From rafael.espindola at gmail.com Fri Jan 14 11:01:20 2011 From: rafael.espindola at gmail.com (Rafael Espindola) Date: Fri, 14 Jan 2011 17:01:20 -0000 Subject: [llvm-commits] [llvm] r123447 - in /llvm/trunk/test/FrontendC: 2005-02-27-MarkGlobalConstant.c 2006-03-03-MissingInitializer.c 2007-02-16-WritableStrings.c hidden-visibility.c pr3518.c pr4349.c Message-ID: <20110114170120.B89712A6C12C@llvm.org> Author: rafael Date: Fri Jan 14 11:01:20 2011 New Revision: 123447 URL: http://llvm.org/viewvc/llvm-project?rev=123447&view=rev Log: Update llvm-gcc's tests. Modified: llvm/trunk/test/FrontendC/2005-02-27-MarkGlobalConstant.c llvm/trunk/test/FrontendC/2006-03-03-MissingInitializer.c llvm/trunk/test/FrontendC/2007-02-16-WritableStrings.c llvm/trunk/test/FrontendC/hidden-visibility.c llvm/trunk/test/FrontendC/pr3518.c llvm/trunk/test/FrontendC/pr4349.c Modified: llvm/trunk/test/FrontendC/2005-02-27-MarkGlobalConstant.c URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC/2005-02-27-MarkGlobalConstant.c?rev=123447&r1=123446&r2=123447&view=diff ============================================================================== --- llvm/trunk/test/FrontendC/2005-02-27-MarkGlobalConstant.c (original) +++ llvm/trunk/test/FrontendC/2005-02-27-MarkGlobalConstant.c Fri Jan 14 11:01:20 2011 @@ -1,4 +1,4 @@ -// RUN: %llvmgcc -xc %s -S -o - | grep {private constant } +// RUN: %llvmgcc -xc %s -S -o - | grep {private unnamed_addr constant } // The synthetic global made by the CFE for big initializer should be marked // constant. Modified: llvm/trunk/test/FrontendC/2006-03-03-MissingInitializer.c URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC/2006-03-03-MissingInitializer.c?rev=123447&r1=123446&r2=123447&view=diff ============================================================================== --- llvm/trunk/test/FrontendC/2006-03-03-MissingInitializer.c (original) +++ llvm/trunk/test/FrontendC/2006-03-03-MissingInitializer.c Fri Jan 14 11:01:20 2011 @@ -1,5 +1,5 @@ // RUN: %llvmgcc %s -S -o - | opt -std-compile-opts | \ -// RUN: llvm-dis | grep {@nate.*internal global i32 0} +// RUN: llvm-dis | grep {@nate.*internal unnamed_addr global i32 0} struct X { int *XX; int Y;}; Modified: llvm/trunk/test/FrontendC/2007-02-16-WritableStrings.c URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC/2007-02-16-WritableStrings.c?rev=123447&r1=123446&r2=123447&view=diff ============================================================================== --- llvm/trunk/test/FrontendC/2007-02-16-WritableStrings.c (original) +++ llvm/trunk/test/FrontendC/2007-02-16-WritableStrings.c Fri Jan 14 11:01:20 2011 @@ -1,7 +1,7 @@ // Test the -fwritable-strings option. // RUN: %llvmgcc -O3 -S -o - -fwritable-strings %s | \ -// RUN: grep {internal global} -// RUN: %llvmgcc -O3 -S -o - %s | grep {private constant} +// RUN: grep {internal unnamed_addr global} +// RUN: %llvmgcc -O3 -S -o - %s | grep {private unnamed_addr constant} char *X = "foo"; Modified: llvm/trunk/test/FrontendC/hidden-visibility.c URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC/hidden-visibility.c?rev=123447&r1=123446&r2=123447&view=diff ============================================================================== --- llvm/trunk/test/FrontendC/hidden-visibility.c (original) +++ llvm/trunk/test/FrontendC/hidden-visibility.c Fri Jan 14 11:01:20 2011 @@ -1,3 +1,3 @@ -// RUN: %llvmgcc %s -S -o - | grep {hidden global} +// RUN: %llvmgcc %s -S -o - | grep {hidden unnamed_addr global} int X __attribute__ ((__visibility__ ("hidden"))) = 123; Modified: llvm/trunk/test/FrontendC/pr3518.c URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC/pr3518.c?rev=123447&r1=123446&r2=123447&view=diff ============================================================================== --- llvm/trunk/test/FrontendC/pr3518.c (original) +++ llvm/trunk/test/FrontendC/pr3518.c Fri Jan 14 11:01:20 2011 @@ -1,4 +1,4 @@ -// RUN: %llvmgcc %s -S -O0 -o - | grep {= internal global} | count 4 +// RUN: %llvmgcc %s -S -O0 -o - | grep {= internal unnamed_addr global} | count 4 // PR 3518 // Some of the objects were coming out as unintialized (external) before 3518 // was fixed. Internal names are different between llvm-gcc and clang so they Modified: llvm/trunk/test/FrontendC/pr4349.c URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/FrontendC/pr4349.c?rev=123447&r1=123446&r2=123447&view=diff ============================================================================== --- llvm/trunk/test/FrontendC/pr4349.c (original) +++ llvm/trunk/test/FrontendC/pr4349.c Fri Jan 14 11:01:20 2011 @@ -16,22 +16,22 @@ { void *ptr; }; -// CHECK: @svars1 = global [1 x %struct.svar] [%struct.svar { i8* bitcast (%struct.cpu* @cpu to i8*) }] +// CHECK: @svars1 = unnamed_addr global [1 x %struct.svar] [%struct.svar { i8* bitcast (%struct.cpu* @cpu to i8*) }] struct svar svars1[] = { { &((cpu.pc).w[0]) } }; -// CHECK: @svars2 = global [1 x %struct.svar] [%struct.svar { i8* getelementptr ([2 x i8]* bitcast (%struct.cpu* @cpu to [2 x i8]*), i{{[0-9]+}} 0, i{{[0-9]+}} 1) }] +// CHECK: @svars2 = unnamed_addr global [1 x %struct.svar] [%struct.svar { i8* getelementptr ([2 x i8]* bitcast (%struct.cpu* @cpu to [2 x i8]*), i{{[0-9]+}} 0, i{{[0-9]+}} 1) }] struct svar svars2[] = { { &((cpu.pc).b[0][1]) } }; -// CHECK: @svars3 = global [1 x %struct.svar] [%struct.svar { i8* bitcast (i16* getelementptr ([2 x i16]* bitcast (%struct.cpu* @cpu to [2 x i16]*), i{{[0-9]+}} 0, i{{[0-9]+}} 1) to i8*) }] +// CHECK: @svars3 = unnamed_addr global [1 x %struct.svar] [%struct.svar { i8* bitcast (i16* getelementptr ([2 x i16]* bitcast (%struct.cpu* @cpu to [2 x i16]*), i{{[0-9]+}} 0, i{{[0-9]+}} 1) to i8*) }] struct svar svars3[] = { { &((cpu.pc).w[1]) } }; -// CHECK: @svars4 = global [1 x %struct.svar] [%struct.svar { i8* getelementptr ([2 x [2 x i8]]* bitcast (%struct.cpu* @cpu to [2 x [2 x i8]]*), i{{[0-9]+}} 0, i{{[0-9]+}} 1, i{{[0-9]+}} 1) }] +// CHECK: @svars4 = unnamed_addr global [1 x %struct.svar] [%struct.svar { i8* getelementptr ([2 x [2 x i8]]* bitcast (%struct.cpu* @cpu to [2 x [2 x i8]]*), i{{[0-9]+}} 0, i{{[0-9]+}} 1, i{{[0-9]+}} 1) }] struct svar svars4[] = { { &((cpu.pc).b[1][1]) } From criswell at uiuc.edu Fri Jan 14 11:24:15 2011 From: criswell at uiuc.edu (John Criswell) Date: Fri, 14 Jan 2011 17:24:15 -0000 Subject: [llvm-commits] [poolalloc] r123450 - /poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp Message-ID: <20110114172415.B67DD2A6C12C@llvm.org> Author: criswell Date: Fri Jan 14 11:24:15 2011 New Revision: 123450 URL: http://llvm.org/viewvc/llvm-project?rev=123450&view=rev Log: Disabled code using preprocessor macros instead of comments. Modified: poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp Modified: poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp URL: http://llvm.org/viewvc/llvm-project/poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp?rev=123450&r1=123449&r2=123450&view=diff ============================================================================== --- poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp (original) +++ poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp Fri Jan 14 11:24:15 2011 @@ -996,14 +996,16 @@ #ifndef NDEBUG // Verify that all potential callees at call site have the same DS graph. - /*DSCallGraph::callee_iterator E = PAInfo.getCallGraph().callee_end(OrigInst); +#if 0 + DSCallGraph::callee_iterator E = PAInfo.getCallGraph().callee_end(OrigInst); for (; I != E; ++I) { const Function * F = *I; assert (F); if (!(F)->isDeclaration()) assert(CalleeGraph == Graphs.getDSGraph(**I) && "Callees at call site do not have a common graph!"); - }*/ + } +#endif #endif // Find the DS nodes for the arguments that need to be added, if any. From clattner at apple.com Fri Jan 14 12:25:17 2011 From: clattner at apple.com (Chris Lattner) Date: Fri, 14 Jan 2011 10:25:17 -0800 Subject: [llvm-commits] [PATCH] remove OperandTraits<>::Layout In-Reply-To: References: Message-ID: <043DDEDD-606A-4256-93C2-24ACED14BD18@apple.com> On Jan 14, 2011, at 8:15 AM, Jay Foad wrote: > OperandTraits<>::Layout doesn't seem to be used for anything. How > about removing it? > > Tested with "make check". OK to commit? Seems obvious, please do! -Chris From clattner at apple.com Fri Jan 14 12:26:00 2011 From: clattner at apple.com (Chris Lattner) Date: Fri, 14 Jan 2011 10:26:00 -0800 Subject: [llvm-commits] [llvm] r123405 - /llvm/trunk/lib/Target/README.txt In-Reply-To: <4D306D04.8020205@gmail.com> References: <20110113220815.5EA392A6C12C@llvm.org> <4D306D04.8020205@gmail.com> Message-ID: <3EF93CAD-E06D-4ED1-81E4-626B3B6F1D4B@apple.com> On Jan 14, 2011, at 7:34 AM, Rafael ?vila de Esp?ndola wrote: > On 11-01-13 5:08 PM, Chris Lattner wrote: >> Author: lattner >> Date: Thu Jan 13 16:08:15 2011 >> New Revision: 123405 >> >> URL: http://llvm.org/viewvc/llvm-project?rev=123405&view=rev >> Log: >> memcpy + metadata = bliss :) > > > Would adding stores of undefs to the holes have the same effect? It's a lot harder for memcpy lowering and SRoA to usefully use that, also stores of undef get removed. -Chris From fvbommel at gmail.com Fri Jan 14 12:28:42 2011 From: fvbommel at gmail.com (Frits van Bommel) Date: Fri, 14 Jan 2011 19:28:42 +0100 Subject: [llvm-commits] [llvm] r123405 - /llvm/trunk/lib/Target/README.txt In-Reply-To: <4D306D04.8020205@gmail.com> References: <20110113220815.5EA392A6C12C@llvm.org> <4D306D04.8020205@gmail.com> Message-ID: 2011/1/14 Rafael ?vila de Esp?ndola : > On 11-01-13 5:08 PM, Chris Lattner wrote: >> memcpy + metadata = bliss :) > > Would adding stores of undefs to the holes have the same effect? (This was discussed on IRC a few days ago) Stores of undefs are deleted by e.g. -instcombine, long before -scalarrepl ever sees them. Metadata is much more likely to survive until it's needed. From jay.foad at gmail.com Fri Jan 14 12:41:56 2011 From: jay.foad at gmail.com (Jay Foad) Date: Fri, 14 Jan 2011 18:41:56 -0000 Subject: [llvm-commits] [llvm] r123452 - in /llvm/trunk/include/llvm: OperandTraits.h User.h Message-ID: <20110114184156.DFBF52A6C12C@llvm.org> Author: foad Date: Fri Jan 14 12:41:56 2011 New Revision: 123452 URL: http://llvm.org/viewvc/llvm-project?rev=123452&view=rev Log: OperandTraits<>::Layout isn't used for anything. Remove it. Modified: llvm/trunk/include/llvm/OperandTraits.h llvm/trunk/include/llvm/User.h Modified: llvm/trunk/include/llvm/OperandTraits.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/OperandTraits.h?rev=123452&r1=123451&r2=123452&view=diff ============================================================================== --- llvm/trunk/include/llvm/OperandTraits.h (original) +++ llvm/trunk/include/llvm/OperandTraits.h Fri Jan 14 12:41:56 2011 @@ -38,16 +38,6 @@ static unsigned operands(const User*) { return ARITY; } - struct prefix { - Use Ops[ARITY]; - prefix(); // DO NOT IMPLEMENT - }; - template - struct Layout { - struct overlay : public prefix, public U { - overlay(); // DO NOT IMPLEMENT - }; - }; }; //===----------------------------------------------------------------------===// Modified: llvm/trunk/include/llvm/User.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/User.h?rev=123452&r1=123451&r2=123452&view=diff ============================================================================== --- llvm/trunk/include/llvm/User.h (original) +++ llvm/trunk/include/llvm/User.h Fri Jan 14 12:41:56 2011 @@ -37,10 +37,6 @@ static inline Use *op_begin(User*); static inline Use *op_end(User*); static inline unsigned operands(const User*); - template - struct Layout { - typedef U overlay; - }; }; class User : public Value { From aggarwa4 at illinois.edu Fri Jan 14 13:04:26 2011 From: aggarwa4 at illinois.edu (Arushi Aggarwal) Date: Fri, 14 Jan 2011 19:04:26 -0000 Subject: [llvm-commits] [poolalloc] r123453 - in /poolalloc/trunk/lib/PoolAllocate: AllNodesHeuristic.cpp Heuristic.cpp Message-ID: <20110114190426.226972A6C12C@llvm.org> Author: aggarwa4 Date: Fri Jan 14 13:04:25 2011 New Revision: 123453 URL: http://llvm.org/viewvc/llvm-project?rev=123453&view=rev Log: Do not poolallocate nodes that are marked PtrToInt. These pointers could escape to external code. Modified: poolalloc/trunk/lib/PoolAllocate/AllNodesHeuristic.cpp poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp Modified: poolalloc/trunk/lib/PoolAllocate/AllNodesHeuristic.cpp URL: http://llvm.org/viewvc/llvm-project/poolalloc/trunk/lib/PoolAllocate/AllNodesHeuristic.cpp?rev=123453&r1=123452&r2=123453&view=diff ============================================================================== --- poolalloc/trunk/lib/PoolAllocate/AllNodesHeuristic.cpp (original) +++ poolalloc/trunk/lib/PoolAllocate/AllNodesHeuristic.cpp Fri Jan 14 13:04:25 2011 @@ -77,14 +77,14 @@ for (DenseSet::iterator I = NodesFromGlobals.begin(), E = NodesFromGlobals.end(); I != E; ) { DenseSet::iterator Last = I; ++I; - // - // Nodes that escape to external code could be reachable from globals. - // Nodes that are incomplete could be heap nodes. - // Unknown nodes could be anything. - // + const DSNode *tmp = *Last; if (!(tmp->isHeapNode())) toRemove.push_back (tmp); + // Do not poolallocate nodes that are cast to Int. + // As we do not track through ints, these could be escaping + if (tmp->isPtrToIntNode()) + toRemove.push_back(tmp); } // @@ -182,14 +182,6 @@ //assert (!GGN || GlobalHeapNodes.count (GGN)); if (GGN && GlobalHeapNodes.count (GGN)) PoolMap[GGN].NodesInPool.push_back (N); - else if (N->isHeapNode() && N->isPtrToIntNode()){ - //FIXME: This is needed to fix failures in 164.gzip and - // 197.parser. I am not fully sure this is the right fix. - if( !N->isAllocaNode() && !N->isUnknownNode()) { - PoolMap[N]= OnePool(N); - GlobalHeapNodes.insert(N); - } - } } } } Modified: poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp URL: http://llvm.org/viewvc/llvm-project/poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp?rev=123453&r1=123452&r2=123453&view=diff ============================================================================== --- poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp (original) +++ poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp Fri Jan 14 13:04:25 2011 @@ -429,8 +429,10 @@ if (!(GlobalPoolNodes.count (N) || GlobalPoolNodes.count (GGN))) { // Otherwise, if it was not passed in from outside the function, it must // be a local pool! - assert(!N->isGlobalNode() && "Should be in global mapping!"); - Nodes.push_back (N); + assert(!N->isGlobalNode() || N->isPtrToIntNode() && "Should be in global mapping!"); + if(!N->isPtrToIntNode()) { + Nodes.push_back (N); + } } } } From criswell at cs.uiuc.edu Fri Jan 14 13:22:20 2011 From: criswell at cs.uiuc.edu (John Criswell) Date: Fri, 14 Jan 2011 13:22:20 -0600 Subject: [llvm-commits] CVS: llvm-www/safecode/people.html funding.html Message-ID: <201101141922.p0EJMK2F013850@zion.cs.uiuc.edu> Changes in directory llvm-www/safecode: people.html updated: 1.7 -> 1.8 funding.html updated: 1.2 -> 1.3 --- Log message: Added Aman. Resorted funding sources to be in roughly reverse chronological order. --- Diffs of the changes: (+10 -6) funding.html | 12 ++++++------ people.html | 4 ++++ 2 files changed, 10 insertions(+), 6 deletions(-) Index: llvm-www/safecode/people.html diff -u llvm-www/safecode/people.html:1.7 llvm-www/safecode/people.html:1.8 --- llvm-www/safecode/people.html:1.7 Fri Jan 14 10:57:15 2011 +++ llvm-www/safecode/people.html Fri Jan 14 13:21:18 2011 @@ -111,6 +111,10 @@

Undergraduate Student Alumni

+ Aman Bhatia +
Billy Lau

- National Science Foundation Grants CNS 07-16768 and CNS 07-09122 -
- A gift from Apple Computer, Inc. + IARPA/AFRL Contract# FA8650-10-C-7022
@@ -69,7 +65,11 @@
- IARPA/AFRL Contract# FA8650-10-C-7022 + A gift from Apple Computer, Inc. +
+ National Science Foundation Grants CNS 07-16768 and CNS 07-09122

&Insts, + SSAUpdater *SSA = 0); + + + /// Return true if the specified instruction is in the Inst list (which was + /// passed into the run method). Clients should implement this with a more + /// efficient version if possible. + virtual bool isInstInList(Instruction *I, + const SmallVectorImpl &Insts) const { + for (unsigned i = 0, e = Insts.size(); i != e; ++i) + if (Insts[i] == I) + return true; + return false; + } +}; } // End llvm namespace Modified: llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp?rev=123456&r1=123455&r2=123456&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp (original) +++ llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp Fri Jan 14 13:36:13 2011 @@ -343,3 +343,157 @@ SSAUpdaterImpl Impl(this, &AvailableVals, InsertedPHIs); return Impl.GetValue(BB); } + +//===----------------------------------------------------------------------===// +// LoadAndStorePromoter Implementation +//===----------------------------------------------------------------------===// + +void LoadAndStorePromoter::run(StringRef BaseName, + const SmallVectorImpl &Insts, + SSAUpdater *SSA) { + if (Insts.empty()) return; + + // If no SSAUpdater was provided, use a default one. This allows the client + // to capture inserted PHI nodes etc if they want. + SSAUpdater DefaultSSA; + if (SSA == 0) SSA = &DefaultSSA; + + const Type *ValTy; + if (LoadInst *LI = dyn_cast(Insts[0])) + ValTy = LI->getType(); + else + ValTy = cast(Insts[0])->getOperand(0)->getType(); + + SSA->Initialize(ValTy, BaseName); + + // First step: bucket up uses of the alloca by the block they occur in. + // This is important because we have to handle multiple defs/uses in a block + // ourselves: SSAUpdater is purely for cross-block references. + // FIXME: Want a TinyVector since there is often 0/1 element. + DenseMap > UsesByBlock; + + for (unsigned i = 0, e = Insts.size(); i != e; ++i) { + Instruction *User = Insts[i]; + UsesByBlock[User->getParent()].push_back(User); + } + + // Okay, now we can iterate over all the blocks in the function with uses, + // processing them. Keep track of which loads are loading a live-in value. + // Walk the uses in the use-list order to be determinstic. + SmallVector LiveInLoads; + DenseMap ReplacedLoads; + + for (unsigned i = 0, e = Insts.size(); i != e; ++i) { + Instruction *User = Insts[i]; + BasicBlock *BB = User->getParent(); + std::vector &BlockUses = UsesByBlock[BB]; + + // If this block has already been processed, ignore this repeat use. + if (BlockUses.empty()) continue; + + // Okay, this is the first use in the block. If this block just has a + // single user in it, we can rewrite it trivially. + if (BlockUses.size() == 1) { + // If it is a store, it is a trivial def of the value in the block. + if (StoreInst *SI = dyn_cast(User)) + SSA->AddAvailableValue(BB, SI->getOperand(0)); + else + // Otherwise it is a load, queue it to rewrite as a live-in load. + LiveInLoads.push_back(cast(User)); + BlockUses.clear(); + continue; + } + + // Otherwise, check to see if this block is all loads. + bool HasStore = false; + for (unsigned i = 0, e = BlockUses.size(); i != e; ++i) { + if (isa(BlockUses[i])) { + HasStore = true; + break; + } + } + + // If so, we can queue them all as live in loads. We don't have an + // efficient way to tell which on is first in the block and don't want to + // scan large blocks, so just add all loads as live ins. + if (!HasStore) { + for (unsigned i = 0, e = BlockUses.size(); i != e; ++i) + LiveInLoads.push_back(cast(BlockUses[i])); + BlockUses.clear(); + continue; + } + + // Otherwise, we have mixed loads and stores (or just a bunch of stores). + // Since SSAUpdater is purely for cross-block values, we need to determine + // the order of these instructions in the block. If the first use in the + // block is a load, then it uses the live in value. The last store defines + // the live out value. We handle this by doing a linear scan of the block. + Value *StoredValue = 0; + for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) { + if (LoadInst *L = dyn_cast(II)) { + // If this is a load from an unrelated pointer, ignore it. + if (!isInstInList(L, Insts)) continue; + + // If we haven't seen a store yet, this is a live in use, otherwise + // use the stored value. + if (StoredValue) { + L->replaceAllUsesWith(StoredValue); + ReplacedLoads[L] = StoredValue; + } else { + LiveInLoads.push_back(L); + } + continue; + } + + if (StoreInst *S = dyn_cast(II)) { + // If this is a store to an unrelated pointer, ignore it. + if (!isInstInList(S, Insts)) continue; + + // Remember that this is the active value in the block. + StoredValue = S->getOperand(0); + } + } + + // The last stored value that happened is the live-out for the block. + assert(StoredValue && "Already checked that there is a store in block"); + SSA->AddAvailableValue(BB, StoredValue); + BlockUses.clear(); + } + + // Okay, now we rewrite all loads that use live-in values in the loop, + // inserting PHI nodes as necessary. + for (unsigned i = 0, e = LiveInLoads.size(); i != e; ++i) { + LoadInst *ALoad = LiveInLoads[i]; + Value *NewVal = SSA->GetValueInMiddleOfBlock(ALoad->getParent()); + ALoad->replaceAllUsesWith(NewVal); + ReplacedLoads[ALoad] = NewVal; + } + + // Now that everything is rewritten, delete the old instructions from the + // function. They should all be dead now. + for (unsigned i = 0, e = Insts.size(); i != e; ++i) { + Instruction *User = Insts[i]; + + // If this is a load that still has uses, then the load must have been added + // as a live value in the SSAUpdate data structure for a block (e.g. because + // the loaded value was stored later). In this case, we need to recursively + // propagate the updates until we get to the real value. + if (!User->use_empty()) { + Value *NewVal = ReplacedLoads[User]; + assert(NewVal && "not a replaced load?"); + + // Propagate down to the ultimate replacee. The intermediately loads + // could theoretically already have been deleted, so we don't want to + // dereference the Value*'s. + DenseMap::iterator RLI = ReplacedLoads.find(NewVal); + while (RLI != ReplacedLoads.end()) { + NewVal = RLI->second; + RLI = ReplacedLoads.find(NewVal); + } + + User->replaceAllUsesWith(NewVal); + } + + User->eraseFromParent(); + } +} From sabre at nondot.org Fri Jan 14 13:50:47 2011 From: sabre at nondot.org (Chris Lattner) Date: Fri, 14 Jan 2011 19:50:47 -0000 Subject: [llvm-commits] [llvm] r123457 - /llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp Message-ID: <20110114195047.C6F2E2A6C12C@llvm.org> Author: lattner Date: Fri Jan 14 13:50:47 2011 New Revision: 123457 URL: http://llvm.org/viewvc/llvm-project?rev=123457&view=rev Log: switch SRoA to use LoadAndStorePromoter instead of its own copy of the code. Modified: llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp Modified: llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp?rev=123457&r1=123456&r2=123457&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp Fri Jan 14 13:50:47 2011 @@ -840,144 +840,35 @@ return Changed; } -/// PromoteAlloca - Promote an alloca to registers, using SSAUpdater. -static void PromoteAlloca(AllocaInst *AI, SSAUpdater &SSA) { - SSA.Initialize(AI->getType()->getElementType(), AI->getName()); - - // First step: bucket up uses of the alloca by the block they occur in. - // This is important because we have to handle multiple defs/uses in a block - // ourselves: SSAUpdater is purely for cross-block references. - // FIXME: Want a TinyVector since there is often 0/1 element. - DenseMap > UsesByBlock; +namespace { +class AllocaPromoter : public LoadAndStorePromoter { + AllocaInst *AI; +public: + AllocaPromoter() : AI(0) {} - for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); - UI != E; ++UI) { - Instruction *User = cast(*UI); - UsesByBlock[User->getParent()].push_back(User); - } - - // Okay, now we can iterate over all the blocks in the function with uses, - // processing them. Keep track of which loads are loading a live-in value. - // Walk the uses in the use-list order to be determinstic. - SmallVector LiveInLoads; - DenseMap ReplacedLoads; - - for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); - UI != E; ++UI) { - Instruction *User = cast(*UI); - BasicBlock *BB = User->getParent(); - std::vector &BlockUses = UsesByBlock[BB]; - - // If this block has already been processed, ignore this repeat use. - if (BlockUses.empty()) continue; - - // Okay, this is the first use in the block. If this block just has a - // single user in it, we can rewrite it trivially. - if (BlockUses.size() == 1) { - // If it is a store, it is a trivial def of the value in the block. - if (StoreInst *SI = dyn_cast(User)) - SSA.AddAvailableValue(BB, SI->getOperand(0)); - else - // Otherwise it is a load, queue it to rewrite as a live-in load. - LiveInLoads.push_back(cast(User)); - BlockUses.clear(); - continue; - } - - // Otherwise, check to see if this block is all loads. - bool HasStore = false; - for (unsigned i = 0, e = BlockUses.size(); i != e; ++i) { - if (isa(BlockUses[i])) { - HasStore = true; - break; - } - } - - // If so, we can queue them all as live in loads. We don't have an - // efficient way to tell which on is first in the block and don't want to - // scan large blocks, so just add all loads as live ins. - if (!HasStore) { - for (unsigned i = 0, e = BlockUses.size(); i != e; ++i) - LiveInLoads.push_back(cast(BlockUses[i])); - BlockUses.clear(); - continue; - } + void run(AllocaInst *AI, SSAUpdater &SSA) { + // Remember which alloca we're promoting (for isInstInList). + this->AI = AI; + + // Build the list of instructions to promote. + SmallVector Insts; + for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); + UI != E; ++UI) + Insts.push_back(cast(*UI)); - // Otherwise, we have mixed loads and stores (or just a bunch of stores). - // Since SSAUpdater is purely for cross-block values, we need to determine - // the order of these instructions in the block. If the first use in the - // block is a load, then it uses the live in value. The last store defines - // the live out value. We handle this by doing a linear scan of the block. - Value *StoredValue = 0; - for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) { - if (LoadInst *L = dyn_cast(II)) { - // If this is a load from an unrelated pointer, ignore it. - if (L->getOperand(0) != AI) continue; - - // If we haven't seen a store yet, this is a live in use, otherwise - // use the stored value. - if (StoredValue) { - L->replaceAllUsesWith(StoredValue); - ReplacedLoads[L] = StoredValue; - } else { - LiveInLoads.push_back(L); - } - continue; - } - - if (StoreInst *S = dyn_cast(II)) { - // If this is a store to an unrelated pointer, ignore it. - if (S->getPointerOperand() != AI) continue; - - // Remember that this is the active value in the block. - StoredValue = S->getOperand(0); - } - } + LoadAndStorePromoter::run(AI->getName(), Insts, &SSA); - // The last stored value that happened is the live-out for the block. - assert(StoredValue && "Already checked that there is a store in block"); - SSA.AddAvailableValue(BB, StoredValue); - BlockUses.clear(); + AI->eraseFromParent(); } - // Okay, now we rewrite all loads that use live-in values in the loop, - // inserting PHI nodes as necessary. - for (unsigned i = 0, e = LiveInLoads.size(); i != e; ++i) { - LoadInst *ALoad = LiveInLoads[i]; - Value *NewVal = SSA.GetValueInMiddleOfBlock(ALoad->getParent()); - ALoad->replaceAllUsesWith(NewVal); - ReplacedLoads[ALoad] = NewVal; + virtual bool isInstInList(Instruction *I, + const SmallVectorImpl &Insts) const { + if (LoadInst *LI = dyn_cast(I)) + return LI->getOperand(0) == AI; + return cast(I)->getPointerOperand() == AI; } - - // Now that everything is rewritten, delete the old instructions from the - // function. They should all be dead now. - for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E; ) { - Instruction *User = cast(*UI++); - - // If this is a load that still has uses, then the load must have been added - // as a live value in the SSAUpdate data structure for a block (e.g. because - // the loaded value was stored later). In this case, we need to recursively - // propagate the updates until we get to the real value. - if (!User->use_empty()) { - Value *NewVal = ReplacedLoads[User]; - assert(NewVal && "not a replaced load?"); - - // Propagate down to the ultimate replacee. The intermediately loads - // could theoretically already have been deleted, so we don't want to - // dereference the Value*'s. - DenseMap::iterator RLI = ReplacedLoads.find(NewVal); - while (RLI != ReplacedLoads.end()) { - NewVal = RLI->second; - RLI = ReplacedLoads.find(NewVal); - } - - User->replaceAllUsesWith(NewVal); - } - - User->eraseFromParent(); - } -} - +}; +} // end anon namespace bool SROA::performPromotion(Function &F) { std::vector Allocas; @@ -1008,10 +899,9 @@ PromoteMemToReg(Allocas, *DT, *DF); else { SSAUpdater SSA; - for (unsigned i = 0, e = Allocas.size(); i != e; ++i) { - PromoteAlloca(Allocas[i], SSA); - Allocas[i]->eraseFromParent(); - } + AllocaPromoter Promoter; + for (unsigned i = 0, e = Allocas.size(); i != e; ++i) + Promoter.run(Allocas[i], SSA); } NumPromoted += Allocas.size(); Changed = true; From jasonwkim at google.com Fri Jan 14 14:20:37 2011 From: jasonwkim at google.com (Jason Kim) Date: Fri, 14 Jan 2011 12:20:37 -0800 Subject: [llvm-commits] [llvm] r123424 - in /llvm/trunk: include/llvm/Object/MachOFormat.h lib/MC/ELFObjectWriter.cpp lib/MC/MachObjectWriter.cpp lib/Target/ARM/ARMAsmBackend.cpp lib/Target/ARM/ARMFixupKinds.h lib/Target/ARM/ARMMCCodeEmitter.cpp test/MC/A Message-ID: On Thu, Jan 13, 2011 at 6:38 PM, Evan Cheng wrote: > Author: evancheng > Date: Thu Jan 13 20:38:49 2011 > New Revision: 123424 > > URL: http://llvm.org/viewvc/llvm-project?rev=123424&view=rev > Log: > Completed :lower16: / :upper16: support for movw / movt pairs on Darwin. > - Fixed :upper16: fix up routine. It should be shifting down the top 16 bits first. > - Added support for Thumb2 :lower16: and :upper16: fix up. > - Added :upper16: and :lower16: relocation support to mach-o object writer. > > Modified: > ? ?llvm/trunk/include/llvm/Object/MachOFormat.h > ? ?llvm/trunk/lib/MC/ELFObjectWriter.cpp > ? ?llvm/trunk/lib/MC/MachObjectWriter.cpp > ? ?llvm/trunk/lib/Target/ARM/ARMAsmBackend.cpp > ? ?llvm/trunk/lib/Target/ARM/ARMFixupKinds.h > ? ?llvm/trunk/lib/Target/ARM/ARMMCCodeEmitter.cpp > ? ?llvm/trunk/test/MC/ARM/elf-movt.s > ? ?llvm/trunk/test/MC/ARM/hilo-16bit-relocations.s > > Modified: llvm/trunk/include/llvm/Object/MachOFormat.h > URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Object/MachOFormat.h?rev=123424&r1=123423&r2=123424&view=diff > ============================================================================== > --- llvm/trunk/include/llvm/Object/MachOFormat.h (original) > +++ llvm/trunk/include/llvm/Object/MachOFormat.h Thu Jan 13 20:38:49 2011 > @@ -353,7 +353,10 @@ > ? ? RIT_ARM_PreboundLazyPointer = 4, > ? ? RIT_ARM_Branch24Bit = 5, > ? ? RIT_ARM_ThumbBranch22Bit = 6, > - ? ?RIT_ARM_ThumbBranch32Bit = 7 > + ? ?RIT_ARM_ThumbBranch32Bit = 7, > + ? ?RIT_ARM_Half = 8, > + ? ?RIT_ARM_HalfDifference = 9 > + > ? }; > > ?} // end namespace macho > > Modified: llvm/trunk/lib/MC/ELFObjectWriter.cpp > URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/MC/ELFObjectWriter.cpp?rev=123424&r1=123423&r2=123424&view=diff > ============================================================================== > --- llvm/trunk/lib/MC/ELFObjectWriter.cpp (original) > +++ llvm/trunk/lib/MC/ELFObjectWriter.cpp Thu Jan 13 20:38:49 2011 > @@ -1508,6 +1508,14 @@ > ? ? case ARM::fixup_arm_movw_lo16_pcrel: > ? ? ? Type = ELF::R_ARM_MOVW_PREL_NC; > ? ? ? break; > + ? ?case ARM::fixup_t2_movt_hi16: > + ? ?case ARM::fixup_t2_movt_hi16_pcrel: > + ? ? ?Type = ELF::R_ARM_THM_MOVT_PREL; > + ? ? ?break; > + ? ?case ARM::fixup_t2_movw_lo16: > + ? ?case ARM::fixup_t2_movw_lo16_pcrel: > + ? ? ?Type = ELF::R_ARM_THM_MOVW_PREL_NC; > + ? ? ?break; > ? ? } > ? } else { > ? ? switch ((unsigned)Fixup.getKind()) { > @@ -1555,6 +1563,12 @@ > ? ? case ARM::fixup_arm_movw_lo16: > ? ? ? Type = ELF::R_ARM_MOVW_ABS_NC; > ? ? ? break; > + ? ?case ARM::fixup_t2_movt_hi16: > + ? ? ?Type = ELF::R_ARM_THM_MOVT_ABS; > + ? ? ?break; > + ? ?case ARM::fixup_t2_movw_lo16: > + ? ? ?Type = ELF::R_ARM_THM_MOVW_ABS_NC; > + ? ? ?break; > ? ? } > ? } > > > Modified: llvm/trunk/lib/MC/MachObjectWriter.cpp > URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/MC/MachObjectWriter.cpp?rev=123424&r1=123423&r2=123424&view=diff > ============================================================================== > --- llvm/trunk/lib/MC/MachObjectWriter.cpp (original) > +++ llvm/trunk/lib/MC/MachObjectWriter.cpp Thu Jan 13 20:38:49 2011 > @@ -656,10 +656,10 @@ > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?const MCAsmLayout &Layout, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?const MCFragment *Fragment, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?const MCFixup &Fixup, MCValue Target, > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? unsigned Log2Size, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?uint64_t &FixedValue) { > ? ? uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset(); > ? ? unsigned IsPCRel = isFixupKindPCRel(Asm, Fixup.getKind()); > - ? ?unsigned Log2Size = getFixupKindLog2Size(Fixup.getKind()); > ? ? unsigned Type = macho::RIT_Vanilla; > > ? ? // See . > @@ -720,10 +720,10 @@ > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? const MCAsmLayout &Layout, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? const MCFragment *Fragment, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? const MCFixup &Fixup, MCValue Target, > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?unsigned Log2Size, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? uint64_t &FixedValue) { > ? ? uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset(); > ? ? unsigned IsPCRel = isFixupKindPCRel(Asm, Fixup.getKind()); > - ? ?unsigned Log2Size = getFixupKindLog2Size(Fixup.getKind()); > ? ? unsigned Type = macho::RIT_Vanilla; > > ? ? // See . > @@ -775,6 +775,100 @@ > ? ? Relocations[Fragment->getParent()].push_back(MRE); > ? } > > + ?void RecordARMMovwMovtRelocation(const MCAssembler &Asm, > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? const MCAsmLayout &Layout, > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? const MCFragment *Fragment, > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? const MCFixup &Fixup, MCValue Target, > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? uint64_t &FixedValue) { > + ? ?uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset(); > + ? ?unsigned IsPCRel = isFixupKindPCRel(Asm, Fixup.getKind()); > + ? ?unsigned Type = macho::RIT_ARM_Half; > + > + ? ?// See . > + ? ?const MCSymbol *A = &Target.getSymA()->getSymbol(); > + ? ?MCSymbolData *A_SD = &Asm.getSymbolData(*A); > + > + ? ?if (!A_SD->getFragment()) > + ? ? ?report_fatal_error("symbol '" + A->getName() + > + ? ? ? ? ? ? ? ? ? ? ? ?"' can not be undefined in a subtraction expression"); > + > + ? ?uint32_t Value = getSymbolAddress(A_SD, Layout); > + ? ?uint32_t Value2 = 0; > + ? ?uint64_t SecAddr = getSectionAddress(A_SD->getFragment()->getParent()); > + ? ?FixedValue += SecAddr; > + > + ? ?if (const MCSymbolRefExpr *B = Target.getSymB()) { > + ? ? ?MCSymbolData *B_SD = &Asm.getSymbolData(B->getSymbol()); > + > + ? ? ?if (!B_SD->getFragment()) > + ? ? ? ?report_fatal_error("symbol '" + B->getSymbol().getName() + > + ? ? ? ? ? ? ? ? ? ? ? ? ?"' can not be undefined in a subtraction expression"); > + > + ? ? ?// Select the appropriate difference relocation type. > + ? ? ?Type = macho::RIT_ARM_HalfDifference; > + ? ? ?Value2 = getSymbolAddress(B_SD, Layout); > + ? ? ?FixedValue -= getSectionAddress(B_SD->getFragment()->getParent()); > + ? ?} > + > + ? ?// Relocations are written out in reverse order, so the PAIR comes first. > + ? ?// ARM_RELOC_HALF and ARM_RELOC_HALF_SECTDIFF abuse the r_length field: > + ? ?// > + ? ?// For these two r_type relocations they always have a pair following them > + ? ?// and the r_length bits are used differently. ?The encoding of the > + ? ?// r_length is as follows: > + ? ?// low bit of r_length: > + ? ?// ?0 - :lower16: for movw instructions > + ? ?// ?1 - :upper16: for movt instructions > + ? ?// high bit of r_length: > + ? ?// ?0 - arm instructions > + ? ?// ?1 - thumb instructions > + ? ?// the other half of the relocated expression is in the following pair > + ? ?// relocation entry in the the low 16 bits of r_address field. > + ? ?unsigned ThumbBit = 0; > + ? ?unsigned MovtBit = 0; > + ? ?switch (Fixup.getKind()) { > + ? ?default: break; > + ? ?case ARM::fixup_arm_movt_hi16: > + ? ?case ARM::fixup_arm_movt_hi16_pcrel: > + ? ? ?MovtBit = 1; > + ? ? ?break; > + ? ?case ARM::fixup_t2_movt_hi16: > + ? ?case ARM::fixup_t2_movt_hi16_pcrel: > + ? ? ?MovtBit = 1; > + ? ? ?// Fallthrough > + ? ?case ARM::fixup_t2_movw_lo16: > + ? ?case ARM::fixup_t2_movw_lo16_pcrel: > + ? ? ?ThumbBit = 1; > + ? ? ?break; > + ? ?} > + > + > + ? ?if (Type == macho::RIT_ARM_HalfDifference) { > + ? ? ?uint32_t OtherHalf = MovtBit > + ? ? ? ?? (FixedValue & 0xffff) : ((FixedValue & 0xffff0000) >> 16); > + > + ? ? ?macho::RelocationEntry MRE; > + ? ? ?MRE.Word0 = ((OtherHalf ? ? ? << ?0) | > + ? ? ? ? ? ? ? ? ? (macho::RIT_Pair << 24) | > + ? ? ? ? ? ? ? ? ? (MovtBit ? ? ? ? << 28) | > + ? ? ? ? ? ? ? ? ? (ThumbBit ? ? ? ?<< 29) | > + ? ? ? ? ? ? ? ? ? (IsPCRel ? ? ? ? << 30) | > + ? ? ? ? ? ? ? ? ? macho::RF_Scattered); > + ? ? ?MRE.Word1 = Value2; > + ? ? ?Relocations[Fragment->getParent()].push_back(MRE); > + ? ?} > + > + ? ?macho::RelocationEntry MRE; > + ? ?MRE.Word0 = ((FixupOffset << ?0) | > + ? ? ? ? ? ? ? ? (Type ? ? ? ?<< 24) | > + ? ? ? ? ? ? ? ? (MovtBit ? ? << 28) | > + ? ? ? ? ? ? ? ? (ThumbBit ? ?<< 29) | > + ? ? ? ? ? ? ? ? (IsPCRel ? ? << 30) | > + ? ? ? ? ? ? ? ? macho::RF_Scattered); > + ? ?MRE.Word1 = Value; > + ? ?Relocations[Fragment->getParent()].push_back(MRE); > + ?} > + > ? void RecordTLVPRelocation(const MCAssembler &Asm, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? const MCAsmLayout &Layout, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? const MCFragment *Fragment, > @@ -868,6 +962,24 @@ > ? ? ? // Report as 'long', even though that is not quite accurate. > ? ? ? Log2Size = llvm::Log2_32(4); > ? ? ? return true; > + > + ? ?case ARM::fixup_arm_movt_hi16: > + ? ?case ARM::fixup_arm_movt_hi16_pcrel: > + ? ?case ARM::fixup_t2_movt_hi16: > + ? ?case ARM::fixup_t2_movt_hi16_pcrel: > + ? ? ?RelocType = unsigned(macho::RIT_ARM_HalfDifference); > + ? ? ?// Report as 'long', even though that is not quite accurate. > + ? ? ?Log2Size = llvm::Log2_32(4); > + ? ? ?return true; > + > + ? ?case ARM::fixup_arm_movw_lo16: > + ? ?case ARM::fixup_arm_movw_lo16_pcrel: > + ? ?case ARM::fixup_t2_movw_lo16: > + ? ?case ARM::fixup_t2_movw_lo16_pcrel: > + ? ? ?RelocType = unsigned(macho::RIT_ARM_Half); > + ? ? ?// Report as 'long', even though that is not quite accurate. > + ? ? ?Log2Size = llvm::Log2_32(4); > + ? ? ?return true; > ? ? } > ? } > ? void RecordARMRelocation(const MCAssembler &Asm, const MCAsmLayout &Layout, > @@ -884,9 +996,14 @@ > ? ? // If this is a difference or a defined symbol plus an offset, then we need > ? ? // a scattered relocation entry. ?Differences always require scattered > ? ? // relocations. > - ? ?if (Target.getSymB()) > - ? ? ? ?return RecordARMScatteredRelocation(Asm, Layout, Fragment, Fixup, > - ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?Target, FixedValue); > + ? ?if (Target.getSymB()) { > + ? ? ?if (RelocType == macho::RIT_ARM_Half || > + ? ? ? ? ?RelocType == macho::RIT_ARM_HalfDifference) > + ? ? ? ?return RecordARMMovwMovtRelocation(Asm, Layout, Fragment, Fixup, > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? Target, FixedValue); > + ? ? ?return RecordARMScatteredRelocation(Asm, Layout, Fragment, Fixup, > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?Target, Log2Size, FixedValue); > + ? ?