>>> -; X64:                 movaps  %xmm0, %xmm1
>>>> +; X64:                 pextrw  $6, %xmm0, %edx
>>>>  ; X64:                 movlhps %xmm1, %xmm1
>>>>  ; X64:                 pshuflw $8, %xmm1, %xmm1
>>>>  ; X64:                 pinsrw  $2, %eax, %xmm1
>>>> -; X64:                 pextrw  $6, %xmm0, %eax
>>>> -; X64:                 pinsrw  $3, %eax, %xmm1
>>>> -; X64:                 movq    _g2 at GOTPCREL(%rip), %rax
>>>> -; X64:                 movq    %xmm1, (%rax)
>>>> -; X64:                 ret
>>>> +; X64:                 pinsrw  $3, %edx, %xmm1
>>>>  }
>>>> 
>>>> 
>>>> @@ -189,8 +183,8 @@
>>>>        ret <8 x i16> %tmp7
>>>> 
>>>>  ; X64: t11:
>>>> -; X64: movd    %xmm1, %eax
>>>>  ; X64: movlhps %xmm0, %xmm0
>>>> +; X64: movd    %xmm1, %eax
>>>>  ; X64: pshuflw $1, %xmm0, %xmm0
>>>>  ; X64: pinsrw  $1, %eax, %xmm0
>>>>  ; X64: ret
>>>> @@ -203,8 +197,8 @@
>>>>        ret <8 x i16> %tmp9
>>>> 
>>>>  ; X64: t12:
>>>> -; X64:         pextrw  $3, %xmm1, %eax
>>>>  ; X64:         movlhps %xmm0, %xmm0
>>>> +; X64:         pextrw  $3, %xmm1, %eax
>>>>  ; X64:         pshufhw $3, %xmm0, %xmm0
>>>>  ; X64:         pinsrw  $5, %eax, %xmm0
>>>>  ; X64:         ret
>>>> @@ -256,18 +250,12 @@
>>>>         %tmp9 = shufflevector <16 x i8> %tmp8, <16 x i8> %T0,  <16 x i32> < i32 0, i32 1, i32 2, i32 17,  i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef , i32 undef >
>>>>         ret <16 x i8> %tmp9
>>>>  ; X64:         t16:
>>>> -; X64:                 movaps  LCPI17_0(%rip), %xmm1
>>>> -; X64:                 movd    %xmm1, %eax
>>>>  ; X64:                 pinsrw  $0, %eax, %xmm1
>>>>  ; X64:                 pextrw  $8, %xmm0, %eax
>>>>  ; X64:                 pinsrw  $1, %eax, %xmm1
>>>>  ; X64:                 pextrw  $1, %xmm1, %ecx
>>>>  ; X64:                 movd    %xmm1, %edx
>>>>  ; X64:                 pinsrw  $0, %edx, %xmm1
>>>> -; X64:                 movzbl  %cl, %ecx
>>>> -; X64:                 andw    $-256, %ax
>>>> -; X64:                 orw     %cx, %ax
>>>> -; X64:                 movaps  %xmm1, %xmm0
>>>>  ; X64:                 pinsrw  $1, %eax, %xmm0
>>>>  ; X64:                 ret
>>>>  }
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/tailcallstack64.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/tailcallstack64.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/tailcallstack64.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/tailcallstack64.ll Sun Oct 18 14:57:27 2009
>>>> @@ -3,19 +3,18 @@
>>>>  ; Check that lowered arguments on the stack do not overwrite each other.
>>>>  ; Add %in1 %p1 to a different temporary register (%eax).
>>>>  ; CHECK: movl  %edi, %eax
>>>> -; CHECK: addl  32(%rsp), %eax
>>>>  ; Move param %in1 to temp register (%r10d).
>>>>  ; CHECK: movl  40(%rsp), %r10d
>>>> -; Move result of addition to stack.
>>>> -; CHECK: movl  %eax, 40(%rsp)
>>>>  ; Move param %in2 to stack.
>>>>  ; CHECK: movl  %r10d, 32(%rsp)
>>>> +; Move result of addition to stack.
>>>> +; CHECK: movl  %eax, 40(%rsp)
>>>>  ; Eventually, do a TAILCALL
>>>>  ; CHECK: TAILCALL
>>>> 
>>>> -declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %a, i32 %b)
>>>> +declare fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %a, i32 %b) nounwind
>>>> 
>>>> -define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in1, i32 %in2) {
>>>> +define fastcc i32 @tailcaller(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in1, i32 %in2) nounwind {
>>>>  entry:
>>>>         %tmp = add i32 %in1, %p1
>>>>         %retval = tail call fastcc i32 @tailcallee(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5, i32 %p6, i32 %in2,i32 %tmp)
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/widen_arith-3.ll
>>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/widen_arith-3.ll?rev=84431&r1=84430&r2=84431&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/widen_arith-3.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/widen_arith-3.ll Sun Oct 18 14:57:27 2009
>>>> @@ -1,7 +1,7 @@
>>>>  ; RUN: llc < %s -march=x86 -mattr=+sse42 -disable-mmx | FileCheck %s
>>>>  ; CHECK: paddw
>>>> -; CHECK: movd
>>>>  ; CHECK: pextrw
>>>> +; CHECK: movd
>>>> 
>>>>  ; Widen a v3i16 to v8i16 to do a vector add
>>>> 
>>>> 
>>>> 
>>>> _______________________________________________
>>>> llvm-commits mailing list
>>>> llvm-commits at cs.uiuc.edu
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>> 
>> 
>> 




From evan.cheng at apple.com  Fri Oct 23 00:57:36 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Fri, 23 Oct 2009 05:57:36 -0000
Subject: [llvm-commits] [llvm] r84931 -
	/llvm/trunk/lib/Target/X86/X86Subtarget.h
Message-ID: <200910230557.n9N5vaOD003929@zion.cs.uiuc.edu>

Author: evancheng
Date: Fri Oct 23 00:57:35 2009
New Revision: 84931

URL: http://llvm.org/viewvc/llvm-project?rev=84931&view=rev
Log:
X86 needs critical path anti-dependency breaking.

Modified:
    llvm/trunk/lib/Target/X86/X86Subtarget.h

Modified: llvm/trunk/lib/Target/X86/X86Subtarget.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Subtarget.h?rev=84931&r1=84930&r2=84931&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86Subtarget.h (original)
+++ llvm/trunk/lib/Target/X86/X86Subtarget.h Fri Oct 23 00:57:35 2009
@@ -220,7 +220,7 @@
   /// at 'More' optimization level.
   bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,
                              TargetSubtarget::AntiDepBreakMode& mode) const {
-    mode = TargetSubtarget::ANTIDEP_NONE;
+    mode = TargetSubtarget::ANTIDEP_CRITICAL;
     return OptLevel >= CodeGenOpt::Default;
   }
 };




From evan.cheng at apple.com  Fri Oct 23 00:58:34 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Fri, 23 Oct 2009 05:58:34 -0000
Subject: [llvm-commits] [llvm] r84932 - in /llvm/trunk/test/CodeGen/X86:
 2007-01-08-InstrSched.ll sse2.ll sse3.ll vshift-1.ll vshift-2.ll
 vshift-3.ll vshift-5.ll
Message-ID: <200910230558.n9N5wZTT003999@zion.cs.uiuc.edu>

Author: evancheng
Date: Fri Oct 23 00:58:34 2009
New Revision: 84932

URL: http://llvm.org/viewvc/llvm-project?rev=84932&view=rev
Log:
Update tests for 84931.

Modified:
    llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
    llvm/trunk/test/CodeGen/X86/sse2.ll
    llvm/trunk/test/CodeGen/X86/sse3.ll
    llvm/trunk/test/CodeGen/X86/vshift-1.ll
    llvm/trunk/test/CodeGen/X86/vshift-2.ll
    llvm/trunk/test/CodeGen/X86/vshift-3.ll
    llvm/trunk/test/CodeGen/X86/vshift-5.ll

Modified: llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll?rev=84932&r1=84931&r2=84932&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll (original)
+++ llvm/trunk/test/CodeGen/X86/2007-01-08-InstrSched.ll Fri Oct 23 00:58:34 2009
@@ -11,12 +11,12 @@
     %tmp14 = fadd float %tmp12, %tmp7
     ret float %tmp14
 
-; CHECK:      mulss	LCPI1_0(%rip)
-; CHECK-NEXT: mulss	LCPI1_1(%rip)
-; CHECK-NEXT: addss
 ; CHECK:      mulss	LCPI1_3(%rip)
+; CHECK-NEXT: mulss	LCPI1_0(%rip)
+; CHECK-NEXT: mulss	LCPI1_1(%rip)
 ; CHECK-NEXT: mulss	LCPI1_2(%rip)
 ; CHECK-NEXT: addss
 ; CHECK-NEXT: addss
+; CHECK-NEXT: addss
 ; CHECK-NEXT: ret
 }

Modified: llvm/trunk/test/CodeGen/X86/sse2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse2.ll?rev=84932&r1=84931&r2=84932&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/sse2.ll (original)
+++ llvm/trunk/test/CodeGen/X86/sse2.ll Fri Oct 23 00:58:34 2009
@@ -10,10 +10,10 @@
         
 ; CHECK: t1:
 ; CHECK: 	movl	8(%esp), %eax
+; CHECK-NEXT: 	movl	4(%esp), %ecx
 ; CHECK-NEXT: 	movapd	(%eax), %xmm0
-; CHECK-NEXT: 	movl	4(%esp), %eax
 ; CHECK-NEXT: 	movlpd	12(%esp), %xmm0
-; CHECK-NEXT: 	movapd	%xmm0, (%eax)
+; CHECK-NEXT: 	movapd	%xmm0, (%ecx)
 ; CHECK-NEXT: 	ret
 }
 
@@ -26,9 +26,9 @@
         
 ; CHECK: t2:
 ; CHECK: 	movl	8(%esp), %eax
+; CHECK-NEXT: 	movl	4(%esp), %ecx
 ; CHECK-NEXT: 	movapd	(%eax), %xmm0
-; CHECK-NEXT: 	movl	4(%esp), %eax
 ; CHECK-NEXT: 	movhpd	12(%esp), %xmm0
-; CHECK-NEXT: 	movapd	%xmm0, (%eax)
+; CHECK-NEXT: 	movapd	%xmm0, (%ecx)
 ; CHECK-NEXT: 	ret
 }

Modified: llvm/trunk/test/CodeGen/X86/sse3.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sse3.ll?rev=84932&r1=84931&r2=84932&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/sse3.ll (original)
+++ llvm/trunk/test/CodeGen/X86/sse3.ll Fri Oct 23 00:58:34 2009
@@ -168,11 +168,11 @@
         ret void
 ; X64: 	t10:
 ; X64: 		pextrw	$4, %xmm0, %eax
+; X64: 		pextrw	$6, %xmm0, %edx
 ; X64: 		movlhps	%xmm1, %xmm1
 ; X64: 		pshuflw	$8, %xmm1, %xmm1
 ; X64: 		pinsrw	$2, %eax, %xmm1
-; X64: 		pextrw	$6, %xmm0, %eax
-; X64: 		pinsrw	$3, %eax, %xmm1
+; X64: 		pinsrw	$3, %edx, %xmm1
 }
 
 

Modified: llvm/trunk/test/CodeGen/X86/vshift-1.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vshift-1.ll?rev=84932&r1=84931&r2=84932&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/vshift-1.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vshift-1.ll Fri Oct 23 00:58:34 2009
@@ -63,7 +63,7 @@
 ; CHECK: shift3b:
 ; CHECK: movzwl
 ; CHECK: movd
-; CHECK: psllw
+; CHECK-NEXT: psllw
   %0 = insertelement <8 x i16> undef, i16 %amt, i32 0
   %1 = insertelement <8 x i16> %0, i16 %amt, i32 1
   %2 = insertelement <8 x i16> %0, i16 %amt, i32 2

Modified: llvm/trunk/test/CodeGen/X86/vshift-2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vshift-2.ll?rev=84932&r1=84931&r2=84932&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/vshift-2.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vshift-2.ll Fri Oct 23 00:58:34 2009
@@ -63,7 +63,7 @@
 ; CHECK: shift3b:
 ; CHECK: movzwl
 ; CHECK: movd
-; CHECK: psrlw
+; CHECK-NEXT: psrlw
   %0 = insertelement <8 x i16> undef, i16 %amt, i32 0
   %1 = insertelement <8 x i16> %0, i16 %amt, i32 1
   %2 = insertelement <8 x i16> %0, i16 %amt, i32 2

Modified: llvm/trunk/test/CodeGen/X86/vshift-3.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vshift-3.ll?rev=84932&r1=84931&r2=84932&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/vshift-3.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vshift-3.ll Fri Oct 23 00:58:34 2009
@@ -52,7 +52,7 @@
 ; CHECK: shift3b:
 ; CHECK: movzwl
 ; CHECK: movd
-; CHECK: psraw
+; CHECK-NEXT: psraw
   %0 = insertelement <8 x i16> undef, i16 %amt, i32 0
   %1 = insertelement <8 x i16> %0, i16 %amt, i32 1
   %2 = insertelement <8 x i16> %0, i16 %amt, i32 2

Modified: llvm/trunk/test/CodeGen/X86/vshift-5.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vshift-5.ll?rev=84932&r1=84931&r2=84932&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/vshift-5.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vshift-5.ll Fri Oct 23 00:58:34 2009
@@ -6,7 +6,7 @@
 entry:
 ; CHECK: shift5a:
 ; CHECK: movd
-; CHECK: pslld
+; CHECK-NEXT: pslld
   %amt = load i32* %pamt 
   %tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
   %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer 
@@ -20,7 +20,7 @@
 entry:
 ; CHECK: shift5b:
 ; CHECK: movd
-; CHECK: psrad
+; CHECK-NEXT: psrad
   %amt = load i32* %pamt 
   %tmp0 = insertelement <4 x i32> undef, i32 %amt, i32 0
   %shamt = shufflevector <4 x i32> %tmp0, <4 x i32> undef, <4 x i32> zeroinitializer 




From tonic at nondot.org  Fri Oct 23 01:20:06 2009
From: tonic at nondot.org (Tanya Lattner)
Date: Fri, 23 Oct 2009 06:20:06 -0000
Subject: [llvm-commits] [llvm] r84934 - /llvm/trunk/docs/GettingStarted.html
Message-ID: <200910230620.n9N6K7Bs005362@zion.cs.uiuc.edu>

Author: tbrethou
Date: Fri Oct 23 01:20:06 2009
New Revision: 84934

URL: http://llvm.org/viewvc/llvm-project?rev=84934&view=rev
Log:
Add 2.6 release note.

Modified:
    llvm/trunk/docs/GettingStarted.html

Modified: llvm/trunk/docs/GettingStarted.html
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/GettingStarted.html?rev=84934&r1=84933&r2=84934&view=diff

==============================================================================
--- llvm/trunk/docs/GettingStarted.html (original)
+++ llvm/trunk/docs/GettingStarted.html Fri Oct 23 01:20:06 2009
@@ -724,6 +724,7 @@
 subdirectories of the '<tt>tags</tt>' directory:</p>
 
 <ul>
+<li>Release 2.6: <b>RELEASE_26</b></li>
 <li>Release 2.5: <b>RELEASE_25</b></li>
 <li>Release 2.4: <b>RELEASE_24</b></li>
 <li>Release 2.3: <b>RELEASE_23</b></li>




From sabre at nondot.org  Fri Oct 23 01:23:49 2009
From: sabre at nondot.org (Chris Lattner)
Date: Fri, 23 Oct 2009 06:23:49 -0000
Subject: [llvm-commits] [llvm] r84936 - in /llvm/trunk:
 lib/Analysis/ConstantFolding.cpp test/Transforms/ConstProp/loads.ll
Message-ID: <200910230623.n9N6NnTT005646@zion.cs.uiuc.edu>

Author: lattner
Date: Fri Oct 23 01:23:49 2009
New Revision: 84936

URL: http://llvm.org/viewvc/llvm-project?rev=84936&view=rev
Log:
teach libanalysis to fold int and fp loads from almost arbitrary
non-type-safe constant initializers.  This sort of thing happens
quite a bit for 4-byte loads out of string constants, unions, 
bitfields, and an interesting endianness check from sqlite, which
is something like this:

const int sqlite3one = 1;
# define SQLITE_BIGENDIAN    (*(char *)(&sqlite3one)==0)
# define SQLITE_LITTLEENDIAN (*(char *)(&sqlite3one)==1)
# define SQLITE_UTF16NATIVE (SQLITE_BIGENDIAN?SQLITE_UTF16BE:SQLITE_UTF16LE)

all of these macros now constant fold away.

This implements PR3152 and is based on a patch started by Eli, but heavily
modified and extended.


Added:
    llvm/trunk/test/Transforms/ConstProp/loads.ll
Modified:
    llvm/trunk/lib/Analysis/ConstantFolding.cpp

Modified: llvm/trunk/lib/Analysis/ConstantFolding.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ConstantFolding.cpp?rev=84936&r1=84935&r2=84936&view=diff

==============================================================================
--- llvm/trunk/lib/Analysis/ConstantFolding.cpp (original)
+++ llvm/trunk/lib/Analysis/ConstantFolding.cpp Fri Oct 23 01:23:49 2009
@@ -93,6 +93,178 @@
   return false;
 }
 
+/// ReadDataFromGlobal - Recursive helper to read bits out of global.  C is the
+/// constant being copied out of. ByteOffset is an offset into C.  CurPtr is the
+/// pointer to copy results into and BytesLeft is the number of bytes left in
+/// the CurPtr buffer.  TD is the target data.
+static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset,
+                               unsigned char *CurPtr, unsigned BytesLeft,
+                               const TargetData &TD) {
+  assert(ByteOffset <= TD.getTypeAllocSize(C->getType()) &&
+         "Out of range access");
+  
+  if (isa<ConstantAggregateZero>(C) || isa<UndefValue>(C))
+    return true;
+  
+  if (ConstantInt *CI = dyn_cast<ConstantInt>(C)) {
+    if (CI->getBitWidth() > 64 ||
+        (CI->getBitWidth() & 7) != 0)
+      return false;
+    
+    uint64_t Val = CI->getZExtValue();
+    unsigned IntBytes = unsigned(CI->getBitWidth()/8);
+    
+    for (unsigned i = 0; i != BytesLeft && ByteOffset != IntBytes; ++i) {
+      CurPtr[i] = (unsigned char)(Val >> ByteOffset * 8);
+      ++ByteOffset;
+    }
+    return true;
+  }
+  
+  if (ConstantFP *CFP = dyn_cast<ConstantFP>(C)) {
+    if (CFP->getType()->isDoubleTy()) {
+      C = ConstantExpr::getBitCast(C, Type::getInt64Ty(C->getContext()));
+      return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, TD);
+    }
+    if (CFP->getType()->isFloatTy()){
+      C = ConstantExpr::getBitCast(C, Type::getInt32Ty(C->getContext()));
+      return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, TD);
+    }
+  }
+
+  if (ConstantStruct *CS = dyn_cast<ConstantStruct>(C)) {
+    const StructLayout *SL = TD.getStructLayout(CS->getType());
+    unsigned Index = SL->getElementContainingOffset(ByteOffset);
+    uint64_t CurEltOffset = SL->getElementOffset(Index);
+    ByteOffset -= CurEltOffset;
+    
+    while (1) {
+      // If the element access is to the element itself and not to tail padding,
+      // read the bytes from the element.
+      uint64_t EltSize = TD.getTypeAllocSize(CS->getOperand(Index)->getType());
+
+      if (ByteOffset < EltSize &&
+          !ReadDataFromGlobal(CS->getOperand(Index), ByteOffset, CurPtr,
+                              BytesLeft, TD))
+        return false;
+      
+      ++Index;
+      
+      // Check to see if we read from the last struct element, if so we're done.
+      if (Index == CS->getType()->getNumElements())
+        return true;
+
+      // If we read all of the bytes we needed from this element we're done.
+      uint64_t NextEltOffset = SL->getElementOffset(Index);
+
+      if (BytesLeft <= NextEltOffset-CurEltOffset-ByteOffset)
+        return true;
+
+      // Move to the next element of the struct.
+      BytesLeft -= NextEltOffset-CurEltOffset-ByteOffset;
+      ByteOffset = 0;
+      CurEltOffset = NextEltOffset;
+    }
+    // not reached.
+  }
+
+  if (ConstantArray *CA = dyn_cast<ConstantArray>(C)) {
+    uint64_t EltSize = TD.getTypeAllocSize(CA->getType()->getElementType());
+    uint64_t Index = ByteOffset / EltSize;
+    uint64_t Offset = ByteOffset - Index * EltSize;
+    for (; Index != CA->getType()->getNumElements(); ++Index) {
+      if (!ReadDataFromGlobal(CA->getOperand(Index), Offset, CurPtr,
+                              BytesLeft, TD))
+        return false;
+      if (EltSize >= BytesLeft)
+        return true;
+      
+      Offset = 0;
+      BytesLeft -= EltSize;
+      CurPtr += EltSize;
+    }
+    return true;
+  }
+  
+  if (ConstantVector *CV = dyn_cast<ConstantVector>(C)) {
+    uint64_t EltSize = TD.getTypeAllocSize(CV->getType()->getElementType());
+    uint64_t Index = ByteOffset / EltSize;
+    uint64_t Offset = ByteOffset - Index * EltSize;
+    for (; Index != CV->getType()->getNumElements(); ++Index) {
+      if (!ReadDataFromGlobal(CV->getOperand(Index), Offset, CurPtr,
+                              BytesLeft, TD))
+        return false;
+      if (EltSize >= BytesLeft)
+        return true;
+      
+      Offset = 0;
+      BytesLeft -= EltSize;
+      CurPtr += EltSize;
+    }
+    return true;
+  }
+  
+  // Otherwise, unknown initializer type.
+  return false;
+}
+
+static Constant *FoldReinterpretLoadFromConstPtr(Constant *C,
+                                                 const TargetData &TD) {
+  const Type *InitializerTy = cast<PointerType>(C->getType())->getElementType();
+  const IntegerType *IntType = dyn_cast<IntegerType>(InitializerTy);
+  
+  // If this isn't an integer load we can't fold it directly.
+  if (!IntType) {
+    // If this is a float/double load, we can try folding it as an int32/64 load
+    // and then bitcast the result.  This can be useful for union cases.
+    const Type *MapTy;
+    if (InitializerTy->isFloatTy())
+      MapTy = Type::getInt32PtrTy(C->getContext());
+    else if (InitializerTy->isDoubleTy())
+      MapTy = Type::getInt64PtrTy(C->getContext());
+    else
+      return 0;
+
+    C = ConstantExpr::getBitCast(C, MapTy);
+    if (Constant *Res = FoldReinterpretLoadFromConstPtr(C, TD))
+      return ConstantExpr::getBitCast(Res, InitializerTy);
+    return 0;
+  }
+  
+  unsigned BytesLoaded = (IntType->getBitWidth() + 7) / 8;
+  if (BytesLoaded > 8 || BytesLoaded == 0) return 0;
+  
+  GlobalValue *GVal;
+  int64_t Offset;
+  if (!IsConstantOffsetFromGlobal(C, GVal, Offset, TD))
+    return 0;
+  
+  GlobalVariable *GV = dyn_cast<GlobalVariable>(GVal);
+  if (!GV || !GV->isConstant() || !GV->hasInitializer() ||
+      !GV->hasDefinitiveInitializer() ||
+      !GV->getInitializer()->getType()->isSized())
+    return 0;
+
+  // If we're loading off the beginning of the global, some bytes may be valid,
+  // but we don't try to handle this.
+  if (Offset < 0) return 0;
+  
+  // If we're not accessing anything in this constant, the result is undefined.
+  if (uint64_t(Offset) >= TD.getTypeAllocSize(GV->getInitializer()->getType()))
+    return UndefValue::get(IntType);
+  
+  unsigned char RawBytes[8] = {0};
+  if (!ReadDataFromGlobal(GV->getInitializer(), Offset, RawBytes,
+                          BytesLoaded, TD))
+    return 0;
+
+  uint64_t ResultVal = 0;
+  for (unsigned i = 0; i != BytesLoaded; ++i)
+    ResultVal |= (uint64_t)RawBytes[i] << (i * 8);
+
+  return ConstantInt::get(IntType, ResultVal);
+}
+
 /// ConstantFoldLoadFromConstPtr - Return the value that a load from C would
 /// produce if it is constant and determinable.  If this is not determinable,
 /// return null.
@@ -119,20 +291,20 @@
   // directly if string length is small enough.
   std::string Str;
   if (TD && GetConstantStringInfo(CE->getOperand(0), Str) && !Str.empty()) {
-    unsigned len = Str.length();
+    unsigned StrLen = Str.length();
     const Type *Ty = cast<PointerType>(CE->getType())->getElementType();
-    unsigned numBits = Ty->getPrimitiveSizeInBits();
+    unsigned NumBits = Ty->getPrimitiveSizeInBits();
     // Replace LI with immediate integer store.
-    if ((numBits >> 3) == len + 1) {
-      APInt StrVal(numBits, 0);
-      APInt SingleChar(numBits, 0);
+    if ((NumBits >> 3) == StrLen + 1) {
+      APInt StrVal(NumBits, 0);
+      APInt SingleChar(NumBits, 0);
       if (TD->isLittleEndian()) {
-        for (signed i = len-1; i >= 0; i--) {
+        for (signed i = StrLen-1; i >= 0; i--) {
           SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
           StrVal = (StrVal << 8) | SingleChar;
         }
       } else {
-        for (unsigned i = 0; i < len; i++) {
+        for (unsigned i = 0; i < StrLen; i++) {
           SingleChar = (uint64_t) Str[i] & UCHAR_MAX;
           StrVal = (StrVal << 8) | SingleChar;
         }
@@ -156,6 +328,11 @@
     }
   }
   
+  // Try hard to fold loads from bitcasted strange and non-type-safe things.  We
+  // currently don't do any of this for big endian systems.  It can be
+  // generalized in the future if someone is interested.
+  if (TD && TD->isLittleEndian())
+    return FoldReinterpretLoadFromConstPtr(CE, *TD);
   return 0;
 }
 
@@ -164,7 +341,7 @@
   
   if (Constant *C = dyn_cast<Constant>(LI->getOperand(0)))
     return ConstantFoldLoadFromConstPtr(C, TD);
-    
+
   return 0;
 }
 

Added: llvm/trunk/test/Transforms/ConstProp/loads.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/ConstProp/loads.ll?rev=84936&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/ConstProp/loads.ll (added)
+++ llvm/trunk/test/Transforms/ConstProp/loads.ll Fri Oct 23 01:23:49 2009
@@ -0,0 +1,68 @@
+; RUN: opt < %s -instcombine -S | FileCheck %s 
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
+
+ at test1 = constant {{i32,i8},i32} {{i32,i8} { i32 -559038737, i8 186 }, i32 -889275714 }
+ at test2 = constant double 1.0
+
+; Simple load
+define i32 @test1() {
+  %r = load i32* getelementptr ({{i32,i8},i32}* @test1, i32 0, i32 0, i32 0)
+  ret i32 %r
+; @test1
+; CHECK: ret i32 -559038737
+}
+
+; PR3152
+; Load of first 16 bits of 32-bit value.
+define i16 @test2() {
+  %r = load i16* bitcast(i32* getelementptr ({{i32,i8},i32}* @test1, i32 0, i32 0, i32 0) to i16*)
+  ret i16 %r
+
+; @test2
+; CHECK: ret i16 -16657 
+}
+
+; Load of second 16 bits of 32-bit value.
+define i16 @test3() {
+  %r = load i16* getelementptr(i16* bitcast(i32* getelementptr ({{i32,i8},i32}* @test1, i32 0, i32 0, i32 0) to i16*), i32 1)
+  ret i16 %r
+
+; @test3
+; CHECK: ret i16 -8531
+}
+
+; Load of 8 bit field + tail padding.
+define i16 @test4() {
+  %r = load i16* getelementptr(i16* bitcast(i32* getelementptr ({{i32,i8},i32}* @test1, i32 0, i32 0, i32 0) to i16*), i32 2)
+  ret i16 %r
+; @test4
+; CHECK: ret i16 186
+}
+
+; Load of double bits.
+define i64 @test6() {
+  %r = load i64* bitcast(double* @test2 to i64*)
+  ret i64 %r
+
+; @test6
+; CHECK: ret i64 4607182418800017408
+}
+
+; Load of double bits.
+define i16 @test7() {
+  %r = load i16* bitcast(double* @test2 to i16*)
+  ret i16 %r
+
+; @test7
+; CHECK: ret i16 0
+}
+
+; Double load.
+define double @test8() {
+  %r = load double* bitcast({{i32,i8},i32}* @test1 to double*)
+  ret double %r
+
+; @test8
+; CHECK: ret double 0xDEADBEBA
+}




From tonic at nondot.org  Fri Oct 23 01:31:01 2009
From: tonic at nondot.org (Tanya Lattner)
Date: Fri, 23 Oct 2009 06:31:01 -0000
Subject: [llvm-commits] [test-suite] r84939 - /test-suite/tags/RELEASE_26/
Message-ID: <200910230631.n9N6V1Ff006139@zion.cs.uiuc.edu>

Author: tbrethou
Date: Fri Oct 23 01:31:01 2009
New Revision: 84939

URL: http://llvm.org/viewvc/llvm-project?rev=84939&view=rev
Log:
Tag 2.6


Added:
    test-suite/tags/RELEASE_26/
      - copied from r84938, test-suite/branches/release_26/



From sabre at nondot.org  Fri Oct 23 01:50:38 2009
From: sabre at nondot.org (Chris Lattner)
Date: Fri, 23 Oct 2009 06:50:38 -0000
Subject: [llvm-commits] [llvm] r84941 - in /llvm/trunk:
 lib/Analysis/ConstantFolding.cpp test/Transforms/ConstProp/loads.ll
Message-ID: <200910230650.n9N6oc8l007320@zion.cs.uiuc.edu>

Author: lattner
Date: Fri Oct 23 01:50:36 2009
New Revision: 84941

URL: http://llvm.org/viewvc/llvm-project?rev=84941&view=rev
Log:
enhance FoldReinterpretLoadFromConstPtr to handle loads of up to 32 
bytes (i256).

Modified:
    llvm/trunk/lib/Analysis/ConstantFolding.cpp
    llvm/trunk/test/Transforms/ConstProp/loads.ll

Modified: llvm/trunk/lib/Analysis/ConstantFolding.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ConstantFolding.cpp?rev=84941&r1=84940&r2=84941&view=diff

==============================================================================
--- llvm/trunk/lib/Analysis/ConstantFolding.cpp (original)
+++ llvm/trunk/lib/Analysis/ConstantFolding.cpp Fri Oct 23 01:50:36 2009
@@ -232,7 +232,7 @@
   }
   
   unsigned BytesLoaded = (IntType->getBitWidth() + 7) / 8;
-  if (BytesLoaded > 8 || BytesLoaded == 0) return 0;
+  if (BytesLoaded > 32 || BytesLoaded == 0) return 0;
   
   GlobalValue *GVal;
   int64_t Offset;
@@ -253,16 +253,18 @@
   if (uint64_t(Offset) >= TD.getTypeAllocSize(GV->getInitializer()->getType()))
     return UndefValue::get(IntType);
   
-  unsigned char RawBytes[8] = {0};
+  unsigned char RawBytes[32] = {0};
   if (!ReadDataFromGlobal(GV->getInitializer(), Offset, RawBytes,
                           BytesLoaded, TD))
     return 0;
 
-  uint64_t ResultVal = 0;
-  for (unsigned i = 0; i != BytesLoaded; ++i)
-    ResultVal |= (uint64_t)RawBytes[i] << (i * 8);
+  APInt ResultVal(IntType->getBitWidth(), 0);
+  for (unsigned i = 0; i != BytesLoaded; ++i) {
+    ResultVal <<= 8;
+    ResultVal |= APInt(IntType->getBitWidth(), RawBytes[BytesLoaded-1-i]);
+  }
 
-  return ConstantInt::get(IntType, ResultVal);
+  return ConstantInt::get(IntType->getContext(), ResultVal);
 }
 
 /// ConstantFoldLoadFromConstPtr - Return the value that a load from C would

Modified: llvm/trunk/test/Transforms/ConstProp/loads.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/ConstProp/loads.ll?rev=84941&r1=84940&r2=84941&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/ConstProp/loads.ll (original)
+++ llvm/trunk/test/Transforms/ConstProp/loads.ll Fri Oct 23 01:50:36 2009
@@ -4,6 +4,7 @@
 
 @test1 = constant {{i32,i8},i32} {{i32,i8} { i32 -559038737, i8 186 }, i32 -889275714 }
 @test2 = constant double 1.0
+ at test3 = constant {i64, i64} { i64 123, i64 112312312 }
 
 ; Simple load
 define i32 @test1() {
@@ -66,3 +67,13 @@
 ; @test8
 ; CHECK: ret double 0xDEADBEBA
 }
+
+
+; i128 load.
+define i128 @test9() {
+  %r = load i128* bitcast({i64, i64}* @test3 to i128*)
+  ret i128 %r
+
+; @test9
+; CHECK: ret i128 112312312
+}




From sabre at nondot.org  Fri Oct 23 01:57:37 2009
From: sabre at nondot.org (Chris Lattner)
Date: Fri, 23 Oct 2009 06:57:37 -0000
Subject: [llvm-commits] [llvm] r84942 - in /llvm/trunk:
 lib/Analysis/ConstantFolding.cpp test/Transforms/ConstProp/loads.ll
Message-ID: <200910230657.n9N6vbrf016695@zion.cs.uiuc.edu>

Author: lattner
Date: Fri Oct 23 01:57:37 2009
New Revision: 84942

URL: http://llvm.org/viewvc/llvm-project?rev=84942&view=rev
Log:
teach libanalysis to simplify vector loads with bitcast sources.  This
implements something out of Target/README.txt producing:

_foo:                                                       ## @foo
	movl	4(%esp), %eax
	movapd	LCPI1_0, %xmm0
	movapd	%xmm0, (%eax)
	ret	$4

instead of:

_foo:                                                       ## @foo
	movl	4(%esp), %eax
	movapd	_b, %xmm0
	mulpd	LCPI1_0, %xmm0
	addpd	_a, %xmm0
	movapd	%xmm0, (%eax)
	ret	$4


Modified:
    llvm/trunk/lib/Analysis/ConstantFolding.cpp
    llvm/trunk/test/Transforms/ConstProp/loads.ll

Modified: llvm/trunk/lib/Analysis/ConstantFolding.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ConstantFolding.cpp?rev=84942&r1=84941&r2=84942&view=diff

==============================================================================
--- llvm/trunk/lib/Analysis/ConstantFolding.cpp (original)
+++ llvm/trunk/lib/Analysis/ConstantFolding.cpp Fri Oct 23 01:57:37 2009
@@ -210,24 +210,30 @@
 
 static Constant *FoldReinterpretLoadFromConstPtr(Constant *C,
                                                  const TargetData &TD) {
-  const Type *InitializerTy = cast<PointerType>(C->getType())->getElementType();
-  const IntegerType *IntType = dyn_cast<IntegerType>(InitializerTy);
+  const Type *LoadTy = cast<PointerType>(C->getType())->getElementType();
+  const IntegerType *IntType = dyn_cast<IntegerType>(LoadTy);
   
   // If this isn't an integer load we can't fold it directly.
   if (!IntType) {
     // If this is a float/double load, we can try folding it as an int32/64 load
-    // and then bitcast the result.  This can be useful for union cases.
+    // and then bitcast the result.  This can be useful for union cases.  Note
+    // that address spaces don't matter here since we're not going to result in
+    // an actual new load.
     const Type *MapTy;
-    if (InitializerTy->isFloatTy())
+    if (LoadTy->isFloatTy())
       MapTy = Type::getInt32PtrTy(C->getContext());
-    else if (InitializerTy->isDoubleTy())
+    else if (LoadTy->isDoubleTy())
       MapTy = Type::getInt64PtrTy(C->getContext());
-    else
+    else if (isa<VectorType>(LoadTy)) {
+      MapTy = IntegerType::get(C->getContext(),
+                               TD.getTypeAllocSizeInBits(LoadTy));
+      MapTy = PointerType::getUnqual(MapTy);
+    } else
       return 0;
 
     C = ConstantExpr::getBitCast(C, MapTy);
     if (Constant *Res = FoldReinterpretLoadFromConstPtr(C, TD))
-      return ConstantExpr::getBitCast(Res, InitializerTy);
+      return ConstantExpr::getBitCast(Res, LoadTy);
     return 0;
   }
   

Modified: llvm/trunk/test/Transforms/ConstProp/loads.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/ConstProp/loads.ll?rev=84942&r1=84941&r2=84942&view=diff

==============================================================================
--- llvm/trunk/test/Transforms/ConstProp/loads.ll (original)
+++ llvm/trunk/test/Transforms/ConstProp/loads.ll Fri Oct 23 01:57:37 2009
@@ -77,3 +77,13 @@
 ; @test9
 ; CHECK: ret i128 112312312
 }
+
+; vector load.
+define <2 x i64> @test10() {
+  %r = load <2 x i64>* bitcast({i64, i64}* @test3 to <2 x i64>*)
+  ret <2 x i64> %r
+
+; @test10
+; CHECK: ret <2 x i64> <i64 112312312, i64 0>
+}
+




From sabre at nondot.org  Fri Oct 23 02:00:56 2009
From: sabre at nondot.org (Chris Lattner)
Date: Fri, 23 Oct 2009 07:00:56 -0000
Subject: [llvm-commits] [llvm] r84943 - /llvm/trunk/lib/Target/README.txt
Message-ID: <200910230700.n9N70u6k016918@zion.cs.uiuc.edu>

Author: lattner
Date: Fri Oct 23 02:00:55 2009
New Revision: 84943

URL: http://llvm.org/viewvc/llvm-project?rev=84943&view=rev
Log:
some stuff is done, we still have constantexpr simplification to do.

Modified:
    llvm/trunk/lib/Target/README.txt

Modified: llvm/trunk/lib/Target/README.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/README.txt?rev=84943&r1=84942&r2=84943&view=diff

==============================================================================
--- llvm/trunk/lib/Target/README.txt (original)
+++ llvm/trunk/lib/Target/README.txt Fri Oct 23 02:00:55 2009
@@ -1594,23 +1594,8 @@
 
 //===---------------------------------------------------------------------===//
 
-Instcombine should replace the load with a constant in:
-
-  static const char x[4] = {'a', 'b', 'c', 'd'};
-  
-  unsigned int y(void) {
-    return *(unsigned int *)x;
-  }
-
-It currently only does this transformation when the size of the constant 
-is the same as the size of the integer (so, try x[5]) and the last byte 
-is a null (making it a C string). There's no need for these restrictions.
-
-//===---------------------------------------------------------------------===//
-
-InstCombine's "turn load from constant into constant" optimization should be
-more aggressive in the presence of bitcasts.  For example, because of unions,
-this code:
+libanalysis is not aggressively folding vector bitcasts.  For example, the
+constant expressions generated when compiling this code:
 
 union vec2d {
     double e[2];
@@ -1624,24 +1609,23 @@
     return (vec2d){ .v = a.v + b.v * (vec2d){{5,5}}.v };
 }
 
-Compiles into:
+in X86-32 end up being:
 
- at a = internal constant %0 { [2 x double] 
-           [double 1.000000e+00, double 2.000000e+00] }, align 16
- at b = internal constant %0 { [2 x double]
-           [double 3.000000e+00, double 4.000000e+00] }, align 16
-...
-define void @foo(%struct.vec2d* noalias nocapture sret %agg.result) nounwind {
+define void @foo(%union.vec2d* noalias nocapture sret %agg.result) nounwind ssp {
 entry:
-	%0 = load <2 x double>* getelementptr (%struct.vec2d* 
-           bitcast (%0* @a to %struct.vec2d*), i32 0, i32 0), align 16
-	%1 = load <2 x double>* getelementptr (%struct.vec2d* 
-           bitcast (%0* @b to %struct.vec2d*), i32 0, i32 0), align 16
-
+  %agg.result.0 = getelementptr %union.vec2d* %agg.result, i32 0, i32 0 ; <<2 x double>*> [#uses=1]
+  store <2 x double> fadd (<2 x double> bitcast (<1 x i128> <i128 85070591730234615870450834276742070272> to <2 x double>), <2 x double> fmul (<2 x double> bitcast (<1 x i128> <i128 85153668479971173112514077617450647552> to <2 x double>), <2 x double> <double 5.000000e+00, double 5.000000e+00>)), <2 x double>* %agg.result.0, align 16
+  ret void
+}
 
-Instcombine should be able to optimize away the loads (and thus the globals).
+and in X86-64 mode:
 
-See also PR4973
+define %0 @foo() nounwind readnone ssp {
+entry:
+  %mrv5 = insertvalue %0 undef, double extractelement (<2 x double> fadd (<2 x double> bitcast (<1 x i128> <i128 85070591730234615870450834276742070272> to <2 x double>), <2 x double> fmul (<2 x double> bitcast (<1 x i128> <i128 85153668479971173112514077617450647552> to <2 x double>), <2 x double> bitcast (<1 x i128> <i128 85174437667405312423031577302488055808> to <2 x double>))), i32 0), 0 ; <%0> [#uses=1]
+  %mrv6 = insertvalue %0 %mrv5, double extractelement (<2 x double> fadd (<2 x double> bitcast (<1 x i128> <i128 85070591730234615870450834276742070272> to <2 x double>), <2 x double> fmul (<2 x double> bitcast (<1 x i128> <i128 85153668479971173112514077617450647552> to <2 x double>), <2 x double> bitcast (<1 x i128> <i128 85174437667405312423031577302488055808> to <2 x double>))), i32 1), 1 ; <%0> [#uses=1]
+  ret %0 %mrv6
+}
 
 //===---------------------------------------------------------------------===//
 




From evan.cheng at apple.com  Fri Oct 23 02:05:32 2009
From: evan.cheng at apple.com (Evan Cheng)
Date: Fri, 23 Oct 2009 07:05:32 -0000
Subject: [llvm-commits] [test-suite] r84944 -
 /test-suite/trunk/MultiSource/Benchmarks/BitBench/uuencode/uuencode.c
Message-ID: <200910230705.n9N75Wlj017291@zion.cs.uiuc.edu>

Author: evancheng
Date: Fri Oct 23 02:05:32 2009
New Revision: 84944

URL: http://llvm.org/viewvc/llvm-project?rev=84944&view=rev
Log:
Make the test bigger.

Modified:
    test-suite/trunk/MultiSource/Benchmarks/BitBench/uuencode/uuencode.c

Modified: test-suite/trunk/MultiSource/Benchmarks/BitBench/uuencode/uuencode.c
URL: http://llvm.org/viewvc/llvm-project/test-suite/trunk/MultiSource/Benchmarks/BitBench/uuencode/uuencode.c?rev=84944&r1=84943&r2=84944&view=diff

==============================================================================
--- test-suite/trunk/MultiSource/Benchmarks/BitBench/uuencode/uuencode.c (original)
+++ test-suite/trunk/MultiSource/Benchmarks/BitBench/uuencode/uuencode.c Fri Oct 23 02:05:32 2009
@@ -180,7 +180,7 @@
   }
   size = read_data(in, inbuf);
   gettimeofday(&pre,0);
-  for(i=0;i<100;i++){
+  for(i=0;i<1000;i++){
     outsize = do_encode(inbuf,outbuf, size, argv[1]);
    
   }




From baldrick at free.fr  Fri Oct 23 07:06:58 2009
From: baldrick at free.fr (Duncan Sands)
Date: Fri, 23 Oct 2009 14:06:58 +0200
Subject: [llvm-commits] [llvm] r84936 - in /llvm/trunk:
 lib/Analysis/ConstantFolding.cpp test/Transforms/ConstProp/loads.ll
In-Reply-To: <200910230623.n9N6NnTT005646@zion.cs.uiuc.edu>
References: <200910230623.n9N6NnTT005646@zion.cs.uiuc.edu>
Message-ID: <4AE19C62.5090709@free.fr>

Hi Chris,

> +  if (isa<ConstantAggregateZero>(C) || isa<UndefValue>(C))
> +    return true;

here you use that RawBytes is zero initialized, but that requirement
is not documented.

> +      CurPtr[i] = (unsigned char)(Val >> ByteOffset * 8);

You should probably bracket ByteOffset * 8 for weenies like me who
don't know C precedence rules.

> +  if (ConstantFP *CFP = dyn_cast<ConstantFP>(C)) {
> +    if (CFP->getType()->isDoubleTy()) {
> +      C = ConstantExpr::getBitCast(C, Type::getInt64Ty(C->getContext()));
> +      return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, TD);
> +    }
> +    if (CFP->getType()->isFloatTy()){
> +      C = ConstantExpr::getBitCast(C, Type::getInt32Ty(C->getContext()));
> +      return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, TD);
> +    }

Probably you should return false here.

> +  }

> +  GlobalVariable *GV = dyn_cast<GlobalVariable>(GVal);
> +  if (!GV || !GV->isConstant() || !GV->hasInitializer() ||
> +      !GV->hasDefinitiveInitializer() ||
> +      !GV->getInitializer()->getType()->isSized())
> +    return 0;

Testing hasDefinitiveInitializer also tests hasInitializer.

Ciao,

Duncan.


From echristo at apple.com  Fri Oct 23 12:05:49 2009
From: echristo at apple.com (Eric Christopher)
Date: Fri, 23 Oct 2009 10:05:49 -0700
Subject: [llvm-commits] [llvm] r84900 -
	/llvm/trunk/include/llvm/ADT/StringRef.h
In-Reply-To: <4AE12BA2.8020503@free.fr>
References: <200910222206.n9MM6oVI030595@zion.cs.uiuc.edu>
	<4AE12BA2.8020503@free.fr>
Message-ID: <7ED41D5A-EC6E-45A8-A5DA-BB4A5F635483@apple.com>


On Oct 22, 2009, at 9:05 PM, Duncan Sands wrote:

> Hi Eric,
> 
>> size_t, not unsigned here to silence a warning.
> 
> while I'm all for white space cleanups, can you please not mix them
> in with real changes since it makes it harder to see what the real
> changes were.

Yup. Normally I don't forget to fix them separately. Do apologize.

-eric


From gohman at apple.com  Fri Oct 23 12:10:02 2009
From: gohman at apple.com (Dan Gohman)
Date: Fri, 23 Oct 2009 17:10:02 -0000
Subject: [llvm-commits] [llvm] r84952 - in /llvm/trunk:
 lib/Transforms/Scalar/LoopDeletion.cpp
 test/Transforms/LoopDeletion/multiple-exit-conditions.ll
Message-ID: <200910231710.n9NHA3T2031176@zion.cs.uiuc.edu>

Author: djg
Date: Fri Oct 23 12:10:01 2009
New Revision: 84952

URL: http://llvm.org/viewvc/llvm-project?rev=84952&view=rev
Log:
Make LoopDeletion check the maximum backedge taken count, rather than the
exact backedge taken count, when checking for infinite loops. This allows
it to delete loops with multiple exit conditions.

Added:
    llvm/trunk/test/Transforms/LoopDeletion/multiple-exit-conditions.ll
Modified:
    llvm/trunk/lib/Transforms/Scalar/LoopDeletion.cpp

Modified: llvm/trunk/lib/Transforms/Scalar/LoopDeletion.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopDeletion.cpp?rev=84952&r1=84951&r2=84952&view=diff

==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/LoopDeletion.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/LoopDeletion.cpp Fri Oct 23 12:10:01 2009
@@ -167,7 +167,7 @@
   // Don't remove loops for which we can't solve the trip count.
   // They could be infinite, in which case we'd be changing program behavior.
   ScalarEvolution& SE = getAnalysis<ScalarEvolution>();
-  const SCEV *S = SE.getBackedgeTakenCount(L);
+  const SCEV *S = SE.getMaxBackedgeTakenCount(L);
   if (isa<SCEVCouldNotCompute>(S))
     return Changed;
   

Added: llvm/trunk/test/Transforms/LoopDeletion/multiple-exit-conditions.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopDeletion/multiple-exit-conditions.ll?rev=84952&view=auto

==============================================================================
--- llvm/trunk/test/Transforms/LoopDeletion/multiple-exit-conditions.ll (added)
+++ llvm/trunk/test/Transforms/LoopDeletion/multiple-exit-conditions.ll Fri Oct 23 12:10:01 2009
@@ -0,0 +1,27 @@
+; RUN: opt < %s -loop-deletion -S | FileCheck %s
+
+; ScalarEvolution can prove the loop iteration is finite, even though
+; it can't represent the exact trip count as an expression. That's
+; good enough to let the loop be deleted.
+
+; CHECK:      entry:
+; CHECK-NEXT:   br label %return
+
+; CHECK:      return:
+; CHECK-NEXT:   ret void
+
+define void @foo(i64 %n, i64 %m) nounwind {
+entry:
+  br label %bb
+
+bb:
+  %x.0 = phi i64 [ 0, %entry ], [ %t0, %bb ]
+  %t0 = add i64 %x.0, 1
+  %t1 = icmp slt i64 %x.0, %n
+  %t3 = icmp sgt i64 %x.0, %m
+  %t4 = and i1 %t1, %t3
+  br i1 %t4, label %bb, label %return
+
+return:
+  ret void
+}




From gohman at apple.com  Fri Oct 23 12:57:43 2009
From: gohman at apple.com (Dan Gohman)
Date: Fri, 23 Oct 2009 17:57:43 -0000
Subject: [llvm-commits] [llvm] r84954 - in /llvm/trunk:
 lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp
 test/CodeGen/X86/large-gep-scale.ll
Message-ID: <200910231757.n9NHvhKc000995@zion.cs.uiuc.edu>

Author: djg
Date: Fri Oct 23 12:57:43 2009
New Revision: 84954

URL: http://llvm.org/viewvc/llvm-project?rev=84954&view=rev
Log:
APInt-ify the gep scaling code, so that it correctly handles the case where
the scale overflows pointer-sized arithmetic. This fixes PR5281.

Added:
    llvm/trunk/test/CodeGen/X86/large-gep-scale.ll
Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp?rev=84954&r1=84953&r2=84954&view=diff

==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp Fri Oct 23 12:57:43 2009
@@ -2666,7 +2666,8 @@
       }
 
       // N = N + Idx * ElementSize;
-      uint64_t ElementSize = TD->getTypeAllocSize(Ty);
+      APInt ElementSize = APInt(TLI.getPointerTy().getSizeInBits(),
+                                TD->getTypeAllocSize(Ty));
       SDValue IdxN = getValue(Idx);
 
       // If the index is smaller or larger than intptr_t, truncate or extend
@@ -2676,13 +2677,13 @@
       // If this is a multiply by a power of two, turn it into a shl
       // immediately.  This is a very common case.
       if (ElementSize != 1) {
-        if (isPowerOf2_64(ElementSize)) {
-          unsigned Amt = Log2_64(ElementSize);
+        if (ElementSize.isPowerOf2()) {
+          unsigned Amt = ElementSize.logBase2();
           IdxN = DAG.getNode(ISD::SHL, getCurDebugLoc(),
                              N.getValueType(), IdxN,
                              DAG.getConstant(Amt, TLI.getPointerTy()));
         } else {
-          SDValue Scale = DAG.getIntPtrConstant(ElementSize);
+          SDValue Scale = DAG.getConstant(ElementSize, TLI.getPointerTy());
           IdxN = DAG.getNode(ISD::MUL, getCurDebugLoc(),
                              N.getValueType(), IdxN, Scale);
         }

Added: llvm/trunk/test/CodeGen/X86/large-gep-scale.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/large-gep-scale.ll?rev=84954&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/X86/large-gep-scale.ll (added)
+++ llvm/trunk/test/CodeGen/X86/large-gep-scale.ll Fri Oct 23 12:57:43 2009
@@ -0,0 +1,12 @@
+; RUN: llc < %s -march=x86 | FileCheck %s
+; PR5281
+
+; After scaling, this type doesn't fit in memory. Codegen should generate
+; correct addressing still.
+
+; CHECK: shll $2, %edx
+
+define fastcc i32* @_ada_smkr([2147483647 x i32]* %u, i32 %t) nounwind {
+  %x = getelementptr [2147483647 x i32]* %u, i32 %t, i32 0
+  ret i32* %x
+}




From tbrethou at cs.uiuc.edu  Fri Oct 23 13:52:15 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:52:15 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/
Message-ID: <200910231852.n9NIqFsj003808@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6 added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/
Message-ID: <200910231853.n9NIrIdA004100@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/CommandGuide/html/
Message-ID: <200910231853.n9NIrIY3004106@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/CommandGuide/html:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/CommandGuide/html added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS:
	llvm-www/releases/2.6/docs/CommandGuide/man/man1/
Message-ID: <200910231853.n9NIrIC2004115@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/CommandGuide/man/man1:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/CommandGuide/man/man1 added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/CommandGuide/man/
Message-ID: <200910231853.n9NIrIWq004110@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/CommandGuide/man:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/CommandGuide/man added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/img/
Message-ID: <200910231853.n9NIrIqF004130@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/img:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/img added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/doxygen/html/
Message-ID: <200910231853.n9NIrItf004143@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/doxygen/html:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/doxygen/html added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/CommandGuide/
Message-ID: <200910231853.n9NIrIXB004103@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/CommandGuide:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/CommandGuide added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/CommandGuide/ps/
Message-ID: <200910231853.n9NIrIug004118@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/CommandGuide/ps:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/CommandGuide/ps added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/doxygen/
Message-ID: <200910231853.n9NIrI3G004138@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/doxygen:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/doxygen added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/tutorial/
Message-ID: <200910231853.n9NIrIk2004135@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/tutorial:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/tutorial added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tbrethou at cs.uiuc.edu  Fri Oct 23 13:53:18 2009
From: tbrethou at cs.uiuc.edu (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:53:18 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/HistoricalNotes/
Message-ID: <200910231853.n9NIrIjG004124@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/HistoricalNotes:

---
Log message:

Directory /home/vadve/shared/PublicCVS/llvm-www/releases/2.6/docs/HistoricalNotes added to the repository


---
Diffs of the changes:  (+0 -0)

 0 files changed





From tonic at nondot.org  Fri Oct 23 13:57:46 2009
From: tonic at nondot.org (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:57:46 -0500
Subject: [llvm-commits] CVS: llvm-www/releases/2.6/docs/img/Debugging.gif
	libdeps.gif lines.gif objdeps.gif venusflytrap.jpg
Message-ID: <200910231857.n9NIvkek004607@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/img:

Debugging.gif added (r1.1)
libdeps.gif added (r1.1)
lines.gif added (r1.1)
objdeps.gif added (r1.1)
venusflytrap.jpg added (r1.1)
---
Log message:

Add 2.6



---
Diffs of the changes:  (+0 -0)

 Debugging.gif    |    0 
 libdeps.gif      |    0 
 lines.gif        |    0 
 objdeps.gif      |    0 
 venusflytrap.jpg |    0 
 5 files changed


Index: llvm-www/releases/2.6/docs/img/Debugging.gif


Index: llvm-www/releases/2.6/docs/img/libdeps.gif


Index: llvm-www/releases/2.6/docs/img/lines.gif


Index: llvm-www/releases/2.6/docs/img/objdeps.gif


Index: llvm-www/releases/2.6/docs/img/venusflytrap.jpg





From tonic at nondot.org  Fri Oct 23 13:57:46 2009
From: tonic at nondot.org (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:57:46 -0500
Subject: [llvm-commits] CVS:
	llvm-www/releases/2.6/docs/HistoricalNotes/2000-11-18-EarlyDesignIdeas.txt
	2000-11-18-EarlyDesignIdeasResp.txt
	2000-12-06-EncodingIdea.txt 2000-12-06-MeetingSummary.txt
	2001-01-31-UniversalIRIdea.txt 2001-02-06-TypeNotationDebate.txt
	2001-02-06-TypeNotationDebateResp1.txt
	2001-02-06-TypeNotationDebateResp2.txt
	2001-02-06-TypeNotationDebateResp4.txt 2001-02-09-AdveComments.txt
	2001-02-09-AdveCommentsResponse.txt 2001-02-13-Reference-Memory.txt
	2001-02-13-Reference-MemoryResponse.txt
	2001-04-16-DynamicCompilation.txt 2001-05-18-ExceptionHandling.txt
	2001-05-19-ExceptionResponse.txt 2001-06-01-GCCOptimizations.txt
	2001-06-01-GCCOptimizations2.txt 2001-06-20-.NET-Differences.txt
	2001-07-06-LoweringIRForCodeGen.txt
	2001-07-08-InstructionSelection.txt
	2001-07-08-InstructionSelection2.txt
	2001-09-18-OptimizeExceptions.txt
	2002-05-12-InstListChange.txt 2002-06-25-MegaPatchInfo.txt
	2003-01-23-CygwinNotes.txt 2003-06-25-Reoptimizer1.txt
	2003-06-26-Reoptimizer2.txt
Message-ID: <200910231857.n9NIvkIB004802@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/HistoricalNotes:

2000-11-18-EarlyDesignIdeas.txt added (r1.1)
2000-11-18-EarlyDesignIdeasResp.txt added (r1.1)
2000-12-06-EncodingIdea.txt added (r1.1)
2000-12-06-MeetingSummary.txt added (r1.1)
2001-01-31-UniversalIRIdea.txt added (r1.1)
2001-02-06-TypeNotationDebate.txt added (r1.1)
2001-02-06-TypeNotationDebateResp1.txt added (r1.1)
2001-02-06-TypeNotationDebateResp2.txt added (r1.1)
2001-02-06-TypeNotationDebateResp4.txt added (r1.1)
2001-02-09-AdveComments.txt added (r1.1)
2001-02-09-AdveCommentsResponse.txt added (r1.1)
2001-02-13-Reference-Memory.txt added (r1.1)
2001-02-13-Reference-MemoryResponse.txt added (r1.1)
2001-04-16-DynamicCompilation.txt added (r1.1)
2001-05-18-ExceptionHandling.txt added (r1.1)
2001-05-19-ExceptionResponse.txt added (r1.1)
2001-06-01-GCCOptimizations.txt added (r1.1)
2001-06-01-GCCOptimizations2.txt added (r1.1)
2001-06-20-.NET-Differences.txt added (r1.1)
2001-07-06-LoweringIRForCodeGen.txt added (r1.1)
2001-07-08-InstructionSelection.txt added (r1.1)
2001-07-08-InstructionSelection2.txt added (r1.1)
2001-09-18-OptimizeExceptions.txt added (r1.1)
2002-05-12-InstListChange.txt added (r1.1)
2002-06-25-MegaPatchInfo.txt added (r1.1)
2003-01-23-CygwinNotes.txt added (r1.1)
2003-06-25-Reoptimizer1.txt added (r1.1)
2003-06-26-Reoptimizer2.txt added (r1.1)
---
Log message:

Add 2.6



---
Diffs of the changes:  (+2185 -0)

 2000-11-18-EarlyDesignIdeas.txt         |   74 +++++++++
 2000-11-18-EarlyDesignIdeasResp.txt     |  199 +++++++++++++++++++++++++
 2000-12-06-EncodingIdea.txt             |   30 +++
 2000-12-06-MeetingSummary.txt           |   83 ++++++++++
 2001-01-31-UniversalIRIdea.txt          |   39 +++++
 2001-02-06-TypeNotationDebate.txt       |   67 ++++++++
 2001-02-06-TypeNotationDebateResp1.txt  |   75 +++++++++
 2001-02-06-TypeNotationDebateResp2.txt  |   53 ++++++
 2001-02-06-TypeNotationDebateResp4.txt  |   89 +++++++++++
 2001-02-09-AdveComments.txt             |  120 +++++++++++++++
 2001-02-09-AdveCommentsResponse.txt     |  245 ++++++++++++++++++++++++++++++++
 2001-02-13-Reference-Memory.txt         |   39 +++++
 2001-02-13-Reference-MemoryResponse.txt |   47 ++++++
 2001-04-16-DynamicCompilation.txt       |   49 ++++++
 2001-05-18-ExceptionHandling.txt        |  202 ++++++++++++++++++++++++++
 2001-05-19-ExceptionResponse.txt        |   45 +++++
 2001-06-01-GCCOptimizations.txt         |   63 ++++++++
 2001-06-01-GCCOptimizations2.txt        |   71 +++++++++
 2001-06-20-.NET-Differences.txt         |   30 +++
 2001-07-06-LoweringIRForCodeGen.txt     |   31 ++++
 2001-07-08-InstructionSelection.txt     |   51 ++++++
 2001-07-08-InstructionSelection2.txt    |   25 +++
 2001-09-18-OptimizeExceptions.txt       |   56 +++++++
 2002-05-12-InstListChange.txt           |   55 +++++++
 2002-06-25-MegaPatchInfo.txt            |   72 +++++++++
 2003-01-23-CygwinNotes.txt              |   28 +++
 2003-06-25-Reoptimizer1.txt             |  137 +++++++++++++++++
 2003-06-26-Reoptimizer2.txt             |  110 ++++++++++++++
 28 files changed, 2185 insertions(+)