} > > ? ? // Get the symbol data, if any. > ? ? MCSymbolData *SD = 0; > @@ -902,8 +1019,8 @@ > ? ? if (IsPCRel && RelocType == macho::RIT_Vanilla) > ? ? ? Offset += 1 << Log2Size; > ? ? if (Offset && SD && !doesSymbolRequireExternRelocation(SD)) > - ? ? ?return RecordARMScatteredRelocation(Asm, Layout, Fragment, Fixup, > - ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?Target, FixedValue); > + ? ? ?return RecordARMScatteredRelocation(Asm, Layout, Fragment, Fixup, Target, > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?Log2Size, FixedValue); > > ? ? // See . > ? ? uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset(); > @@ -986,7 +1103,7 @@ > ? ? // Differences always require scattered relocations. > ? ? if (Target.getSymB()) > ? ? ? ? return RecordScatteredRelocation(Asm, Layout, Fragment, Fixup, > - ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? Target, FixedValue); > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? Target, Log2Size, FixedValue); > > ? ? // Get the symbol data, if any. > ? ? MCSymbolData *SD = 0; > @@ -1000,7 +1117,7 @@ > ? ? ? Offset += 1 << Log2Size; > ? ? if (Offset && SD && !doesSymbolRequireExternRelocation(SD)) > ? ? ? return RecordScatteredRelocation(Asm, Layout, Fragment, Fixup, > - ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? Target, FixedValue); > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? Target, Log2Size, FixedValue); > > ? ? // See . > ? ? uint32_t FixupOffset = Layout.getFragmentOffset(Fragment)+Fixup.getOffset(); > > Modified: llvm/trunk/lib/Target/ARM/ARMAsmBackend.cpp > URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMAsmBackend.cpp?rev=123424&r1=123423&r2=123424&view=diff > ============================================================================== > --- llvm/trunk/lib/Target/ARM/ARMAsmBackend.cpp (original) > +++ llvm/trunk/lib/Target/ARM/ARMAsmBackend.cpp Thu Jan 13 20:38:49 2011 > @@ -76,10 +76,15 @@ > ?{ "fixup_arm_thumb_cb", ? ? ?0, ? ? ? ? ? ?16, ?MCFixupKindInfo::FKF_IsPCRel }, > ?{ "fixup_arm_thumb_cp", ? ? ?1, ? ? ? ? ? ? 8, ?MCFixupKindInfo::FKF_IsPCRel }, > ?{ "fixup_arm_thumb_bcc", ? ? 1, ? ? ? ? ? ? 8, ?MCFixupKindInfo::FKF_IsPCRel }, > -{ "fixup_arm_movt_hi16", ? ? 0, ? ? ? ? ? ?16, ?0 }, > -{ "fixup_arm_movw_lo16", ? ? 0, ? ? ? ? ? ?16, ?0 }, > -{ "fixup_arm_movt_hi16_pcrel", 0, ? ? ? ? ?16, ?MCFixupKindInfo::FKF_IsPCRel }, > -{ "fixup_arm_movw_lo16_pcrel", 0, ? ? ? ? ?16, ?MCFixupKindInfo::FKF_IsPCRel }, > +// movw / movt: 16-bits immediate but scattered into two chunks 0 - 12, 16 - 19. > +{ "fixup_arm_movt_hi16", ? ? 0, ? ? ? ? ? ?20, ?0 }, > +{ "fixup_arm_movw_lo16", ? ? 0, ? ? ? ? ? ?20, ?0 }, > +{ "fixup_t2_movt_hi16", ? ? ?0, ? ? ? ? ? ?20, ?0 }, > +{ "fixup_t2_movw_lo16", ? ? ?0, ? ? ? ? ? ?20, ?0 }, > +{ "fixup_arm_movt_hi16_pcrel", 0, ? ? ? ? ?20, ?MCFixupKindInfo::FKF_IsPCRel }, > +{ "fixup_arm_movw_lo16_pcrel", 0, ? ? ? ? ?20, ?MCFixupKindInfo::FKF_IsPCRel }, > +{ "fixup_t2_movt_hi16_pcrel", 0, ? ? ? ? ? 20, ?MCFixupKindInfo::FKF_IsPCRel }, > +{ "fixup_t2_movw_lo16_pcrel", 0, ? ? ? ? ? 20, ?MCFixupKindInfo::FKF_IsPCRel }, > ? ? }; > > ? ? if (Kind < FirstTargetFixupKind) > @@ -158,8 +163,10 @@ > ? case FK_Data_4: > ? ? return Value; > ? case ARM::fixup_arm_movt_hi16: > - ?case ARM::fixup_arm_movw_lo16: > ? case ARM::fixup_arm_movt_hi16_pcrel: > + ? ?Value >>= 16; > + ? ?// Fallthrough > + ?case ARM::fixup_arm_movw_lo16: > ? case ARM::fixup_arm_movw_lo16_pcrel: { > ? ? unsigned Hi4 = (Value & 0xF000) >> 12; > ? ? unsigned Lo12 = Value & 0x0FFF; > @@ -168,6 +175,26 @@ > ? ? Value = (Hi4 << 16) | (Lo12); > ? ? return Value; > ? } > + ?case ARM::fixup_t2_movt_hi16: > + ?case ARM::fixup_t2_movt_hi16_pcrel: > + ? ?Value >>= 16; > + ? ?// Fallthrough > + ?case ARM::fixup_t2_movw_lo16: > + ?case ARM::fixup_t2_movw_lo16_pcrel: { > + ? ?unsigned Hi4 = (Value & 0xF000) >> 12; > + ? ?unsigned i = (Value & 0x800) >> 11; > + ? ?unsigned Mid3 = (Value & 0x700) >> 8; > + ? ?unsigned Lo8 = Value & 0x0FF; > + ? ?// inst{19-16} = Hi4; > + ? ?// inst{26} = i; > + ? ?// inst{14-12} = Mid3; > + ? ?// inst{7-0} = Lo8; > + ? ?Value = (Hi4 << 16) | (i << 26) | (Mid3 << 12) | (Lo8); > + > + ? ?uint64_t swapped = (Value & 0xFFFF0000) >> 16; > + ? ?swapped |= (Value & 0x0000FFFF) << 16; > + ? ?return swapped; > + ?} > ? case ARM::fixup_arm_ldst_pcrel_12: > ? ? // ARM PC-relative values are offset by 8. > ? ? Value -= 4; > @@ -438,6 +465,14 @@ > ? case ARM::fixup_t2_adr_pcrel_12: > ? case ARM::fixup_arm_thumb_bl: > ? case ARM::fixup_arm_thumb_blx: > + ?case ARM::fixup_arm_movt_hi16: > + ?case ARM::fixup_arm_movw_lo16: > + ?case ARM::fixup_arm_movt_hi16_pcrel: > + ?case ARM::fixup_arm_movw_lo16_pcrel: > + ?case ARM::fixup_t2_movt_hi16: > + ?case ARM::fixup_t2_movw_lo16: > + ?case ARM::fixup_t2_movt_hi16_pcrel: > + ?case ARM::fixup_t2_movw_lo16_pcrel: > ? ? return 4; > ? } > ?} > > Modified: llvm/trunk/lib/Target/ARM/ARMFixupKinds.h > URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMFixupKinds.h?rev=123424&r1=123423&r2=123424&view=diff > ============================================================================== > --- llvm/trunk/lib/Target/ARM/ARMFixupKinds.h (original) > +++ llvm/trunk/lib/Target/ARM/ARMFixupKinds.h Thu Jan 13 20:38:49 2011 > @@ -70,9 +70,10 @@ > > ? // The next two are for the movt/movw pair > ? // the 16bit imm field are split into imm{15-12} and imm{11-0} > - ?// Fixme: We need new ones for Thumb. > ? fixup_arm_movt_hi16, // :upper16: > ? fixup_arm_movw_lo16, // :lower16: > + ?fixup_t2_movt_hi16, // :upper16: > + ?fixup_t2_movw_lo16, // :lower16: > > ? // It is possible to create an "immediate" that happens to be pcrel. > ? // movw r0, :lower16:Foo-(Bar+8) and movt ?r0, :upper16:Foo-(Bar+8) > @@ -80,6 +81,8 @@ > ? // Needed to support ELF::R_ARM_MOVT_PREL and ELF::R_ARM_MOVW_PREL_NC > ? fixup_arm_movt_hi16_pcrel, // :upper16: > ? fixup_arm_movw_lo16_pcrel, // :lower16: > + ?fixup_t2_movt_hi16_pcrel, // :upper16: > + ?fixup_t2_movw_lo16_pcrel, // :lower16: > > ? // Marker > ? LastTargetFixupKind, > > Modified: llvm/trunk/lib/Target/ARM/ARMMCCodeEmitter.cpp > URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMMCCodeEmitter.cpp?rev=123424&r1=123423&r2=123424&view=diff > ============================================================================== > --- llvm/trunk/lib/Target/ARM/ARMMCCodeEmitter.cpp (original) > +++ llvm/trunk/lib/Target/ARM/ARMMCCodeEmitter.cpp Thu Jan 13 20:38:49 2011 > @@ -17,6 +17,7 @@ > ?#include "ARMFixupKinds.h" > ?#include "ARMInstrInfo.h" > ?#include "ARMMCExpr.h" > +#include "ARMSubtarget.h" > ?#include "llvm/MC/MCCodeEmitter.h" > ?#include "llvm/MC/MCExpr.h" > ?#include "llvm/MC/MCInst.h" > @@ -33,11 +34,13 @@ > ? void operator=(const ARMMCCodeEmitter &); // DO NOT IMPLEMENT > ? const TargetMachine &TM; > ? const TargetInstrInfo &TII; > + ?const ARMSubtarget *Subtarget; > ? MCContext &Ctx; > > ?public: > ? ARMMCCodeEmitter(TargetMachine &tm, MCContext &ctx) > - ? ?: TM(tm), TII(*TM.getInstrInfo()), Ctx(ctx) { > + ? ?: TM(tm), TII(*TM.getInstrInfo()), > + ? ? ?Subtarget(&TM.getSubtarget()), Ctx(ctx) { > ? } > > ? ~ARMMCCodeEmitter() {} > @@ -306,8 +309,7 @@ > ?/// Thumb2 mode. > ?unsigned ARMMCCodeEmitter::NEONThumb2DataIPostEncoder(const MCInst &MI, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?unsigned EncodedValue) const { > - ?const ARMSubtarget &Subtarget = TM.getSubtarget(); > - ?if (Subtarget.isThumb2()) { > + ?if (Subtarget->isThumb2()) { > ? ? // NEON Thumb2 data-processsing encodings are very simple: bit 24 is moved > ? ? // to bit 12 of the high half-word (i.e. bit 28), and bits 27-24 are > ? ? // set to 1111. > @@ -326,8 +328,7 @@ > ?/// Thumb2 mode. > ?unsigned ARMMCCodeEmitter::NEONThumb2LoadStorePostEncoder(const MCInst &MI, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?unsigned EncodedValue) const { > - ?const ARMSubtarget &Subtarget = TM.getSubtarget(); > - ?if (Subtarget.isThumb2()) { > + ?if (Subtarget->isThumb2()) { > ? ? EncodedValue &= 0xF0FFFFFF; > ? ? EncodedValue |= 0x09000000; > ? } > @@ -340,8 +341,7 @@ > ?/// Thumb2 mode. > ?unsigned ARMMCCodeEmitter::NEONThumb2DupPostEncoder(const MCInst &MI, > ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?unsigned EncodedValue) const { > - ?const ARMSubtarget &Subtarget = TM.getSubtarget(); > - ?if (Subtarget.isThumb2()) { > + ?if (Subtarget->isThumb2()) { > ? ? EncodedValue &= 0x00FFFFFF; > ? ? EncodedValue |= 0xEE000000; > ? } > @@ -353,7 +353,7 @@ > ?/// them to their Thumb2 form if we are currently in Thumb2 mode. > ?unsigned ARMMCCodeEmitter:: > ?VFPThumb2PostEncoder(const MCInst &MI, unsigned EncodedValue) const { > - ?if (TM.getSubtarget().isThumb2()) { > + ?if (Subtarget->isThumb2()) { > ? ? EncodedValue &= 0x0FFFFFFF; > ? ? EncodedValue |= 0xE0000000; > ? } > @@ -477,8 +477,7 @@ > ? ? ? ? ? ? ? ? ? ? ? ?SmallVectorImpl &Fixups) const { > ? // FIXME: This really, really shouldn't use TargetMachine. We don't want > ? // coupling between MC and TM anywhere we can help it. > - ?const ARMSubtarget &Subtarget = TM.getSubtarget(); > - ?if (Subtarget.isThumb2()) > + ?if (Subtarget->isThumb2()) > ? ? return > ? ? ? ::getBranchTargetOpValue(MI, OpIdx, ARM::fixup_t2_condbranch, Fixups); > ? return ::getBranchTargetOpValue(MI, OpIdx, ARM::fixup_arm_branch, Fixups); > @@ -575,9 +574,8 @@ > ? ? else > ? ? ? Expr = MO2.getExpr(); > > - ? ?const ARMSubtarget &Subtarget = TM.getSubtarget(); > ? ? MCFixupKind Kind; > - ? ?if (Subtarget.isThumb2()) > + ? ?if (Subtarget->isThumb2()) > ? ? ? Kind = MCFixupKind(ARM::fixup_t2_ldst_pcrel_12); > ? ? else > ? ? ? Kind = MCFixupKind(ARM::fixup_arm_ldst_pcrel_12); > @@ -662,14 +660,24 @@ > ? ? switch (ARM16Expr->getKind()) { > ? ? default: assert(0 && "Unsupported ARMFixup"); > ? ? case ARMMCExpr::VK_ARM_HI16: > - ? ? ?Kind = MCFixupKind(ARM::fixup_arm_movt_hi16); > - ? ? ?if (EvaluateAsPCRel(E)) > - ? ? ? ?Kind = MCFixupKind(ARM::fixup_arm_movt_hi16_pcrel); > + ? ? ?if (!Subtarget->isTargetDarwin() && EvaluateAsPCRel(E)) > + ? ? ? ?Kind = MCFixupKind(Subtarget->isThumb2() > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ARM::fixup_t2_movt_hi16_pcrel > + ? ? ? ? ? ? ? ? ? ? ? ? ? : ARM::fixup_arm_movt_hi16_pcrel); > + ? ? ?else > + ? ? ? ?Kind = MCFixupKind(Subtarget->isThumb2() > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ARM::fixup_t2_movt_hi16 > + ? ? ? ? ? ? ? ? ? ? ? ? ? : ARM::fixup_arm_movt_hi16); > ? ? ? break; > ? ? case ARMMCExpr::VK_ARM_LO16: > - ? ? ?Kind = MCFixupKind(ARM::fixup_arm_movw_lo16); > - ? ? ?if (EvaluateAsPCRel(E)) > - ? ? ? ?Kind = MCFixupKind(ARM::fixup_arm_movw_lo16_pcrel); > + ? ? ?if (!Subtarget->isTargetDarwin() && EvaluateAsPCRel(E)) > + ? ? ? ?Kind = MCFixupKind(Subtarget->isThumb2() > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ARM::fixup_t2_movw_lo16_pcrel > + ? ? ? ? ? ? ? ? ? ? ? ? ? : ARM::fixup_arm_movw_lo16_pcrel); > + ? ? ?else > + ? ? ? ?Kind = MCFixupKind(Subtarget->isThumb2() > + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ARM::fixup_t2_movw_lo16 > + ? ? ? ? ? ? ? ? ? ? ? ? ? : ARM::fixup_arm_movw_lo16); > ? ? ? break; > ? ? } > ? ? Fixups.push_back(MCFixup::Create(0, E, Kind)); > @@ -841,8 +849,7 @@ > ? ? assert(MO.isExpr() && "Unexpected machine operand type!"); > ? ? const MCExpr *Expr = MO.getExpr(); > ? ? MCFixupKind Kind; > - ? ?const ARMSubtarget &Subtarget = TM.getSubtarget(); > - ? ?if (Subtarget.isThumb2()) > + ? ?if (Subtarget->isThumb2()) > ? ? ? Kind = MCFixupKind(ARM::fixup_t2_pcrel_10); > ? ? else > ? ? ? Kind = MCFixupKind(ARM::fixup_arm_pcrel_10); > @@ -1151,7 +1158,6 @@ > ?void ARMMCCodeEmitter:: > ?EncodeInstruction(const MCInst &MI, raw_ostream &OS, > ? ? ? ? ? ? ? ? ? SmallVectorImpl &Fixups) const { > - ?const ARMSubtarget &Subtarget = TM.getSubtarget(); > ? // Pseudo instructions don't get encoded. > ? const TargetInstrDesc &Desc = TII.get(MI.getOpcode()); > ? uint64_t TSFlags = Desc.TSFlags; > @@ -1167,7 +1173,7 @@ > ? uint32_t Binary = getBinaryCodeForInstr(MI, Fixups); > ? // Thumb 32-bit wide instructions need to emit the high order halfword > ? // first. > - ?if (Subtarget.isThumb() && Size == 4) { > + ?if (Subtarget->isThumb() && Size == 4) { > ? ? EmitConstant(Binary >> 16, 2, OS); > ? ? EmitConstant(Binary & 0xffff, 2, OS); > ? } else > > Modified: llvm/trunk/test/MC/ARM/elf-movt.s > URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/MC/ARM/elf-movt.s?rev=123424&r1=123423&r2=123424&view=diff > ============================================================================== > --- llvm/trunk/test/MC/ARM/elf-movt.s (original) > +++ llvm/trunk/test/MC/ARM/elf-movt.s Thu Jan 13 20:38:49 2011 > @@ -25,7 +25,7 @@ > ?@ OBJ-NEXT: ? ? ? ? ? ?'sh_info', 0x00000000 > ?@ OBJ-NEXT: ? ? ? ? ? ?'sh_addralign', 0x00000004 > ?@ OBJ-NEXT: ? ? ? ? ? ?'sh_entsize', 0x00000000 > -@ OBJ-NEXT: ? ? ? ? ? ?'_section_data', 'f00f0fe3 ec0f4fe3' > +@ OBJ-NEXT: ? ? ? ? ? ?'_section_data', 'f00f0fe3 ff0f4fe3' Hi Evan, FYI, the movt hi16 fixup you put in does not match the output of GNU as - Let me look into this and figure out why. I'll reply soon. -jason > > ?@ OBJ: ? ? ? ? ? ? ?Relocation 0x00000000 > ?@ OBJ-NEXT: ? ? ? ? 'r_offset', 0x00000000 > > Modified: llvm/trunk/test/MC/ARM/hilo-16bit-relocations.s > URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/MC/ARM/hilo-16bit-relocations.s?rev=123424&r1=123423&r2=123424&view=diff > ============================================================================== > --- llvm/trunk/test/MC/ARM/hilo-16bit-relocations.s (original) > +++ llvm/trunk/test/MC/ARM/hilo-16bit-relocations.s Thu Jan 13 20:38:49 2011 > @@ -1,14 +1,13 @@ > -@ RUN: llvm-mc %s -triple armv7-apple-darwin -show-encoding | FileCheck %s > +@ RUN: llvm-mc %s -triple armv7-apple-darwin | FileCheck %s > +@ RUN: llvm-mc %s -triple armv7-apple-darwin | FileCheck %s > > ?_t: > ? ? ? ? movw ? ?r0, :lower16:(L_foo$non_lazy_ptr - (L1 + 8)) > ? ? ? ? movt ? ?r0, :upper16:(L_foo$non_lazy_ptr - (L1 + 8)) > ?L1: > > -@ CHECK: movw ?r0, :lower16:(L_foo$non_lazy_ptr-(L1+8)) @ encoding: [A,A,0x00,0xe3] > -@ CHECK: ?@ ? fixup A - offset: 0, value: L_foo$non_lazy_ptr-(L1+8), kind: fixup_arm_movw_lo16_pcrel > -@ CHECK: movt ?r0, :upper16:(L_foo$non_lazy_ptr-(L1+8)) @ encoding: [A,A,0x40,0xe3] > -@ CHECK: ?@ ? fixup A - offset: 0, value: L_foo$non_lazy_ptr-(L1+8), kind: fixup_arm_movt_hi16_pcrel > +@ CHECK: movw ?r0, :lower16:(L_foo$non_lazy_ptr-(L1+8)) > +@ CHECK: movt ?r0, :upper16:(L_foo$non_lazy_ptr-(L1+8)) > > ? ? ? ? .comm ?_foo,4,2 > > > > _______________________________________________ > llvm-commits mailing list > llvm-commits at cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits > From atrick at apple.com Fri Jan 14 15:11:41 2011 From: atrick at apple.com (Andrew Trick) Date: Fri, 14 Jan 2011 21:11:41 -0000 Subject: [llvm-commits] [llvm] r123468 - /llvm/trunk/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp Message-ID: <20110114211141.7CC252A6C12C@llvm.org> Author: atrick Date: Fri Jan 14 15:11:41 2011 New Revision: 123468 URL: http://llvm.org/viewvc/llvm-project?rev=123468&view=rev Log: Support for precise scheduling of the instruction selection DAG, disabled in this checkin. Sorry for the large diffs due to refactoring. New functionality is all guarded by EnableSchedCycles. Scheduling the isel DAG is inherently imprecise, but we give it a best effort: - Added MayReduceRegPressure to allow stalled nodes in the queue only if there is a regpressure need. - Added BUHasStall to allow checking for either dependence stalls due to latency or resource stalls due to pipeline hazards. - Added BUCompareLatency to encapsulate and standardize the heuristics for minimizing stall cycles (vs. reducing register pressure). - Modified the bottom-up heuristic (now in BUCompareLatency) to prioritize nodes by their depth rather than height. As long as it doesn't stall, height is irrelevant. Depth represents the critical path to the DAG root. - Added hybrid_ls_rr_sort::isReady to filter stalled nodes before adding them to the available queue. Related Cleanup: most of the register reduction routines do not need to be templates. Modified: llvm/trunk/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp Modified: llvm/trunk/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp?rev=123468&r1=123467&r2=123468&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp (original) +++ llvm/trunk/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp Fri Jan 14 15:11:41 2011 @@ -137,6 +137,8 @@ void Schedule(); + ScheduleHazardRecognizer *getHazardRec() { return HazardRec; } + /// IsReachable - Checks if SU is reachable from TargetSU. bool IsReachable(const SUnit *SU, const SUnit *TargetSU) { return Topo.IsReachable(SU, TargetSU); @@ -1246,101 +1248,286 @@ //===----------------------------------------------------------------------===// -// RegReductionPriorityQueue Implementation +// RegReductionPriorityQueue Definition //===----------------------------------------------------------------------===// // // This is a SchedulingPriorityQueue that schedules using Sethi Ullman numbers // to reduce register pressure. // namespace { - template - class RegReductionPriorityQueue; +class RegReductionPQBase; + +struct queue_sort : public std::binary_function { + bool isReady(SUnit* SU, unsigned CurCycle) const { return true; } +}; - struct queue_sort : public std::binary_function { - bool isReady(SUnit* SU, unsigned CurCycle) const { return true; } +/// bu_ls_rr_sort - Priority function for bottom up register pressure +// reduction scheduler. +struct bu_ls_rr_sort : public queue_sort { + enum { + IsBottomUp = true, + HasReadyFilter = false }; - /// bu_ls_rr_sort - Priority function for bottom up register pressure - // reduction scheduler. - struct bu_ls_rr_sort : public queue_sort { - enum { - IsBottomUp = true, - HasReadyFilter = false - }; - - RegReductionPriorityQueue *SPQ; - bu_ls_rr_sort(RegReductionPriorityQueue *spq) : SPQ(spq) {} - bu_ls_rr_sort(const bu_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {} + RegReductionPQBase *SPQ; + bu_ls_rr_sort(RegReductionPQBase *spq) : SPQ(spq) {} + bu_ls_rr_sort(const bu_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {} + + bool operator()(SUnit* left, SUnit* right) const; +}; - bool operator()(const SUnit* left, const SUnit* right) const; +// td_ls_rr_sort - Priority function for top down register pressure reduction +// scheduler. +struct td_ls_rr_sort : public queue_sort { + enum { + IsBottomUp = false, + HasReadyFilter = false }; - // td_ls_rr_sort - Priority function for top down register pressure reduction - // scheduler. - struct td_ls_rr_sort : public queue_sort { - enum { - IsBottomUp = false, - HasReadyFilter = false - }; - - RegReductionPriorityQueue *SPQ; - td_ls_rr_sort(RegReductionPriorityQueue *spq) : SPQ(spq) {} - td_ls_rr_sort(const td_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {} + RegReductionPQBase *SPQ; + td_ls_rr_sort(RegReductionPQBase *spq) : SPQ(spq) {} + td_ls_rr_sort(const td_ls_rr_sort &RHS) : SPQ(RHS.SPQ) {} + + bool operator()(const SUnit* left, const SUnit* right) const; +}; - bool operator()(const SUnit* left, const SUnit* right) const; +// src_ls_rr_sort - Priority function for source order scheduler. +struct src_ls_rr_sort : public queue_sort { + enum { + IsBottomUp = true, + HasReadyFilter = false }; - // src_ls_rr_sort - Priority function for source order scheduler. - struct src_ls_rr_sort : public queue_sort { - enum { - IsBottomUp = true, - HasReadyFilter = false - }; - - RegReductionPriorityQueue *SPQ; - src_ls_rr_sort(RegReductionPriorityQueue *spq) - : SPQ(spq) {} - src_ls_rr_sort(const src_ls_rr_sort &RHS) - : SPQ(RHS.SPQ) {} + RegReductionPQBase *SPQ; + src_ls_rr_sort(RegReductionPQBase *spq) + : SPQ(spq) {} + src_ls_rr_sort(const src_ls_rr_sort &RHS) + : SPQ(RHS.SPQ) {} + + bool operator()(SUnit* left, SUnit* right) const; +}; - bool operator()(const SUnit* left, const SUnit* right) const; +// hybrid_ls_rr_sort - Priority function for hybrid scheduler. +struct hybrid_ls_rr_sort : public queue_sort { + enum { + IsBottomUp = true, + HasReadyFilter = true }; - // hybrid_ls_rr_sort - Priority function for hybrid scheduler. - struct hybrid_ls_rr_sort : public queue_sort { - enum { - IsBottomUp = true, - HasReadyFilter = false - }; - - RegReductionPriorityQueue *SPQ; - hybrid_ls_rr_sort(RegReductionPriorityQueue *spq) - : SPQ(spq) {} - hybrid_ls_rr_sort(const hybrid_ls_rr_sort &RHS) - : SPQ(RHS.SPQ) {} + RegReductionPQBase *SPQ; + hybrid_ls_rr_sort(RegReductionPQBase *spq) + : SPQ(spq) {} + hybrid_ls_rr_sort(const hybrid_ls_rr_sort &RHS) + : SPQ(RHS.SPQ) {} + + bool isReady(SUnit *SU, unsigned CurCycle) const; + + bool operator()(SUnit* left, SUnit* right) const; +}; - bool operator()(const SUnit* left, const SUnit* right) const; +// ilp_ls_rr_sort - Priority function for ILP (instruction level parallelism) +// scheduler. +struct ilp_ls_rr_sort : public queue_sort { + enum { + IsBottomUp = true, + HasReadyFilter = true }; - // ilp_ls_rr_sort - Priority function for ILP (instruction level parallelism) - // scheduler. - struct ilp_ls_rr_sort : public queue_sort { - enum { - IsBottomUp = true, - HasReadyFilter = true - }; - - RegReductionPriorityQueue *SPQ; - ilp_ls_rr_sort(RegReductionPriorityQueue *spq) - : SPQ(spq) {} - ilp_ls_rr_sort(const ilp_ls_rr_sort &RHS) - : SPQ(RHS.SPQ) {} + RegReductionPQBase *SPQ; + ilp_ls_rr_sort(RegReductionPQBase *spq) + : SPQ(spq) {} + ilp_ls_rr_sort(const ilp_ls_rr_sort &RHS) + : SPQ(RHS.SPQ) {} - bool isReady(SUnit *SU, unsigned CurCycle) const; + bool isReady(SUnit *SU, unsigned CurCycle) const; - bool operator()(const SUnit* left, const SUnit* right) const; - }; -} // end anonymous namespace + bool operator()(SUnit* left, SUnit* right) const; +}; + +class RegReductionPQBase : public SchedulingPriorityQueue { +protected: + std::vector Queue; + unsigned CurQueueId; + bool TracksRegPressure; + + // SUnits - The SUnits for the current graph. + std::vector *SUnits; + + MachineFunction &MF; + const TargetInstrInfo *TII; + const TargetRegisterInfo *TRI; + const TargetLowering *TLI; + ScheduleDAGRRList *scheduleDAG; + + // SethiUllmanNumbers - The SethiUllman number for each node. + std::vector SethiUllmanNumbers; + + /// RegPressure - Tracking current reg pressure per register class. + /// + std::vector RegPressure; + + /// RegLimit - Tracking the number of allocatable registers per register + /// class. + std::vector RegLimit; + +public: + RegReductionPQBase(MachineFunction &mf, + bool hasReadyFilter, + bool tracksrp, + const TargetInstrInfo *tii, + const TargetRegisterInfo *tri, + const TargetLowering *tli) + : SchedulingPriorityQueue(hasReadyFilter), + CurQueueId(0), TracksRegPressure(tracksrp), + MF(mf), TII(tii), TRI(tri), TLI(tli), scheduleDAG(NULL) { + if (TracksRegPressure) { + unsigned NumRC = TRI->getNumRegClasses(); + RegLimit.resize(NumRC); + RegPressure.resize(NumRC); + std::fill(RegLimit.begin(), RegLimit.end(), 0); + std::fill(RegPressure.begin(), RegPressure.end(), 0); + for (TargetRegisterInfo::regclass_iterator I = TRI->regclass_begin(), + E = TRI->regclass_end(); I != E; ++I) + RegLimit[(*I)->getID()] = tli->getRegPressureLimit(*I, MF); + } + } + + void setScheduleDAG(ScheduleDAGRRList *scheduleDag) { + scheduleDAG = scheduleDag; + } + + ScheduleHazardRecognizer* getHazardRec() { + return scheduleDAG->getHazardRec(); + } + + void initNodes(std::vector &sunits); + + void addNode(const SUnit *SU); + + void updateNode(const SUnit *SU); + + void releaseState() { + SUnits = 0; + SethiUllmanNumbers.clear(); + std::fill(RegPressure.begin(), RegPressure.end(), 0); + } + + unsigned getNodePriority(const SUnit *SU) const; + + unsigned getNodeOrdering(const SUnit *SU) const { + return scheduleDAG->DAG->GetOrdering(SU->getNode()); + } + + bool empty() const { return Queue.empty(); } + + void push(SUnit *U) { + assert(!U->NodeQueueId && "Node in the queue already"); + U->NodeQueueId = ++CurQueueId; + Queue.push_back(U); + } + + void remove(SUnit *SU) { + assert(!Queue.empty() && "Queue is empty!"); + assert(SU->NodeQueueId != 0 && "Not in queue!"); + std::vector::iterator I = std::find(Queue.begin(), Queue.end(), + SU); + if (I != prior(Queue.end())) + std::swap(*I, Queue.back()); + Queue.pop_back(); + SU->NodeQueueId = 0; + } + + void dumpRegPressure() const; + + bool HighRegPressure(const SUnit *SU) const; + + bool MayReduceRegPressure(SUnit *SU); + + void ScheduledNode(SUnit *SU); + + void UnscheduledNode(SUnit *SU); + +protected: + bool canClobber(const SUnit *SU, const SUnit *Op); + void AddPseudoTwoAddrDeps(); + void PrescheduleNodesWithMultipleUses(); + void CalculateSethiUllmanNumbers(); +}; + +template +class RegReductionPriorityQueue : public RegReductionPQBase { + static SUnit *popFromQueue(std::vector &Q, SF &Picker) { + std::vector::iterator Best = Q.begin(); + for (std::vector::iterator I = llvm::next(Q.begin()), + E = Q.end(); I != E; ++I) + if (Picker(*Best, *I)) + Best = I; + SUnit *V = *Best; + if (Best != prior(Q.end())) + std::swap(*Best, Q.back()); + Q.pop_back(); + return V; + } + + SF Picker; + +public: + RegReductionPriorityQueue(MachineFunction &mf, + bool tracksrp, + const TargetInstrInfo *tii, + const TargetRegisterInfo *tri, + const TargetLowering *tli) + : RegReductionPQBase(mf, SF::HasReadyFilter, tracksrp, tii, tri, tli), + Picker(this) {} + + bool isBottomUp() const { return SF::IsBottomUp; } + + bool isReady(SUnit *U) const { + return Picker.HasReadyFilter && Picker.isReady(U, getCurCycle()); + } + + SUnit *pop() { + if (Queue.empty()) return NULL; + + SUnit *V = popFromQueue(Queue, Picker); + V->NodeQueueId = 0; + return V; + } + + void dump(ScheduleDAG *DAG) const { + // Emulate pop() without clobbering NodeQueueIds. + std::vector DumpQueue = Queue; + SF DumpPicker = Picker; + while (!DumpQueue.empty()) { + SUnit *SU = popFromQueue(DumpQueue, DumpPicker); + if (isBottomUp()) + dbgs() << "Height " << SU->getHeight() << ": "; + else + dbgs() << "Depth " << SU->getDepth() << ": "; + SU->dump(DAG); + } + } +}; + +typedef RegReductionPriorityQueue +BURegReductionPriorityQueue; + +typedef RegReductionPriorityQueue +TDRegReductionPriorityQueue; + +typedef RegReductionPriorityQueue +SrcRegReductionPriorityQueue; + +typedef RegReductionPriorityQueue +HybridBURRPriorityQueue; + +typedef RegReductionPriorityQueue +ILPBURRPriorityQueue; +} // end anonymous namespace + +//===----------------------------------------------------------------------===// +// Static Node Priority for Register Pressure Reduction +//===----------------------------------------------------------------------===// /// CalcNodeSethiUllmanNumber - Compute Sethi Ullman number. /// Smaller number is the higher priority. @@ -1371,436 +1558,324 @@ return SethiUllmanNumber; } -namespace { - template - class RegReductionPriorityQueue : public SchedulingPriorityQueue { - static SUnit *popFromQueue(std::vector &Q, SF &Picker) { - std::vector::iterator Best = Q.begin(); - for (std::vector::iterator I = llvm::next(Q.begin()), - E = Q.end(); I != E; ++I) - if (Picker(*Best, *I)) - Best = I; - SUnit *V = *Best; - if (Best != prior(Q.end())) - std::swap(*Best, Q.back()); - Q.pop_back(); - return V; - } - - std::vector Queue; - SF Picker; - unsigned CurQueueId; - bool TracksRegPressure; - - protected: - // SUnits - The SUnits for the current graph. - std::vector *SUnits; - - MachineFunction &MF; - const TargetInstrInfo *TII; - const TargetRegisterInfo *TRI; - const TargetLowering *TLI; - ScheduleDAGRRList *scheduleDAG; - - // SethiUllmanNumbers - The SethiUllman number for each node. - std::vector SethiUllmanNumbers; - - /// RegPressure - Tracking current reg pressure per register class. - /// - std::vector RegPressure; - - /// RegLimit - Tracking the number of allocatable registers per register - /// class. - std::vector RegLimit; - - public: - RegReductionPriorityQueue(MachineFunction &mf, - bool tracksrp, - const TargetInstrInfo *tii, - const TargetRegisterInfo *tri, - const TargetLowering *tli) - : SchedulingPriorityQueue(SF::HasReadyFilter), Picker(this), - CurQueueId(0), TracksRegPressure(tracksrp), - MF(mf), TII(tii), TRI(tri), TLI(tli), scheduleDAG(NULL) { - if (TracksRegPressure) { - unsigned NumRC = TRI->getNumRegClasses(); - RegLimit.resize(NumRC); - RegPressure.resize(NumRC); - std::fill(RegLimit.begin(), RegLimit.end(), 0); - std::fill(RegPressure.begin(), RegPressure.end(), 0); - for (TargetRegisterInfo::regclass_iterator I = TRI->regclass_begin(), - E = TRI->regclass_end(); I != E; ++I) - RegLimit[(*I)->getID()] = tli->getRegPressureLimit(*I, MF); - } - } +/// CalculateSethiUllmanNumbers - Calculate Sethi-Ullman numbers of all +/// scheduling units. +void RegReductionPQBase::CalculateSethiUllmanNumbers() { + SethiUllmanNumbers.assign(SUnits->size(), 0); - bool isBottomUp() const { return SF::IsBottomUp; } + for (unsigned i = 0, e = SUnits->size(); i != e; ++i) + CalcNodeSethiUllmanNumber(&(*SUnits)[i], SethiUllmanNumbers); +} - void initNodes(std::vector &sunits) { - SUnits = &sunits; - // Add pseudo dependency edges for two-address nodes. - AddPseudoTwoAddrDeps(); - // Reroute edges to nodes with multiple uses. - PrescheduleNodesWithMultipleUses(); - // Calculate node priorities. - CalculateSethiUllmanNumbers(); - } +void RegReductionPQBase::initNodes(std::vector &sunits) { + SUnits = &sunits; + // Add pseudo dependency edges for two-address nodes. + AddPseudoTwoAddrDeps(); + // Reroute edges to nodes with multiple uses. + PrescheduleNodesWithMultipleUses(); + // Calculate node priorities. + CalculateSethiUllmanNumbers(); +} + +void RegReductionPQBase::addNode(const SUnit *SU) { + unsigned SUSize = SethiUllmanNumbers.size(); + if (SUnits->size() > SUSize) + SethiUllmanNumbers.resize(SUSize*2, 0); + CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers); +} + +void RegReductionPQBase::updateNode(const SUnit *SU) { + SethiUllmanNumbers[SU->NodeNum] = 0; + CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers); +} + +unsigned RegReductionPQBase::getNodePriority(const SUnit *SU) const { + assert(SU->NodeNum < SethiUllmanNumbers.size()); + unsigned Opc = SU->getNode() ? SU->getNode()->getOpcode() : 0; + if (Opc == ISD::TokenFactor || Opc == ISD::CopyToReg) + // CopyToReg should be close to its uses to facilitate coalescing and + // avoid spilling. + return 0; + if (Opc == TargetOpcode::EXTRACT_SUBREG || + Opc == TargetOpcode::SUBREG_TO_REG || + Opc == TargetOpcode::INSERT_SUBREG) + // EXTRACT_SUBREG, INSERT_SUBREG, and SUBREG_TO_REG nodes should be + // close to their uses to facilitate coalescing. + return 0; + if (SU->NumSuccs == 0 && SU->NumPreds != 0) + // If SU does not have a register use, i.e. it doesn't produce a value + // that would be consumed (e.g. store), then it terminates a chain of + // computation. Give it a large SethiUllman number so it will be + // scheduled right before its predecessors that it doesn't lengthen + // their live ranges. + return 0xffff; + if (SU->NumPreds == 0 && SU->NumSuccs != 0) + // If SU does not have a register def, schedule it close to its uses + // because it does not lengthen any live ranges. + return 0; + return SethiUllmanNumbers[SU->NodeNum]; +} - void addNode(const SUnit *SU) { - unsigned SUSize = SethiUllmanNumbers.size(); - if (SUnits->size() > SUSize) - SethiUllmanNumbers.resize(SUSize*2, 0); - CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers); - } +//===----------------------------------------------------------------------===// +// Register Pressure Tracking +//===----------------------------------------------------------------------===// - void updateNode(const SUnit *SU) { - SethiUllmanNumbers[SU->NodeNum] = 0; - CalcNodeSethiUllmanNumber(SU, SethiUllmanNumbers); - } +void RegReductionPQBase::dumpRegPressure() const { + for (TargetRegisterInfo::regclass_iterator I = TRI->regclass_begin(), + E = TRI->regclass_end(); I != E; ++I) { + const TargetRegisterClass *RC = *I; + unsigned Id = RC->getID(); + unsigned RP = RegPressure[Id]; + if (!RP) continue; + DEBUG(dbgs() << RC->getName() << ": " << RP << " / " << RegLimit[Id] + << '\n'); + } +} - void releaseState() { - SUnits = 0; - SethiUllmanNumbers.clear(); - std::fill(RegPressure.begin(), RegPressure.end(), 0); - } +bool RegReductionPQBase::HighRegPressure(const SUnit *SU) const { + if (!TLI) + return false; - unsigned getNodePriority(const SUnit *SU) const { - assert(SU->NodeNum < SethiUllmanNumbers.size()); - unsigned Opc = SU->getNode() ? SU->getNode()->getOpcode() : 0; - if (Opc == ISD::TokenFactor || Opc == ISD::CopyToReg) - // CopyToReg should be close to its uses to facilitate coalescing and - // avoid spilling. - return 0; - if (Opc == TargetOpcode::EXTRACT_SUBREG || - Opc == TargetOpcode::SUBREG_TO_REG || - Opc == TargetOpcode::INSERT_SUBREG) - // EXTRACT_SUBREG, INSERT_SUBREG, and SUBREG_TO_REG nodes should be - // close to their uses to facilitate coalescing. - return 0; - if (SU->NumSuccs == 0 && SU->NumPreds != 0) - // If SU does not have a register use, i.e. it doesn't produce a value - // that would be consumed (e.g. store), then it terminates a chain of - // computation. Give it a large SethiUllman number so it will be - // scheduled right before its predecessors that it doesn't lengthen - // their live ranges. - return 0xffff; - if (SU->NumPreds == 0 && SU->NumSuccs != 0) - // If SU does not have a register def, schedule it close to its uses - // because it does not lengthen any live ranges. - return 0; - return SethiUllmanNumbers[SU->NodeNum]; - } - - unsigned getNodeOrdering(const SUnit *SU) const { - return scheduleDAG->DAG->GetOrdering(SU->getNode()); - } - - bool empty() const { return Queue.empty(); } - - bool isReady(SUnit *U) const { - return Picker.HasReadyFilter && Picker.isReady(U, getCurCycle()); - } - - void push(SUnit *U) { - assert(!U->NodeQueueId && "Node in the queue already"); - U->NodeQueueId = ++CurQueueId; - Queue.push_back(U); - } - - SUnit *pop() { - if (Queue.empty()) return NULL; - - SUnit *V = popFromQueue(Queue, Picker); - V->NodeQueueId = 0; - return V; - } - - void remove(SUnit *SU) { - assert(!Queue.empty() && "Queue is empty!"); - assert(SU->NodeQueueId != 0 && "Not in queue!"); - std::vector::iterator I = std::find(Queue.begin(), Queue.end(), - SU); - if (I != prior(Queue.end())) - std::swap(*I, Queue.back()); - Queue.pop_back(); - SU->NodeQueueId = 0; - } - - bool HighRegPressure(const SUnit *SU) const { - if (!TLI) - return false; - - for (SUnit::const_pred_iterator I = SU->Preds.begin(),E = SU->Preds.end(); - I != E; ++I) { - if (I->isCtrl()) - continue; - SUnit *PredSU = I->getSUnit(); - const SDNode *PN = PredSU->getNode(); - if (!PN->isMachineOpcode()) { - if (PN->getOpcode() == ISD::CopyFromReg) { - EVT VT = PN->getValueType(0); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - unsigned Cost = TLI->getRepRegClassCostFor(VT); - if ((RegPressure[RCId] + Cost) >= RegLimit[RCId]) - return true; - } - continue; - } - unsigned POpc = PN->getMachineOpcode(); - if (POpc == TargetOpcode::IMPLICIT_DEF) - continue; - if (POpc == TargetOpcode::EXTRACT_SUBREG) { - EVT VT = PN->getOperand(0).getValueType(); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - unsigned Cost = TLI->getRepRegClassCostFor(VT); - // Check if this increases register pressure of the specific register - // class to the point where it would cause spills. - if ((RegPressure[RCId] + Cost) >= RegLimit[RCId]) - return true; - continue; - } else if (POpc == TargetOpcode::INSERT_SUBREG || - POpc == TargetOpcode::SUBREG_TO_REG) { - EVT VT = PN->getValueType(0); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - unsigned Cost = TLI->getRepRegClassCostFor(VT); - // Check if this increases register pressure of the specific register - // class to the point where it would cause spills. - if ((RegPressure[RCId] + Cost) >= RegLimit[RCId]) - return true; - continue; - } - unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs(); - for (unsigned i = 0; i != NumDefs; ++i) { - EVT VT = PN->getValueType(i); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - if (RegPressure[RCId] >= RegLimit[RCId]) - return true; // Reg pressure already high. - unsigned Cost = TLI->getRepRegClassCostFor(VT); - if (!PN->hasAnyUseOfValue(i)) - continue; - // Check if this increases register pressure of the specific register - // class to the point where it would cause spills. - if ((RegPressure[RCId] + Cost) >= RegLimit[RCId]) - return true; - } + for (SUnit::const_pred_iterator I = SU->Preds.begin(),E = SU->Preds.end(); + I != E; ++I) { + if (I->isCtrl()) + continue; + SUnit *PredSU = I->getSUnit(); + const SDNode *PN = PredSU->getNode(); + if (!PN->isMachineOpcode()) { + if (PN->getOpcode() == ISD::CopyFromReg) { + EVT VT = PN->getValueType(0); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + unsigned Cost = TLI->getRepRegClassCostFor(VT); + if ((RegPressure[RCId] + Cost) >= RegLimit[RCId]) + return true; } - - return false; + continue; } + unsigned POpc = PN->getMachineOpcode(); + if (POpc == TargetOpcode::IMPLICIT_DEF) + continue; + if (POpc == TargetOpcode::EXTRACT_SUBREG) { + EVT VT = PN->getOperand(0).getValueType(); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + unsigned Cost = TLI->getRepRegClassCostFor(VT); + // Check if this increases register pressure of the specific register + // class to the point where it would cause spills. + if ((RegPressure[RCId] + Cost) >= RegLimit[RCId]) + return true; + continue; + } else if (POpc == TargetOpcode::INSERT_SUBREG || + POpc == TargetOpcode::SUBREG_TO_REG) { + EVT VT = PN->getValueType(0); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + unsigned Cost = TLI->getRepRegClassCostFor(VT); + // Check if this increases register pressure of the specific register + // class to the point where it would cause spills. + if ((RegPressure[RCId] + Cost) >= RegLimit[RCId]) + return true; + continue; + } + unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs(); + for (unsigned i = 0; i != NumDefs; ++i) { + EVT VT = PN->getValueType(i); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + if (RegPressure[RCId] >= RegLimit[RCId]) + return true; // Reg pressure already high. + unsigned Cost = TLI->getRepRegClassCostFor(VT); + if (!PN->hasAnyUseOfValue(i)) + continue; + // Check if this increases register pressure of the specific register + // class to the point where it would cause spills. + if ((RegPressure[RCId] + Cost) >= RegLimit[RCId]) + return true; + } + } - void ScheduledNode(SUnit *SU) { - if (!TracksRegPressure) - return; - - const SDNode *N = SU->getNode(); - if (!N->isMachineOpcode()) { - if (N->getOpcode() != ISD::CopyToReg) - return; - } else { - unsigned Opc = N->getMachineOpcode(); - if (Opc == TargetOpcode::EXTRACT_SUBREG || - Opc == TargetOpcode::INSERT_SUBREG || - Opc == TargetOpcode::SUBREG_TO_REG || - Opc == TargetOpcode::REG_SEQUENCE || - Opc == TargetOpcode::IMPLICIT_DEF) - return; - } + return false; +} - for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end(); - I != E; ++I) { - if (I->isCtrl()) - continue; - SUnit *PredSU = I->getSUnit(); - if (PredSU->NumSuccsLeft != PredSU->NumSuccs) - continue; - const SDNode *PN = PredSU->getNode(); - if (!PN->isMachineOpcode()) { - if (PN->getOpcode() == ISD::CopyFromReg) { - EVT VT = PN->getValueType(0); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); - } - continue; - } - unsigned POpc = PN->getMachineOpcode(); - if (POpc == TargetOpcode::IMPLICIT_DEF) - continue; - if (POpc == TargetOpcode::EXTRACT_SUBREG) { - EVT VT = PN->getOperand(0).getValueType(); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); - continue; - } else if (POpc == TargetOpcode::INSERT_SUBREG || - POpc == TargetOpcode::SUBREG_TO_REG) { - EVT VT = PN->getValueType(0); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); - continue; - } - unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs(); - for (unsigned i = 0; i != NumDefs; ++i) { - EVT VT = PN->getValueType(i); - if (!PN->hasAnyUseOfValue(i)) - continue; - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); - } - } +bool RegReductionPQBase::MayReduceRegPressure(SUnit *SU) { + const SDNode *N = SU->getNode(); - // Check for isMachineOpcode() as PrescheduleNodesWithMultipleUses() - // may transfer data dependencies to CopyToReg. - if (SU->NumSuccs && N->isMachineOpcode()) { - unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs(); - for (unsigned i = 0; i != NumDefs; ++i) { - EVT VT = N->getValueType(i); - if (!N->hasAnyUseOfValue(i)) - continue; - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - if (RegPressure[RCId] < TLI->getRepRegClassCostFor(VT)) - // Register pressure tracking is imprecise. This can happen. - RegPressure[RCId] = 0; - else - RegPressure[RCId] -= TLI->getRepRegClassCostFor(VT); - } - } + if (!N->isMachineOpcode() || !SU->NumSuccs) + return false; - dumpRegPressure(); - } + unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs(); + for (unsigned i = 0; i != NumDefs; ++i) { + EVT VT = N->getValueType(i); + if (!N->hasAnyUseOfValue(i)) + continue; + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + if (RegPressure[RCId] >= RegLimit[RCId]) + return true; + } + return false; +} - void UnscheduledNode(SUnit *SU) { - if (!TracksRegPressure) - return; - - const SDNode *N = SU->getNode(); - if (!N->isMachineOpcode()) { - if (N->getOpcode() != ISD::CopyToReg) - return; - } else { - unsigned Opc = N->getMachineOpcode(); - if (Opc == TargetOpcode::EXTRACT_SUBREG || - Opc == TargetOpcode::INSERT_SUBREG || - Opc == TargetOpcode::SUBREG_TO_REG || - Opc == TargetOpcode::REG_SEQUENCE || - Opc == TargetOpcode::IMPLICIT_DEF) - return; - } +void RegReductionPQBase::ScheduledNode(SUnit *SU) { + if (!TracksRegPressure) + return; - for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end(); - I != E; ++I) { - if (I->isCtrl()) - continue; - SUnit *PredSU = I->getSUnit(); - if (PredSU->NumSuccsLeft != PredSU->NumSuccs) - continue; - const SDNode *PN = PredSU->getNode(); - if (!PN->isMachineOpcode()) { - if (PN->getOpcode() == ISD::CopyFromReg) { - EVT VT = PN->getValueType(0); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); - } - continue; - } - unsigned POpc = PN->getMachineOpcode(); - if (POpc == TargetOpcode::IMPLICIT_DEF) - continue; - if (POpc == TargetOpcode::EXTRACT_SUBREG) { - EVT VT = PN->getOperand(0).getValueType(); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); - continue; - } else if (POpc == TargetOpcode::INSERT_SUBREG || - POpc == TargetOpcode::SUBREG_TO_REG) { - EVT VT = PN->getValueType(0); - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); - continue; - } - unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs(); - for (unsigned i = 0; i != NumDefs; ++i) { - EVT VT = PN->getValueType(i); - if (!PN->hasAnyUseOfValue(i)) - continue; - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - if (RegPressure[RCId] < TLI->getRepRegClassCostFor(VT)) - // Register pressure tracking is imprecise. This can happen. - RegPressure[RCId] = 0; - else - RegPressure[RCId] -= TLI->getRepRegClassCostFor(VT); - } - } + const SDNode *N = SU->getNode(); + if (!N->isMachineOpcode()) { + if (N->getOpcode() != ISD::CopyToReg) + return; + } else { + unsigned Opc = N->getMachineOpcode(); + if (Opc == TargetOpcode::EXTRACT_SUBREG || + Opc == TargetOpcode::INSERT_SUBREG || + Opc == TargetOpcode::SUBREG_TO_REG || + Opc == TargetOpcode::REG_SEQUENCE || + Opc == TargetOpcode::IMPLICIT_DEF) + return; + } - // Check for isMachineOpcode() as PrescheduleNodesWithMultipleUses() - // may transfer data dependencies to CopyToReg. - if (SU->NumSuccs && N->isMachineOpcode()) { - unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs(); - for (unsigned i = NumDefs, e = N->getNumValues(); i != e; ++i) { - EVT VT = N->getValueType(i); - if (VT == MVT::Glue || VT == MVT::Other) - continue; - if (!N->hasAnyUseOfValue(i)) - continue; - unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); - RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); - } + for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end(); + I != E; ++I) { + if (I->isCtrl()) + continue; + SUnit *PredSU = I->getSUnit(); + if (PredSU->NumSuccsLeft != PredSU->NumSuccs) + continue; + const SDNode *PN = PredSU->getNode(); + if (!PN->isMachineOpcode()) { + if (PN->getOpcode() == ISD::CopyFromReg) { + EVT VT = PN->getValueType(0); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); } - - dumpRegPressure(); + continue; } - - void setScheduleDAG(ScheduleDAGRRList *scheduleDag) { - scheduleDAG = scheduleDag; + unsigned POpc = PN->getMachineOpcode(); + if (POpc == TargetOpcode::IMPLICIT_DEF) + continue; + if (POpc == TargetOpcode::EXTRACT_SUBREG) { + EVT VT = PN->getOperand(0).getValueType(); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); + continue; + } else if (POpc == TargetOpcode::INSERT_SUBREG || + POpc == TargetOpcode::SUBREG_TO_REG) { + EVT VT = PN->getValueType(0); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); + continue; } - - void dumpRegPressure() const { - for (TargetRegisterInfo::regclass_iterator I = TRI->regclass_begin(), - E = TRI->regclass_end(); I != E; ++I) { - const TargetRegisterClass *RC = *I; - unsigned Id = RC->getID(); - unsigned RP = RegPressure[Id]; - if (!RP) continue; - DEBUG(dbgs() << RC->getName() << ": " << RP << " / " << RegLimit[Id] - << '\n'); - } + unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs(); + for (unsigned i = 0; i != NumDefs; ++i) { + EVT VT = PN->getValueType(i); + if (!PN->hasAnyUseOfValue(i)) + continue; + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); } + } - void dump(ScheduleDAG *DAG) const { - // Emulate pop() without clobbering NodeQueueIds. - std::vector DumpQueue = Queue; - SF DumpPicker = Picker; - while (!DumpQueue.empty()) { - SUnit *SU = popFromQueue(DumpQueue, DumpPicker); - if (isBottomUp()) - dbgs() << "Height " << SU->getHeight() << ": "; - else - dbgs() << "Depth " << SU->getDepth() << ": "; - SU->dump(DAG); - } + // Check for isMachineOpcode() as PrescheduleNodesWithMultipleUses() + // may transfer data dependencies to CopyToReg. + if (SU->NumSuccs && N->isMachineOpcode()) { + unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs(); + for (unsigned i = 0; i != NumDefs; ++i) { + EVT VT = N->getValueType(i); + if (!N->hasAnyUseOfValue(i)) + continue; + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + if (RegPressure[RCId] < TLI->getRepRegClassCostFor(VT)) + // Register pressure tracking is imprecise. This can happen. + RegPressure[RCId] = 0; + else + RegPressure[RCId] -= TLI->getRepRegClassCostFor(VT); } + } - protected: - bool canClobber(const SUnit *SU, const SUnit *Op); - void AddPseudoTwoAddrDeps(); - void PrescheduleNodesWithMultipleUses(); - void CalculateSethiUllmanNumbers(); - }; + dumpRegPressure(); +} - typedef RegReductionPriorityQueue - BURegReductionPriorityQueue; +void RegReductionPQBase::UnscheduledNode(SUnit *SU) { + if (!TracksRegPressure) + return; - typedef RegReductionPriorityQueue - TDRegReductionPriorityQueue; + const SDNode *N = SU->getNode(); + if (!N->isMachineOpcode()) { + if (N->getOpcode() != ISD::CopyToReg) + return; + } else { + unsigned Opc = N->getMachineOpcode(); + if (Opc == TargetOpcode::EXTRACT_SUBREG || + Opc == TargetOpcode::INSERT_SUBREG || + Opc == TargetOpcode::SUBREG_TO_REG || + Opc == TargetOpcode::REG_SEQUENCE || + Opc == TargetOpcode::IMPLICIT_DEF) + return; + } - typedef RegReductionPriorityQueue - SrcRegReductionPriorityQueue; + for (SUnit::pred_iterator I = SU->Preds.begin(), E = SU->Preds.end(); + I != E; ++I) { + if (I->isCtrl()) + continue; + SUnit *PredSU = I->getSUnit(); + if (PredSU->NumSuccsLeft != PredSU->NumSuccs) + continue; + const SDNode *PN = PredSU->getNode(); + if (!PN->isMachineOpcode()) { + if (PN->getOpcode() == ISD::CopyFromReg) { + EVT VT = PN->getValueType(0); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); + } + continue; + } + unsigned POpc = PN->getMachineOpcode(); + if (POpc == TargetOpcode::IMPLICIT_DEF) + continue; + if (POpc == TargetOpcode::EXTRACT_SUBREG) { + EVT VT = PN->getOperand(0).getValueType(); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); + continue; + } else if (POpc == TargetOpcode::INSERT_SUBREG || + POpc == TargetOpcode::SUBREG_TO_REG) { + EVT VT = PN->getValueType(0); + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); + continue; + } + unsigned NumDefs = TII->get(PN->getMachineOpcode()).getNumDefs(); + for (unsigned i = 0; i != NumDefs; ++i) { + EVT VT = PN->getValueType(i); + if (!PN->hasAnyUseOfValue(i)) + continue; + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + if (RegPressure[RCId] < TLI->getRepRegClassCostFor(VT)) + // Register pressure tracking is imprecise. This can happen. + RegPressure[RCId] = 0; + else + RegPressure[RCId] -= TLI->getRepRegClassCostFor(VT); + } + } - typedef RegReductionPriorityQueue - HybridBURRPriorityQueue; + // Check for isMachineOpcode() as PrescheduleNodesWithMultipleUses() + // may transfer data dependencies to CopyToReg. + if (SU->NumSuccs && N->isMachineOpcode()) { + unsigned NumDefs = TII->get(N->getMachineOpcode()).getNumDefs(); + for (unsigned i = NumDefs, e = N->getNumValues(); i != e; ++i) { + EVT VT = N->getValueType(i); + if (VT == MVT::Glue || VT == MVT::Other) + continue; + if (!N->hasAnyUseOfValue(i)) + continue; + unsigned RCId = TLI->getRepRegClassFor(VT)->getID(); + RegPressure[RCId] += TLI->getRepRegClassCostFor(VT); + } + } - typedef RegReductionPriorityQueue - ILPBURRPriorityQueue; + dumpRegPressure(); } +//===----------------------------------------------------------------------===// +// Dynamic Node Priority for Register Pressure Reduction +//===----------------------------------------------------------------------===// + /// closestSucc - Returns the scheduled cycle of the successor which is /// closest to the current cycle. static unsigned closestSucc(const SUnit *SU) { @@ -1872,9 +1947,81 @@ return false; } -template -static bool BURRSort(const SUnit *left, const SUnit *right, - const RegReductionPriorityQueue *SPQ) { +// Check for either a dependence (latency) or resource (hazard) stall. +// +// Note: The ScheduleHazardRecognizer interface requires a non-const SU. +static bool BUHasStall(SUnit *SU, int Height, RegReductionPQBase *SPQ) { + if ((int)SPQ->getCurCycle() < Height) return true; + if (SPQ->getHazardRec()->getHazardType(SU, 0) + != ScheduleHazardRecognizer::NoHazard) + return true; + return false; +} + +// Return -1 if left has higher priority, 1 if right has higher priority. +// Return 0 if latency-based priority is equivalent. +static int BUCompareLatency(SUnit *left, SUnit *right, bool checkPref, + RegReductionPQBase *SPQ) { + // If the two nodes share an operand and one of them has a single + // use that is a live out copy, favor the one that is live out. Otherwise + // it will be difficult to eliminate the copy if the instruction is a + // loop induction variable update. e.g. + // BB: + // sub r1, r3, #1 + // str r0, [r2, r3] + // mov r3, r1 + // cmp + // bne BB + bool SharePred = UnitsSharePred(left, right); + // FIXME: Only adjust if BB is a loop back edge. + // FIXME: What's the cost of a copy? + int LBonus = (SharePred && hasOnlyLiveOutUses(left)) ? 1 : 0; + int RBonus = (SharePred && hasOnlyLiveOutUses(right)) ? 1 : 0; + int LHeight = (int)left->getHeight() - LBonus; + int RHeight = (int)right->getHeight() - RBonus; + + bool LStall = (!checkPref || left->SchedulingPref == Sched::Latency) && + BUHasStall(left, LHeight, SPQ); + bool RStall = (!checkPref || right->SchedulingPref == Sched::Latency) && + BUHasStall(right, RHeight, SPQ); + + // If scheduling one of the node will cause a pipeline stall, delay it. + // If scheduling either one of the node will cause a pipeline stall, sort + // them according to their height. + if (LStall) { + if (!RStall) + return 1; + if (LHeight != RHeight) + return LHeight > RHeight ? 1 : -1; + } else if (RStall) + return -1; + + // If either node is scheduling for latency, sort them by depth + // and latency. + if (!checkPref || (left->SchedulingPref == Sched::Latency || + right->SchedulingPref == Sched::Latency)) { + int LDepth = (int)left->getDepth(); + int RDepth = (int)right->getDepth(); + + DEBUG(dbgs() << " Comparing latency of SU #" << left->NodeNum + << " depth " << LDepth << " vs SU #" << right->NodeNum + << " depth " << RDepth << "\n"); + + if (EnableSchedCycles) { + if (LDepth != RDepth) + return LDepth < RDepth ? 1 : -1; + } + else { + if (LHeight != RHeight) + return LHeight > RHeight ? 1 : -1; + } + if (left->Latency != right->Latency) + return left->Latency > right->Latency ? 1 : -1; + } + return 0; +} + +static bool BURRSort(SUnit *left, SUnit *right, RegReductionPQBase *SPQ) { unsigned LPriority = SPQ->getNodePriority(left); unsigned RPriority = SPQ->getNodePriority(right); if (LPriority != RPriority) @@ -1908,12 +2055,18 @@ if (LScratch != RScratch) return LScratch > RScratch; - // Note: with a bottom-up ready filter, the height check may be redundant. - if (left->getHeight() != right->getHeight()) - return left->getHeight() > right->getHeight(); + if (EnableSchedCycles) { + int result = BUCompareLatency(left, right, false /*checkPref*/, SPQ); + if (result != 0) + return result > 0; + } + else { + if (left->getHeight() != right->getHeight()) + return left->getHeight() > right->getHeight(); - if (left->getDepth() != right->getDepth()) - return left->getDepth() < right->getDepth(); + if (left->getDepth() != right->getDepth()) + return left->getDepth() < right->getDepth(); + } assert(left->NodeQueueId && right->NodeQueueId && "NodeQueueId cannot be zero"); @@ -1921,12 +2074,12 @@ } // Bottom up -bool bu_ls_rr_sort::operator()(const SUnit *left, const SUnit *right) const { +bool bu_ls_rr_sort::operator()(SUnit *left, SUnit *right) const { return BURRSort(left, right, SPQ); } // Source order, otherwise bottom up. -bool src_ls_rr_sort::operator()(const SUnit *left, const SUnit *right) const { +bool src_ls_rr_sort::operator()(SUnit *left, SUnit *right) const { unsigned LOrder = SPQ->getNodeOrdering(left); unsigned ROrder = SPQ->getNodeOrdering(right); @@ -1938,7 +2091,26 @@ return BURRSort(left, right, SPQ); } -bool hybrid_ls_rr_sort::operator()(const SUnit *left, const SUnit *right) const{ +// If the time between now and when the instruction will be ready can cover +// the spill code, then avoid adding it to the ready queue. This gives long +// stalls highest priority and allows hoisting across calls. It should also +// speed up processing the available queue. +bool hybrid_ls_rr_sort::isReady(SUnit *SU, unsigned CurCycle) const { + static const unsigned ReadyDelay = 3; + + if (SPQ->MayReduceRegPressure(SU)) return true; + + if (SU->getHeight() > (CurCycle + ReadyDelay)) return false; + + if (SPQ->getHazardRec()->getHazardType(SU, -ReadyDelay) + != ScheduleHazardRecognizer::NoHazard) + return false; + + return true; +} + +// Return true if right should be scheduled with higher priority than left. +bool hybrid_ls_rr_sort::operator()(SUnit *left, SUnit *right) const { if (left->isCall || right->isCall) // No way to compute latency of calls. return BURRSort(left, right, SPQ); @@ -1952,62 +2124,26 @@ else if (!LHigh && RHigh) return false; else if (!LHigh && !RHigh) { - // If the two nodes share an operand and one of them has a single - // use that is a live out copy, favor the one that is live out. Otherwise - // it will be difficult to eliminate the copy if the instruction is a - // loop induction variable update. e.g. - // BB: - // sub r1, r3, #1 - // str r0, [r2, r3] - // mov r3, r1 - // cmp - // bne BB - bool SharePred = UnitsSharePred(left, right); - // FIXME: Only adjust if BB is a loop back edge. - // FIXME: What's the cost of a copy? - int LBonus = (SharePred && hasOnlyLiveOutUses(left)) ? 1 : 0; - int RBonus = (SharePred && hasOnlyLiveOutUses(right)) ? 1 : 0; - int LHeight = (int)left->getHeight() - LBonus; - int RHeight = (int)right->getHeight() - RBonus; - - // Low register pressure situation, schedule for latency if possible. - bool LStall = left->SchedulingPref == Sched::Latency && - (int)SPQ->getCurCycle() < LHeight; - bool RStall = right->SchedulingPref == Sched::Latency && - (int)SPQ->getCurCycle() < RHeight; - // If scheduling one of the node will cause a pipeline stall, delay it. - // If scheduling either one of the node will cause a pipeline stall, sort - // them according to their height. - if (LStall) { - if (!RStall) - return true; - if (LHeight != RHeight) - return LHeight > RHeight; - } else if (RStall) - return false; - - // If either node is scheduling for latency, sort them by height - // and latency. - if (left->SchedulingPref == Sched::Latency || - right->SchedulingPref == Sched::Latency) { - if (LHeight != RHeight) - return LHeight > RHeight; - if (left->Latency != right->Latency) - return left->Latency > right->Latency; - } + int result = BUCompareLatency(left, right, true /*checkPref*/, SPQ); + if (result != 0) + return result > 0; } - return BURRSort(left, right, SPQ); } // Schedule as many instructions in each cycle as possible. So don't make an // instruction available unless it is ready in the current cycle. bool ilp_ls_rr_sort::isReady(SUnit *SU, unsigned CurCycle) const { + if (SU->getHeight() > CurCycle) return false; + + if (SPQ->getHazardRec()->getHazardType(SU, 0) + != ScheduleHazardRecognizer::NoHazard) + return false; + return SU->getHeight() <= CurCycle; } -bool ilp_ls_rr_sort::operator()(const SUnit *left, - const SUnit *right) const { +bool ilp_ls_rr_sort::operator()(SUnit *left, SUnit *right) const { if (left->isCall || right->isCall) // No way to compute latency of calls. return BURRSort(left, right, SPQ); @@ -2032,9 +2168,11 @@ return BURRSort(left, right, SPQ); } -template -bool -RegReductionPriorityQueue::canClobber(const SUnit *SU, const SUnit *Op) { +//===----------------------------------------------------------------------===// +// Preschedule for Register Pressure +//===----------------------------------------------------------------------===// + +bool RegReductionPQBase::canClobber(const SUnit *SU, const SUnit *Op) { if (SU->isTwoAddress) { unsigned Opc = SU->getNode()->getMachineOpcode(); const TargetInstrDesc &TID = TII->get(Opc); @@ -2117,8 +2255,7 @@ /// after N, which shortens the U->N live range, reducing /// register pressure. /// -template -void RegReductionPriorityQueue::PrescheduleNodesWithMultipleUses() { +void RegReductionPQBase::PrescheduleNodesWithMultipleUses() { // Visit all the nodes in topological order, working top-down. for (unsigned i = 0, e = SUnits->size(); i != e; ++i) { SUnit *SU = &(*SUnits)[i]; @@ -2210,8 +2347,7 @@ /// one that has a CopyToReg use (more likely to be a loop induction update). /// If both are two-address, but one is commutable while the other is not /// commutable, favor the one that's not commutable. -template -void RegReductionPriorityQueue::AddPseudoTwoAddrDeps() { +void RegReductionPQBase::AddPseudoTwoAddrDeps() { for (unsigned i = 0, e = SUnits->size(); i != e; ++i) { SUnit *SU = &(*SUnits)[i]; if (!SU->isTwoAddress) @@ -2286,16 +2422,6 @@ } } -/// CalculateSethiUllmanNumbers - Calculate Sethi-Ullman numbers of all -/// scheduling units. -template -void RegReductionPriorityQueue::CalculateSethiUllmanNumbers() { - SethiUllmanNumbers.assign(SUnits->size(), 0); - - for (unsigned i = 0, e = SUnits->size(); i != e; ++i) - CalcNodeSethiUllmanNumber(&(*SUnits)[i], SethiUllmanNumbers); -} - /// LimitedSumOfUnscheduledPredsOfSuccs - Compute the sum of the unscheduled /// predecessors of the successors of the SUnit SU. Stop when the provided /// limit is exceeded. From asl at math.spbu.ru Fri Jan 14 15:57:39 2011 From: asl at math.spbu.ru (Anton Korobeynikov) Date: Fri, 14 Jan 2011 21:57:39 -0000 Subject: [llvm-commits] [llvm] r123472 - /llvm/trunk/lib/MC/MCAsmStreamer.cpp Message-ID: <20110114215739.CA79F2A6C12C@llvm.org> Author: asl Date: Fri Jan 14 15:57:39 2011 New Revision: 123472 URL: http://llvm.org/viewvc/llvm-project?rev=123472&view=rev Log: Use common style for .cfi directives Modified: llvm/trunk/lib/MC/MCAsmStreamer.cpp Modified: llvm/trunk/lib/MC/MCAsmStreamer.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/MC/MCAsmStreamer.cpp?rev=123472&r1=123471&r2=123472&view=diff ============================================================================== --- llvm/trunk/lib/MC/MCAsmStreamer.cpp (original) +++ llvm/trunk/lib/MC/MCAsmStreamer.cpp Fri Jan 14 15:57:39 2011 @@ -712,7 +712,7 @@ if (this->MCStreamer::EmitCFIStartProc()) return true; - OS << ".cfi_startproc"; + OS << "\t.cfi_startproc"; EmitEOL(); return false; @@ -722,7 +722,7 @@ if (this->MCStreamer::EmitCFIEndProc()) return true; - OS << ".cfi_endproc"; + OS << "\t.cfi_endproc"; EmitEOL(); return false; @@ -732,7 +732,7 @@ if (this->MCStreamer::EmitCFIDefCfaOffset(Offset)) return true; - OS << ".cfi_def_cfa_offset " << Offset; + OS << "\t.cfi_def_cfa_offset " << Offset; EmitEOL(); return false; @@ -742,7 +742,7 @@ if (this->MCStreamer::EmitCFIDefCfaRegister(Register)) return true; - OS << ".cfi_def_cfa_register " << Register; + OS << "\t.cfi_def_cfa_register " << Register; EmitEOL(); return false; @@ -752,7 +752,7 @@ if (this->MCStreamer::EmitCFIOffset(Register, Offset)) return true; - OS << ".cfi_offset " << Register << ", " << Offset; + OS << "\t.cfi_offset " << Register << ", " << Offset; EmitEOL(); return false; @@ -763,7 +763,7 @@ if (this->MCStreamer::EmitCFIPersonality(Sym, Encoding)) return true; - OS << ".cfi_personality " << Encoding << ", " << *Sym; + OS << "\t.cfi_personality " << Encoding << ", " << *Sym; EmitEOL(); return false; @@ -773,7 +773,7 @@ if (this->MCStreamer::EmitCFILsda(Sym, Encoding)) return true; - OS << ".cfi_lsda " << Encoding << ", " << *Sym; + OS << "\t.cfi_lsda " << Encoding << ", " << *Sym; EmitEOL(); return false; From asl at math.spbu.ru Fri Jan 14 15:57:45 2011 From: asl at math.spbu.ru (Anton Korobeynikov) Date: Fri, 14 Jan 2011 21:57:45 -0000 Subject: [llvm-commits] [llvm] r123473 - in /llvm/trunk/lib/CodeGen/AsmPrinter: AsmPrinter.cpp DwarfException.cpp DwarfException.h DwarfTableException.cpp Message-ID: <20110114215746.060622A6C12D@llvm.org> Author: asl Date: Fri Jan 14 15:57:45 2011 New Revision: 123473 URL: http://llvm.org/viewvc/llvm-project?rev=123473&view=rev Log: Split stuff as a preparation for CFI directives-based frame information emission Added: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp Modified: llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h Modified: llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp?rev=123473&r1=123472&r2=123473&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp Fri Jan 14 15:57:45 2011 @@ -186,9 +186,9 @@ if (MAI->doesSupportDebugInformation()) DD = new DwarfDebug(this, &M); - + if (MAI->doesSupportExceptionHandling()) - DE = new DwarfException(this); + DE = new DwarfTableException(this); return false; } Modified: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp?rev=123473&r1=123472&r2=123473&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp Fri Jan 14 15:57:45 2011 @@ -39,239 +39,10 @@ using namespace llvm; DwarfException::DwarfException(AsmPrinter *A) - : Asm(A), MMI(Asm->MMI), shouldEmitTable(false), shouldEmitMoves(false), - shouldEmitTableModule(false), shouldEmitMovesModule(false) {} + : Asm(A), MMI(Asm->MMI) {} DwarfException::~DwarfException() {} -/// EmitCIE - Emit a Common Information Entry (CIE). This holds information that -/// is shared among many Frame Description Entries. There is at least one CIE -/// in every non-empty .debug_frame section. -void DwarfException::EmitCIE(const Function *PersonalityFn, unsigned Index) { - // Size and sign of stack growth. - int stackGrowth = Asm->getTargetData().getPointerSize(); - if (Asm->TM.getFrameLowering()->getStackGrowthDirection() == - TargetFrameLowering::StackGrowsDown) - stackGrowth *= -1; - - const TargetLoweringObjectFile &TLOF = Asm->getObjFileLowering(); - - // Begin eh frame section. - Asm->OutStreamer.SwitchSection(TLOF.getEHFrameSection()); - - MCSymbol *EHFrameSym; - if (TLOF.isFunctionEHFrameSymbolPrivate()) - EHFrameSym = Asm->GetTempSymbol("EH_frame", Index); - else - EHFrameSym = Asm->OutContext.GetOrCreateSymbol(Twine("EH_frame") + - Twine(Index)); - Asm->OutStreamer.EmitLabel(EHFrameSym); - - Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("section_eh_frame", Index)); - - // Define base labels. - Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_common", Index)); - - // Define the eh frame length. - Asm->OutStreamer.AddComment("Length of Common Information Entry"); - Asm->EmitLabelDifference(Asm->GetTempSymbol("eh_frame_common_end", Index), - Asm->GetTempSymbol("eh_frame_common_begin", Index), - 4); - - // EH frame header. - Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_common_begin",Index)); - Asm->OutStreamer.AddComment("CIE Identifier Tag"); - Asm->OutStreamer.EmitIntValue(0, 4/*size*/, 0/*addrspace*/); - Asm->OutStreamer.AddComment("DW_CIE_VERSION"); - Asm->OutStreamer.EmitIntValue(dwarf::DW_CIE_VERSION, 1/*size*/, 0/*addr*/); - - // The personality presence indicates that language specific information will - // show up in the eh frame. Find out how we are supposed to lower the - // personality function reference: - - unsigned LSDAEncoding = TLOF.getLSDAEncoding(); - unsigned FDEEncoding = TLOF.getFDEEncoding(); - unsigned PerEncoding = TLOF.getPersonalityEncoding(); - - char Augmentation[6] = { 0 }; - unsigned AugmentationSize = 0; - char *APtr = Augmentation + 1; - - if (PersonalityFn) { - // There is a personality function. - *APtr++ = 'P'; - AugmentationSize += 1 + Asm->GetSizeOfEncodedValue(PerEncoding); - } - - if (UsesLSDA[Index]) { - // An LSDA pointer is in the FDE augmentation. - *APtr++ = 'L'; - ++AugmentationSize; - } - - if (FDEEncoding != dwarf::DW_EH_PE_absptr) { - // A non-default pointer encoding for the FDE. - *APtr++ = 'R'; - ++AugmentationSize; - } - - if (APtr != Augmentation + 1) - Augmentation[0] = 'z'; - - Asm->OutStreamer.AddComment("CIE Augmentation"); - Asm->OutStreamer.EmitBytes(StringRef(Augmentation, strlen(Augmentation)+1),0); - - // Round out reader. - Asm->EmitULEB128(1, "CIE Code Alignment Factor"); - Asm->EmitSLEB128(stackGrowth, "CIE Data Alignment Factor"); - Asm->OutStreamer.AddComment("CIE Return Address Column"); - - const TargetRegisterInfo *RI = Asm->TM.getRegisterInfo(); - const TargetFrameLowering *TFI = Asm->TM.getFrameLowering(); - Asm->EmitInt8(RI->getDwarfRegNum(RI->getRARegister(), true)); - - if (Augmentation[0]) { - Asm->EmitULEB128(AugmentationSize, "Augmentation Size"); - - // If there is a personality, we need to indicate the function's location. - if (PersonalityFn) { - Asm->EmitEncodingByte(PerEncoding, "Personality"); - Asm->OutStreamer.AddComment("Personality"); - Asm->EmitReference(PersonalityFn, PerEncoding); - } - if (UsesLSDA[Index]) - Asm->EmitEncodingByte(LSDAEncoding, "LSDA"); - if (FDEEncoding != dwarf::DW_EH_PE_absptr) - Asm->EmitEncodingByte(FDEEncoding, "FDE"); - } - - // Indicate locations of general callee saved registers in frame. - std::vector Moves; - TFI->getInitialFrameState(Moves); - Asm->EmitFrameMoves(Moves, 0, true); - - // On Darwin the linker honors the alignment of eh_frame, which means it must - // be 8-byte on 64-bit targets to match what gcc does. Otherwise you get - // holes which confuse readers of eh_frame. - Asm->EmitAlignment(Asm->getTargetData().getPointerSize() == 4 ? 2 : 3); - Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_common_end", Index)); -} - -/// EmitFDE - Emit the Frame Description Entry (FDE) for the function. -void DwarfException::EmitFDE(const FunctionEHFrameInfo &EHFrameInfo) { - assert(!EHFrameInfo.function->hasAvailableExternallyLinkage() && - "Should not emit 'available externally' functions at all"); - - const Function *TheFunc = EHFrameInfo.function; - const TargetLoweringObjectFile &TLOF = Asm->getObjFileLowering(); - - unsigned LSDAEncoding = TLOF.getLSDAEncoding(); - unsigned FDEEncoding = TLOF.getFDEEncoding(); - - Asm->OutStreamer.SwitchSection(TLOF.getEHFrameSection()); - - // Externally visible entry into the functions eh frame info. If the - // corresponding function is static, this should not be externally visible. - if (!TheFunc->hasLocalLinkage() && TLOF.isFunctionEHSymbolGlobal()) - Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym,MCSA_Global); - - // If corresponding function is weak definition, this should be too. - if (TheFunc->isWeakForLinker() && Asm->MAI->getWeakDefDirective()) - Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym, - MCSA_WeakDefinition); - - // If corresponding function is hidden, this should be too. - if (TheFunc->hasHiddenVisibility()) - if (MCSymbolAttr HiddenAttr = Asm->MAI->getHiddenVisibilityAttr()) - Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym, - HiddenAttr); - - // If there are no calls then you can't unwind. This may mean we can omit the - // EH Frame, but some environments do not handle weak absolute symbols. If - // UnwindTablesMandatory is set we cannot do this optimization; the unwind - // info is to be available for non-EH uses. - if (!EHFrameInfo.adjustsStack && !UnwindTablesMandatory && - (!TheFunc->isWeakForLinker() || - !Asm->MAI->getWeakDefDirective() || - TLOF.getSupportsWeakOmittedEHFrame())) { - Asm->OutStreamer.EmitAssignment(EHFrameInfo.FunctionEHSym, - MCConstantExpr::Create(0, Asm->OutContext)); - // This name has no connection to the function, so it might get - // dead-stripped when the function is not, erroneously. Prohibit - // dead-stripping unconditionally. - if (Asm->MAI->hasNoDeadStrip()) - Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym, - MCSA_NoDeadStrip); - } else { - Asm->OutStreamer.EmitLabel(EHFrameInfo.FunctionEHSym); - - // EH frame header. - Asm->OutStreamer.AddComment("Length of Frame Information Entry"); - Asm->EmitLabelDifference( - Asm->GetTempSymbol("eh_frame_end", EHFrameInfo.Number), - Asm->GetTempSymbol("eh_frame_begin", EHFrameInfo.Number), 4); - - Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_begin", - EHFrameInfo.Number)); - - Asm->OutStreamer.AddComment("FDE CIE offset"); - Asm->EmitLabelDifference( - Asm->GetTempSymbol("eh_frame_begin", EHFrameInfo.Number), - Asm->GetTempSymbol("eh_frame_common", - EHFrameInfo.PersonalityIndex), 4); - - MCSymbol *EHFuncBeginSym = - Asm->GetTempSymbol("eh_func_begin", EHFrameInfo.Number); - - Asm->OutStreamer.AddComment("FDE initial location"); - Asm->EmitReference(EHFuncBeginSym, FDEEncoding); - - Asm->OutStreamer.AddComment("FDE address range"); - Asm->EmitLabelDifference(Asm->GetTempSymbol("eh_func_end", - EHFrameInfo.Number), - EHFuncBeginSym, - Asm->GetSizeOfEncodedValue(FDEEncoding)); - - // If there is a personality and landing pads then point to the language - // specific data area in the exception table. - if (MMI->getPersonalities()[0] != NULL) { - unsigned Size = Asm->GetSizeOfEncodedValue(LSDAEncoding); - - Asm->EmitULEB128(Size, "Augmentation size"); - Asm->OutStreamer.AddComment("Language Specific Data Area"); - if (EHFrameInfo.hasLandingPads) - Asm->EmitReference(Asm->GetTempSymbol("exception", EHFrameInfo.Number), - LSDAEncoding); - else - Asm->OutStreamer.EmitIntValue(0, Size/*size*/, 0/*addrspace*/); - - } else { - Asm->EmitULEB128(0, "Augmentation size"); - } - - // Indicate locations of function specific callee saved registers in frame. - Asm->EmitFrameMoves(EHFrameInfo.Moves, EHFuncBeginSym, true); - - // On Darwin the linker honors the alignment of eh_frame, which means it - // must be 8-byte on 64-bit targets to match what gcc does. Otherwise you - // get holes which confuse readers of eh_frame. - Asm->EmitAlignment(Asm->getTargetData().getPointerSize() == 4 ? 2 : 3); - Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_end", - EHFrameInfo.Number)); - - // If the function is marked used, this table should be also. We cannot - // make the mark unconditional in this case, since retaining the table also - // retains the function in this case, and there is code around that depends - // on unused functions (calling undefined externals) being dead-stripped to - // link correctly. Yes, there really is. - if (MMI->isUsedFunction(EHFrameInfo.function)) - if (Asm->MAI->hasNoDeadStrip()) - Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym, - MCSA_NoDeadStrip); - } - Asm->OutStreamer.AddBlankLine(); -} - /// SharedTypeIds - How many leading type ids two landing pads have in common. unsigned DwarfException::SharedTypeIds(const LandingPadInfo *L, const LandingPadInfo *R) { @@ -423,7 +194,7 @@ const MachineOperand &MO = MI->getOperand(I); if (!MO.isGlobal()) continue; - + const Function *F = dyn_cast(MO.getGlobal()); if (F == 0) continue; @@ -431,7 +202,7 @@ // Be conservative. If we have more than one function operand for this // call, then we can't make the assumption that it's the callee and // not a parameter to the call. - // + // // FIXME: Determine if there's a way to say that `F' is the callee or // parameter. MarkedNoUnwind = false; @@ -621,7 +392,7 @@ // Call sites. bool IsSJLJ = Asm->MAI->getExceptionHandlingType() == ExceptionHandling::SjLj; bool HaveTTData = IsSJLJ ? (!TypeInfos.empty() || !FilterIds.empty()) : true; - + unsigned CallSiteTableLength; if (IsSJLJ) CallSiteTableLength = 0; @@ -629,7 +400,7 @@ unsigned SiteStartSize = 4; // dwarf::DW_EH_PE_udata4 unsigned SiteLengthSize = 4; // dwarf::DW_EH_PE_udata4 unsigned LandingPadSize = 4; // dwarf::DW_EH_PE_udata4 - CallSiteTableLength = + CallSiteTableLength = CallSites.size() * (SiteStartSize + SiteLengthSize + LandingPadSize); } @@ -657,15 +428,15 @@ // mode, this reference will require a relocation by the dynamic linker. // // Because of this, we have a couple of options: - // + // // 1) If we are in -static mode, we can always use an absolute reference // from the LSDA, because the static linker will resolve it. - // + // // 2) Otherwise, if the LSDA section is writable, we can output the direct // reference to the typeinfo and allow the dynamic linker to relocate // it. Since it is in a writable section, the dynamic linker won't // have a problem. - // + // // 3) Finally, if we're in PIC mode and the LDSA section isn't writable, // we need to use some form of indirection. For example, on Darwin, // we can output a statically-relocatable reference to a dyld stub. The @@ -687,7 +458,7 @@ Asm->EmitAlignment(2); // Emit the LSDA. - MCSymbol *GCCETSym = + MCSymbol *GCCETSym = Asm->OutContext.GetOrCreateSymbol(Twine("GCC_except_table")+ Twine(Asm->getFunctionNumber())); Asm->OutStreamer.EmitLabel(GCCETSym); @@ -794,23 +565,23 @@ for (SmallVectorImpl::const_iterator I = CallSites.begin(), E = CallSites.end(); I != E; ++I) { const CallSiteEntry &S = *I; - + MCSymbol *EHFuncBeginSym = Asm->GetTempSymbol("eh_func_begin", Asm->getFunctionNumber()); - + MCSymbol *BeginLabel = S.BeginLabel; if (BeginLabel == 0) BeginLabel = EHFuncBeginSym; MCSymbol *EndLabel = S.EndLabel; if (EndLabel == 0) EndLabel = Asm->GetTempSymbol("eh_func_end", Asm->getFunctionNumber()); - + // Offset of the call site relative to the previous call site, counted in // number of 16-byte bundles. The first call site is counted relative to // the start of the procedure fragment. Asm->OutStreamer.AddComment("Region start"); Asm->EmitLabelDifference(BeginLabel, EHFuncBeginSym, 4); - + Asm->OutStreamer.AddComment("Region length"); Asm->EmitLabelDifference(EndLabel, BeginLabel, 4); @@ -835,7 +606,7 @@ Asm->OutStreamer.AddComment("-- Action Record Table --"); Asm->OutStreamer.AddBlankLine(); } - + for (SmallVectorImpl::const_iterator I = Actions.begin(), E = Actions.end(); I != E; ++I) { const ActionEntry &Action = *I; @@ -889,73 +660,17 @@ /// EndModule - Emit all exception information that should come after the /// content. void DwarfException::EndModule() { - if (Asm->MAI->getExceptionHandlingType() != ExceptionHandling::Dwarf) - return; - - if (!shouldEmitMovesModule && !shouldEmitTableModule) - return; - - const std::vector &Personalities = MMI->getPersonalities(); - - for (unsigned I = 0, E = Personalities.size(); I < E; ++I) - EmitCIE(Personalities[I], I); - - for (std::vector::iterator - I = EHFrames.begin(), E = EHFrames.end(); I != E; ++I) - EmitFDE(*I); + assert(0 && "Should be implemented"); } /// BeginFunction - Gather pre-function exception information. Assumes it's /// being emitted immediately after the function entry point. void DwarfException::BeginFunction(const MachineFunction *MF) { - shouldEmitTable = shouldEmitMoves = false; - - // If any landing pads survive, we need an EH table. - shouldEmitTable = !MMI->getLandingPads().empty(); - - // See if we need frame move info. - shouldEmitMoves = - !Asm->MF->getFunction()->doesNotThrow() || UnwindTablesMandatory; - - if (shouldEmitMoves || shouldEmitTable) - // Assumes in correct section after the entry point. - Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_func_begin", - Asm->getFunctionNumber())); - - shouldEmitTableModule |= shouldEmitTable; - shouldEmitMovesModule |= shouldEmitMoves; + assert(0 && "Should be implemented"); } /// EndFunction - Gather and emit post-function exception information. /// void DwarfException::EndFunction() { - if (!shouldEmitMoves && !shouldEmitTable) return; - - Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_func_end", - Asm->getFunctionNumber())); - - // Record if this personality index uses a landing pad. - bool HasLandingPad = !MMI->getLandingPads().empty(); - UsesLSDA[MMI->getPersonalityIndex()] |= HasLandingPad; - - // Map all labels and get rid of any dead landing pads. - MMI->TidyLandingPads(); - - if (HasLandingPad) - EmitExceptionTable(); - - const TargetLoweringObjectFile &TLOF = Asm->getObjFileLowering(); - MCSymbol *FunctionEHSym = - Asm->GetSymbolWithGlobalValueBase(Asm->MF->getFunction(), ".eh", - TLOF.isFunctionEHFrameSymbolPrivate()); - - // Save EH frame information - EHFrames. - push_back(FunctionEHFrameInfo(FunctionEHSym, - Asm->getFunctionNumber(), - MMI->getPersonalityIndex(), - Asm->MF->getFrameInfo()->adjustsStack(), - !MMI->getLandingPads().empty(), - MMI->getFrameMoves(), - Asm->MF->getFunction())); + assert(0 && "Should be implemented"); } Modified: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h?rev=123473&r1=123472&r2=123473&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h Fri Jan 14 15:57:45 2011 @@ -35,60 +35,13 @@ /// DwarfException - Emits Dwarf exception handling directives. /// class DwarfException { +protected: /// Asm - Target of Dwarf emission. AsmPrinter *Asm; /// MMI - Collected machine module information. MachineModuleInfo *MMI; - struct FunctionEHFrameInfo { - MCSymbol *FunctionEHSym; // L_foo.eh - unsigned Number; - unsigned PersonalityIndex; - bool adjustsStack; - bool hasLandingPads; - std::vector Moves; - const Function *function; - - FunctionEHFrameInfo(MCSymbol *EHSym, unsigned Num, unsigned P, - bool hC, bool hL, - const std::vector &M, - const Function *f): - FunctionEHSym(EHSym), Number(Num), PersonalityIndex(P), - adjustsStack(hC), hasLandingPads(hL), Moves(M), function (f) { } - }; - - std::vector EHFrames; - - /// UsesLSDA - Indicates whether an FDE that uses the CIE at the given index - /// uses an LSDA. If so, then we need to encode that information in the CIE's - /// augmentation. - DenseMap UsesLSDA; - - /// shouldEmitTable - Per-function flag to indicate if EH tables should - /// be emitted. - bool shouldEmitTable; - - /// shouldEmitMoves - Per-function flag to indicate if frame moves info - /// should be emitted. - bool shouldEmitMoves; - - /// shouldEmitTableModule - Per-module flag to indicate if EH tables - /// should be emitted. - bool shouldEmitTableModule; - - /// shouldEmitFrameModule - Per-module flag to indicate if frame moves - /// should be emitted. - bool shouldEmitMovesModule; - - /// EmitCIE - Emit a Common Information Entry (CIE). This holds information - /// that is shared among many Frame Description Entries. There is at least - /// one CIE in every non-empty .debug_frame section. - void EmitCIE(const Function *Personality, unsigned Index); - - /// EmitFDE - Emit the Frame Description Entry (FDE) for the function. - void EmitFDE(const FunctionEHFrameInfo &EHFrameInfo); - /// EmitExceptionTable - Emit landing pads and actions. /// /// The general organization of the table is complex, but the basic concepts @@ -172,18 +125,85 @@ // Main entry points. // DwarfException(AsmPrinter *A); - ~DwarfException(); + virtual ~DwarfException(); + + /// EndModule - Emit all exception information that should come after the + /// content. + virtual void EndModule(); + + /// BeginFunction - Gather pre-function exception information. Assumes being + /// emitted immediately after the function entry point. + virtual void BeginFunction(const MachineFunction *MF); + + /// EndFunction - Gather and emit post-function exception information. + virtual void EndFunction(); +}; + +class DwarfTableException : public DwarfException { + /// shouldEmitTable - Per-function flag to indicate if EH tables should + /// be emitted. + bool shouldEmitTable; + + /// shouldEmitMoves - Per-function flag to indicate if frame moves info + /// should be emitted. + bool shouldEmitMoves; + + /// shouldEmitTableModule - Per-module flag to indicate if EH tables + /// should be emitted. + bool shouldEmitTableModule; + + /// shouldEmitMovesModule - Per-module flag to indicate if frame moves + /// should be emitted. + bool shouldEmitMovesModule; + + struct FunctionEHFrameInfo { + MCSymbol *FunctionEHSym; // L_foo.eh + unsigned Number; + unsigned PersonalityIndex; + bool adjustsStack; + bool hasLandingPads; + std::vector Moves; + const Function *function; + + FunctionEHFrameInfo(MCSymbol *EHSym, unsigned Num, unsigned P, + bool hC, bool hL, + const std::vector &M, + const Function *f): + FunctionEHSym(EHSym), Number(Num), PersonalityIndex(P), + adjustsStack(hC), hasLandingPads(hL), Moves(M), function (f) { } + }; + + std::vector EHFrames; + + /// UsesLSDA - Indicates whether an FDE that uses the CIE at the given index + /// uses an LSDA. If so, then we need to encode that information in the CIE's + /// augmentation. + DenseMap UsesLSDA; + + /// EmitCIE - Emit a Common Information Entry (CIE). This holds information + /// that is shared among many Frame Description Entries. There is at least + /// one CIE in every non-empty .debug_frame section. + void EmitCIE(const Function *Personality, unsigned Index); + + /// EmitFDE - Emit the Frame Description Entry (FDE) for the function. + void EmitFDE(const FunctionEHFrameInfo &EHFrameInfo); +public: + //===--------------------------------------------------------------------===// + // Main entry points. + // + DwarfTableException(AsmPrinter *A); + virtual ~DwarfTableException(); /// EndModule - Emit all exception information that should come after the /// content. - void EndModule(); + virtual void EndModule(); /// BeginFunction - Gather pre-function exception information. Assumes being /// emitted immediately after the function entry point. - void BeginFunction(const MachineFunction *MF); + virtual void BeginFunction(const MachineFunction *MF); /// EndFunction - Gather and emit post-function exception information. - void EndFunction(); + virtual void EndFunction(); }; } // End of namespace llvm Added: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp?rev=123473&view=auto ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp (added) +++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp Fri Jan 14 15:57:45 2011 @@ -0,0 +1,349 @@ +//===-- CodeGen/AsmPrinter/DwarfTableException.cpp - Dwarf Exception Impl --==// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file contains support for writing DWARF exception info into asm files. +// The implementation emits all the necessary tables "by hands". +// +//===----------------------------------------------------------------------===// + +#include "DwarfException.h" +#include "llvm/Module.h" +#include "llvm/CodeGen/AsmPrinter.h" +#include "llvm/CodeGen/MachineModuleInfo.h" +#include "llvm/CodeGen/MachineFrameInfo.h" +#include "llvm/CodeGen/MachineFunction.h" +#include "llvm/CodeGen/MachineLocation.h" +#include "llvm/MC/MCAsmInfo.h" +#include "llvm/MC/MCContext.h" +#include "llvm/MC/MCExpr.h" +#include "llvm/MC/MCSection.h" +#include "llvm/MC/MCStreamer.h" +#include "llvm/MC/MCSymbol.h" +#include "llvm/Target/Mangler.h" +#include "llvm/Target/TargetData.h" +#include "llvm/Target/TargetFrameLowering.h" +#include "llvm/Target/TargetLoweringObjectFile.h" +#include "llvm/Target/TargetMachine.h" +#include "llvm/Target/TargetOptions.h" +#include "llvm/Target/TargetRegisterInfo.h" +#include "llvm/Support/Dwarf.h" +#include "llvm/Support/FormattedStream.