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2000-11-18-EarlyDesignIdeas.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2000-11-18-EarlyDesignIdeas.txt:1.1
*** /dev/null	Fri Oct 23 13:56:36 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2000-11-18-EarlyDesignIdeas.txt	Fri Oct 23 13:56:23 2009
***************
*** 0 ****
--- 1,74 ----
+ Date: Sat, 18 Nov 2000 09:19:35 -0600 (CST)
+ From: Vikram Adve <vadve at cs.uiuc.edu>
+ To: Chris Lattner <lattner at cs.uiuc.edu>
+ Subject: a few thoughts
+ 
+ I've been mulling over the virtual machine problem and I had some
+ thoughts about some things for us to think about discuss:
+ 
+ 1. We need to be clear on our goals for the VM.  Do we want to emphasize
+    portability and safety like the Java VM?  Or shall we focus on the
+    architecture interface first (i.e., consider the code generation and
+    processor issues), since the architecture interface question is also
+    important for portable Java-type VMs?
+ 
+    This is important because the audiences for these two goals are very
+    different.  Architects and many compiler people care much more about
+    the second question.  The Java compiler and OS community care much more
+    about the first one.
+ 
+    Also, while the architecture interface question is important for
+    Java-type VMs, the design constraints are very different.
+ 
+ 
+ 2. Design issues to consider (an initial list that we should continue
+    to modify).  Note that I'm not trying to suggest actual solutions here,
+    but just various directions we can pursue:
+ 
+    a. A single-assignment VM, which we've both already been thinking about.
+ 
+    b. A strongly-typed VM.  One question is do we need the types to be
+       explicitly declared or should they be inferred by the dynamic compiler?
+ 
+    c. How do we get more high-level information into the VM while keeping
+       to a low-level VM design?
+ 
+         o  Explicit array references as operands?  An alternative is
+            to have just an array type, and let the index computations be
+            separate 3-operand instructions.
+ 
+         o  Explicit instructions to handle aliasing, e.g.s:
+            -- an instruction to say "I speculate that these two values are not
+               aliased, but check at runtime", like speculative execution in
+               EPIC?
+            -- or an instruction to check whether two values are aliased and
+               execute different code depending on the answer, somewhat like
+               predicated code in EPIC
+ 
+         o  (This one is a difficult but powerful idea.)
+            A "thread-id" field on every instruction that allows the static
+            compiler to generate a set of parallel threads, and then have
+            the runtime compiler and hardware do what they please with it.
+            This has very powerful uses, but thread-id on every instruction
+            is expensive in terms of instruction size and code size.
+            We would need to compactly encode it somehow.
+ 
+            Also, this will require some reading on at least two other
+            projects:
+                 -- Multiscalar architecture from Wisconsin
+                 -- Simultaneous multithreading architecture from Washington
+ 
+         o  Or forget all this and stick to a traditional instruction set?
+ 
+ 
+ BTW, on an unrelated note, after the meeting yesterday, I did remember
+ that you had suggested doing instruction scheduling on SSA form instead
+ of a dependence DAG earlier in the semester.  When we talked about
+ it yesterday, I didn't remember where the idea had come from but I
+ remembered later.  Just giving credit where its due...
+ 
+ Perhaps you can save the above as a file under RCS so you and I can
+ continue to expand on this.
+ 
+ --Vikram
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2000-11-18-EarlyDesignIdeasResp.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2000-11-18-EarlyDesignIdeasResp.txt:1.1
*** /dev/null	Fri Oct 23 13:57:43 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2000-11-18-EarlyDesignIdeasResp.txt	Fri Oct 23 13:56:24 2009
***************
*** 0 ****
--- 1,199 ----
+ Date: Sun, 19 Nov 2000 16:23:57 -0600 (CST)
+ From: Chris Lattner <sabre at nondot.org>
+ To: Vikram Adve <vadve at cs.uiuc.edu>
+ Subject: Re: a few thoughts
+ 
+ Okay... here are a few of my thoughts on this (it's good to know that we
+ think so alike!):
+ 
+ > 1. We need to be clear on our goals for the VM.  Do we want to emphasize
+ >    portability and safety like the Java VM?  Or shall we focus on the
+ >    architecture interface first (i.e., consider the code generation and
+ >    processor issues), since the architecture interface question is also
+ >    important for portable Java-type VMs?
+ 
+ I forsee the architecture looking kinda like this: (which is completely
+ subject to change)
+ 
+ 1. The VM code is NOT guaranteed safe in a java sense.  Doing so makes it
+    basically impossible to support C like languages.  Besides that,
+    certifying a register based language as safe at run time would be a
+    pretty expensive operation to have to do.  Additionally, we would like
+    to be able to statically eliminate many bounds checks in Java
+    programs... for example.
+ 
+  2. Instead, we can do the following (eventually): 
+    * Java bytecode is used as our "safe" representation (to avoid
+      reinventing something that we don't add much value to).  When the
+      user chooses to execute Java bytecodes directly (ie, not
+      precompiled) the runtime compiler can do some very simple
+      transformations (JIT style) to convert it into valid input for our
+      VM.  Performance is not wonderful, but it works right.
+    * The file is scheduled to be compiled (rigorously) at a later
+      time.  This could be done by some background process or by a second
+      processor in the system during idle time or something...
+    * To keep things "safe" ie to enforce a sandbox on Java/foreign code,
+      we could sign the generated VM code with a host specific private
+      key.  Then before the code is executed/loaded, we can check to see if
+      the trusted compiler generated the code.  This would be much quicker
+      than having to validate consistency (especially if bounds checks have
+      been removed, for example)
+ 
+ >    This is important because the audiences for these two goals are very
+ >    different.  Architects and many compiler people care much more about
+ >    the second question.  The Java compiler and OS community care much more
+ >    about the first one.
+ 
+ 3. By focusing on a more low level virtual machine, we have much more room
+    for value add.  The nice safe "sandbox" VM can be provided as a layer
+    on top of it.  It also lets us focus on the more interesting compilers
+    related projects.
+ 
+ > 2. Design issues to consider (an initial list that we should continue
+ >    to modify).  Note that I'm not trying to suggest actual solutions here,
+ >    but just various directions we can pursue:
+ 
+ Understood.  :)
+ 
+ >    a. A single-assignment VM, which we've both already been thinking
+ >       about.
+ 
+ Yup, I think that this makes a lot of sense.  I am still intrigued,
+ however, by the prospect of a minimally allocated VM representation... I
+ think that it could have definate advantages for certain applications
+ (think very small machines, like PDAs).  I don't, however, think that our
+ initial implementations should focus on this.  :)
+ 
+ Here are some other auxilliary goals that I think we should consider:
+ 
+ 1. Primary goal: Support a high performance dynamic compilation
+    system.  This means that we have an "ideal" division of labor between
+    the runtime and static compilers.  Of course, the other goals of the
+    system somewhat reduce the importance of this point (f.e. portability
+    reduces performance, but hopefully not much)
+ 2. Portability to different processors.  Since we are most familiar with
+    x86 and solaris, I think that these two are excellent candidates when
+    we get that far...
+ 3. Support for all languages & styles of programming (general purpose
+    VM).  This is the point that disallows java style bytecodes, where all
+    array refs are checked for bounds, etc...
+ 4. Support linking between different language families.  For example, call
+    C functions directly from Java without using the nasty/slow/gross JNI
+    layer.  This involves several subpoints:
+   A. Support for languages that require garbage collectors and integration
+      with languages that don't.  As a base point, we could insist on
+      always using a conservative GC, but implement free as a noop, f.e.
+ 
+ >    b. A strongly-typed VM.  One question is do we need the types to be
+ >       explicitly declared or should they be inferred by the dynamic
+ >       compiler?
+ 
+   B. This is kind of similar to another idea that I have: make OOP
+      constructs (virtual function tables, class heirarchies, etc) explicit
+      in the VM representation.  I believe that the number of additional
+      constructs would be fairly low, but would give us lots of important
+      information... something else that would/could be important is to
+      have exceptions as first class types so that they would be handled in
+      a uniform way for the entire VM... so that C functions can call Java
+      functions for example...
+ 
+ >    c. How do we get more high-level information into the VM while keeping
+ >       to a low-level VM design?
+ >       o  Explicit array references as operands?  An alternative is
+ >          to have just an array type, and let the index computations be
+ >          separate 3-operand instructions.
+ 
+    C. In the model I was thinking of (subject to change of course), we
+       would just have an array type (distinct from the pointer
+       types).  This would allow us to have arbitrarily complex index
+       expressions, while still distinguishing "load" from "Array load",
+       for example.  Perhaps also, switch jump tables would be first class
+       types as well?  This would allow better reasoning about the program.
+ 
+ 5. Support dynamic loading of code from various sources.  Already
+    mentioned above was the example of loading java bytecodes, but we want
+    to support dynamic loading of VM code as well.  This makes the job of
+    the runtime compiler much more interesting:  it can do interprocedural
+    optimizations that the static compiler can't do, because it doesn't
+    have all of the required information (for example, inlining from
+    shared libraries, etc...)
+ 
+ 6. Define a set of generally useful annotations to add to the VM
+    representation.  For example, a function can be analysed to see if it
+    has any sideeffects when run... also, the MOD/REF sets could be
+    calculated, etc... we would have to determine what is reasonable.  This
+    would generally be used to make IP optimizations cheaper for the
+    runtime compiler...
+ 
+ >       o  Explicit instructions to handle aliasing, e.g.s:
+ >            -- an instruction to say "I speculate that these two values are not
+ >               aliased, but check at runtime", like speculative execution in
+ >             EPIC?
+ >          -- or an instruction to check whether two values are aliased and
+ >             execute different code depending on the answer, somewhat like
+ >             predicated code in EPIC
+ 
+ These are also very good points... if this can be determined at compile
+ time.  I think that an epic style of representation (not the instruction
+ packing, just the information presented) could be a very interesting model
+ to use... more later...
+ 
+ >         o  (This one is a difficult but powerful idea.)
+ >          A "thread-id" field on every instruction that allows the static
+ >          compiler to generate a set of parallel threads, and then have
+ >          the runtime compiler and hardware do what they please with it.
+ >          This has very powerful uses, but thread-id on every instruction
+ >          is expensive in terms of instruction size and code size.
+ >          We would need to compactly encode it somehow.
+ 
+ Yes yes yes!  :)  I think it would be *VERY* useful to include this kind
+ of information (which EPIC architectures *implicitly* encode.  The trend
+ that we are seeing supports this greatly:
+ 
+ 1. Commodity processors are getting massive SIMD support:
+    * Intel/Amd MMX/MMX2
+    * AMD's 3Dnow!
+    * Intel's SSE/SSE2
+    * Sun's VIS
+ 2. SMP is becoming much more common, especially in the server space.
+ 3. Multiple processors on a die are right around the corner.
+ 
+ If nothing else, not designing this in would severely limit our future
+ expansion of the project...
+ 
+ >          Also, this will require some reading on at least two other
+ >          projects:
+ >               -- Multiscalar architecture from Wisconsin
+ >               -- Simultaneous multithreading architecture from Washington
+ >
+ >       o  Or forget all this and stick to a traditional instruction set?
+ 
+ Heh... :)  Well, from a pure research point of view, it is almost more
+ attactive to go with the most extreme/different ISA possible.  On one axis
+ you get safety and conservatism, and on the other you get degree of
+ influence that the results have.  Of course the problem with pure research
+ is that often times there is no concrete product of the research... :)
+ 
+ > BTW, on an unrelated note, after the meeting yesterday, I did remember
+ > that you had suggested doing instruction scheduling on SSA form instead
+ > of a dependence DAG earlier in the semester.  When we talked about
+ > it yesterday, I didn't remember where the idea had come from but I
+ > remembered later.  Just giving credit where its due...
+ 
+ :) Thanks.  
+ 
+ > Perhaps you can save the above as a file under RCS so you and I can
+ > continue to expand on this.
+ 
+ I think it makes sense to do so when we get our ideas more formalized and
+ bounce it back and forth a couple of times... then I'll do a more formal
+ writeup of our goals and ideas.  Obviously our first implementation will
+ not want to do all of the stuff that I pointed out above... be we will
+ want to design the project so that we do not artificially limit ourselves
+ at sometime in the future...
+ 
+ Anyways, let me know what you think about these ideas... and if they sound
+ reasonable...
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2000-12-06-EncodingIdea.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2000-12-06-EncodingIdea.txt:1.1
*** /dev/null	Fri Oct 23 13:57:43 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2000-12-06-EncodingIdea.txt	Fri Oct 23 13:56:24 2009
***************
*** 0 ****
--- 1,30 ----
+ From: Chris Lattner [mailto:sabre at nondot.org]
+ Sent: Wednesday, December 06, 2000 6:41 PM
+ To: Vikram S. Adve
+ Subject: Additional idea with respect to encoding
+ 
+ Here's another idea with respect to keeping the common case instruction
+ size down (less than 32 bits ideally):
+ 
+ Instead of encoding an instruction to operate on two register numbers,
+ have it operate on two negative offsets based on the current register
+ number.  Therefore, instead of using:
+ 
+ r57 = add r55, r56  (r57 is the implicit dest register, of course)
+ 
+ We could use:
+ 
+ r57 = add -2, -1
+ 
+ My guess is that most SSA references are to recent values (especially if
+ they correspond to expressions like (x+y*z+p*q/ ...), so the negative
+ numbers would tend to stay small, even at the end of the procedure (where
+ the implicit register destination number could be quite large).  Of course
+ the negative sign is reduntant, so you would be storing small integers
+ almost all of the time, and 5-6 bits worth of register number would be
+ plenty for most cases...
+ 
+ What do you think?
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2000-12-06-MeetingSummary.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2000-12-06-MeetingSummary.txt:1.1
*** /dev/null	Fri Oct 23 13:57:43 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2000-12-06-MeetingSummary.txt	Fri Oct 23 13:56:24 2009
***************
*** 0 ****
--- 1,83 ----
+ SUMMARY
+ -------
+ 
+ We met to discuss the LLVM instruction format and bytecode representation:
+ 
+ ISSUES RESOLVED
+ ---------------
+ 
+ 1. We decided that we shall use a flat namespace to represent our 
+    variables in SSA form, as opposed to having a two dimensional namespace
+    of the original variable and the SSA instance subscript.
+ 
+ ARGUMENT AGAINST:
+    * A two dimensional namespace would be valuable when doing alias 
+      analysis because the extra information can help limit the scope of
+      analysis.
+ 
+ ARGUMENT FOR:
+    * Including this information would require that all users of the LLVM
+      bytecode would have to parse and handle it.  This would slow down the
+      common case and inflate the instruction representation with another
+      infinite variable space.
+ 
+ REASONING:
+    * It was decided that because original variable sources could be
+      reconstructed from SSA form in linear time, that it would be an
+      unjustified expense for the common case to include the extra
+      information for one optimization.  Alias analysis itself is typically
+      greater than linear in asymptotic complexity, so this extra analaysis
+      would not affect the runtime of the optimization in a significant
+      way.  Additionally, this would be an unlikely optimization to do at
+      runtime.
+ 
+ 
+ IDEAS TO CONSIDER
+ -----------------
+ 
+ 1. Including dominator information in the LLVM bytecode
+    representation.  This is one example of an analysis result that may be
+    packaged with the bytecodes themselves.  As a conceptual implementation 
+    idea, we could include an immediate dominator number for each basic block
+    in the LLVM bytecode program.  Basic blocks could be numbered according
+    to the order of occurance in the bytecode representation.
+ 
+ 2. Including loop header and body information.  This would facilitate
+    detection of intervals and natural loops.
+ 
+ UNRESOLVED ISSUES 
+ ----------------- 
+ 
+ 1. Will oSUIF provide enough of an infrastructure to support the research
+    that we will be doing?  We know that it has less than stellar
+    performance, but hope that this will be of little importance for our
+    static compiler.  This could affect us if we decided to do some IP
+    research.  Also we do not yet understand the level of exception support
+    currently implemented.
+ 
+ 2. Should we consider the requirements of a direct hardware implementation
+    of the LLVM when we design it?  If so, several design issues should
+    have their priorities shifted.  The other option is to focus on a
+    software layer interpreting the LLVM in all cases.
+ 
+ 3. Should we use some form of packetized format to improve forward
+    compatibility?  For example, we could design the system to encode a
+    packet type and length field before analysis information, to allow a
+    runtime to skip information that it didn't understand in a bytecode
+    stream.  The obvious benefit would be for compatibility, the drawback
+    is that it would tend to splinter that 'standard' LLVM definition.
+ 
+ 4. Should we use fixed length instructions or variable length
+    instructions?  Fetching variable length instructions is expensive (for
+    either hardware or software based LLVM runtimes), but we have several
+    'infinite' spaces that instructions operate in (SSA register numbers,
+    type spaces, or packet length [if packets were implemented]).  Several
+    options were mentioned including: 
+      A. Using 16 or 32 bit numbers, which would be 'big enough'
+      B. A scheme similar to how UTF-8 works, to encode infinite numbers
+         while keeping small number small.
+      C. Use something similar to Huffman encoding, so that the most common
+         numbers are the smallest.
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-01-31-UniversalIRIdea.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-01-31-UniversalIRIdea.txt:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-01-31-UniversalIRIdea.txt	Fri Oct 23 13:56:24 2009
***************
*** 0 ****
--- 1,39 ----
+ Date: Wed, 31 Jan 2001 12:04:33 -0600
+ From: Vikram S. Adve <vadve at cs.uiuc.edu>
+ To: Chris Lattner <lattner at cs.uiuc.edu>
+ Subject: another thought
+ 
+ I have a budding idea about making LLVM a little more ambitious: a
+ customizable runtime system that can be used to implement language-specific
+ virtual machines for many different languages.  E.g., a C vm, a C++ vm, a
+ Java vm, a Lisp vm, ..
+ 
+ The idea would be that LLVM would provide a standard set of runtime features
+ (some low-level like standard assembly instructions with code generation and
+ static and runtime optimization; some higher-level like type-safety and
+ perhaps a garbage collection library).  Each language vm would select the
+ runtime features needed for that language, extending or customizing them as
+ needed.  Most of the machine-dependent code-generation and optimization
+ features as well as low-level machine-independent optimizations (like PRE)
+ could be provided by LLVM and should be sufficient for any language,
+ simplifying the language compiler.  (This would also help interoperability
+ between languages.)  Also, some or most of the higher-level
+ machine-independent features like type-safety and access safety should be
+ reusable by different languages, with minor extensions.  The language
+ compiler could then focus on language-specific analyses and optimizations.
+ 
+ The risk is that this sounds like a universal IR -- something that the
+ compiler community has tried and failed to develop for decades, and is
+ universally skeptical about.  No matter what we say, we won't be able to
+ convince anyone that we have a universal IR that will work.  We need to
+ think about whether LLVM is different or if has something novel that might
+ convince people.  E.g., the idea of providing a package of separable
+ features that different languages select from.  Also, using SSA with or
+ without type-safety as the intermediate representation.
+ 
+ One interesting starting point would be to discuss how a JVM would be
+ implemented on top of LLVM a bit more.  That might give us clues on how to
+ structure LLVM to support one or more language VMs.
+ 
+ --Vikram
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebate.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebate.txt:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebate.txt	Fri Oct 23 13:56:24 2009
***************
*** 0 ****
--- 1,67 ----
+ Date: Tue, 6 Feb 2001 20:27:37 -0600 (CST)
+ From: Chris Lattner <sabre at nondot.org>
+ To: Vikram S. Adve <vadve at cs.uiuc.edu>
+ Subject: Type notation debate...
+ 
+ This is the way that I am currently planning on implementing types:
+ 
+ Primitive Types:        
+ type ::= void|bool|sbyte|ubyte|short|ushort|int|uint|long|ulong
+ 
+ Method:
+ typelist ::= typelisth | /*empty*/
+ typelisth ::= type | typelisth ',' type
+ type ::= type (typelist)
+ 
+ Arrays (without and with size):
+ type ::= '[' type ']' | '[' INT ',' type ']'
+ 
+ Pointer:
+ type ::= type '*'
+ 
+ Structure:
+ type ::= '{' typelist '}'
+ 
+ Packed:
+ type ::= '<' INT ',' type '>'
+ 
+ Simple examples:
+ 
+ [[ %4, int ]]   - array of (array of 4 (int))
+ [ { int, int } ] - Array of structure
+ [ < %4, int > ] - Array of 128 bit SIMD packets
+ int (int, [[int, %4]])  - Method taking a 2d array and int, returning int
+ 
+ 
+ Okay before you comment, please look at:
+ 
+ http://www.research.att.com/~bs/devXinterview.html
+ 
+ Search for "In another interview, you defined the C declarator syntax as
+ an experiment that failed. However, this syntactic construct has been
+ around for 27 years and perhaps more; why do you consider it problematic
+ (except for its cumbersome syntax)?" and read that response for me.  :)
+ 
+ Now with this syntax, his example would be represented as:
+ 
+ [ %10, bool (int, int) * ] *
+ 
+ vs 
+ 
+ bool (*(*)[10])(int, int)
+ 
+ in C.
+ 
+ Basically, my argument for this type construction system is that it is
+ VERY simple to use and understand (although it IS different than C, it is
+ very simple and straightforward, which C is NOT).  In fact, I would assert
+ that most programmers TODAY do not understand pointers to member
+ functions, and have to look up an example when they have to write them.
+ 
+ In my opinion, it is critically important to have clear and concise type
+ specifications, because types are going to be all over the programs.
+ 
+ Let me know your thoughts on this.  :)
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebateResp1.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebateResp1.txt:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebateResp1.txt	Fri Oct 23 13:56:24 2009
***************
*** 0 ****
--- 1,75 ----
+ Date: Thu, 8 Feb 2001 08:42:04 -0600
+ From: Vikram S. Adve <vadve at cs.uiuc.edu>
+ To: Chris Lattner <sabre at nondot.org>
+ Subject: RE: Type notation debate...
+ 
+ Chris,
+ 
+ > Okay before you comment, please look at:
+ >
+ > http://www.research.att.com/~bs/devXinterview.html
+ 
+ I read this argument.  Even before that, I was already in agreement with you
+ and him that the C declarator syntax is difficult and confusing.
+ 
+ But in fact, if you read the entire answer carefully, he came to the same
+ conclusion I do: that you have to go with familiar syntax over logical
+ syntax because familiarity is such a strong force:
+ 
+         "However, familiarity is a strong force. To compare, in English, we
+ live
+ more or less happily with the absurd rules for "to be" (am, are, is, been,
+ was, were, ...) and all attempts to simplify are treated with contempt or
+ (preferably) humor. It be a curious world and it always beed."
+ 
+ > Basically, my argument for this type construction system is that it is
+ > VERY simple to use and understand (although it IS different than C, it is
+ > very simple and straightforward, which C is NOT).  In fact, I would assert
+ > that most programmers TODAY do not understand pointers to member
+ > functions, and have to look up an example when they have to write them.
+ 
+ Again, I don't disagree with this at all.  But to some extent this
+ particular problem is inherently difficult.  Your syntax for the above
+ example may be easier for you to read because this is the way you have been
+ thinking about it.  Honestly, I don't find it much easier than the C syntax.
+ In either case, I would have to look up an example to write pointers to
+ member functions.
+ 
+ But pointers to member functions are nowhere near as common as arrays.  And
+ the old array syntax:
+         type [ int, int, ...]
+ is just much more familiar and clear to people than anything new you
+ introduce, no matter how logical it is.  Introducing a new syntax that may
+ make function pointers easier but makes arrays much more difficult seems
+ very risky to me.
+ 
+ > In my opinion, it is critically important to have clear and concise type
+ > specifications, because types are going to be all over the programs.
+ 
+ I absolutely agree.  But the question is, what is more clear and concise?
+ The syntax programmers are used to out of years of experience or a new
+ syntax that they have never seen that has a more logical structure.  I think
+ the answer is the former.  Sometimes, you have to give up a better idea
+ because you can't overcome sociological barriers to it.  Qwerty keyboards
+ and Windows are two classic examples of bad technology that are difficult to
+ root out.
+ 
+ P.S.  Also, while I agree that most your syntax is more logical, there is
+ one part that isn't:
+ 
+ Arrays (without and with size):
+ type ::= '[' type ']' | '[' INT ',' type ']'.
+ 
+ The arrays with size lists the dimensions and the type in a single list.
+ That is just too confusing:
+         [10, 40, int]
+ This seems to be a 3-D array where the third dimension is something strange.
+ It is too confusing to have a list of 3 things, some of which are dimensions
+ and one is a type.  Either of the following would be better:
+ 
+         array [10, 40] of int
+ or
+         int [10, 40]
+ 
+ --Vikram
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebateResp2.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebateResp2.txt:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebateResp2.txt	Fri Oct 23 13:56:24 2009
***************
*** 0 ****
--- 1,53 ----
+ Date: Thu, 8 Feb 2001 14:31:05 -0600 (CST)
+ From: Chris Lattner <sabre at nondot.org>
+ To: Vikram S. Adve <vadve at cs.uiuc.edu>
+ Subject: RE: Type notation debate...
+ 
+ > Arrays (without and with size):
+ > type ::= '[' type ']' | '[' INT ',' type ']'.
+ > 
+ > The arrays with size lists the dimensions and the type in a single list.
+ > That is just too confusing:
+ 
+ >       [10, 40, int]
+ > This seems to be a 3-D array where the third dimension is something strange.
+ > It is too confusing to have a list of 3 things, some of which are dimensions
+ > and one is a type. 
+ 
+ The above grammar indicates that there is only one integer parameter, ie
+ the upper bound.  The lower bound is always implied to be zero, for
+ several reasons:
+ 
+ * As a low level VM, we want to expose addressing computations
+   explicitly.  Since the lower bound must always be known in a high level
+   language statically, the language front end can do the translation
+   automatically.
+ * This fits more closely with what Java needs, ie what we need in the
+   short term.  Java arrays are always zero based.
+ 
+ If a two element list is too confusing, I would recommend an alternate
+ syntax of:
+ 
+ type ::= '[' type ']' | '[' INT 'x' type ']'.
+ 
+ For example:
+   [12 x int]
+   [12x int]
+   [ 12 x [ 4x int ]]
+ 
+ Which is syntactically nicer, and more explicit.
+ 
+ > Either of the following would be better:
+ >       array [10, 40] of int
+ 
+ I considered this approach for arrays in general (ie array of int/ array
+ of 12 int), but found that it made declarations WAY too long.  Remember
+ that because of the nature of llvm, you get a lot of types strewn all over
+ the program, and using the 'typedef' like facility is not a wonderful
+ option, because then types aren't explicit anymore.
+ 
+ I find this email interesting, because you contradict the previous email
+ you sent, where you recommend that we stick to C syntax....
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebateResp4.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebateResp4.txt:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-06-TypeNotationDebateResp4.txt	Fri Oct 23 13:56:24 2009
***************
*** 0 ****
--- 1,89 ----
+ > But in fact, if you read the entire answer carefully, he came to the same
+ > conclusion I do: that you have to go with familiar syntax over logical
+ > syntax because familiarity is such a strong force:
+ >       "However, familiarity is a strong force. To compare, in English, we
+ live
+ > more or less happily with the absurd rules for "to be" (am, are, is, been,
+ > was, were, ...) and all attempts to simplify are treated with contempt or
+ > (preferably) humor. It be a curious world and it always beed."
+ 
+ Although you have to remember that his situation was considerably
+ different than ours.  He was in a position where he was designing a high
+ level language that had to be COMPATIBLE with C.  Our language is such
+ that a new person would have to learn the new, different, syntax
+ anyways.  Making them learn about the type system does not seem like much
+ of a stretch from learning the opcodes and how SSA form works, and how
+ everything ties together...
+ 
+ > > Basically, my argument for this type construction system is that it is
+ > > VERY simple to use and understand (although it IS different than C, it is
+ > > very simple and straightforward, which C is NOT).  In fact, I would assert
+ > > that most programmers TODAY do not understand pointers to member
+ > > functions, and have to look up an example when they have to write them.
+ 
+ > Again, I don't disagree with this at all.  But to some extent this
+ > particular problem is inherently difficult.  Your syntax for the above
+ > example may be easier for you to read because this is the way you have been
+ > thinking about it.  Honestly, I don't find it much easier than the C syntax.
+ > In either case, I would have to look up an example to write pointers to
+ > member functions.
+ 
+ I would argue that because the lexical structure of the language is self
+ consistent, any person who spent a significant amount of time programming
+ in LLVM directly would understand how to do it without looking it up in a
+ manual.  The reason this does not work for C is because you rarely have to
+ declare these pointers, and the syntax is inconsistent with the method
+ declaration and calling syntax.
+ 
+ > But pointers to member functions are nowhere near as common as arrays.
+ 
+ Very true.  If you're implementing an object oriented language, however,
+ remember that you have to do all the pointer to member function stuff
+ yourself.... so everytime you invoke a virtual method one is involved
+ (instead of having C++ hide it for you behind "syntactic sugar").
+ 
+ > And the old array syntax:
+ >       type [ int, int, ...]
+ > is just much more familiar and clear to people than anything new you
+ > introduce, no matter how logical it is.  
+ 
+ Erm... excuse me but how is this the "old array syntax"?  If you are
+ arguing for consistency with C, you should be asking for 'type int []',
+ which is significantly different than the above (beside the above
+ introduces a new operator and duplicates information
+ needlessly).  Basically what I am suggesting is exactly the above without
+ the fluff.  So instead of:
+ 
+        type [ int, int, ...]
+ 
+ you use:
+ 
+        type [ int ]
+ 
+ > Introducing a new syntax that may
+ > make function pointers easier but makes arrays much more difficult seems
+ > very risky to me.
+ 
+ This is not about function pointers.  This is about consistency in the
+ type system, and consistency with the rest of the language.  The point
+ above does not make arrays any more difficult to use, and makes the
+ structure of types much more obvious than the "c way".
+ 
+ > > In my opinion, it is critically important to have clear and concise type
+ > > specifications, because types are going to be all over the programs.
+ > 
+ > I absolutely agree.  But the question is, what is more clear and concise?
+ > The syntax programmers are used to out of years of experience or a new
+ > syntax that they have never seen that has a more logical structure.  I think
+ > the answer is the former.  Sometimes, you have to give up a better idea
+ > because you can't overcome sociological barriers to it.  Qwerty keyboards
+ > and Windows are two classic examples of bad technology that are difficult to
+ > root out.
+ 
+ Very true, but you seem to be advocating a completely different Type
+ system than C has, in addition to it not offering the advantages of clear
+ structure that the system I recommended does... so you seem to not have a
+ problem with changing this, just with what I change it to.  :)
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-09-AdveComments.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-09-AdveComments.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-09-AdveComments.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,120 ----
+ Ok, here are my comments and suggestions about the LLVM instruction set.
+ We should discuss some now, but can discuss many of them later, when we
+ revisit synchronization, type inference, and other issues.
+ (We have discussed some of the comments already.)
+ 
+ 
+ o  We should consider eliminating the type annotation in cases where it is
+    essentially obvious from the instruction type, e.g., in br, it is obvious
+    that the first arg. should be a bool and the other args should be labels:
+ 
+ 	br bool <cond>, label <iftrue>, label <iffalse>
+ 
+    I think your point was that making all types explicit improves clarity
+    and readability.  I agree to some extent, but it also comes at the cost
+    of verbosity.  And when the types are obvious from people's experience
+    (e.g., in the br instruction), it doesn't seem to help as much.
+ 
+ 
+ o  On reflection, I really like your idea of having the two different switch
+    types (even though they encode implementation techniques rather than
+    semantics).  It should simplify building the CFG and my guess is it could
+    enable some significant optimizations, though we should think about which.
+ 
+ 
+ o  In the lookup-indirect form of the switch, is there a reason not to make
+    the val-type uint?  Most HLL switch statements (including Java and C++)
+    require that anyway.  And it would also make the val-type uniform 
+    in the two forms of the switch.
+ 
+    I did see the switch-on-bool examples and, while cute, we can just use
+    the branch instructions in that particular case.
+ 
+ 
+ o  I agree with your comment that we don't need 'neg'.
+ 
+ 
+ o  There's a trade-off with the cast instruction:
+    +  it avoids having to define all the upcasts and downcasts that are
+       valid for the operands of each instruction  (you probably have thought
+       of other benefits also)
+    -  it could make the bytecode significantly larger because there could
+       be a lot of cast operations
+ 
+ 
+ o  Making the second arg. to 'shl' a ubyte seems good enough to me.
+    255 positions seems adequate for several generations of machines
+    and is more compact than uint.
+ 
+ 
+ o  I still have some major concerns about including malloc and free in the
+    language (either as builtin functions or instructions).  LLVM must be
+    able to represent code from many different languages.  Languages such as
+    C, C++ Java and Fortran 90 would not be able to use our malloc anyway
+    because each of them will want to provide a library implementation of it.
+ 
+    This gets even worse when code from different languages is linked
+    into a single executable (which is fairly common in large apps).
+    Having a single malloc would just not suffice, and instead would simply
+    complicate the picture further because it adds an extra variant in
+    addition to the one each language provides.
+ 
+    Instead, providing a default library version of malloc and free
+    (and perhaps a malloc_gc with garbage collection instead of free)
+    would make a good implementation available to anyone who wants it.
+ 
+    I don't recall all your arguments in favor so let's discuss this again,
+    and soon.
+ 
+ 
+ o  'alloca' on the other hand sounds like a good idea, and the
+    implementation seems fairly language-independent so it doesn't have the
+    problems with malloc listed above.
+ 
+ 
+ o  About indirect call:
+    Your option #2 sounded good to me.  I'm not sure I understand your
+    concern about an explicit 'icall' instruction?
+ 
+ 
+ o  A pair of important synchronization instr'ns to think about:
+      load-linked
+      store-conditional
+ 
+ 
+ o  Other classes of instructions that are valuable for pipeline performance:
+      conditional-move		 
+      predicated instructions
+ 
+ 
+ o  I believe tail calls are relatively easy to identify; do you know why
+    .NET has a tailcall instruction?
+ 
+ 
+ o  I agree that we need a static data space.  Otherwise, emulating global
+    data gets unnecessarily complex.
+ 
+ 
+ o  About explicit parallelism:
+ 
+    We once talked about adding a symbolic thread-id field to each
+    instruction.  (It could be optional so single-threaded codes are
+    not penalized.)  This could map well to multi-threaded architectures
+    while providing easy ILP for single-threaded onces.  But it is probably
+    too radical an idea to include in a base version of LLVM.  Instead, it
+    could a great topic for a separate study.
+ 
+    What is the semantics of the IA64 stop bit?
+ 
+ 
+ 
+ 
+ o  And finally, another thought about the syntax for arrays :-)
+ 
+    Although this syntax:
+ 	  array <dimension-list> of <type>
+    is verbose, it will be used only in the human-readable assembly code so
+    size should not matter.  I think we should consider it because I find it
+    to be the clearest syntax.  It could even make arrays of function
+    pointers somewhat readable.
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-09-AdveCommentsResponse.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-09-AdveCommentsResponse.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-09-AdveCommentsResponse.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,245 ----
+ From: Chris Lattner <sabre at nondot.org>
+ To: "Vikram S. Adve" <vadve at cs.uiuc.edu>
+ Subject: Re: LLVM Feedback
+ 
+ I've included your feedback in the /home/vadve/lattner/llvm/docs directory
+ so that it will live in CVS eventually with the rest of LLVM.  I've
+ significantly updated the documentation to reflect the changes you
+ suggested, as specified below:
+ 
+ > We should consider eliminating the type annotation in cases where it is
+ > essentially obvious from the instruction type:
+ >        br bool <cond>, label <iftrue>, label <iffalse>
+ > I think your point was that making all types explicit improves clarity
+ > and readability.  I agree to some extent, but it also comes at the
+ > cost of verbosity.  And when the types are obvious from people's
+ > experience (e.g., in the br instruction), it doesn't seem to help as
+ > much.
+ 
+ Very true.  We should discuss this more, but my reasoning is more of a
+ consistency argument.  There are VERY few instructions that can have all
+ of the types eliminated, and doing so when available unnecesarily makes
+ the language more difficult to handle.  Especially when you see 'int
+ %this' and 'bool %that' all over the place, I think it would be
+ disorienting to see:
+ 
+   br %predicate, %iftrue, %iffalse
+ 
+ for branches.  Even just typing that once gives me the creeps. ;)  Like I
+ said, we should probably discuss this further in person...
+ 
+ > On reflection, I really like your idea of having the two different
+ > switch types (even though they encode implementation techniques rather
+ > than semantics).  It should simplify building the CFG and my guess is it
+ > could enable some significant optimizations, though we should think
+ > about which.
+ 
+ Great.  I added a note to the switch section commenting on how the VM
+ should just use the instruction type as a hint, and that the
+ implementation may choose altermate representations (such as predicated
+ branches).
+ 
+ > In the lookup-indirect form of the switch, is there a reason not to
+ > make the val-type uint?
+ 
+ No.  This was something I was debating for a while, and didn't really feel
+ strongly about either way.  It is common to switch on other types in HLL's
+ (for example signed int's are particually common), but in this case, all
+ that will be added is an additional 'cast' instruction.  I removed that
+ from the spec.
+ 
+ > I agree with your comment that we don't need 'neg'
+ 
+ Removed.
+ 
+ > There's a trade-off with the cast instruction:
+ >  +  it avoids having to define all the upcasts and downcasts that are
+ >     valid for the operands of each instruction  (you probably have
+ >     thought of other benefits also)
+ >  -  it could make the bytecode significantly larger because there could
+ >     be a lot of cast operations
+ 
+  + You NEED casts to represent things like:
+     void foo(float);
+     ...
+     int x;
+     ...
+     foo(x);
+    in a language like C.  Even in a Java like language, you need upcasts
+    and some way to implement dynamic downcasts.
+  + Not all forms of instructions take every type (for example you can't
+    shift by a floating point number of bits), thus SOME programs will need
+    implicit casts.
+ 
+ To be efficient and to avoid your '-' point above, we just have to be
+ careful to specify that the instructions shall operate on all common
+ types, therefore casting should be relatively uncommon.  For example all
+ of the arithmetic operations work on almost all data types.
+ 
+ > Making the second arg. to 'shl' a ubyte seems good enough to me.
+ > 255 positions seems adequate for several generations of machines
+ 
+ Okay, that comment is removed.
+ 
+ > and is more compact than uint.
+ 
+ No, it isn't.  Remember that the bytecode encoding saves value slots into
+ the bytecode instructions themselves, not constant values.  This is
+ another case where we may introduce more cast instructions (but we will
+ also reduce the number of opcode variants that must be supported by a
+ virtual machine).  Because most shifts are by constant values, I don't
+ think that we'll have to cast many shifts.  :)
+ 
+ > I still have some major concerns about including malloc and free in the
+ > language (either as builtin functions or instructions).
+ 
+ Agreed.  How about this proposal:
+ 
+ malloc/free are either built in functions or actual opcodes.  They provide
+ all of the type safety that the document would indicate, blah blah
+ blah. :)
+ 
+ Now, because of all of the excellent points that you raised, an
+ implementation may want to override the default malloc/free behavior of
+ the program.  To do this, they simply implement a "malloc" and
+ "free" function.  The virtual machine will then be defined to use the user
+ defined malloc/free function (which return/take void*'s, not type'd
+ pointers like the builtin function would) if one is available, otherwise
+ fall back on a system malloc/free.
+ 
+ Does this sound like a good compromise?  It would give us all of the
+ typesafety/elegance in the language while still allowing the user to do
+ all the cool stuff they want to...
+ 
+ >  'alloca' on the other hand sounds like a good idea, and the
+ >  implementation seems fairly language-independent so it doesn't have the
+ >  problems with malloc listed above.
+ 
+ Okay, once we get the above stuff figured out, I'll put it all in the
+ spec.
+ 
+ >  About indirect call:
+ >  Your option #2 sounded good to me.  I'm not sure I understand your
+ >  concern about an explicit 'icall' instruction?
+ 
+ I worry too much.  :)  The other alternative has been removed. 'icall' is
+ now up in the instruction list next to 'call'.
+ 
+ > I believe tail calls are relatively easy to identify; do you know why
+ > .NET has a tailcall instruction?
+ 
+ Although I am just guessing, I believe it probably has to do with the fact
+ that they want languages like Haskell and lisp to be efficiently runnable
+ on their VM.  Of course this means that the VM MUST implement tail calls
+ 'correctly', or else life will suck.  :)  I would put this into a future
+ feature bin, because it could be pretty handy...
+ 
+ >  A pair of important synchronization instr'ns to think about:
+ >    load-linked
+ >    store-conditional
+ 
+ What is 'load-linked'?  I think that (at least for now) I should add these
+ to the 'possible extensions' section, because they are not immediately
+ needed...
+ 
+ > Other classes of instructions that are valuable for pipeline
+ > performance:
+ >    conditional-move            
+ >    predicated instructions
+ 
+ Conditional move is effectly a special case of a predicated
+ instruction... and I think that all predicated instructions can possibly
+ be implemented later in LLVM.  It would significantly change things, and
+ it doesn't seem to be very necessary right now.  It would seem to
+ complicate flow control analysis a LOT in the virtual machine.  I would
+ tend to prefer that a predicated architecture like IA64 convert from a
+ "basic block" representation to a predicated rep as part of it's dynamic
+ complication phase.  Also, if a basic block contains ONLY a move, then
+ that can be trivally translated into a conditional move...
+ 
+ > I agree that we need a static data space.  Otherwise, emulating global
+ > data gets unnecessarily complex.
+ 
+ Definately.  Also a later item though.  :)
+ 
+ > We once talked about adding a symbolic thread-id field to each
+ > ..
+ > Instead, it could a great topic for a separate study.
+ 
+ Agreed.  :)
+ 
+ > What is the semantics of the IA64 stop bit?
+ 
+ Basically, the IA64 writes instructions like this:
+ mov ...
+ add ...
+ sub ...
+ op xxx
+ op xxx
+ ;;
+ mov ...
+ add ...
+ sub ...
+ op xxx
+ op xxx
+ ;;
+ 
+ Where the ;; delimits a group of instruction with no dependencies between
+ them, which can all be executed concurrently (to the limits of the
+ available functional units).  The ;; gets translated into a bit set in one
+ of the opcodes.
+ 
+ The advantages of this representation is that you don't have to do some
+ kind of 'thread id scheduling' pass by having to specify ahead of time how
+ many threads to use, and the representation doesn't have a per instruction
+ overhead...
+ 
+ > And finally, another thought about the syntax for arrays :-)
+ >  Although this syntax:
+ >         array <dimension-list> of <type>
+ >  is verbose, it will be used only in the human-readable assembly code so
+ >  size should not matter.  I think we should consider it because I find it
+ >  to be the clearest syntax.  It could even make arrays of function
+ >  pointers somewhat readable.
+ 
+ My only comment will be to give you an example of why this is a bad
+ idea.  :)
+ 
+ Here is an example of using the switch statement (with my recommended
+ syntax):
+ 
+ switch uint %val, label %otherwise, 
+        [%3 x {uint, label}] [ { uint %57, label %l1 }, 
+                               { uint %20, label %l2 }, 
+                               { uint %14, label %l3 } ]
+ 
+ Here it is with the syntax you are proposing:
+ 
+ switch uint %val, label %otherwise, 
+        array %3 of {uint, label} 
+               array of {uint, label}
+                               { uint %57, label %l1 },
+                               { uint %20, label %l2 },
+                               { uint %14, label %l3 }
+ 
+ Which is ambiguous and very verbose. It would be possible to specify
+ constants with [] brackets as in my syntax, which would look like this:
+ 
+ switch uint %val, label %otherwise,
+        array %3 of {uint, label}  [ { uint %57, label %l1 },
+                                     { uint %20, label %l2 },
+                                     { uint %14, label %l3 } ]
+ 
+ But then the syntax is inconsistent between type definition and constant
+ definition (why do []'s enclose the constants but not the types??).  
+ 
+ Anyways, I'm sure that there is much debate still to be had over
+ this... :)
+ 
+ -Chris
+ 
+ http://www.nondot.org/~sabre/os/
+ http://www.nondot.org/MagicStats/
+ http://korbit.sourceforge.net/
+ 
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-13-Reference-Memory.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-13-Reference-Memory.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-13-Reference-Memory.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,39 ----
+ Date: Tue, 13 Feb 2001 13:29:52 -0600 (CST)
+ From: Chris Lattner <sabre at nondot.org>
+ To: Vikram S. Adve <vadve at cs.uiuc.edu>
+ Subject: LLVM Concerns...
+ 
+ 
+ I've updated the documentation to include load store and allocation
+ instructions (please take a look and let me know if I'm on the right
+ track):
+ 
+ file:/home/vadve/lattner/llvm/docs/LangRef.html#memoryops
+ 
+ I have a couple of concerns I would like to bring up:
+ 
+ 1. Reference types
+    Right now, I've spec'd out the language to have a pointer type, which
+    works fine for lots of stuff... except that Java really has
+    references: constrained pointers that cannot be manipulated: added and
+    subtracted, moved, etc... Do we want to have a type like this?  It
+    could be very nice for analysis (pointer always points to the start of
+    an object, etc...) and more closely matches Java semantics.  The
+    pointer type would be kept for C++ like semantics.  Through analysis,
+    C++ pointers could be promoted to references in the LLVM
+    representation.
+ 
+ 2. Our "implicit" memory references in assembly language:
+    After thinking about it, this model has two problems:
+       A. If you do pointer analysis and realize that two stores are
+          independent and can share the same memory source object, there is
+          no way to represent this in either the bytecode or assembly.
+       B. When parsing assembly/bytecode, we effectively have to do a full
+          SSA generation/PHI node insertion pass to build the dependencies
+          when we don't want the "pinned" representation.  This is not
+          cool.
+    I'm tempted to make memory references explicit in both the assembly and
+    bytecode to get around this... what do you think?
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-13-Reference-MemoryResponse.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-13-Reference-MemoryResponse.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-02-13-Reference-MemoryResponse.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,47 ----
+ Date: Tue, 13 Feb 2001 18:25:42 -0600
+ From: Vikram S. Adve <vadve at cs.uiuc.edu>
+ To: Chris Lattner <sabre at nondot.org>
+ Subject: RE: LLVM Concerns...
+ 
+ > 1. Reference types
+ >    Right now, I've spec'd out the language to have a pointer type, which
+ >    works fine for lots of stuff... except that Java really has
+ >    references: constrained pointers that cannot be manipulated: added and
+ >    subtracted, moved, etc... Do we want to have a type like this?  It
+ >    could be very nice for analysis (pointer always points to the start of
+ >    an object, etc...) and more closely matches Java semantics.  The
+ >    pointer type would be kept for C++ like semantics.  Through analysis,
+ >    C++ pointers could be promoted to references in the LLVM
+ >    representation.
+ 
+ 
+ You're right, having references would be useful.  Even for C++ the *static*
+ compiler could generate references instead of pointers with fairly
+ straightforward analysis.  Let's include a reference type for now.  But I'm
+ also really concerned that LLVM is becoming big and complex and (perhaps)
+ too high-level.  After we get some initial performance results, we may have
+ a clearer idea of what our goals should be and we should revisit this
+ question then.
+ 
+ > 2. Our "implicit" memory references in assembly language:
+ >    After thinking about it, this model has two problems:
+ >       A. If you do pointer analysis and realize that two stores are
+ >          independent and can share the same memory source object,
+ 
+ not sure what you meant by "share the same memory source object"
+ 
+ > there is
+ >          no way to represent this in either the bytecode or assembly.
+ >       B. When parsing assembly/bytecode, we effectively have to do a full
+ >          SSA generation/PHI node insertion pass to build the dependencies
+ >          when we don't want the "pinned" representation.  This is not
+ >          cool.
+ 
+ I understand the concern.  But again, let's focus on the performance first
+ and then look at the language design issues.  E.g., it would be good to know
+ how big the bytecode files are before expanding them further.  I am pretty
+ keen to explore the implications of LLVM for mobile devices.  Both bytecode
+ size and power consumption are important to consider there.
+ 
+ --Vikram
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-04-16-DynamicCompilation.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-04-16-DynamicCompilation.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-04-16-DynamicCompilation.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,49 ----
+ By Chris:
+ 
+ LLVM has been designed with two primary goals in mind.  First we strive to 
+ enable the best possible division of labor between static and dynamic 
+ compilers, and second, we need a flexible and powerful interface 
+ between these two complementary stages of compilation.  We feel that 
+ providing a solution to these two goals will yield an excellent solution 
+ to the performance problem faced by modern architectures and programming 
+ languages.
+ 
+ A key insight into current compiler and runtime systems is that a 
+ compiler may fall in anywhere in a "continuum of compilation" to do its 
+ job.  On one side, scripting languages statically compile nothing and 
+ dynamically compile (or equivalently, interpret) everything.  On the far 
+ other side, traditional static compilers process everything statically and 
+ nothing dynamically.  These approaches have typically been seen as a 
+ tradeoff between performance and portability.  On a deeper level, however, 
+ there are two reasons that optimal system performance may be obtained by a
+ system somewhere in between these two extremes: Dynamic application 
+ behavior and social constraints.
+ 
+ From a technical perspective, pure static compilation cannot ever give 
+ optimal performance in all cases, because applications have varying dynamic
+ behavior that the static compiler cannot take into consideration.  Even 
+ compilers that support profile guided optimization generate poor code in 
+ the real world, because using such optimization tunes that application 
+ to one particular usage pattern, whereas real programs (as opposed to 
+ benchmarks) often have several different usage patterns.
+ 
+ On a social level, static compilation is a very shortsighted solution to 
+ the performance problem.  Instruction set architectures (ISAs) continuously 
+ evolve, and each implementation of an ISA (a processor) must choose a set 
+ of tradeoffs that make sense in the market context that it is designed for.  
+ With every new processor introduced, the vendor faces two fundamental 
+ problems: First, there is a lag time between when a processor is introduced 
+ to when compilers generate quality code for the architecture.  Secondly, 
+ even when compilers catch up to the new architecture there is often a large 
+ body of legacy code that was compiled for previous generations and will 
+ not or can not be upgraded.  Thus a large percentage of code running on a 
+ processor may be compiled quite sub-optimally for the current 
+ characteristics of the dynamic execution environment.
+ 
+ For these reasons, LLVM has been designed from the beginning as a long-term 
+ solution to these problems.  Its design allows the large body of platform 
+ independent, static, program optimizations currently in compilers to be 
+ reused unchanged in their current form.  It also provides important static 
+ type information to enable powerful dynamic and link time optimizations 
+ to be performed quickly and efficiently.  This combination enables an 
+ increase in effective system performance for real world environments.


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-05-18-ExceptionHandling.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-05-18-ExceptionHandling.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-05-18-ExceptionHandling.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,202 ----
+ Meeting notes: Implementation idea: Exception Handling in C++/Java
+ 
+ The 5/18/01 meeting discussed ideas for implementing exceptions in LLVM.
+ We decided that the best solution requires a set of library calls provided by
+ the VM, as well as an extension to the LLVM function invocation syntax.
+ 
+ The LLVM function invocation instruction previously looks like this (ignoring
+ types):
+ 
+   call func(arg1, arg2, arg3)
+ 
+ The extension discussed today adds an optional "with" clause that 
+ associates a label with the call site.  The new syntax looks like this:
+ 
+   call func(arg1, arg2, arg3) with funcCleanup
+ 
+ This funcHandler always stays tightly associated with the call site (being
+ encoded directly into the call opcode itself), and should be used whenever
+ there is cleanup work that needs to be done for the current function if 
+ an exception is thrown by func (or if we are in a try block).
+ 
+ To support this, the VM/Runtime provide the following simple library 
+ functions (all syntax in this document is very abstract):
+ 
+ typedef struct { something } %frame;
+   The VM must export a "frame type", that is an opaque structure used to 
+   implement different types of stack walking that may be used by various
+   language runtime libraries. We imagine that it would be typical to 
+   represent a frame with a PC and frame pointer pair, although that is not 
+   required.
+ 
+ %frame getStackCurrentFrame();
+   Get a frame object for the current function.  Note that if the current
+   function was inlined into its caller, the "current" frame will belong to
+   the "caller".
+ 
+ bool isFirstFrame(%frame f);
+   Returns true if the specified frame is the top level (first activated) frame
+   for this thread.  For the main thread, this corresponds to the main() 
+   function, for a spawned thread, it corresponds to the thread function.
+ 
+ %frame getNextFrame(%frame f);
+   Return the previous frame on the stack.  This function is undefined if f
+   satisfies the predicate isFirstFrame(f).
+ 
+ Label *getFrameLabel(%frame f);
+   If a label was associated with f (as discussed below), this function returns
+   it.  Otherwise, it returns a null pointer.
+ 
+ doNonLocalBranch(Label *L);
+   At this point, it is not clear whether this should be a function or 
+   intrinsic.  It should probably be an intrinsic in LLVM, but we'll deal with
+   this issue later.
+ 
+ 
+ Here is a motivating example that illustrates how these facilities could be
+ used to implement the C++ exception model:
+ 
+ void TestFunction(...) {
+   A a; B b;
+   foo();        // Any function call may throw
+   bar();
+   C c;
+ 
+   try {
+     D d;
+     baz();
+   } catch (int) {
+     ...int Stuff...
+     // execution continues after the try block: the exception is consumed
+   } catch (double) {
+     ...double stuff...
+    throw;            // Exception is propogated
+   }
+ }
+ 
+ This function would compile to approximately the following code (heavy 
+ pseudo code follows):
+ 
+ Func:
+   %a = alloca A
+   A::A(%a)        // These ctors & dtors could throw, but we ignore this 
+   %b = alloca B   // minor detail for this example
+   B::B(%b)
+ 
+   call foo() with fooCleanup // An exception in foo is propogated to fooCleanup
+   call bar() with barCleanup // An exception in bar is propogated to barCleanup
+ 
+   %c = alloca C
+   C::C(c)
+   %d = alloca D
+   D::D(d)
+   call baz() with bazCleanup // An exception in baz is propogated to bazCleanup
+   d->~D();
+ EndTry:                   // This label corresponds to the end of the try block
+   c->~C()       // These could also throw, these are also ignored
+   b->~B()
+   a->~A()
+   return
+ 
+ Note that this is a very straight forward and literal translation: exactly
+ what we want for zero cost (when unused) exception handling.  Especially on
+ platforms with many registers (ie, the IA64) setjmp/longjmp style exception
+ handling is *very* impractical.  Also, the "with" clauses describe the 
+ control flow paths explicitly so that analysis is not adversly effected.
+ 
+ The foo/barCleanup labels are implemented as:
+ 
+ TryCleanup:          // Executed if an exception escapes the try block  
+   c->~C()
+ barCleanup:          // Executed if an exception escapes from bar()
+   // fall through
+ fooCleanup:          // Executed if an exception escapes from foo()
+   b->~B()
+   a->~A()
+   Exception *E = getThreadLocalException()
+   call throw(E)      // Implemented by the C++ runtime, described below
+ 
+ Which does the work one would expect.  getThreadLocalException is a function
+ implemented by the C++ support library.  It returns the current exception 
+ object for the current thread.  Note that we do not attempt to recycle the 
+ shutdown code from before, because performance of the mainline code is 
+ critically important.  Also, obviously fooCleanup and barCleanup may be 
+ merged and one of them eliminated.  This just shows how the code generator 
+ would most likely emit code.
+ 
+ The bazCleanup label is more interesting.  Because the exception may be caught
+ by the try block, we must dispatch to its handler... but it does not exist
+ on the call stack (it does not have a VM Call->Label mapping installed), so 
+ we must dispatch statically with a goto.  The bazHandler thus appears as:
+ 
+ bazHandler:
+   d->~D();    // destruct D as it goes out of scope when entering catch clauses
+   goto TryHandler
+ 
+ In general, TryHandler is not the same as bazHandler, because multiple 
+ function calls could be made from the try block.  In this case, trivial 
+ optimization could merge the two basic blocks.  TryHandler is the code 
+ that actually determines the type of exception, based on the Exception object
+ itself.  For this discussion, assume that the exception object contains *at
+ least*:
+ 
+ 1. A pointer to the RTTI info for the contained object
+ 2. A pointer to the dtor for the contained object
+ 3. The contained object itself
+ 
+ Note that it is necessary to maintain #1 & #2 in the exception object itself
+ because objects without virtual function tables may be thrown (as in this 
+ example).  Assuming this, TryHandler would look something like this:
+ 
+ TryHandler: 
+   Exception *E = getThreadLocalException();
+   switch (E->RTTIType) {
+   case IntRTTIInfo:
+     ...int Stuff...       // The action to perform from the catch block
+     break;
+   case DoubleRTTIInfo:
+     ...double Stuff...    // The action to perform from the catch block
+     goto TryCleanup       // This catch block rethrows the exception
+     break;                // Redundant, eliminated by the optimizer
+   default:
+     goto TryCleanup       // Exception not caught, rethrow
+   }
+ 
+   // Exception was consumed
+   if (E->dtor)
+     E->dtor(E->object)    // Invoke the dtor on the object if it exists
+   goto EndTry             // Continue mainline code...
+ 
+ And that is all there is to it.
+ 
+ The throw(E) function would then be implemented like this (which may be 
+ inlined into the caller through standard optimization):
+ 
+ function throw(Exception *E) {
+   // Get the start of the stack trace...
+   %frame %f = call getStackCurrentFrame()
+ 
+   // Get the label information that corresponds to it
+   label * %L = call getFrameLabel(%f)
+   while (%L == 0 && !isFirstFrame(%f)) {
+     // Loop until a cleanup handler is found
+     %f = call getNextFrame(%f)
+     %L = call getFrameLabel(%f)
+   }
+ 
+   if (%L != 0) {
+     call setThreadLocalException(E)   // Allow handlers access to this...
+     call doNonLocalBranch(%L)
+   }
+   // No handler found!
+   call BlowUp()         // Ends up calling the terminate() method in use
+ }
+ 
+ That's a brief rundown of how C++ exception handling could be implemented in
+ llvm.  Java would be very similar, except it only uses destructors to unlock
+ synchronized blocks, not to destroy data.  Also, it uses two stack walks: a
+ nondestructive walk that builds a stack trace, then a destructive walk that
+ unwinds the stack as shown here. 
+ 
+ It would be trivial to get exception interoperability between C++ and Java.
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-05-19-ExceptionResponse.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-05-19-ExceptionResponse.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-05-19-ExceptionResponse.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,45 ----
+ Date: Sat, 19 May 2001 19:09:13 -0500 (CDT)
+ From: Chris Lattner <sabre at nondot.org>
+ To: Vikram S. Adve <vadve at cs.uiuc.edu>
+ Subject: RE: Meeting writeup
+ 
+ > I read it through and it looks great!
+ 
+ Thanks!
+ 
+ > The finally clause in Java may need more thought.  The code for this clause
+ > is like a subroutine because it needs to be entered from many points (end of
+ > try block and beginning of each catch block), and then needs to *return to
+ > the place from where the code was entered*.  That's why JVM has the
+ > jsr/jsr_w instruction.
+ 
+ Hrm... I guess that is an implementation decision.  It can either be
+ modelled as a subroutine (as java bytecodes do), which is really
+ gross... or it can be modelled as code duplication (emitted once inline,
+ then once in the exception path).  Because this could, at worst,
+ slightly less than double the amount of code in a function (it is
+ bounded) I don't think this is a big deal.  One of the really nice things
+ about the LLVM representation is that it still allows for runtime code
+ generation for exception paths (exceptions paths are not compiled until
+ needed).  Obviously a static compiler couldn't do this though.  :)
+ 
+ In this case, only one copy of the code would be compiled... until the
+ other one is needed on demand.  Also this strategy fits with the "zero
+ cost" exception model... the standard case is not burdened with extra
+ branches or "call"s.
+ 
+ > I suppose you could save the return address in a particular register
+ > (specific to this finally block), jump to the finally block, and then at the
+ > end of the finally block, jump back indirectly through this register.  It
+ > will complicate building the CFG but I suppose that can be handled.  It is
+ > also unsafe in terms of checking where control returns (which is I suppose
+ > why the JVM doesn't use this).
+ 
+ I think that a code duplication method would be cleaner, and would avoid
+ the caveats that you mention.  Also, it does not slow down the normal case
+ with an indirect branch...
+ 
+ Like everything, we can probably defer a final decision until later.  :)
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-06-01-GCCOptimizations.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-06-01-GCCOptimizations.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-06-01-GCCOptimizations.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,63 ----
+ Date: Fri, 1 Jun 2001 16:38:17 -0500 (CDT)
+ From: Chris Lattner <sabre at nondot.org>
+ To: Vikram S. Adve <vadve at cs.uiuc.edu>
+ Subject: Interesting: GCC passes
+ 
+ 
+ Take a look at this document (which describes the order of optimizations
+ that GCC performs):
+ 
+ http://gcc.gnu.org/onlinedocs/gcc_17.html
+ 
+ The rundown is that after RTL generation, the following happens:
+ 
+ 1 . [t] jump optimization (jumps to jumps, etc)
+ 2 . [t] Delete unreachable code
+ 3 .     Compute live ranges for CSE
+ 4 . [t] Jump threading (jumps to jumps with identical or inverse conditions)
+ 5 . [t] CSE
+ 6 . *** Conversion to SSA 
+ 7 . [t] SSA Based DCE
+ 8 . *** Conversion to LLVM
+ 9 .     UnSSA
+ 10.     GCSE
+ 11.     LICM
+ 12.     Strength Reduction
+ 13.     Loop unrolling
+ 14. [t] CSE
+ 15. [t] DCE
+ 16.     Instruction combination, register movement, scheduling... etc.
+ 
+ I've marked optimizations with a [t] to indicate things that I believe to
+ be relatively trivial to implement in LLVM itself.  The time consuming
+ things to reimplement would be SSA based PRE, Strength reduction & loop
+ unrolling... these would be the major things we would miss out on if we
+ did LLVM creation from tree code [inlining and other high level
+ optimizations are done on the tree representation].
+ 
+ Given the lack of "strong" optimizations that would take a long time to
+ reimplement, I am leaning a bit more towards creating LLVM from the tree
+ code.  Especially given that SGI has GPL'd their compiler, including many
+ SSA based optimizations that could be adapted (besides the fact that their
+ code looks MUCH nicer than GCC :)
+ 
+ Even if we choose to do LLVM code emission from RTL, we will almost
+ certainly want to move LLVM emission from step 8 down until at least CSE
+ has been rerun... which causes me to wonder if the SSA generation code
+ will still work (due to global variable dependencies and stuff).  I assume
+ that it can be made to work, but might be a little more involved than we
+ would like.
+ 
+ I'm continuing to look at the Tree -> RTL code.  It is pretty gross
+ because they do some of the translation a statement at a time, and some
+ of it a function at a time...  I'm not quite clear why and how the
+ distinction is drawn, but it does not appear that there is a wonderful
+ place to attach extra info.
+ 
+ Anyways, I'm proceeding with the RTL -> LLVM conversion phase for now.  We
+ can talk about this more on Monday.
+ 
+ Wouldn't it be nice if there were a obvious decision to be made?  :)
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-06-01-GCCOptimizations2.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-06-01-GCCOptimizations2.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-06-01-GCCOptimizations2.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,71 ----
+ Date: Fri, 1 Jun 2001 17:08:44 -0500 (CDT)
+ From: Chris Lattner <sabre at nondot.org>
+ To: Vikram S. Adve <vadve at cs.uiuc.edu>
+ Subject: RE: Interesting: GCC passes
+ 
+ > That is very interesting.  I agree that some of these could be done on LLVM
+ > at link-time, but it is the extra time required that concerns me.  Link-time
+ > optimization is severely time-constrained.
+ 
+ If we were to reimplement any of these optimizations, I assume that we
+ could do them a translation unit at a time, just as GCC does now.  This
+ would lead to a pipeline like this:
+ 
+ Static optimizations, xlation unit at a time:
+ .c --GCC--> .llvm --llvmopt--> .llvm 
+ 
+ Link time optimizations:
+ .llvm --llvm-ld--> .llvm --llvm-link-opt--> .llvm 
+ 
+ Of course, many optimizations could be shared between llvmopt and
+ llvm-link-opt, but the wouldn't need to be shared...  Thus compile time
+ could be faster, because we are using a "smarter" IR (SSA based).
+ 
+ > BTW, about SGI, "borrowing" SSA-based optimizations from one compiler and
+ > putting it into another is not necessarily easier than re-doing it.
+ > Optimization code is usually heavily tied in to the specific IR they use.
+ 
+ Understood.  The only reason that I brought this up is because SGI's IR is
+ more similar to LLVM than it is different in many respects (SSA based,
+ relatively low level, etc), and could be easily adapted.  Also their
+ optimizations are written in C++ and are actually somewhat
+ structured... of course it would be no walk in the park, but it would be
+ much less time consuming to adapt, say, SSA-PRE than to rewrite it.
+ 
+ > But your larger point is valid that adding SSA based optimizations is
+ > feasible and should be fun.  (Again, link time cost is the issue.)
+ 
+ Assuming linktime cost wasn't an issue, the question is: 
+ Does using GCC's backend buy us anything?
+ 
+ > It also occurs to me that GCC is probably doing quite a bit of back-end
+ > optimization (step 16 in your list).  Do you have a breakdown of that?
+ 
+ Not really.  The irritating part of GCC is that it mixes it all up and
+ doesn't have a clean seperation of concerns.  A lot of the "back end
+ optimization" happens right along with other data optimizations (ie, CSE
+ of machine specific things).
+ 
+ As far as REAL back end optimizations go, it looks something like this:
+ 
+ 1. Instruction combination: try to make CISCy instructions, if available
+ 2. Register movement: try to get registers in the right places for the
+ architecture to avoid register to register moves.  For example, try to get
+ the first argument of a function to naturally land in %o0 for sparc.
+ 3. Instruction scheduling: 'nuff said :)
+ 4. Register class preferencing: ??
+ 5. Local register allocation
+ 6. global register allocation
+ 7. Spilling
+ 8. Local regalloc
+ 9. Jump optimization
+ 10. Delay slot scheduling
+ 11. Branch shorting for CISC machines
+ 12. Instruction selection & peephole optimization
+ 13. Debug info output
+ 
+ But none of this would be usable for LLVM anyways, unless we were using
+ GCC as a static compiler.
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-06-20-.NET-Differences.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-06-20-.NET-Differences.txt:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-06-20-.NET-Differences.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,30 ----
+ Date: Wed, 20 Jun 2001 12:32:22 -0500
+ From: Vikram Adve <vadve at cs.uiuc.edu>
+ To: Chris Lattner <lattner at cs.uiuc.edu>
+ Subject: .NET vs. our VM
+ 
+ One significant difference between .NET CLR and our VM is that the CLR
+ includes full information about classes and inheritance.  In fact, I just
+ sat through the paper on adding templates to .NET CLR, and the speaker
+ indicated that the goal seems to be to do simple static compilation (very
+ little lowering or optimization).  Also, the templates implementation in CLR
+ "relies on dynamic class loading and JIT compilation".
+ 
+ This is an important difference because I think there are some significant
+ advantages to have a much lower level VM layer, and do significant static
+ analysis and optimization.
+ 
+ I also talked to the lead guy for KAI's C++ compiler (Arch Robison) and he
+ said that SGI and other commercial compilers have included options to export
+ their *IR* next to the object code (i.e., .il files) and use them for
+ link-time code generation.  In fact, he said that the .o file was nearly
+ empty and was entirely generated from the .il at link-time.  But he agreed
+ that this limited the link-time interprocedural optimization to modules
+ compiled by the same compiler, whereas our approach allows us to link and
+ optimize modules from multiple different compilers.  (Also, of course, they
+ don't do anything for runtime optimization).
+ 
+ All issues to bring up in Related Work.
+ 
+ --Vikram
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-07-06-LoweringIRForCodeGen.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-07-06-LoweringIRForCodeGen.txt:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-07-06-LoweringIRForCodeGen.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,31 ----
+ Date: Fri, 6 Jul 2001 16:56:56 -0500
+ From: Vikram S. Adve <vadve at cs.uiuc.edu>
+ To: Chris Lattner <lattner at cs.uiuc.edu>
+ Subject: lowering the IR
+ 
+ BTW, I do think that we should consider lowering the IR as you said.  I
+ didn't get time to raise it today, but it comes up with the SPARC
+ move-conditional instruction.  I don't think we want to put that in the core
+ VM -- it is a little too specialized.  But without a corresponding
+ conditional move instruction in the VM, it is pretty difficult to maintain a
+ close mapping between VM and machine code.  Other architectures may have
+ other such instructions.
+ 
+ What I was going to suggest was that for a particular processor, we define
+ additional VM instructions that match some of the unusual opcodes on the
+ processor but have VM semantics otherwise, i.e., all operands are in SSA
+ form and typed.  This means that we can re-generate core VM code from the
+ more specialized code any time we want (so that portability is not lost).
+ 
+ Typically, a static compiler like gcc would generate just the core VM, which
+ is relatively portable.  Anyone (an offline tool, the linker, etc., or even
+ the static compiler itself if it chooses) can transform that into more
+ specialized target-specific VM code for a particular architecture.  If the
+ linker does it, it can do it after all machine-independent optimizations.
+ This would be the most convenient, but not necessary.
+ 
+ The main benefit of lowering will be that we will be able to retain a close
+ mapping between VM and machine code.
+ 
+ --Vikram
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-07-08-InstructionSelection.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-07-08-InstructionSelection.txt:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-07-08-InstructionSelection.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,51 ----
+ Date: Sun, 8 Jul 2001 09:37:22 -0500
+ From: Vikram S. Adve <vadve at cs.uiuc.edu>
+ To: Ruchira Sasanka <sasanka at students.uiuc.edu>
+ Cc: Chris Lattner <lattner at cs.uiuc.edu>
+ Subject: machine instruction operands
+ 
+ Ruchira,
+ 
+ When generating machine instructions, I have to make several choices about
+ operands.  For cases were a register is required, there are 3 cases:
+ 
+ 1. The register is for a Value* that is already in the VM code.
+ 
+ 2. The register is for a value that is not in the VM code, usually because 2
+ machine instructions get generated for a single VM instruction (and the
+ register holds the result of the first m/c instruction and is used by the
+ second m/c instruction).
+ 
+ 3. The register is a pre-determined machine register.
+ 
+ E.g, for this VM instruction:
+         ptr = alloca type, numElements
+ I have to generate 2 machine instructions:
+         reg = mul constant, numElements
+         ptr = add %sp, reg
+ 
+ Each machine instruction is of class MachineInstr.
+ It has a vector of operands.  All register operands have type MO_REGISTER.
+ The 3 types of register operands are marked using this enum:
+ 
+  enum VirtualRegisterType {
+     MO_VMVirtualReg,            // virtual register for *value
+     MO_MInstrVirtualReg,        // virtual register for result of *minstr
+     MO_MachineReg               // pre-assigned machine register `regNum'
+   } vregType;
+ 
+ Here's how this affects register allocation:
+ 
+ 1. MO_VMVirtualReg is the standard case: you just do the register
+ allocation.
+ 
+ 2. MO_MInstrVirtualReg is the case where there is a hidden register being
+ used.  You should decide how you want to handle it, e.g., do you want do
+ create a Value object during the preprocessing phase to make the value
+ explicit (like for address register for the RETURN instruction).
+ 
+ 3. For case MO_MachineReg, you don't need to do anything, at least for
+ SPARC. The only machine regs I am using so far are %g0 and %sp.
+ 
+ --Vikram
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-07-08-InstructionSelection2.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-07-08-InstructionSelection2.txt:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-07-08-InstructionSelection2.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,25 ----
+ Date: Sun, 8 Jul 2001 10:02:20 -0500
+ From: Vikram S. Adve <vadve at cs.uiuc.edu>
+ To: vadve at cs.uiuc.edu, Ruchira Sasanka <sasanka at students.uiuc.edu>
+ Cc: Chris Lattner <lattner at cs.uiuc.edu>
+ Subject: RE: machine instruction operands
+ 
+ I got interrupted and forgot to explain the example.  In that case:
+ 
+         reg will be the 3rd operand of MUL and it will be of type
+ MO_MInstrVirtualReg.  The field MachineInstr* minstr will point to the
+ instruction that computes reg.
+ 
+         numElements will be an immediate constant, not a register.
+ 
+         %sp will be operand 1 of ADD and it will be of type MO_MachineReg.  The
+ field regNum identifies the register.
+ 
+         numElements will be operand 2 of ADD and it will be of type
+ MO_VMVirtualReg.  The field Value* value identifies the value.
+ 
+         ptr will be operand 3 of ADD will also be %sp, i.e., of
+         type MO_MachineReg. regNum identifies the register.
+ 
+ --Vikram
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2001-09-18-OptimizeExceptions.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2001-09-18-OptimizeExceptions.txt:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2001-09-18-OptimizeExceptions.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,56 ----
+ Date: Tue, 18 Sep 2001 00:38:37 -0500 (CDT)
+ From: Chris Lattner <sabre at nondot.org>
+ To: Vikram S. Adve <vadve at cs.uiuc.edu>
+ Subject: Idea for a simple, useful link time optimization
+ 
+ 
+ In C++ programs, exceptions suck, and here's why:
+ 
+ 1. In virtually all function calls, you must assume that the function
+    throws an exception, unless it is defined as 'nothrow'.  This means
+    that every function call has to have code to invoke dtors on objects
+    locally if one is thrown by the function.  Most functions don't throw
+    exceptions, so this code is dead [with all the bad effects of dead
+    code, including icache pollution].
+ 2. Declaring a function nothrow causes catch blocks to be added to every
+    call that isnot  provably nothrow.  This makes them very slow.
+ 3. Extra extraneous exception edges reduce the opportunity for code
+    motion.
+ 4. EH is typically implemented with large lookup tables.  Ours is going to
+    be much smaller (than the "standard" way of doing it) to start with,
+    but eliminating it entirely would be nice. :)
+ 5. It is physically impossible to correctly put (accurate, correct)
+    exception specifications on generic, templated code.  But it is trivial
+    to analyze instantiations of said code.
+ 6. Most large C++ programs throw few exceptions.  Most well designed
+    programs only throw exceptions in specific planned portions of the
+    code.
+ 
+ Given our _planned_ model of handling exceptions, all of this would be
+ pretty trivial to eliminate through some pretty simplistic interprocedural
+ analysis.  The DCE factor alone could probably be pretty significant.  The
+ extra code motion opportunities could also be exploited though...
+ 
+ Additionally, this optimization can be implemented in a straight forward
+ conservative manner, allowing libraries to be optimized or individual
+ files even (if there are leaf functions visible in the translation unit
+ that are called).
+ 
+ I think it's a reasonable optimization that hasn't really been addressed
+ (because assembly is way too low level for this), and could have decent
+ payoffs... without being a overly complex optimization.
+ 
+ After I wrote all of that, I found this page that is talking about
+ basically the same thing I just wrote, except that it is translation unit
+ at a time, tree based approach:
+ http://www.ocston.org/~jls/ehopt.html
+ 
+ but is very useful from "expected gain" and references perspective.  Note
+ that their compiler is apparently unable to inline functions that use
+ exceptions, so there numbers are pretty worthless... also our results
+ would (hopefully) be better because it's interprocedural...
+ 
+ What do you think?
+ 
+ -Chris
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2002-05-12-InstListChange.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2002-05-12-InstListChange.txt:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2002-05-12-InstListChange.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,55 ----
+ Date: Sun, 12 May 2002 17:12:53 -0500 (CDT)
+ From: Chris Lattner <sabre at nondot.org>
+ To: "Vikram S. Adve" <vadve at cs.uiuc.edu>
+ Subject: LLVM change
+ 
+ There is a fairly fundemental change that I would like to make to the LLVM 
+ infrastructure, but I'd like to know if you see any drawbacks that I 
+ don't...
+ 
+ Basically right now at the basic block level, each basic block contains an 
+ instruction list (returned by getInstList()) that is a ValueHolder of 
+ instructions.  To iterate over instructions, we must actually iterate over 
+ the instlist, and access the instructions through the instlist.
+ 
+ To add or remove an instruction from a basic block, we need to get an 
+ iterator to an instruction, which, given just an Instruction*, requires a 
+ linear search of the basic block the instruction is contained in... just 
+ to insert an instruction before another instruction, or to delete an 
+ instruction!  This complicates algorithms that should be very simple (like 
+ simple constant propogation), because they aren't actually sparse anymore, 
+ they have to traverse basic blocks to remove constant propogated 
+ instructions.
+ 
+ Additionally, adding or removing instructions to a basic block 
+ _invalidates all iterators_ pointing into that block, which is really 
+ irritating.
+ 
+ To fix these problems (and others), I would like to make the ordering of
+ the instructions be represented with a doubly linked list in the
+ instructions themselves, instead of an external data structure.  This is 
+ how many other representations do it, and frankly I can't remember why I 
+ originally implemented it the way I did.
+ 
+ Long term, all of the code that depends on the nasty features in the 
+ instruction list (which can be found by grep'ing for getInstList()) will 
+ be changed to do nice local transformations.  In the short term, I'll 
+ change the representation, but preserve the interface (including 
+ getInstList()) so that all of the code doesn't have to change.
+ 
+ Iteration over the instructions in a basic block remains the simple:
+ for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) ...
+ 
+ But we will also support:
+ for (Instruction *I = BB->front(); I; I = I->getNext()) ...
+ 
+ After converting instructions over, I'll convert basic blocks and 
+ functions to have a similar interface.
+ 
+ The only negative aspect of this change that I see is that it increases 
+ the amount of memory consumed by one pointer per instruction.  Given the 
+ benefits, I think this is a very reasonable tradeoff. 
+ 
+ What do you think?
+ 
+ -Chris


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2002-06-25-MegaPatchInfo.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2002-06-25-MegaPatchInfo.txt:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2002-06-25-MegaPatchInfo.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,72 ----
+ Changes:
+ * Change the casting code to be const correct.  Now, doing this is invalid:
+      const Value *V = ...;
+      Instruction *I = dyn_cast<Instruction>(V);
+   instead, the second line should be:
+      const Instruction *I = dyn_cast<Instruction>(V);
+ 
+ * Change the casting code to allow casting a reference value thus:
+      const Value &V = ...;
+      Instruction &I = cast<Instruction>(V);
+ 
+   dyn_cast does not work with references, because it must return a null pointer
+   on failure.
+ 
+ * Fundamentally change how instructions and other values are represented.
+   Before, every llvm container was an instance of the ValueHolder template,
+   instantiated for each container type.  This ValueHolder was effectively a
+   wrapper around a vector of pointers to the sub-objects.
+ 
+   Now, instead of having a vector to pointers of objects, the objects are
+   maintained in a doubly linked list of values (ie each Instruction now has
+   Next & Previous fields).  The containers are now instances of ilist (intrusive
+   linked list class), which use the next and previous fields to chain them
+   together.  The advantage of this implementation is that iterators can be
+   formed directly from pointers to the LLVM value, and invalidation is much
+   easier to handle.
+ 
+ * As part of the above change, dereferencing an iterator (for example:
+   BasicBlock::iterator) now produces a reference to the underlying type (same
+   example: Instruction&) instead of a pointer to the underlying object.  This
+   makes it much easier to write nested loops that iterator over things, changing
+   this:
+ 
+     for (Function::iterator BI = Func->begin(); BI != Func->end(); ++BI)
+       for (BasicBlock::iterator II = (*BI)->begin(); II != (*BI)->end(); ++II)
+         (*II)->dump();
+ 
+   into:
+ 
+     for (Function::iterator BI = Func->begin(); BI != Func->end(); ++BI)
+       for (BasicBlock::iterator II = BI->begin(); II != BI->end(); ++II)
+         II->dump();
+ 
+   which is much more natural and what users expect.
+ 
+ * Simplification of #include's: Before, it was necessary for a .cpp file to
+   include every .h file that it used.  Now things are batched a little bit more
+   to make it easier to use.  Specifically, the include graph now includes these
+   edges:
+     Module.h -> Function.h, GlobalVariable.h
+     Function.h -> BasicBlock.h, Argument.h
+     BasicBlock.h -> Instruction.h
+ 
+   Which means that #including Function.h is usually sufficient for getting the
+   lower level #includes.
+ 
+ * Printing out a Value* has now changed: Printing a Value* will soon print out
+   the address of the value instead of the contents of the Value.  To print out
+   the contents, you must convert it to a reference with (for example)
+   'cout << *I' instead of 'cout << I;'.  This conversion is not yet complete,
+   but will be eventually.  In the mean time, both forms print out the contents.
+ 
+ * References are used much more throughout the code base.  In general, if a
+   pointer is known to never be null, it is passed in as a reference instead of a
+   pointer.  For example, the instruction visitor class uses references instead
+   of pointers, and that Pass subclasses now all receive references to Values
+   instead of pointers, because they may never be null.
+ 
+ * The Function class now has helper functions for accessing the Arguments list.
+   Instead of having to go through getArgumentList for simple things like
+   iterator over the arguments, now the a*() methods can be used to access them.
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2003-01-23-CygwinNotes.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2003-01-23-CygwinNotes.txt:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2003-01-23-CygwinNotes.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,28 ----
+ Date: Mon, 20 Jan 2003 00:00:28 -0600
+ From: Brian R. Gaeke <gaeke at uiuc.edu>
+ Subject: windows vs. llvm
+ 
+ If you're interested, here are some of the major problems compiling LLVM
+ under Cygwin and/or Mingw.
+ 
+ 1. Cygwin doesn't have <inttypes.h> or <stdint.h>, so all the INT*_MAX
+    symbols and standard int*_t types are off in limbo somewhere. Mingw has
+    <stdint.h>, but Cygwin doesn't like it.
+ 
+ 2. Mingw doesn't have <dlfcn.h> (because Windows doesn't have it.)
+ 
+ 3. SA_SIGINFO and friends are not around; only signal() seems to work.
+ 
+ 4. Relink, aka ld -r, doesn't work (probably an ld bug); you need
+    DONT_BUILD_RELINKED. This breaks all the tools makefiles; you just need to
+    change them to have .a's.
+ 
+ 5. There isn't a <values.h>.
+ 
+ 6. There isn't a mallinfo() (or, at least, it's documented, but it doesn't seem
+    to link).
+ 
+ 7. The version of Bison that cygwin (and newer Linux versions) comes with
+    does not like = signs in rules. Burg's gram.yc source file uses them. I think
+    you can just take them out.
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2003-06-25-Reoptimizer1.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2003-06-25-Reoptimizer1.txt:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2003-06-25-Reoptimizer1.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,137 ----
+ Wed Jun 25 15:13:51 CDT 2003
+ 
+ First-level instrumentation
+ ---------------------------
+ 
+ We use opt to do Bytecode-to-bytecode instrumentation. Look at
+ back-edges and insert llvm_first_trigger() function call which takes
+ no arguments and no return value. This instrumentation is designed to
+ be easy to remove, for instance by writing a NOP over the function
+ call instruction.
+ 
+ Keep count of every call to llvm_first_trigger(), and maintain
+ counters in a map indexed by return address. If the trigger count
+ exceeds a threshold, we identify a hot loop and perform second-level
+ instrumentation on the hot loop region (the instructions between the
+ target of the back-edge and the branch that causes the back-edge).  We
+ do not move code across basic-block boundaries.
+ 
+ 
+ Second-level instrumentation
+ ---------------------------
+ 
+ We remove the first-level instrumentation by overwriting the CALL to
+ llvm_first_trigger() with a NOP.
+ 
+ The reoptimizer maintains a map between machine-code basic blocks and
+ LLVM BasicBlock*s.  We only keep track of paths that start at the
+ first machine-code basic block of the hot loop region.
+ 
+ How do we keep track of which edges to instrument, and which edges are
+ exits from the hot region? 3 step process.
+ 
+ 1) Do a DFS from the first machine-code basic block of the hot loop
+ region and mark reachable edges.
+ 
+ 2) Do a DFS from the last machine-code basic block of the hot loop
+ region IGNORING back edges, and mark the edges which are reachable in
+ 1) and also in 2) (i.e., must be reachable from both the start BB and
+ the end BB of the hot region).
+ 
+ 3) Mark BBs which end in edges that exit the hot region; we need to
+ instrument these differently.
+ 
+ Assume that there is 1 free register. On SPARC we use %g1, which LLC
+ has agreed not to use.  Shift a 1 into it at the beginning. At every
+ edge which corresponds to a conditional branch, we shift 0 for not
+ taken and 1 for taken into a register. This uniquely numbers the paths
+ through the hot region. Silently fail if we need more than 64 bits.
+ 
+ At the end BB we call countPath and increment the counter based on %g1
+ and the return address of the countPath call.  We keep track of the
+ number of iterations and the number of paths.  We only run this
+ version 30 or 40 times.
+ 
+ Find the BBs that total 90% or more of execution, and aggregate them
+ together to form our trace. But we do not allow more than 5 paths; if
+ we have more than 5 we take the ones that are executed the most.  We
+ verify our assumption that we picked a hot back-edge in first-level
+ instrumentation, by making sure that the number of times we took an
+ exit edge from the hot trace is less than 10% of the number of
+ iterations.
+ 
+ LLC has been taught to recognize llvm_first_trigger() calls and NOT
+ generate saves and restores of caller-saved registers around these
+ calls.
+ 
+ 
+ Phase behavior
+ --------------
+ 
+ We turn off llvm_first_trigger() calls with NOPs, but this would hide
+ phase behavior from us (when some funcs/traces stop being hot and
+ others become hot.)
+ 
+ We have a SIGALRM timer that counts time for us. Every time we get a
+ SIGALRM we look at our priority queue of locations where we have
+ removed llvm_first_trigger() calls. Each location is inserted along
+ with a time when we will next turn instrumentation back on for that
+ call site. If the time has arrived for a particular call site, we pop
+ that off the prio. queue and turn instrumentation back on for that
+ call site.
+ 
+ 
+ Generating traces
+ -----------------
+ 
+ When we finally generate an optimized trace we first copy the code
+ into the trace cache. This leaves us with 3 copies of the code: the
+ original code, the instrumented code, and the optimized trace. The
+ optimized trace does not have instrumentation. The original code and
+ the instrumented code are modified to have a branch to the trace
+ cache, where the optimized traces are kept.
+ 
+ We copy the code from the original to the instrumentation version
+ by tracing the LLVM-to-Machine code basic block map and then copying
+ each machine code basic block we think is in the hot region into the
+ trace cache. Then we instrument that code. The process is similar for
+ generating the final optimized trace; we copy the same basic blocks
+ because we might need to put in fixup code for exit BBs.
+ 
+ LLVM basic blocks are not typically used in the Reoptimizer except
+ for the mapping information.
+ 
+ We are restricted to using single instructions to branch between the
+ original code, trace, and instrumented code. So we have to keep the
+ code copies in memory near the original code (they can't be far enough
+ away that a single pc-relative branch would not work.) Malloc() or
+ data region space is too far away. this impacts the design of the 
+ trace cache.
+ 
+ We use a dummy function that is full of a bunch of for loops which we
+ overwrite with trace-cache code. The trace manager keeps track of
+ whether or not we have enough space in the trace cache, etc.
+ 
+ The trace insertion routine takes an original start address, a vector
+ of machine instructions representing the trace, index of branches and
+ their corresponding absolute targets, and index of calls and their
+ corresponding absolute targets.
+ 
+ The trace insertion routine is responsible for inserting branches from
+ the beginning of the original code to the beginning of the optimized
+ trace. This is because at some point the trace cache may run out of
+ space and it may have to evict a trace, at which point the branch to
+ the trace would also have to be removed. It uses a round-robin
+ replacement policy; we have found that this is almost as good as LRU
+ and better than random (especially because of problems fitting the new
+ trace in.)
+ 
+ We cannot deal with discontiguous trace cache areas.  The trace cache
+ is supposed to be cache-line-aligned, but it is not page-aligned.
+ 
+ We generate instrumentation traces and optimized traces into separate
+ trace caches. We keep the instrumented code around because you don't
+ want to delete a trace when you still might have to return to it
+ (i.e., return from a llvm_first_trigger() or countPath() call.)
+ 
+ 


Index: llvm-www/releases/2.6/docs/HistoricalNotes/2003-06-26-Reoptimizer2.txt
diff -c /dev/null llvm-www/releases/2.6/docs/HistoricalNotes/2003-06-26-Reoptimizer2.txt:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/HistoricalNotes/2003-06-26-Reoptimizer2.txt	Fri Oct 23 13:56:25 2009
***************
*** 0 ****
--- 1,110 ----
+ Thu Jun 26 14:43:04 CDT 2003
+ 
+ Information about BinInterface
+ ------------------------------
+ 
+ Take in a set of instructions with some particular register
+ allocation. It allows you to add, modify, or delete some instructions,
+ in SSA form (kind of like LLVM's MachineInstrs.) Then re-allocate
+ registers. It assumes that the transformations you are doing are safe.
+ It does not update the mapping information or the LLVM representation
+ for the modified trace (so it would not, for instance, support
+ multiple optimization passes; passes have to be aware of and update
+ manually the mapping information.)
+ 
+ The way you use it is you take the original code and provide it to
+ BinInterface; then you do optimizations to it, then you put it in the
+ trace cache.
+ 
+ The BinInterface tries to find live-outs for traces so that it can do
+ register allocation on just the trace, and stitch the trace back into
+ the original code. It has to preserve the live-ins and live-outs when
+ it does its register allocation.  (On exits from the trace we have
+ epilogues that copy live-outs back into the right registers, but
+ live-ins have to be in the right registers.)
+ 
+ 
+ Limitations of BinInterface
+ ---------------------------
+ 
+ It does copy insertions for PHIs, which it infers from the machine
+ code. The mapping info inserted by LLC is not sufficient to determine
+ the PHIs.
+ 
+ It does not handle integer or floating-point condition codes and it
+ does not handle floating-point register allocation.
+ 
+ It is not aggressively able to use lots of registers.
+ 
+ There is a problem with alloca: we cannot find our spill space for
+ spilling registers, normally allocated on the stack, if the trace
+ follows an alloca(). What might be an acceptable solution would be to
+ disable trace generation on functions that have variable-sized
+ alloca()s. Variable-sized allocas in the trace would also probably
+ screw things up.
+ 
+ Because of the FP and alloca limitations, the BinInterface is
+ completely disabled right now.
+ 
+ 
+ Demo
+ ----
+ 
+ This is a demo of the Ball & Larus version that does NOT use 2-level
+ profiling.
+ 
+ 1. Compile program with llvm-gcc.
+ 2. Run opt -lowerswitch -paths -emitfuncs on the bytecode.
+    -lowerswitch change switch statements to branches
+    -paths       Ball & Larus path-profiling algorithm
+    -emitfuncs   emit the table of functions
+ 3. Run llc to generate SPARC assembly code for the result of step 2.
+ 4. Use g++ to link the (instrumented) assembly code.
+ 
+ We use a script to do all this:
+ ------------------------------------------------------------------------------
+ #!/bin/sh
+ llvm-gcc $1.c -o $1
+ opt -lowerswitch -paths -emitfuncs $1.bc > $1.run.bc
+ llc -f $1.run.bc 
+ LIBS=$HOME/llvm_sparc/lib/Debug
+ GXX=/usr/dcs/software/evaluation/bin/g++
+ $GXX -g -L $LIBS $1.run.s -o $1.run.llc \
+ $LIBS/tracecache.o \
+ $LIBS/mapinfo.o \
+ $LIBS/trigger.o \
+ $LIBS/profpaths.o \
+ $LIBS/bininterface.o \
+ $LIBS/support.o \
+ $LIBS/vmcore.o \
+ $LIBS/transformutils.o \
+ $LIBS/bcreader.o \
+ -lscalaropts -lscalaropts -lanalysis \
+ -lmalloc -lcpc -lm -ldl
+ ------------------------------------------------------------------------------
+ 
+ 5. Run the resulting binary.  You will see output from BinInterface
+ (described below) intermixed with the output from the program.
+ 
+ 
+ Output from BinInterface
+ ------------------------
+ 
+ BinInterface's debugging code prints out the following stuff in order:
+ 
+ 1. Initial code provided to BinInterface with original register
+ allocation.
+ 
+ 2. Section 0 is the trace prolog, consisting mainly of live-ins and
+ register saves which will be restored in epilogs.
+ 
+ 3. Section 1 is the trace itself, in SSA form used by BinInterface,
+ along with the PHIs that are inserted.
+ PHIs are followed by the copies that implement them.
+ Each branch (i.e., out of the trace) is annotated with the
+ section number that represents the epilog it branches to.
+ 
+ 4. All the other sections starting with Section 2 are trace epilogs.
+ Every branch from the trace has to go to some epilog.
+ 
+ 5. After the last section is the register allocation output.





From tonic at nondot.org  Fri Oct 23 13:57:46 2009
From: tonic at nondot.org (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:57:46 -0500
Subject: [llvm-commits] CVS:
	llvm-www/releases/2.6/docs/CommandGuide/FileCheck.pod
	Makefile bugpoint.pod index.html llc.pod lli.pod llvm-ar.pod
	llvm-as.pod llvm-bcanalyzer.pod llvm-config.pod llvm-db.pod
	llvm-dis.pod llvm-extract.pod llvm-ld.pod llvm-link.pod
	llvm-nm.pod llvm-prof.pod llvm-ranlib.pod llvmc.pod
	llvmgcc.pod llvmgxx.pod manpage.css opt.pod pod2htmd.tmp
	pod2htmi.tmp tblgen.pod
Message-ID: <200910231857.n9NIvkNT004764@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/CommandGuide:

FileCheck.pod added (r1.1)
Makefile added (r1.1)
bugpoint.pod added (r1.1)
index.html added (r1.1)
llc.pod added (r1.1)
lli.pod added (r1.1)
llvm-ar.pod added (r1.1)
llvm-as.pod added (r1.1)
llvm-bcanalyzer.pod added (r1.1)
llvm-config.pod added (r1.1)
llvm-db.pod added (r1.1)
llvm-dis.pod added (r1.1)
llvm-extract.pod added (r1.1)
llvm-ld.pod added (r1.1)
llvm-link.pod added (r1.1)
llvm-nm.pod added (r1.1)
llvm-prof.pod added (r1.1)
llvm-ranlib.pod added (r1.1)
llvmc.pod added (r1.1)
llvmgcc.pod added (r1.1)
llvmgxx.pod added (r1.1)
manpage.css added (r1.1)
opt.pod added (r1.1)
pod2htmd.tmp added (r1.1)
pod2htmi.tmp added (r1.1)
tblgen.pod added (r1.1)
---
Log message:

Add 2.6



---
Diffs of the changes:  (+3387 -0)

 FileCheck.pod       |   65 ++++++++
 Makefile            |  103 +++++++++++++
 bugpoint.pod        |  167 +++++++++++++++++++++
 index.html          |  153 +++++++++++++++++++
 llc.pod             |  197 +++++++++++++++++++++++++
 lli.pod             |  216 +++++++++++++++++++++++++++
 llvm-ar.pod         |  406 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 llvm-as.pod         |   77 +++++++++
 llvm-bcanalyzer.pod |  315 ++++++++++++++++++++++++++++++++++++++++
 llvm-config.pod     |  131 ++++++++++++++++
 llvm-db.pod         |   16 ++
 llvm-dis.pod        |   60 +++++++
 llvm-extract.pod    |   63 ++++++++
 llvm-ld.pod         |  234 +++++++++++++++++++++++++++++
 llvm-link.pod       |   74 +++++++++
 llvm-nm.pod         |  122 +++++++++++++++
 llvm-prof.pod       |   57 +++++++
 llvm-ranlib.pod     |   52 ++++++
 llvmc.pod           |  179 ++++++++++++++++++++++
 llvmgcc.pod         |   76 +++++++++
 llvmgxx.pod         |   85 ++++++++++
 manpage.css         |  256 ++++++++++++++++++++++++++++++++
 opt.pod             |  138 +++++++++++++++++
 pod2htmd.tmp        |   28 +++
 pod2htmi.tmp        |    2 
 tblgen.pod          |  115 ++++++++++++++
 26 files changed, 3387 insertions(+)


Index: llvm-www/releases/2.6/docs/CommandGuide/FileCheck.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/FileCheck.pod:1.1
*** /dev/null	Fri Oct 23 13:56:31 2009
--- llvm-www/releases/2.6/docs/CommandGuide/FileCheck.pod	Fri Oct 23 13:56:20 2009
***************
*** 0 ****
--- 1,65 ----
+ 
+ =pod
+ 
+ =head1 NAME
+ 
+ FileCheck - Flexible pattern matching file verifier
+ 
+ =head1 SYNOPSIS
+ 
+ B<FileCheck> I<match-filename> [I<--check-prefix=XXX>] [I<--strict-whitespace>]
+ 
+ =head1 DESCRIPTION
+ 
+ B<FileCheck> reads two files (one from standard input, and one specified on the
+ command line) and uses one to verify the other.  This behavior is particularly
+ useful for the testsuite, which wants to verify that the output of some tool
+ (e.g. llc) contains the expected information (for example, a movsd from esp or
+ whatever is interesting).  This is similar to using grep, but it is optimized
+ for matching multiple different inputs in one file in a specific order.
+ 
+ The I<match-filename> file specifies the file that contains the patterns to
+ match.  The file to verify is always read from standard input.
+ 
+ The input and output of B<FileCheck> is beyond the scope of this short
+ introduction. Please see the I<TestingGuide> page in the LLVM documentation.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<--check-prefix> I<prefix>
+ 
+ FileCheck searches the contents of I<match-filename> for patterns to match.  By
+ default, these patterns are prefixed with "CHECK:".  If you'd like to use a
+ different prefix (e.g. because the same input file is checking multiple
+ different tool or options), the B<--check-prefix> argument allows you to specify
+ a specific prefix to match.
+ 
+ =item B<--strict-whitespace>
+ 
+ By default, FileCheck canonicalizes input horizontal whitespace (spaces and
+ tabs) which causes it to ignore these differences (a space will match a tab).
+ The --strict-whitespace argument disables this behavior.
+ 
+ =item B<-version>
+ 
+ Show the version number of this program.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<FileCheck> verifies that the file matches the expected contents, it exits
+ with 0.  Otherwise, if not, or if an error occurs, it will exit with a non-zero
+ value.
+ 
+ =head1 AUTHORS
+ 
+ Maintained by The LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/Makefile
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/Makefile:1.1
*** /dev/null	Fri Oct 23 13:57:42 2009
--- llvm-www/releases/2.6/docs/CommandGuide/Makefile	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,103 ----
+ ##===- docs/CommandGuide/Makefile --------------------------*- Makefile -*-===##
+ # 
+ #                     The LLVM Compiler Infrastructure
+ #
+ # This file is distributed under the University of Illinois Open Source
+ # License. See LICENSE.TXT for details.
+ # 
+ ##===----------------------------------------------------------------------===##
+ 
+ ifdef BUILD_FOR_WEBSITE
+ # This special case is for keeping the CommandGuide on the LLVM web site
+ # up to date automatically as the documents are checked in. It must build
+ # the POD files to HTML only and keep them in the src directories. It must also
+ # build in an unconfigured tree, hence the ifdef. To use this, run
+ # make -s BUILD_FOR_WEBSITE=1 inside the cvs commit script.
+ SRC_DOC_DIR=
+ DST_HTML_DIR=html/
+ DST_MAN_DIR=man/man1/
+ DST_PS_DIR=ps/
+ 
+ # If we are in BUILD_FOR_WEBSITE mode, default to the all target.
+ all:: html man ps
+ 
+ clean:
+ 	rm -f pod2htm*.*~~ $(HTML) $(MAN) $(PS)
+ 
+ # To create other directories, as needed, and timestamp their creation
+ %/.dir:
+ 	-mkdir $* > /dev/null
+ 	date > $@
+ 
+ else
+ 
+ # Otherwise, if not in BUILD_FOR_WEBSITE mode, use the project info.
+ LEVEL := ../..
+ include $(LEVEL)/Makefile.common
+ 
+ SRC_DOC_DIR=$(PROJ_SRC_DIR)/
+ DST_HTML_DIR=$(PROJ_OBJ_DIR)/
+ DST_MAN_DIR=$(PROJ_OBJ_DIR)/
+ DST_PS_DIR=$(PROJ_OBJ_DIR)/
+ 
+ endif
+ 
+ 
+ POD  := $(wildcard $(SRC_DOC_DIR)*.pod)
+ HTML := $(patsubst $(SRC_DOC_DIR)%.pod, $(DST_HTML_DIR)%.html, $(POD))
+ MAN  := $(patsubst $(SRC_DOC_DIR)%.pod, $(DST_MAN_DIR)%.1, $(POD))
+ PS   := $(patsubst $(SRC_DOC_DIR)%.pod, $(DST_PS_DIR)%.ps, $(POD))
+ 
+ # The set of man pages we will not install
+ NO_INSTALL_MANS = $(DST_MAN_DIR)FileCheck.1
+ 
+ # The set of man pages that we will install
+ INSTALL_MANS = $(filter-out $(NO_INSTALL_MANS), $(MAN))
+ 
+ .SUFFIXES:
+ .SUFFIXES: .html .pod .1 .ps
+ 
+ $(DST_HTML_DIR)%.html: %.pod $(DST_HTML_DIR)/.dir
+ 	pod2html --css=manpage.css --htmlroot=. \
+ 	  --podpath=. --noindex --infile=$< --outfile=$@ --title=$*
+ 
+ $(DST_MAN_DIR)%.1: %.pod $(DST_MAN_DIR)/.dir
+ 	pod2man --release=CVS --center="LLVM Command Guide" $< $@
+ 
+ $(DST_PS_DIR)%.ps: $(DST_MAN_DIR)%.1 $(DST_PS_DIR)/.dir
+ 	groff -Tps -man $< > $@
+ 
+ 
+ html: $(HTML)
+ man: $(MAN)
+ ps: $(PS)
+ 
+ EXTRA_DIST := $(POD) index.html
+ 
+ clean-local::
+ 	$(Verb) $(RM) -f pod2htm*.*~~ $(HTML) $(MAN) $(PS)
+ 
+ HTML_DIR := $(PROJ_docsdir)/html/CommandGuide
+ MAN_DIR  := $(PROJ_mandir)/man1
+ PS_DIR   := $(PROJ_docsdir)/ps
+ 
+ install-local:: $(HTML) $(INSTALL_MANS) $(PS)
+ 	$(Echo) Installing HTML CommandGuide Documentation
+ 	$(Verb) $(MKDIR) $(HTML_DIR)
+ 	$(Verb) $(DataInstall) $(HTML) $(HTML_DIR)
+ 	$(Verb) $(DataInstall) $(PROJ_SRC_DIR)/index.html $(HTML_DIR)
+ 	$(Verb) $(DataInstall) $(PROJ_SRC_DIR)/manpage.css $(HTML_DIR)
+ 	$(Echo) Installing MAN CommandGuide Documentation
+ 	$(Verb) $(MKDIR) $(MAN_DIR)
+ 	$(Verb) $(DataInstall) $(INSTALL_MANS) $(MAN_DIR)
+ 	$(Echo) Installing PS CommandGuide Documentation
+ 	$(Verb) $(MKDIR) $(PS_DIR)
+ 	$(Verb) $(DataInstall) $(PS) $(PS_DIR)
+ 
+ uninstall-local::
+ 	$(Echo) Uninstalling CommandGuide Documentation
+ 	$(Verb) $(RM) -rf $(HTML_DIR) $(MAN_DIR) $(PS_DIR)
+ 
+ printvars::
+ 	$(Echo) "POD            : " '$(POD)'
+ 	$(Echo) "HTML           : " '$(HTML)'