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/StringExtras.h" +#include "llvm/ADT/Twine.h" +using namespace llvm; + +DwarfTableException::DwarfTableException(AsmPrinter *A) + : DwarfException(A), + shouldEmitTable(false), shouldEmitMoves(false), + shouldEmitTableModule(false), shouldEmitMovesModule(false) {} + +DwarfTableException::~DwarfTableException() {} + +/// EmitCIE - Emit a Common Information Entry (CIE). This holds information that +/// is shared among many Frame Description Entries. There is at least one CIE +/// in every non-empty .debug_frame section. +void DwarfTableException::EmitCIE(const Function *PersonalityFn, unsigned Index) { + // Size and sign of stack growth. + int stackGrowth = Asm->getTargetData().getPointerSize(); + if (Asm->TM.getFrameLowering()->getStackGrowthDirection() == + TargetFrameLowering::StackGrowsDown) + stackGrowth *= -1; + + const TargetLoweringObjectFile &TLOF = Asm->getObjFileLowering(); + + // Begin eh frame section. + Asm->OutStreamer.SwitchSection(TLOF.getEHFrameSection()); + + MCSymbol *EHFrameSym; + if (TLOF.isFunctionEHFrameSymbolPrivate()) + EHFrameSym = Asm->GetTempSymbol("EH_frame", Index); + else + EHFrameSym = Asm->OutContext.GetOrCreateSymbol(Twine("EH_frame") + + Twine(Index)); + Asm->OutStreamer.EmitLabel(EHFrameSym); + + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("section_eh_frame", Index)); + + // Define base labels. + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_common", Index)); + + // Define the eh frame length. + Asm->OutStreamer.AddComment("Length of Common Information Entry"); + Asm->EmitLabelDifference(Asm->GetTempSymbol("eh_frame_common_end", Index), + Asm->GetTempSymbol("eh_frame_common_begin", Index), + 4); + + // EH frame header. + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_common_begin",Index)); + Asm->OutStreamer.AddComment("CIE Identifier Tag"); + Asm->OutStreamer.EmitIntValue(0, 4/*size*/, 0/*addrspace*/); + Asm->OutStreamer.AddComment("DW_CIE_VERSION"); + Asm->OutStreamer.EmitIntValue(dwarf::DW_CIE_VERSION, 1/*size*/, 0/*addr*/); + + // The personality presence indicates that language specific information will + // show up in the eh frame. Find out how we are supposed to lower the + // personality function reference: + + unsigned LSDAEncoding = TLOF.getLSDAEncoding(); + unsigned FDEEncoding = TLOF.getFDEEncoding(); + unsigned PerEncoding = TLOF.getPersonalityEncoding(); + + char Augmentation[6] = { 0 }; + unsigned AugmentationSize = 0; + char *APtr = Augmentation + 1; + + if (PersonalityFn) { + // There is a personality function. + *APtr++ = 'P'; + AugmentationSize += 1 + Asm->GetSizeOfEncodedValue(PerEncoding); + } + + if (UsesLSDA[Index]) { + // An LSDA pointer is in the FDE augmentation. + *APtr++ = 'L'; + ++AugmentationSize; + } + + if (FDEEncoding != dwarf::DW_EH_PE_absptr) { + // A non-default pointer encoding for the FDE. + *APtr++ = 'R'; + ++AugmentationSize; + } + + if (APtr != Augmentation + 1) + Augmentation[0] = 'z'; + + Asm->OutStreamer.AddComment("CIE Augmentation"); + Asm->OutStreamer.EmitBytes(StringRef(Augmentation, strlen(Augmentation)+1),0); + + // Round out reader. + Asm->EmitULEB128(1, "CIE Code Alignment Factor"); + Asm->EmitSLEB128(stackGrowth, "CIE Data Alignment Factor"); + Asm->OutStreamer.AddComment("CIE Return Address Column"); + + const TargetRegisterInfo *RI = Asm->TM.getRegisterInfo(); + const TargetFrameLowering *TFI = Asm->TM.getFrameLowering(); + Asm->EmitInt8(RI->getDwarfRegNum(RI->getRARegister(), true)); + + if (Augmentation[0]) { + Asm->EmitULEB128(AugmentationSize, "Augmentation Size"); + + // If there is a personality, we need to indicate the function's location. + if (PersonalityFn) { + Asm->EmitEncodingByte(PerEncoding, "Personality"); + Asm->OutStreamer.AddComment("Personality"); + Asm->EmitReference(PersonalityFn, PerEncoding); + } + if (UsesLSDA[Index]) + Asm->EmitEncodingByte(LSDAEncoding, "LSDA"); + if (FDEEncoding != dwarf::DW_EH_PE_absptr) + Asm->EmitEncodingByte(FDEEncoding, "FDE"); + } + + // Indicate locations of general callee saved registers in frame. + std::vector Moves; + TFI->getInitialFrameState(Moves); + Asm->EmitFrameMoves(Moves, 0, true); + + // On Darwin the linker honors the alignment of eh_frame, which means it must + // be 8-byte on 64-bit targets to match what gcc does. Otherwise you get + // holes which confuse readers of eh_frame. + Asm->EmitAlignment(Asm->getTargetData().getPointerSize() == 4 ? 2 : 3); + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_common_end", Index)); +} + +/// EmitFDE - Emit the Frame Description Entry (FDE) for the function. +void DwarfTableException::EmitFDE(const FunctionEHFrameInfo &EHFrameInfo) { + assert(!EHFrameInfo.function->hasAvailableExternallyLinkage() && + "Should not emit 'available externally' functions at all"); + + const Function *TheFunc = EHFrameInfo.function; + const TargetLoweringObjectFile &TLOF = Asm->getObjFileLowering(); + + unsigned LSDAEncoding = TLOF.getLSDAEncoding(); + unsigned FDEEncoding = TLOF.getFDEEncoding(); + + Asm->OutStreamer.SwitchSection(TLOF.getEHFrameSection()); + + // Externally visible entry into the functions eh frame info. If the + // corresponding function is static, this should not be externally visible. + if (!TheFunc->hasLocalLinkage() && TLOF.isFunctionEHSymbolGlobal()) + Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym,MCSA_Global); + + // If corresponding function is weak definition, this should be too. + if (TheFunc->isWeakForLinker() && Asm->MAI->getWeakDefDirective()) + Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym, + MCSA_WeakDefinition); + + // If corresponding function is hidden, this should be too. + if (TheFunc->hasHiddenVisibility()) + if (MCSymbolAttr HiddenAttr = Asm->MAI->getHiddenVisibilityAttr()) + Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym, + HiddenAttr); + + // If there are no calls then you can't unwind. This may mean we can omit the + // EH Frame, but some environments do not handle weak absolute symbols. If + // UnwindTablesMandatory is set we cannot do this optimization; the unwind + // info is to be available for non-EH uses. + if (!EHFrameInfo.adjustsStack && !UnwindTablesMandatory && + (!TheFunc->isWeakForLinker() || + !Asm->MAI->getWeakDefDirective() || + TLOF.getSupportsWeakOmittedEHFrame())) { + Asm->OutStreamer.EmitAssignment(EHFrameInfo.FunctionEHSym, + MCConstantExpr::Create(0, Asm->OutContext)); + // This name has no connection to the function, so it might get + // dead-stripped when the function is not, erroneously. Prohibit + // dead-stripping unconditionally. + if (Asm->MAI->hasNoDeadStrip()) + Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym, + MCSA_NoDeadStrip); + } else { + Asm->OutStreamer.EmitLabel(EHFrameInfo.FunctionEHSym); + + // EH frame header. + Asm->OutStreamer.AddComment("Length of Frame Information Entry"); + Asm->EmitLabelDifference( + Asm->GetTempSymbol("eh_frame_end", EHFrameInfo.Number), + Asm->GetTempSymbol("eh_frame_begin", EHFrameInfo.Number), 4); + + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_begin", + EHFrameInfo.Number)); + + Asm->OutStreamer.AddComment("FDE CIE offset"); + Asm->EmitLabelDifference( + Asm->GetTempSymbol("eh_frame_begin", EHFrameInfo.Number), + Asm->GetTempSymbol("eh_frame_common", + EHFrameInfo.PersonalityIndex), 4); + + MCSymbol *EHFuncBeginSym = + Asm->GetTempSymbol("eh_func_begin", EHFrameInfo.Number); + + Asm->OutStreamer.AddComment("FDE initial location"); + Asm->EmitReference(EHFuncBeginSym, FDEEncoding); + + Asm->OutStreamer.AddComment("FDE address range"); + Asm->EmitLabelDifference(Asm->GetTempSymbol("eh_func_end", + EHFrameInfo.Number), + EHFuncBeginSym, + Asm->GetSizeOfEncodedValue(FDEEncoding)); + + // If there is a personality and landing pads then point to the language + // specific data area in the exception table. + if (MMI->getPersonalities()[0] != NULL) { + unsigned Size = Asm->GetSizeOfEncodedValue(LSDAEncoding); + + Asm->EmitULEB128(Size, "Augmentation size"); + Asm->OutStreamer.AddComment("Language Specific Data Area"); + if (EHFrameInfo.hasLandingPads) + Asm->EmitReference(Asm->GetTempSymbol("exception", EHFrameInfo.Number), + LSDAEncoding); + else + Asm->OutStreamer.EmitIntValue(0, Size/*size*/, 0/*addrspace*/); + + } else { + Asm->EmitULEB128(0, "Augmentation size"); + } + + // Indicate locations of function specific callee saved registers in frame. + Asm->EmitFrameMoves(EHFrameInfo.Moves, EHFuncBeginSym, true); + + // On Darwin the linker honors the alignment of eh_frame, which means it + // must be 8-byte on 64-bit targets to match what gcc does. Otherwise you + // get holes which confuse readers of eh_frame. + Asm->EmitAlignment(Asm->getTargetData().getPointerSize() == 4 ? 2 : 3); + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_frame_end", + EHFrameInfo.Number)); + + // If the function is marked used, this table should be also. We cannot + // make the mark unconditional in this case, since retaining the table also + // retains the function in this case, and there is code around that depends + // on unused functions (calling undefined externals) being dead-stripped to + // link correctly. Yes, there really is. + if (MMI->isUsedFunction(EHFrameInfo.function)) + if (Asm->MAI->hasNoDeadStrip()) + Asm->OutStreamer.EmitSymbolAttribute(EHFrameInfo.FunctionEHSym, + MCSA_NoDeadStrip); + } + Asm->OutStreamer.AddBlankLine(); +} + +/// EndModule - Emit all exception information that should come after the +/// content. +void DwarfTableException::EndModule() { + if (Asm->MAI->getExceptionHandlingType() != ExceptionHandling::Dwarf) + return; + + if (!shouldEmitMovesModule && !shouldEmitTableModule) + return; + + const std::vector &Personalities = MMI->getPersonalities(); + + for (unsigned I = 0, E = Personalities.size(); I < E; ++I) + EmitCIE(Personalities[I], I); + + for (std::vector::iterator + I = EHFrames.begin(), E = EHFrames.end(); I != E; ++I) + EmitFDE(*I); +} + +/// BeginFunction - Gather pre-function exception information. Assumes it's +/// being emitted immediately after the function entry point. +void DwarfTableException::BeginFunction(const MachineFunction *MF) { + shouldEmitTable = shouldEmitMoves = false; + + // If any landing pads survive, we need an EH table. + shouldEmitTable = !MMI->getLandingPads().empty(); + + // See if we need frame move info. + shouldEmitMoves = + !Asm->MF->getFunction()->doesNotThrow() || UnwindTablesMandatory; + + if (shouldEmitMoves || shouldEmitTable) + // Assumes in correct section after the entry point. + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_func_begin", + Asm->getFunctionNumber())); + + shouldEmitTableModule |= shouldEmitTable; + shouldEmitMovesModule |= shouldEmitMoves; +} + +/// EndFunction - Gather and emit post-function exception information. +/// +void DwarfTableException::EndFunction() { + if (!shouldEmitMoves && !shouldEmitTable) return; + + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_func_end", + Asm->getFunctionNumber())); + + // Record if this personality index uses a landing pad. + bool HasLandingPad = !MMI->getLandingPads().empty(); + UsesLSDA[MMI->getPersonalityIndex()] |= HasLandingPad; + + // Map all labels and get rid of any dead landing pads. + MMI->TidyLandingPads(); + + if (HasLandingPad) + EmitExceptionTable(); + + const TargetLoweringObjectFile &TLOF = Asm->getObjFileLowering(); + MCSymbol *FunctionEHSym = + Asm->GetSymbolWithGlobalValueBase(Asm->MF->getFunction(), ".eh", + TLOF.isFunctionEHFrameSymbolPrivate()); + + // Save EH frame information + EHFrames. + push_back(FunctionEHFrameInfo(FunctionEHSym, + Asm->getFunctionNumber(), + MMI->getPersonalityIndex(), + Asm->MF->getFrameInfo()->adjustsStack(), + !MMI->getLandingPads().empty(), + MMI->getFrameMoves(), + Asm->MF->getFunction())); +} From asl at math.spbu.ru Fri Jan 14 15:57:53 2011 From: asl at math.spbu.ru (Anton Korobeynikov) Date: Fri, 14 Jan 2011 21:57:53 -0000 Subject: [llvm-commits] [llvm] r123474 - in /llvm/trunk: include/llvm/CodeGen/AsmPrinter.h lib/CodeGen/AsmPrinter/AsmPrinterDwarf.cpp lib/CodeGen/AsmPrinter/DwarfCFIException.cpp lib/CodeGen/AsmPrinter/DwarfException.h Message-ID: <20110114215753.5D7252A6C12C@llvm.org> Author: asl Date: Fri Jan 14 15:57:53 2011 New Revision: 123474 URL: http://llvm.org/viewvc/llvm-project?rev=123474&view=rev Log: Add CFI directives-based frame information emission. Not hooked yet. Added: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp Modified: llvm/trunk/include/llvm/CodeGen/AsmPrinter.h llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinterDwarf.cpp llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h Modified: llvm/trunk/include/llvm/CodeGen/AsmPrinter.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/CodeGen/AsmPrinter.h?rev=123474&r1=123473&r2=123474&view=diff ============================================================================== --- llvm/trunk/include/llvm/CodeGen/AsmPrinter.h (original) +++ llvm/trunk/include/llvm/CodeGen/AsmPrinter.h Fri Jan 14 15:57:53 2011 @@ -389,7 +389,7 @@ /// frame. void EmitFrameMoves(const std::vector &Moves, MCSymbol *BaseLabel, bool isEH) const; - + void EmitCFIFrameMoves(const std::vector &Moves) const; //===------------------------------------------------------------------===// // Inline Asm Support Modified: llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinterDwarf.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinterDwarf.cpp?rev=123474&r1=123473&r2=123474&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinterDwarf.cpp (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinterDwarf.cpp Fri Jan 14 15:57:53 2011 @@ -276,3 +276,43 @@ } } } + +/// EmitFrameMoves - Emit frame instructions to describe the layout of the +/// frame. +void AsmPrinter::EmitCFIFrameMoves(const std::vector &Moves) const { + const TargetRegisterInfo *RI = TM.getRegisterInfo(); + + int stackGrowth = TM.getTargetData()->getPointerSize(); + if (TM.getFrameLowering()->getStackGrowthDirection() != + TargetFrameLowering::StackGrowsUp) + stackGrowth *= -1; + + for (unsigned i = 0, N = Moves.size(); i < N; ++i) { + const MachineMove &Move = Moves[i]; + MCSymbol *Label = Move.getLabel(); + // Throw out move if the label is invalid. + if (Label && !Label->isDefined()) continue; // Not emitted, in dead code. + + const MachineLocation &Dst = Move.getDestination(); + const MachineLocation &Src = Move.getSource(); + + // If advancing cfa. + if (Dst.isReg() && Dst.getReg() == MachineLocation::VirtualFP) { + assert(!Src.isReg() && "Machine move not supported yet."); + + if (Src.getReg() == MachineLocation::VirtualFP) { + OutStreamer.EmitCFIDefCfaOffset(-Src.getOffset()); + } else { + assert("Machine move not supported yet"); + // Reg + Offset + } + } else if (Src.isReg() && Src.getReg() == MachineLocation::VirtualFP) { + assert(Dst.isReg() && "Machine move not supported yet."); + OutStreamer.EmitCFIDefCfaRegister(RI->getDwarfRegNum(Dst.getReg(), true)); + } else { + assert(!Dst.isReg() && "Machine move not supported yet."); + OutStreamer.EmitCFIOffset(RI->getDwarfRegNum(Src.getReg(), true), + Dst.getOffset()); + } + } +} Added: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp?rev=123474&view=auto ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp (added) +++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp Fri Jan 14 15:57:53 2011 @@ -0,0 +1,138 @@ +//===-- CodeGen/AsmPrinter/DwarfException.cpp - Dwarf Exception Impl ------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file contains support for writing DWARF exception info into asm files. +// +//===----------------------------------------------------------------------===// + +#include "DwarfException.h" +#include "llvm/Module.h" +#include "llvm/CodeGen/AsmPrinter.h" +#include "llvm/CodeGen/MachineModuleInfo.h" +#include "llvm/CodeGen/MachineFrameInfo.h" +#include "llvm/CodeGen/MachineFunction.h" +#include "llvm/CodeGen/MachineLocation.h" +#include "llvm/MC/MCAsmInfo.h" +#include "llvm/MC/MCContext.h" +#include "llvm/MC/MCExpr.h" +#include "llvm/MC/MCSection.h" +#include "llvm/MC/MCStreamer.h" +#include "llvm/MC/MCSymbol.h" +#include "llvm/Target/Mangler.h" +#include "llvm/Target/TargetData.h" +#include "llvm/Target/TargetFrameLowering.h" +#include "llvm/Target/TargetLoweringObjectFile.h" +#include "llvm/Target/TargetMachine.h" +#include "llvm/Target/TargetOptions.h" +#include "llvm/Target/TargetRegisterInfo.h" +#include "llvm/Support/Dwarf.h" +#include "llvm/Support/FormattedStream.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/StringExtras.h" +#include "llvm/ADT/Twine.h" +using namespace llvm; + +DwarfCFIException::DwarfCFIException(AsmPrinter *A) + : DwarfException(A), + shouldEmitTable(false), shouldEmitMoves(false), shouldEmitTableModule(false) + {} + +DwarfCFIException::~DwarfCFIException() {} + +/// EndModule - Emit all exception information that should come after the +/// content. +void DwarfCFIException::EndModule() { + if (Asm->MAI->getExceptionHandlingType() != ExceptionHandling::Dwarf) + return; + + if (!shouldEmitTableModule) + return; + + const TargetLoweringObjectFile &TLOF = Asm->getObjFileLowering(); + unsigned PerEncoding = TLOF.getPersonalityEncoding(); + + // Begin eh frame section. + Asm->OutStreamer.SwitchSection(TLOF.getEHFrameSection()); + + // Emit references to all used personality functions + const std::vector &Personalities = MMI->getPersonalities(); + for (size_t i = 0, e = Personalities.size(); i != e; ++i) { + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("personality", i)); + Asm->EmitReference(Personalities[i], PerEncoding); + } +} + +/// BeginFunction - Gather pre-function exception information. Assumes it's +/// being emitted immediately after the function entry point. +void DwarfCFIException::BeginFunction(const MachineFunction *MF) { + shouldEmitTable = shouldEmitMoves = false; + + // If any landing pads survive, we need an EH table. + shouldEmitTable = !MMI->getLandingPads().empty(); + + // See if we need frame move info. + shouldEmitMoves = + !Asm->MF->getFunction()->doesNotThrow() || UnwindTablesMandatory; + + if (shouldEmitMoves || shouldEmitTable) + // Assumes in correct section after the entry point. + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_func_begin", + Asm->getFunctionNumber())); + + shouldEmitTableModule |= shouldEmitTable; + + if (shouldEmitMoves) { + const TargetFrameLowering *TFL = Asm->TM.getFrameLowering(); + Asm->OutStreamer.EmitCFIStartProc(); + + // Indicate locations of general callee saved registers in frame. + std::vector Moves; + TFL->getInitialFrameState(Moves); + Asm->EmitCFIFrameMoves(Moves); + Asm->EmitCFIFrameMoves(MMI->getFrameMoves()); + } + + if (!shouldEmitTable) + return; + + const TargetLoweringObjectFile &TLOF = Asm->getObjFileLowering(); + + // Provide LSDA information. + unsigned LSDAEncoding = TLOF.getLSDAEncoding(); + if (LSDAEncoding != dwarf::DW_EH_PE_omit) + Asm->OutStreamer.EmitCFILsda(Asm->GetTempSymbol("exception", + Asm->getFunctionNumber()), + LSDAEncoding); + + // Indicate personality routine, if any. + unsigned PerEncoding = TLOF.getPersonalityEncoding(); + if (PerEncoding != dwarf::DW_EH_PE_omit && + MMI->getPersonalities()[MMI->getPersonalityIndex()]) + Asm->OutStreamer.EmitCFIPersonality(Asm->GetTempSymbol("personality", + MMI->getPersonalityIndex()), + PerEncoding); +} + +/// EndFunction - Gather and emit post-function exception information. +/// +void DwarfCFIException::EndFunction() { + if (!shouldEmitMoves && !shouldEmitTable) return; + + if (shouldEmitMoves) + Asm->OutStreamer.EmitCFIEndProc(); + + Asm->OutStreamer.EmitLabel(Asm->GetTempSymbol("eh_func_end", + Asm->getFunctionNumber())); + + // Map all labels and get rid of any dead landing pads. + MMI->TidyLandingPads(); + + if (shouldEmitTable) + EmitExceptionTable(); +} Modified: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h?rev=123474&r1=123473&r2=123474&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.h Fri Jan 14 15:57:53 2011 @@ -139,6 +139,37 @@ virtual void EndFunction(); }; +class DwarfCFIException : public DwarfException { + /// shouldEmitTable - Per-function flag to indicate if EH tables should + /// be emitted. + bool shouldEmitTable; + + /// shouldEmitMoves - Per-function flag to indicate if frame moves info + /// should be emitted. + bool shouldEmitMoves; + + /// shouldEmitTableModule - Per-module flag to indicate if EH tables + /// should be emitted. + bool shouldEmitTableModule; +public: + //===--------------------------------------------------------------------===// + // Main entry points. + // + DwarfCFIException(AsmPrinter *A); + virtual ~DwarfCFIException(); + + /// EndModule - Emit all exception information that should come after the + /// content. + virtual void EndModule(); + + /// BeginFunction - Gather pre-function exception information. Assumes being + /// emitted immediately after the function entry point. + virtual void BeginFunction(const MachineFunction *MF); + + /// EndFunction - Gather and emit post-function exception information. + virtual void EndFunction(); +}; + class DwarfTableException : public DwarfException { /// shouldEmitTable - Per-function flag to indicate if EH tables should /// be emitted. From asl at math.spbu.ru Fri Jan 14 15:57:58 2011 From: asl at math.spbu.ru (Anton Korobeynikov) Date: Fri, 14 Jan 2011 21:57:58 -0000 Subject: [llvm-commits] [llvm] r123475 - /llvm/trunk/lib/Target/X86/X86FrameLowering.cpp Message-ID: <20110114215758.8BDEE2A6C12D@llvm.org> Author: asl Date: Fri Jan 14 15:57:58 2011 New Revision: 123475 URL: http://llvm.org/viewvc/llvm-project?rev=123475&view=rev Log: Cleanup Modified: llvm/trunk/lib/Target/X86/X86FrameLowering.cpp Modified: llvm/trunk/lib/Target/X86/X86FrameLowering.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86FrameLowering.cpp?rev=123475&r1=123474&r2=123475&view=diff ============================================================================== --- llvm/trunk/lib/Target/X86/X86FrameLowering.cpp (original) +++ llvm/trunk/lib/Target/X86/X86FrameLowering.cpp Fri Jan 14 15:57:58 2011 @@ -291,10 +291,7 @@ bool HasFP = hasFP(MF); // Calculate amount of bytes used for return address storing. - int stackGrowth = - (TM.getFrameLowering()->getStackGrowthDirection() == - TargetFrameLowering::StackGrowsUp ? - TD->getPointerSize() : -TD->getPointerSize()); + int stackGrowth = -TD->getPointerSize(); // FIXME: This is dirty hack. The code itself is pretty mess right now. // It should be rewritten from scratch and generalized sometimes. @@ -467,7 +464,6 @@ MachineLocation SPSrc(MachineLocation::VirtualFP, 2 * stackGrowth); Moves.push_back(MachineMove(FrameLabel, SPDst, SPSrc)); } else { - // FIXME: Verify & implement for FP MachineLocation SPDst(StackPtr); MachineLocation SPSrc(StackPtr, stackGrowth); Moves.push_back(MachineMove(FrameLabel, SPDst, SPSrc)); @@ -627,7 +623,6 @@ -StackSize + stackGrowth); Moves.push_back(MachineMove(Label, SPDst, SPSrc)); } else { - // FIXME: Verify & implement for FP MachineLocation SPDst(StackPtr); MachineLocation SPSrc(StackPtr, stackGrowth); Moves.push_back(MachineMove(Label, SPDst, SPSrc)); From asl at math.spbu.ru Fri Jan 14 15:58:08 2011 From: asl at math.spbu.ru (Anton Korobeynikov) Date: Fri, 14 Jan 2011 21:58:08 -0000 Subject: [llvm-commits] [llvm] r123476 - in /llvm/trunk: include/llvm/MC/MCAsmInfo.h lib/CodeGen/AsmPrinter/AsmPrinter.cpp lib/CodeGen/AsmPrinter/DwarfCFIException.cpp lib/CodeGen/AsmPrinter/DwarfException.cpp lib/CodeGen/AsmPrinter/DwarfTableException.cpp lib/CodeGen/LLVMTargetMachine.cpp lib/Target/PowerPC/PPCMCAsmInfo.cpp lib/Target/X86/X86MCAsmInfo.cpp Message-ID: <20110114215808.ECCBC2A6C12C@llvm.org> Author: asl Date: Fri Jan 14 15:58:08 2011 New Revision: 123476 URL: http://llvm.org/viewvc/llvm-project?rev=123476&view=rev Log: Add a possibility to switch between CFI directives- and table-based frame description emission. Currently all the backends use table-based stuff. Modified: llvm/trunk/include/llvm/MC/MCAsmInfo.h llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp llvm/trunk/lib/CodeGen/LLVMTargetMachine.cpp llvm/trunk/lib/Target/PowerPC/PPCMCAsmInfo.cpp llvm/trunk/lib/Target/X86/X86MCAsmInfo.cpp Modified: llvm/trunk/include/llvm/MC/MCAsmInfo.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/MC/MCAsmInfo.h?rev=123476&r1=123475&r2=123476&view=diff ============================================================================== --- llvm/trunk/include/llvm/MC/MCAsmInfo.h (original) +++ llvm/trunk/include/llvm/MC/MCAsmInfo.h Fri Jan 14 15:58:08 2011 @@ -25,7 +25,9 @@ /// MCAsmInfo - This class is intended to be used as a base class for asm /// properties and features specific to the target. - namespace ExceptionHandling { enum ExceptionsType { None, Dwarf, SjLj }; } + namespace ExceptionHandling { + enum ExceptionsType { None, DwarfTable, DwarfCFI, SjLj }; + } class MCAsmInfo { protected: @@ -438,6 +440,12 @@ ExceptionHandling::ExceptionsType getExceptionHandlingType() const { return ExceptionsType; } + bool isExceptionHandlingDwarf() const { + return + (ExceptionsType == ExceptionHandling::DwarfTable || + ExceptionsType == ExceptionHandling::DwarfCFI); + } + bool doesDwarfRequireFrameSection() const { return DwarfRequiresFrameSection; } Modified: llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp?rev=123476&r1=123475&r2=123476&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp Fri Jan 14 15:58:08 2011 @@ -188,7 +188,15 @@ DD = new DwarfDebug(this, &M); if (MAI->doesSupportExceptionHandling()) - DE = new DwarfTableException(this); + switch (MAI->getExceptionHandlingType()) { + default: + case ExceptionHandling::DwarfTable: + DE = new DwarfTableException(this); + break; + case ExceptionHandling::DwarfCFI: + DE = new DwarfCFIException(this); + break; + } return false; } Modified: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp?rev=123476&r1=123475&r2=123476&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfCFIException.cpp Fri Jan 14 15:58:08 2011 @@ -48,7 +48,7 @@ /// EndModule - Emit all exception information that should come after the /// content. void DwarfCFIException::EndModule() { - if (Asm->MAI->getExceptionHandlingType() != ExceptionHandling::Dwarf) + if (!Asm->MAI->isExceptionHandlingDwarf()) return; if (!shouldEmitTableModule) Modified: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp?rev=123476&r1=123475&r2=123476&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfException.cpp Fri Jan 14 15:58:08 2011 @@ -269,8 +269,7 @@ // instruction between the previous try-range and this one may throw, // create a call-site entry with no landing pad for the region between the // try-ranges. - if (SawPotentiallyThrowing && - Asm->MAI->getExceptionHandlingType() == ExceptionHandling::Dwarf) { + if (SawPotentiallyThrowing && Asm->MAI->isExceptionHandlingDwarf()) { CallSiteEntry Site = { LastLabel, BeginLabel, 0, 0 }; CallSites.push_back(Site); PreviousIsInvoke = false; @@ -292,8 +291,7 @@ }; // Try to merge with the previous call-site. SJLJ doesn't do this - if (PreviousIsInvoke && - Asm->MAI->getExceptionHandlingType() == ExceptionHandling::Dwarf) { + if (PreviousIsInvoke && Asm->MAI->isExceptionHandlingDwarf()) { CallSiteEntry &Prev = CallSites.back(); if (Site.PadLabel == Prev.PadLabel && Site.Action == Prev.Action) { // Extend the range of the previous entry. @@ -303,7 +301,7 @@ } // Otherwise, create a new call-site. - if (Asm->MAI->getExceptionHandlingType() == ExceptionHandling::Dwarf) + if (Asm->MAI->isExceptionHandlingDwarf()) CallSites.push_back(Site); else { // SjLj EH must maintain the call sites in the order assigned @@ -321,8 +319,7 @@ // If some instruction between the previous try-range and the end of the // function may throw, create a call-site entry with no landing pad for the // region following the try-range. - if (SawPotentiallyThrowing && - Asm->MAI->getExceptionHandlingType() == ExceptionHandling::Dwarf) { + if (SawPotentiallyThrowing && Asm->MAI->isExceptionHandlingDwarf()) { CallSiteEntry Site = { LastLabel, 0, 0, 0 }; CallSites.push_back(Site); } @@ -536,7 +533,7 @@ } } else { // DWARF Exception handling - assert(Asm->MAI->getExceptionHandlingType() == ExceptionHandling::Dwarf); + assert(Asm->MAI->isExceptionHandlingDwarf()); // The call-site table is a list of all call sites that may throw an // exception (including C++ 'throw' statements) in the procedure Modified: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp?rev=123476&r1=123475&r2=123476&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfTableException.cpp Fri Jan 14 15:58:08 2011 @@ -277,7 +277,7 @@ /// EndModule - Emit all exception information that should come after the /// content. void DwarfTableException::EndModule() { - if (Asm->MAI->getExceptionHandlingType() != ExceptionHandling::Dwarf) + if (!Asm->MAI->isExceptionHandlingDwarf()) return; if (!shouldEmitMovesModule && !shouldEmitTableModule) Modified: llvm/trunk/lib/CodeGen/LLVMTargetMachine.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/LLVMTargetMachine.cpp?rev=123476&r1=123475&r2=123476&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/LLVMTargetMachine.cpp (original) +++ llvm/trunk/lib/CodeGen/LLVMTargetMachine.cpp Fri Jan 14 15:58:08 2011 @@ -301,7 +301,8 @@ // edge from elsewhere. PM.add(createSjLjEHPass(getTargetLowering())); // FALLTHROUGH - case ExceptionHandling::Dwarf: + case ExceptionHandling::DwarfCFI: + case ExceptionHandling::DwarfTable: PM.add(createDwarfEHPass(this)); break; case ExceptionHandling::None: Modified: llvm/trunk/lib/Target/PowerPC/PPCMCAsmInfo.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCMCAsmInfo.cpp?rev=123476&r1=123475&r2=123476&view=diff ============================================================================== --- llvm/trunk/lib/Target/PowerPC/PPCMCAsmInfo.cpp (original) +++ llvm/trunk/lib/Target/PowerPC/PPCMCAsmInfo.cpp Fri Jan 14 15:58:08 2011 @@ -17,7 +17,7 @@ PPCMCAsmInfoDarwin::PPCMCAsmInfoDarwin(bool is64Bit) { PCSymbol = "."; CommentString = ";"; - ExceptionsType = ExceptionHandling::Dwarf; + ExceptionsType = ExceptionHandling::DwarfTable; if (!is64Bit) Data64bitsDirective = 0; // We can't emit a 64-bit unit in PPC32 mode. @@ -48,7 +48,7 @@ // Exceptions handling if (!is64Bit) - ExceptionsType = ExceptionHandling::Dwarf; + ExceptionsType = ExceptionHandling::DwarfTable; ZeroDirective = "\t.space\t"; Data64bitsDirective = is64Bit ? "\t.quad\t" : 0; Modified: llvm/trunk/lib/Target/X86/X86MCAsmInfo.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86MCAsmInfo.cpp?rev=123476&r1=123475&r2=123476&view=diff ============================================================================== --- llvm/trunk/lib/Target/X86/X86MCAsmInfo.cpp (original) +++ llvm/trunk/lib/Target/X86/X86MCAsmInfo.cpp Fri Jan 14 15:58:08 2011 @@ -68,7 +68,7 @@ DwarfUsesInlineInfoSection = true; // Exceptions handling - ExceptionsType = ExceptionHandling::Dwarf; + ExceptionsType = ExceptionHandling::DwarfTable; } X86ELFMCAsmInfo::X86ELFMCAsmInfo(const Triple &T) { @@ -88,8 +88,8 @@ SupportsDebugInformation = true; // Exceptions handling - ExceptionsType = ExceptionHandling::Dwarf; - + ExceptionsType = ExceptionHandling::DwarfTable; + // OpenBSD has buggy support for .quad in 32-bit mode, just split into two // .words. if (T.getOS() == Triple::OpenBSD && T.getArch() == Triple::x86) From resistor at mac.com Fri Jan 14 16:19:20 2011 From: resistor at mac.com (Owen Anderson) Date: Fri, 14 Jan 2011 22:19:20 -0000 Subject: [llvm-commits] [llvm] r123477 - /llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Message-ID: <20110114221920.6E9742A6C12C@llvm.org> Author: resistor Date: Fri Jan 14 16:19:20 2011 New Revision: 123477 URL: http://llvm.org/viewvc/llvm-project?rev=123477&view=rev Log: Enhance GlobalOpt to be able evaluate initializers that involve stores through bitcasts, at least in simple cases. This fixes clang's CodeGenCXX/virtual-base-dtor.cpp Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp?rev=123477&r1=123476&r2=123477&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Fri Jan 14 16:19:20 2011 @@ -2128,8 +2128,9 @@ /// isSimpleEnoughPointerToCommit - Return true if this constant is simple -/// enough for us to understand. In particular, if it is a cast of something, -/// we punt. We basically just support direct accesses to globals and GEP's of +/// enough for us to understand. In particular, if it is a cast to anything +/// other than from one pointer type to another pointer type, we punt. +/// We basically just support direct accesses to globals and GEP's of /// globals. This should be kept up to date with CommitValueTo. static bool isSimpleEnoughPointerToCommit(Constant *C) { // Conservatively, avoid aggregate types. This is because we don't @@ -2163,6 +2164,20 @@ return false; return ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE); + + // A constantexpr bitcast from a pointer to another pointer is a no-op, + // and we know how to evaluate it by moving the bitcast from the pointer + // operand to the value operand. + } else if (CE->getOpcode() == Instruction::BitCast && + CE->getType()->isPointerTy() && + CE->getOperand(0)->getType()->isPointerTy()) { + GlobalVariable *GV = cast(CE->getOperand(0)); + // Do not allow weak/*_odr/linkonce/dllimport/dllexport linkage or + // external globals. + if (!GV->hasUniqueInitializer()) + return false; + + return true; } return false; } @@ -2336,6 +2351,38 @@ // of one global variable divided by another) then we can't commit it. if (!isSimpleEnoughValueToCommit(Val, SimpleConstants)) return false; + + if (ConstantExpr *CE = dyn_cast(Ptr)) + if (CE->getOpcode() == Instruction::BitCast) { + // If we're evaluating a store through a bitcast, then we need + // to pull the bitcast off the pointer type and push it onto the + // stored value. + Ptr = dyn_cast(Ptr->getOperand(0)); + if (!Ptr) return false; + + const PointerType *Ty = dyn_cast(Ptr->getType()); + if (!Ty) return false; + const Type *NewTy = Ty->getElementType(); + + // A bitcast'd pointer implicitly points to the first field of a + // struct. Insert implicity "gep @x, 0, 0, ..." until we get down + // to the first concrete member. + // FIXME: This could be extended to work for arrays as well. + while (const StructType *STy = dyn_cast(NewTy)) { + NewTy = STy->getTypeAtIndex(0U); + + const IntegerType *IdxTy = + IntegerType::get(NewTy->getContext(), 32); + Constant *IdxZero = ConstantInt::get(IdxTy, 0, false); + Constant * const IdxList[] = {IdxZero, IdxZero}; + + Ptr = ConstantExpr::getGetElementPtr(Ptr, IdxList, 2); + } + + if (!isa(NewTy)) return false; + Val = ConstantExpr::getBitCast(Val, NewTy); + } + MutatedMemory[Ptr] = Val; } else if (BinaryOperator *BO = dyn_cast(CurInst)) { InstResult = ConstantExpr::get(BO->getOpcode(), From gohman at apple.com Fri Jan 14 16:26:16 2011 From: gohman at apple.com (Dan Gohman) Date: Fri, 14 Jan 2011 22:26:16 -0000 Subject: [llvm-commits] [llvm] r123479 - /llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp Message-ID: <20110114222616.603F82A6C12C@llvm.org> Author: djg Date: Fri Jan 14 16:26:16 2011 New Revision: 123479 URL: http://llvm.org/viewvc/llvm-project?rev=123479&view=rev Log: Delete an assignment to ThisBB which isn't needed, and tidy up some comments. Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp?rev=123479&r1=123478&r2=123479&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp (original) +++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp Fri Jan 14 16:26:16 2011 @@ -1135,7 +1135,7 @@ // additional DAGs necessary. for (unsigned i = 0, e = SDB->SwitchCases.size(); i != e; ++i) { // Set the current basic block to the mbb we wish to insert the code into - MachineBasicBlock *ThisBB = FuncInfo->MBB = SDB->SwitchCases[i].ThisBB; + FuncInfo->MBB = SDB->SwitchCases[i].ThisBB; FuncInfo->InsertPt = FuncInfo->MBB->end(); // Determine the unique successors. @@ -1144,13 +1144,15 @@ if (SDB->SwitchCases[i].TrueBB != SDB->SwitchCases[i].FalseBB) Succs.push_back(SDB->SwitchCases[i].FalseBB); - // Emit the code. Note that this could result in ThisBB being split, so - // we need to check for updates. + // Emit the code. Note that this could result in FuncInfo->MBB being split. SDB->visitSwitchCase(SDB->SwitchCases[i], FuncInfo->MBB); CurDAG->setRoot(SDB->getRoot()); SDB->clear(); CodeGenAndEmitDAG(); - ThisBB = FuncInfo->MBB; + + // Remember the last block, now that any splitting is done, for use in + // populating PHI nodes in successors. + MachineBasicBlock *ThisBB = FuncInfo->MBB; // Handle any PHI nodes in successors of this chunk, as if we were coming // from the original BB before switch expansion. Note that PHI nodes can From resistor at mac.com Fri Jan 14 16:31:13 2011 From: resistor at mac.com (Owen Anderson) Date: Fri, 14 Jan 2011 22:31:13 -0000 Subject: [llvm-commits] [llvm] r123480 - /llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Message-ID: <20110114223113.5BE5B2A6C12C@llvm.org> Author: resistor Date: Fri Jan 14 16:31:13 2011 New Revision: 123480 URL: http://llvm.org/viewvc/llvm-project?rev=123480&view=rev Log: Fix a false-positive warning. Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp?rev=123480&r1=123479&r2=123480&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Fri Jan 14 16:31:13 2011 @@ -2143,7 +2143,7 @@ // external globals. return GV->hasUniqueInitializer(); - if (ConstantExpr *CE = dyn_cast(C)) + if (ConstantExpr *CE = dyn_cast(C)) { // Handle a constantexpr gep. if (CE->getOpcode() == Instruction::GetElementPtr && isa(CE->getOperand(0)) && @@ -2179,6 +2179,8 @@ return true; } + } + return false; } From kremenek at apple.com Fri Jan 14 16:34:14 2011 From: kremenek at apple.com (Ted Kremenek) Date: Fri, 14 Jan 2011 22:34:14 -0000 Subject: [llvm-commits] [llvm] r123486 - /llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp Message-ID: <20110114223414.0C99C2A6C12C@llvm.org> Author: kremenek Date: Fri Jan 14 16:34:13 2011 New Revision: 123486 URL: http://llvm.org/viewvc/llvm-project?rev=123486&view=rev Log: 'HiReg' is written but never read. Nuke its declaration and its assignments. Found by clang static analyzer. Modified: llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp Modified: llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp?rev=123486&r1=123485&r2=123486&view=diff ============================================================================== --- llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp (original) +++ llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp Fri Jan 14 16:34:13 2011 @@ -1607,13 +1607,13 @@ SDValue N0 = Node->getOperand(0); SDValue N1 = Node->getOperand(1); - unsigned LoReg, HiReg; + unsigned LoReg; switch (NVT.getSimpleVT().SimpleTy) { default: llvm_unreachable("Unsupported VT!"); - case MVT::i8: LoReg = X86::AL; HiReg = X86::AH; Opc = X86::MUL8r; break; - case MVT::i16: LoReg = X86::AX; HiReg = X86::DX; Opc = X86::MUL16r; break; - case MVT::i32: LoReg = X86::EAX; HiReg = X86::EDX; Opc = X86::MUL32r; break; - case MVT::i64: LoReg = X86::RAX; HiReg = X86::RDX; Opc = X86::MUL64r; break; + case MVT::i8: LoReg = X86::AL; Opc = X86::MUL8r; break; + case MVT::i16: LoReg = X86::AX; Opc = X86::MUL16r; break; + case MVT::i32: LoReg = X86::EAX; Opc = X86::MUL32r; break; + case MVT::i64: LoReg = X86::RAX; Opc = X86::MUL64r; break; } SDValue InFlag = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, LoReg, From kremenek at apple.com Fri Jan 14 16:34:17 2011 From: kremenek at apple.com (Ted Kremenek) Date: Fri, 14 Jan 2011 22:34:17 -0000 Subject: [llvm-commits] [llvm] r123487 - /llvm/trunk/tools/llvm-ld/llvm-ld.cpp Message-ID: <20110114223417.DC68F2A6C12D@llvm.org> Author: kremenek Date: Fri Jan 14 16:34:17 2011 New Revision: 123487 URL: http://llvm.org/viewvc/llvm-project?rev=123487&view=rev Log: Fix memory leak found by clang static analyzer. Modified: llvm/trunk/tools/llvm-ld/llvm-ld.cpp 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=123487&r1=123486&r2=123487&view=diff ============================================================================== --- llvm/trunk/tools/llvm-ld/llvm-ld.cpp (original) +++ llvm/trunk/tools/llvm-ld/llvm-ld.cpp Fri Jan 14 16:34:17 2011 @@ -178,7 +178,7 @@ // Allocate a new environment list. char **newenv = new char* [entries]; - if ((newenv = new char* [entries]) == NULL) + if (newenv == NULL) return NULL; // Make a copy of the list. Don't forget the NULL that ends the list. From bob.wilson at apple.com Fri Jan 14 16:58:09 2011 From: bob.wilson at apple.com (Bob Wilson) Date: Fri, 14 Jan 2011 22:58:09 -0000 Subject: [llvm-commits] [llvm] r123490 - in /llvm/trunk/utils/TableGen: AsmMatcherEmitter.cpp CodeGenInstruction.cpp Message-ID: <20110114225810.038C22A6C12C@llvm.org> Author: bwilson Date: Fri Jan 14 16:58:09 2011 New Revision: 123490 URL: http://llvm.org/viewvc/llvm-project?rev=123490&view=rev Log: Fix some tablegen issues to allow using zero_reg for InstAlias definitions. This is needed to allow an InstAlias for an instruction with an "OptionalDef" result register (like ARM's cc_out) where you want to set the optional register to reg0. Modified: llvm/trunk/utils/TableGen/AsmMatcherEmitter.cpp llvm/trunk/utils/TableGen/CodeGenInstruction.cpp Modified: llvm/trunk/utils/TableGen/AsmMatcherEmitter.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/TableGen/AsmMatcherEmitter.cpp?rev=123490&r1=123489&r2=123490&view=diff ============================================================================== --- llvm/trunk/utils/TableGen/AsmMatcherEmitter.cpp (original) +++ llvm/trunk/utils/TableGen/AsmMatcherEmitter.cpp Fri Jan 14 16:58:09 2011 @@ -1380,9 +1380,14 @@ break; } case MatchableInfo::ResOperand::RegOperand: { - std::string N = getQualifiedName(OpInfo.Register); - CaseOS << " Inst.addOperand(MCOperand::CreateReg(" << N << "));\n"; - Signature += "__reg" + OpInfo.Register->getName(); + if (OpInfo.Register == 0) { + CaseOS << " Inst.addOperand(MCOperand::CreateReg(0));\n"; + Signature += "__reg0"; + } else { + std::string N = getQualifiedName(OpInfo.Register); + CaseOS << " Inst.addOperand(MCOperand::CreateReg(" << N << "));\n"; + Signature += "__reg" + OpInfo.Register->getName(); + } } } } Modified: llvm/trunk/utils/TableGen/CodeGenInstruction.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/TableGen/CodeGenInstruction.cpp?rev=123490&r1=123489&r2=123490&view=diff ============================================================================== --- llvm/trunk/utils/TableGen/CodeGenInstruction.cpp (original) +++ llvm/trunk/utils/TableGen/CodeGenInstruction.cpp Fri Jan 14 16:58:09 2011 @@ -442,6 +442,21 @@ ++AliasOpNo; continue; } + if (ADI->getDef()->getName() == "zero_reg") { + if (!Result->getArgName(AliasOpNo).empty()) + throw TGError(R->getLoc(), "result fixed register argument must " + "not have a name!"); + + // Check if this is an optional def. + if (!ResultOpRec->isSubClassOf("OptionalDefOperand")) + throw TGError(R->getLoc(), "reg0 used for result that is not an " + "OptionalDefOperand!"); + + // Now that it is validated, add it. + ResultOperands.push_back(ResultOperand(static_cast(0))); + ++AliasOpNo; + continue; + } } // If the operand is a record, it must have a name, and the record type must From kremenek at apple.com Fri Jan 14 16:58:11 2011 From: kremenek at apple.com (Ted Kremenek) Date: Fri, 14 Jan 2011 22:58:11 -0000 Subject: [llvm-commits] [llvm] r123491 - /llvm/trunk/lib/CodeGen/AsmPrinter/CMakeLists.txt Message-ID: <20110114225812.011322A6C12D@llvm.org> Author: kremenek Date: Fri Jan 14 16:58:11 2011 New Revision: 123491 URL: http://llvm.org/viewvc/llvm-project?rev=123491&view=rev Log: Update CMake build. Modified: llvm/trunk/lib/CodeGen/AsmPrinter/CMakeLists.txt Modified: llvm/trunk/lib/CodeGen/AsmPrinter/CMakeLists.txt URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/CMakeLists.txt?rev=123491&r1=123490&r2=123491&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/AsmPrinter/CMakeLists.txt (original) +++ llvm/trunk/lib/CodeGen/AsmPrinter/CMakeLists.txt Fri Jan 14 16:58:11 2011 @@ -3,8 +3,10 @@ AsmPrinterDwarf.cpp AsmPrinterInlineAsm.cpp DIE.cpp + DwarfCFIException.cpp DwarfDebug.cpp DwarfException.cpp + DwarfTableException.cpp OcamlGCPrinter.cpp ) From evan.cheng at apple.com Fri Jan 14 17:38:20 2011 From: evan.cheng at apple.com (Evan Cheng) Date: Fri, 14 Jan 2011 15:38:20 -0800 Subject: [llvm-commits] [llvm] r123424 - in /llvm/trunk: include/llvm/Object/MachOFormat.h lib/MC/ELFObjectWriter.cpp lib/MC/MachObjectWriter.cpp lib/Target/ARM/ARMAsmBackend.cpp lib/Target/ARM/ARMFixupKinds.h lib/Target/ARM/ARMMCCodeEmitter.cpp test/MC/A In-Reply-To: References: Message-ID: <1A3DF11A-CCFD-4209-9AE0-09060C9D861F@apple.com> On Jan 14, 2011, at 12:20 PM, Jason Kim wrote: >> > > Hi Evan, > > FYI, the movt hi16 fixup you put in does not match the output of GNU as - > Let me look into this and figure out why. I'll reply soon. > > -jason Interesting. That's probably due to this: > case ARM::fixup_arm_movt_hi16: > - case ARM::fixup_arm_movw_lo16: > case ARM::fixup_arm_movt_hi16_pcrel: > + Value >>= 16; > + // Fallthrough > + case ARM::fixup_arm_movw_lo16: > case ARM::fixup_arm_movw_lo16_pcrel: { > unsigned Hi4 = (Value & 0xF000) >> 12; > unsigned Lo12 = Value & 0x0FFF; > @@ -168,6 +175,26 @@ > Value = (Hi4 << 16) | (Lo12); > return Value; > } Notice the Value is shifted down 16 bits for movt. I'm fairly sure this is right, at least for Darwin. It's possible there is a bug elsewhere. Evan -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110114/2e96ac50/attachment-0001.html From echristo at apple.com Fri Jan 14 17:50:53 2011 From: echristo at apple.com (Eric Christopher) Date: Fri, 14 Jan 2011 23:50:53 -0000 Subject: [llvm-commits] [llvm] r123494 - /llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp Message-ID: <20110114235053.829252A6C12C@llvm.org> Author: echristo Date: Fri Jan 14 17:50:53 2011 New Revision: 123494 URL: http://llvm.org/viewvc/llvm-project?rev=123494&view=rev Log: Fix 80-cols. 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=123494&r1=123493&r2=123494&view=diff ============================================================================== --- llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp (original) +++ llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp Fri Jan 14 17:50:53 2011 @@ -3652,7 +3652,8 @@ // Gather data to see if the operation can be modelled as a // shuffle in combination with VEXTs. -SDValue ARMTargetLowering::ReconstructShuffle(SDValue Op, SelectionDAG &DAG) const { +SDValue ARMTargetLowering::ReconstructShuffle(SDValue Op, + SelectionDAG &DAG) const { DebugLoc dl = Op.getDebugLoc(); EVT VT = Op.getValueType(); unsigned NumElts = VT.getVectorNumElements(); @@ -3729,19 +3730,23 @@ if (MinElts[i] >= NumElts) { // The extraction can just take the second half VEXTOffsets[i] = NumElts; - ShuffleSrcs[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, SourceVecs[i], + ShuffleSrcs[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, + SourceVecs[i], DAG.getIntPtrConstant(NumElts)); } else if (MaxElts[i] < NumElts) { // The extraction can just take the first half VEXTOffsets[i] = 0; - ShuffleSrcs[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, SourceVecs[i], + ShuffleSrcs[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, + SourceVecs[i], DAG.getIntPtrConstant(0)); } else { // An actual VEXT is needed VEXTOffsets[i] = MinElts[i]; - SDValue VEXTSrc1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, SourceVecs[i], + SDValue VEXTSrc1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, + SourceVecs[i], DAG.getIntPtrConstant(0)); - SDValue VEXTSrc2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, SourceVecs[i], + SDValue VEXTSrc2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, VT, + SourceVecs[i], DAG.getIntPtrConstant(NumElts)); ShuffleSrcs[i] = DAG.getNode(ARMISD::VEXT, dl, VT, VEXTSrc1, VEXTSrc2, DAG.getConstant(VEXTOffsets[i], MVT::i32)); @@ -3758,7 +3763,8 @@ } SDValue ExtractVec = Entry.getOperand(0); - int ExtractElt = cast(Op.getOperand(i).getOperand(1))->getSExtValue(); + int ExtractElt = cast(Op.getOperand(i) + .getOperand(1))->getSExtValue(); if (ExtractVec == SourceVecs[0]) { Mask.push_back(ExtractElt - VEXTOffsets[0]); } else { @@ -3768,7 +3774,8 @@ // Final check before we try to produce nonsense... if (isShuffleMaskLegal(Mask, VT)) - return DAG.getVectorShuffle(VT, dl, ShuffleSrcs[0], ShuffleSrcs[1], &Mask[0]); + return DAG.getVectorShuffle(VT, dl, ShuffleSrcs[0], ShuffleSrcs[1], + &Mask[0]); return SDValue(); } From bob.wilson at apple.com Fri Jan 14 18:09:18 2011 From: bob.wilson at apple.com (Bob Wilson) Date: Sat, 15 Jan 2011 00:09:18 -0000 Subject: [llvm-commits] [llvm] r123497 - /llvm/trunk/lib/Target/ARM/ARMMachineFunctionInfo.h Message-ID: <20110115000918.5B1812A6C12C@llvm.org> Author: bwilson Date: Fri Jan 14 18:09:18 2011 New Revision: 123497 URL: http://llvm.org/viewvc/llvm-project?rev=123497&view=rev Log: Fix a comment. Modified: llvm/trunk/lib/Target/ARM/ARMMachineFunctionInfo.h Modified: llvm/trunk/lib/Target/ARM/ARMMachineFunctionInfo.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMMachineFunctionInfo.h?rev=123497&r1=123496&r2=123497&view=diff ============================================================================== --- llvm/trunk/lib/Target/ARM/ARMMachineFunctionInfo.h (original) +++ llvm/trunk/lib/Target/ARM/ARMMachineFunctionInfo.h Fri Jan 14 18:09:18 2011 @@ -22,8 +22,8 @@ namespace llvm { -/// ARMFunctionInfo - This class is derived from MachineFunction private -/// ARM target-specific information for each MachineFunction. +/// ARMFunctionInfo - This class is derived from MachineFunctionInfo and +/// contains private ARM-specific information for each MachineFunction. class ARMFunctionInfo : public MachineFunctionInfo { /// isThumb - True if this function is compiled under Thumb mode. From sabre at nondot.org Fri Jan 14 18:12:35 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 00:12:35 -0000 Subject: [llvm-commits] [llvm] r123501 - in /llvm/trunk: include/llvm/Transforms/Utils/SSAUpdater.h lib/Transforms/Scalar/LICM.cpp lib/Transforms/Scalar/ScalarReplAggregates.cpp lib/Transforms/Utils/SSAUpdater.cpp Message-ID: <20110115001235.758832A6C12C@llvm.org> Author: lattner Date: Fri Jan 14 18:12:35 2011 New Revision: 123501 URL: http://llvm.org/viewvc/llvm-project?rev=123501&view=rev Log: Generalize LoadAndStorePromoter a bit and switch LICM to use it. Modified: llvm/trunk/include/llvm/Transforms/Utils/SSAUpdater.h llvm/trunk/lib/Transforms/Scalar/LICM.cpp llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp Modified: llvm/trunk/include/llvm/Transforms/Utils/SSAUpdater.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/Utils/SSAUpdater.h?rev=123501&r1=123500&r2=123501&view=diff ============================================================================== --- llvm/trunk/include/llvm/Transforms/Utils/SSAUpdater.h (original) +++ llvm/trunk/include/llvm/Transforms/Utils/SSAUpdater.h Fri Jan 14 18:12:35 2011 @@ -118,15 +118,17 @@ /// virtual methods. /// class LoadAndStorePromoter { +protected: + SSAUpdater &SSA; public: - LoadAndStorePromoter() {} + LoadAndStorePromoter(const SmallVectorImpl &Insts, + SSAUpdater &S, StringRef Name = StringRef()); virtual ~LoadAndStorePromoter() {} /// run - This does the promotion. Insts is a list of loads and stores to /// promote, and Name is the basename for the PHIs to insert. After this is /// complete, the loads and stores are removed from the code. - void run(StringRef Name, const SmallVectorImpl &Insts, - SSAUpdater *SSA = 0); + void run(const SmallVectorImpl &Insts) const; /// Return true if the specified instruction is in the Inst list (which was @@ -139,6 +141,21 @@ return true; return false; } + + /// doExtraRewritesBeforeFinalDeletion - This hook is invoked after all the + /// stores are found and inserted as available values, but + virtual void doExtraRewritesBeforeFinalDeletion() const { + } + + /// replaceLoadWithValue - Clients can choose to implement this to get + /// notified right before a load is RAUW'd another value. + virtual void replaceLoadWithValue(LoadInst *LI, Value *V) const { + } + + /// This is called before each instruction is deleted. + virtual void instructionDeleted(Instruction *I) const { + } + }; } // End llvm namespace Modified: llvm/trunk/lib/Transforms/Scalar/LICM.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LICM.cpp?rev=123501&r1=123500&r2=123501&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/LICM.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/LICM.cpp Fri Jan 14 18:12:35 2011 @@ -595,6 +595,53 @@ return true; } +namespace { + class LoopPromoter : public LoadAndStorePromoter { + Value *SomePtr; // Designated pointer to store to. + SmallPtrSet &PointerMustAliases; + SmallVectorImpl &LoopExitBlocks; + AliasSetTracker &AST; + public: + LoopPromoter(Value *SP, + const SmallVectorImpl &Insts, SSAUpdater &S, + SmallPtrSet &PMA, + SmallVectorImpl &LEB, AliasSetTracker &ast) + : LoadAndStorePromoter(Insts, S), SomePtr(SP), PointerMustAliases(PMA), + LoopExitBlocks(LEB), AST(ast) {} + + virtual bool isInstInList(Instruction *I, + const SmallVectorImpl &) const { + Value *Ptr; + if (LoadInst *LI = dyn_cast(I)) + Ptr = LI->getOperand(0); + else + Ptr = cast(I)->getPointerOperand(); + return PointerMustAliases.count(Ptr); + } + + virtual void doExtraRewritesBeforeFinalDeletion() const { + // Insert stores after in the loop exit blocks. Each exit block gets a + // store of the live-out values that feed them. Since we've already told + // the SSA updater about the defs in the loop and the preheader + // definition, it is all set and we can start using it. + for (unsigned i = 0, e = LoopExitBlocks.size(); i != e; ++i) { + BasicBlock *ExitBlock = LoopExitBlocks[i]; + Value *LiveInValue = SSA.GetValueInMiddleOfBlock(ExitBlock); + Instruction *InsertPos = ExitBlock->getFirstNonPHI(); + new StoreInst(LiveInValue, SomePtr, InsertPos); + } + } + + virtual void replaceLoadWithValue(LoadInst *LI, Value *V) const { + // Update alias analysis. + AST.copyValue(LI, V); + } + virtual void instructionDeleted(Instruction *I) const { + AST.deleteValue(I); + } + }; +} // end anon namespace + /// PromoteAliasSet - Try to promote memory values to scalars by sinking /// stores out of the loop and moving loads to before the loop. We do this by /// looping over the stores in the loop, looking for stores to Must pointers @@ -679,179 +726,43 @@ Changed = true; ++NumPromoted; + SmallVector ExitBlocks; + CurLoop->getUniqueExitBlocks(ExitBlocks); + // We use the SSAUpdater interface to insert phi nodes as required. SmallVector NewPHIs; SSAUpdater SSA(&NewPHIs); + LoopPromoter Promoter(SomePtr, LoopUses, SSA, PointerMustAliases, ExitBlocks, + *CurAST); - // It wants to know some value of the same type as what we'll be inserting. - Value *SomeValue; - if (isa(LoopUses[0])) - SomeValue = LoopUses[0]; - else - SomeValue = cast(LoopUses[0])->getOperand(0); - SSA.Initialize(SomeValue->getType(), SomeValue->getName()); - - // First step: bucket up uses of the pointers by the block they occur in. - // This is important because we have to handle multiple defs/uses in a block - // ourselves: SSAUpdater is purely for cross-block references. - // FIXME: Want a TinyVector since there is usually 0/1 element. - DenseMap > UsesByBlock; - for (unsigned i = 0, e = LoopUses.size(); i != e; ++i) { - Instruction *User = LoopUses[i]; - UsesByBlock[User->getParent()].push_back(User); - } - - // Okay, now we can iterate over all the blocks in the loop with uses, - // processing them. Keep track of which loads are loading a live-in value. - SmallVector LiveInLoads; - DenseMap ReplacedLoads; - - for (unsigned LoopUse = 0, e = LoopUses.size(); LoopUse != e; ++LoopUse) { - Instruction *User = LoopUses[LoopUse]; - std::vector &BlockUses = UsesByBlock[User->getParent()]; - - // If this block has already been processed, ignore this repeat use. - if (BlockUses.empty()) continue; - - // Okay, this is the first use in the block. If this block just has a - // single user in it, we can rewrite it trivially. - if (BlockUses.size() == 1) { - // If it is a store, it is a trivial def of the value in the block. - if (isa(User)) { - SSA.AddAvailableValue(User->getParent(), - cast(User)->getOperand(0)); - } else { - // Otherwise it is a load, queue it to rewrite as a live-in load. - LiveInLoads.push_back(cast(User)); - } - BlockUses.clear(); - continue; - } - - // Otherwise, check to see if this block is all loads. If so, we can queue - // them all as live in loads. - bool HasStore = false; - for (unsigned i = 0, e = BlockUses.size(); i != e; ++i) { - if (isa(BlockUses[i])) { - HasStore = true; - break; - } - } - - if (!HasStore) { - for (unsigned i = 0, e = BlockUses.size(); i != e; ++i) - LiveInLoads.push_back(cast(BlockUses[i])); - BlockUses.clear(); - continue; - } - - // Otherwise, we have mixed loads and stores (or just a bunch of stores). - // Since SSAUpdater is purely for cross-block values, we need to determine - // the order of these instructions in the block. If the first use in the - // block is a load, then it uses the live in value. The last store defines - // the live out value. We handle this by doing a linear scan of the block. - BasicBlock *BB = User->getParent(); - Value *StoredValue = 0; - for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) { - if (LoadInst *L = dyn_cast(II)) { - // If this is a load from an unrelated pointer, ignore it. - if (!PointerMustAliases.count(L->getOperand(0))) continue; - - // If we haven't seen a store yet, this is a live in use, otherwise - // use the stored value. - if (StoredValue) { - L->replaceAllUsesWith(StoredValue); - ReplacedLoads[L] = StoredValue; - } else { - LiveInLoads.push_back(L); - } - continue; - } - - if (StoreInst *S = dyn_cast(II)) { - // If this is a store to an unrelated pointer, ignore it. - if (!PointerMustAliases.count(S->getOperand(1))) continue; - - // Remember that this is the active value in the block. - StoredValue = S->getOperand(0); - } - } - - // The last stored value that happened is the live-out for the block. - assert(StoredValue && "Already checked that there is a store in block"); - SSA.AddAvailableValue(BB, StoredValue); - BlockUses.clear(); - } - - // Now that all the intra-loop values are classified, set up the preheader. - // It gets a load of the pointer we're promoting, and it is the live-out value - // from the preheader. - LoadInst *PreheaderLoad = new LoadInst(SomePtr,SomePtr->getName()+".promoted", - Preheader->getTerminator()); + // Set up the preheader to have a definition of the value. It is the live-out + // value from the preheader that uses in the loop will use. + LoadInst *PreheaderLoad = + new LoadInst(SomePtr, SomePtr->getName()+".promoted", + Preheader->getTerminator()); SSA.AddAvailableValue(Preheader, PreheaderLoad); - // Now that the preheader is good to go, set up the exit blocks. Each exit - // block gets a store of the live-out values that feed them. Since we've - // already told the SSA updater about the defs in the loop and the preheader - // definition, it is all set and we can start using it. - SmallVector ExitBlocks; - CurLoop->getUniqueExitBlocks(ExitBlocks); - for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) { - BasicBlock *ExitBlock = ExitBlocks[i]; - Value *LiveInValue = SSA.GetValueInMiddleOfBlock(ExitBlock); - Instruction *InsertPos = ExitBlock->getFirstNonPHI(); - new StoreInst(LiveInValue, SomePtr, InsertPos); + // Copy any value stored to or loaded from a must-alias of the pointer. + if (PreheaderLoad->getType()->isPointerTy()) { + Value *SomeValue; + if (LoadInst *LI = dyn_cast(LoopUses[0])) + SomeValue = LI; + else + SomeValue = cast(LoopUses[0])->getValueOperand(); + + CurAST->copyValue(SomeValue, PreheaderLoad); } - // Okay, now we rewrite all loads that use live-in values in the loop, - // inserting PHI nodes as necessary. - for (unsigned i = 0, e = LiveInLoads.size(); i != e; ++i) { - LoadInst *ALoad = LiveInLoads[i]; - Value *NewVal = SSA.GetValueInMiddleOfBlock(ALoad->getParent()); - ALoad->replaceAllUsesWith(NewVal); - CurAST->copyValue(ALoad, NewVal); - ReplacedLoads[ALoad] = NewVal; - } + // Rewrite all the loads in the loop and remember all the definitions from + // stores in the loop. + Promoter.run(LoopUses); // If the preheader load is itself a pointer, we need to tell alias analysis // about the new pointer we created in the preheader block and about any PHI // nodes that just got inserted. if (PreheaderLoad->getType()->isPointerTy()) { - // Copy any value stored to or loaded from a must-alias of the pointer. - CurAST->copyValue(SomeValue, PreheaderLoad); - for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i) - CurAST->copyValue(SomeValue, NewPHIs[i]); - } - - // Now that everything is rewritten, delete the old instructions from the body - // of the loop. They should all be dead now. - for (unsigned i = 0, e = LoopUses.size(); i != e; ++i) { - Instruction *User = LoopUses[i]; - - // If this is a load that still has uses, then the load must have been added - // as a live value in the SSAUpdate data structure for a block (e.g. because - // the loaded value was stored later). In this case, we need to recursively - // propagate the updates until we get to the real value. - if (!User->use_empty()) { - Value *NewVal = ReplacedLoads[User]; - assert(NewVal && "not a replaced load?"); - - // Propagate down to the ultimate replacee. The intermediately loads - // could theoretically already have been deleted, so we don't want to - // dereference the Value*'s. - DenseMap::iterator RLI = ReplacedLoads.find(NewVal); - while (RLI != ReplacedLoads.end()) { - NewVal = RLI->second; - RLI = ReplacedLoads.find(NewVal); - } - - User->replaceAllUsesWith(NewVal); - CurAST->copyValue(User, NewVal); - } - - CurAST->deleteValue(User); - User->eraseFromParent(); + CurAST->copyValue(PreheaderLoad, NewPHIs[i]); } // fwew, we're done! Modified: llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp?rev=123501&r1=123500&r2=123501&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp Fri Jan 14 18:12:35 2011 @@ -844,20 +844,13 @@ class AllocaPromoter : public LoadAndStorePromoter { AllocaInst *AI; public: - AllocaPromoter() : AI(0) {} + AllocaPromoter(const SmallVectorImpl &Insts, SSAUpdater &S) + : LoadAndStorePromoter(Insts, S), AI(0) {} - void run(AllocaInst *AI, SSAUpdater &SSA) { + void run(AllocaInst *AI, const SmallVectorImpl &Insts) { // Remember which alloca we're promoting (for isInstInList). this->AI = AI; - - // Build the list of instructions to promote. - SmallVector Insts; - for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); - UI != E; ++UI) - Insts.push_back(cast(*UI)); - - LoadAndStorePromoter::run(AI->getName(), Insts, &SSA); - + LoadAndStorePromoter::run(Insts); AI->eraseFromParent(); } @@ -882,7 +875,7 @@ BasicBlock &BB = F.getEntryBlock(); // Get the entry node for the function bool Changed = false; - + SmallVector Insts; while (1) { Allocas.clear(); @@ -899,9 +892,17 @@ PromoteMemToReg(Allocas, *DT, *DF); else { SSAUpdater SSA; - AllocaPromoter Promoter; - for (unsigned i = 0, e = Allocas.size(); i != e; ++i) - Promoter.run(Allocas[i], SSA); + for (unsigned i = 0, e = Allocas.size(); i != e; ++i) { + AllocaInst *AI = Allocas[i]; + + // Build list of instructions to promote. + for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); + UI != E; ++UI) + Insts.push_back(cast(*UI)); + + AllocaPromoter(Insts, SSA).run(AI, Insts); + Insts.clear(); + } } NumPromoted += Allocas.size(); Changed = true; Modified: llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp?rev=123501&r1=123500&r2=123501&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp (original) +++ llvm/trunk/lib/Transforms/Utils/SSAUpdater.cpp Fri Jan 14 18:12:35 2011 @@ -348,23 +348,25 @@ // LoadAndStorePromoter Implementation //===----------------------------------------------------------------------===// -void LoadAndStorePromoter::run(StringRef BaseName, - const SmallVectorImpl &Insts, - SSAUpdater *SSA) { +LoadAndStorePromoter:: +LoadAndStorePromoter(const SmallVectorImpl &Insts, + SSAUpdater &S, StringRef BaseName) : SSA(S) { if (Insts.empty()) return; - // If no SSAUpdater was provided, use a default one. This allows the client - // to capture inserted PHI nodes etc if they want. - SSAUpdater DefaultSSA; - if (SSA == 0) SSA = &DefaultSSA; - - const Type *ValTy; + Value *SomeVal; if (LoadInst *LI = dyn_cast(Insts[0])) - ValTy = LI->getType(); + SomeVal = LI; else - ValTy = cast(Insts[0])->getOperand(0)->getType(); - - SSA->Initialize(ValTy, BaseName); + SomeVal = cast(Insts[0])->getOperand(0); + + if (BaseName.empty()) + BaseName = SomeVal->getName(); + SSA.Initialize(SomeVal->getType(), BaseName); +} + + +void LoadAndStorePromoter:: +run(const SmallVectorImpl &Insts) const { // First step: bucket up uses of the alloca by the block they occur in. // This is important because we have to handle multiple defs/uses in a block @@ -396,7 +398,7 @@ if (BlockUses.size() == 1) { // If it is a store, it is a trivial def of the value in the block. if (StoreInst *SI = dyn_cast(User)) - SSA->AddAvailableValue(BB, SI->getOperand(0)); + SSA.AddAvailableValue(BB, SI->getOperand(0)); else // Otherwise it is a load, queue it to rewrite as a live-in load. LiveInLoads.push_back(cast(User)); @@ -437,6 +439,7 @@ // If we haven't seen a store yet, this is a live in use, otherwise // use the stored value. if (StoredValue) { + replaceLoadWithValue(L, StoredValue); L->replaceAllUsesWith(StoredValue); ReplacedLoads[L] = StoredValue; } else { @@ -456,7 +459,7 @@ // The last stored value that happened is the live-out for the block. assert(StoredValue && "Already checked that there is a store in block"); - SSA->AddAvailableValue(BB, StoredValue); + SSA.AddAvailableValue(BB, StoredValue); BlockUses.clear(); } @@ -464,11 +467,15 @@ // inserting PHI nodes as necessary. for (unsigned i = 0, e = LiveInLoads.size(); i != e; ++i) { LoadInst *ALoad = LiveInLoads[i]; - Value *NewVal = SSA->GetValueInMiddleOfBlock(ALoad->getParent()); + Value *NewVal = SSA.GetValueInMiddleOfBlock(ALoad->getParent()); + replaceLoadWithValue(ALoad, NewVal); ALoad->replaceAllUsesWith(NewVal); ReplacedLoads[ALoad] = NewVal; } + // Allow the client to do stuff before we start nuking things. + doExtraRewritesBeforeFinalDeletion(); + // Now that everything is rewritten, delete the old instructions from the // function. They should all be dead now. for (unsigned i = 0, e = Insts.size(); i != e; ++i) { @@ -491,9 +498,11 @@ RLI = ReplacedLoads.find(NewVal); } + replaceLoadWithValue(cast(User), NewVal); User->replaceAllUsesWith(NewVal); } + instructionDeleted(User); User->eraseFromParent(); } } From echristo at apple.com Fri Jan 14 18:25:09 2011 From: echristo at apple.com (Eric Christopher) Date: Sat, 15 Jan 2011 00:25:09 -0000 Subject: [llvm-commits] [llvm] r123505 - /llvm/trunk/lib/Target/ARM/ARMInstrInfo.td Message-ID: <20110115002509.E42752A6C12C@llvm.org> Author: echristo Date: Fri Jan 14 18:25:09 2011 New Revision: 123505 URL: http://llvm.org/viewvc/llvm-project?rev=123505&view=rev Log: 80-col. Modified: llvm/trunk/lib/Target/ARM/ARMInstrInfo.td Modified: llvm/trunk/lib/Target/ARM/ARMInstrInfo.td URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrInfo.td?rev=123505&r1=123504&r2=123505&view=diff ============================================================================== --- llvm/trunk/lib/Target/ARM/ARMInstrInfo.td (original) +++ llvm/trunk/lib/Target/ARM/ARMInstrInfo.td Fri Jan 14 18:25:09 2011 @@ -1175,7 +1175,8 @@ Size4Bytes, IIC_iStore_r, [(store GPR:$src, addrmodepc:$addr)]>; def PICSTRH : ARMPseudoInst<(outs), (ins GPR:$src, addrmodepc:$addr, pred:$p), - Size4Bytes, IIC_iStore_bh_r, [(truncstorei16 GPR:$src, addrmodepc:$addr)]>; + Size4Bytes, IIC_iStore_bh_r, [(truncstorei16 GPR:$src, + addrmodepc:$addr)]>; def PICSTRB : ARMPseudoInst<(outs), (ins GPR:$src, addrmodepc:$addr, pred:$p), Size4Bytes, IIC_iStore_bh_r, [(truncstorei8 GPR:$src, addrmodepc:$addr)]>; @@ -2517,7 +2518,8 @@ let isCommutable = 1 in { let Constraints = "@earlyclobber $Rd" in -def MULv5: PseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, pred:$p, cc_out:$s), +def MULv5: PseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, + pred:$p, cc_out:$s), IIC_iMUL32, [(set GPR:$Rd, (mul GPR:$Rn, GPR:$Rm))]>, Requires<[IsARM, NoV6]>; From rafael.espindola at gmail.com Fri Jan 14 19:12:50 2011 From: rafael.espindola at gmail.com (=?ISO-8859-1?Q?Rafael_=C1vila_de_Esp=EDndola?=) Date: Fri, 14 Jan 2011 20:12:50 -0500 Subject: [llvm-commits] [patch] Enable unnamed_addr in declarations again Message-ID: <4D30F492.9030509@gmail.com> I was convinced that we should allow unnamed_addr in declarations. This patch does it. I am just not entirely convinced what the behavior should be when resolving a declaration to a definition. This patch implements what I understood is sabre's preference: bool HasUnnamedAddr = SGV->hasUnnamedAddr(); if (DGV && (!LinkFromSrc || DGV->isDeclaration())) HasUnnamedAddr = HasUnnamedAddr && DGV->hasUnnamedAddr(); i.e, set drop the unnamed_addr unless both have it except for weak defs, where we keep the strong def. I find bool HasUnnamedAddr = SGV->hasUnnamedAddr(); if (DGV && !LinkFromSrc) HasUnnamedAddr = HasUnnamedAddr && DGV->hasUnnamedAddr(); to be a bit more natural (in the included testcase, this changes the result of @d). Cheers, Rafael -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: unnamed.patch Url: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110114/5f3f1661/attachment.pl From rafael.espindola at gmail.com Fri Jan 14 22:45:43 2011 From: rafael.espindola at gmail.com (=?ISO-8859-1?Q?Rafael_=C1vila_de_Esp=EDndola?=) Date: Fri, 14 Jan 2011 23:45:43 -0500 Subject: [llvm-commits] [patch] Enable unnamed_addr in declarations again In-Reply-To: <4D30F492.9030509@gmail.com> References: <4D30F492.9030509@gmail.com> Message-ID: <4D312677.2010907@gmail.com> And attached are the corresponding clang and llvm-gcc patches. Cheers, Rafael -------------- next part -------------- A non-text attachment was scrubbed... Name: clang.patch Type: text/x-patch Size: 3307 bytes Desc: not available Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110114/ef1f658e/attachment.bin -------------- next part -------------- A non-text attachment was scrubbed... Name: llvm-gcc.patch Type: text/x-patch Size: 917 bytes Desc: not available Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110114/ef1f658e/attachment-0001.bin From sabre at nondot.org Fri Jan 14 23:40:30 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 05:40:30 -0000 Subject: [llvm-commits] [llvm] r123514 - /llvm/trunk/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp Message-ID: <20110115054030.25F922A6C12C@llvm.org> Author: lattner Date: Fri Jan 14 23:40:29 2011 New Revision: 123514 URL: http://llvm.org/viewvc/llvm-project?rev=123514&view=rev Log: reduce indentation Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp?rev=123514&r1=123513&r2=123514&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp Fri Jan 14 23:40:29 2011 @@ -1920,37 +1920,37 @@ // fold (or (cast A), (cast B)) -> (cast (or A, B)) if (CastInst *Op0C = dyn_cast(Op0)) { - if (CastInst *Op1C = dyn_cast(Op1)) - if (Op0C->getOpcode() == Op1C->getOpcode()) {// same cast kind ? - const Type *SrcTy = Op0C->getOperand(0)->getType(); - if (SrcTy == Op1C->getOperand(0)->getType() && - SrcTy->isIntOrIntVectorTy()) { - Value *Op0COp = Op0C->getOperand(0), *Op1COp = Op1C->getOperand(0); - - if ((!isa(Op0COp) || !isa(Op1COp)) && - // Only do this if the casts both really cause code to be - // generated. - ShouldOptimizeCast(Op0C->getOpcode(), Op0COp, I.getType()) && - ShouldOptimizeCast(Op1C->getOpcode(), Op1COp, I.getType())) { - Value *NewOp = Builder->CreateOr(Op0COp, Op1COp, I.getName()); - return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType()); - } - - // If this is or(cast(icmp), cast(icmp)), try to fold this even if the - // cast is otherwise not optimizable. This happens for vector sexts. - if (ICmpInst *RHS = dyn_cast(Op1COp)) - if (ICmpInst *LHS = dyn_cast(Op0COp)) - if (Value *Res = FoldOrOfICmps(LHS, RHS)) - return CastInst::Create(Op0C->getOpcode(), Res, I.getType()); - - // If this is or(cast(fcmp), cast(fcmp)), try to fold this even if the - // cast is otherwise not optimizable. This happens for vector sexts. - if (FCmpInst *RHS = dyn_cast(Op1COp)) - if (FCmpInst *LHS = dyn_cast(Op0COp)) - if (Value *Res = FoldOrOfFCmps(LHS, RHS)) - return CastInst::Create(Op0C->getOpcode(), Res, I.getType()); + CastInst *Op1C = dyn_cast(Op1); + if (Op1C && Op0C->getOpcode() == Op1C->getOpcode()) {// same cast kind ? + const Type *SrcTy = Op0C->getOperand(0)->getType(); + if (SrcTy == Op1C->getOperand(0)->getType() && + SrcTy->isIntOrIntVectorTy()) { + Value *Op0COp = Op0C->getOperand(0), *Op1COp = Op1C->getOperand(0); + + if ((!isa(Op0COp) || !isa(Op1COp)) && + // Only do this if the casts both really cause code to be + // generated. + ShouldOptimizeCast(Op0C->getOpcode(), Op0COp, I.getType()) && + ShouldOptimizeCast(Op1C->getOpcode(), Op1COp, I.getType())) { + Value *NewOp = Builder->CreateOr(Op0COp, Op1COp, I.getName()); + return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType()); } + + // If this is or(cast(icmp), cast(icmp)), try to fold this even if the + // cast is otherwise not optimizable. This happens for vector sexts. + if (ICmpInst *RHS = dyn_cast(Op1COp)) + if (ICmpInst *LHS = dyn_cast(Op0COp)) + if (Value *Res = FoldOrOfICmps(LHS, RHS)) + return CastInst::Create(Op0C->getOpcode(), Res, I.getType()); + + // If this is or(cast(fcmp), cast(fcmp)), try to fold this even if the + // cast is otherwise not optimizable. This happens for vector sexts. + if (FCmpInst *RHS = dyn_cast(Op1COp)) + if (FCmpInst *LHS = dyn_cast(Op0COp)) + if (Value *Res = FoldOrOfFCmps(LHS, RHS)) + return CastInst::Create(Op0C->getOpcode(), Res, I.getType()); } + } } // Note: If we've gotten to the point of visiting the outer OR, then the From sabre at nondot.org Fri Jan 14 23:41:33 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 05:41:33 -0000 Subject: [llvm-commits] [llvm] r123515 - /llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp Message-ID: <20110115054133.A10A82A6C12C@llvm.org> Author: lattner Date: Fri Jan 14 23:41:33 2011 New Revision: 123515 URL: http://llvm.org/viewvc/llvm-project?rev=123515&view=rev Log: Catch ~x < cst just like ~x < ~y, we currently handle this through means that are about to disappear. Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp?rev=123515&r1=123514&r2=123515&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp Fri Jan 14 23:41:33 2011 @@ -2284,11 +2284,15 @@ } } - // ~x < ~y --> y < x { Value *A, *B; - if (match(Op0, m_Not(m_Value(A))) && - match(Op1, m_Not(m_Value(B)))) - return new ICmpInst(I.getPredicate(), B, A); + // ~x < ~y --> y < x + // ~x < cst --> ~cst < x + if (match(Op0, m_Not(m_Value(A)))) { + if (match(Op1, m_Not(m_Value(B)))) + return new ICmpInst(I.getPredicate(), B, A); + if (ConstantInt *RHSC = dyn_cast(B)) + return new ICmpInst(I.getPredicate(), ConstantExpr::getNot(RHSC), A); + } // (a+b) llvm.uadd.with.overflow. // (a+b) llvm.uadd.with.overflow. From sabre at nondot.org Fri Jan 14 23:42:47 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 05:42:47 -0000 Subject: [llvm-commits] [llvm] r123516 - /llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp Message-ID: <20110115054247.B50BD2A6C12C@llvm.org> Author: lattner Date: Fri Jan 14 23:42:47 2011 New Revision: 123516 URL: http://llvm.org/viewvc/llvm-project?rev=123516&view=rev Log: fix typo Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp?rev=123516&r1=123515&r2=123516&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp Fri Jan 14 23:42:47 2011 @@ -2290,7 +2290,7 @@ if (match(Op0, m_Not(m_Value(A)))) { if (match(Op1, m_Not(m_Value(B)))) return new ICmpInst(I.getPredicate(), B, A); - if (ConstantInt *RHSC = dyn_cast(B)) + if (ConstantInt *RHSC = dyn_cast(Op1)) return new ICmpInst(I.getPredicate(), ConstantExpr::getNot(RHSC), A); } From sabre at nondot.org Fri Jan 14 23:50:18 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 05:50:18 -0000 Subject: [llvm-commits] [llvm] r123517 - /llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp Message-ID: <20110115055018.F29FB2A6C12C@llvm.org> Author: lattner Date: Fri Jan 14 23:50:18 2011 New Revision: 123517 URL: http://llvm.org/viewvc/llvm-project?rev=123517&view=rev Log: one more instcombine variant that is needed to work with future changes, no functionality change currently. Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp?rev=123517&r1=123516&r2=123517&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstCombineAddSub.cpp Fri Jan 14 23:50:18 2011 @@ -672,6 +672,15 @@ Value *NewMul = Builder->CreateMul(A, B); return BinaryOperator::CreateAdd(Op0, NewMul); } + + // X - A*Cst -> X + A*-Cst + // X - Cst*A -> X + A*-Cst + ConstantInt *BCst; + if (match(Op1I, m_Mul(m_Value(A), m_ConstantInt(BCst))) || + match(Op1I, m_Mul(m_ConstantInt(BCst), m_Value(A)))) { + Value *NewMul = Builder->CreateMul(A, ConstantExpr::getNeg(BCst)); + return BinaryOperator::CreateAdd(Op0, NewMul); + } } } From sabre at nondot.org Fri Jan 14 23:52:27 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 05:52:27 -0000 Subject: [llvm-commits] [llvm] r123518 - /llvm/trunk/include/llvm/Support/PatternMatch.h Message-ID: <20110115055227.6C82B2A6C12C@llvm.org> Author: lattner Date: Fri Jan 14 23:52:27 2011 New Revision: 123518 URL: http://llvm.org/viewvc/llvm-project?rev=123518&view=rev Log: Fix m_Not and m_Neg to not match random ConstantInt's. Before these would try hard to match constants by inverting the bits and recursively matching. There are two problems with this: 1) some patterns would match when we didn't want them to (theoretical) 2) this is insanely expensive to do, and most often pointless. This was apparently useful in just 2 instcombine cases, which I added code to handle explicitly. This change speeds up 'opt' time on 176.gcc by 1% and produces bitwise identical code. Modified: llvm/trunk/include/llvm/Support/PatternMatch.h Modified: llvm/trunk/include/llvm/Support/PatternMatch.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/PatternMatch.h?rev=123518&r1=123517&r2=123518&view=diff ============================================================================== --- llvm/trunk/include/llvm/Support/PatternMatch.h (original) +++ llvm/trunk/include/llvm/Support/PatternMatch.h Fri Jan 14 23:52:27 2011 @@ -521,8 +521,6 @@ if (ConstantExpr *CE = dyn_cast(V)) if (CE->getOpcode() == Instruction::Xor) return matchIfNot(CE->getOperand(0), CE->getOperand(1)); - if (ConstantInt *CI = dyn_cast(V)) - return L.match(ConstantExpr::getNot(CI)); return false; } private: @@ -557,8 +555,6 @@ if (ConstantExpr *CE = dyn_cast(V)) if (CE->getOpcode() == Instruction::Sub) return matchIfNeg(CE->getOperand(0), CE->getOperand(1)); - if (ConstantInt *CI = dyn_cast(V)) - return L.match(ConstantExpr::getNeg(CI)); return false; } private: From sabre at nondot.org Sat Jan 15 00:27:35 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 06:27:35 -0000 Subject: [llvm-commits] [llvm] r123519 - /llvm/trunk/include/llvm/Target/TargetData.h Message-ID: <20110115062735.CE5C82A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 00:27:35 2011 New Revision: 123519 URL: http://llvm.org/viewvc/llvm-project?rev=123519&view=rev Log: fix typo Modified: llvm/trunk/include/llvm/Target/TargetData.h Modified: llvm/trunk/include/llvm/Target/TargetData.