Index: llvm-www/releases/2.6/docs/CommandGuide/bugpoint.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/bugpoint.pod:1.1
*** /dev/null	Fri Oct 23 13:57:42 2009
--- llvm-www/releases/2.6/docs/CommandGuide/bugpoint.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,167 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ bugpoint - automatic test case reduction tool
+ 
+ =head1 SYNOPSIS
+ 
+ B<bugpoint> [I<options>] [I<input LLVM ll/bc files>] [I<LLVM passes>] B<--args>
+ I<program arguments>
+ 
+ =head1 DESCRIPTION
+ 
+ B<bugpoint> narrows down the source of problems in LLVM tools and passes.  It
+ can be used to debug three types of failures: optimizer crashes, miscompilations
+ by optimizers, or bad native code generation (including problems in the static
+ and JIT compilers).  It aims to reduce large test cases to small, useful ones.
+ For more information on the design and inner workings of B<bugpoint>, as well as
+ advice for using bugpoint, see F<llvm/docs/Bugpoint.html> in the LLVM
+ distribution.
+ 
+ =head1 OPTIONS
+ 
+ =over 
+ 
+ =item B<--additional-so> F<library>
+ 
+ Load the dynamic shared object F<library> into the test program whenever it is
+ run.  This is useful if you are debugging programs which depend on non-LLVM
+ libraries (such as the X or curses libraries) to run.
+ 
+ =item B<--append-exit-code>=I<{true,false}>
+ 
+ Append the test programs exit code to the output file so that a change in exit
+ code is considered a test failure. Defaults to false.
+ 
+ =item B<--args> I<program args>
+ 
+ Pass all arguments specified after -args to the test program whenever it runs.
+ Note that if any of the I<program args> start with a '-', you should use:
+ 
+     bugpoint [bugpoint args] --args -- [program args]
+ 
+ The "--" right after the B<--args> option tells B<bugpoint> to consider any
+ options starting with C<-> to be part of the B<--args> option, not as options to
+ B<bugpoint> itself.
+ 
+ =item B<--tool-args> I<tool args>
+ 
+ Pass all arguments specified after --tool-args to the LLVM tool under test
+ (B<llc>, B<lli>, etc.) whenever it runs.  You should use this option in the
+ following way:
+ 
+     bugpoint [bugpoint args] --tool-args -- [tool args]
+ 
+ The "--" right after the B<--tool-args> option tells B<bugpoint> to consider any
+ options starting with C<-> to be part of the B<--tool-args> option, not as
+ options to B<bugpoint> itself. (See B<--args>, above.)
+ 
+ =item B<--safe-tool-args> I<tool args>
+ 
+ Pass all arguments specified after B<--safe-tool-args> to the "safe" execution
+ tool.
+ 
+ =item B<--gcc-tool-args> I<gcc tool args>
+ 
+ Pass all arguments specified after B<--gcc-tool-args> to the invocation of
+ B<gcc>.
+ 
+ =item B<--disable-{dce,simplifycfg}>
+ 
+ Do not run the specified passes to clean up and reduce the size of the test
+ program. By default, B<bugpoint> uses these passes internally when attempting to
+ reduce test programs.  If you're trying to find a bug in one of these passes,
+ B<bugpoint> may crash.
+ 
+ =item B<--enable-valgrind>
+ 
+ Use valgrind to find faults in the optimization phase. This will allow
+ bugpoint to find otherwise asymptomatic problems caused by memory
+ mis-management.
+ 
+ =item B<-find-bugs>
+ 
+ Continually randomize the specified passes and run them on the test program
+ until a bug is found or the user kills B<bugpoint>. 
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<--input> F<filename>
+ 
+ Open F<filename> and redirect the standard input of the test program, whenever
+ it runs, to come from that file.
+ 
+ =item B<--load> F<plugin>
+ 
+ Load the dynamic object F<plugin> into B<bugpoint> itself.  This object should
+ register new optimization passes.  Once loaded, the object will add new command
+ line options to enable various optimizations.  To see the new complete list of
+ optimizations, use the B<--help> and B<--load> options together; for example:
+ 
+     bugpoint --load myNewPass.so --help
+ 
+ =item B<--mlimit> F<megabytes>
+ 
+ Specifies an upper limit on memory usage of the optimization and codegen. Set
+ to zero to disable the limit.
+ 
+ =item B<--output> F<filename>
+ 
+ Whenever the test program produces output on its standard output stream, it
+ should match the contents of F<filename> (the "reference output"). If you
+ do not use this option, B<bugpoint> will attempt to generate a reference output
+ by compiling the program with the "safe" backend and running it.
+ 
+ =item B<--profile-info-file> F<filename>
+ 
+ Profile file loaded by B<--profile-loader>.
+ 
+ =item B<--run-{int,jit,llc,cbe,custom}>
+ 
+ Whenever the test program is compiled, B<bugpoint> should generate code for it
+ using the specified code generator.  These options allow you to choose the
+ interpreter, the JIT compiler, the static native code compiler, the C
+ backend, or a custom command (see B<--exec-command>) respectively.
+ 
+ =item B<--safe-{llc,cbe,custom}>
+ 
+ When debugging a code generator, B<bugpoint> should use the specified code
+ generator as the "safe" code generator. This is a known-good code generator
+ used to generate the "reference output" if it has not been provided, and to
+ compile portions of the program that as they are excluded from the testcase.
+ These options allow you to choose the
+ static native code compiler, the C backend, or a custom command,
+ (see B<--exec-command>) respectively. The interpreter and the JIT backends
+ cannot currently be used as the "safe" backends.
+ 
+ =item B<--exec-command> I<command>
+ 
+ This option defines the command to use with the B<--run-custom> and
+ B<--safe-custom> options to execute the bitcode testcase. This can
+ be useful for cross-compilation.
+ 
+ =item B<--safe-path> I<path>
+ 
+ This option defines the path to the command to execute with the
+ B<--safe-{int,jit,llc,cbe,custom}>
+ option.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<bugpoint> succeeds in finding a problem, it will exit with 0.  Otherwise,
+ if an error occurs, it will exit with a non-zero value.
+ 
+ =head1 SEE ALSO
+ 
+ L<opt|opt>
+ 
+ =head1 AUTHOR
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/index.html
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/index.html:1.1
*** /dev/null	Fri Oct 23 13:57:43 2009
--- llvm-www/releases/2.6/docs/CommandGuide/index.html	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,153 ----
+ <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
+                       "http://www.w3.org/TR/html4/strict.dtd">
+ <html>
+ <head>
+   <title>LLVM Command Guide</title>
+   <link rel="stylesheet" href="/docs/llvm.css" type="text/css">
+ </head>
+ <body>
+ 
+ <div class="doc_title">
+   LLVM Command Guide
+ </div>
+ 
+ <div class="doc_text">
+ 
+ <p>These documents are HTML versions of the <a href="man/man1/">man pages</a>
+ for all of the LLVM tools.  These pages describe how to use the LLVM commands
+ and what their options are.  Note that these pages do not describe all of the
+ options available for all tools.  To get a complete listing, pass the
+ <tt>--help</tt> (general options) or <tt>--help-hidden</tt> (general+debugging
+ options) arguments to the tool you are interested in.</p>
+ 
+ </div>
+ 
+ <!-- *********************************************************************** -->
+ <div class="doc_section">
+   <a name="basic">Basic Commands</a>
+ </div>
+ <!-- *********************************************************************** -->
+ 
+ <div class="doc_text">
+ 
+ <ul>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-as.html"><b>llvm-as</b></a> -
+     assemble a human-readable .ll file into bytecode</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-dis.html"><b>llvm-dis</b></a> -
+     disassemble a bytecode file into a human-readable .ll file</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/opt.html"><b>opt</b></a> -
+     run a series of LLVM-to-LLVM optimizations on a bytecode file</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llc.html"><b>llc</b></a> -
+     generate native machine code for a bytecode file</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/lli.html"><b>lli</b></a> -
+     directly run a program compiled to bytecode using a JIT compiler or
+     interpreter</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-link.html"><b>llvm-link</b></a> -
+     link several bytecode files into one</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-ar.html"><b>llvm-ar</b></a> -
+     archive bytecode files</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-ranlib.html"><b>llvm-ranlib</b></a> -
+     create an index for archives made with llvm-ar</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-nm.html"><b>llvm-nm</b></a> -
+     print out the names and types of symbols in a bytecode file</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-prof.html"><b>llvm-prof</b></a> -
+     format raw `<tt>llvmprof.out</tt>' data into a human-readable report</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-ld.html"><b>llvm-ld</b></a> -
+     general purpose linker with loadable runtime optimization support</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-config.html"><b>llvm-config</b></a> -
+     print out LLVM compilation options, libraries, etc. as configured</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvmc.html"><b>llvmc</b></a> -
+     a generic customizable compiler driver</li>
+ 
+ </ul>
+ 
+ </div>
+ 
+ <!-- *********************************************************************** -->
+ <div class="doc_section">
+   <a name="frontend">C and C++ Front-end Commands</a>
+ </div>
+ <!-- *********************************************************************** -->
+ 
+ <div class="doc_text">
+ <ul>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvmgcc.html"><b>llvm-gcc</b></a> -
+     GCC-based C front-end for LLVM
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvmgxx.html"><b>llvm-g++</b></a> -
+     GCC-based C++ front-end for LLVM</li>
+ 
+ </ul>
+ 
+ </div>
+ 
+ <!-- *********************************************************************** -->
+ <div class="doc_section">
+   <a name="debug">Debugging Tools</a>
+ </div>
+ <!-- *********************************************************************** -->
+ 
+ 
+ <div class="doc_text">
+ 
+ <ul>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/bugpoint.html"><b>bugpoint</b></a> -
+     automatic test-case reducer</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-extract.html"><b>llvm-extract</b></a> -
+     extract a function from an LLVM bytecode file</li>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/llvm-bcanalyzer.html"><b>llvm-bcanalyzer</b></a> -
+     bytecode analyzer (analyzes the binary encoding itself, not the program it
+     represents)</li>
+ 
+ </ul>
+ </div>
+ 
+ <!-- *********************************************************************** -->
+ <div class="doc_section">
+   <a name="internal">Internal Tools</a>
+ </div>
+ <!-- *********************************************************************** -->
+ 
+ <div class="doc_text">
+ <ul>
+ 
+ <li><a href="/releases/2.6/docs/CommandGuide/html/FileCheck.html"><b>FileCheck</b></a> -
+     Flexible file verifier used extensively by the testing harness</li>
+ <li><a href="/releases/2.6/docs/CommandGuide/html/tblgen.html"><b>tblgen</b></a> -
+     target description reader and generator</li>
+ 
+ </ul>
+ </div>
+ 
+ <!-- *********************************************************************** -->
+ 
+ <hr>
+ <address>
+   <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
+   src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
+   <a href="http://validator.w3.org/check/referer"><img
+   src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
+ 
+   <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
+   Last modified: $Date: 2009/10/23 18:56:21 $
+ </address>
+ 
+ </body>
+ </html>


Index: llvm-www/releases/2.6/docs/CommandGuide/llc.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llc.pod:1.1
*** /dev/null	Fri Oct 23 13:57:43 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llc.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,197 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llc - LLVM static compiler
+ 
+ =head1 SYNOPSIS
+ 
+ B<llc> [I<options>] [I<filename>]
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llc> command compiles LLVM bitcode into assembly language for a
+ specified architecture.  The assembly language output can then be passed through
+ a native assembler and linker to generate a native executable.
+ 
+ The choice of architecture for the output assembly code is automatically
+ determined from the input bitcode file, unless the B<-march> option is used to
+ override the default.
+ 
+ =head1 OPTIONS
+ 
+ If I<filename> is - or omitted, B<llc> reads LLVM bitcode from standard input.
+ Otherwise, it will read LLVM bitcode from I<filename>.
+ 
+ If the B<-o> option is omitted, then B<llc> will send its output to standard
+ output if the input is from standard input.  If the B<-o> option specifies -,
+ then the output will also be sent to standard output.
+ 
+ If no B<-o> option is specified and an input file other than - is specified,
+ then B<llc> creates the output filename by taking the input filename,
+ removing any existing F<.bc> extension, and adding a F<.s> suffix.
+ 
+ Other B<llc> options are as follows:
+ 
+ =head2 End-user Options
+ 
+ =over
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-O>=I<uint>
+ 
+ Generate code at different optimization levels. These correspond to the I<-O0>,
+ I<-O1>, I<-O2>, I<-O3>, and I<-O4> optimization levels used by B<llvm-gcc> and
+ B<clang>.
+ 
+ =item B<-f>
+ 
+ Overwrite output files. By default, B<llc> will refuse to overwrite
+ an output file which already exists.
+ 
+ =item B<-mtriple>=I<target triple>
+ 
+ Override the target triple specified in the input bitcode file with the 
+ specified string.
+ 
+ =item B<-march>=I<arch>
+ 
+ Specify the architecture for which to generate assembly, overriding the target
+ encoded in the bitcode file.  See the output of B<llc --help> for a list of
+ valid architectures.  By default this is inferred from the target triple or
+ autodetected to the current architecture.
+ 
+ =item B<-mcpu>=I<cpuname>
+ 
+ Specify a specific chip in the current architecture to generate code for.
+ By default this is inferred from the target triple and autodetected to 
+ the current architecture.  For a list of available CPUs, use:
+ B<llvm-as E<lt> /dev/null | llc -march=xyz -mcpu=help>
+ 
+ =item B<-mattr>=I<a1,+a2,-a3,...>
+ 
+ Override or control specific attributes of the target, such as whether SIMD
+ operations are enabled or not.  The default set of attributes is set by the
+ current CPU.  For a list of available attributes, use:
+ B<llvm-as E<lt> /dev/null | llc -march=xyz -mattr=help>
+ 
+ =item B<--disable-fp-elim>
+ 
+ Disable frame pointer elimination optimization.
+ 
+ =item B<--disable-excess-fp-precision>
+ 
+ Disable optimizations that may produce excess precision for floating point.
+ Note that this option can dramatically slow down code on some systems
+ (e.g. X86).
+ 
+ =item B<--enable-unsafe-fp-math>
+ 
+ Enable optimizations that make unsafe assumptions about IEEE math (e.g. that
+ addition is associative) or may not work for all input ranges.  These
+ optimizations allow the code generator to make use of some instructions which
+ would otherwise not be usable (such as fsin on X86).
+ 
+ =item B<--enable-correct-eh-support>
+ 
+ Instruct the B<lowerinvoke> pass to insert code for correct exception handling
+ support.  This is expensive and is by default omitted for efficiency.
+ 
+ =item B<--stats>
+ 
+ Print statistics recorded by code-generation passes.
+ 
+ =item B<--time-passes>
+ 
+ Record the amount of time needed for each pass and print a report to standard
+ error.
+ 
+ =item B<--load>=F<dso_path>
+ 
+ Dynamically load F<dso_path> (a path to a dynamically shared object) that
+ implements an LLVM target. This will permit the target name to be used with the
+ B<-march> option so that code can be generated for that target.
+ 
+ =back
+ 
+ =head2 Tuning/Configuration Options
+ 
+ =over
+ 
+ =item B<--print-machineinstrs>
+ 
+ Print generated machine code between compilation phases (useful for debugging).
+ 
+ =item B<--regalloc>=I<allocator>
+ 
+ Specify the register allocator to use. The default I<allocator> is I<local>.
+ Valid register allocators are:
+ 
+ =over
+ 
+ =item I<simple>
+ 
+ Very simple "always spill" register allocator
+ 
+ =item I<local>
+ 
+ Local register allocator
+ 
+ =item I<linearscan>
+ 
+ Linear scan global register allocator
+ 
+ =item I<iterativescan>
+ 
+ Iterative scan global register allocator
+ 
+ =back
+ 
+ =item B<--spiller>=I<spiller>
+ 
+ Specify the spiller to use for register allocators that support it.  Currently
+ this option is used only by the linear scan register allocator. The default
+ I<spiller> is I<local>.  Valid spillers are:
+ 
+ =over
+ 
+ =item I<simple>
+ 
+ Simple spiller
+ 
+ =item I<local>
+ 
+ Local spiller
+ 
+ =back
+ 
+ =back
+ 
+ =head2 Intel IA-32-specific Options
+ 
+ =over
+ 
+ =item B<--x86-asm-syntax=att|intel>
+ 
+ Specify whether to emit assembly code in AT&T syntax (the default) or intel
+ syntax.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llc> succeeds, it will exit with 0.  Otherwise, if an error occurs,
+ it will exit with a non-zero value.
+ 
+ =head1 SEE ALSO
+ 
+ L<lli|lli>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/lli.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/lli.pod:1.1
*** /dev/null	Fri Oct 23 13:57:43 2009
--- llvm-www/releases/2.6/docs/CommandGuide/lli.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,216 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ lli - directly execute programs from LLVM bitcode
+ 
+ =head1 SYNOPSIS
+ 
+ B<lli> [I<options>] [I<filename>] [I<program args>]
+ 
+ =head1 DESCRIPTION
+ 
+ B<lli> directly executes programs in LLVM bitcode format.  It takes a program
+ in LLVM bitcode format and executes it using a just-in-time compiler, if one is
+ available for the current architecture, or an interpreter.  B<lli> takes all of
+ the same code generator options as L<llc|llc>, but they are only effective when
+ B<lli> is using the just-in-time compiler.
+ 
+ If I<filename> is not specified, then B<lli> reads the LLVM bitcode for the
+ program from standard input.
+ 
+ The optional I<args> specified on the command line are passed to the program as
+ arguments.
+ 
+ =head1 GENERAL OPTIONS
+ 
+ =over
+ 
+ =item B<-fake-argv0>=I<executable>
+ 
+ Override the C<argv[0]> value passed into the executing program.
+ 
+ =item B<-force-interpreter>=I<{false,true}>
+ 
+ If set to true, use the interpreter even if a just-in-time compiler is available
+ for this architecture. Defaults to false.
+ 
+ =item B<-help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-load>=I<puginfilename>
+ 
+ Causes B<lli> to load the plugin (shared object) named I<pluginfilename> and use
+ it for optimization.
+ 
+ =item B<-stats>
+ 
+ Print statistics from the code-generation passes. This is only meaningful for
+ the just-in-time compiler, at present.
+ 
+ =item B<-time-passes>
+ 
+ Record the amount of time needed for each code-generation pass and print it to
+ standard error.
+ 
+ =item B<-version>
+ 
+ Print out the version of B<lli> and exit without doing anything else.
+ 
+ =back 
+ 
+ =head1 TARGET OPTIONS
+ 
+ =over 
+ 
+ =item B<-mtriple>=I<target triple>
+ 
+ Override the target triple specified in the input bitcode file with the 
+ specified string.  This may result in a crash if you pick an
+ architecture which is not compatible with the current system.
+ 
+ =item B<-march>=I<arch>
+ 
+ Specify the architecture for which to generate assembly, overriding the target
+ encoded in the bitcode file.  See the output of B<llc --help> for a list of
+ valid architectures.  By default this is inferred from the target triple or
+ autodetected to the current architecture.
+ 
+ =item B<-mcpu>=I<cpuname>
+ 
+ Specify a specific chip in the current architecture to generate code for.
+ By default this is inferred from the target triple and autodetected to 
+ the current architecture.  For a list of available CPUs, use:
+ B<llvm-as E<lt> /dev/null | llc -march=xyz -mcpu=help>
+ 
+ =item B<-mattr>=I<a1,+a2,-a3,...>
+ 
+ Override or control specific attributes of the target, such as whether SIMD
+ operations are enabled or not.  The default set of attributes is set by the
+ current CPU.  For a list of available attributes, use:
+ B<llvm-as E<lt> /dev/null | llc -march=xyz -mattr=help>
+ 
+ =back
+ 
+ 
+ =head1 FLOATING POINT OPTIONS
+ 
+ =over 
+ 
+ =item B<-disable-excess-fp-precision>
+ 
+ Disable optimizations that may increase floating point precision.
+ 
+ =item B<-enable-finite-only-fp-math>
+ 
+ Enable optimizations that assumes only finite floating point math. That is,
+ there is no NAN or Inf values.
+ 
+ =item B<-enable-unsafe-fp-math>
+ 
+ Causes B<lli> to enable optimizations that may decrease floating point
+ precision.
+ 
+ =item B<-soft-float>
+ 
+ Causes B<lli> to generate software floating point library calls instead of
+ equivalent hardware instructions.
+ 
+ =back
+ 
+ =head1 CODE GENERATION OPTIONS
+ 
+ =over
+ 
+ =item B<-code-model>=I<model>
+ 
+ Choose the code model from:
+ 
+     default: Target default code model
+     small: Small code model
+     kernel: Kernel code model
+     medium: Medium code model
+     large: Large code model
+ 
+ =item B<-disable-post-RA-scheduler>
+ 
+ Disable scheduling after register allocation.
+ 
+ =item B<-disable-spill-fusing>
+ 
+ Disable fusing of spill code into instructions.
+ 
+ =item B<-enable-correct-eh-support> 
+ 
+ Make the -lowerinvoke pass insert expensive, but correct, EH code.
+ 
+ =item B<-enable-eh> 
+ 
+ Exception handling should be emitted.
+ 
+ =item B<-join-liveintervals> 
+ 
+ Coalesce copies (default=true).
+ 
+ =item B<-nozero-initialized-in-bss>
+ Don't place zero-initialized symbols into the BSS section.
+ 
+ =item B<-pre-RA-sched>=I<scheduler>
+ 
+ Instruction schedulers available (before register allocation):
+ 
+     =default: Best scheduler for the target 
+     =none: No scheduling: breadth first sequencing 
+     =simple: Simple two pass scheduling: minimize critical path and maximize processor utilization 
+     =simple-noitin: Simple two pass scheduling: Same as simple except using generic latency 
+     =list-burr: Bottom-up register reduction list scheduling 
+     =list-tdrr: Top-down register reduction list scheduling 
+     =list-td: Top-down list scheduler -print-machineinstrs - Print generated machine code
+ 
+ =item B<-regalloc>=I<allocator>
+ 
+ Register allocator to use: (default = linearscan)
+ 
+     =bigblock: Big-block register allocator 
+     =linearscan: linear scan register allocator =local -   local register allocator 
+     =simple: simple register allocator 
+ 
+ =item B<-relocation-model>=I<model> 
+ 
+ Choose relocation model from:
+ 
+     =default: Target default relocation model 
+     =static: Non-relocatable code =pic -   Fully relocatable, position independent code 
+     =dynamic-no-pic: Relocatable external references, non-relocatable code 
+ 
+ =item B<-spiller>
+ 
+ Spiller to use: (default: local) 
+ 
+     =simple: simple spiller 
+     =local: local spiller 
+ 
+ =item B<-x86-asm-syntax>=I<syntax>
+ 
+ Choose style of code to emit from X86 backend: 
+ 
+     =att: Emit AT&T-style assembly 
+     =intel: Emit Intel-style assembly
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<lli> fails to load the program, it will exit with an exit code of 1.
+ Otherwise, it will return the exit code of the program it executes.
+ 
+ =head1 SEE ALSO
+ 
+ L<llc|llc>
+ 
+ =head1 AUTHOR
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-ar.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-ar.pod:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-ar.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,406 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-ar - LLVM archiver
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-ar> [-]{dmpqrtx}[Rabfikouz] [relpos] [count] <archive> [files...]
+ 
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-ar> command is similar to the common Unix utility, C<ar>. It 
+ archives several files together into a single file. The intent for this is
+ to produce archive libraries by LLVM bitcode that can be linked into an
+ LLVM program. However, the archive can contain any kind of file. By default,
+ B<llvm-ar> generates a symbol table that makes linking faster because
+ only the symbol table needs to be consulted, not each individual file member
+ of the archive. 
+ 
+ The B<llvm-ar> command can be used to I<read> both SVR4 and BSD style archive
+ files. However, it cannot be used to write them.  While the B<llvm-ar> command 
+ produces files that are I<almost> identical to the format used by other C<ar> 
+ implementations, it has two significant departures in order to make the 
+ archive appropriate for LLVM. The first departure is that B<llvm-ar> only
+ uses BSD4.4 style long path names (stored immediately after the header) and
+ never contains a string table for long names. The second departure is that the
+ symbol table is formated for efficient construction of an in-memory data
+ structure that permits rapid (red-black tree) lookups. Consequently, archives 
+ produced with B<llvm-ar> usually won't be readable or editable with any
+ C<ar> implementation or useful for linking.  Using the C<f> modifier to flatten
+ file names will make the archive readable by other C<ar> implementations
+ but not for linking because the symbol table format for LLVM is unique. If an
+ SVR4 or BSD style archive is used with the C<r> (replace) or C<q> (quick
+ update) operations, the archive will be reconstructed in LLVM format. This 
+ means that the string table will be dropped (in deference to BSD 4.4 long names)
+ and an LLVM symbol table will be added (by default). The system symbol table
+ will be retained.
+ 
+ Here's where B<llvm-ar> departs from previous C<ar> implementations:
+ 
+ =over
+ 
+ =item I<Symbol Table>
+ 
+ Since B<llvm-ar> is intended to archive bitcode files, the symbol table
+ won't make much sense to anything but LLVM. Consequently, the symbol table's
+ format has been simplified. It consists simply of a sequence of pairs
+ of a file member index number as an LSB 4byte integer and a null-terminated 
+ string.
+ 
+ =item I<Long Paths>
+ 
+ Some C<ar> implementations (SVR4) use a separate file member to record long
+ path names (> 15 characters). B<llvm-ar> takes the BSD 4.4 and Mac OS X 
+ approach which is to simply store the full path name immediately preceding
+ the data for the file. The path name is null terminated and may contain the
+ slash (/) character. 
+ 
+ =item I<Compression>
+ 
+ B<llvm-ar> can compress the members of an archive to save space. The 
+ compression used depends on what's available on the platform and what choices
+ the LLVM Compressor utility makes. It generally favors bzip2 but will select
+ between "no compression" or bzip2 depending on what makes sense for the
+ file's content.
+ 
+ =item I<Directory Recursion>
+ 
+ Most C<ar> implementations do not recurse through directories but simply
+ ignore directories if they are presented to the program in the F<files> 
+ option. B<llvm-ar>, however, can recurse through directory structures and
+ add all the files under a directory, if requested.
+ 
+ =item I<TOC Verbose Output>
+ 
+ When B<llvm-ar> prints out the verbose table of contents (C<tv> option), it
+ precedes the usual output with a character indicating the basic kind of 
+ content in the file. A blank means the file is a regular file. A 'Z' means
+ the file is compressed. A 'B' means the file is an LLVM bitcode file. An
+ 'S' means the file is the symbol table.
+ 
+ =back
+ 
+ =head1 OPTIONS
+ 
+ The options to B<llvm-ar> are compatible with other C<ar> implementations.
+ However, there are a few modifiers (F<zR>) that are not found in other
+ C<ar>s. The options to B<llvm-ar> specify a single basic operation to 
+ perform on the archive, a variety of modifiers for that operation, the
+ name of the archive file, and an optional list of file names. These options
+ are used to determine how B<llvm-ar> should process the archive file.
+ 
+ The Operations and Modifiers are explained in the sections below. The minimal
+ set of options is at least one operator and the name of the archive. Typically
+ archive files end with a C<.a> suffix, but this is not required. Following
+ the F<archive-name> comes a list of F<files> that indicate the specific members
+ of the archive to operate on. If the F<files> option is not specified, it
+ generally means either "none" or "all" members, depending on the operation.
+ 
+ =head2 Operations
+ 
+ =over
+ 
+ =item d
+ 
+ Delete files from the archive. No modifiers are applicable to this operation.
+ The F<files> options specify which members should be removed from the
+ archive. It is not an error if a specified file does not appear in the archive.
+ If no F<files> are specified, the archive is not modified.
+ 
+ =item m[abi]
+ 
+ Move files from one location in the archive to another. The F<a>, F<b>, and 
+ F<i> modifiers apply to this operation. The F<files> will all be moved
+ to the location given by the modifiers. If no modifiers are used, the files
+ will be moved to the end of the archive. If no F<files> are specified, the
+ archive is not modified.
+ 
+ =item p[k]
+ 
+ Print files to the standard output. The F<k> modifier applies to this
+ operation. This operation simply prints the F<files> indicated to the
+ standard output. If no F<files> are specified, the entire archive is printed.
+ Printing bitcode files is ill-advised as they might confuse your terminal
+ settings. The F<p> operation never modifies the archive.
+ 
+ =item q[Rfz]
+ 
+ Quickly append files to the end of the archive. The F<R>, F<f>, and F<z>
+ modifiers apply to this operation.  This operation quickly adds the 
+ F<files> to the archive without checking for duplicates that should be 
+ removed first. If no F<files> are specified, the archive is not modified. 
+ Because of the way that B<llvm-ar> constructs the archive file, its dubious 
+ whether the F<q> operation is any faster than the F<r> operation.
+ 
+ =item r[Rabfuz]
+ 
+ Replace or insert file members. The F<R>, F<a>, F<b>, F<f>, F<u>, and F<z>
+ modifiers apply to this operation. This operation will replace existing
+ F<files> or insert them at the end of the archive if they do not exist. If no
+ F<files> are specified, the archive is not modified.
+ 
+ =item t[v]
+ 
+ Print the table of contents. Without any modifiers, this operation just prints
+ the names of the members to the standard output. With the F<v> modifier,
+ B<llvm-ar> also prints out the file type (B=bitcode, Z=compressed, S=symbol
+ table, blank=regular file), the permission mode, the owner and group, the
+ size, and the date. If any F<files> are specified, the listing is only for
+ those files. If no F<files> are specified, the table of contents for the
+ whole archive is printed.
+ 
+ =item x[oP]
+ 
+ Extract archive members back to files. The F<o> modifier applies to this
+ operation. This operation retrieves the indicated F<files> from the archive 
+ and writes them back to the operating system's file system. If no 
+ F<files> are specified, the entire archive is extract. 
+ 
+ =back
+ 
+ =head2 Modifiers (operation specific)
+ 
+ The modifiers below are specific to certain operations. See the Operations
+ section (above) to determine which modifiers are applicable to which operations.
+ 
+ =over
+ 
+ =item [a] 
+ 
+ When inserting or moving member files, this option specifies the destination of
+ the new files as being C<a>fter the F<relpos> member. If F<relpos> is not found,
+ the files are placed at the end of the archive.
+ 
+ =item [b] 
+ 
+ When inserting or moving member files, this option specifies the destination of
+ the new files as being C<b>efore the F<relpos> member. If F<relpos> is not 
+ found, the files are placed at the end of the archive. This modifier is 
+ identical to the the F<i> modifier.
+ 
+ =item [f] 
+ 
+ Normally, B<llvm-ar> stores the full path name to a file as presented to it on
+ the command line. With this option, truncated (15 characters max) names are
+ used. This ensures name compatibility with older versions of C<ar> but may also
+ thwart correct extraction of the files (duplicates may overwrite). If used with
+ the F<R> option, the directory recursion will be performed but the file names
+ will all be C<f>lattened to simple file names.
+ 
+ =item [i] 
+ 
+ A synonym for the F<b> option.
+ 
+ =item [k] 
+ 
+ Normally, B<llvm-ar> will not print the contents of bitcode files when the 
+ F<p> operation is used. This modifier defeats the default and allows the 
+ bitcode members to be printed.
+ 
+ =item [N] 
+ 
+ This option is ignored by B<llvm-ar> but provided for compatibility.
+ 
+ =item [o] 
+ 
+ When extracting files, this option will cause B<llvm-ar> to preserve the
+ original modification times of the files it writes. 
+ 
+ =item [P] 
+ 
+ use full path names when matching
+ 
+ =item [R]
+ 
+ This modifier instructions the F<r> option to recursively process directories.
+ Without F<R>, directories are ignored and only those F<files> that refer to
+ files will be added to the archive. When F<R> is used, any directories specified
+ with F<files> will be scanned (recursively) to find files to be added to the
+ archive. Any file whose name begins with a dot will not be added.
+ 
+ =item [u] 
+ 
+ When replacing existing files in the archive, only replace those files that have
+ a time stamp than the time stamp of the member in the archive.
+ 
+ =item [z] 
+ 
+ When inserting or replacing any file in the archive, compress the file first.
+ This
+ modifier is safe to use when (previously) compressed bitcode files are added to
+ the archive; the compressed bitcode files will not be doubly compressed.
+ 
+ =back
+ 
+ =head2 Modifiers (generic)
+ 
+ The modifiers below may be applied to any operation.
+ 
+ =over
+ 
+ =item [c]
+ 
+ For all operations, B<llvm-ar> will always create the archive if it doesn't 
+ exist. Normally, B<llvm-ar> will print a warning message indicating that the
+ archive is being created. Using this modifier turns off that warning.
+ 
+ =item [s]
+ 
+ This modifier requests that an archive index (or symbol table) be added to the
+ archive. This is the default mode of operation. The symbol table will contain
+ all the externally visible functions and global variables defined by all the
+ bitcode files in the archive. Using this modifier is more efficient that using
+ L<llvm-ranlib|llvm-ranlib> which also creates the symbol table.
+ 
+ =item [S]
+ 
+ This modifier is the opposite of the F<s> modifier. It instructs B<llvm-ar> to
+ not build the symbol table. If both F<s> and F<S> are used, the last modifier to
+ occur in the options will prevail. 
+ 
+ =item [v]
+ 
+ This modifier instructs B<llvm-ar> to be verbose about what it is doing. Each
+ editing operation taken against the archive will produce a line of output saying
+ what is being done.
+ 
+ =back
+ 
+ =head1 STANDARDS
+ 
+ The B<llvm-ar> utility is intended to provide a superset of the IEEE Std 1003.2
+ (POSIX.2) functionality for C<ar>. B<llvm-ar> can read both SVR4 and BSD4.4 (or
+ Mac OS X) archives. If the C<f> modifier is given to the C<x> or C<r> operations
+ then B<llvm-ar> will write SVR4 compatible archives. Without this modifier, 
+ B<llvm-ar> will write BSD4.4 compatible archives that have long names
+ immediately after the header and indicated using the "#1/ddd" notation for the
+ name in the header.
+ 
+ =head1 FILE FORMAT
+ 
+ The file format for LLVM Archive files is similar to that of BSD 4.4 or Mac OSX
+ archive files. In fact, except for the symbol table, the C<ar> commands on those
+ operating systems should be able to read LLVM archive files. The details of the
+ file format follow.
+ 
+ Each archive begins with the archive magic number which is the eight printable
+ characters "!<arch>\n" where \n represents the newline character (0x0A). 
+ Following the magic number, the file is composed of even length members that 
+ begin with an archive header and end with a \n padding character if necessary 
+ (to make the length even). Each file member is composed of a header (defined 
+ below), an optional newline-terminated "long file name" and the contents of 
+ the file. 
+ 
+ The fields of the header are described in the items below. All fields of the
+ header contain only ASCII characters, are left justified and are right padded 
+ with space characters.
+ 
+ =over
+ 
+ =item name - char[16]
+ 
+ This field of the header provides the name of the archive member. If the name is
+ longer than 15 characters or contains a slash (/) character, then this field
+ contains C<#1/nnn> where C<nnn> provides the length of the name and the C<#1/>
+ is literal.  In this case, the actual name of the file is provided in the C<nnn>
+ bytes immediately following the header. If the name is 15 characters or less, it
+ is contained directly in this field and terminated with a slash (/) character.
+ 
+ =item date - char[12]
+ 
+ This field provides the date of modification of the file in the form of a
+ decimal encoded number that provides the number of seconds since the epoch 
+ (since 00:00:00 Jan 1, 1970) per Posix specifications.
+ 
+ =item uid - char[6]
+ 
+ This field provides the user id of the file encoded as a decimal ASCII string.
+ This field might not make much sense on non-Unix systems. On Unix, it is the
+ same value as the st_uid field of the stat structure returned by the stat(2)
+ operating system call.
+ 
+ =item gid - char[6]
+ 
+ This field provides the group id of the file encoded as a decimal ASCII string.
+ This field might not make much sense on non-Unix systems. On Unix, it is the
+ same value as the st_gid field of the stat structure returned by the stat(2)
+ operating system call.
+ 
+ =item mode - char[8]
+ 
+ This field provides the access mode of the file encoded as an octal ASCII 
+ string. This field might not make much sense on non-Unix systems. On Unix, it 
+ is the same value as the st_mode field of the stat structure returned by the 
+ stat(2) operating system call.
+ 
+ =item size - char[10]
+ 
+ This field provides the size of the file, in bytes, encoded as a decimal ASCII
+ string. If the size field is negative (starts with a minus sign, 0x02D), then
+ the archive member is stored in compressed form. The first byte of the archive
+ member's data indicates the compression type used. A value of 0 (0x30) indicates
+ that no compression was used. A value of 2 (0x32) indicates that bzip2
+ compression was used.
+ 
+ =item fmag - char[2]
+ 
+ This field is the archive file member magic number. Its content is always the
+ two characters back tick (0x60) and newline (0x0A). This provides some measure 
+ utility in identifying archive files that have been corrupted.
+ 
+ =back 
+ 
+ The LLVM symbol table has the special name "#_LLVM_SYM_TAB_#". It is presumed
+ that no regular archive member file will want this name. The LLVM symbol table 
+ is simply composed of a sequence of triplets: byte offset, length of symbol, 
+ and the symbol itself. Symbols are not null or newline terminated. Here are 
+ the details on each of these items:
+ 
+ =over
+ 
+ =item offset - vbr encoded 32-bit integer
+ 
+ The offset item provides the offset into the archive file where the bitcode
+ member is stored that is associated with the symbol. The offset value is 0
+ based at the start of the first "normal" file member. To derive the actual
+ file offset of the member, you must add the number of bytes occupied by the file
+ signature (8 bytes) and the symbol tables. The value of this item is encoded
+ using variable bit rate encoding to reduce the size of the symbol table.
+ Variable bit rate encoding uses the high bit (0x80) of each byte to indicate 
+ if there are more bytes to follow. The remaining 7 bits in each byte carry bits
+ from the value. The final byte does not have the high bit set.
+ 
+ =item length - vbr encoded 32-bit integer
+ 
+ The length item provides the length of the symbol that follows. Like this
+ I<offset> item, the length is variable bit rate encoded.
+ 
+ =item symbol - character array
+ 
+ The symbol item provides the text of the symbol that is associated with the
+ I<offset>. The symbol is not terminated by any character. Its length is provided
+ by the I<length> field. Note that is allowed (but unwise) to use non-printing
+ characters (even 0x00) in the symbol. This allows for multiple encodings of 
+ symbol names.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-ar> succeeds, it will exit with 0.  A usage error, results
+ in an exit code of 1. A hard (file system typically) error results in an
+ exit code of 2. Miscellaneous or unknown errors result in an
+ exit code of 3.
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-ranlib|llvm-ranlib>, ar(1)
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-as.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-as.pod:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-as.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,77 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-as - LLVM assembler
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-as> [I<options>] [I<filename>]
+ 
+ =head1 DESCRIPTION
+ 
+ B<llvm-as> is the LLVM assembler.  It reads a file containing human-readable
+ LLVM assembly language, translates it to LLVM bitcode, and writes the result
+ into a file or to standard output.
+ 
+ If F<filename> is omitted or is C<->, then B<llvm-as> reads its input from
+ standard input.
+ 
+ If an output file is not specified with the B<-o> option, then
+ B<llvm-as> sends its output to a file or standard output by following
+ these rules:
+ 
+ =over 
+ 
+ =item *
+ 
+ If the input is standard input, then the output is standard output.
+ 
+ =item *
+ 
+ If the input is a file that ends with C<.ll>, then the output file is of
+ the same name, except that the suffix is changed to C<.bc>.
+ 
+ =item *
+ 
+ If the input is a file that does not end with the C<.ll> suffix, then the
+ output file has the same name as the input file, except that the C<.bc>
+ suffix is appended.
+ 
+ =back
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<-f>
+ 
+ Force overwrite.  Normally, B<llvm-as> will refuse to overwrite an
+ output file that already exists.  With this option, B<llvm-as>
+ will overwrite the output file and replace it with new bitcode.
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-o> F<filename>
+ 
+ Specify the output file name.  If F<filename> is C<->, then B<llvm-as>
+ sends its output to standard output.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-as> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-dis|llvm-dis>, L<gccas|gccas>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-bcanalyzer.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-bcanalyzer.pod:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-bcanalyzer.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,315 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-bcanalyzer - LLVM bitcode analyzer
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-bcanalyzer> [I<options>] [F<filename>]
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-bcanalyzer> command is a small utility for analyzing bitcode files.
+ The tool reads a bitcode file (such as generated with the B<llvm-as> tool) and
+ produces a statistical report on the contents of the bitcode file.  The tool
+ can also dump a low level but human readable version of the bitcode file. 
+ This tool is probably not of much interest or utility except for those working 
+ directly with the bitcode file format. Most LLVM users can just ignore
+ this tool.
+ 
+ If F<filename> is omitted or is C<->, then B<llvm-bcanalyzer> reads its input 
+ from standard input. This is useful for combining the tool into a pipeline.
+ Output is written to the standard output.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<-nodetails>
+ 
+ Causes B<llvm-bcanalyzer> to abbreviate its output by writing out only a module 
+ level summary. The details for individual functions are not displayed.
+ 
+ =item B<-dump>
+ 
+ Causes B<llvm-bcanalyzer> to dump the bitcode in a human readable format. This 
+ format is significantly different from LLVM assembly and provides details about 
+ the encoding of the bitcode file.
+ 
+ =item B<-verify>
+ 
+ Causes B<llvm-bcanalyzer> to verify the module produced by reading the 
+ bitcode. This ensures that the statistics generated are based on a consistent
+ module.
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-bcanalyzer> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value, usually 1.
+ 
+ =head1 SUMMARY OUTPUT DEFINITIONS
+ 
+ The following items are always printed by llvm-bcanalyzer. They comprize the
+ summary output.
+ 
+ =over
+ 
+ =item B<Bitcode Analysis Of Module>
+ 
+ This just provides the name of the module for which bitcode analysis is being
+ generated.
+ 
+ =item B<Bitcode Version Number>
+ 
+ The bitcode version (not LLVM version) of the file read by the analyzer.
+ 
+ =item B<File Size>
+ 
+ The size, in bytes, of the entire bitcode file.
+ 
+ =item B<Module Bytes>
+ 
+ The size, in bytes, of the module block. Percentage is relative to File Size.
+ 
+ =item B<Function Bytes>
+ 
+ The size, in bytes, of all the function blocks. Percentage is relative to File
+ Size.
+ 
+ =item B<Global Types Bytes>
+ 
+ The size, in bytes, of the Global Types Pool. Percentage is relative to File
+ Size. This is the size of the definitions of all types in the bitcode file.
+ 
+ =item B<Constant Pool Bytes>
+ 
+ The size, in bytes, of the Constant Pool Blocks Percentage is relative to File
+ Size.
+ 
+ =item B<Module Globals Bytes>
+ 
+ Ths size, in bytes, of the Global Variable Definitions and their initializers.
+ Percentage is relative to File Size.
+ 
+ =item B<Instruction List Bytes>
+ 
+ The size, in bytes, of all the instruction lists in all the functions.
+ Percentage is relative to File Size. Note that this value is also included in
+ the Function Bytes.
+ 
+ =item B<Compaction Table Bytes>
+ 
+ The size, in bytes, of all the compaction tables in all the functions.
+ Percentage is relative to File Size. Note that this value is also included in
+ the Function Bytes.
+ 
+ =item B<Symbol Table Bytes>
+ 
+ The size, in bytes, of all the symbol tables in all the functions. Percentage is
+ relative to File Size. Note that this value is also included in the Function
+ Bytes.
+ 
+ =item B<Dependent Libraries Bytes>
+ 
+ The size, in bytes, of the list of dependent libraries in the module. Percentage
+ is relative to File Size. Note that this value is also included in the Module
+ Global Bytes.
+ 
+ =item B<Number Of Bitcode Blocks>
+ 
+ The total number of blocks of any kind in the bitcode file.
+ 
+ =item B<Number Of Functions>
+ 
+ The total number of function definitions in the bitcode file.
+ 
+ =item B<Number Of Types>
+ 
+ The total number of types defined in the Global Types Pool.
+ 
+ =item B<Number Of Constants>
+ 
+ The total number of constants (of any type) defined in the Constant Pool.
+ 
+ =item B<Number Of Basic Blocks>
+ 
+ The total number of basic blocks defined in all functions in the bitcode file.
+ 
+ =item B<Number Of Instructions>
+ 
+ The total number of instructions defined in all functions in the bitcode file.
+ 
+ =item B<Number Of Long Instructions>
+ 
+ The total number of long instructions defined in all functions in the bitcode
+ file. Long instructions are those taking greater than 4 bytes. Typically long
+ instructions are GetElementPtr with several indices, PHI nodes, and calls to
+ functions with large numbers of arguments.
+ 
+ =item B<Number Of Operands>
+ 
+ The total number of operands used in all instructions in the bitcode file.
+ 
+ =item B<Number Of Compaction Tables>
+ 
+ The total number of compaction tables in all functions in the bitcode file.
+ 
+ =item B<Number Of Symbol Tables>
+ 
+ The total number of symbol tables in all functions in the bitcode file.
+ 
+ =item B<Number Of Dependent Libs>
+ 
+ The total number of dependent libraries found in the bitcode file.
+ 
+ =item B<Total Instruction Size>
+ 
+ The total size of the instructions in all functions in the bitcode file.
+ 
+ =item B<Average Instruction Size>
+ 
+ The average number of bytes per instruction across all functions in the bitcode
+ file. This value is computed by dividing Total Instruction Size by Number Of
+ Instructions.
+ 
+ =item B<Maximum Type Slot Number>
+ 
+ The maximum value used for a type's slot number. Larger slot number values take 
+ more bytes to encode.
+ 
+ =item B<Maximum Value Slot Number>
+ 
+ The maximum value used for a value's slot number. Larger slot number values take 
+ more bytes to encode.
+ 
+ =item B<Bytes Per Value>
+ 
+ The average size of a Value definition (of any type). This is computed by
+ dividing File Size by the total number of values of any type.
+ 
+ =item B<Bytes Per Global>
+ 
+ The average size of a global definition (constants and global variables).
+ 
+ =item B<Bytes Per Function>
+ 
+ The average number of bytes per function definition. This is computed by
+ dividing Function Bytes by Number Of Functions.
+ 
+ =item B<# of VBR 32-bit Integers>
+ 
+ The total number of 32-bit integers encoded using the Variable Bit Rate
+ encoding scheme.
+ 
+ =item B<# of VBR 64-bit Integers>
+ 
+ The total number of 64-bit integers encoded using the Variable Bit Rate encoding
+ scheme.
+ 
+ =item B<# of VBR Compressed Bytes>
+ 
+ The total number of bytes consumed by the 32-bit and 64-bit integers that use
+ the Variable Bit Rate encoding scheme.
+ 
+ =item B<# of VBR Expanded Bytes>
+ 
+ The total number of bytes that would have been consumed by the 32-bit and 64-bit
+ integers had they not been compressed with the Variable Bit Rage encoding
+ scheme.
+ 
+ =item B<Bytes Saved With VBR>
+ 
+ The total number of bytes saved by using the Variable Bit Rate encoding scheme.
+ The percentage is relative to # of VBR Expanded Bytes.
+ 
+ =back
+ 
+ =head1 DETAILED OUTPUT DEFINITIONS
+ 
+ The following definitions occur only if the -nodetails option was not given.
+ The detailed output provides additional information on a per-function basis.
+ 
+ =over
+ 
+ =item B<Type>
+ 
+ The type signature of the function.
+ 
+ =item B<Byte Size>
+ 
+ The total number of bytes in the function's block.
+ 
+ =item B<Basic Blocks>
+ 
+ The number of basic blocks defined by the function.
+ 
+ =item B<Instructions>
+ 
+ The number of instructions defined by the function.
+ 
+ =item B<Long Instructions>
+ 
+ The number of instructions using the long instruction format in the function.
+ 
+ =item B<Operands>
+ 
+ The number of operands used by all instructions in the function.
+ 
+ =item B<Instruction Size>
+ 
+ The number of bytes consumed by instructions in the function.
+ 
+ =item B<Average Instruction Size>
+ 
+ The average number of bytes consumed by the instructions in the funtion. This
+ value is computed by dividing Instruction Size by Instructions.
+ 
+ =item B<Bytes Per Instruction>
+ 
+ The average number of bytes used by the function per instruction. This value is
+ computed by dividing Byte Size by Instructions. Note that this is not the same
+ as Average Instruction Size. It computes a number relative to the total function
+ size not just the size of the instruction list.
+ 
+ =item B<Number of VBR 32-bit Integers>
+ 
+ The total number of 32-bit integers found in this function (for any use).
+ 
+ =item B<Number of VBR 64-bit Integers>
+ 
+ The total number of 64-bit integers found in this function (for any use).
+ 
+ =item B<Number of VBR Compressed Bytes>
+ 
+ The total number of bytes in this function consumed by the 32-bit and 64-bit 
+ integers that use the Variable Bit Rate encoding scheme.
+ 
+ =item B<Number of VBR Expanded Bytes>
+ 
+ The total number of bytes in this function that would have been consumed by 
+ the 32-bit and 64-bit integers had they not been compressed with the Variable 
+ Bit Rate encoding scheme.
+ 
+ =item B<Bytes Saved With VBR>
+ 
+ The total number of bytes saved in this function by using the Variable Bit 
+ Rate encoding scheme. The percentage is relative to # of VBR Expanded Bytes.
+ 
+ =back
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-dis|llvm-dis>, L<http://llvm.org/docs/BitCodeFormat.html>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-config.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-config.pod:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-config.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,131 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-config - Print LLVM compilation options
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-config> I<option> [I<components>...]
+ 
+ =head1 DESCRIPTION
+ 
+ B<llvm-config> makes it easier to build applications that use LLVM.  It can
+ print the compiler flags, linker flags and object libraries needed to link
+ against LLVM.
+ 
+ =head1 EXAMPLES
+ 
+ To link against the JIT:
+ 
+   g++ `llvm-config --cxxflags` -o HowToUseJIT.o -c HowToUseJIT.cpp
+   g++ `llvm-config --ldflags` -o HowToUseJIT HowToUseJIT.o \
+       `llvm-config --libs engine bcreader scalaropts`
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<--version>
+ 
+ Print the version number of LLVM.
+ 
+ =item B<--help>
+ 
+ Print a summary of B<llvm-config> arguments.
+ 
+ =item B<--prefix>
+ 
+ Print the installation prefix for LLVM.
+ 
+ =item B<--src-root>
+ 
+ Print the source root from which LLVM was built.
+ 
+ =item B<--obj-root>
+ 
+ Print the object root used to build LLVM.
+ 
+ =item B<--bindir>
+ 
+ Print the installation directory for LLVM binaries.
+ 
+ =item B<--includedir>
+ 
+ Print the installation directory for LLVM headers.
+ 
+ =item B<--libdir>
+ 
+ Print the installation directory for LLVM libraries.
+ 
+ =item B<--cxxflags>
+ 
+ Print the C++ compiler flags needed to use LLVM headers.
+ 
+ =item B<--ldflags>
+ 
+ Print the flags needed to link against LLVM libraries.
+ 
+ =item B<--libs>
+ 
+ Print all the libraries needed to link against the specified LLVM
+ I<components>, including any dependencies.
+ 
+ =item B<--libnames>
+ 
+ Similar to B<--libs>, but prints the bare filenames of the libraries
+ without B<-l> or pathnames.  Useful for linking against a not-yet-installed
+ copy of LLVM.
+ 
+ =item B<--libfiles>
+ 
+ Similar to B<--libs>, but print the full path to each library file.  This is
+ useful when creating makefile dependencies, to ensure that a tool is relinked if
+ any library it uses changes.
+ 
+ =item B<--components>
+ 
+ Print all valid component names.
+ 
+ =item B<--targets-built>
+ 
+ Print the component names for all targets supported by this copy of LLVM.
+ 
+ =item B<--build-mode>
+ 
+ Print the build mode used when LLVM was built (e.g. Debug or Release)
+ 
+ =back
+ 
+ =head1 COMPONENTS
+ 
+ To print a list of all available components, run B<llvm-config
+ --components>.  In most cases, components correspond directly to LLVM
+ libraries.  Useful "virtual" components include:
+ 
+ =over
+ 
+ =item B<all>
+ 
+ Includes all LLVM libaries.  The default if no components are specified.
+ 
+ =item B<backend>
+ 
+ Includes either a native backend or the C backend.
+ 
+ =item B<engine>
+ 
+ Includes either a native JIT or the bitcode interpreter.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-config> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-db.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-db.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-db.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,16 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-db - LLVM debugger (alpha)
+ 
+ =head1 SYNOPSIS
+ 
+ Details coming soon. Please see 
+ L<http://llvm.org/docs/SourceLevelDebugging.html> in the meantime.
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-dis.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-dis.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-dis.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,60 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-dis - LLVM disassembler
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-dis> [I<options>] [I<filename>]
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-dis> command is the LLVM disassembler.  It takes an LLVM
+ bitcode file and converts it into human-readable LLVM assembly language.
+ 
+ If filename is omitted or specified as C<->, B<llvm-dis> reads its
+ input from standard input.
+ 
+ If the input is being read from standard input, then B<llvm-dis>
+ will send its output to standard output by default.  Otherwise, the
+ output will be written to a file named after the input file, with
+ a C<.ll> suffix added (any existing C<.bc> suffix will first be
+ removed).  You can override the choice of output file using the
+ B<-o> option.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<-f>
+ 
+ Force overwrite.  Normally, B<llvm-dis> will refuse to overwrite
+ an output file that already exists.  With this option, B<llvm-dis>
+ will overwrite the output file.
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-o> F<filename>
+ 
+ Specify the output file name.  If F<filename> is -, then the output is sent
+ to standard output.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-dis> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-as|llvm-as>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-extract.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-extract.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-extract.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,63 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-extract - extract a function from an LLVM module
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-extract> [I<options>] B<--func> I<function-name> [I<filename>]
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-extract> command takes the name of a function and extracts it from
+ the specified LLVM bitcode file.  It is primarily used as a debugging tool to
+ reduce test cases from larger programs that are triggering a bug.
+ 
+ In addition to extracting the bitcode of the specified function,
+ B<llvm-extract> will also remove unreachable global variables, prototypes, and
+ unused types.
+ 
+ The B<llvm-extract> command reads its input from standard input if filename is
+ omitted or if filename is -.  The output is always written to standard output,
+ unless the B<-o> option is specified (see below).
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<-f>
+ 
+ Force overwrite.  Normally, B<llvm-extract> will refuse to overwrite an
+ output file that already exists.  With this option, B<llvm-extract>
+ will overwrite the output file and replace it with new bitcode.
+ 
+ =item B<--func> I<function-name>
+ 
+ Extract the function named I<function-name> from the LLVM bitcode.
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-o> I<filename>
+ 
+ Specify the output filename.  If filename is "-" (the default), then
+ B<llvm-extract> sends its output to standard output.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-extract> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.
+ 
+ =head1 SEE ALSO
+ 
+ L<bugpoint|bugpoint>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-ld.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-ld.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-ld.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,234 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-ld - LLVM linker
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-ld> <options> <files>
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-ld> tool takes a set of LLVM bitcode files and links them
+ together into a single LLVM bitcode file.  The output bitcode file can be
+ another bitcode file or an executable bitcode program.  Using additional
+ options, B<llvm-ld> is able to produce native code executables.
+ 
+ The B<llvm-ld> tool is the main linker for LLVM. It is used to link together
+ the output of LLVM front-end compilers and run "link time" optimizations (mostly
+ the inter-procedural kind).
+ 
+ The B<llvm-ld> tools attempts to mimic the interface provided by the default
+ system linker so that it can act as a I<drop-in> replacement.
+ 
+ =head2 Search Order
+ 
+ When looking for objects specified on the command line, B<llvm-ld> will search 
+ for the object first in the current directory and then in the directory 
+ specified by the B<LLVM_LIB_SEARCH_PATH> environment variable.  If it cannot 
+ find the object, it fails.
+ 
+ When looking for a library specified with the B<-l> option, B<llvm-ld> first
+ attempts to load a file with that name from the current directory.  If that
+ fails, it looks for libI<library>.bc, libI<library>.a, or libI<library>.I<shared
+ library extension>, in that order, in each directory added to the library search
+ path with the B<-L> option.  These directories are searched in the order they
+ are specified.  If the library cannot be located, then B<llvm-ld> looks in the
+ directory specified by the B<LLVM_LIB_SEARCH_PATH> environment variable.  If it
+ does not find a library there, it fails.
+ 
+ The I<shared library extension> may be I<.so>, I<.dyld>, I<.dll>, or something
+ different, depending upon the system.
+ 
+ The B<-L> option is global.  It does not matter where it is specified in the
+ list of command line arguments; the directory is simply added to the search path
+ and is applied to all libraries, preceding or succeeding, in the command line.
+ 
+ =head2 Link order
+ 
+ All object and bitcode files are linked first in the order they were 
+ specified on the command line.  All library files are linked next.  
+ Some libraries may not be linked into the object program; see below.
+ 
+ =head2 Library Linkage
+ 
+ Object files and static bitcode objects are always linked into the output
+ file.  Library archives (.a files) load only the objects within the archive
+ that define symbols needed by the output file.  Hence, libraries should be
+ listed after the object files and libraries which need them; otherwise, the
+ library may not be linked in, and the dependent library will not have its
+ undefined symbols defined.
+ 
+ =head2 Native code generation
+ 
+ The B<llvm-ld> program has limited support for native code generation, when
+ using the B<-native> or B<-native-cbe> options. Native code generation is
+ performed by converting the linked bitcode into native assembly (.s) or C code
+ and running the system compiler (typically gcc) on the result.
+ 
+ =head1 OPTIONS
+ 
+ =head2 General Options
+ 
+ =over 
+ 
+ =item B<-help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-v>
+ 
+ Specifies verbose mode. In this mode the linker will print additional
+ information about the actions it takes, programs it executes, etc. 
+ 
+ =item B<-stats>
+ 
+ Print statistics.
+ 
+ =item B<-time-passes>
+ 
+ Record the amount of time needed for each pass and print it to standard
+ error.
+ 
+ =back 
+ 
+ =head2 Input/Output Options
+ 
+ =over
+ 
+ =item B<-o> F<filename>
+ 
+ This overrides the default output file and specifies the name of the file that
+ should be generated by the linker. By default, B<llvm-ld> generates a file named
+ F<a.out> for compatibility with B<ld>. The output will be written to
+ F<filename>.
+ 
+ =item B<-b> F<filename>
+ 
+ This option can be used to override the output bitcode file name. By default, 
+ the name of the bitcode output file is one more ".bc" suffix added to the name 
+ specified by B<-o filename> option.
+ 
+ =item B<-l>F<name>
+ 
+ This option specifies the F<name> of a library to search when resolving symbols
+ for the program. Only the base name should be specified as F<name>, without a
+ F<lib> prefix or any suffix. 
+ 
+ =item B<-L>F<Path>
+ 
+ This option tells B<llvm-ld> to look in F<Path> to find any library subsequently
+ specified with the B<-l> option. The paths will be searched in the order in
+ which they are specified on the command line. If the library is still not found,
+ a small set of system specific directories will also be searched. Note that
+ libraries specified with the B<-l> option that occur I<before> any B<-L> options
+ will not search the paths given by the B<-L> options following it.
+ 
+ =item B<-link-as-library>
+ 
+ Link the bitcode files together as a library, not an executable. In this mode,
+ undefined symbols will be permitted.
+ 
+ =item B<-r>
+ 
+ An alias for -link-as-library.
+ 
+ =item B<-native>
+ 
+ Generate a native machine code executable.
+ 
+ When generating native executables, B<llvm-ld> first checks for a bitcode
+ version of the library and links it in, if necessary.  If the library is
+ missing, B<llvm-ld> skips it.  Then, B<llvm-ld> links in the same
+ libraries as native code.
+ 
+ In this way, B<llvm-ld> should be able to link in optimized bitcode
+ subsets of common libraries and then link in any part of the library that
+ hasn't been converted to bitcode.
+ 
+ =item B<-native-cbe>
+ 
+ Generate a native machine code executable with the LLVM C backend.
+       
+ This option is identical to the B<-native> option, but uses the
+ C backend to generate code for the program instead of an LLVM native
+ code generator.
+ 
+ =back
+ 
+ =head2 Optimization Options
+ 
+ =over 
+ 
+ =item B<-disable-inlining>
+ 
+ Do not run the inlining pass. Functions will not be inlined into other
+ functions.
+ 
+ =item B<-disable-opt>
+ 
+ Completely disable optimization.
+ 
+ =item B<-disable-internalize>
+ 
+ Do not mark all symbols as internal.
+ 
+ =item B<-verify-each>
+ 
+ Run the verification pass after each of the passes to verify intermediate
+ results.
+ 
+ =item B<-strip-all>
+ 
+ Strip all debug and symbol information from the executable to make it smaller.
+ 
+ =item B<-strip-debug>
+ 
+ Strip all debug information from the executable to make it smaller.
+ 
+ =item B<-s>
+ 
+ An alias for B<-strip-all>.
+ 
+ =item B<-S>
+ 
+ An alias for B<-strip-debug>.
+ 
+ =item B<-export-dynamic>
+ 
+ An alias for B<-disable-internalize>
+ 
+ =item B<-post-link-opt>F<Path>
+ 
+ Run post-link optimization program. After linking is completed a bitcode file
+ will be generated. It will be passed to the program specified by F<Path> as the
+ first argument. The second argument to the program will be the name of a
+ temporary file into which the program should place its optimized output. For
+ example, the "no-op optimization" would be a simple shell script:
+ 
+     #!/bin/bash
+     cp $1 $2
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-ld> succeeds, it will exit with 0 return code.  If an error occurs,
+ it will exit with a non-zero return code.
+ 
+ =head1 ENVIRONMENT
+ 
+ The C<LLVM_LIB_SEARCH_PATH> environment variable is used to find bitcode
+ libraries. Any paths specified in this variable will be searched after the C<-L>
+ options.
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-link|llvm-link>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-link.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-link.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-link.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,74 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-link - LLVM linker
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-link> [I<options>] I<filename ...>
+ 
+ =head1 DESCRIPTION
+ 
+ B<llvm-link> takes several LLVM bitcode files and links them together into a
+ single LLVM bitcode file.  It writes the output file to standard output, unless
+ the B<-o> option is used to specify a filename.
+ 
+ B<llvm-link> attempts to load the input files from the current directory.  If
+ that fails, it looks for each file in each of the directories specified by the
+ B<-L> options on the command line.  The library search paths are global; each
+ one is searched for every input file if necessary.  The directories are searched
+ in the order they were specified on the command line.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<-L> F<directory>
+ 
+ Add the specified F<directory> to the library search path.  When looking for
+ libraries, B<llvm-link> will look in path name for libraries.  This option can be
+ specified multiple times; B<llvm-link> will search inside these directories in
+ the order in which they were specified on the command line.
+ 
+ =item B<-f>
+ 
+ Overwrite output files.  By default, B<llvm-link> will not overwrite an output
+ file if it already exists.
+ 
+ =item B<-o> F<filename>
+ 
+ Specify the output file name.  If F<filename> is C<->, then B<llvm-link> will
+ write its output to standard output.
+ 
+ =item B<-d>
+ 
+ If specified, B<llvm-link> prints a human-readable version of the output
+ bitcode file to standard error.
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-v>
+ 
+ Verbose mode.  Print information about what B<llvm-link> is doing.  This
+ typically includes a message for each bitcode file linked in and for each
+ library found.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-link> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.
+ 
+ =head1 SEE ALSO
+ 
+ L<gccld|gccld>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-nm.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-nm.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-nm.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,122 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-nm - list LLVM bitcode file's symbol table
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-nm> [I<options>] [I<filenames...>]
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-nm> utility lists the names of symbols from the LLVM bitcode files,
+ or B<ar> archives containing LLVM bitcode files, named on the command line.
+ Each symbol is listed along with some simple information about its provenance.
+ If no file name is specified, or I<-> is used as a file name, B<llvm-nm> will
+ process a bitcode file on its standard input stream.
+ 
+ B<llvm-nm>'s default output format is the traditional BSD B<nm> output format.
+ Each such output record consists of an (optional) 8-digit hexadecimal address,
+ followed by a type code character, followed by a name, for each symbol. One
+ record is printed per line; fields are separated by spaces. When the address is
+ omitted, it is replaced by 8 spaces.
+ 
+ Type code characters currently supported, and their meanings, are as follows:
+ 
+ =over
+ 
+ =item U
+ 
+ Named object is referenced but undefined in this bitcode file
+ 
+ =item C
+ 
+ Common (multiple definitions link together into one def)
+ 
+ =item W
+ 
+ Weak reference (multiple definitions link together into zero or one definitions)
+ 
+ =item t
+ 
+ Local function (text) object
+ 
+ =item T
+ 
+ Global function (text) object
+ 
+ =item d
+ 
+ Local data object
+ 
+ =item D
+ 
+ Global data object
+ 
+ =item ?
+ 
+ Something unrecognizable
+ 
+ =back
+ 
+ Because LLVM bitcode files typically contain objects that are not considered to
+ have addresses until they are linked into an executable image or dynamically
+ compiled "just-in-time", B<llvm-nm> does not print an address for any symbol,
+ even symbols which are defined in the bitcode file.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<-P>
+ 
+ Use POSIX.2 output format. Alias for B<--format=posix>.
+ 
+ =item B<-B>    (default)
+ 
+ Use BSD output format. Alias for B<--format=bsd>.
+ 
+ =item B<--help>
+ 
+ Print a summary of command-line options and their meanings.
+ 
+ =item B<--defined-only>
+ 
+ Print only symbols defined in this bitcode file (as opposed to
+ symbols which may be referenced by objects in this file, but not
+ defined in this file.)
+ 
+ =item B<--extern-only>, B<-g>
+ 
+ Print only symbols whose definitions are external; that is, accessible
+ from other bitcode files.
+ 
+ =item B<--undefined-only>, B<-u>
+ 
+ Print only symbols referenced but not defined in this bitcode file.
+ 
+ =item B<--format=>I<fmt>, B<-f>
+ 
+ Select an output format; I<fmt> may be I<sysv>, I<posix>, or I<bsd>. The
+ default is I<bsd>.
+ 
+ =back
+ 
+ =head1 BUGS
+ 
+ B<llvm-nm> cannot demangle C++ mangled names, like GNU B<nm> can.
+ 
+ =head1 EXIT STATUS
+ 
+ B<llvm-nm> exits with an exit code of zero.
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-dis|llvm-dis>, ar(1), nm(1)
+ 
+ =head1 AUTHOR
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-prof.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-prof.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-prof.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,57 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-prof - print execution profile of LLVM program
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-prof> [I<options>] [I<bitcode file>] [I<llvmprof.out>]
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-prof> tool reads in an F<llvmprof.out> file (which can
+ optionally use a specific file with the third program argument), a bitcode file
+ for the program, and produces a human readable report, suitable for determining
+ where the program hotspots are.
+ 
+ This program is often used in conjunction with the F<utils/profile.pl>
+ script.  This script automatically instruments a program, runs it with the JIT,
+ then runs B<llvm-prof> to format a report.  To get more information about
+ F<utils/profile.pl>, execute it with the B<--help> option.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<--annotated-llvm> or B<-A>
+ 
+ In addition to the normal report printed, print out the code for the
+ program, annotated with execution frequency information. This can be
+ particularly useful when trying to visualize how frequently basic blocks
+ are executed.  This is most useful with basic block profiling
+ information or better.
+ 
+ =item B<--print-all-code>
+ 
+ Using this option enables the B<--annotated-llvm> option, but it
+ prints the entire module, instead of just the most commonly executed
+ functions.
+ 
+ =item B<--time-passes>
+ 
+ Record the amount of time needed for each pass and print it to standard
+ error.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ B<llvm-prof> returns 1 if it cannot load the bitcode file or the profile
+ information. Otherwise, it exits with zero.
+ 
+ =head1 AUTHOR
+ 
+ B<llvm-prof> is maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvm-ranlib.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvm-ranlib.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvm-ranlib.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,52 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-ranlib - Generate index for LLVM archive
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-ranlib> [--version] [--help] <archive-file>
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-ranlib> command is similar to the common Unix utility, C<ranlib>. It
+ adds or updates the symbol table in an LLVM archive file. Note that using the
+ B<llvm-ar> modifier F<s> is usually more efficient than running B<llvm-ranlib>
+ which is only provided only for completness and compatibility. Unlike other 
+ implementations of C<ranlib>, B<llvm-ranlib> indexes LLVM bitcode files, not
+ native object modules. You can list the contents of the symbol table with the
+ C<llvm-nm -s> command.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item F<archive-file>
+ 
+ Specifies the archive-file to which the symbol table is added or updated.
+ 
+ =item F<--version>
+ 
+ Print the version of B<llvm-ranlib> and exit without building a symbol table.
+ 
+ =item F<--help>
+ 
+ Print usage help for B<llvm-ranlib> and exit without building a symbol table.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-ranlib> succeeds, it will exit with 0.  If an error occurs, a non-zero
+ exit code will be returned.
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-ar|llvm-ar>, ranlib(1)
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvmc.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvmc.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvmc.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,179 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvmc - The LLVM Compiler Driver (WIP)
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvmc> [I<options>] I<filenames...>
+ 
+ =head1 DESCRIPTION
+ 
+ B<llvmc> is a configurable driver for invoking other LLVM (and non-LLVM) tools
+ in order to compile, optimize and link software for multiple languages. For
+ those familiar with FSF's B<gcc> tool, it is very similar.  Please note that
+ B<llvmc> is considered an experimental tool.
+ 
+ =head1 OPTIONS
+ 
+ =head2 Built-in Options
+ 
+ LLVMC has some built-in options that can't be overridden in the
+ configuration libraries.
+ 
+ =over
+ 
+ =item B<-o> I<filename>
+ 
+ Output file name.
+ 
+ =item B<-x> I<language>
+ 
+ Specify the language of the following input files until the next B<-x>
+ option.
+ 
+ =item B<-load> I<plugin_name>
+ 
+ Load the specified plugin DLL. Example:
+ S<-load $LLVM_DIR/Release/lib/LLVMCSimple.so>.
+ 
+ =item B<-v> or B<--verbose>
+ 
+ Enable verbose mode, i.e. print out all executed commands.
+ 
+ =item B<--check-graph>
+ 
+ Check the compilation for common errors like mismatched output/input language
+ names, multiple default edges and cycles. Because of plugins, these checks can't
+ be performed at compile-time. Exit with code zero if no errors were found, and
+ return the number of found errors otherwise. Hidden option, useful for debugging
+ LLVMC plugins.
+ 
+ =item B<--view-graph>
+ 
+ Show a graphical representation of the compilation graph and exit. Requires that
+ you have I<dot> and I<gv> programs installed. Hidden option, useful for
+ debugging LLVMC plugins.
+ 
+ =item B<--write-graph>
+ 
+ Write a I<compilation-graph.dot> file in the current directory with the
+ compilation graph description in Graphviz format (identical to the file used by
+ the B<--view-graph> option). The B<-o> option can be used to set the output file
+ name. Hidden option, useful for debugging LLVMC plugins.
+ 
+ =item B<--save-temps>
+ 
+ Write temporary files to the current directory and do not delete them on
+ exit. This option can also take an argument: the I<--save-temps=obj> switch will
+ write files into the directory specified with the I<-o> option. The
+ I<--save-temps=cwd> and I<--save-temps> switches are both synonyms for the
+ default behaviour.
+ 
+ =item B<--temp-dir> I<directory>
+ 
+ Store temporary files in the given directory. This directory is deleted on exit
+ unless I<--save-temps> is specified. If I<--save-temps=obj> is also specified,
+ I<--temp-dir> is given the precedence.
+ 
+ =item B<--help>
+ 
+ Print a summary of command-line options and exit.
+ 
+ =item B<--help-hidden>
+ 
+ Print a summary of command-line options and exit. Print help even for
+ options intended for developers.
+ 
+ =item B<--version>
+ 
+ Print version information and exit.
+ 
+ =item B<@>I<file>
+ 
+ Read command-line options from I<file>. The options read are inserted
+ in place of the original @I<file> option. If I<file> does not exist, or
+ cannot be read, then the option will be treated literally, and not
+ removed.
+ 
+ Options in I<file> are separated by whitespace. A whitespace character
+ may be included in an option by surrounding the entire option in
+ either single or double quotes. Any character (including a backslash)
+ may be included by prefixing the character to be included with a
+ backslash. The file may itself contain additional @I<file> options;
+ any such options will be processed recursively.
+ 
+ 
+ =back
+ 
+ 
+ =head2 Control Options
+ 
+ By default, LLVMC is built with some standard configuration libraries
+ that define the following options:
+ 
+ =over
+ 
+ =item B<-clang>
+ 
+ Use Clang instead of llvm-gcc.
+ 
+ =item B<-opt>
+ 
+ Enable optimization passes with B<opt>. To pass options to the B<opt> program
+ use the B<-Wo,> option.
+ 
+ =item B<-I> I<directory>
+ 
+ Add a directory to the header file search path.  This option can be
+ repeated.
+ 
+ =item B<-L> I<directory>
+ 
+ Add I<directory> to the library search path.  This option can be
+ repeated.
+ 
+ =item B<-l>I<name>
+ 
+ Link in the library libI<name>.[bc | a | so].  This library should
+ be a bitcode library.
+ 
+ =item B<-emit-llvm>
+ 
+ Make the output be LLVM bitcode (with B<-c>) or assembly (with B<-S>) instead
+ of native object (or assembly).  If B<-emit-llvm> is given without either B<-c>
+ or B<-S> it has no effect.
+ 
+ =item B<-Wa>
+ 
+ Pass options to assembler.
+ 
+ =item B<-Wl>
+ 
+ Pass options to linker.
+ 
+ =item B<-Wo>
+ 
+ Pass options to opt.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvmc> succeeds, it will exit with code 0.  Otherwise, if an
+ error occurs, it will exit with a non-zero value. If one of the
+ compilation tools returns a non-zero status, pending actions will be
+ discarded and B<llvmc> will return the same result code as the failing
+ compilation tool.
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-gcc|llvmgcc>, L<llvm-g++|llvmgxx>, L<llvm-as|llvm-as>,
+ L<llvm-dis|llvm-dis>, L<llc|llc>, L<llvm-link|llvm-link>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/llvmgcc.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvmgcc.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvmgcc.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,76 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-gcc - LLVM C front-end
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-gcc> [I<options>] I<filename>
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-gcc> command is the LLVM C front end.  It is a modified
+ version of gcc that compiles C/ObjC programs into native objects, LLVM
+ bitcode or LLVM assembly language, depending upon the options.
+ 
+ By default, B<llvm-gcc> compiles to native objects just like GCC does. If the
+ B<-emit-llvm> and B<-c> options are given then it will generate LLVM bitcode files
+ instead. If B<-emit-llvm> and B<-S> are given, then it will generate LLVM
+ assembly.
+ 
+ Being derived from the GNU Compiler Collection, B<llvm-gcc> has many
+ of gcc's features and accepts most of gcc's options.  It handles a
+ number of gcc's extensions to the C programming language.  See the gcc
+ documentation for details.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-o> I<filename>
+ 
+ Specify the output file to be I<filename>.
+ 
+ =item B<-I> I<directory>
+ 
+ Add a directory to the header file search path.  This option can be
+ repeated.
+ 
+ =item B<-L> I<directory>
+ 
+ Add I<directory> to the library search path.  This option can be
+ repeated.
+ 
+ =item B<-l>I<name>
+ 
+ Link in the library libI<name>.[bc | a | so].  This library should
+ be a bitcode library.
+ 
+ =item B<-emit-llvm>
+ 
+ Make the output be LLVM bitcode (with B<-c>) or assembly (with B<-s>) instead
+ of native object (or assembly).  If B<-emit-llvm> is given without either B<-c>
+ or B<-S> it has no effect.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-gcc> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-g++|llvmgxx>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut
+ 


Index: llvm-www/releases/2.6/docs/CommandGuide/llvmgxx.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/llvmgxx.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/llvmgxx.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,85 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ llvm-g++ - LLVM C++ front-end
+ 
+ =head1 SYNOPSIS
+ 
+ B<llvm-g++> [I<options>] I<filename>
+ 
+ =head1 DESCRIPTION
+ 
+ The B<llvm-g++> command is the LLVM C++ front end.  It is a modified
+ version of g++ that compiles C++/ObjC++ programs into native code, 
+ LLVM bitcode or assembly language, depending upon the options.
+ 
+ By default, B<llvm-g++> compiles to native objects just like GCC does. If the
+ B<-emit-llvm> option is given then it will generate LLVM bitcode files instead.
+ If B<-S> (assembly) is also given, then it will generate LLVM assembly. 
+ 
+ Being derived from the GNU Compiler Collection, B<llvm-g++> has many
+ of g++'s features and accepts most of g++'s options.  It handles a
+ number of g++'s extensions to the C++ programming language.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-S>
+ 
+ Do not generate an LLVM bitcode file.  Rather, compile the source
+ file into an LLVM assembly language file.
+ 
+ =item B<-c>
+ 
+ Do not generate a linked executable.  Rather, compile the source
+ file into an LLVM bitcode file.  This bitcode file can then be
+ linked with other bitcode files later on to generate a full LLVM
+ executable.
+ 
+ =item B<-o> I<filename>
+ 
+ Specify the output file to be I<filename>.
+ 
+ =item B<-I> I<directory>
+ 
+ Add a directory to the header file search path.  This option can be
+ repeated.
+ 
+ =item B<-L> I<directory>
+ 
+ Add I<directory> to the library search path.  This option can be
+ repeated.
+ 
+ =item B<-l>I<name>
+ 
+ Link in the library libI<name>.[bc | a | so].  This library should
+ be a bitcode library.
+ 
+ =item B<-emit-llvm>
+ 
+ Make the output be LLVM bitcode (or assembly) instead of native object (or
+ assembly).
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<llvm-g++> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.
+ 
+ =head1 SEE ALSO
+ 
+ L<llvm-gcc|llvmgcc>
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut
+ 


Index: llvm-www/releases/2.6/docs/CommandGuide/manpage.css
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/manpage.css:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/manpage.css	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,256 ----
+ /* Based on http://www.perldoc.com/css/perldoc.css */
+ 
+ @import url("../llvm.css");
+ 
+ body { font-family: Arial,Helvetica; }
+ 
+ blockquote { margin: 10pt;  }
+ 
+ h1, a { color: #336699; }
+ 
+ 
+ /*** Top menu style ****/
+ .mmenuon { 
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #ff6600; font-size: 10pt;
+ }
+ .mmenuoff { 
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #ffffff; font-size: 10pt;
+ }	  
+ .cpyright {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #ffffff; font-size: xx-small;
+ }
+ .cpyrightText {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #ffffff; font-size: xx-small;
+ }
+ .sections { 
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #336699; font-size: 11pt;
+ }	 
+ .dsections { 
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #336699; font-size: 12pt;
+ }	
+ .slink { 
+  font-family: Arial,Helvetica; font-weight: normal; text-decoration: none;
+  color: #000000; font-size: 9pt;
+ }	 
+ 
+ .slink2 { font-family: Arial,Helvetica; text-decoration: none; color: #336699; }	 
+ 
+ .maintitle { 
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #336699; font-size: 18pt;
+ }	 
+ .dblArrow {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #336699; font-size: small;
+ }
+ .menuSec {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #336699; font-size: small;
+ }
+ 
+ .newstext {
+  font-family: Arial,Helvetica; font-size: small;
+ }
+ 
+ .linkmenu {
+  font-family: Arial,Helvetica; color: #000000; font-weight: bold;
+  text-decoration: none;
+ }
+ 
+ P {
+  font-family: Arial,Helvetica;
+ }
+ 
+ PRE {
+     font-size: 10pt;
+ }
+ .quote { 
+  font-family: Times; text-decoration: none;
+  color: #000000; font-size: 9pt; font-style: italic;
+ }	
+ .smstd { font-family: Arial,Helvetica; color: #000000; font-size: x-small; } 
+ .std { font-family: Arial,Helvetica; color: #000000; } 
+ .meerkatTitle { 
+  font-family: sans-serif; font-size: x-small;  color: black;    }
+ 
+ .meerkatDescription { font-family: sans-serif; font-size: 10pt; color: black }
+ .meerkatCategory { 
+  font-family: sans-serif; font-size: 9pt; font-weight: bold; font-style: italic; 
+  color: brown; }
+ .meerkatChannel { 
+  font-family: sans-serif; font-size: 9pt; font-style: italic; color: brown; }
+ .meerkatDate { font-family: sans-serif; font-size: xx-small; color: #336699; }
+ 
+ .tocTitle {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #333333; font-size: 10pt;
+ }
+ 
+ .toc-item {
+  font-family: Arial,Helvetica; font-weight: bold; 
+  color: #336699; font-size: 10pt; text-decoration: underline;
+ }
+ 
+ .perlVersion {
+  font-family: Arial,Helvetica; font-weight: bold; 
+  color: #336699; font-size: 10pt; text-decoration: none;
+ }
+ 
+ .podTitle {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #000000;
+ }
+ 
+ .docTitle {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #000000; font-size: 10pt;
+ }
+ .dotDot {
+  font-family: Arial,Helvetica; font-weight: bold; 
+  color: #000000; font-size: 9pt;
+ }
+ 
+ .docSec {
+  font-family: Arial,Helvetica; font-weight: normal; 
+  color: #333333; font-size: 9pt;
+ }
+ .docVersion {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #336699; font-size: 10pt;
+ }
+ 
+ .docSecs-on {
+  font-family: Arial,Helvetica; font-weight: normal; text-decoration: none;
+  color: #ff0000; font-size: 10pt;
+ }
+ .docSecs-off {
+  font-family: Arial,Helvetica; font-weight: normal; text-decoration: none;
+  color: #333333; font-size: 10pt;
+ }
+ 
+ h2 {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #336699; font-size: medium;
+ }
+ h1 {
+  font-family: Verdana,Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #336699; font-size: large;
+ }
+ 
+ DL {
+  font-family: Arial,Helvetica; font-weight: normal; text-decoration: none;
+  color: #333333; font-size: 10pt;
+ }
+ 
+ UL > LI > A {
+  font-family: Arial,Helvetica; font-weight: bold;
+  color: #336699; font-size: 10pt;
+ }
+ 
+ .moduleInfo {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #333333; font-size: 11pt;
+ }
+ 
+ .moduleInfoSec {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none;
+  color: #336699; font-size: 10pt;
+ }
+ 
+ .moduleInfoVal {
+  font-family: Arial,Helvetica; font-weight: normal; text-decoration: underline;
+  color: #000000; font-size: 10pt;
+ }
+ 
+ .cpanNavTitle {
+  font-family: Arial,Helvetica; font-weight: bold; 
+  color: #ffffff; font-size: 10pt;
+ }
+ .cpanNavLetter {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none; 
+  color: #333333; font-size: 9pt;
+ }
+ .cpanCat {
+  font-family: Arial,Helvetica; font-weight: bold; text-decoration: none; 
+  color: #336699; font-size: 9pt;
+ }
+ 
+ .bttndrkblue-bkgd-top {
+ 	background-color: #225688;
+ 	background-image: url(/global/mvc_objects/images/bttndrkblue_bgtop.gif);
+ }
+ .bttndrkblue-bkgd-left {
+ 	background-color: #225688;
+ 	background-image: url(/global/mvc_objects/images/bttndrkblue_bgleft.gif);
+ }
+ .bttndrkblue-bkgd {
+ 	padding-top: 0px;
+ 	padding-bottom: 0px;
+ 	margin-bottom: 0px;
+ 	margin-top: 0px;
+ 	background-repeat: no-repeat;
+ 	background-color: #225688;
+ 	background-image: url(/global/mvc_objects/images/bttndrkblue_bgmiddle.gif);
+ 	vertical-align: top;
+ }
+ .bttndrkblue-bkgd-right {
+ 	background-color: #225688;
+ 	background-image: url(/global/mvc_objects/images/bttndrkblue_bgright.gif);
+ }
+ .bttndrkblue-bkgd-bottom {
+ 	background-color: #225688;
+ 	background-image: url(/global/mvc_objects/images/bttndrkblue_bgbottom.gif);
+ }
+ .bttndrkblue-text a {
+ 	color: #ffffff;
+ 	text-decoration: none;
+ }
+ a.bttndrkblue-text:hover {
+ 	color: #ffDD3C;
+ 	text-decoration: none;
+ }
+ .bg-ltblue {
+ 	background-color: #f0f5fa;
+ } 
+ 
+ .border-left-b {
+ 	background: #f0f5fa url(/i/corner-leftline.gif) repeat-y;
+ } 
+ 
+ .border-right-b {
+ 	background: #f0f5fa url(/i/corner-rightline.gif) repeat-y;
+ } 
+ 
+ .border-top-b {
+ 	background: #f0f5fa url(/i/corner-topline.gif) repeat-x;
+ } 
+ 
+ .border-bottom-b {
+ 	background: #f0f5fa url(/i/corner-botline.gif) repeat-x;
+ } 
+ 
+ .border-right-w {
+ 	background: #ffffff url(/i/corner-rightline.gif) repeat-y;
+ } 
+ 
+ .border-top-w {
+ 	background: #ffffff url(/i/corner-topline.gif) repeat-x;
+ } 
+ 
+ .border-bottom-w {
+ 	background: #ffffff url(/i/corner-botline.gif) repeat-x;
+ } 
+ 
+ .bg-white {
+ 	background-color: #ffffff;
+ } 
+ 
+ .border-left-w {
+ 	background: #ffffff url(/i/corner-leftline.gif) repeat-y;
+ } 


Index: llvm-www/releases/2.6/docs/CommandGuide/opt.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/opt.pod:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/opt.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,138 ----
+ =pod
+ 
+ =head1 NAME
+ 
+ opt - LLVM optimizer
+ 
+ =head1 SYNOPSIS
+ 
+ B<opt> [I<options>] [I<filename>]
+ 
+ =head1 DESCRIPTION
+ 
+ The B<opt> command is the modular LLVM optimizer and analyzer.  It takes LLVM 
+ bitcode as input, runs the specified optimizations or analyses on it, and then
+ outputs the optimized LLVM bitcode or the analysis results.  The function of 
+ B<opt> depends on whether the B<-analyze> option is given. 
+ 
+ When B<-analyze> is specified, B<opt> performs various analyses of LLVM 
+ bitcode.  It will usually print the results on standard output, but in a few 
+ cases, it will print output to standard error or generate a file with the 
+ analysis output, which is usually done when the output is meant for another 
+ program.  
+ 
+ While B<-analyze> is I<not> given, B<opt> attempts to produce an optimized 
+ bitcode file.  The optimizations available via B<opt> depend upon what 
+ libraries were linked into it as well as any additional libraries that have 
+ been loaded with the B<-load> option.  Use the B<-help> option to determine 
+ what optimizations you can use.
+ 
+ If I<filename> is omitted from the command line or is I<->, B<opt> reads its
+ input from standard input. The input must be an LLVM bitcode file.
+ 
+ If an output filename is not specified with the B<-o> option, B<opt>
+ writes its output to the standard output.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<-f>
+ 
+ Force overwrite.  Normally, B<opt> will refuse to overwrite an
+ output file that already exists.  With this option, B<opt> will
+ overwrite the output file and replace it with new bitcode.
+ 
+ =item B<-help>
+ 
+ Print a summary of command line options. 
+ 
+ =item B<-o> I<filename>
+ 
+ Specify the output filename.
+ 
+ =item B<-{passname}>
+ 
+ B<opt> provides the ability to run any of LLVM's optimization or analysis passes
+ in any order. The B<-help> option lists all the passes available. The order in
+ which the options occur on the command line are the order in which they are
+ executed (within pass constraints). 
+ 
+ =item B<-std-compile-opts>
+ 
+ This is short hand for a standard list of I<compile time optimization> passes.
+ This is typically used to optimize the output from the llvm-gcc front end. It
+ might be useful for other front end compilers as well. To discover the full set
+ of options available, use the following command:
+ 
+    llvm-as < /dev/null | opt -std-compile-opts -disable-output -debug-pass=Arguments
+ 
+ =item B<-disable-inlining>
+ 
+ This option is only meaningful when B<-std-compile-opts> is given. It simply
+ removes the inlining pass from the standard list.
+ 
+ =item B<-disable-opt>
+ 
+ This option is only meaningful when B<-std-compile-opts> is given. It disables
+ most, but not all, of the B<-std-compile-opts>. The ones that remain are
+ B<-verify>, B<-lower-setjmp>, and B<-funcresolve>.
+ 
+ =item B<-strip-debug>
+ 
+ This option causes opt to strip debug information from the module before 
+ applying other optimizations. It is essentially the same as B<-strip> but it
+ ensures that stripping of debug information is done first.
+ 
+ =item B<-verify-each>
+ 
+ This option causes opt to add a verify pass after every pass otherwise specified
+ on the command line (including B<-verify>).  This is useful for cases where it 
+ is suspected that a pass is creating an invalid module but it is not clear which
+ pass is doing it. The combination of B<-std-compile-opts> and B<-verify-each>
+ can quickly track down this kind of problem.
+ 
+ =item B<-profile-info-file> I<filename>
+ 
+ Specify the name of the file loaded by the -profile-loader option.
+ 
+ =item B<-stats>
+ 
+ Print statistics.
+ 
+ =item B<-time-passes>
+ 
+ Record the amount of time needed for each pass and print it to standard
+ error.
+ 
+ =item B<-debug>
+ 
+ If this is a debug build, this option will enable debug printouts
+ from passes which use the I<DEBUG()> macro.  See the B<LLVM Programmer's
+ Manual>, section I<#DEBUG> for more information.
+ 
+ =item B<-load>=I<plugin>
+ 
+ Load the dynamic object I<plugin>.  This object should register new optimization
+ or analysis passes. Once loaded, the object will add new command line options to
+ enable various optimizations or analyses.  To see the new complete list of 
+ optimizations, use the B<-help> and B<-load> options together. For example:
+ 
+    opt -load=plugin.so -help
+ 
+ =item B<-p>
+ 
+ Print module after each transformation.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<opt> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.
+ 
+ =head1 AUTHORS
+ 
+ Maintained by the LLVM Team (L<http://llvm.org>).
+ 
+ =cut


Index: llvm-www/releases/2.6/docs/CommandGuide/pod2htmd.tmp
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/pod2htmd.tmp:1.1
*** /dev/null	Fri Oct 23 13:57:45 2009
--- llvm-www/releases/2.6/docs/CommandGuide/pod2htmd.tmp	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,28 ----
+ .
+ .
+ llvm-bcanalyzer ./llvm-bcanalyzer.pod:
+ lli ./lli.pod:
+ llvmgxx ./llvmgxx.pod:
+ llvm-nm ./llvm-nm.pod:
+ llvm-db ./llvm-db.pod:
+ llvmgcc ./llvmgcc.pod:
+ opt ./opt.pod:
+ html ./html:
+ llvm-link ./llvm-link.pod:
+ llvm-ar ./llvm-ar.pod:
+ tblgen ./tblgen.pod:
+ llc ./llc.pod:
+ llvm-config ./llvm-config.pod:
+ llvmc ./llvmc.pod:
+ llvm-extract ./llvm-extract.pod:
+ llvm-as ./llvm-as.pod:
+ llvm-prof ./llvm-prof.pod:
+ bugpoint ./bugpoint.pod:./html/bugpoint.pod:
+ man ./man:
+ index ./index.pod:
+ llvm-ranlib ./llvm-ranlib.pod:
+ FileCheck ./FileCheck.pod:
+ man1 ./man/man1:
+ ps ./ps:
+ llvm-dis ./llvm-dis.pod:
+ llvm-ld ./llvm-ld.pod:


Index: llvm-www/releases/2.6/docs/CommandGuide/pod2htmi.tmp
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/pod2htmi.tmp:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/CommandGuide/pod2htmi.tmp	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,2 ----
+ .
+ .