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Target/TargetData.h?rev=123519&r1=123518&r2=123519&view=diff ============================================================================== --- llvm/trunk/include/llvm/Target/TargetData.h (original) +++ llvm/trunk/include/llvm/Target/TargetData.h Sat Jan 15 00:27:35 2011 @@ -144,7 +144,7 @@ std::string getStringRepresentation() const; /// isLegalInteger - This function returns true if the specified type is - /// known tobe a native integer type supported by the CPU. For example, + /// known to be a native integer type supported by the CPU. For example, /// i64 is not native on most 32-bit CPUs and i37 is not native on any known /// one. This returns false if the integer width is not legal. /// From nicholas at mxc.ca Sat Jan 15 00:33:12 2011 From: nicholas at mxc.ca (Nick Lewycky) Date: Fri, 14 Jan 2011 22:33:12 -0800 Subject: [llvm-commits] [patch] Enable unnamed_addr in declarations again In-Reply-To: <4D30F492.9030509@gmail.com> References: <4D30F492.9030509@gmail.com> Message-ID: <4D313FA8.7000308@mxc.ca> Rafael ?vila de Esp?ndola wrote: > I was convinced that we should allow unnamed_addr in declarations. This > patch does it. I am just not entirely convinced what the behavior should > be when resolving a declaration to a definition. > > This patch implements what I understood is sabre's preference: > > bool HasUnnamedAddr = SGV->hasUnnamedAddr(); > if (DGV && (!LinkFromSrc || DGV->isDeclaration())) > HasUnnamedAddr = HasUnnamedAddr && DGV->hasUnnamedAddr(); > > i.e, set drop the unnamed_addr unless both have it except for weak defs, > where we keep the strong def. I don't see what makes the weak def'n exemption correct. If the weak global is used in ways where its address matters, but the strong global isn't, linking them would produce uses of the strong global where the address matters. What am I missing? Otherwise it sounds correct! Nick > I find > > bool HasUnnamedAddr = SGV->hasUnnamedAddr(); > if (DGV && !LinkFromSrc) > HasUnnamedAddr = HasUnnamedAddr && DGV->hasUnnamedAddr(); > > to be a bit more natural (in the included testcase, this changes the > result of @d). > > Cheers, > Rafael > > > > _______________________________________________ > llvm-commits mailing list > llvm-commits at cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits From sabre at nondot.org Sat Jan 15 00:32:33 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 06:32:33 -0000 Subject: [llvm-commits] [llvm] r123520 - in /llvm/trunk: lib/Transforms/InstCombine/InstCombineCasts.cpp test/Transforms/InstCombine/trunc.ll Message-ID: <20110115063233.377AC2A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 00:32:33 2011 New Revision: 123520 URL: http://llvm.org/viewvc/llvm-project?rev=123520&view=rev Log: implement an instcombine xform that canonicalizes casts outside of and-with-constant operations. This fixes rdar://8808586 which observed that we used to compile: union xy { struct x { _Bool b[15]; } x; __attribute__((packed)) struct y { __attribute__((packed)) unsigned long b0to7; __attribute__((packed)) unsigned int b8to11; __attribute__((packed)) unsigned short b12to13; __attribute__((packed)) unsigned char b14; } y; }; struct x foo(union xy *xy) { return xy->x; } into: _foo: ## @foo movq (%rdi), %rax movabsq $1095216660480, %rcx ## imm = 0xFF00000000 andq %rax, %rcx movabsq $-72057594037927936, %rdx ## imm = 0xFF00000000000000 andq %rax, %rdx movzbl %al, %esi orq %rdx, %rsi movq %rax, %rdx andq $65280, %rdx ## imm = 0xFF00 orq %rsi, %rdx movq %rax, %rsi andq $16711680, %rsi ## imm = 0xFF0000 orq %rdx, %rsi movl %eax, %edx andl $-16777216, %edx ## imm = 0xFFFFFFFFFF000000 orq %rsi, %rdx orq %rcx, %rdx movabsq $280375465082880, %rcx ## imm = 0xFF0000000000 movq %rax, %rsi andq %rcx, %rsi orq %rdx, %rsi movabsq $71776119061217280, %r8 ## imm = 0xFF000000000000 andq %r8, %rax orq %rsi, %rax movzwl 12(%rdi), %edx movzbl 14(%rdi), %esi shlq $16, %rsi orl %edx, %esi movq %rsi, %r9 shlq $32, %r9 movl 8(%rdi), %edx orq %r9, %rdx andq %rdx, %rcx movzbl %sil, %esi shlq $32, %rsi orq %rcx, %rsi movl %edx, %ecx andl $-16777216, %ecx ## imm = 0xFFFFFFFFFF000000 orq %rsi, %rcx movq %rdx, %rsi andq $16711680, %rsi ## imm = 0xFF0000 orq %rcx, %rsi movq %rdx, %rcx andq $65280, %rcx ## imm = 0xFF00 orq %rsi, %rcx movzbl %dl, %esi orq %rcx, %rsi andq %r8, %rdx orq %rsi, %rdx ret We now compile this into: _foo: ## @foo ## BB#0: ## %entry movzwl 12(%rdi), %eax movzbl 14(%rdi), %ecx shlq $16, %rcx orl %eax, %ecx shlq $32, %rcx movl 8(%rdi), %edx orq %rcx, %rdx movq (%rdi), %rax ret A small improvement :-) Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp llvm/trunk/test/Transforms/InstCombine/trunc.ll Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp?rev=123520&r1=123519&r2=123520&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstCombineCasts.cpp Sat Jan 15 00:32:33 2011 @@ -462,8 +462,8 @@ // Transform trunc(lshr (zext A), Cst) to eliminate one type conversion. Value *A = 0; ConstantInt *Cst = 0; - if (match(Src, m_LShr(m_ZExt(m_Value(A)), m_ConstantInt(Cst))) && - Src->hasOneUse()) { + if (Src->hasOneUse() && + match(Src, m_LShr(m_ZExt(m_Value(A)), m_ConstantInt(Cst)))) { // We have three types to worry about here, the type of A, the source of // the truncate (MidSize), and the destination of the truncate. We know that // ASize < MidSize and MidSize > ResultSize, but don't know the relation @@ -482,6 +482,16 @@ Shift->takeName(Src); return CastInst::CreateIntegerCast(Shift, CI.getType(), false); } + + // Transform "trunc (and X, cst)" -> "and (trunc X), cst" so long as the dest + // type isn't non-native. + if (Src->hasOneUse() && isa(Src->getType()) && + ShouldChangeType(Src->getType(), CI.getType()) && + match(Src, m_And(m_Value(A), m_ConstantInt(Cst)))) { + Value *NewTrunc = Builder->CreateTrunc(A, CI.getType(), A->getName()+".tr"); + return BinaryOperator::CreateAnd(NewTrunc, + ConstantExpr::getTrunc(Cst, CI.getType())); + } return 0; } Modified: llvm/trunk/test/Transforms/InstCombine/trunc.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/trunc.ll?rev=123520&r1=123519&r2=123520&view=diff ============================================================================== --- llvm/trunk/test/Transforms/InstCombine/trunc.ll (original) +++ llvm/trunk/test/Transforms/InstCombine/trunc.ll Sat Jan 15 00:32:33 2011 @@ -97,3 +97,23 @@ ; CHECK: ret i64 %ins35 } +define i8 @test9(i32 %X) { + %Y = and i32 %X, 42 + %Z = trunc i32 %Y to i8 + ret i8 %Z +; CHECK: @test9 +; CHECK: trunc +; CHECK: and +; CHECK: ret +} + +; rdar://8808586 +define i8 @test10(i32 %X) { + %Y = trunc i32 %X to i8 + %Z = and i8 %Y, 42 + ret i8 %Z +; CHECK: @test10 +; CHECK: trunc +; CHECK: and +; CHECK: ret +} From sabre at nondot.org Sat Jan 15 01:14:54 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 07:14:54 -0000 Subject: [llvm-commits] [llvm] r123523 - /llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Message-ID: <20110115071454.512402A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 01:14:54 2011 New Revision: 123523 URL: http://llvm.org/viewvc/llvm-project?rev=123523&view=rev Log: make the current instruction iterator an ivar, allowing xforms that potentially invalidate it (like inline asm lowering) to be sunk into their proper place, cleaning up a ton of code. Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp?rev=123523&r1=123522&r2=123523&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Sat Jan 15 01:14:54 2011 @@ -69,6 +69,11 @@ const TargetLowering *TLI; DominatorTree *DT; ProfileInfo *PFI; + + /// CurInstIterator - As we scan instructions optimizing them, this is the + /// next instruction to optimize. Xforms that can invalidate this should + /// update it. + BasicBlock::iterator CurInstIterator; /// BackEdges - Keep a set of all the loop back edges. /// @@ -104,8 +109,7 @@ bool OptimizeInst(Instruction *I); bool OptimizeMemoryInst(Instruction *I, Value *Addr, const Type *AccessTy, DenseMap &SunkAddrs); - bool OptimizeInlineAsmInst(Instruction *I, CallSite CS, - DenseMap &SunkAddrs); + bool OptimizeInlineAsmInst(CallInst *CS); bool OptimizeCallInst(CallInst *CI); bool MoveExtToFormExtLoad(Instruction *I); bool OptimizeExtUses(Instruction *I); @@ -605,6 +609,25 @@ } // end anonymous namespace bool CodeGenPrepare::OptimizeCallInst(CallInst *CI) { + BasicBlock *BB = CI->getParent(); + + // Lower inline assembly if we can. + // If we found an inline asm expession, and if the target knows how to + // lower it to normal LLVM code, do so now. + if (TLI && isa(CI->getCalledValue())) { + if (TLI->ExpandInlineAsm(CI)) { + // Avoid invalidating the iterator. + CurInstIterator = BB->begin(); + // Avoid processing instructions out of order, which could cause + // reuse before a value is defined. + SunkAddrs.clear(); + return true; + } + // Sink address computing for memory operands into the block. + if (OptimizeInlineAsmInst(CI)) + return true; + } + // Lower all uses of llvm.objectsize.* IntrinsicInst *II = dyn_cast(CI); if (II && II->getIntrinsicID() == Intrinsic::objectsize) { @@ -833,11 +856,11 @@ /// OptimizeInlineAsmInst - If there are any memory operands, use /// OptimizeMemoryInst to sink their address computing into the block when /// possible / profitable. -bool CodeGenPrepare::OptimizeInlineAsmInst(Instruction *I, CallSite CS, - DenseMap &SunkAddrs) { +bool CodeGenPrepare::OptimizeInlineAsmInst(CallInst *CS) { bool MadeChange = false; - TargetLowering::AsmOperandInfoVector TargetConstraints = TLI->ParseConstraints(CS); + TargetLowering::AsmOperandInfoVector + TargetConstraints = TLI->ParseConstraints(CS); unsigned ArgNo = 0; for (unsigned i = 0, e = TargetConstraints.size(); i != e; ++i) { TargetLowering::AsmOperandInfo &OpInfo = TargetConstraints[i]; @@ -847,8 +870,8 @@ if (OpInfo.ConstraintType == TargetLowering::C_Memory && OpInfo.isIndirect) { - Value *OpVal = const_cast(CS.getArgument(ArgNo++)); - MadeChange |= OptimizeMemoryInst(I, OpVal, OpVal->getType(), SunkAddrs); + Value *OpVal = CS->getArgOperand(ArgNo++); + MadeChange |= OptimizeMemoryInst(CS, OpVal, OpVal->getType(), SunkAddrs); } else if (OpInfo.Type == InlineAsm::isInput) ArgNo++; } @@ -1026,12 +1049,7 @@ OptimizeInst(NC); } } else if (CallInst *CI = dyn_cast(I)) { - if (TLI && isa(CI->getCalledValue())) { - // Sink address computing for memory operands into the block. - MadeChange |= OptimizeInlineAsmInst(I, &(*CI), SunkAddrs); - } else { - MadeChange |= OptimizeCallInst(CI); - } + MadeChange |= OptimizeCallInst(CI); } return MadeChange; @@ -1057,29 +1075,14 @@ SunkAddrs.clear(); - for (BasicBlock::iterator BBI = BB.begin(), E = BB.end(); BBI != E; ) { - Instruction *I = BBI++; + CurInstIterator = BB.begin(); + for (BasicBlock::iterator E = BB.end(); CurInstIterator != E; ) { + Instruction *I = CurInstIterator++; - if (CallInst *CI = dyn_cast(I)) { - // If we found an inline asm expession, and if the target knows how to - // lower it to normal LLVM code, do so now. - if (TLI && isa(CI->getCalledValue())) { - if (TLI->ExpandInlineAsm(CI)) { - BBI = BB.begin(); - // Avoid processing instructions out of order, which could cause - // reuse before a value is defined. - SunkAddrs.clear(); - } else - // Sink address computing for memory operands into the block. - MadeChange |= OptimizeInlineAsmInst(I, &(*CI), SunkAddrs); - } else { - // Other CallInst optimizations that don't need to muck with the - // enclosing iterator here. - MadeChange |= OptimizeCallInst(CI); - } - } else { + if (CallInst *CI = dyn_cast(I)) + MadeChange |= OptimizeCallInst(CI); + else MadeChange |= OptimizeInst(I); - } } return MadeChange; From sabre at nondot.org Sat Jan 15 01:25:29 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 07:25:29 -0000 Subject: [llvm-commits] [llvm] r123524 - in /llvm/trunk: lib/Transforms/Scalar/CodeGenPrepare.cpp test/Transforms/CodeGenPrepare/basic.ll Message-ID: <20110115072529.EE1542A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 01:25:29 2011 New Revision: 123524 URL: http://llvm.org/viewvc/llvm-project?rev=123524&view=rev Log: Now that instruction optzns can update the iterator as they go, we can have objectsize folding recursively simplify away their result when it folds. It is important to catch this here, because otherwise we won't eliminate the cross-block values at isel and other times. Added: llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp?rev=123524&r1=123523&r2=123524&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Sat Jan 15 01:25:29 2011 @@ -42,6 +42,7 @@ #include "llvm/Support/PatternMatch.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Support/IRBuilder.h" +#include "llvm/Support/ValueHandle.h" using namespace llvm; using namespace llvm::PatternMatch; @@ -634,8 +635,18 @@ bool Min = (cast(II->getArgOperand(1))->getZExtValue() == 1); const Type *ReturnTy = CI->getType(); Constant *RetVal = ConstantInt::get(ReturnTy, Min ? 0 : -1ULL); - CI->replaceAllUsesWith(RetVal); - CI->eraseFromParent(); + + // Substituting this can cause recursive simplifications, which can + // invalidate our iterator. Use a WeakVH to hold onto it in case this + // happens. + WeakVH IterHandle(CurInstIterator); + + ReplaceAndSimplifyAllUses(CI, RetVal, TLI ? TLI->getTargetData() : 0, DT); + + // If the iterator instruction was recursively deleted, start over at the + // start of the block. + if (IterHandle != CurInstIterator) + CurInstIterator = BB->begin(); return true; } @@ -653,6 +664,7 @@ CodeGenPrepareFortifiedLibCalls Simplifier; return Simplifier.fold(CI, TD); } + //===----------------------------------------------------------------------===// // Memory Optimization //===----------------------------------------------------------------------===// @@ -1076,14 +1088,8 @@ SunkAddrs.clear(); CurInstIterator = BB.begin(); - for (BasicBlock::iterator E = BB.end(); CurInstIterator != E; ) { - Instruction *I = CurInstIterator++; - - if (CallInst *CI = dyn_cast(I)) - MadeChange |= OptimizeCallInst(CI); - else - MadeChange |= OptimizeInst(I); - } + for (BasicBlock::iterator E = BB.end(); CurInstIterator != E; ) + MadeChange |= OptimizeInst(CurInstIterator++); return MadeChange; } Added: llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll?rev=123524&view=auto ============================================================================== --- llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll (added) +++ llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll Sat Jan 15 01:25:29 2011 @@ -0,0 +1,24 @@ +; RUN: opt -codegenprepare %s -S -o - | 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-n8:16:32:64" +target triple = "x86_64-apple-darwin10.0.0" + +; CHECK: @test1 +define i32 @test1(i8* %ptr) nounwind ssp noredzone align 2 { +entry: + %0 = tail call i64 @llvm.objectsize.i64(i8* %ptr, i1 false) + %1 = icmp ugt i64 %0, 3 + br i1 %1, label %T, label %trap + +; CHECK: br i1 true, label +trap: ; preds = %0, %entry + tail call void @llvm.trap() noreturn nounwind + unreachable + +T: + ret i32 4 +} + +declare i64 @llvm.objectsize.i64(i8*, i1) nounwind readonly + +declare void @llvm.trap() nounwind From rafael.espindola at gmail.com Sat Jan 15 01:29:48 2011 From: rafael.espindola at gmail.com (=?ISO-8859-1?Q?Rafael_=C1vila_de_Esp=EDndola?=) Date: Sat, 15 Jan 2011 02:29:48 -0500 Subject: [llvm-commits] [patch] Enable unnamed_addr in declarations again In-Reply-To: <4D313FA8.7000308@mxc.ca> References: <4D30F492.9030509@gmail.com> <4D313FA8.7000308@mxc.ca> Message-ID: <4D314CEC.9060508@gmail.com> > I don't see what makes the weak def'n exemption correct. If the weak > global is used in ways where its address matters, but the strong global > isn't, linking them would produce uses of the strong global where the > address matters. What am I missing? If the linker is deciding to keep an unnamed_addr or a non unnamed_addr definition, it is possible that you have already lost. Consider the case: * module 1 has strong const unnamed_addr symbol foo and an internal unnamed_add const bar with the same init as foo * module 2 has weak symbols foo and its address matters We optimize 1 and constmerge does its job and bar is merged with foo. We now link 1 and 2, and 2's assumptions are already wrong. It is for an analogous reason that I was proposing to not special case declarations, but maybe the correct thing to do is to be really strict and disallow the replacement of regular symbols with unnamed_addr ones. A unnamed_addr reference or weak symbol could be replaced with a non unnamed_addr definition, but not the other way around since making the resulting symbol non unnamed_addr now might be too late. Sounds reasonable? > Otherwise it sounds correct! > > Nick Cheers, Rafael From sabre at nondot.org Sat Jan 15 01:29:01 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 07:29:01 -0000 Subject: [llvm-commits] [llvm] r123525 - /llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Message-ID: <20110115072901.78B0A2A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 01:29:01 2011 New Revision: 123525 URL: http://llvm.org/viewvc/llvm-project?rev=123525&view=rev Log: simplify code, no functionality change. Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp?rev=123525&r1=123524&r2=123525&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Sat Jan 15 01:29:01 2011 @@ -108,8 +108,7 @@ void EliminateMostlyEmptyBlock(BasicBlock *BB); bool OptimizeBlock(BasicBlock &BB); bool OptimizeInst(Instruction *I); - bool OptimizeMemoryInst(Instruction *I, Value *Addr, const Type *AccessTy, - DenseMap &SunkAddrs); + bool OptimizeMemoryInst(Instruction *I, Value *Addr, const Type *AccessTy); bool OptimizeInlineAsmInst(CallInst *CS); bool OptimizeCallInst(CallInst *CI); bool MoveExtToFormExtLoad(Instruction *I); @@ -687,8 +686,7 @@ /// This method is used to optimize both load/store and inline asms with memory /// operands. bool CodeGenPrepare::OptimizeMemoryInst(Instruction *MemoryInst, Value *Addr, - const Type *AccessTy, - DenseMap &SunkAddrs) { + const Type *AccessTy) { Value *Repl = Addr; // Try to collapse single-value PHI nodes. This is necessary to undo @@ -883,7 +881,7 @@ if (OpInfo.ConstraintType == TargetLowering::C_Memory && OpInfo.isIndirect) { Value *OpVal = CS->getArgOperand(ArgNo++); - MadeChange |= OptimizeMemoryInst(CS, OpVal, OpVal->getType(), SunkAddrs); + MadeChange |= OptimizeMemoryInst(CS, OpVal, OpVal->getType()); } else if (OpInfo.Type == InlineAsm::isInput) ArgNo++; } @@ -1007,8 +1005,6 @@ } bool CodeGenPrepare::OptimizeInst(Instruction *I) { - bool MadeChange = false; - if (PHINode *P = dyn_cast(I)) { // It is possible for very late stage optimizations (such as SimplifyCFG) // to introduce PHI nodes too late to be cleaned up. If we detect such a @@ -1017,8 +1013,12 @@ P->replaceAllUsesWith(V); P->eraseFromParent(); ++NumPHIsElim; + return true; } - } else if (CastInst *CI = dyn_cast(I)) { + return false; + } + + if (CastInst *CI = dyn_cast(I)) { // If the source of the cast is a constant, then this should have // already been constant folded. The only reason NOT to constant fold // it is if something (e.g. LSR) was careful to place the constant @@ -1028,28 +1028,33 @@ if (isa(CI->getOperand(0))) return false; - bool Change = false; - if (TLI) { - Change = OptimizeNoopCopyExpression(CI, *TLI); - MadeChange |= Change; - } + if (TLI && OptimizeNoopCopyExpression(CI, *TLI)) + return true; - if (!Change && (isa(I) || isa(I))) { - MadeChange |= MoveExtToFormExtLoad(I); - MadeChange |= OptimizeExtUses(I); - } - } else if (CmpInst *CI = dyn_cast(I)) { - MadeChange |= OptimizeCmpExpression(CI); - } else if (LoadInst *LI = dyn_cast(I)) { + if (isa(I) || isa(I)) { + bool MadeChange = MoveExtToFormExtLoad(I); + return MadeChange | OptimizeExtUses(I); + } + return false; + } + + if (CmpInst *CI = dyn_cast(I)) + return OptimizeCmpExpression(CI); + + if (LoadInst *LI = dyn_cast(I)) { if (TLI) - MadeChange |= OptimizeMemoryInst(I, I->getOperand(0), LI->getType(), - SunkAddrs); - } else if (StoreInst *SI = dyn_cast(I)) { + return OptimizeMemoryInst(I, I->getOperand(0), LI->getType()); + return false; + } + + if (StoreInst *SI = dyn_cast(I)) { if (TLI) - MadeChange |= OptimizeMemoryInst(I, SI->getOperand(1), - SI->getOperand(0)->getType(), - SunkAddrs); - } else if (GetElementPtrInst *GEPI = dyn_cast(I)) { + return OptimizeMemoryInst(I, SI->getOperand(1), + SI->getOperand(0)->getType()); + return false; + } + + if (GetElementPtrInst *GEPI = dyn_cast(I)) { if (GEPI->hasAllZeroIndices()) { /// The GEP operand must be a pointer, so must its result -> BitCast Instruction *NC = new BitCastInst(GEPI->getOperand(0), GEPI->getType(), @@ -1057,14 +1062,16 @@ GEPI->replaceAllUsesWith(NC); GEPI->eraseFromParent(); ++NumGEPsElim; - MadeChange = true; OptimizeInst(NC); + return true; } - } else if (CallInst *CI = dyn_cast(I)) { - MadeChange |= OptimizeCallInst(CI); + return false; } + + if (CallInst *CI = dyn_cast(I)) + return OptimizeCallInst(CI); - return MadeChange; + return false; } // In this pass we look for GEP and cast instructions that are used From sabre at nondot.org Sat Jan 15 01:36:13 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 07:36:13 -0000 Subject: [llvm-commits] [llvm] r123526 - in /llvm/trunk: lib/Transforms/Scalar/CodeGenPrepare.cpp test/Transforms/CodeGenPrepare/basic.ll Message-ID: <20110115073613.A23A92A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 01:36:13 2011 New Revision: 123526 URL: http://llvm.org/viewvc/llvm-project?rev=123526&view=rev Log: fix rdar://8785296 - -fcatch-undefined-behavior generates inefficient code The basic issue is that isel (very reasonably!) expects conditional branches to be folded, so CGP leaving around a bunch dead computation feeding conditional branches isn't such a good idea. Just fold branches on constants into unconditional branches. Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp?rev=123526&r1=123525&r2=123526&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Sat Jan 15 01:36:13 2011 @@ -1071,6 +1071,9 @@ if (CallInst *CI = dyn_cast(I)) return OptimizeCallInst(CI); + if (isa(I)) + return ConstantFoldTerminator(I->getParent()); + return false; } Modified: llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll?rev=123526&r1=123525&r2=123526&view=diff ============================================================================== --- llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll (original) +++ llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll Sat Jan 15 01:36:13 2011 @@ -4,13 +4,16 @@ target triple = "x86_64-apple-darwin10.0.0" ; CHECK: @test1 +; objectsize should fold to a constant, which causes the branch to fold to an +; uncond branch. +; rdar://8785296 define i32 @test1(i8* %ptr) nounwind ssp noredzone align 2 { entry: %0 = tail call i64 @llvm.objectsize.i64(i8* %ptr, i1 false) %1 = icmp ugt i64 %0, 3 br i1 %1, label %T, label %trap -; CHECK: br i1 true, label +; CHECK: br label %T trap: ; preds = %0, %entry tail call void @llvm.trap() noreturn nounwind unreachable From rafael.espindola at gmail.com Sat Jan 15 01:47:57 2011 From: rafael.espindola at gmail.com (=?ISO-8859-1?Q?Rafael_=C1vila_de_Esp=EDndola?=) Date: Sat, 15 Jan 2011 02:47:57 -0500 Subject: [llvm-commits] [patch] Enable unnamed_addr in declarations again In-Reply-To: <4D314CEC.9060508@gmail.com> References: <4D30F492.9030509@gmail.com> <4D313FA8.7000308@mxc.ca> <4D314CEC.9060508@gmail.com> Message-ID: <4D31512D.50105@gmail.com> > * module 1 has strong const unnamed_addr symbol foo and an internal > unnamed_add const bar with the same init as foo > * module 2 has weak symbols foo and its address matters > > We optimize 1 and constmerge does its job and bar is merged with foo. We > now link 1 and 2, and 2's assumptions are already wrong. Sorry for replying to myself, but this is still a bit confusing to me. I think I understand you now. If we say that a symbol whose address matters is only guaranteed to have an address different from all other symbols whose address matters, we get a much simpler model. Maybe that is what you were thinking. Under these rules * Constmerge *can* merge two constants as long as *one* of them is unnamed_addr. The resulting constant is unnamed_addr if and only if both original ones are. * Weak symbols are really not an exception and the linker always uses an 'and' in all cases. I will give that a try, thanks a lot. Cheers, Rafael From sabre at nondot.org Sat Jan 15 01:51:19 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 07:51:19 -0000 Subject: [llvm-commits] [llvm] r123527 - in /llvm/trunk: lib/Transforms/Scalar/CodeGenPrepare.cpp test/Transforms/CodeGenPrepare/basic.ll Message-ID: <20110115075120.54B692A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 01:51:19 2011 New Revision: 123527 URL: http://llvm.org/viewvc/llvm-project?rev=123527&view=rev Log: temporarily revert r123526. While working on a follow-on patch I realize that ConstantFoldTerminator doesn't preserve dominfo. Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll Modified: llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp?rev=123527&r1=123526&r2=123527&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/CodeGenPrepare.cpp Sat Jan 15 01:51:19 2011 @@ -1071,9 +1071,6 @@ if (CallInst *CI = dyn_cast(I)) return OptimizeCallInst(CI); - if (isa(I)) - return ConstantFoldTerminator(I->getParent()); - return false; } Modified: llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll?rev=123527&r1=123526&r2=123527&view=diff ============================================================================== --- llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll (original) +++ llvm/trunk/test/Transforms/CodeGenPrepare/basic.ll Sat Jan 15 01:51:19 2011 @@ -13,7 +13,9 @@ %1 = icmp ugt i64 %0, 3 br i1 %1, label %T, label %trap -; CHECK: br label %T +; CHECK: entry: +; HECK-NEXT: ret i32 4 + trap: ; preds = %0, %entry tail call void @llvm.trap() noreturn nounwind unreachable From aggarwa4 at illinois.edu Sat Jan 15 01:53:47 2011 From: aggarwa4 at illinois.edu (Arushi Aggarwal) Date: Sat, 15 Jan 2011 07:53:47 -0000 Subject: [llvm-commits] [poolalloc] r123528 - /poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp Message-ID: <20110115075348.01BBA2A6C12C@llvm.org> Author: aggarwa4 Date: Sat Jan 15 01:53:47 2011 New Revision: 123528 URL: http://llvm.org/viewvc/llvm-project?rev=123528&view=rev Log: Fixed error in build. Modified: poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp Modified: poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp URL: http://llvm.org/viewvc/llvm-project/poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp?rev=123528&r1=123527&r2=123528&view=diff ============================================================================== --- poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp (original) +++ poolalloc/trunk/lib/PoolAllocate/Heuristic.cpp Sat Jan 15 01:53:47 2011 @@ -429,7 +429,7 @@ if (!(GlobalPoolNodes.count (N) || GlobalPoolNodes.count (GGN))) { // Otherwise, if it was not passed in from outside the function, it must // be a local pool! - assert(!N->isGlobalNode() || N->isPtrToIntNode() && "Should be in global mapping!"); + assert((!N->isGlobalNode() || N->isPtrToIntNode()) && "Should be in global mapping!"); if(!N->isPtrToIntNode()) { Nodes.push_back (N); } From nicholas at mxc.ca Sat Jan 15 02:01:08 2011 From: nicholas at mxc.ca (Nick Lewycky) Date: Sat, 15 Jan 2011 00:01:08 -0800 Subject: [llvm-commits] [patch] Enable unnamed_addr in declarations again In-Reply-To: <4D31512D.50105@gmail.com> References: <4D30F492.9030509@gmail.com> <4D313FA8.7000308@mxc.ca> <4D314CEC.9060508@gmail.com> <4D31512D.50105@gmail.com> Message-ID: <4D315444.9060001@mxc.ca> Rafael ?vila de Esp?ndola wrote: >> * module 1 has strong const unnamed_addr symbol foo and an internal >> unnamed_add const bar with the same init as foo >> * module 2 has weak symbols foo and its address matters >> >> We optimize 1 and constmerge does its job and bar is merged with foo. We >> now link 1 and 2, and 2's assumptions are already wrong. > > Sorry for replying to myself, but this is still a bit confusing to me. > > I think I understand you now. If we say that a symbol whose address > matters is only guaranteed to have an address different from all other > symbols whose address matters, we get a much simpler model. Maybe that > is what you were thinking. Yes exactly. The only gotcha is that for non-local-linkage unnamed_addr, the address has to be irrelevant for everyone we link to too. Under these rules > > * Constmerge *can* merge two constants as long as *one* of them is > unnamed_addr. The resulting constant is unnamed_addr if and only if both > original ones are. > * Weak symbols are really not an exception and the linker always uses an > 'and' in all cases. Yep! Nick > > I will give that a try, thanks a lot. > > Cheers, > Rafael > From rafael.espindola at gmail.com Sat Jan 15 02:15:01 2011 From: rafael.espindola at gmail.com (Rafael Espindola) Date: Sat, 15 Jan 2011 08:15:01 -0000 Subject: [llvm-commits] [llvm] r123529 - in /llvm/trunk: lib/AsmParser/LLParser.cpp lib/Linker/LinkModules.cpp lib/VMCore/Verifier.cpp test/Assembler/declare-unnamed-addr.ll test/Assembler/external-unnamed-addr.ll test/Linker/unnamed-addr1-a.ll test/Linker/unnamed-addr1-b.ll test/Linker/unnamed-addr2-a.ll test/Linker/unnamed-addr2-b.ll unittests/VMCore/VerifierTest.cpp Message-ID: <20110115081501.3A9E42A6C12C@llvm.org> Author: rafael Date: Sat Jan 15 02:15:00 2011 New Revision: 123529 URL: http://llvm.org/viewvc/llvm-project?rev=123529&view=rev Log: Allow unnamed_addr on declarations. Removed: llvm/trunk/test/Assembler/declare-unnamed-addr.ll llvm/trunk/test/Assembler/external-unnamed-addr.ll llvm/trunk/test/Linker/unnamed-addr2-a.ll llvm/trunk/test/Linker/unnamed-addr2-b.ll Modified: llvm/trunk/lib/AsmParser/LLParser.cpp llvm/trunk/lib/Linker/LinkModules.cpp llvm/trunk/lib/VMCore/Verifier.cpp llvm/trunk/test/Linker/unnamed-addr1-a.ll llvm/trunk/test/Linker/unnamed-addr1-b.ll llvm/trunk/unittests/VMCore/VerifierTest.cpp Modified: llvm/trunk/lib/AsmParser/LLParser.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/AsmParser/LLParser.cpp?rev=123529&r1=123528&r2=123529&view=diff ============================================================================== --- llvm/trunk/lib/AsmParser/LLParser.cpp (original) +++ llvm/trunk/lib/AsmParser/LLParser.cpp Sat Jan 15 02:15:00 2011 @@ -716,9 +716,6 @@ return true; } - if (!Init && UnnamedAddr) - return Error(UnnamedAddrLoc, "only definitions can have unnamed_addr"); - if (Ty->isFunctionTy() || Ty->isLabelTy()) return Error(TyLoc, "invalid type for global variable"); @@ -2687,9 +2684,6 @@ ParseType(RetType, RetTypeLoc, true /*void allowed*/)) return true; - if (!isDefine && UnnamedAddr) - return Error(UnnamedAddrLoc, "only definitions can have unnamed_addr"); - // Verify that the linkage is ok. switch ((GlobalValue::LinkageTypes)Linkage) { case GlobalValue::ExternalLinkage: Modified: llvm/trunk/lib/Linker/LinkModules.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Linker/LinkModules.cpp?rev=123529&r1=123528&r2=123529&view=diff ============================================================================== --- llvm/trunk/lib/Linker/LinkModules.cpp (original) +++ llvm/trunk/lib/Linker/LinkModules.cpp Sat Jan 15 02:15:00 2011 @@ -352,8 +352,6 @@ unsigned Alignment = std::max(DestGV->getAlignment(), SrcGV->getAlignment()); DestGV->copyAttributesFrom(SrcGV); DestGV->setAlignment(Alignment); - if (SrcGV->hasUnnamedAddr()) - DestGV->setUnnamedAddr(true); } /// GetLinkageResult - This analyzes the two global values and determines what @@ -521,6 +519,8 @@ continue; } + bool HasUnnamedAddr = SGV->hasUnnamedAddr() && DGV->hasUnnamedAddr(); + // If the visibilities of the symbols disagree and the destination is a // prototype, take the visibility of its input. if (DGV->isDeclaration()) @@ -565,6 +565,9 @@ DGV->getName(), 0, false, SGV->getType()->getAddressSpace()); + // Set the unnamed_addr. + NewDGV->setUnnamedAddr(HasUnnamedAddr); + // Propagate alignment, section, and visibility info. CopyGVAttributes(NewDGV, SGV); DGV->replaceAllUsesWith(ConstantExpr::getBitCast(NewDGV, @@ -609,8 +612,9 @@ "': symbol multiple defined"); } - // Set calculated linkage + // Set calculated linkage and unnamed_addr DGV->setLinkage(NewLinkage); + DGV->setUnnamedAddr(HasUnnamedAddr); // Make sure to remember this mapping... ValueMap[SGV] = ConstantExpr::getBitCast(DGV, SGV->getType()); Modified: llvm/trunk/lib/VMCore/Verifier.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/VMCore/Verifier.cpp?rev=123529&r1=123528&r2=123529&view=diff ============================================================================== --- llvm/trunk/lib/VMCore/Verifier.cpp (original) +++ llvm/trunk/lib/VMCore/Verifier.cpp Sat Jan 15 02:15:00 2011 @@ -469,8 +469,6 @@ Assert1(GV.hasExternalLinkage() || GV.hasDLLImportLinkage() || GV.hasExternalWeakLinkage(), "invalid linkage type for global declaration", &GV); - Assert1(!GV.hasUnnamedAddr(), "only definitions can have unnamed_addr", - &GV); } visitGlobalValue(GV); @@ -727,7 +725,6 @@ Assert1(F.hasExternalLinkage() || F.hasDLLImportLinkage() || F.hasExternalWeakLinkage(), "invalid linkage type for function declaration", &F); - Assert1(!F.hasUnnamedAddr(), "only definitions can have unnamed_addr", &F); } else { // Verify that this function (which has a body) is not named "llvm.*". It // is not legal to define intrinsics. Removed: llvm/trunk/test/Assembler/declare-unnamed-addr.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Assembler/declare-unnamed-addr.ll?rev=123528&view=auto ============================================================================== --- llvm/trunk/test/Assembler/declare-unnamed-addr.ll (original) +++ llvm/trunk/test/Assembler/declare-unnamed-addr.ll (removed) @@ -1,8 +0,0 @@ -; RUN: not llvm-as %s -o /dev/null 2>%t -; RUN: FileCheck -input-file=%t %s - -declare unnamed_addr i32 @zed() - -// CHECK: error: only definitions can have unnamed_addr -// CHECK: declare unnamed_addr i32 @zed() -// CHECK: ^ Removed: llvm/trunk/test/Assembler/external-unnamed-addr.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Assembler/external-unnamed-addr.ll?rev=123528&view=auto ============================================================================== --- llvm/trunk/test/Assembler/external-unnamed-addr.ll (original) +++ llvm/trunk/test/Assembler/external-unnamed-addr.ll (removed) @@ -1,8 +0,0 @@ -; RUN: not llvm-as %s -o /dev/null 2>%t -; RUN: FileCheck -input-file=%t %s - - at foo = external unnamed_addr global i8* - -// CHECK: error: only definitions can have unnamed_addr -// CHECK: @foo = external unnamed_addr global i8* -// CHECK: ^ Modified: llvm/trunk/test/Linker/unnamed-addr1-a.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Linker/unnamed-addr1-a.ll?rev=123529&r1=123528&r2=123529&view=diff ============================================================================== --- llvm/trunk/test/Linker/unnamed-addr1-a.ll (original) +++ llvm/trunk/test/Linker/unnamed-addr1-a.ll Sat Jan 15 02:15:00 2011 @@ -1,11 +1,27 @@ -; RUN: llvm-link %s %p/unnamed-addr1-b.ll -S -o - | FileCheck %s +; RUN: llvm-link %s %p/unnamed-addr1-b.ll -S -o - | sort | FileCheck %s - at foo = external global i32 +; Only in this file + at a = common global i32 0 +; CHECK: @a = common global i32 0 + at b = common unnamed_addr global i32 0 +; CHECK: @b = common unnamed_addr global i32 0 -define i32 @bar() { -entry: - %tmp = load i32* @foo, align 4 - ret i32 %tmp -} +; Other file has unnamed_addr definition + at c = common unnamed_addr global i32 0 +; CHECK: @c = common unnamed_addr global i32 0 + at d = external global i32 +; CHECK: @d = global i32 42 + at e = external unnamed_addr global i32 +; CHECK: @e = unnamed_addr global i32 42 + at f = weak global i32 42 +; CHECK: @f = global i32 42 -; CHECK: @foo = common unnamed_addr global i32 0, align 4 +; Other file has non-unnamed_addr definition + at g = common unnamed_addr global i32 0 +; CHECK: @g = common global i32 0 + at h = external global i32 +; CHECK: @h = global i32 42 + at i = external unnamed_addr global i32 +; CHECK: @i = global i32 42 + at j = weak global i32 42 +; CHECK: @j = global i32 42 Modified: llvm/trunk/test/Linker/unnamed-addr1-b.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Linker/unnamed-addr1-b.ll?rev=123529&r1=123528&r2=123529&view=diff ============================================================================== --- llvm/trunk/test/Linker/unnamed-addr1-b.ll (original) +++ llvm/trunk/test/Linker/unnamed-addr1-b.ll Sat Jan 15 02:15:00 2011 @@ -1,4 +1,12 @@ ; This file is for use with unnamed-addr1-a.ll ; RUN: true - at foo = common unnamed_addr global i32 0, align 4 + at c = common unnamed_addr global i32 42 + at d = unnamed_addr global i32 42 + at e = unnamed_addr global i32 42 + at f = unnamed_addr global i32 42 + + at g = common global i32 42 + at h = global i32 42 + at i = global i32 42 + at j = global i32 42 Removed: llvm/trunk/test/Linker/unnamed-addr2-a.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Linker/unnamed-addr2-a.ll?rev=123528&view=auto ============================================================================== --- llvm/trunk/test/Linker/unnamed-addr2-a.ll (original) +++ llvm/trunk/test/Linker/unnamed-addr2-a.ll (removed) @@ -1,11 +0,0 @@ -; RUN: llvm-link %s %p/unnamed-addr2-b.ll -S -o - | FileCheck %s - -define i32 @bar() { -entry: - %call = tail call i32 @foo() - ret i32 %call -} - -declare i32 @foo() - -; CHECK: define unnamed_addr i32 @foo() Removed: llvm/trunk/test/Linker/unnamed-addr2-b.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Linker/unnamed-addr2-b.ll?rev=123528&view=auto ============================================================================== --- llvm/trunk/test/Linker/unnamed-addr2-b.ll (original) +++ llvm/trunk/test/Linker/unnamed-addr2-b.ll (removed) @@ -1,7 +0,0 @@ -; This file is for use with unnamed-addr2-a.ll -; RUN: true - -define unnamed_addr i32 @foo() { -entry: - ret i32 42 -} Modified: llvm/trunk/unittests/VMCore/VerifierTest.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/VMCore/VerifierTest.cpp?rev=123529&r1=123528&r2=123529&view=diff ============================================================================== --- llvm/trunk/unittests/VMCore/VerifierTest.cpp (original) +++ llvm/trunk/unittests/VMCore/VerifierTest.cpp Sat Jan 15 02:15:00 2011 @@ -60,32 +60,5 @@ EXPECT_TRUE(verifyModule(M, ReturnStatusAction, &Error)); EXPECT_TRUE(StringRef(Error).startswith("Alias cannot have unnamed_addr")); } - -TEST(VerifierTest, ExternalUnnamedAddr) { - LLVMContext &C = getGlobalContext(); - Module M("M", C); - const Type *Ty = Type::getInt8Ty(C); - GlobalVariable *GV = new GlobalVariable(M, Ty, true, - GlobalValue::ExternalLinkage, - NULL, "foo"); - GV->setUnnamedAddr(true); - std::string Error; - EXPECT_TRUE(verifyModule(M, ReturnStatusAction, &Error)); - EXPECT_TRUE(StringRef(Error) - .startswith("only definitions can have unnamed_addr")); -} - -TEST(VerifierTest, DeclarationUnnamedAddr) { - LLVMContext &C = getGlobalContext(); - Module M("M", C); - FunctionType *FTy = FunctionType::get(Type::getVoidTy(C), /*isVarArg=*/false); - Function *F = Function::Create(FTy, GlobalValue::ExternalLinkage, - "foo", &M); - F->setUnnamedAddr(true); - std::string Error; - EXPECT_TRUE(verifyModule(M, ReturnStatusAction, &Error)); - EXPECT_TRUE(StringRef(Error) - .startswith("only definitions can have unnamed_addr")); -} } } From rafael.espindola at gmail.com Sat Jan 15 02:20:57 2011 From: rafael.espindola at gmail.com (Rafael Espindola) Date: Sat, 15 Jan 2011 08:20:57 -0000 Subject: [llvm-commits] [llvm] r123530 - /llvm/trunk/docs/LangRef.html Message-ID: <20110115082057.6BA2C2A6C12C@llvm.org> Author: rafael Date: Sat Jan 15 02:20:57 2011 New Revision: 123530 URL: http://llvm.org/viewvc/llvm-project?rev=123530&view=rev Log: Add a clarification about merging constants with and without unnamed_addr. Modified: llvm/trunk/docs/LangRef.html Modified: llvm/trunk/docs/LangRef.html URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/docs/LangRef.html?rev=123530&r1=123529&r2=123530&view=diff ============================================================================== --- llvm/trunk/docs/LangRef.html (original) +++ llvm/trunk/docs/LangRef.html Sat Jan 15 02:20:57 2011 @@ -848,7 +848,10 @@

Global variables can be marked with unnamed_addr which indicates that the address is not significant, only the content. Constants marked - like this can be merged if they have the same content.

+ like this can be merged with other constants if they have the same + initializer. Note that a constant with significant address can + be merged with a unnamed_addr constant, the result being a + constant whose address is significant.