Index: llvm-www/releases/2.6/docs/CommandGuide/tblgen.pod
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/tblgen.pod:1.1
*** /dev/null	Fri Oct 23 13:57:46 2009
--- llvm-www/releases/2.6/docs/CommandGuide/tblgen.pod	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,115 ----
+ 
+ =pod
+ 
+ =head1 NAME
+ 
+ tblgen - Target Description To C++ Code Generator
+ 
+ =head1 SYNOPSIS
+ 
+ B<tblgen> [I<options>] [I<filename>]
+ 
+ =head1 DESCRIPTION
+ 
+ B<tblgen> translates from target description (.td) files into C++ code that can
+ be included in the definition of an LLVM target library. Most users of LLVM will
+ not need to use this program. It is only for assisting with writing an LLVM
+ target backend.
+ 
+ The input and output of B<tblgen> is beyond the scope of this short
+ introduction. Please see the I<CodeGeneration> page in the LLVM documentation.
+ 
+ The F<filename> argument specifies the name of a Target Description (.td) file
+ to read as input.
+ 
+ =head1 OPTIONS
+ 
+ =over
+ 
+ =item B<--help>
+ 
+ Print a summary of command line options.
+ 
+ =item B<-o> F<filename>
+ 
+ Specify the output file name.  If F<filename> is C<->, then B<tblgen>
+ sends its output to standard output.
+ 
+ =item B<-I> F<directory>
+ 
+ Specify where to find other target description files for inclusion. The
+ F<directory> value should be a full or partial path to a directory that contains
+ target description files.
+ 
+ =item B<-asmwriternum> F<N>
+ 
+ Make -gen-asm-writer emit assembly writer number F<N>.
+ 
+ =item B<-class> F<class Name>
+ 
+ Print the enumeration list for this class.
+ 
+ =item B<-print-records>
+ 
+ Print all records to standard output (default).
+ 
+ =item B<-print-enums>
+ 
+ Print enumeration values for a class
+ 
+ =item B<-gen-emitter>
+ 
+ Generate machine code emitter.
+ 
+ =item B<-gen-register-enums>
+ 
+ Generate the enumeration values for all registers.
+ 
+ =item B<-gen-register-desc>
+ 
+ Generate a register info description for each register.
+ 
+ =item B<-gen-register-desc-header>
+ 
+ Generate a register info description header for each register.
+ 
+ =item B<-gen-instr-enums>
+ 
+ Generate enumeration values for instructions.
+ 
+ =item B<-gen-instr-desc>
+ 
+ Generate instruction descriptions.
+ 
+ =item B<-gen-asm-writer>
+ 
+ Generate the assembly writer.
+ 
+ =item B<-gen-dag-isel>
+ 
+ Generate a DAG (Directed Acycle Graph) instruction selector.
+ 
+ =item B<-gen-subtarget>
+ 
+ Generate subtarget enumerations.
+ 
+ =item B<-gen-intrinsic>
+ 
+ Generate intrinsic information.
+ 
+ =item B<-version>
+ 
+ Show the version number of this program.
+ 
+ =back
+ 
+ =head1 EXIT STATUS
+ 
+ If B<tblgen> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.
+ 
+ =head1 AUTHORS
+ 
+ Maintained by The LLVM Team (L<http://llvm.org>).
+ 
+ =cut





From tonic at nondot.org  Fri Oct 23 13:57:46 2009
From: tonic at nondot.org (Tanya Lattner)
Date: Fri, 23 Oct 2009 13:57:46 -0500
Subject: [llvm-commits] CVS:
	llvm-www/releases/2.6/docs/CommandGuide/html/FileCheck.html
	bugpoint.html llc.html lli.html llvm-ar.html llvm-as.html
	llvm-bcanalyzer.html llvm-config.html llvm-db.html
	llvm-dis.html llvm-extract.html llvm-ld.html llvm-link.html
	llvm-nm.html llvm-prof.html llvm-ranlib.html llvmc.html
	llvmgcc.html llvmgxx.html manpage.css opt.html tblgen.html
Message-ID: <200910231857.n9NIvkO1004776@zion.cs.uiuc.edu>



Changes in directory llvm-www/releases/2.6/docs/CommandGuide/html:

FileCheck.html added (r1.1)
bugpoint.html added (r1.1)
llc.html added (r1.1)
lli.html added (r1.1)
llvm-ar.html added (r1.1)
llvm-as.html added (r1.1)
llvm-bcanalyzer.html added (r1.1)
llvm-config.html added (r1.1)
llvm-db.html added (r1.1)
llvm-dis.html added (r1.1)
llvm-extract.html added (r1.1)
llvm-ld.html added (r1.1)
llvm-link.html added (r1.1)
llvm-nm.html added (r1.1)
llvm-prof.html added (r1.1)
llvm-ranlib.html added (r1.1)
llvmc.html added (r1.1)
llvmgcc.html added (r1.1)
llvmgxx.html added (r1.1)
manpage.css added (r1.1)
opt.html added (r1.1)
tblgen.html added (r1.1)
---
Log message:

Add 2.6



---
Diffs of the changes:  (+4383 -0)

 FileCheck.html       |  101 ++++++++++
 bugpoint.html        |  246 ++++++++++++++++++++++++
 llc.html             |  270 +++++++++++++++++++++++++++
 lli.html             |  310 +++++++++++++++++++++++++++++++
 llvm-ar.html         |  502 +++++++++++++++++++++++++++++++++++++++++++++++++++
 llvm-as.html         |  108 ++++++++++
 llvm-bcanalyzer.html |  451 +++++++++++++++++++++++++++++++++++++++++++++
 llvm-config.html     |  201 ++++++++++++++++++++
 llvm-db.html         |   43 ++++
 llvm-dis.html        |   97 +++++++++
 llvm-extract.html    |  102 ++++++++++
 llvm-ld.html         |  328 +++++++++++++++++++++++++++++++++
 llvm-link.html       |  118 +++++++++++
 llvm-nm.html         |  181 ++++++++++++++++++
 llvm-prof.html       |   93 +++++++++
 llvm-ranlib.html     |   91 +++++++++
 llvmc.html           |  258 ++++++++++++++++++++++++++
 llvmgcc.html         |  118 +++++++++++
 llvmgxx.html         |  131 +++++++++++++
 manpage.css          |  256 ++++++++++++++++++++++++++
 opt.html             |  199 ++++++++++++++++++++
 tblgen.html          |  179 ++++++++++++++++++
 22 files changed, 4383 insertions(+)


Index: llvm-www/releases/2.6/docs/CommandGuide/html/FileCheck.html
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/html/FileCheck.html:1.1
*** /dev/null	Fri Oct 23 13:56:32 2009
--- llvm-www/releases/2.6/docs/CommandGuide/html/FileCheck.html	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,101 ----
+ <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
+ <html xmlns="http://www.w3.org/1999/xhtml">
+ <head>
+ <title>FileCheck</title>
+ <link rel="stylesheet" href="manpage.css" type="text/css" />
+ <link rev="made" href="mailto:root at localhost" />
+ </head>
+ 
+ <body>
+ 
+ <p><a name="__index__"></a></p>
+ <!-- INDEX BEGIN -->
+ <!--
+ 
+ <ul>
+ 
+ 	<li><a href="#name">NAME</a></li>
+ 	<li><a href="#synopsis">SYNOPSIS</a></li>
+ 	<li><a href="#description">DESCRIPTION</a></li>
+ 	<li><a href="#options">OPTIONS</a></li>
+ 	<li><a href="#exit_status">EXIT STATUS</a></li>
+ 	<li><a href="#authors">AUTHORS</a></li>
+ </ul>
+ -->
+ <!-- INDEX END -->
+ 
+ <p>
+ </p>
+ <hr />
+ <h1><a name="name">NAME</a></h1>
+ <p>FileCheck - Flexible pattern matching file verifier</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="synopsis">SYNOPSIS</a></h1>
+ <p><strong>FileCheck</strong> <em>match-filename</em> [<em>--check-prefix=XXX</em>] [<em>--strict-whitespace</em>]</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="description">DESCRIPTION</a></h1>
+ <p><strong>FileCheck</strong> reads two files (one from standard input, and one specified on the
+ command line) and uses one to verify the other.  This behavior is particularly
+ useful for the testsuite, which wants to verify that the output of some tool
+ (e.g. llc) contains the expected information (for example, a movsd from esp or
+ whatever is interesting).  This is similar to using grep, but it is optimized
+ for matching multiple different inputs in one file in a specific order.</p>
+ <p>The <em>match-filename</em> file specifies the file that contains the patterns to
+ match.  The file to verify is always read from standard input.</p>
+ <p>The input and output of <strong>FileCheck</strong> is beyond the scope of this short
+ introduction. Please see the <em>TestingGuide</em> page in the LLVM documentation.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="options">OPTIONS</a></h1>
+ <dl>
+ <dt><strong><a name="item__2d_2dhelp"><strong>--help</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print a summary of command line options.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dcheck_2dprefix_prefix"><strong>--check-prefix</strong> <em>prefix</em></a></strong><br />
+ </dt>
+ <dd>
+ FileCheck searches the contents of <em>match-filename</em> for patterns to match.  By
+ default, these patterns are prefixed with ``CHECK:''.  If you'd like to use a
+ different prefix (e.g. because the same input file is checking multiple
+ different tool or options), the <strong>--check-prefix</strong> argument allows you to specify
+ a specific prefix to match.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dstrict_2dwhitespace"><strong>--strict-whitespace</strong></a></strong><br />
+ </dt>
+ <dd>
+ By default, FileCheck canonicalizes input horizontal whitespace (spaces and
+ tabs) which causes it to ignore these differences (a space will match a tab).
+ The --strict-whitespace argument disables this behavior.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dversion"><strong>-version</strong></a></strong><br />
+ </dt>
+ <dd>
+ Show the version number of this program.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="exit_status">EXIT STATUS</a></h1>
+ <p>If <strong>FileCheck</strong> verifies that the file matches the expected contents, it exits
+ with 0.  Otherwise, if not, or if an error occurs, it will exit with a non-zero
+ value.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="authors">AUTHORS</a></h1>
+ <p>Maintained by The LLVM Team (<a href="http://llvm.org">http://llvm.org</a>).</p>
+ 
+ </body>
+ 
+ </html>


Index: llvm-www/releases/2.6/docs/CommandGuide/html/bugpoint.html
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/html/bugpoint.html:1.1
*** /dev/null	Fri Oct 23 13:57:42 2009
--- llvm-www/releases/2.6/docs/CommandGuide/html/bugpoint.html	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,246 ----
+ <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
+ <html xmlns="http://www.w3.org/1999/xhtml">
+ <head>
+ <title>bugpoint</title>
+ <link rel="stylesheet" href="manpage.css" type="text/css" />
+ <link rev="made" href="mailto:root at localhost" />
+ </head>
+ 
+ <body>
+ 
+ <p><a name="__index__"></a></p>
+ <!-- INDEX BEGIN -->
+ <!--
+ 
+ <ul>
+ 
+ 	<li><a href="#name">NAME</a></li>
+ 	<li><a href="#synopsis">SYNOPSIS</a></li>
+ 	<li><a href="#description">DESCRIPTION</a></li>
+ 	<li><a href="#options">OPTIONS</a></li>
+ 	<li><a href="#exit_status">EXIT STATUS</a></li>
+ 	<li><a href="#see_also">SEE ALSO</a></li>
+ 	<li><a href="#author">AUTHOR</a></li>
+ </ul>
+ -->
+ <!-- INDEX END -->
+ 
+ <p>
+ </p>
+ <hr />
+ <h1><a name="name">NAME</a></h1>
+ <p>bugpoint - automatic test case reduction tool</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="synopsis">SYNOPSIS</a></h1>
+ <p><strong>bugpoint</strong> [<em>options</em>] [<em>input LLVM ll/bc files</em>] [<em>LLVM passes</em>] <strong>--args</strong>
+ <em>program arguments</em></p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="description">DESCRIPTION</a></h1>
+ <p><strong>bugpoint</strong> narrows down the source of problems in LLVM tools and passes.  It
+ can be used to debug three types of failures: optimizer crashes, miscompilations
+ by optimizers, or bad native code generation (including problems in the static
+ and JIT compilers).  It aims to reduce large test cases to small, useful ones.
+ For more information on the design and inner workings of <strong>bugpoint</strong>, as well as
+ advice for using bugpoint, see <em>llvm/docs/Bugpoint.html</em> in the LLVM
+ distribution.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="options">OPTIONS</a></h1>
+ <dl>
+ <dt><strong><a name="item__2d_2dadditional_2dso_library"><strong>--additional-so</strong> <em>library</em></a></strong><br />
+ </dt>
+ <dd>
+ Load the dynamic shared object <em>library</em> into the test program whenever it is
+ run.  This is useful if you are debugging programs which depend on non-LLVM
+ libraries (such as the X or curses libraries) to run.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dappend_2dexit_2dcode_3d_7btrue_2cfalse_7d"><strong>--append-exit-code</strong>=<em>{true,false}</em></a></strong><br />
+ </dt>
+ <dd>
+ Append the test programs exit code to the output file so that a change in exit
+ code is considered a test failure. Defaults to false.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dargs_program_args"><strong>--args</strong> <em>program args</em></a></strong><br />
+ </dt>
+ <dd>
+ Pass all arguments specified after -args to the test program whenever it runs.
+ Note that if any of the <em>program args</em> start with a '-', you should use:
+ </dd>
+ <dd>
+ <pre>
+     bugpoint [bugpoint args] --args -- [program args]</pre>
+ </dd>
+ <dd>
+ <p>The ``--'' right after the <strong>--args</strong> option tells <strong>bugpoint</strong> to consider any
+ options starting with <code>-</code> to be part of the <strong>--args</strong> option, not as options to
+ <strong>bugpoint</strong> itself.</p>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dtool_2dargs_tool_args"><strong>--tool-args</strong> <em>tool args</em></a></strong><br />
+ </dt>
+ <dd>
+ Pass all arguments specified after --tool-args to the LLVM tool under test
+ (<strong>llc</strong>, <strong>lli</strong>, etc.) whenever it runs.  You should use this option in the
+ following way:
+ </dd>
+ <dd>
+ <pre>
+     bugpoint [bugpoint args] --tool-args -- [tool args]</pre>
+ </dd>
+ <dd>
+ <p>The ``--'' right after the <strong>--tool-args</strong> option tells <strong>bugpoint</strong> to consider any
+ options starting with <code>-</code> to be part of the <strong>--tool-args</strong> option, not as
+ options to <strong>bugpoint</strong> itself. (See <strong>--args</strong>, above.)</p>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dsafe_2dtool_2dargs_tool_args"><strong>--safe-tool-args</strong> <em>tool args</em></a></strong><br />
+ </dt>
+ <dd>
+ Pass all arguments specified after <strong>--safe-tool-args</strong> to the ``safe'' execution
+ tool.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dgcc_2dtool_2dargs_gcc_tool_args"><strong>--gcc-tool-args</strong> <em>gcc tool args</em></a></strong><br />
+ </dt>
+ <dd>
+ Pass all arguments specified after <strong>--gcc-tool-args</strong> to the invocation of
+ <strong>gcc</strong>.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2ddisable_2d_7bdce_2csimplifycfg_7d"><strong>--disable-{dce,simplifycfg}</strong></a></strong><br />
+ </dt>
+ <dd>
+ Do not run the specified passes to clean up and reduce the size of the test
+ program. By default, <strong>bugpoint</strong> uses these passes internally when attempting to
+ reduce test programs.  If you're trying to find a bug in one of these passes,
+ <strong>bugpoint</strong> may crash.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2denable_2dvalgrind"><strong>--enable-valgrind</strong></a></strong><br />
+ </dt>
+ <dd>
+ Use valgrind to find faults in the optimization phase. This will allow
+ bugpoint to find otherwise asymptomatic problems caused by memory
+ mis-management.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dfind_2dbugs"><strong>-find-bugs</strong></a></strong><br />
+ </dt>
+ <dd>
+ Continually randomize the specified passes and run them on the test program
+ until a bug is found or the user kills <strong>bugpoint</strong>.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dhelp"><strong>--help</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print a summary of command line options.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dinput_filename"><strong>--input</strong> <em>filename</em></a></strong><br />
+ </dt>
+ <dd>
+ Open <em>filename</em> and redirect the standard input of the test program, whenever
+ it runs, to come from that file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dload_plugin"><strong>--load</strong> <em>plugin</em></a></strong><br />
+ </dt>
+ <dd>
+ Load the dynamic object <em>plugin</em> into <strong>bugpoint</strong> itself.  This object should
+ register new optimization passes.  Once loaded, the object will add new command
+ line options to enable various optimizations.  To see the new complete list of
+ optimizations, use the <strong>--help</strong> and <strong>--load</strong> options together; for example:
+ </dd>
+ <dd>
+ <pre>
+     bugpoint --load myNewPass.so --help</pre>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dmlimit_megabytes"><strong>--mlimit</strong> <em>megabytes</em></a></strong><br />
+ </dt>
+ <dd>
+ Specifies an upper limit on memory usage of the optimization and codegen. Set
+ to zero to disable the limit.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2doutput_filename"><strong>--output</strong> <em>filename</em></a></strong><br />
+ </dt>
+ <dd>
+ Whenever the test program produces output on its standard output stream, it
+ should match the contents of <em>filename</em> (the ``reference output''). If you
+ do not use this option, <strong>bugpoint</strong> will attempt to generate a reference output
+ by compiling the program with the ``safe'' backend and running it.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dprofile_2dinfo_2dfile_filename"><strong>--profile-info-file</strong> <em>filename</em></a></strong><br />
+ </dt>
+ <dd>
+ Profile file loaded by <strong>--profile-loader</strong>.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2drun_2d_7bint_2cjit_2cllc_2ccbe_2ccustom_7d"><strong>--run-{int,jit,llc,cbe,custom}</strong></a></strong><br />
+ </dt>
+ <dd>
+ Whenever the test program is compiled, <strong>bugpoint</strong> should generate code for it
+ using the specified code generator.  These options allow you to choose the
+ interpreter, the JIT compiler, the static native code compiler, the C
+ backend, or a custom command (see <strong>--exec-command</strong>) respectively.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dsafe_2d_7bllc_2ccbe_2ccustom_7d"><strong>--safe-{llc,cbe,custom}</strong></a></strong><br />
+ </dt>
+ <dd>
+ When debugging a code generator, <strong>bugpoint</strong> should use the specified code
+ generator as the ``safe'' code generator. This is a known-good code generator
+ used to generate the ``reference output'' if it has not been provided, and to
+ compile portions of the program that as they are excluded from the testcase.
+ These options allow you to choose the
+ static native code compiler, the C backend, or a custom command,
+ (see <strong>--exec-command</strong>) respectively. The interpreter and the JIT backends
+ cannot currently be used as the ``safe'' backends.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dexec_2dcommand_command"><strong>--exec-command</strong> <em>command</em></a></strong><br />
+ </dt>
+ <dd>
+ This option defines the command to use with the <strong>--run-custom</strong> and
+ <strong>--safe-custom</strong> options to execute the bitcode testcase. This can
+ be useful for cross-compilation.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dsafe_2dpath_path"><strong>--safe-path</strong> <em>path</em></a></strong><br />
+ </dt>
+ <dd>
+ This option defines the path to the command to execute with the
+ <strong>--safe-{int,jit,llc,cbe,custom}</strong>
+ option.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="exit_status">EXIT STATUS</a></h1>
+ <p>If <strong>bugpoint</strong> succeeds in finding a problem, it will exit with 0.  Otherwise,
+ if an error occurs, it will exit with a non-zero value.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="see_also">SEE ALSO</a></h1>
+ <p><a href="././opt.html">opt</a></p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="author">AUTHOR</a></h1>
+ <p>Maintained by the LLVM Team (<a href="http://llvm.org">http://llvm.org</a>).</p>
+ 
+ </body>
+ 
+ </html>


Index: llvm-www/releases/2.6/docs/CommandGuide/html/llc.html
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/html/llc.html:1.1
*** /dev/null	Fri Oct 23 13:57:43 2009
--- llvm-www/releases/2.6/docs/CommandGuide/html/llc.html	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,270 ----
+ <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
+ <html xmlns="http://www.w3.org/1999/xhtml">
+ <head>
+ <title>llc</title>
+ <link rel="stylesheet" href="manpage.css" type="text/css" />
+ <link rev="made" href="mailto:root at localhost" />
+ </head>
+ 
+ <body>
+ 
+ <p><a name="__index__"></a></p>
+ <!-- INDEX BEGIN -->
+ <!--
+ 
+ <ul>
+ 
+ 	<li><a href="#name">NAME</a></li>
+ 	<li><a href="#synopsis">SYNOPSIS</a></li>
+ 	<li><a href="#description">DESCRIPTION</a></li>
+ 	<li><a href="#options">OPTIONS</a></li>
+ 	<ul>
+ 
+ 		<li><a href="#enduser_options">End-user Options</a></li>
+ 		<li><a href="#tuning_configuration_options">Tuning/Configuration Options</a></li>
+ 		<li><a href="#intel_ia32specific_options">Intel IA-32-specific Options</a></li>
+ 	</ul>
+ 
+ 	<li><a href="#exit_status">EXIT STATUS</a></li>
+ 	<li><a href="#see_also">SEE ALSO</a></li>
+ 	<li><a href="#authors">AUTHORS</a></li>
+ </ul>
+ -->
+ <!-- INDEX END -->
+ 
+ <p>
+ </p>
+ <hr />
+ <h1><a name="name">NAME</a></h1>
+ <p>llc - LLVM static compiler</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="synopsis">SYNOPSIS</a></h1>
+ <p><strong>llc</strong> [<em>options</em>] [<em>filename</em>]</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="description">DESCRIPTION</a></h1>
+ <p>The <strong>llc</strong> command compiles LLVM bitcode into assembly language for a
+ specified architecture.  The assembly language output can then be passed through
+ a native assembler and linker to generate a native executable.</p>
+ <p>The choice of architecture for the output assembly code is automatically
+ determined from the input bitcode file, unless the <strong>-march</strong> option is used to
+ override the default.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="options">OPTIONS</a></h1>
+ <p>If <em>filename</em> is - or omitted, <strong>llc</strong> reads LLVM bitcode from standard input.
+ Otherwise, it will read LLVM bitcode from <em>filename</em>.</p>
+ <p>If the <strong>-o</strong> option is omitted, then <strong>llc</strong> will send its output to standard
+ output if the input is from standard input.  If the <strong>-o</strong> option specifies -,
+ then the output will also be sent to standard output.</p>
+ <p>If no <strong>-o</strong> option is specified and an input file other than - is specified,
+ then <strong>llc</strong> creates the output filename by taking the input filename,
+ removing any existing <em>.bc</em> extension, and adding a <em>.s</em> suffix.</p>
+ <p>Other <strong>llc</strong> options are as follows:</p>
+ <p>
+ </p>
+ <h2><a name="enduser_options">End-user Options</a></h2>
+ <dl>
+ <dt><strong><a name="item__2d_2dhelp"><strong>--help</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print a summary of command line options.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2do_3duint"><strong>-O</strong>=<em>uint</em></a></strong><br />
+ </dt>
+ <dd>
+ Generate code at different optimization levels. These correspond to the <em>-O0</em>,
+ <em>-O1</em>, <em>-O2</em>, <em>-O3</em>, and <em>-O4</em> optimization levels used by <strong>llvm-gcc</strong> and
+ <strong>clang</strong>.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2df"><strong>-f</strong></a></strong><br />
+ </dt>
+ <dd>
+ Overwrite output files. By default, <strong>llc</strong> will refuse to overwrite
+ an output file which already exists.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dmtriple_3dtarget_triple"><strong>-mtriple</strong>=<em>target triple</em></a></strong><br />
+ </dt>
+ <dd>
+ Override the target triple specified in the input bitcode file with the 
+ specified string.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dmarch_3darch"><strong>-march</strong>=<em>arch</em></a></strong><br />
+ </dt>
+ <dd>
+ Specify the architecture for which to generate assembly, overriding the target
+ encoded in the bitcode file.  See the output of <strong>llc --help</strong> for a list of
+ valid architectures.  By default this is inferred from the target triple or
+ autodetected to the current architecture.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dmcpu_3dcpuname"><strong>-mcpu</strong>=<em>cpuname</em></a></strong><br />
+ </dt>
+ <dd>
+ Specify a specific chip in the current architecture to generate code for.
+ By default this is inferred from the target triple and autodetected to 
+ the current architecture.  For a list of available CPUs, use:
+ <strong>llvm-as &lt; /dev/null | llc -march=xyz -mcpu=help</strong>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dmattr_3da1_2c_2ba2_2c_2da3_2c_2e_2e_2e"><strong>-mattr</strong>=<em>a1,+a2,-a3,...</em></a></strong><br />
+ </dt>
+ <dd>
+ Override or control specific attributes of the target, such as whether SIMD
+ operations are enabled or not.  The default set of attributes is set by the
+ current CPU.  For a list of available attributes, use:
+ <strong>llvm-as &lt; /dev/null | llc -march=xyz -mattr=help</strong>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2ddisable_2dfp_2delim"><strong>--disable-fp-elim</strong></a></strong><br />
+ </dt>
+ <dd>
+ Disable frame pointer elimination optimization.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2ddisable_2dexcess_2dfp_2dprecision"><strong>--disable-excess-fp-precision</strong></a></strong><br />
+ </dt>
+ <dd>
+ Disable optimizations that may produce excess precision for floating point.
+ Note that this option can dramatically slow down code on some systems
+ (e.g. X86).
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2denable_2dunsafe_2dfp_2dmath"><strong>--enable-unsafe-fp-math</strong></a></strong><br />
+ </dt>
+ <dd>
+ Enable optimizations that make unsafe assumptions about IEEE math (e.g. that
+ addition is associative) or may not work for all input ranges.  These
+ optimizations allow the code generator to make use of some instructions which
+ would otherwise not be usable (such as fsin on X86).
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2denable_2dcorrect_2deh_2dsupport"><strong>--enable-correct-eh-support</strong></a></strong><br />
+ </dt>
+ <dd>
+ Instruct the <strong>lowerinvoke</strong> pass to insert code for correct exception handling
+ support.  This is expensive and is by default omitted for efficiency.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dstats"><strong>--stats</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print statistics recorded by code-generation passes.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dtime_2dpasses"><strong>--time-passes</strong></a></strong><br />
+ </dt>
+ <dd>
+ Record the amount of time needed for each pass and print a report to standard
+ error.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dload_3ddso_path"><strong>--load</strong>=<em>dso_path</em></a></strong><br />
+ </dt>
+ <dd>
+ Dynamically load <em>dso_path</em> (a path to a dynamically shared object) that
+ implements an LLVM target. This will permit the target name to be used with the
+ <strong>-march</strong> option so that code can be generated for that target.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <h2><a name="tuning_configuration_options">Tuning/Configuration Options</a></h2>
+ <dl>
+ <dt><strong><a name="item__2d_2dprint_2dmachineinstrs"><strong>--print-machineinstrs</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print generated machine code between compilation phases (useful for debugging).
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dregalloc_3dallocator"><strong>--regalloc</strong>=<em>allocator</em></a></strong><br />
+ </dt>
+ <dd>
+ Specify the register allocator to use. The default <em>allocator</em> is <em>local</em>.
+ Valid register allocators are:
+ </dd>
+ <dl>
+ <dt><strong><a name="item_simple"><em>simple</em></a></strong><br />
+ </dt>
+ <dd>
+ Very simple ``always spill'' register allocator
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_local"><em>local</em></a></strong><br />
+ </dt>
+ <dd>
+ Local register allocator
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_linearscan"><em>linearscan</em></a></strong><br />
+ </dt>
+ <dd>
+ Linear scan global register allocator
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_iterativescan"><em>iterativescan</em></a></strong><br />
+ </dt>
+ <dd>
+ Iterative scan global register allocator
+ </dd>
+ <p></p></dl>
+ <dt><strong><a name="item__2d_2dspiller_3dspiller"><strong>--spiller</strong>=<em>spiller</em></a></strong><br />
+ </dt>
+ <dd>
+ Specify the spiller to use for register allocators that support it.  Currently
+ this option is used only by the linear scan register allocator. The default
+ <em>spiller</em> is <em>local</em>.  Valid spillers are:
+ </dd>
+ <dl>
+ <dt><strong><em>simple</em></strong><br />
+ </dt>
+ <dd>
+ Simple spiller
+ </dd>
+ <p></p>
+ <dt><strong><em>local</em></strong><br />
+ </dt>
+ <dd>
+ Local spiller
+ </dd>
+ <p></p></dl>
+ </dl>
+ <p>
+ </p>
+ <h2><a name="intel_ia32specific_options">Intel IA-32-specific Options</a></h2>
+ <dl>
+ <dt><strong><a name="item__2d_2dx86_2dasm_2dsyntax_3datt_7cintel"><strong>--x86-asm-syntax=att|intel</strong></a></strong><br />
+ </dt>
+ <dd>
+ Specify whether to emit assembly code in AT&amp;T syntax (the default) or intel
+ syntax.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="exit_status">EXIT STATUS</a></h1>
+ <p>If <strong>llc</strong> succeeds, it will exit with 0.  Otherwise, if an error occurs,
+ it will exit with a non-zero value.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="see_also">SEE ALSO</a></h1>
+ <p><a href="././lli.html">lli</a></p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="authors">AUTHORS</a></h1>
+ <p>Maintained by the LLVM Team (<a href="http://llvm.org">http://llvm.org</a>).</p>
+ 
+ </body>
+ 
+ </html>


Index: llvm-www/releases/2.6/docs/CommandGuide/html/lli.html
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/html/lli.html:1.1
*** /dev/null	Fri Oct 23 13:57:43 2009
--- llvm-www/releases/2.6/docs/CommandGuide/html/lli.html	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,310 ----
+ <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
+ <html xmlns="http://www.w3.org/1999/xhtml">
+ <head>
+ <title>lli</title>
+ <link rel="stylesheet" href="manpage.css" type="text/css" />
+ <link rev="made" href="mailto:root at localhost" />
+ </head>
+ 
+ <body>
+ 
+ <p><a name="__index__"></a></p>
+ <!-- INDEX BEGIN -->
+ <!--
+ 
+ <ul>
+ 
+ 	<li><a href="#name">NAME</a></li>
+ 	<li><a href="#synopsis">SYNOPSIS</a></li>
+ 	<li><a href="#description">DESCRIPTION</a></li>
+ 	<li><a href="#general_options">GENERAL OPTIONS</a></li>
+ 	<li><a href="#target_options">TARGET OPTIONS</a></li>
+ 	<li><a href="#floating_point_options">FLOATING POINT OPTIONS</a></li>
+ 	<li><a href="#code_generation_options">CODE GENERATION OPTIONS</a></li>
+ 	<li><a href="#exit_status">EXIT STATUS</a></li>
+ 	<li><a href="#see_also">SEE ALSO</a></li>
+ 	<li><a href="#author">AUTHOR</a></li>
+ </ul>
+ -->
+ <!-- INDEX END -->
+ 
+ <p>
+ </p>
+ <hr />
+ <h1><a name="name">NAME</a></h1>
+ <p>lli - directly execute programs from LLVM bitcode</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="synopsis">SYNOPSIS</a></h1>
+ <p><strong>lli</strong> [<em>options</em>] [<em>filename</em>] [<em>program args</em>]</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="description">DESCRIPTION</a></h1>
+ <p><strong>lli</strong> directly executes programs in LLVM bitcode format.  It takes a program
+ in LLVM bitcode format and executes it using a just-in-time compiler, if one is
+ available for the current architecture, or an interpreter.  <strong>lli</strong> takes all of
+ the same code generator options as <a href="././llc.html">llc</a>, but they are only effective when
+ <strong>lli</strong> is using the just-in-time compiler.</p>
+ <p>If <em>filename</em> is not specified, then <strong>lli</strong> reads the LLVM bitcode for the
+ program from standard input.</p>
+ <p>The optional <em>args</em> specified on the command line are passed to the program as
+ arguments.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="general_options">GENERAL OPTIONS</a></h1>
+ <dl>
+ <dt><strong><a name="item__2dfake_2dargv0_3dexecutable"><strong>-fake-argv0</strong>=<em>executable</em></a></strong><br />
+ </dt>
+ <dd>
+ Override the <code>argv[0]</code> value passed into the executing program.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dforce_2dinterpreter_3d_7bfalse_2ctrue_7d"><strong>-force-interpreter</strong>=<em>{false,true}</em></a></strong><br />
+ </dt>
+ <dd>
+ If set to true, use the interpreter even if a just-in-time compiler is available
+ for this architecture. Defaults to false.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dhelp"><strong>-help</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print a summary of command line options.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dload_3dpuginfilename"><strong>-load</strong>=<em>puginfilename</em></a></strong><br />
+ </dt>
+ <dd>
+ Causes <strong>lli</strong> to load the plugin (shared object) named <em>pluginfilename</em> and use
+ it for optimization.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dstats"><strong>-stats</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print statistics from the code-generation passes. This is only meaningful for
+ the just-in-time compiler, at present.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dtime_2dpasses"><strong>-time-passes</strong></a></strong><br />
+ </dt>
+ <dd>
+ Record the amount of time needed for each code-generation pass and print it to
+ standard error.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dversion"><strong>-version</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print out the version of <strong>lli</strong> and exit without doing anything else.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="target_options">TARGET OPTIONS</a></h1>
+ <dl>
+ <dt><strong><a name="item__2dmtriple_3dtarget_triple"><strong>-mtriple</strong>=<em>target triple</em></a></strong><br />
+ </dt>
+ <dd>
+ Override the target triple specified in the input bitcode file with the 
+ specified string.  This may result in a crash if you pick an
+ architecture which is not compatible with the current system.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dmarch_3darch"><strong>-march</strong>=<em>arch</em></a></strong><br />
+ </dt>
+ <dd>
+ Specify the architecture for which to generate assembly, overriding the target
+ encoded in the bitcode file.  See the output of <strong>llc --help</strong> for a list of
+ valid architectures.  By default this is inferred from the target triple or
+ autodetected to the current architecture.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dmcpu_3dcpuname"><strong>-mcpu</strong>=<em>cpuname</em></a></strong><br />
+ </dt>
+ <dd>
+ Specify a specific chip in the current architecture to generate code for.
+ By default this is inferred from the target triple and autodetected to 
+ the current architecture.  For a list of available CPUs, use:
+ <strong>llvm-as &lt; /dev/null | llc -march=xyz -mcpu=help</strong>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dmattr_3da1_2c_2ba2_2c_2da3_2c_2e_2e_2e"><strong>-mattr</strong>=<em>a1,+a2,-a3,...</em></a></strong><br />
+ </dt>
+ <dd>
+ Override or control specific attributes of the target, such as whether SIMD
+ operations are enabled or not.  The default set of attributes is set by the
+ current CPU.  For a list of available attributes, use:
+ <strong>llvm-as &lt; /dev/null | llc -march=xyz -mattr=help</strong>
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="floating_point_options">FLOATING POINT OPTIONS</a></h1>
+ <dl>
+ <dt><strong><a name="item__2ddisable_2dexcess_2dfp_2dprecision"><strong>-disable-excess-fp-precision</strong></a></strong><br />
+ </dt>
+ <dd>
+ Disable optimizations that may increase floating point precision.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2denable_2dfinite_2donly_2dfp_2dmath"><strong>-enable-finite-only-fp-math</strong></a></strong><br />
+ </dt>
+ <dd>
+ Enable optimizations that assumes only finite floating point math. That is,
+ there is no NAN or Inf values.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2denable_2dunsafe_2dfp_2dmath"><strong>-enable-unsafe-fp-math</strong></a></strong><br />
+ </dt>
+ <dd>
+ Causes <strong>lli</strong> to enable optimizations that may decrease floating point
+ precision.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dsoft_2dfloat"><strong>-soft-float</strong></a></strong><br />
+ </dt>
+ <dd>
+ Causes <strong>lli</strong> to generate software floating point library calls instead of
+ equivalent hardware instructions.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="code_generation_options">CODE GENERATION OPTIONS</a></h1>
+ <dl>
+ <dt><strong><a name="item__2dcode_2dmodel_3dmodel"><strong>-code-model</strong>=<em>model</em></a></strong><br />
+ </dt>
+ <dd>
+ Choose the code model from:
+ </dd>
+ <dd>
+ <pre>
+     default: Target default code model
+     small: Small code model
+     kernel: Kernel code model
+     medium: Medium code model
+     large: Large code model</pre>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2ddisable_2dpost_2dra_2dscheduler"><strong>-disable-post-RA-scheduler</strong></a></strong><br />
+ </dt>
+ <dd>
+ Disable scheduling after register allocation.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2ddisable_2dspill_2dfusing"><strong>-disable-spill-fusing</strong></a></strong><br />
+ </dt>
+ <dd>
+ Disable fusing of spill code into instructions.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2denable_2dcorrect_2deh_2dsupport"><strong>-enable-correct-eh-support</strong></a></strong><br />
+ </dt>
+ <dd>
+ Make the -lowerinvoke pass insert expensive, but correct, EH code.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2denable_2deh"><strong>-enable-eh</strong></a></strong><br />
+ </dt>
+ <dd>
+ Exception handling should be emitted.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2djoin_2dliveintervals"><strong>-join-liveintervals</strong></a></strong><br />
+ </dt>
+ <dd>
+ Coalesce copies (default=true).
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dnozero_2dinitialized_2din_2dbss_don_27t_place_z"><strong>-nozero-initialized-in-bss</strong>
+ Don't place zero-initialized symbols into the BSS section.</a></strong><br />
+ </dt>
+ <dt><strong><a name="item__2dpre_2dra_2dsched_3dscheduler"><strong>-pre-RA-sched</strong>=<em>scheduler</em></a></strong><br />
+ </dt>
+ <dd>
+ Instruction schedulers available (before register allocation):
+ </dd>
+ <dd>
+ <pre>
+     =default: Best scheduler for the target 
+     =none: No scheduling: breadth first sequencing 
+     =simple: Simple two pass scheduling: minimize critical path and maximize processor utilization 
+     =simple-noitin: Simple two pass scheduling: Same as simple except using generic latency 
+     =list-burr: Bottom-up register reduction list scheduling 
+     =list-tdrr: Top-down register reduction list scheduling 
+     =list-td: Top-down list scheduler -print-machineinstrs - Print generated machine code</pre>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dregalloc_3dallocator"><strong>-regalloc</strong>=<em>allocator</em></a></strong><br />
+ </dt>
+ <dd>
+ Register allocator to use: (default = linearscan)
+ </dd>
+ <dd>
+ <pre>
+     =bigblock: Big-block register allocator 
+     =linearscan: linear scan register allocator =local -   local register allocator 
+     =simple: simple register allocator</pre>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2drelocation_2dmodel_3dmodel"><strong>-relocation-model</strong>=<em>model</em></a></strong><br />
+ </dt>
+ <dd>
+ Choose relocation model from:
+ </dd>
+ <dd>
+ <pre>
+     =default: Target default relocation model 
+     =static: Non-relocatable code =pic -   Fully relocatable, position independent code 
+     =dynamic-no-pic: Relocatable external references, non-relocatable code</pre>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dspiller"><strong>-spiller</strong></a></strong><br />
+ </dt>
+ <dd>
+ Spiller to use: (default: local)
+ </dd>
+ <dd>
+ <pre>
+     =simple: simple spiller 
+     =local: local spiller</pre>
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dx86_2dasm_2dsyntax_3dsyntax"><strong>-x86-asm-syntax</strong>=<em>syntax</em></a></strong><br />
+ </dt>
+ <dd>
+ Choose style of code to emit from X86 backend:
+ </dd>
+ <dd>
+ <pre>
+     =att: Emit AT&amp;T-style assembly 
+     =intel: Emit Intel-style assembly</pre>
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="exit_status">EXIT STATUS</a></h1>
+ <p>If <strong>lli</strong> fails to load the program, it will exit with an exit code of 1.
+ Otherwise, it will return the exit code of the program it executes.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="see_also">SEE ALSO</a></h1>
+ <p><a href="././llc.html">llc</a></p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="author">AUTHOR</a></h1>
+ <p>Maintained by the LLVM Team (<a href="http://llvm.org">http://llvm.org</a>).</p>
+ 
+ </body>
+ 
+ </html>


Index: llvm-www/releases/2.6/docs/CommandGuide/html/llvm-ar.html
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/html/llvm-ar.html:1.1
*** /dev/null	Fri Oct 23 13:57:43 2009
--- llvm-www/releases/2.6/docs/CommandGuide/html/llvm-ar.html	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,502 ----
+ <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
+ <html xmlns="http://www.w3.org/1999/xhtml">
+ <head>
+ <title>llvm-ar</title>
+ <link rel="stylesheet" href="manpage.css" type="text/css" />
+ <link rev="made" href="mailto:root at localhost" />
+ </head>
+ 
+ <body>
+ 
+ <p><a name="__index__"></a></p>
+ <!-- INDEX BEGIN -->
+ <!--
+ 
+ <ul>
+ 
+ 	<li><a href="#name">NAME</a></li>
+ 	<li><a href="#synopsis">SYNOPSIS</a></li>
+ 	<li><a href="#description">DESCRIPTION</a></li>
+ 	<li><a href="#options">OPTIONS</a></li>
+ 	<ul>
+ 
+ 		<li><a href="#operations">Operations</a></li>
+ 		<li><a href="#modifiers__operation_specific_">Modifiers (operation specific)</a></li>
+ 		<li><a href="#modifiers__generic_">Modifiers (generic)</a></li>
+ 	</ul>
+ 
+ 	<li><a href="#standards">STANDARDS</a></li>
+ 	<li><a href="#file_format">FILE FORMAT</a></li>
+ 	<li><a href="#exit_status">EXIT STATUS</a></li>
+ 	<li><a href="#see_also">SEE ALSO</a></li>
+ 	<li><a href="#authors">AUTHORS</a></li>
+ </ul>
+ -->
+ <!-- INDEX END -->
+ 
+ <p>
+ </p>
+ <hr />
+ <h1><a name="name">NAME</a></h1>
+ <p>llvm-ar - LLVM archiver</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="synopsis">SYNOPSIS</a></h1>
+ <p><strong>llvm-ar</strong> [-]{dmpqrtx}[Rabfikouz] [relpos] [count] &lt;archive&gt; [files...]</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="description">DESCRIPTION</a></h1>
+ <p>The <strong>llvm-ar</strong> command is similar to the common Unix utility, <code>ar</code>. It 
+ archives several files together into a single file. The intent for this is
+ to produce archive libraries by LLVM bitcode that can be linked into an
+ LLVM program. However, the archive can contain any kind of file. By default,
+ <strong>llvm-ar</strong> generates a symbol table that makes linking faster because
+ only the symbol table needs to be consulted, not each individual file member
+ of the archive.</p>
+ <p>The <strong>llvm-ar</strong> command can be used to <em>read</em> both SVR4 and BSD style archive
+ files. However, it cannot be used to write them.  While the <strong>llvm-ar</strong> command 
+ produces files that are <em>almost</em> identical to the format used by other <code>ar</code> 
+ implementations, it has two significant departures in order to make the 
+ archive appropriate for LLVM. The first departure is that <strong>llvm-ar</strong> only
+ uses BSD4.4 style long path names (stored immediately after the header) and
+ never contains a string table for long names. The second departure is that the
+ symbol table is formated for efficient construction of an in-memory data
+ structure that permits rapid (red-black tree) lookups. Consequently, archives 
+ produced with <strong>llvm-ar</strong> usually won't be readable or editable with any
+ <code>ar</code> implementation or useful for linking.  Using the <code>f</code> modifier to flatten
+ file names will make the archive readable by other <code>ar</code> implementations
+ but not for linking because the symbol table format for LLVM is unique. If an
+ SVR4 or BSD style archive is used with the <code>r</code> (replace) or <code>q</code> (quick
+ update) operations, the archive will be reconstructed in LLVM format. This 
+ means that the string table will be dropped (in deference to BSD 4.4 long names)
+ and an LLVM symbol table will be added (by default). The system symbol table
+ will be retained.</p>
+ <p>Here's where <strong>llvm-ar</strong> departs from previous <code>ar</code> implementations:</p>
+ <dl>
+ <dt><strong><a name="item_symbol_table"><em>Symbol Table</em></a></strong><br />
+ </dt>
+ <dd>
+ Since <strong>llvm-ar</strong> is intended to archive bitcode files, the symbol table
+ won't make much sense to anything but LLVM. Consequently, the symbol table's
+ format has been simplified. It consists simply of a sequence of pairs
+ of a file member index number as an LSB 4byte integer and a null-terminated 
+ string.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_long_paths"><em>Long Paths</em></a></strong><br />
+ </dt>
+ <dd>
+ Some <code>ar</code> implementations (SVR4) use a separate file member to record long
+ path names (&gt; 15 characters). <strong>llvm-ar</strong> takes the BSD 4.4 and Mac OS X 
+ approach which is to simply store the full path name immediately preceding
+ the data for the file. The path name is null terminated and may contain the
+ slash (/) character.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_compression"><em>Compression</em></a></strong><br />
+ </dt>
+ <dd>
+ <strong>llvm-ar</strong> can compress the members of an archive to save space. The 
+ compression used depends on what's available on the platform and what choices
+ the LLVM Compressor utility makes. It generally favors bzip2 but will select
+ between ``no compression'' or bzip2 depending on what makes sense for the
+ file's content.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_directory_recursion"><em>Directory Recursion</em></a></strong><br />
+ </dt>
+ <dd>
+ Most <code>ar</code> implementations do not recurse through directories but simply
+ ignore directories if they are presented to the program in the <em>files</em> 
+ option. <strong>llvm-ar</strong>, however, can recurse through directory structures and
+ add all the files under a directory, if requested.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_toc_verbose_output"><em>TOC Verbose Output</em></a></strong><br />
+ </dt>
+ <dd>
+ When <strong>llvm-ar</strong> prints out the verbose table of contents (<code>tv</code> option), it
+ precedes the usual output with a character indicating the basic kind of 
+ content in the file. A blank means the file is a regular file. A 'Z' means
+ the file is compressed. A 'B' means the file is an LLVM bitcode file. An
+ 'S' means the file is the symbol table.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="options">OPTIONS</a></h1>
+ <p>The options to <strong>llvm-ar</strong> are compatible with other <code>ar</code> implementations.
+ However, there are a few modifiers (<em>zR</em>) that are not found in other
+ <code>ar</code>s. The options to <strong>llvm-ar</strong> specify a single basic operation to 
+ perform on the archive, a variety of modifiers for that operation, the
+ name of the archive file, and an optional list of file names. These options
+ are used to determine how <strong>llvm-ar</strong> should process the archive file.</p>
+ <p>The Operations and Modifiers are explained in the sections below. The minimal
+ set of options is at least one operator and the name of the archive. Typically
+ archive files end with a <code>.a</code> suffix, but this is not required. Following
+ the <em>archive-name</em> comes a list of <em>files</em> that indicate the specific members
+ of the archive to operate on. If the <em>files</em> option is not specified, it
+ generally means either ``none'' or ``all'' members, depending on the operation.</p>
+ <p>
+ </p>
+ <h2><a name="operations">Operations</a></h2>
+ <dl>
+ <dt><strong><a name="item_d">d</a></strong><br />
+ </dt>
+ <dd>
+ Delete files from the archive. No modifiers are applicable to this operation.
+ The <em>files</em> options specify which members should be removed from the
+ archive. It is not an error if a specified file does not appear in the archive.
+ If no <em>files</em> are specified, the archive is not modified.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_m_5babi_5d">m[abi]</a></strong><br />
+ </dt>
+ <dd>
+ Move files from one location in the archive to another. The <em>a</em>, <em>b</em>, and 
+ <em>i</em> modifiers apply to this operation. The <em>files</em> will all be moved
+ to the location given by the modifiers. If no modifiers are used, the files
+ will be moved to the end of the archive. If no <em>files</em> are specified, the
+ archive is not modified.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_p_5bk_5d">p[k]</a></strong><br />
+ </dt>
+ <dd>
+ Print files to the standard output. The <em>k</em> modifier applies to this
+ operation. This operation simply prints the <em>files</em> indicated to the
+ standard output. If no <em>files</em> are specified, the entire archive is printed.
+ Printing bitcode files is ill-advised as they might confuse your terminal
+ settings. The <em>p</em> operation never modifies the archive.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_q_5brfz_5d">q[Rfz]</a></strong><br />
+ </dt>
+ <dd>
+ Quickly append files to the end of the archive. The <em>R</em>, <em>f</em>, and <em>z</em>
+ modifiers apply to this operation.  This operation quickly adds the 
+ <em>files</em> to the archive without checking for duplicates that should be 
+ removed first. If no <em>files</em> are specified, the archive is not modified. 
+ Because of the way that <strong>llvm-ar</strong> constructs the archive file, its dubious 
+ whether the <em>q</em> operation is any faster than the <em>r</em> operation.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_r_5brabfuz_5d">r[Rabfuz]</a></strong><br />
+ </dt>
+ <dd>
+ Replace or insert file members. The <em>R</em>, <em>a</em>, <em>b</em>, <em>f</em>, <em>u</em>, and <em>z</em>
+ modifiers apply to this operation. This operation will replace existing
+ <em>files</em> or insert them at the end of the archive if they do not exist. If no
+ <em>files</em> are specified, the archive is not modified.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_t_5bv_5d">t[v]</a></strong><br />
+ </dt>
+ <dd>
+ Print the table of contents. Without any modifiers, this operation just prints
+ the names of the members to the standard output. With the <em>v</em> modifier,
+ <strong>llvm-ar</strong> also prints out the file type (B=bitcode, Z=compressed, S=symbol
+ table, blank=regular file), the permission mode, the owner and group, the
+ size, and the date. If any <em>files</em> are specified, the listing is only for
+ those files. If no <em>files</em> are specified, the table of contents for the
+ whole archive is printed.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_x_5bop_5d">x[oP]</a></strong><br />
+ </dt>
+ <dd>
+ Extract archive members back to files. The <em>o</em> modifier applies to this
+ operation. This operation retrieves the indicated <em>files</em> from the archive 
+ and writes them back to the operating system's file system. If no 
+ <em>files</em> are specified, the entire archive is extract.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <h2><a name="modifiers__operation_specific_">Modifiers (operation specific)</a></h2>
+ <p>The modifiers below are specific to certain operations. See the Operations
+ section (above) to determine which modifiers are applicable to which operations.</p>
+ <dl>
+ <dt><strong><a name="item__5ba_5d">[a]</a></strong><br />
+ </dt>
+ <dd>
+ When inserting or moving member files, this option specifies the destination of
+ the new files as being <code>a</code>fter the <em>relpos</em> member. If <em>relpos</em> is not found,
+ the files are placed at the end of the archive.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bb_5d">[b]</a></strong><br />
+ </dt>
+ <dd>
+ When inserting or moving member files, this option specifies the destination of
+ the new files as being <code>b</code>efore the <em>relpos</em> member. If <em>relpos</em> is not 
+ found, the files are placed at the end of the archive. This modifier is 
+ identical to the the <em>i</em> modifier.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bf_5d">[f]</a></strong><br />
+ </dt>
+ <dd>
+ Normally, <strong>llvm-ar</strong> stores the full path name to a file as presented to it on
+ the command line. With this option, truncated (15 characters max) names are
+ used. This ensures name compatibility with older versions of <code>ar</code> but may also
+ thwart correct extraction of the files (duplicates may overwrite). If used with
+ the <em>R</em> option, the directory recursion will be performed but the file names
+ will all be <code>f</code>lattened to simple file names.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bi_5d">[i]</a></strong><br />
+ </dt>
+ <dd>
+ A synonym for the <em>b</em> option.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bk_5d">[k]</a></strong><br />
+ </dt>
+ <dd>
+ Normally, <strong>llvm-ar</strong> will not print the contents of bitcode files when the 
+ <em>p</em> operation is used. This modifier defeats the default and allows the 
+ bitcode members to be printed.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bn_5d">[N]</a></strong><br />
+ </dt>
+ <dd>
+ This option is ignored by <strong>llvm-ar</strong> but provided for compatibility.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bo_5d">[o]</a></strong><br />
+ </dt>
+ <dd>
+ When extracting files, this option will cause <strong>llvm-ar</strong> to preserve the
+ original modification times of the files it writes.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bp_5d">[P]</a></strong><br />
+ </dt>
+ <dd>
+ use full path names when matching
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5br_5d">[R]</a></strong><br />
+ </dt>
+ <dd>
+ This modifier instructions the <em>r</em> option to recursively process directories.
+ Without <em>R</em>, directories are ignored and only those <em>files</em> that refer to
+ files will be added to the archive. When <em>R</em> is used, any directories specified
+ with <em>files</em> will be scanned (recursively) to find files to be added to the
+ archive. Any file whose name begins with a dot will not be added.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bu_5d">[u]</a></strong><br />
+ </dt>
+ <dd>
+ When replacing existing files in the archive, only replace those files that have
+ a time stamp than the time stamp of the member in the archive.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bz_5d">[z]</a></strong><br />
+ </dt>
+ <dd>
+ When inserting or replacing any file in the archive, compress the file first.
+ This
+ modifier is safe to use when (previously) compressed bitcode files are added to
+ the archive; the compressed bitcode files will not be doubly compressed.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <h2><a name="modifiers__generic_">Modifiers (generic)</a></h2>
+ <p>The modifiers below may be applied to any operation.</p>
+ <dl>
+ <dt><strong><a name="item__5bc_5d">[c]</a></strong><br />
+ </dt>
+ <dd>
+ For all operations, <strong>llvm-ar</strong> will always create the archive if it doesn't 
+ exist. Normally, <strong>llvm-ar</strong> will print a warning message indicating that the
+ archive is being created. Using this modifier turns off that warning.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bs_5d">[s]</a></strong><br />
+ </dt>
+ <dd>
+ This modifier requests that an archive index (or symbol table) be added to the
+ archive. This is the default mode of operation. The symbol table will contain
+ all the externally visible functions and global variables defined by all the
+ bitcode files in the archive. Using this modifier is more efficient that using
+ <a href="././llvm-ranlib.html">llvm-ranlib</a> which also creates the symbol table.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bs_5d">[S]</a></strong><br />
+ </dt>
+ <dd>
+ This modifier is the opposite of the <em>s</em> modifier. It instructs <strong>llvm-ar</strong> to
+ not build the symbol table. If both <em>s</em> and <em>S</em> are used, the last modifier to
+ occur in the options will prevail.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__5bv_5d">[v]</a></strong><br />
+ </dt>
+ <dd>
+ This modifier instructs <strong>llvm-ar</strong> to be verbose about what it is doing. Each
+ editing operation taken against the archive will produce a line of output saying
+ what is being done.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="standards">STANDARDS</a></h1>
+ <p>The <strong>llvm-ar</strong> utility is intended to provide a superset of the IEEE Std 1003.2
+ (POSIX.2) functionality for <code>ar</code>. <strong>llvm-ar</strong> can read both SVR4 and BSD4.4 (or
+ Mac OS X) archives. If the <code>f</code> modifier is given to the <code>x</code> or <code>r</code> operations
+ then <strong>llvm-ar</strong> will write SVR4 compatible archives. Without this modifier, 
+ <strong>llvm-ar</strong> will write BSD4.4 compatible archives that have long names
+ immediately after the header and indicated using the ``#1/ddd'' notation for the
+ name in the header.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="file_format">FILE FORMAT</a></h1>
+ <p>The file format for LLVM Archive files is similar to that of BSD 4.4 or Mac OSX
+ archive files. In fact, except for the symbol table, the <code>ar</code> commands on those
+ operating systems should be able to read LLVM archive files. The details of the
+ file format follow.</p>
+ <p>Each archive begins with the archive magic number which is the eight printable
+ characters ``!&lt;arch&gt;\n'' where \n represents the newline character (0x0A). 
+ Following the magic number, the file is composed of even length members that 
+ begin with an archive header and end with a \n padding character if necessary 
+ (to make the length even). Each file member is composed of a header (defined 
+ below), an optional newline-terminated ``long file name'' and the contents of 
+ the file.</p>
+ <p>The fields of the header are described in the items below. All fields of the
+ header contain only ASCII characters, are left justified and are right padded 
+ with space characters.</p>
+ <dl>
+ <dt><strong><a name="item_name__2d_char_5b16_5d">name - char[16]</a></strong><br />
+ </dt>
+ <dd>
+ This field of the header provides the name of the archive member. If the name is
+ longer than 15 characters or contains a slash (/) character, then this field
+ contains <code>#1/nnn</code> where <code>nnn</code> provides the length of the name and the <code>#1/</code>
+ is literal.  In this case, the actual name of the file is provided in the <code>nnn</code>
+ bytes immediately following the header. If the name is 15 characters or less, it
+ is contained directly in this field and terminated with a slash (/) character.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_date__2d_char_5b12_5d">date - char[12]</a></strong><br />
+ </dt>
+ <dd>
+ This field provides the date of modification of the file in the form of a
+ decimal encoded number that provides the number of seconds since the epoch 
+ (since 00:00:00 Jan 1, 1970) per Posix specifications.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_uid__2d_char_5b6_5d">uid - char[6]</a></strong><br />
+ </dt>
+ <dd>
+ This field provides the user id of the file encoded as a decimal ASCII string.
+ This field might not make much sense on non-Unix systems. On Unix, it is the
+ same value as the st_uid field of the stat structure returned by the <code>stat(2)</code>
+ operating system call.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_gid__2d_char_5b6_5d">gid - char[6]</a></strong><br />
+ </dt>
+ <dd>
+ This field provides the group id of the file encoded as a decimal ASCII string.
+ This field might not make much sense on non-Unix systems. On Unix, it is the
+ same value as the st_gid field of the stat structure returned by the <code>stat(2)</code>
+ operating system call.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_mode__2d_char_5b8_5d">mode - char[8]</a></strong><br />
+ </dt>
+ <dd>
+ This field provides the access mode of the file encoded as an octal ASCII 
+ string. This field might not make much sense on non-Unix systems. On Unix, it 
+ is the same value as the st_mode field of the stat structure returned by the 
+ <code>stat(2)</code> operating system call.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_size__2d_char_5b10_5d">size - char[10]</a></strong><br />
+ </dt>
+ <dd>
+ This field provides the size of the file, in bytes, encoded as a decimal ASCII
+ string. If the size field is negative (starts with a minus sign, 0x02D), then
+ the archive member is stored in compressed form. The first byte of the archive
+ member's data indicates the compression type used. A value of 0 (0x30) indicates
+ that no compression was used. A value of 2 (0x32) indicates that bzip2
+ compression was used.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_fmag__2d_char_5b2_5d">fmag - char[2]</a></strong><br />
+ </dt>
+ <dd>
+ This field is the archive file member magic number. Its content is always the
+ two characters back tick (0x60) and newline (0x0A). This provides some measure 
+ utility in identifying archive files that have been corrupted.
+ </dd>
+ <p></p></dl>
+ <p>The LLVM symbol table has the special name ``#_LLVM_SYM_TAB_#''. It is presumed
+ that no regular archive member file will want this name. The LLVM symbol table 
+ is simply composed of a sequence of triplets: byte offset, length of symbol, 
+ and the symbol itself. Symbols are not null or newline terminated. Here are 
+ the details on each of these items:</p>
+ <dl>
+ <dt><strong><a name="item_offset__2d_vbr_encoded_32_2dbit_integer">offset - vbr encoded 32-bit integer</a></strong><br />
+ </dt>
+ <dd>
+ The offset item provides the offset into the archive file where the bitcode
+ member is stored that is associated with the symbol. The offset value is 0
+ based at the start of the first ``normal'' file member. To derive the actual
+ file offset of the member, you must add the number of bytes occupied by the file
+ signature (8 bytes) and the symbol tables. The value of this item is encoded
+ using variable bit rate encoding to reduce the size of the symbol table.
+ Variable bit rate encoding uses the high bit (0x80) of each byte to indicate 
+ if there are more bytes to follow. The remaining 7 bits in each byte carry bits
+ from the value. The final byte does not have the high bit set.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_length__2d_vbr_encoded_32_2dbit_integer">length - vbr encoded 32-bit integer</a></strong><br />
+ </dt>
+ <dd>
+ The length item provides the length of the symbol that follows. Like this
+ <em>offset</em> item, the length is variable bit rate encoded.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_symbol__2d_character_array">symbol - character array</a></strong><br />
+ </dt>
+ <dd>
+ The symbol item provides the text of the symbol that is associated with the
+ <em>offset</em>. The symbol is not terminated by any character. Its length is provided
+ by the <em>length</em> field. Note that is allowed (but unwise) to use non-printing
+ characters (even 0x00) in the symbol. This allows for multiple encodings of 
+ symbol names.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="exit_status">EXIT STATUS</a></h1>
+ <p>If <strong>llvm-ar</strong> succeeds, it will exit with 0.  A usage error, results
+ in an exit code of 1. A hard (file system typically) error results in an
+ exit code of 2. Miscellaneous or unknown errors result in an
+ exit code of 3.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="see_also">SEE ALSO</a></h1>
+ <p><a href="././llvm-ranlib.html">llvm-ranlib</a>, <code>ar(1)</code></p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="authors">AUTHORS</a></h1>
+ <p>Maintained by the LLVM Team (<a href="http://llvm.org">http://llvm.org</a>).</p>
+ 
+ </body>
+ 
+ </html>


Index: llvm-www/releases/2.6/docs/CommandGuide/html/llvm-as.html
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/html/llvm-as.html:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/CommandGuide/html/llvm-as.html	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,108 ----
+ <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
+ <html xmlns="http://www.w3.org/1999/xhtml">
+ <head>
+ <title>llvm-as</title>
+ <link rel="stylesheet" href="manpage.css" type="text/css" />
+ <link rev="made" href="mailto:root at localhost" />
+ </head>
+ 
+ <body>
+ 
+ <p><a name="__index__"></a></p>
+ <!-- INDEX BEGIN -->
+ <!--
+ 
+ <ul>
+ 
+ 	<li><a href="#name">NAME</a></li>
+ 	<li><a href="#synopsis">SYNOPSIS</a></li>
+ 	<li><a href="#description">DESCRIPTION</a></li>
+ 	<li><a href="#options">OPTIONS</a></li>
+ 	<li><a href="#exit_status">EXIT STATUS</a></li>
+ 	<li><a href="#see_also">SEE ALSO</a></li>
+ 	<li><a href="#authors">AUTHORS</a></li>
+ </ul>
+ -->
+ <!-- INDEX END -->
+ 
+ <p>
+ </p>
+ <hr />
+ <h1><a name="name">NAME</a></h1>
+ <p>llvm-as - LLVM assembler</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="synopsis">SYNOPSIS</a></h1>
+ <p><strong>llvm-as</strong> [<em>options</em>] [<em>filename</em>]</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="description">DESCRIPTION</a></h1>
+ <p><strong>llvm-as</strong> is the LLVM assembler.  It reads a file containing human-readable
+ LLVM assembly language, translates it to LLVM bitcode, and writes the result
+ into a file or to standard output.</p>
+ <p>If <em>filename</em> is omitted or is <code>-</code>, then <strong>llvm-as</strong> reads its input from
+ standard input.</p>
+ <p>If an output file is not specified with the <strong>-o</strong> option, then
+ <strong>llvm-as</strong> sends its output to a file or standard output by following
+ these rules:</p>
+ <ul>
+ <li></li>
+ If the input is standard input, then the output is standard output.
+ <p></p>
+ <li></li>
+ If the input is a file that ends with <code>.ll</code>, then the output file is of
+ the same name, except that the suffix is changed to <code>.bc</code>.
+ <p></p>
+ <li></li>
+ If the input is a file that does not end with the <code>.ll</code> suffix, then the
+ output file has the same name as the input file, except that the <code>.bc</code>
+ suffix is appended.
+ <p></p></ul>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="options">OPTIONS</a></h1>
+ <dl>
+ <dt><strong><a name="item__2df"><strong>-f</strong></a></strong><br />
+ </dt>
+ <dd>
+ Force overwrite.  Normally, <strong>llvm-as</strong> will refuse to overwrite an
+ output file that already exists.  With this option, <strong>llvm-as</strong>
+ will overwrite the output file and replace it with new bitcode.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dhelp"><strong>--help</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print a summary of command line options.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2do_filename"><strong>-o</strong> <em>filename</em></a></strong><br />
+ </dt>
+ <dd>
+ Specify the output file name.  If <em>filename</em> is <code>-</code>, then <strong>llvm-as</strong>
+ sends its output to standard output.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="exit_status">EXIT STATUS</a></h1>
+ <p>If <strong>llvm-as</strong> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="see_also">SEE ALSO</a></h1>
+ <p><a href="././llvm-dis.html">llvm-dis</a>, <em>gccas</em></p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="authors">AUTHORS</a></h1>
+ <p>Maintained by the LLVM Team (<a href="http://llvm.org">http://llvm.org</a>).</p>
+ 
+ </body>
+ 
+ </html>


Index: llvm-www/releases/2.6/docs/CommandGuide/html/llvm-bcanalyzer.html
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/html/llvm-bcanalyzer.html:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/CommandGuide/html/llvm-bcanalyzer.html	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,451 ----
+ <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
+ <html xmlns="http://www.w3.org/1999/xhtml">
+ <head>
+ <title>llvm-bcanalyzer</title>
+ <link rel="stylesheet" href="manpage.css" type="text/css" />
+ <link rev="made" href="mailto:root at localhost" />
+ </head>
+ 
+ <body>
+ 
+ <p><a name="__index__"></a></p>
+ <!-- INDEX BEGIN -->
+ <!--
+ 
+ <ul>
+ 
+ 	<li><a href="#name">NAME</a></li>
+ 	<li><a href="#synopsis">SYNOPSIS</a></li>
+ 	<li><a href="#description">DESCRIPTION</a></li>
+ 	<li><a href="#options">OPTIONS</a></li>
+ 	<li><a href="#exit_status">EXIT STATUS</a></li>
+ 	<li><a href="#summary_output_definitions">SUMMARY OUTPUT DEFINITIONS</a></li>
+ 	<li><a href="#detailed_output_definitions">DETAILED OUTPUT DEFINITIONS</a></li>
+ 	<li><a href="#see_also">SEE ALSO</a></li>
+ 	<li><a href="#authors">AUTHORS</a></li>
+ </ul>
+ -->
+ <!-- INDEX END -->
+ 
+ <p>
+ </p>
+ <hr />
+ <h1><a name="name">NAME</a></h1>
+ <p>llvm-bcanalyzer - LLVM bitcode analyzer</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="synopsis">SYNOPSIS</a></h1>
+ <p><strong>llvm-bcanalyzer</strong> [<em>options</em>] [<em>filename</em>]</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="description">DESCRIPTION</a></h1>
+ <p>The <strong>llvm-bcanalyzer</strong> command is a small utility for analyzing bitcode files.
+ The tool reads a bitcode file (such as generated with the <strong>llvm-as</strong> tool) and
+ produces a statistical report on the contents of the bitcode file.  The tool
+ can also dump a low level but human readable version of the bitcode file. 
+ This tool is probably not of much interest or utility except for those working 
+ directly with the bitcode file format. Most LLVM users can just ignore
+ this tool.</p>
+ <p>If <em>filename</em> is omitted or is <code>-</code>, then <strong>llvm-bcanalyzer</strong> reads its input 
+ from standard input. This is useful for combining the tool into a pipeline.
+ Output is written to the standard output.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="options">OPTIONS</a></h1>
+ <dl>
+ <dt><strong><a name="item__2dnodetails"><strong>-nodetails</strong></a></strong><br />
+ </dt>
+ <dd>
+ Causes <strong>llvm-bcanalyzer</strong> to abbreviate its output by writing out only a module 
+ level summary. The details for individual functions are not displayed.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2ddump"><strong>-dump</strong></a></strong><br />
+ </dt>
+ <dd>
+ Causes <strong>llvm-bcanalyzer</strong> to dump the bitcode in a human readable format. This 
+ format is significantly different from LLVM assembly and provides details about 
+ the encoding of the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2dverify"><strong>-verify</strong></a></strong><br />
+ </dt>
+ <dd>
+ Causes <strong>llvm-bcanalyzer</strong> to verify the module produced by reading the 
+ bitcode. This ensures that the statistics generated are based on a consistent
+ module.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__2d_2dhelp"><strong>--help</strong></a></strong><br />
+ </dt>
+ <dd>
+ Print a summary of command line options.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="exit_status">EXIT STATUS</a></h1>
+ <p>If <strong>llvm-bcanalyzer</strong> succeeds, it will exit with 0.  Otherwise, if an error
+ occurs, it will exit with a non-zero value, usually 1.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="summary_output_definitions">SUMMARY OUTPUT DEFINITIONS</a></h1>
+ <p>The following items are always printed by llvm-bcanalyzer. They comprize the
+ summary output.</p>
+ <dl>
+ <dt><strong><a name="item_bitcode_analysis_of_module"><strong>Bitcode Analysis Of Module</strong></a></strong><br />
+ </dt>
+ <dd>
+ This just provides the name of the module for which bitcode analysis is being
+ generated.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_bitcode_version_number"><strong>Bitcode Version Number</strong></a></strong><br />
+ </dt>
+ <dd>
+ The bitcode version (not LLVM version) of the file read by the analyzer.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_file_size"><strong>File Size</strong></a></strong><br />
+ </dt>
+ <dd>
+ The size, in bytes, of the entire bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_module_bytes"><strong>Module Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The size, in bytes, of the module block. Percentage is relative to File Size.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_function_bytes"><strong>Function Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The size, in bytes, of all the function blocks. Percentage is relative to File
+ Size.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_global_types_bytes"><strong>Global Types Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The size, in bytes, of the Global Types Pool. Percentage is relative to File
+ Size. This is the size of the definitions of all types in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_constant_pool_bytes"><strong>Constant Pool Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The size, in bytes, of the Constant Pool Blocks Percentage is relative to File
+ Size.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_module_globals_bytes"><strong>Module Globals Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ Ths size, in bytes, of the Global Variable Definitions and their initializers.
+ Percentage is relative to File Size.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_instruction_list_bytes"><strong>Instruction List Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The size, in bytes, of all the instruction lists in all the functions.
+ Percentage is relative to File Size. Note that this value is also included in
+ the Function Bytes.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_compaction_table_bytes"><strong>Compaction Table Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The size, in bytes, of all the compaction tables in all the functions.
+ Percentage is relative to File Size. Note that this value is also included in
+ the Function Bytes.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_symbol_table_bytes"><strong>Symbol Table Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The size, in bytes, of all the symbol tables in all the functions. Percentage is
+ relative to File Size. Note that this value is also included in the Function
+ Bytes.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_dependent_libraries_bytes"><strong>Dependent Libraries Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The size, in bytes, of the list of dependent libraries in the module. Percentage
+ is relative to File Size. Note that this value is also included in the Module
+ Global Bytes.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_bitcode_blocks"><strong>Number Of Bitcode Blocks</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of blocks of any kind in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_functions"><strong>Number Of Functions</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of function definitions in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_types"><strong>Number Of Types</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of types defined in the Global Types Pool.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_constants"><strong>Number Of Constants</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of constants (of any type) defined in the Constant Pool.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_basic_blocks"><strong>Number Of Basic Blocks</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of basic blocks defined in all functions in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_instructions"><strong>Number Of Instructions</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of instructions defined in all functions in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_long_instructions"><strong>Number Of Long Instructions</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of long instructions defined in all functions in the bitcode
+ file. Long instructions are those taking greater than 4 bytes. Typically long
+ instructions are GetElementPtr with several indices, PHI nodes, and calls to
+ functions with large numbers of arguments.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_operands"><strong>Number Of Operands</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of operands used in all instructions in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_compaction_tables"><strong>Number Of Compaction Tables</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of compaction tables in all functions in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_symbol_tables"><strong>Number Of Symbol Tables</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of symbol tables in all functions in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_dependent_libs"><strong>Number Of Dependent Libs</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of dependent libraries found in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_total_instruction_size"><strong>Total Instruction Size</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total size of the instructions in all functions in the bitcode file.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_average_instruction_size"><strong>Average Instruction Size</strong></a></strong><br />
+ </dt>
+ <dd>
+ The average number of bytes per instruction across all functions in the bitcode
+ file. This value is computed by dividing Total Instruction Size by Number Of
+ Instructions.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_maximum_type_slot_number"><strong>Maximum Type Slot Number</strong></a></strong><br />
+ </dt>
+ <dd>
+ The maximum value used for a type's slot number. Larger slot number values take 
+ more bytes to encode.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_maximum_value_slot_number"><strong>Maximum Value Slot Number</strong></a></strong><br />
+ </dt>
+ <dd>
+ The maximum value used for a value's slot number. Larger slot number values take 
+ more bytes to encode.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_bytes_per_value"><strong>Bytes Per Value</strong></a></strong><br />
+ </dt>
+ <dd>
+ The average size of a Value definition (of any type). This is computed by
+ dividing File Size by the total number of values of any type.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_bytes_per_global"><strong>Bytes Per Global</strong></a></strong><br />
+ </dt>
+ <dd>
+ The average size of a global definition (constants and global variables).
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_bytes_per_function"><strong>Bytes Per Function</strong></a></strong><br />
+ </dt>
+ <dd>
+ The average number of bytes per function definition. This is computed by
+ dividing Function Bytes by Number Of Functions.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__23_of_vbr_32_2dbit_integers"><strong># of VBR 32-bit Integers</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of 32-bit integers encoded using the Variable Bit Rate
+ encoding scheme.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__23_of_vbr_64_2dbit_integers"><strong># of VBR 64-bit Integers</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of 64-bit integers encoded using the Variable Bit Rate encoding
+ scheme.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__23_of_vbr_compressed_bytes"><strong># of VBR Compressed Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of bytes consumed by the 32-bit and 64-bit integers that use
+ the Variable Bit Rate encoding scheme.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item__23_of_vbr_expanded_bytes"><strong># of VBR Expanded Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of bytes that would have been consumed by the 32-bit and 64-bit
+ integers had they not been compressed with the Variable Bit Rage encoding
+ scheme.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_bytes_saved_with_vbr"><strong>Bytes Saved With VBR</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of bytes saved by using the Variable Bit Rate encoding scheme.
+ The percentage is relative to # of VBR Expanded Bytes.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="detailed_output_definitions">DETAILED OUTPUT DEFINITIONS</a></h1>
+ <p>The following definitions occur only if the -nodetails option was not given.
+ The detailed output provides additional information on a per-function basis.</p>
+ <dl>
+ <dt><strong><a name="item_type"><strong>Type</strong></a></strong><br />
+ </dt>
+ <dd>
+ The type signature of the function.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_byte_size"><strong>Byte Size</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of bytes in the function's block.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_basic_blocks"><strong>Basic Blocks</strong></a></strong><br />
+ </dt>
+ <dd>
+ The number of basic blocks defined by the function.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_instructions"><strong>Instructions</strong></a></strong><br />
+ </dt>
+ <dd>
+ The number of instructions defined by the function.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_long_instructions"><strong>Long Instructions</strong></a></strong><br />
+ </dt>
+ <dd>
+ The number of instructions using the long instruction format in the function.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_operands"><strong>Operands</strong></a></strong><br />
+ </dt>
+ <dd>
+ The number of operands used by all instructions in the function.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_instruction_size"><strong>Instruction Size</strong></a></strong><br />
+ </dt>
+ <dd>
+ The number of bytes consumed by instructions in the function.
+ </dd>
+ <p></p>
+ <dt><strong><strong>Average Instruction Size</strong></strong><br />
+ </dt>
+ <dd>
+ The average number of bytes consumed by the instructions in the funtion. This
+ value is computed by dividing Instruction Size by Instructions.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_bytes_per_instruction"><strong>Bytes Per Instruction</strong></a></strong><br />
+ </dt>
+ <dd>
+ The average number of bytes used by the function per instruction. This value is
+ computed by dividing Byte Size by Instructions. Note that this is not the same
+ as Average Instruction Size. It computes a number relative to the total function
+ size not just the size of the instruction list.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_vbr_32_2dbit_integers"><strong>Number of VBR 32-bit Integers</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of 32-bit integers found in this function (for any use).
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_vbr_64_2dbit_integers"><strong>Number of VBR 64-bit Integers</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of 64-bit integers found in this function (for any use).
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_vbr_compressed_bytes"><strong>Number of VBR Compressed Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of bytes in this function consumed by the 32-bit and 64-bit 
+ integers that use the Variable Bit Rate encoding scheme.
+ </dd>
+ <p></p>
+ <dt><strong><a name="item_number_of_vbr_expanded_bytes"><strong>Number of VBR Expanded Bytes</strong></a></strong><br />
+ </dt>
+ <dd>
+ The total number of bytes in this function that would have been consumed by 
+ the 32-bit and 64-bit integers had they not been compressed with the Variable 
+ Bit Rate encoding scheme.
+ </dd>
+ <p></p>
+ <dt><strong><strong>Bytes Saved With VBR</strong></strong><br />
+ </dt>
+ <dd>
+ The total number of bytes saved in this function by using the Variable Bit 
+ Rate encoding scheme. The percentage is relative to # of VBR Expanded Bytes.
+ </dd>
+ <p></p></dl>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="see_also">SEE ALSO</a></h1>
+ <p><a href="././llvm-dis.html">llvm-dis</a>, <a href="http://llvm.org/docs/BitCodeFormat.html">http://llvm.org/docs/BitCodeFormat.html</a></p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="authors">AUTHORS</a></h1>
+ <p>Maintained by the LLVM Team (<a href="http://llvm.org">http://llvm.org</a>).</p>
+ 
+ </body>
+ 
+ </html>


Index: llvm-www/releases/2.6/docs/CommandGuide/html/llvm-config.html
diff -c /dev/null llvm-www/releases/2.6/docs/CommandGuide/html/llvm-config.html:1.1
*** /dev/null	Fri Oct 23 13:57:44 2009
--- llvm-www/releases/2.6/docs/CommandGuide/html/llvm-config.html	Fri Oct 23 13:56:21 2009
***************
*** 0 ****
--- 1,201 ----
+ <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
+ <html xmlns="http://www.w3.org/1999/xhtml">
+ <head>
+ <title>llvm-config</title>
+ <link rel="stylesheet" href="manpage.css" type="text/css" />
+ <link rev="made" href="mailto:root at localhost" />
+ </head>
+ 
+ <body>
+ 
+ <p><a name="__index__"></a></p>
+ <!-- INDEX BEGIN -->
+ <!--
+ 
+ <ul>
+ 
+ 	<li><a href="#name">NAME</a></li>
+ 	<li><a href="#synopsis">SYNOPSIS</a></li>
+ 	<li><a href="#description">DESCRIPTION</a></li>
+ 	<li><a href="#examples">EXAMPLES</a></li>
+ 	<li><a href="#options">OPTIONS</a></li>
+ 	<li><a href="#components">COMPONENTS</a></li>
+ 	<li><a href="#exit_status">EXIT STATUS</a></li>
+ 	<li><a href="#authors">AUTHORS</a></li>
+ </ul>
+ -->
+ <!-- INDEX END -->
+ 
+ <p>
+ </p>
+ <hr />
+ <h1><a name="name">NAME</a></h1>
+ <p>llvm-config - Print LLVM compilation options</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="synopsis">SYNOPSIS</a></h1>
+ <p><strong>llvm-config</strong> <em>option</em> [<em>components</em>...]</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="description">DESCRIPTION</a></h1>
+ <p><strong>llvm-config</strong> makes it easier to build applications that use LLVM.  It can
+ print the compiler flags, linker flags and object libraries needed to link
+ against LLVM.</p>
+ <p>
+ </p>
+ <hr />
+ <h1><a name="examples">EXAMPLES</a></h1>
+ <p>To link against the JIT:</p>
+ <pre>
+   g++ `llvm-config --cxxflags` -o HowToUseJIT.o -c HowToUseJI