A global variable may be declared to reside in a target-specific numbered address space. For targets that support them, address spaces may affect how From rafael.espindola at gmail.com Sat Jan 15 02:24:15 2011 From: rafael.espindola at gmail.com (Rafael Espindola) Date: Sat, 15 Jan 2011 08:24:15 -0000 Subject: [llvm-commits] [llvm-gcc-4.2] r123532 - /llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp Message-ID: <20110115082415.33CE32A6C12C@llvm.org> Author: rafael Date: Sat Jan 15 02:24:15 2011 New Revision: 123532 URL: http://llvm.org/viewvc/llvm-project?rev=123532&view=rev Log: Add unnamed_addr to declarations. Modified: llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp 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=123532&r1=123531&r2=123532&view=diff ============================================================================== --- llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp (original) +++ llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp Sat Jan 15 02:24:15 2011 @@ -1169,6 +1169,7 @@ AliaseeName, TheModule); else assert(0 && "Unsuported global value"); + Aliasee->setUnnamedAddr(true); } else { error ("%J%qD aliased to undefined symbol %qs", decl, decl, AliaseeName); timevar_pop(TV_LLVM_GLOBALS); @@ -1767,6 +1768,7 @@ GV = new GlobalVariable(*TheModule, Ty, false, GlobalValue::ExternalLinkage, 0, Name); + GV->setUnnamedAddr(true); // Check for external weak linkage. if (DECL_EXTERNAL(decl) && DECL_WEAK(decl)) GV->setLinkage(GlobalValue::ExternalWeakLinkage); From nicholas at mxc.ca Sat Jan 15 03:16:12 2011 From: nicholas at mxc.ca (Nick Lewycky) Date: Sat, 15 Jan 2011 09:16:12 -0000 Subject: [llvm-commits] [llvm] r123533 - in /llvm/trunk: include/llvm/IntrinsicInst.h lib/Analysis/LazyValueInfo.cpp Message-ID: <20110115091612.75AB02A6C12D@llvm.org> Author: nicholas Date: Sat Jan 15 03:16:12 2011 New Revision: 123533 URL: http://llvm.org/viewvc/llvm-project?rev=123533&view=rev Log: Teach LazyValueInfo that allocas aren't NULL. Over all of llvm-test, this saves half a million non-local queries, each of which would otherwise have triggered a linear scan over a basic block. Also fix a fixme for memory intrinsics which dereference pointers. With this, we prove that a pointer is non-null because it was dereferenced by an intrinsic 112 times in llvm-test. Modified: llvm/trunk/include/llvm/IntrinsicInst.h llvm/trunk/lib/Analysis/LazyValueInfo.cpp Modified: llvm/trunk/include/llvm/IntrinsicInst.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/IntrinsicInst.h?rev=123533&r1=123532&r2=123533&view=diff ============================================================================== --- llvm/trunk/include/llvm/IntrinsicInst.h (original) +++ llvm/trunk/include/llvm/IntrinsicInst.h Sat Jan 15 03:16:12 2011 @@ -139,6 +139,10 @@ return !getVolatileCst()->isZero(); } + unsigned getAddressSpace() const { + return cast(getRawDest()->getType())->getAddressSpace(); + } + /// getDest - This is just like getRawDest, but it strips off any cast /// instructions that feed it, giving the original input. The returned /// value is guaranteed to be a pointer. Modified: llvm/trunk/lib/Analysis/LazyValueInfo.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/LazyValueInfo.cpp?rev=123533&r1=123532&r2=123533&view=diff ============================================================================== --- llvm/trunk/lib/Analysis/LazyValueInfo.cpp (original) +++ llvm/trunk/lib/Analysis/LazyValueInfo.cpp Sat Jan 15 03:16:12 2011 @@ -17,6 +17,7 @@ #include "llvm/Analysis/ValueTracking.h" #include "llvm/Constants.h" #include "llvm/Instructions.h" +#include "llvm/IntrinsicInst.h" #include "llvm/Analysis/ConstantFolding.h" #include "llvm/Target/TargetData.h" #include "llvm/Support/CFG.h" @@ -544,6 +545,11 @@ return ODCacheUpdater.markResult(solveBlockValuePHINode(BBLV, PN, BB)); } + if (AllocaInst *AI = dyn_cast(BBI)) { + BBLV = LVILatticeVal::getNot(ConstantPointerNull::get(AI->getType())); + return ODCacheUpdater.markResult(true); + } + // We can only analyze the definitions of certain classes of instructions // (integral binops and casts at the moment), so bail if this isn't one. LVILatticeVal Result; @@ -580,7 +586,19 @@ GetUnderlyingObject(S->getPointerOperand()) == GetUnderlyingObject(Ptr); } - // FIXME: llvm.memset, etc. + if (MemIntrinsic *MI = dyn_cast(I)) { + if (MI->isVolatile()) return false; + if (MI->getAddressSpace() != 0) return false; + + // FIXME: check whether it has a valuerange that excludes zero? + ConstantInt *Len = dyn_cast(MI->getLength()); + if (!Len || Len->isZero()) return false; + + if (MI->getRawDest() == Ptr || MI->getDest() == Ptr) + return true; + if (MemTransferInst *MTI = dyn_cast(MI)) + return MTI->getRawSource() == Ptr || MTI->getSource() == Ptr; + } return false; } @@ -592,10 +610,14 @@ // then we know that the pointer can't be NULL. bool NotNull = false; if (Val->getType()->isPointerTy()) { - for (BasicBlock::iterator BI = BB->begin(), BE = BB->end();BI != BE;++BI){ - if (InstructionDereferencesPointer(BI, Val)) { - NotNull = true; - break; + if (isa(Val)) { + NotNull = true; + } else { + for (BasicBlock::iterator BI = BB->begin(), BE = BB->end();BI != BE;++BI){ + if (InstructionDereferencesPointer(BI, Val)) { + NotNull = true; + break; + } } } } From aggarwa4 at illinois.edu Sat Jan 15 03:42:52 2011 From: aggarwa4 at illinois.edu (Arushi Aggarwal) Date: Sat, 15 Jan 2011 09:42:52 -0000 Subject: [llvm-commits] [poolalloc] r123534 - /poolalloc/trunk/lib/DSA/StdLibPass.cpp Message-ID: <20110115094252.915F52A6C12C@llvm.org> Author: aggarwa4 Date: Sat Jan 15 03:42:52 2011 New Revision: 123534 URL: http://llvm.org/viewvc/llvm-project?rev=123534&view=rev Log: Always check if a DSGraph has a DSNode for a Value. Use hasNodeForValue to check for node, instead of getNodeForValue which creates a new node if one does not exist. Modified: poolalloc/trunk/lib/DSA/StdLibPass.cpp Modified: poolalloc/trunk/lib/DSA/StdLibPass.cpp URL: http://llvm.org/viewvc/llvm-project/poolalloc/trunk/lib/DSA/StdLibPass.cpp?rev=123534&r1=123533&r2=123534&view=diff ============================================================================== --- poolalloc/trunk/lib/DSA/StdLibPass.cpp (original) +++ poolalloc/trunk/lib/DSA/StdLibPass.cpp Sat Jan 15 03:42:52 2011 @@ -374,18 +374,18 @@ for (unsigned y = 1; y < CI->getNumOperands(); ++y) if (recFuncs[x].action.read[y]) if (isa(CI->getOperand(y)->getType())) - if (DSNode * Node=Graph->getNodeForValue(CI->getOperand(y)).getNode()) - Node->setReadMarker(); + if (Graph->hasNodeForValue(CI->getOperand(y))) + Graph->getNodeForValue(CI->getOperand(y)).getNode()->setReadMarker(); for (unsigned y = 1; y < CI->getNumOperands(); ++y) if (recFuncs[x].action.write[y]) if (isa(CI->getOperand(y)->getType())) - if (DSNode * Node=Graph->getNodeForValue(CI->getOperand(y)).getNode()) - Node->setModifiedMarker(); + if (Graph->hasNodeForValue(CI->getOperand(y))) + Graph->getNodeForValue(CI->getOperand(y)).getNode()->setModifiedMarker(); for (unsigned y = 1; y < CI->getNumOperands(); ++y) if (recFuncs[x].action.heap[y]) if (isa(CI->getOperand(y)->getType())) - if (DSNode * Node=Graph->getNodeForValue(CI->getOperand(y)).getNode()) - Node->setHeapMarker(); + if (Graph->hasNodeForValue(CI->getOperand(y))) + Graph->getNodeForValue(CI->getOperand(y)).getNode()->setHeapMarker(); // // Merge the DSNoes for return values and parameters as @@ -397,7 +397,8 @@ if (recFuncs[x].action.mergeAllArgs || recFuncs[x].action.mergeWithRet) for (unsigned y = 1; y < CI->getNumOperands(); ++y) if (isa(CI->getOperand(y)->getType())) - toMerge.push_back(Graph->getNodeForValue(CI->getOperand(y))); + if (Graph->hasNodeForValue(CI->getOperand(y))) + toMerge.push_back(Graph->getNodeForValue(CI->getOperand(y))); for (unsigned y = 1; y < toMerge.size(); ++y) toMerge[0].mergeWith(toMerge[y]); @@ -412,8 +413,8 @@ } for (unsigned y = 1; y < CI->getNumOperands(); ++y){ if (isa(CI->getOperand(y)->getType())){ - if (DSNode * Node=Graph->getNodeForValue(CI->getOperand(y)).getNode()){ - Node->foldNodeCompletely(); + if (Graph->hasNodeForValue(CI->getOperand(y))){ + Graph->getNodeForValue(CI->getOperand(y)).getNode()->foldNodeCompletely(); NumNodesFoldedInStdLib++; } } @@ -446,18 +447,18 @@ for (unsigned y = 1; y < CI->getNumOperands(); ++y) if (recFuncs[x].action.read[y]) if (isa(CI->getOperand(y)->getType())) - if (DSNode * Node=Graph->getNodeForValue(CI->getOperand(y)).getNode()) - Node->setReadMarker(); + if (Graph->hasNodeForValue(CI->getOperand(y))) + Graph->getNodeForValue(CI->getOperand(y)).getNode()->setReadMarker(); for (unsigned y = 1; y < CI->getNumOperands(); ++y) if (recFuncs[x].action.write[y]) if (isa(CI->getOperand(y)->getType())) - if (DSNode * Node=Graph->getNodeForValue(CI->getOperand(y)).getNode()) - Node->setModifiedMarker(); + if (Graph->hasNodeForValue(CI->getOperand(y))) + Graph->getNodeForValue(CI->getOperand(y)).getNode()->setModifiedMarker(); for (unsigned y = 1; y < CI->getNumOperands(); ++y) if (recFuncs[x].action.heap[y]) if (isa(CI->getOperand(y)->getType())) - if (DSNode * Node=Graph->getNodeForValue(CI->getOperand(y)).getNode()) - Node->setHeapMarker(); + if (Graph->hasNodeForValue(CI->getOperand(y))) + Graph->getNodeForValue(CI->getOperand(y)).getNode()->setHeapMarker(); // // Merge the DSNoes for return values and parameters as @@ -469,7 +470,8 @@ if (recFuncs[x].action.mergeAllArgs || recFuncs[x].action.mergeWithRet) for (unsigned y = 1; y < CI->getNumOperands(); ++y) if (isa(CI->getOperand(y)->getType())) - toMerge.push_back(Graph->getNodeForValue(CI->getOperand(y))); + if (Graph->hasNodeForValue(CI->getOperand(y))) + toMerge.push_back(Graph->getNodeForValue(CI->getOperand(y))); for (unsigned y = 1; y < toMerge.size(); ++y) toMerge[0].mergeWith(toMerge[y]); @@ -484,7 +486,8 @@ } for (unsigned y = 1; y < CI->getNumOperands(); ++y) if (isa(CI->getOperand(y)->getType())){ - if (DSNode * Node=Graph->getNodeForValue(CI->getOperand(y)).getNode()){ + if (Graph->hasNodeForValue(CI->getOperand(y))){ + DSNode * Node=Graph->getNodeForValue(CI->getOperand(y)).getNode(); Node->foldNodeCompletely(); NumNodesFoldedInStdLib++; } From nicholas at mxc.ca Sat Jan 15 04:16:23 2011 From: nicholas at mxc.ca (Nick Lewycky) Date: Sat, 15 Jan 2011 10:16:23 -0000 Subject: [llvm-commits] [llvm] r123535 - /llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp Message-ID: <20110115101623.D8D5F2A6C12C@llvm.org> Author: nicholas Date: Sat Jan 15 04:16:23 2011 New Revision: 123535 URL: http://llvm.org/viewvc/llvm-project?rev=123535&view=rev Log: Add a cache that protects mergefunc's internals from more surprises in DenseSet. Also, replace tabs with spaces. Yes, it's 2011. Modified: llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp Modified: llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp?rev=123535&r1=123534&r2=123535&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp Sat Jan 15 04:16:23 2011 @@ -109,10 +109,23 @@ Func = NULL; } + bool &getOrInsertCachedComparison(const ComparableFunction &Other, + bool &inserted) const { + typedef DenseMap::iterator iterator; + std::pair p = + CompareResultCache.insert(std::make_pair(Other.getFunc(), false)); + inserted = p.second; + return p.first->second; + } + private: explicit ComparableFunction(unsigned Hash) : Func(NULL), Hash(Hash), TD(NULL) {} + // DenseMap::grow() triggers a recomparison of all keys in the map, which is + // wildly expensive. This cache tries to preserve known results. + mutable DenseMap CompareResultCache; + AssertingVH Func; unsigned Hash; TargetData *TD; @@ -708,10 +721,10 @@ if (Instruction *I = dyn_cast(U.getUser())) { Remove(I->getParent()->getParent()); } else if (isa(U.getUser())) { - // do nothing + // do nothing } else if (Constant *C = dyn_cast(U.getUser())) { - for (Value::use_iterator CUI = C->use_begin(), CUE = C->use_end(); - CUI != CUE; ++CUI) + for (Value::use_iterator CUI = C->use_begin(), CUE = C->use_end(); + CUI != CUE; ++CUI) Worklist.push_back(*CUI); } } @@ -777,6 +790,15 @@ return false; assert(LHS.getTD() == RHS.getTD() && "Comparing functions for different targets"); - return FunctionComparator(LHS.getTD(), - LHS.getFunc(), RHS.getFunc()).Compare(); + + bool inserted; + bool &result1 = LHS.getOrInsertCachedComparison(RHS, inserted); + if (!inserted) + return result1; + bool &result2 = RHS.getOrInsertCachedComparison(LHS, inserted); + if (!inserted) + return result1 = result2; + + return result1 = result2 = FunctionComparator(LHS.getTD(), LHS.getFunc(), + RHS.getFunc()).Compare(); } From aggarwa4 at illinois.edu Sat Jan 15 04:54:23 2011 From: aggarwa4 at illinois.edu (Arushi Aggarwal) Date: Sat, 15 Jan 2011 10:54:23 -0000 Subject: [llvm-commits] [poolalloc] r123536 - /poolalloc/trunk/test/TEST.poolalloc.Makefile Message-ID: <20110115105423.39E392A6C12C@llvm.org> Author: aggarwa4 Date: Sat Jan 15 04:54:23 2011 New Revision: 123536 URL: http://llvm.org/viewvc/llvm-project?rev=123536&view=rev Log: Do not dump output when running tests. Modified: poolalloc/trunk/test/TEST.poolalloc.Makefile Modified: poolalloc/trunk/test/TEST.poolalloc.Makefile URL: http://llvm.org/viewvc/llvm-project/poolalloc/trunk/test/TEST.poolalloc.Makefile?rev=123536&r1=123535&r2=123536&view=diff ============================================================================== --- poolalloc/trunk/test/TEST.poolalloc.Makefile (original) +++ poolalloc/trunk/test/TEST.poolalloc.Makefile Sat Jan 15 04:54:23 2011 @@ -238,7 +238,6 @@ Output/%.mallocrepl.diff-nat \ Output/%.onlyoverhead.diff-nat \ Output/%.LOC.txt - @-cat $< @echo > $@ @echo "---------------------------------------------------------------" >> $@ @echo ">>> ========= '$(RELDIR)/$*' Program" >> $@ From benny.kra at googlemail.com Sat Jan 15 05:25:34 2011 From: benny.kra at googlemail.com (Benjamin Kramer) Date: Sat, 15 Jan 2011 11:25:34 -0000 Subject: [llvm-commits] [llvm] r123537 - /llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Message-ID: <20110115112534.4316C2A6C12C@llvm.org> Author: d0k Date: Sat Jan 15 05:25:34 2011 New Revision: 123537 URL: http://llvm.org/viewvc/llvm-project?rev=123537&view=rev Log: Try to unbreak selfhost. Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp?rev=123537&r1=123536&r2=123537&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Sat Jan 15 05:25:34 2011 @@ -2169,6 +2169,7 @@ // and we know how to evaluate it by moving the bitcast from the pointer // operand to the value operand. } else if (CE->getOpcode() == Instruction::BitCast && + isa(CE->getOperand(0)) && CE->getType()->isPointerTy() && CE->getOperand(0)->getType()->isPointerTy()) { GlobalVariable *GV = cast(CE->getOperand(0)); From 6yearold at gmail.com Sat Jan 15 05:58:52 2011 From: 6yearold at gmail.com (arrowdodger) Date: Sat, 15 Jan 2011 14:58:52 +0300 Subject: [llvm-commits] [PATCH][CMake] include/Config/config.h.cmake reorganization. In-Reply-To: <87fwsvcu4j.fsf@wanadoo.es> References: <87lj2ncvk1.fsf@wanadoo.es> <87fwsvcu4j.fsf@wanadoo.es> Message-ID: Last part of the patch. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/7bd8e8de/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: config.h.cmake.patch Type: application/octet-stream Size: 9174 bytes Desc: not available Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/7bd8e8de/attachment.obj From ofv at wanadoo.es Sat Jan 15 07:35:37 2011 From: ofv at wanadoo.es (Oscar Fuentes) Date: Sat, 15 Jan 2011 13:35:37 -0000 Subject: [llvm-commits] [llvm] r123539 - in /llvm/trunk: cmake/config-ix.cmake include/llvm/Config/config.h.cmake Message-ID: <20110115133537.6E8F02A6C12C@llvm.org> Author: ofv Date: Sat Jan 15 07:35:37 2011 New Revision: 123539 URL: http://llvm.org/viewvc/llvm-project?rev=123539&view=rev Log: Make config.h.cmake similar to config.h.in Patch by arrowdodger! Modified: llvm/trunk/cmake/config-ix.cmake llvm/trunk/include/llvm/Config/config.h.cmake Modified: llvm/trunk/cmake/config-ix.cmake URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/cmake/config-ix.cmake?rev=123539&r1=123538&r2=123539&view=diff ============================================================================== --- llvm/trunk/cmake/config-ix.cmake (original) +++ llvm/trunk/cmake/config-ix.cmake Sat Jan 15 07:35:37 2011 @@ -179,7 +179,11 @@ # Define LLVM_MULTITHREADED if gcc atomic builtins exists. include(CheckAtomic) -set(ENABLE_PIC ${LLVM_ENABLE_PIC}) +if( LLVM_ENABLE_PIC ) + set(ENABLE_PIC 1) +else() + set(ENABLE_PIC 0) +endif() include(CheckCXXCompilerFlag) Modified: llvm/trunk/include/llvm/Config/config.h.cmake URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Config/config.h.cmake?rev=123539&r1=123538&r2=123539&view=diff ============================================================================== --- llvm/trunk/include/llvm/Config/config.h.cmake (original) +++ llvm/trunk/include/llvm/Config/config.h.cmake Sat Jan 15 07:35:37 2011 @@ -8,16 +8,16 @@ /* Relative directory for resource files */ #define CLANG_RESOURCE_DIR "${CLANG_RESOURCE_DIR}" -/* 32 bit multilib directory */ +/* 32 bit multilib directory. */ #define CXX_INCLUDE_32BIT_DIR "${CXX_INCLUDE_32BIT_DIR}" -/* 64 bit multilib directory */ +/* 64 bit multilib directory. */ #define CXX_INCLUDE_64BIT_DIR "${CXX_INCLUDE_64BIT_DIR}" -/* Architecture of libstdc++ headers */ +/* Arch the libstdc++ headers. */ #define CXX_INCLUDE_ARCH "${CXX_INCLUDE_ARCH}" -/* Directory clang will search for libstdc++ headers */ +/* Directory with the libstdc++ headers. */ #define CXX_INCLUDE_ROOT "${CXX_INCLUDE_ROOT}" /* Directories clang will search for headers */ @@ -299,7 +299,7 @@ /* Have pthread_getspecific */ #cmakedefine HAVE_PTHREAD_GETSPECIFIC ${HAVE_PTHREAD_GETSPECIFIC} -/* Have pthread.h */ +/* Define to 1 if you have the header file. */ #cmakedefine HAVE_PTHREAD_H ${HAVE_PTHREAD_H} /* Have pthread_mutex_lock */ @@ -368,18 +368,9 @@ /* Set to 1 if the std::isinf function is found in */ #undef HAVE_STD_ISINF_IN_CMATH -/* Define to 1 if you have the `_chsize_s' function. */ -#cmakedefine HAVE__CHSIZE_S ${HAVE__CHSIZE_S} - -/* define if the compiler implements namespaces */ -#undef HAVE_NAMESPACES - /* Set to 1 if the std::isnan function is found in */ #undef HAVE_STD_ISNAN_IN_CMATH -/* Does not have std namespace iterator */ -#undef HAVE_STD_ITERATOR - /* Define to 1 if you have the `strchr' function. */ #undef HAVE_STRCHR @@ -407,6 +398,9 @@ /* Define to 1 if you have the `strrchr' function. */ #undef HAVE_STRRCHR +/* Define to 1 if you have the `strtof' function. */ +#cmakedefine HAVE_STRTOF ${HAVE_STRTOF} + /* Define to 1 if you have the `strtoll' function. */ #cmakedefine HAVE_STRTOLL ${HAVE_STRTOLL} @@ -416,30 +410,6 @@ /* Define to 1 if you have the `sysconf' function. */ #undef HAVE_SYSCONF -/* Does not have forward iterator */ -#undef HAVE_FWD_ITERATOR - -/* Does not have bi-directional iterator */ -#undef HAVE_BI_ITERATOR - -/* Does not have */ -#undef HAVE_GLOBAL_HASH_MAP - -/* Does not have hash_set in global namespace */ -#undef HAVE_GLOBAL_HASH_SET - -/* Does not have ext/hash_map */ -#undef HAVE_GNU_EXT_HASH_MAP - -/* Does not have hash_set in gnu namespace */ -#undef HAVE_GNU_EXT_HASH_SET - -/* Does not have ext/hash_map> */ -#undef HAVE_STD_EXT_HASH_MAP - -/* Does not have hash_set in std namespace */ -#undef HAVE_STD_EXT_HASH_SET - /* Define to 1 if you have the header file, and it defines `DIR'. */ #cmakedefine HAVE_SYS_DIR_H ${HAVE_SYS_DIR_H} @@ -475,12 +445,6 @@ /* Define to 1 if you have that is POSIX.1 compatible. */ #cmakedefine HAVE_SYS_WAIT_H ${HAVE_SYS_WAIT_H} -/* Define if the neat program is available */ -#cmakedefine HAVE_TWOPI ${HAVE_TWOPI} - -/* Define to 1 if the system has the type `uint64_t'. */ -#cmakedefine HAVE_UINT64_T ${HAVE_UINT64_T} - /* Define to 1 if you have the header file. */ #cmakedefine HAVE_TERMIOS_H ${HAVE_TERMIOS_H} @@ -499,6 +463,15 @@ /* Define to 1 if you have the header file. */ #cmakedefine HAVE_WINDOWS_H ${HAVE_WINDOWS_H} +/* Define if the xdot.py program is available */ +#undef HAVE_XDOT_PY + +/* Define to 1 if you have the `__dso_handle' function. */ +#undef HAVE___DSO_HANDLE + +/* Linker version detected at compile time. */ +#undef HOST_LINK_VERSION + /* Installation directory for binary executables */ #undef LLVM_BINDIR @@ -508,6 +481,9 @@ /* Installation directory for documentation */ #undef LLVM_DATADIR +/* Installation directory for documentation */ +#undef LLVM_DOCSDIR + /* Installation directory for config files */ #undef LLVM_ETCDIR @@ -529,15 +505,24 @@ /* Build multithreading support into LLVM */ #cmakedefine LLVM_MULTITHREADED ${LLVM_MULTITHREADED} +/* LLVM architecture name for the native architecture, if available */ +#cmakedefine LLVM_NATIVE_ARCH ${LLVM_NATIVE_ARCH} + +/* LLVM name for the native AsmPrinter init function, if available */ +#cmakedefine LLVM_NATIVE_ASMPRINTER LLVMInitialize${LLVM_NATIVE_ARCH}AsmPrinter + +/* LLVM name for the native Target init function, if available */ +#cmakedefine LLVM_NATIVE_TARGET LLVMInitialize${LLVM_NATIVE_ARCH}Target + +/* LLVM name for the native TargetInfo init function, if available */ +#cmakedefine LLVM_NATIVE_TARGETINFO LLVMInitialize${LLVM_NATIVE_ARCH}TargetInfo + /* Define if this is Unixish platform */ #cmakedefine LLVM_ON_UNIX ${LLVM_ON_UNIX} /* Define if this is Win32ish platform */ #cmakedefine LLVM_ON_WIN32 ${LLVM_ON_WIN32} -/* Added by Kevin -- Maximum path length */ -#cmakedefine MAXPATHLEN ${MAXPATHLEN} - /* Define to path to circo program if found or 'echo circo' otherwise */ #cmakedefine LLVM_PATH_CIRCO "${LLVM_PATH_CIRCO}" @@ -562,6 +547,9 @@ /* Define to path to twopi program if found or 'echo twopi' otherwise */ #cmakedefine LLVM_PATH_TWOPI "${LLVM_PATH_TWOPI}" +/* Define to path to xdot.py program if found or 'echo xdot.py' otherwise */ +#undef LLVM_PATH_XDOT_PY + /* Installation prefix directory */ #cmakedefine LLVM_PREFIX "${LLVM_PREFIX}" @@ -607,6 +595,9 @@ /* Define as the return type of signal handlers (`int' or `void'). */ #cmakedefine RETSIGTYPE ${RETSIGTYPE} +/* Define to 1 if the `S_IS*' macros in do not work properly. */ +#undef STAT_MACROS_BROKEN + /* If using the C implementation of alloca, define if you know the direction of stack growth for your system; otherwise it will be automatically deduced at runtime. @@ -627,28 +618,37 @@ /* Define to 1 if your declares `struct tm'. */ #undef TM_IN_SYS_TIME +/* Define if we have the oprofile JIT-support library */ +#undef USE_OPROFILE + /* Define if use udis86 library */ #undef USE_UDIS86 -/* Define to 1 if `lex' declares `yytext' as a `char *' by default, not a - `char[]'. */ -#undef YYTEXT_POINTER - /* Define to empty if `const' does not conform to ANSI C. */ #undef const /* Define to a type to use for `error_t' if it is not otherwise available. */ #cmakedefine error_t ${error_t} -/* Define to a type to use for `mode_t' if it is not otherwise available. */ -#cmakedefine mode_t ${mode_t} - /* Define to `int' if does not define. */ #undef pid_t /* Define to `unsigned int' if does not define. */ #undef size_t +/* Define if the neat program is available */ +#cmakedefine HAVE_TWOPI ${HAVE_TWOPI} + +/* Define to 1 if the system has the type `uint64_t'. */ +#cmakedefine HAVE_UINT64_T ${HAVE_UINT64_T} + +/* Define to 1 if `lex' declares `yytext' as a `char *' by default, not a + `char[]'. */ +#undef YYTEXT_POINTER + +/* Define to a type to use for `mode_t' if it is not otherwise available. */ +#cmakedefine mode_t ${mode_t} + /* Define to a function replacing strtoll */ #cmakedefine strtoll ${strtoll} @@ -661,16 +661,40 @@ /* Define to a function implementing strdup */ #cmakedefine strdup ${strdup} -/* LLVM architecture name for the native architecture, if available */ -#cmakedefine LLVM_NATIVE_ARCH ${LLVM_NATIVE_ARCH} - -/* LLVM name for the native Target init function, if available */ -#cmakedefine LLVM_NATIVE_TARGET LLVMInitialize${LLVM_NATIVE_ARCH}Target - -/* LLVM name for the native TargetInfo init function, if available */ -#cmakedefine LLVM_NATIVE_TARGETINFO LLVMInitialize${LLVM_NATIVE_ARCH}TargetInfo - -/* LLVM name for the native AsmPrinter init function, if available */ -#cmakedefine LLVM_NATIVE_ASMPRINTER LLVMInitialize${LLVM_NATIVE_ARCH}AsmPrinter +/* Define to 1 if you have the `_chsize_s' function. */ +#cmakedefine HAVE__CHSIZE_S ${HAVE__CHSIZE_S} + +/* define if the compiler implements namespaces */ +#undef HAVE_NAMESPACES + +/* Does not have std namespace iterator */ +#undef HAVE_STD_ITERATOR + +/* Does not have forward iterator */ +#undef HAVE_FWD_ITERATOR + +/* Does not have bi-directional iterator */ +#undef HAVE_BI_ITERATOR + +/* Does not have */ +#undef HAVE_GLOBAL_HASH_MAP + +/* Does not have hash_set in global namespace */ +#undef HAVE_GLOBAL_HASH_SET + +/* Does not have ext/hash_map */ +#undef HAVE_GNU_EXT_HASH_MAP + +/* Does not have hash_set in gnu namespace */ +#undef HAVE_GNU_EXT_HASH_SET + +/* Does not have ext/hash_map> */ +#undef HAVE_STD_EXT_HASH_MAP + +/* Does not have hash_set in std namespace */ +#undef HAVE_STD_EXT_HASH_SET + +/* Added by Kevin -- Maximum path length */ +#cmakedefine MAXPATHLEN ${MAXPATHLEN} #endif From ofv at wanadoo.es Sat Jan 15 07:39:08 2011 From: ofv at wanadoo.es (=?utf-8?Q?=C3=93scar_Fuentes?=) Date: Sat, 15 Jan 2011 14:39:08 +0100 Subject: [llvm-commits] [PATCH][CMake] include/Config/config.h.cmake reorganization. In-Reply-To: (arrowdodger's message of "Sat, 15 Jan 2011 14:58:52 +0300") References: <87lj2ncvk1.fsf@wanadoo.es> <87fwsvcu4j.fsf@wanadoo.es> Message-ID: <87sjwub81v.fsf@wanadoo.es> arrowdodger <6yearold at gmail.com> writes: > Last part of the patch. Committed. Thanks! From jay.foad at gmail.com Sat Jan 15 07:43:31 2011 From: jay.foad at gmail.com (Jay Foad) Date: Sat, 15 Jan 2011 13:43:31 +0000 Subject: [llvm-commits] [PATCH] remove unnecessary specialization of OperandTraits<> Message-ID: OperandTraits<> is always used with a concrete subclass of User. There is no need to specialize it for User itself. Tested with "make check". I also tested a patch that only removed the out-of-line definitions of the methods declared in OperandTraits, just to double-check that they weren't being used for anything. OK to commit? Thanks, Jay. -------------- next part -------------- A non-text attachment was scrubbed... Name: remove-operandtraits-user Type: application/octet-stream Size: 997 bytes Desc: not available Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/3d9adcd0/attachment-0001.obj From jay.foad at gmail.com Sat Jan 15 08:38:11 2011 From: jay.foad at gmail.com (Jay Foad) Date: Sat, 15 Jan 2011 14:38:11 +0000 Subject: [llvm-commits] [PATCH] new source file User.cpp Message-ID: This patch moves the implementation of User methods out of Use.cpp and Value.cpp, and into a new source file User.cpp. Tested with "make check". Comments? Or, OK to commit? Thanks, Jay. -------------- next part -------------- A non-text attachment was scrubbed... Name: introduce-user-cpp Type: application/octet-stream Size: 8605 bytes Desc: not available Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/f802aaf0/attachment.obj From aggarwa4 at illinois.edu Sat Jan 15 10:05:16 2011 From: aggarwa4 at illinois.edu (Arushi Aggarwal) Date: Sat, 15 Jan 2011 16:05:16 -0000 Subject: [llvm-commits] [poolalloc] r123540 - /poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp Message-ID: <20110115160516.EB9DA2A6C12C@llvm.org> Author: aggarwa4 Date: Sat Jan 15 10:05:16 2011 New Revision: 123540 URL: http://llvm.org/viewvc/llvm-project?rev=123540&view=rev Log: Enable checking of graphs in DEBUG mode. Modified: poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp Modified: poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp URL: http://llvm.org/viewvc/llvm-project/poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp?rev=123540&r1=123539&r2=123540&view=diff ============================================================================== --- poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp (original) +++ poolalloc/trunk/lib/PoolAllocate/TransformFunctionBody.cpp Sat Jan 15 10:05:16 2011 @@ -996,8 +996,7 @@ #ifndef NDEBUG // Verify that all potential callees at call site have the same DS graph. -#if 0 - DSCallGraph::callee_iterator E = PAInfo.getCallGraph().callee_end(OrigInst); + DSCallGraph::callee_iterator E = Graphs.getCallGraph().callee_end(OrigInst); for (; I != E; ++I) { const Function * F = *I; assert (F); @@ -1005,7 +1004,6 @@ assert(CalleeGraph == Graphs.getDSGraph(**I) && "Callees at call site do not have a common graph!"); } -#endif #endif // Find the DS nodes for the arguments that need to be added, if any. From nicholas at mxc.ca Sat Jan 15 12:14:21 2011 From: nicholas at mxc.ca (Nick Lewycky) Date: Sat, 15 Jan 2011 18:14:21 -0000 Subject: [llvm-commits] [llvm] r123541 - in /llvm/trunk: lib/Transforms/IPO/ConstantMerge.cpp test/Transforms/ConstantMerge/2011-01-15-EitherOrder.ll Message-ID: <20110115181421.880EA2A6C12C@llvm.org> Author: nicholas Date: Sat Jan 15 12:14:21 2011 New Revision: 123541 URL: http://llvm.org/viewvc/llvm-project?rev=123541&view=rev Log: Make constmerge a two-pass algorithm so that it won't miss merging opporuntities. Fixes PR8978. Added: llvm/trunk/test/Transforms/ConstantMerge/2011-01-15-EitherOrder.ll Modified: llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp Modified: llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp?rev=123541&r1=123540&r2=123541&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp Sat Jan 15 12:14:21 2011 @@ -101,15 +101,18 @@ continue; } - // Only process constants with initializers in the default addres space. + // Only process constants with initializers in the default address space. if (!GV->isConstant() ||!GV->hasDefinitiveInitializer() || - GV->getType()->getAddressSpace() != 0 || !GV->getSection().empty() || + GV->getType()->getAddressSpace() != 0 || GV->hasSection() || // Don't touch values marked with attribute(used). UsedGlobals.count(GV)) continue; + // Start by filling slots with only the globals we aren't allowed to + // delete because they're externally visible. + if (GV->hasLocalLinkage()) + continue; - Constant *Init = GV->getInitializer(); // Check to see if the initializer is already known. @@ -117,7 +120,32 @@ if (Slot == 0) { // Nope, add it to the map. Slot = GV; - } else if (GV->hasLocalLinkage()) { // Yup, this is a duplicate! + } + } + + for (Module::global_iterator GVI = M.global_begin(), E = M.global_end(); + GVI != E; ) { + GlobalVariable *GV = GVI++; + + // Only process constants with initializers in the default address space. + if (!GV->isConstant() ||!GV->hasDefinitiveInitializer() || + GV->getType()->getAddressSpace() != 0 || GV->hasSection() || + // Don't touch values marked with attribute(used). + UsedGlobals.count(GV)) + continue; + + // Only look at the remaining globals now. + if (!GV->hasLocalLinkage()) + continue; + + Constant *Init = GV->getInitializer(); + + // Check to see if the initializer is already known. + GlobalVariable *&Slot = CMap[Init]; + + if (Slot == 0) { // Nope, add it to the map. + Slot = GV; + } else { // Yup, this is a duplicate! // Make all uses of the duplicate constant use the canonical version. Replacements.push_back(std::make_pair(GV, Slot)); } @@ -135,6 +163,8 @@ Replacements[i].first->replaceAllUsesWith(Replacements[i].second); // Delete the global value from the module. + assert(Replacements[i].first->hasLocalLinkage() && + "Refusing to delete an externally visible global variable."); Replacements[i].first->eraseFromParent(); } Added: llvm/trunk/test/Transforms/ConstantMerge/2011-01-15-EitherOrder.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/ConstantMerge/2011-01-15-EitherOrder.ll?rev=123541&view=auto ============================================================================== --- llvm/trunk/test/Transforms/ConstantMerge/2011-01-15-EitherOrder.ll (added) +++ llvm/trunk/test/Transforms/ConstantMerge/2011-01-15-EitherOrder.ll Sat Jan 15 12:14:21 2011 @@ -0,0 +1,18 @@ +; RUN: opt -constmerge %s -S -o - | FileCheck %s +; PR8978 + +declare i32 @zed(%struct.foobar*, %struct.foobar*) + +%struct.foobar = type { i32 } +; CHECK: bar.d + at bar.d = constant %struct.foobar zeroinitializer, align 4 +; CHECK-NOT: foo.d + at foo.d = internal constant %struct.foobar zeroinitializer, align 4 +define i32 @main() nounwind ssp { +entry: +; CHECK: bar.d + %call2 = tail call i32 @zed(%struct.foobar* @foo.d, %struct.foobar* @bar.d) +nounwind + ret i32 0 +} + From nicholas at mxc.ca Sat Jan 15 12:42:52 2011 From: nicholas at mxc.ca (Nick Lewycky) Date: Sat, 15 Jan 2011 18:42:52 -0000 Subject: [llvm-commits] [llvm] r123543 - /llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp Message-ID: <20110115184252.DB3292A6C12C@llvm.org> Author: nicholas Date: Sat Jan 15 12:42:52 2011 New Revision: 123543 URL: http://llvm.org/viewvc/llvm-project?rev=123543&view=rev Log: Add missing whitespace. Modified: llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp Modified: llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp?rev=123543&r1=123542&r2=123543&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/ConstantMerge.cpp Sat Jan 15 12:42:52 2011 @@ -102,7 +102,7 @@ } // Only process constants with initializers in the default address space. - if (!GV->isConstant() ||!GV->hasDefinitiveInitializer() || + if (!GV->isConstant() || !GV->hasDefinitiveInitializer() || GV->getType()->getAddressSpace() != 0 || GV->hasSection() || // Don't touch values marked with attribute(used). UsedGlobals.count(GV)) @@ -128,7 +128,7 @@ GlobalVariable *GV = GVI++; // Only process constants with initializers in the default address space. - if (!GV->isConstant() ||!GV->hasDefinitiveInitializer() || + if (!GV->isConstant() || !GV->hasDefinitiveInitializer() || GV->getType()->getAddressSpace() != 0 || GV->hasSection() || // Don't touch values marked with attribute(used). UsedGlobals.count(GV)) From bigcheesegs at gmail.com Sat Jan 15 12:52:33 2011 From: bigcheesegs at gmail.com (Michael J. Spencer) Date: Sat, 15 Jan 2011 18:52:33 -0000 Subject: [llvm-commits] [llvm] r123544 - in /llvm/trunk: include/llvm/Support/FileSystem.h lib/Support/Unix/PathV2.inc lib/Support/Windows/PathV2.inc lib/Support/Windows/system_error.inc Message-ID: <20110115185233.5DB5D2A6C12C@llvm.org> Author: mspencer Date: Sat Jan 15 12:52:33 2011 New Revision: 123544 URL: http://llvm.org/viewvc/llvm-project?rev=123544&view=rev Log: Support/PathV2: Implement get_magic. Modified: llvm/trunk/include/llvm/Support/FileSystem.h llvm/trunk/lib/Support/Unix/PathV2.inc llvm/trunk/lib/Support/Windows/PathV2.inc llvm/trunk/lib/Support/Windows/system_error.inc Modified: llvm/trunk/include/llvm/Support/FileSystem.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/FileSystem.h?rev=123544&r1=123543&r2=123544&view=diff ============================================================================== --- llvm/trunk/include/llvm/Support/FileSystem.h (original) +++ llvm/trunk/include/llvm/Support/FileSystem.h Sat Jan 15 12:52:33 2011 @@ -455,9 +455,11 @@ /// @param path Input path. /// @param len Number of magic bytes to get. /// @param result Set to the first \a len bytes in the file pointed to by -/// \a path. -/// @results errc::success if result has been successfully set, otherwise a -/// platform specific error_code. +/// \a path. Or the entire file if file_size(path) < len, in which +/// case result.size() returns the size of the file. +/// @results errc::success if result has been successfully set, +/// errc::value_too_large if len is larger then the file pointed to by +/// \a path, otherwise a platform specific error_code. error_code get_magic(const Twine &path, uint32_t len, SmallVectorImpl &result); Modified: llvm/trunk/lib/Support/Unix/PathV2.inc URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Support/Unix/PathV2.inc?rev=123544&r1=123543&r2=123544&view=diff ============================================================================== --- llvm/trunk/lib/Support/Unix/PathV2.inc (original) +++ llvm/trunk/lib/Support/Unix/PathV2.inc Sat Jan 15 12:52:33 2011 @@ -462,6 +462,37 @@ return success; } +error_code get_magic(const Twine &path, uint32_t len, + SmallVectorImpl &result) { + SmallString<128> PathStorage; + StringRef Path = path.toNullTerminatedStringRef(PathStorage); + result.set_size(0); + + // Open path. + std::FILE *file = std::fopen(Path.data(), "rb"); + if (file == 0) + return error_code(errno, system_category()); + + // Reserve storage. + result.reserve(len); + + // Read magic! + size_t size = std::fread(result.data(), 1, len, file); + if (std::ferror(file) != 0) { + std::fclose(file); + return error_code(errno, system_category()); + } else if (size != result.size()) { + if (std::feof(file) != 0) { + std::fclose(file); + result.set_size(size); + return make_error_code(errc::value_too_large); + } + } + std::fclose(file); + result.set_size(len); + return success; +} + } // end namespace fs } // end namespace sys } // end namespace llvm Modified: llvm/trunk/lib/Support/Windows/PathV2.inc URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Support/Windows/PathV2.inc?rev=123544&r1=123543&r2=123544&view=diff ============================================================================== --- llvm/trunk/lib/Support/Windows/PathV2.inc (original) +++ llvm/trunk/lib/Support/Windows/PathV2.inc Sat Jan 15 12:52:33 2011 @@ -616,6 +616,48 @@ return success; } +error_code get_magic(const Twine &path, uint32_t len, + SmallVectorImpl &result) { + SmallString<128> path_storage; + SmallVector path_utf16; + result.set_size(0); + + // Convert path to UTF-16. + if (error_code ec = UTF8ToUTF16(path.toStringRef(path_storage), + path_utf16)) + return ec; + + // Open file. + HANDLE file = ::CreateFileW(c_str(path_utf16), + GENERIC_READ, + FILE_SHARE_READ, + NULL, + OPEN_EXISTING, + FILE_ATTRIBUTE_READONLY, + NULL); + if (file == INVALID_HANDLE_VALUE) + return windows_error(::GetLastError()); + + // Allocate buffer. + result.reserve(len); + + // Get magic! + DWORD bytes_read = 0; + BOOL read_success = ::ReadFile(file, result.data(), len, &bytes_read, NULL); + error_code ec = windows_error(::GetLastError()); + ::CloseHandle(file); + if (!read_success || (bytes_read != len)) { + // Set result size to the number of bytes read if it's valid. + if (bytes_read >= 0 && bytes_read <= len) + result.set_size(bytes_read); + // ERROR_HANDLE_EOF is mapped to errc::value_too_large. + return ec; + } + + result.set_size(len); + return success; +} + error_code directory_iterator_construct(directory_iterator &it, StringRef path){ SmallVector path_utf16; Modified: llvm/trunk/lib/Support/Windows/system_error.inc URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Support/Windows/system_error.inc?rev=123544&r1=123543&r2=123544&view=diff ============================================================================== --- llvm/trunk/lib/Support/Windows/system_error.inc (original) +++ llvm/trunk/lib/Support/Windows/system_error.inc Sat Jan 15 12:52:33 2011 @@ -79,6 +79,7 @@ MAP_ERR_TO_COND(ERROR_FILE_EXISTS, file_exists); MAP_ERR_TO_COND(ERROR_FILE_NOT_FOUND, no_such_file_or_directory); MAP_ERR_TO_COND(ERROR_HANDLE_DISK_FULL, no_space_on_device); + MAP_ERR_TO_COND(ERROR_HANDLE_EOF, value_too_large); MAP_ERR_TO_COND(ERROR_INVALID_ACCESS, permission_denied); MAP_ERR_TO_COND(ERROR_INVALID_DRIVE, no_such_device); MAP_ERR_TO_COND(ERROR_INVALID_FUNCTION, function_not_supported); From bigcheesegs at gmail.com Sat Jan 15 12:52:41 2011 From: bigcheesegs at gmail.com (Michael J. Spencer) Date: Sat, 15 Jan 2011 18:52:41 -0000 Subject: [llvm-commits] [llvm] r123545 - /llvm/trunk/lib/Support/PathV2.cpp Message-ID: <20110115185241.E08762A6C12C@llvm.org> Author: mspencer Date: Sat Jan 15 12:52:41 2011 New Revision: 123545 URL: http://llvm.org/viewvc/llvm-project?rev=123545&view=rev Log: Support/PathV2: Implement has_magic in terms of get_magic. Modified: llvm/trunk/lib/Support/PathV2.cpp Modified: llvm/trunk/lib/Support/PathV2.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Support/PathV2.cpp?rev=123545&r1=123544&r2=123545&view=diff ============================================================================== --- llvm/trunk/lib/Support/PathV2.cpp (original) +++ llvm/trunk/lib/Support/PathV2.cpp Sat Jan 15 12:52:41 2011 @@ -686,38 +686,20 @@ } error_code has_magic(const Twine &path, const Twine &magic, bool &result) { - SmallString<128> PathStorage; SmallString<32> MagicStorage; - StringRef Path = path.toNullTerminatedStringRef(PathStorage); - StringRef Magic = magic.toNullTerminatedStringRef(MagicStorage); + StringRef Magic = magic.toStringRef(MagicStorage); + SmallString<32> Buffer; - assert(Magic.size() > 0 && "magic must be non-empty!"); - - SmallString<32> BufferStorage; - BufferStorage.reserve(Magic.size()); - - // Open file. - std::FILE *file = std::fopen(Path.data(), "rb"); - if (file == 0) - return error_code(errno, posix_category()); - size_t size = ::fread(BufferStorage.data(), 1, Magic.size(), file); - if (size != Magic.size()) { - int error = errno; - bool eof = std::feof(file) != 0; - std::fclose(file); - if (eof) { - // EOF, return false. + if (error_code ec = get_magic(path, Magic.size(), Buffer)) { + if (ec == errc::value_too_large) { + // Magic.size() > file_size(Path). result = false; return success; } - return error_code(error, posix_category()); + return ec; } - std::fclose(file); - if (std::memcmp(BufferStorage.data(), Magic.data(), Magic.size()) != 0) - result = false; - else - result = true; + result = Magic == Buffer; return success; } From bigcheesegs at gmail.com Sat Jan 15 12:52:49 2011 From: bigcheesegs at gmail.com (Michael J. Spencer) Date: Sat, 15 Jan 2011 18:52:49 -0000 Subject: [llvm-commits] [llvm] r123546 - /llvm/trunk/unittests/Support/Path.cpp Message-ID: <20110115185249.9EDB32A6C12C@llvm.org> Author: mspencer Date: Sat Jan 15 12:52:49 2011 New Revision: 123546 URL: http://llvm.org/viewvc/llvm-project?rev=123546&view=rev Log: Unittests/Support/Path: Tweak test. Modified: llvm/trunk/unittests/Support/Path.cpp Modified: llvm/trunk/unittests/Support/Path.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/unittests/Support/Path.cpp?rev=123546&r1=123545&r2=123546&view=diff ============================================================================== --- llvm/trunk/unittests/Support/Path.cpp (original) +++ llvm/trunk/unittests/Support/Path.cpp Sat Jan 15 12:52:49 2011 @@ -230,7 +230,7 @@ ASSERT_FALSE(file.has_error()); StringRef magic(i->magic_str, i->magic_str_len); file << magic; - file.flush(); + file.close(); bool res = false; ASSERT_NO_ERROR(fs::has_magic(file_pathname.c_str(), magic, res)); EXPECT_TRUE(res); From benny.kra at googlemail.com Sat Jan 15 14:30:31 2011 From: benny.kra at googlemail.com (Benjamin Kramer) Date: Sat, 15 Jan 2011 20:30:31 -0000 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp Message-ID: <20110115203031.19E282A6C12C@llvm.org> Author: d0k Date: Sat Jan 15 14:30:30 2011 New Revision: 123547 URL: http://llvm.org/viewvc/llvm-project?rev=123547&view=rev Log: Reimplement CTPOP legalization with the "best" algorithm from http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel In a silly microbenchmark on a 65 nm core2 this is 1.5x faster than the old code in 32 bit mode and about 2x faster in 64 bit mode. It's also a lot shorter, especially when counting 64 bit population on a 32 bit target. I hope this is fast enough to replace Kernighan-style counting loops even when the input is rather sparse. Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp?rev=123547&r1=123546&r2=123547&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp (original) +++ llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp Sat Jan 15 14:30:30 2011 @@ -2388,6 +2388,17 @@ } } +/// SplatByte - Distribute ByteVal over NumBits bits. +static APInt SplatByte(unsigned NumBits, uint8_t ByteVal) { + APInt Val = APInt(NumBits, ByteVal); + unsigned Shift = 8; + for (unsigned i = NumBits; i > 8; i >>= 1) { + Val = (Val << Shift) | Val; + Shift <<= 1; + } + return Val; +} + /// ExpandBitCount - Expand the specified bitcount instruction into operations. /// SDValue SelectionDAGLegalize::ExpandBitCount(unsigned Opc, SDValue Op, @@ -2395,26 +2406,42 @@ switch (Opc) { default: assert(0 && "Cannot expand this yet!"); case ISD::CTPOP: { - static const uint64_t mask[6] = { - 0x5555555555555555ULL, 0x3333333333333333ULL, - 0x0F0F0F0F0F0F0F0FULL, 0x00FF00FF00FF00FFULL, - 0x0000FFFF0000FFFFULL, 0x00000000FFFFFFFFULL - }; EVT VT = Op.getValueType(); EVT ShVT = TLI.getShiftAmountTy(); - unsigned len = VT.getSizeInBits(); - for (unsigned i = 0; (1U << i) <= (len / 2); ++i) { - //x = (x & mask[i][len/8]) + (x >> (1 << i) & mask[i][len/8]) - unsigned EltSize = VT.isVector() ? - VT.getVectorElementType().getSizeInBits() : len; - SDValue Tmp2 = DAG.getConstant(APInt(EltSize, mask[i]), VT); - SDValue Tmp3 = DAG.getConstant(1ULL << i, ShVT); - Op = DAG.getNode(ISD::ADD, dl, VT, - DAG.getNode(ISD::AND, dl, VT, Op, Tmp2), - DAG.getNode(ISD::AND, dl, VT, - DAG.getNode(ISD::SRL, dl, VT, Op, Tmp3), - Tmp2)); - } + unsigned Len = VT.getSizeInBits(); + + // This is the "best" algorithm from + // http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel + + SDValue Mask55 = DAG.getConstant(SplatByte(Len, 0x55), VT); + SDValue Mask33 = DAG.getConstant(SplatByte(Len, 0x33), VT); + SDValue Mask0F = DAG.getConstant(SplatByte(Len, 0x0F), VT); + SDValue Mask01 = DAG.getConstant(SplatByte(Len, 0x01), VT); + + // v = v - ((v >> 1) & 0x55555555...) + Op = DAG.getNode(ISD::SUB, dl, VT, Op, + DAG.getNode(ISD::AND, dl, VT, + DAG.getNode(ISD::SRL, dl, VT, Op, + DAG.getConstant(1, ShVT)), + Mask55)); + // v = (v & 0x33333333...) + ((v >> 2) & 0x33333333...) + Op = DAG.getNode(ISD::ADD, dl, VT, + DAG.getNode(ISD::AND, dl, VT, Op, Mask33), + DAG.getNode(ISD::AND, dl, VT, + DAG.getNode(ISD::SRL, dl, VT, Op, + DAG.getConstant(2, ShVT)), + Mask33)); + // v = (v + (v >> 4)) & 0x0F0F0F0F... + Op = DAG.getNode(ISD::AND, dl, VT, + DAG.getNode(ISD::ADD, dl, VT, Op, + DAG.getNode(ISD::SRL, dl, VT, Op, + DAG.getConstant(4, ShVT))), + Mask0F); + // v = (v * 0x01010101...) >> (Len - 8) + Op = DAG.getNode(ISD::SRL, dl, VT, + DAG.getNode(ISD::MUL, dl, VT, Op, Mask01), + DAG.getConstant(Len - 8, ShVT)); + return Op; } case ISD::CTLZ: { From bigcheesegs at gmail.com Sat Jan 15 14:39:36 2011 From: bigcheesegs at gmail.com (Michael J. Spencer) Date: Sat, 15 Jan 2011 20:39:36 -0000 Subject: [llvm-commits] [llvm] r123548 - in /llvm/trunk: include/llvm/Support/FileSystem.h lib/Archive/Archive.cpp lib/Archive/ArchiveWriter.cpp lib/Support/Path.cpp lib/Support/PathV2.cpp Message-ID: <20110115203936.F2AD22A6C12C@llvm.org> Author: mspencer Date: Sat Jan 15 14:39:36 2011 New Revision: 123548 URL: http://llvm.org/viewvc/llvm-project?rev=123548&view=rev Log: Support/PathV2: Add identify_magic. Modified: llvm/trunk/include/llvm/Support/FileSystem.h llvm/trunk/lib/Archive/Archive.cpp llvm/trunk/lib/Archive/ArchiveWriter.cpp llvm/trunk/lib/Support/Path.cpp llvm/trunk/lib/Support/PathV2.cpp Modified: llvm/trunk/include/llvm/Support/FileSystem.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/FileSystem.h?rev=123548&r1=123547&r2=123548&view=diff ============================================================================== --- llvm/trunk/include/llvm/Support/FileSystem.h (original) +++ llvm/trunk/include/llvm/Support/FileSystem.h Sat Jan 15 14:39:36 2011 @@ -30,6 +30,7 @@ #include "llvm/ADT/SmallString.h" #include "llvm/ADT/Twine.h" #include "llvm/Support/DataTypes.h" +#include "llvm/Support/PathV1.h" #include "llvm/Support/system_error.h" #include #include @@ -463,6 +464,14 @@ error_code get_magic(const Twine &path, uint32_t len, SmallVectorImpl &result); +/// @brief Get and identify \a path's type based on its content. +/// +/// @param path Input path. +/// @param result Set to the type of file, or LLVMFileType::Unknown_FileType. +/// @results errc::success if result has been successfully set, otherwise a +/// platform specific error_code. +error_code identify_magic(const Twine &path, LLVMFileType &result); + /// @brief Is file bitcode? /// /// @param path Input path. Modified: llvm/trunk/lib/Archive/Archive.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Archive/Archive.cpp?rev=123548&r1=123547&r2=123548&view=diff ============================================================================== --- llvm/trunk/lib/Archive/Archive.cpp (original) +++ llvm/trunk/lib/Archive/Archive.cpp Sat Jan 15 14:39:36 2011 @@ -116,11 +116,10 @@ // Get the signature and status info const char* signature = (const char*) data; - std::string magic; + SmallString<4> magic; if (!signature) { - path.getMagicNumber(magic,4); + sys::fs::get_magic(path.str(), magic.capacity(), magic); signature = magic.c_str(); - std::string err; const sys::FileStatus *FSinfo = path.getFileStatus(false, ErrMsg); if (FSinfo) info = *FSinfo; Modified: llvm/trunk/lib/Archive/ArchiveWriter.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Archive/ArchiveWriter.cpp?rev=123548&r1=123547&r2=123548&view=diff ============================================================================== --- llvm/trunk/lib/Archive/ArchiveWriter.cpp (original) +++ llvm/trunk/lib/Archive/ArchiveWriter.cpp Sat Jan 15 14:39:36 2011 @@ -181,9 +181,11 @@ flags |= ArchiveMember::HasPathFlag; if (hasSlash || filePath.str().length() > 15) flags |= ArchiveMember::HasLongFilenameFlag; - std::string magic; - mbr->path.getMagicNumber(magic,4); - switch (sys::IdentifyFileType(magic.c_str(),4)) { + + sys::LLVMFileType type; + if (sys::fs::identify_magic(mbr->path.str(), type)) + type = sys::Unknown_FileType; + switch (type) { case sys::Bitcode_FileType: flags |= ArchiveMember::BitcodeFlag; break; Modified: llvm/trunk/lib/Support/Path.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Support/Path.cpp?rev=123548&r1=123547&r2=123548&view=diff ============================================================================== --- llvm/trunk/lib/Support/Path.cpp (original) +++ llvm/trunk/lib/Support/Path.cpp Sat Jan 15 14:39:36 2011 @@ -12,6 +12,7 @@ //===----------------------------------------------------------------------===// #include "llvm/Support/Path.h" +#include "llvm/Support/FileSystem.h" #include "llvm/Config/config.h" #include "llvm/Support/FileSystem.h" #include @@ -141,42 +142,33 @@ bool Path::isArchive() const { - std::string Magic; - if (getMagicNumber(Magic, 8)) - if (IdentifyFileType(Magic.c_str(), Magic.length()) == Archive_FileType) - return true; - return false; + LLVMFileType type; + if (fs::identify_magic(str(), type)) + return false; + return type == Archive_FileType; } bool Path::isDynamicLibrary() const { - std::string Magic; - if (getMagicNumber(Magic, 64)) - switch (IdentifyFileType(Magic.c_str(), - static_cast(Magic.length()))) { - default: return false; - case Mach_O_FixedVirtualMemorySharedLib_FileType: - case Mach_O_DynamicallyLinkedSharedLib_FileType: - case Mach_O_DynamicallyLinkedSharedLibStub_FileType: - case ELF_SharedObject_FileType: - case COFF_FileType: return true; - } - - return false; + LLVMFileType type; + if (fs::identify_magic(str(), type)) + return false; + switch (type) { + default: return false; + case Mach_O_FixedVirtualMemorySharedLib_FileType: + case Mach_O_DynamicallyLinkedSharedLib_FileType: + case Mach_O_DynamicallyLinkedSharedLibStub_FileType: + case ELF_SharedObject_FileType: + case COFF_FileType: return true; + } } bool Path::isObjectFile() const { - std::string Magic; - if (getMagicNumber(Magic, 64)) - if (IdentifyFileType(Magic.c_str(), - static_cast(Magic.length())) - != Unknown_FileType) { - // Everything in LLVMFileType is currently an object file. - return true; - } - - return false; + LLVMFileType type; + if (fs::identify_magic(str(), type) || type == Unknown_FileType) + return false; + return true; } Path @@ -210,13 +202,10 @@ bool Path::isBitcodeFile() const { - std::string actualMagic; - if (!getMagicNumber(actualMagic, 4)) + LLVMFileType type; + if (fs::identify_magic(str(), type)) return false; - LLVMFileType FT = - IdentifyFileType(actualMagic.c_str(), - static_cast(actualMagic.length())); - return FT == Bitcode_FileType; + return type == Bitcode_FileType; } bool Path::hasMagicNumber(StringRef Magic) const { Modified: llvm/trunk/lib/Support/PathV2.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Support/PathV2.cpp?rev=123548&r1=123547&r2=123548&view=diff ============================================================================== --- llvm/trunk/lib/Support/PathV2.cpp (original) +++ llvm/trunk/lib/Support/PathV2.cpp Sat Jan 15 14:39:36 2011 @@ -703,6 +703,16 @@ return success; } +error_code identify_magic(const Twine &path, LLVMFileType &result) { + SmallString<32> Magic; + error_code ec = get_magic(path, Magic.capacity(), Magic); + if (ec && ec != errc::value_too_large) + return ec; + + result = IdentifyFileType(Magic.data(), Magic.size()); + return success; +} + namespace { error_code remove_all_r(StringRef path, file_type ft, uint32_t &count) { if (ft == file_type::directory_file) { From fvbommel at gmail.com Sat Jan 15 14:51:32 2011 From: fvbommel at gmail.com (Frits van Bommel) Date: Sat, 15 Jan 2011 21:51:32 +0100 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp In-Reply-To: <20110115203031.19E282A6C12C@llvm.org> References: <20110115203031.19E282A6C12C@llvm.org> Message-ID: On Sat, Jan 15, 2011 at 9:30 PM, Benjamin Kramer wrote: > Author: d0k > Date: Sat Jan 15 14:30:30 2011 > New Revision: 123547 > > URL: http://llvm.org/viewvc/llvm-project?rev=123547&view=rev > Log: > Reimplement CTPOP legalization with the "best" algorithm from > http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel You seem to be using the "generalization of the best bit counting method to integers of bit-widths upto 128" from that site. Did you check whether this works for >128 bits too? It's probably not very relevant in practice (yet), but it should still produce correct code... From fvbommel at gmail.com Sat Jan 15 15:04:44 2011 From: fvbommel at gmail.com (Frits van Bommel) Date: Sat, 15 Jan 2011 22:04:44 +0100 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp In-Reply-To: References: <20110115203031.19E282A6C12C@llvm.org> Message-ID: On Sat, Jan 15, 2011 at 9:51 PM, Frits van Bommel wrote: > On Sat, Jan 15, 2011 at 9:30 PM, Benjamin Kramer > wrote: >> Reimplement CTPOP legalization with the "best" algorithm from >> http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel > > You seem to be using the "generalization of the best bit counting > method to integers of bit-widths upto 128" from that site. Did you > check whether this works for >128 bits too? Thinking about it, this *can't* be correct for >255 bits since the result is a one-byte value. The obvious way to fix it would be to generate this code for every 128-bit chunk of the input, and add the results. (or codegen a loop for especially large values) Maybe there's a better way though? From baldrick at free.fr Sat Jan 15 15:17:30 2011 From: baldrick at free.fr (Duncan Sands) Date: Sat, 15 Jan 2011 22:17:30 +0100 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp In-Reply-To: References: <20110115203031.19E282A6C12C@llvm.org> Message-ID: <4D320EEA.6050906@free.fr> Hi Frits, > On Sat, Jan 15, 2011 at 9:51 PM, Frits van Bommel wrote: >> On Sat, Jan 15, 2011 at 9:30 PM, Benjamin Kramer >> wrote: >>> Reimplement CTPOP legalization with the "best" algorithm from >>> http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel >> >> You seem to be using the "generalization of the best bit counting >> method to integers of bit-widths upto 128" from that site. Did you >> check whether this works for>128 bits too? > > Thinking about it, this *can't* be correct for>255 bits since the > result is a one-byte value. > The obvious way to fix it would be to generate this code for every > 128-bit chunk of the input, and add the results. (or codegen a loop > for especially large values) > Maybe there's a better way though? in LegalizeDAG all types are legal, so it is enough to future proof things by adding an assertion if there are more than 128 bits. Ciao, Duncan. From benny.kra at googlemail.com Sat Jan 15 15:20:40 2011 From: benny.kra at googlemail.com (Benjamin Kramer) Date: Sat, 15 Jan 2011 22:20:40 +0100 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp In-Reply-To: References: <20110115203031.19E282A6C12C@llvm.org> Message-ID: <89B03DA2-6952-4DB8-BDF8-1E584D1D0F75@gmail.com> On 15.01.2011, at 22:04, Frits van Bommel wrote: > On Sat, Jan 15, 2011 at 9:51 PM, Frits van Bommel wrote: >> On Sat, Jan 15, 2011 at 9:30 PM, Benjamin Kramer >> wrote: >>> Reimplement CTPOP legalization with the "best" algorithm from >>> http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel >> >> You seem to be using the "generalization of the best bit counting >> method to integers of bit-widths upto 128" from that site. Did you >> check whether this works for >128 bits too? > > Thinking about it, this *can't* be correct for >255 bits since the > result is a one-byte value. > The obvious way to fix it would be to generate this code for every > 128-bit chunk of the input, and add the results. (or codegen a loop > for especially large values) > Maybe there's a better way though? CodeGen will split up large popcounts. I'll add an assert though so we don't have silent miscompiles when a target with native 256 bit ints gets implemented in LLVM. From benny.kra at googlemail.com Sat Jan 15 15:19:37 2011 From: benny.kra at googlemail.com (Benjamin Kramer) Date: Sat, 15 Jan 2011 21:19:37 -0000 Subject: [llvm-commits] [llvm] r123549 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp Message-ID: <20110115211937.8E9312A6C12C@llvm.org> Author: d0k Date: Sat Jan 15 15:19:37 2011 New Revision: 123549 URL: http://llvm.org/viewvc/llvm-project?rev=123549&view=rev Log: Add an assert so we don't silently miscompile ctpop for bit widths > 128. Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp?rev=123549&r1=123548&r2=123549&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp (original) +++ llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp Sat Jan 15 15:19:37 2011 @@ -2389,6 +2389,7 @@ } /// SplatByte - Distribute ByteVal over NumBits bits. +// FIXME: Move this helper to a common place. static APInt SplatByte(unsigned NumBits, uint8_t ByteVal) { APInt Val = APInt(NumBits, ByteVal); unsigned Shift = 8; @@ -2410,6 +2411,9 @@ EVT ShVT = TLI.getShiftAmountTy(); unsigned Len = VT.getSizeInBits(); + assert(VT.isInteger() && Len <= 128 && Len % 8 == 0 && + "CTPOP not implemented for this type."); + // This is the "best" algorithm from // http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel From bigcheesegs at gmail.com Sat Jan 15 15:43:25 2011 From: bigcheesegs at gmail.com (Michael J. Spencer) Date: Sat, 15 Jan 2011 21:43:25 -0000 Subject: [llvm-commits] [llvm] r123550 - /llvm/trunk/include/llvm/Support/GraphWriter.h Message-ID: <20110115214325.8D5C12A6C12C@llvm.org> Author: mspencer Date: Sat Jan 15 15:43:25 2011 New Revision: 123550 URL: http://llvm.org/viewvc/llvm-project?rev=123550&view=rev Log: Support/GraphWriter: Replace all internal uses of PathV1 with PathV2. The external API still uses PathV1. Modified: llvm/trunk/include/llvm/Support/GraphWriter.h Modified: llvm/trunk/include/llvm/Support/GraphWriter.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Support/GraphWriter.h?rev=123550&r1=123549&r2=123550&view=diff ============================================================================== --- llvm/trunk/include/llvm/Support/GraphWriter.h (original) +++ llvm/trunk/include/llvm/Support/GraphWriter.h Sat Jan 15 15:43:25 2011 @@ -24,6 +24,7 @@ #define LLVM_SUPPORT_GRAPHWRITER_H #include "llvm/Support/DOTGraphTraits.h" +#include "llvm/Support/FileSystem.h" #include "llvm/Support/raw_ostream.h" #include "llvm/ADT/GraphTraits.h" #include "llvm/Support/Path.h" @@ -309,32 +310,21 @@ template sys::Path WriteGraph(const GraphType &G, const std::string &Name, bool ShortNames = false, const std::string &Title = "") { - std::string ErrMsg; - sys::Path Filename = sys::Path::GetTemporaryDirectory(&ErrMsg); - if (Filename.isEmpty()) { - errs() << "Error: " << ErrMsg << "\n"; - return Filename; - } - Filename.appendComponent(Name + ".dot"); - if (Filename.makeUnique(true,&ErrMsg)) { - errs() << "Error: " << ErrMsg << "\n"; + SmallString<128> FilePath; + + int FileFD; + if (error_code ec = sys::fs::unique_file("graph-" + Name + "-%%-%%-%%-%%.dot", + FileFD, FilePath)) { + errs() << "Error creating output file: " << ec.message() << '\n'; return sys::Path(); } - errs() << "Writing '" << Filename.str() << "'... "; - - std::string ErrorInfo; - raw_fd_ostream O(Filename.c_str(), ErrorInfo); - - if (ErrorInfo.empty()) { - llvm::WriteGraph(O, G, ShortNames, Title); - errs() << " done. \n"; - } else { - errs() << "error opening file '" << Filename.str() << "' for writing!\n"; - Filename.clear(); - } + errs() << "Writing '" << FilePath << "'... "; + raw_fd_ostream O(FileFD, true); + llvm::WriteGraph(O, G, ShortNames, Title); + errs() << " done. \n"; - return Filename; + return sys::Path(FilePath.str()); } /// ViewGraph - Emit a dot graph, run 'dot', run gv on the postscript file, From bigcheesegs at gmail.com Sat Jan 15 15:43:37 2011 From: bigcheesegs at gmail.com (Michael J. Spencer) Date: Sat, 15 Jan 2011 21:43:37 -0000 Subject: [llvm-commits] [llvm] r123551 - in /llvm/trunk: include/llvm/Bitcode/Archive.h lib/Archive/ArchiveWriter.cpp Message-ID: <20110115214337.5EB3E2A6C12C@llvm.org> Author: mspencer Date: Sat Jan 15 15:43:37 2011 New Revision: 123551 URL: http://llvm.org/viewvc/llvm-project?rev=123551&view=rev Log: Archive: Replace all internal uses of PathV1 with PathV2. The external API still uses PathV1. Modified: llvm/trunk/include/llvm/Bitcode/Archive.h llvm/trunk/lib/Archive/ArchiveWriter.cpp Modified: llvm/trunk/include/llvm/Bitcode/Archive.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Bitcode/Archive.h?rev=123551&r1=123550&r2=123551&view=diff ============================================================================== --- llvm/trunk/include/llvm/Bitcode/Archive.h (original) +++ llvm/trunk/include/llvm/Bitcode/Archive.h Sat Jan 15 15:43:37 2011 @@ -25,6 +25,7 @@ namespace llvm { class MemoryBuffer; + class raw_ostream; // Forward declare classes class Module; // From VMCore @@ -482,7 +483,7 @@ bool loadSymbolTable(std::string* ErrMessage); /// @brief Write the symbol table to an ofstream. - void writeSymbolTable(std::ofstream& ARFile); + void writeSymbolTable(raw_ostream& ARFile); /// Writes one ArchiveMember to an ofstream. If an error occurs, returns /// false, otherwise true. If an error occurs and error is non-null then @@ -491,7 +492,7 @@ /// @returns true Writing member failed, \p error set to error message bool writeMember( const ArchiveMember& member, ///< The member to be written - std::ofstream& ARFile, ///< The file to write member onto + raw_ostream& ARFile, ///< The file to write member onto bool CreateSymbolTable, ///< Should symbol table be created? bool TruncateNames, ///< Should names be truncated to 11 chars? bool ShouldCompress, ///< Should the member be compressed? Modified: llvm/trunk/lib/Archive/ArchiveWriter.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Archive/ArchiveWriter.cpp?rev=123551&r1=123550&r2=123551&view=diff ============================================================================== --- llvm/trunk/lib/Archive/ArchiveWriter.cpp (original) +++ llvm/trunk/lib/Archive/ArchiveWriter.cpp Sat Jan 15 15:43:37 2011 @@ -18,6 +18,7 @@ #include "llvm/Support/FileSystem.h" #include "llvm/Support/MemoryBuffer.h" #include "llvm/Support/Process.h" +#include "llvm/Support/raw_ostream.h" #include "llvm/Support/Signals.h" #include "llvm/Support/system_error.h" #include @@ -27,7 +28,7 @@ // Write an integer using variable bit rate encoding. This saves a few bytes // per entry in the symbol table. -static inline void writeInteger(unsigned num, std::ofstream& ARFile) { +static inline void writeInteger(unsigned num, raw_ostream& ARFile) { while (1) { if (num < 0x80) { // done? ARFile << (unsigned char)num; @@ -201,14 +202,14 @@ bool Archive::writeMember( const ArchiveMember& member, - std::ofstream& ARFile, + raw_ostream& ARFile, bool CreateSymbolTable, bool TruncateNames, bool ShouldCompress, std::string* ErrMsg ) { - unsigned filepos = ARFile.tellp(); + unsigned filepos = ARFile.tell(); filepos -= 8; // Get the data and its size either from the @@ -281,7 +282,7 @@ ARFile.write(data,fSize); // Make sure the member is an even length - if ((ARFile.tellp() & 1) == 1) + if ((ARFile.tell() & 1) == 1) ARFile << ARFILE_PAD; // Close the mapped file if it was opened @@ -291,7 +292,7 @@ // Write out the LLVM symbol table as an archive member to the file. void -Archive::writeSymbolTable(std::ofstream& ARFile) { +Archive::writeSymbolTable(raw_ostream& ARFile) { // Construct the symbol table's header ArchiveMemberHeader Hdr; @@ -315,7 +316,7 @@ #ifndef NDEBUG // Save the starting position of the symbol tables data content. - unsigned startpos = ARFile.tellp(); + unsigned startpos = ARFile.tell(); #endif // Write out the symbols sequentially @@ -332,7 +333,7 @@ #ifndef NDEBUG // Now that we're done with the symbol table, get the ending file position - unsigned endpos = ARFile.tellp(); + unsigned endpos = ARFile.tell(); #endif // Make sure that the amount we wrote is what we pre-computed. This is @@ -361,25 +362,19 @@ } // Create a temporary file to store the archive in - sys::Path TmpArchive = archPath; - if (TmpArchive.createTemporaryFileOnDisk(ErrMsg)) + SmallString<128> TempArchivePath; + int ArchFD; + if (error_code ec = sys::fs::unique_file("%%-%%-%%-%%" + archPath.str(), + ArchFD, TempArchivePath)) { + if (ErrMsg) *ErrMsg = ec.message(); return true; + } // Make sure the temporary gets removed if we crash - sys::RemoveFileOnSignal(TmpArchive); + sys::RemoveFileOnSignal(sys::Path(TempArchivePath.str())); // Create archive file for output. - std::ios::openmode io_mode = std::ios::out | std::ios::trunc | - std::ios::binary; - std::ofstream ArchiveFile(TmpArchive.c_str(), io_mode); - - // Check for errors opening or creating archive file. - if (!ArchiveFile.is_open() || ArchiveFile.bad()) { - TmpArchive.eraseFromDisk(); - if (ErrMsg) - *ErrMsg = "Error opening archive file: " + archPath.str(); - return true; - } + raw_fd_ostream ArchiveFile(ArchFD, true); // If we're creating a symbol table, reset it now if (CreateSymbolTable) { @@ -395,8 +390,9 @@ for (MembersList::iterator I = begin(), E = end(); I != E; ++I) { if (writeMember(*I, ArchiveFile, CreateSymbolTable, TruncateNames, Compress, ErrMsg)) { - TmpArchive.eraseFromDisk(); ArchiveFile.close(); + bool existed; + sys::fs::remove(TempArchivePath.str(), existed); return true; } } @@ -411,12 +407,12 @@ // ensure compatibility with other archivers we need to put the symbol // table first in the file. Unfortunately, this means mapping the file // we just wrote back in and copying it to the destination file. - sys::Path FinalFilePath = archPath; + SmallString<128> TempArchiveWithSymbolTablePath; // Map in the archive we just wrote. { OwningPtr arch; - if (error_code ec = MemoryBuffer::getFile(TmpArchive.c_str(), arch)) { + if (error_code ec = MemoryBuffer::getFile(TempArchivePath.c_str(), arch)) { if (ErrMsg) *ErrMsg = ec.message(); return true; @@ -425,17 +421,14 @@ // Open another temporary file in order to avoid invalidating the // mmapped data - if (FinalFilePath.createTemporaryFileOnDisk(ErrMsg)) - return true; - sys::RemoveFileOnSignal(FinalFilePath); - - std::ofstream FinalFile(FinalFilePath.c_str(), io_mode); - if (!FinalFile.is_open() || FinalFile.bad()) { - TmpArchive.eraseFromDisk(); - if (ErrMsg) - *ErrMsg = "Error opening archive file: " + FinalFilePath.str(); + if (error_code ec = sys::fs::unique_file("%%-%%-%%-%%" + archPath.str(), + ArchFD, TempArchiveWithSymbolTablePath)) { + if (ErrMsg) *ErrMsg = ec.message(); return true; } + sys::RemoveFileOnSignal(sys::Path(TempArchiveWithSymbolTablePath.str())); + + raw_fd_ostream FinalFile(ArchFD, true); // Write the file magic number FinalFile << ARFILE_MAGIC; @@ -448,7 +441,8 @@ if (foreignST) { if (writeMember(*foreignST, FinalFile, false, false, false, ErrMsg)) { FinalFile.close(); - TmpArchive.eraseFromDisk(); + bool existed; + sys::fs::remove(TempArchiveWithSymbolTablePath.str(), existed); return true; } } @@ -466,8 +460,11 @@ } // free arch. // Move the final file over top of TmpArchive - if (FinalFilePath.renamePathOnDisk(TmpArchive, ErrMsg)) + if (error_code ec = sys::fs::rename(TempArchiveWithSymbolTablePath.str(), + TempArchivePath.str())) { + if (ErrMsg) *ErrMsg = ec.message(); return true; + } } // Before we replace the actual archive, we need to forget all the @@ -475,8 +472,11 @@ // this because we cannot replace an open file on Windows. cleanUpMemory(); - if (TmpArchive.renamePathOnDisk(archPath, ErrMsg)) + if (error_code ec = sys::fs::rename(TempArchivePath.str(), + archPath.str())) { + if (ErrMsg) *ErrMsg = ec.message(); return true; + } // Set correct read and write permissions after temporary file is moved // to final destination path. From bigcheesegs at gmail.com Sat Jan 15 15:43:46 2011 From: bigcheesegs at gmail.com (Michael J. Spencer) Date: Sat, 15 Jan 2011 21:43:46 -0000 Subject: [llvm-commits] [llvm] r123552 - /llvm/trunk/include/llvm/Bitcode/Archive.h Message-ID: <20110115214346.24E6C2A6C12C@llvm.org> Author: mspencer Date: Sat Jan 15 15:43:45 2011 New Revision: 123552 URL: http://llvm.org/viewvc/llvm-project?rev=123552&view=rev Log: Archive: Fix spelling. Modified: llvm/trunk/include/llvm/Bitcode/Archive.h Modified: llvm/trunk/include/llvm/Bitcode/Archive.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Bitcode/Archive.h?rev=123552&r1=123551&r2=123552&view=diff ============================================================================== --- llvm/trunk/include/llvm/Bitcode/Archive.h (original) +++ llvm/trunk/include/llvm/Bitcode/Archive.h Sat Jan 15 15:43:45 2011 @@ -83,7 +83,7 @@ unsigned getGroup() const { return info.getGroup(); } /// The "mode" specifies the access permissions for the file per Unix - /// security. This may not have any applicabiity on non-Unix systems but is + /// security. This may not have any applicability on non-Unix systems but is /// a required component of the "ar" file format. /// @brief Get the permission mode associated with this archive member. unsigned getMode() const { return info.getMode(); } @@ -145,7 +145,7 @@ /// allowed that doesn't have this restriction. This method determines if /// that "long format" is used for this member. /// @returns true iff the file name uses the long form - /// @brief Determin if the member has a long file name + /// @brief Determine if the member has a long file name bool hasLongFilename() const { return flags&HasLongFilenameFlag; } /// This method returns the status info (like Unix stat(2)) for the archive @@ -403,7 +403,7 @@ /// bitcode archive. It first makes sure the symbol table has been loaded /// and has a non-zero size. If it does, then it is an archive. If not, /// then it tries to load all the bitcode modules of the archive. Finally, - /// it returns whether it was successfull. + /// it returns whether it was successful. /// @returns true if the archive is a proper llvm bitcode archive /// @brief Determine whether the archive is a proper llvm bitcode archive. bool isBitcodeArchive(); From sabre at nondot.org Sat Jan 15 18:27:10 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 00:27:10 -0000 Subject: [llvm-commits] [llvm] r123554 - in /llvm/trunk: include/llvm/Analysis/CodeMetrics.h include/llvm/InitializePasses.h include/llvm/LinkAllPasses.h include/llvm/Transforms/IPO.h lib/Transforms/IPO/CMakeLists.txt lib/Transforms/IPO/IPO.cpp lib/Transforms/IPO/PartialSpecialization.cpp test/Transforms/PartialSpecialize/ Message-ID: <20110116002710.A86632A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 18:27:10 2011 New Revision: 123554 URL: http://llvm.org/viewvc/llvm-project?rev=123554&view=rev Log: remove the partial specialization pass. It is unmaintained and has bugs. Removed: llvm/trunk/lib/Transforms/IPO/PartialSpecialization.cpp llvm/trunk/test/Transforms/PartialSpecialize/ Modified: llvm/trunk/include/llvm/Analysis/CodeMetrics.h llvm/trunk/include/llvm/InitializePasses.h llvm/trunk/include/llvm/LinkAllPasses.h llvm/trunk/include/llvm/Transforms/IPO.h llvm/trunk/lib/Transforms/IPO/CMakeLists.txt llvm/trunk/lib/Transforms/IPO/IPO.cpp Modified: llvm/trunk/include/llvm/Analysis/CodeMetrics.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Analysis/CodeMetrics.h?rev=123554&r1=123553&r2=123554&view=diff ============================================================================== --- llvm/trunk/include/llvm/Analysis/CodeMetrics.h (original) +++ llvm/trunk/include/llvm/Analysis/CodeMetrics.h Sat Jan 15 18:27:10 2011 @@ -7,9 +7,8 @@ // //===----------------------------------------------------------------------===// // -// This file implements various weight measurements for a function, helping -// the Inliner and PartialSpecialization decide whether to duplicate its -// contents. +// This file implements various weight measurements for code, helping +// the Inliner and other passes decide whether to duplicate its contents. // //===----------------------------------------------------------------------===// Modified: llvm/trunk/include/llvm/InitializePasses.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/InitializePasses.h?rev=123554&r1=123553&r2=123554&view=diff ============================================================================== --- llvm/trunk/include/llvm/InitializePasses.h (original) +++ llvm/trunk/include/llvm/InitializePasses.h Sat Jan 15 18:27:10 2011 @@ -161,7 +161,6 @@ void initializeOptimizePHIsPass(PassRegistry&); void initializePEIPass(PassRegistry&); void initializePHIEliminationPass(PassRegistry&); -void initializePartSpecPass(PassRegistry&); void initializePartialInlinerPass(PassRegistry&); void initializePeepholeOptimizerPass(PassRegistry&); void initializePostDomOnlyPrinterPass(PassRegistry&); Modified: llvm/trunk/include/llvm/LinkAllPasses.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/LinkAllPasses.h?rev=123554&r1=123553&r2=123554&view=diff ============================================================================== --- llvm/trunk/include/llvm/LinkAllPasses.h (original) +++ llvm/trunk/include/llvm/LinkAllPasses.h Sat Jan 15 18:27:10 2011 @@ -135,7 +135,6 @@ (void) llvm::createPostDomTree(); (void) llvm::createPostDomFrontier(); (void) llvm::createInstructionNamerPass(); - (void) llvm::createPartialSpecializationPass(); (void) llvm::createFunctionAttrsPass(); (void) llvm::createMergeFunctionsPass(); (void) llvm::createPrintModulePass(0); Modified: llvm/trunk/include/llvm/Transforms/IPO.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Transforms/IPO.h?rev=123554&r1=123553&r2=123554&view=diff ============================================================================== --- llvm/trunk/include/llvm/Transforms/IPO.h (original) +++ llvm/trunk/include/llvm/Transforms/IPO.h Sat Jan 15 18:27:10 2011 @@ -189,12 +189,6 @@ ModulePass *createStripDeadPrototypesPass(); //===----------------------------------------------------------------------===// -/// createPartialSpecializationPass - This pass specializes functions for -/// constant arguments. -/// -ModulePass *createPartialSpecializationPass(); - -//===----------------------------------------------------------------------===// /// createFunctionAttrsPass - This pass discovers functions that do not access /// memory, or only read memory, and gives them the readnone/readonly attribute. /// It also discovers function arguments that are not captured by the function Modified: llvm/trunk/lib/Transforms/IPO/CMakeLists.txt URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/CMakeLists.txt?rev=123554&r1=123553&r2=123554&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/CMakeLists.txt (original) +++ llvm/trunk/lib/Transforms/IPO/CMakeLists.txt Sat Jan 15 18:27:10 2011 @@ -17,7 +17,6 @@ LowerSetJmp.cpp MergeFunctions.cpp PartialInlining.cpp - PartialSpecialization.cpp PruneEH.cpp StripDeadPrototypes.cpp StripSymbols.cpp Modified: llvm/trunk/lib/Transforms/IPO/IPO.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/IPO.cpp?rev=123554&r1=123553&r2=123554&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/IPO.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/IPO.cpp Sat Jan 15 18:27:10 2011 @@ -39,7 +39,6 @@ initializeLowerSetJmpPass(Registry); initializeMergeFunctionsPass(Registry); initializePartialInlinerPass(Registry); - initializePartSpecPass(Registry); initializePruneEHPass(Registry); initializeStripDeadPrototypesPassPass(Registry); initializeStripSymbolsPass(Registry); Removed: llvm/trunk/lib/Transforms/IPO/PartialSpecialization.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/PartialSpecialization.cpp?rev=123553&view=auto ============================================================================== --- llvm/trunk/lib/Transforms/IPO/PartialSpecialization.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/PartialSpecialization.cpp (removed) @@ -1,228 +0,0 @@ -//===-- PartialSpecialization.cpp - Specialize for common constants--------===// -// -// The LLVM Compiler Infrastructure -// -// This file is distributed under the University of Illinois Open Source -// License. See LICENSE.TXT for details. -// -//===----------------------------------------------------------------------===// -// -// This pass finds function arguments that are often a common constant and -// specializes a version of the called function for that constant. -// -// This pass simply does the cloning for functions it specializes. It depends -// on IPSCCP and DAE to clean up the results. -// -// The initial heuristic favors constant arguments that are used in control -// flow. -// -//===----------------------------------------------------------------------===// - -#define DEBUG_TYPE "partialspecialization" -#include "llvm/Transforms/IPO.h" -#include "llvm/Constant.h" -#include "llvm/Instructions.h" -#include "llvm/Module.h" -#include "llvm/Pass.h" -#include "llvm/ADT/Statistic.h" -#include "llvm/Analysis/InlineCost.h" -#include "llvm/Transforms/Utils/Cloning.h" -#include "llvm/Support/CallSite.h" -#include "llvm/ADT/DenseSet.h" -#include -using namespace llvm; - -STATISTIC(numSpecialized, "Number of specialized functions created"); -STATISTIC(numReplaced, "Number of callers replaced by specialization"); - -// Maximum number of arguments markable interested -static const int MaxInterests = 6; - -namespace { - typedef SmallVector InterestingArgVector; - class PartSpec : public ModulePass { - void scanForInterest(Function&, InterestingArgVector&); - int scanDistribution(Function&, int, std::map&); - InlineCostAnalyzer CA; - public : - static char ID; // Pass identification, replacement for typeid - PartSpec() : ModulePass(ID) { - initializePartSpecPass(*PassRegistry::getPassRegistry()); - } - bool runOnModule(Module &M); - }; -} - -char PartSpec::ID = 0; -INITIALIZE_PASS(PartSpec, "partialspecialization", - "Partial Specialization", false, false) - -// Specialize F by replacing the arguments (keys) in replacements with the -// constants (values). Replace all calls to F with those constants with -// a call to the specialized function. Returns the specialized function -static Function* -SpecializeFunction(Function* F, - ValueToValueMapTy& replacements) { - // arg numbers of deleted arguments - DenseMap deleted; - for (ValueToValueMapTy::iterator - repb = replacements.begin(), repe = replacements.end(); - repb != repe; ++repb) { - Argument const *arg = cast(repb->first); - deleted[arg->getArgNo()] = arg; - } - - Function* NF = CloneFunction(F, replacements, - /*ModuleLevelChanges=*/false); - NF->setLinkage(GlobalValue::InternalLinkage); - F->getParent()->getFunctionList().push_back(NF); - - // FIXME: Specialized versions getting the same constants should also get - // the same name. That way, specializations for public functions can be - // marked linkonce_odr and reused across modules. - - for (Value::use_iterator ii = F->use_begin(), ee = F->use_end(); - ii != ee; ) { - Value::use_iterator i = ii; - ++ii; - User *U = *i; - CallSite CS(U); - if (CS) { - if (CS.getCalledFunction() == F) { - SmallVector args; - // Assemble the non-specialized arguments for the updated callsite. - // In the process, make sure that the specialized arguments are - // constant and match the specialization. If that's not the case, - // this callsite needs to call the original or some other - // specialization; don't change it here. - CallSite::arg_iterator as = CS.arg_begin(), ae = CS.arg_end(); - for (CallSite::arg_iterator ai = as; ai != ae; ++ai) { - DenseMap::iterator delit = deleted.find( - std::distance(as, ai)); - if (delit == deleted.end()) - args.push_back(cast(ai)); - else { - Constant *ci = dyn_cast(ai); - if (!(ci && ci == replacements[delit->second])) - goto next_use; - } - } - Value* NCall; - if (CallInst *CI = dyn_cast(U)) { - NCall = CallInst::Create(NF, args.begin(), args.end(), - CI->getName(), CI); - cast(NCall)->setTailCall(CI->isTailCall()); - cast(NCall)->setCallingConv(CI->getCallingConv()); - } else { - InvokeInst *II = cast(U); - NCall = InvokeInst::Create(NF, II->getNormalDest(), - II->getUnwindDest(), - args.begin(), args.end(), - II->getName(), II); - cast(NCall)->setCallingConv(II->getCallingConv()); - } - CS.getInstruction()->replaceAllUsesWith(NCall); - CS.getInstruction()->eraseFromParent(); - ++numReplaced; - } - } - next_use:; - } - return NF; -} - - -bool PartSpec::runOnModule(Module &M) { - bool Changed = false; - for (Module::iterator I = M.begin(); I != M.end(); ++I) { - Function &F = *I; - if (F.isDeclaration() || F.mayBeOverridden()) continue; - InterestingArgVector interestingArgs; - scanForInterest(F, interestingArgs); - - // Find the first interesting Argument that we can specialize on - // If there are multiple interesting Arguments, then those will be found - // when processing the cloned function. - bool breakOuter = false; - for (unsigned int x = 0; !breakOuter && x < interestingArgs.size(); ++x) { - std::map distribution; - scanDistribution(F, interestingArgs[x], distribution); - for (std::map::iterator ii = distribution.begin(), - ee = distribution.end(); ii != ee; ++ii) { - // The distribution map might have an entry for NULL (i.e., one or more - // callsites were passing a non-constant there). We allow that to - // happen so that we can see whether any callsites pass a non-constant; - // if none do and the function is internal, we might have an opportunity - // to kill the original function. - if (!ii->first) continue; - int bonus = ii->second; - SmallVector argnos; - argnos.push_back(interestingArgs[x]); - InlineCost cost = CA.getSpecializationCost(&F, argnos); - // FIXME: If this is the last constant entry, and no non-constant - // entries exist, and the target function is internal, the cost should - // be reduced by the original size of the target function, almost - // certainly making it negative and causing a specialization that will - // leave the original function dead and removable. - if (cost.isAlways() || - (cost.isVariable() && cost.getValue() < bonus)) { - ValueToValueMapTy m; - Function::arg_iterator arg = F.arg_begin(); - for (int y = 0; y < interestingArgs[x]; ++y) - ++arg; - m[&*arg] = ii->first; - SpecializeFunction(&F, m); - ++numSpecialized; - breakOuter = true; - Changed = true; - } - } - } - } - return Changed; -} - -/// scanForInterest - This function decides which arguments would be worth -/// specializing on. -void PartSpec::scanForInterest(Function& F, InterestingArgVector& args) { - for(Function::arg_iterator ii = F.arg_begin(), ee = F.arg_end(); - ii != ee; ++ii) { - int argno = std::distance(F.arg_begin(), ii); - SmallVector argnos; - argnos.push_back(argno); - int bonus = CA.getSpecializationBonus(&F, argnos); - if (bonus > 0) { - args.push_back(argno); - } - } -} - -/// scanDistribution - Construct a histogram of constants for arg of F at arg. -/// For each distinct constant, we'll compute the total of the specialization -/// bonus across all callsites passing that constant; if that total exceeds -/// the specialization cost, we will create the specialization. -int PartSpec::scanDistribution(Function& F, int arg, - std::map& dist) { - bool hasIndirect = false; - int total = 0; - for (Value::use_iterator ii = F.use_begin(), ee = F.use_end(); - ii != ee; ++ii) { - User *U = *ii; - CallSite CS(U); - if (CS && CS.getCalledFunction() == &F) { - SmallVector argnos; - argnos.push_back(arg); - dist[dyn_cast(CS.getArgument(arg))] += - CA.getSpecializationBonus(&F, argnos); - ++total; - } else - hasIndirect = true; - } - - // Preserve the original address taken function even if all other uses - // will be specialized. - if (hasIndirect) ++total; - return total; -} - -ModulePass* llvm::createPartialSpecializationPass() { return new PartSpec(); } From clattner at apple.com Sat Jan 15 19:02:34 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 17:02:34 -0800 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp In-Reply-To: <20110115203031.19E282A6C12C@llvm.org> References: <20110115203031.19E282A6C12C@llvm.org> Message-ID: On Jan 15, 2011, at 12:30 PM, Benjamin Kramer wrote: > Author: d0k > Date: Sat Jan 15 14:30:30 2011 > New Revision: 123547 > > URL: http://llvm.org/viewvc/llvm-project?rev=123547&view=rev > Log: > Reimplement CTPOP legalization with the "best" algorithm from > http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel > > In a silly microbenchmark on a 65 nm core2 this is 1.5x faster than the old > code in 32 bit mode and about 2x faster in 64 bit mode. It's also a lot shorter, > especially when counting 64 bit population on a 32 bit target. > > I hope this is fast enough to replace Kernighan-style counting loops even when > the input is rather sparse. Wow, very nice! This is quite a bit faster than the old one. When I replace the manual ctpop loops in crafty with builtin_popcount, the performance still slows down from 5.5s to 5.62s though. This is very close to making me want to turn it on, particularly given that this is almost certainly a win for all targets with popcount and 32-bit platforms without them. One major advantage of recognizing popcount though is that the optimizer has a chance of hacking on it, and there are a lot of instcombine xforms that we could do. Here are some that I note looking at the bc file for crafy (when hacked to use builtin_popcount). I attached the bc file below if you're interested. One interesting thing that I see is that the calls are often of the form: icmp_ugt(ctpop(a), ctpop(b)) I wonder if there is some clever optimization for that case. Another case I see in a few places is: %331 = tail call i64 @llvm.ctpop.i64(i64 %330) nounwind %cast.i137 = trunc i64 %331 to i32 %332 = icmp ugt i32 %cast.i137, 1 Where both the trunc and the ctpop have one use. There are a few other interesting patterns that seem simplifiable: ; :178 ; preds = %176 ... %183 = tail call i64 @llvm.ctpop.i64(i64 %182) nounwind %cast.i107 = trunc i64 %183 to i32 %184 = getelementptr inbounds [64 x i64]* @w_pawn_attacks, i64 0, i64 %179 %185 = load i64* %184, align 8, !tbaa !2 %186 = and i64 %68, %185 %187 = tail call i64 @llvm.ctpop.i64(i64 %186) nounwind %cast.i108 = trunc i64 %187 to i32 %188 = getelementptr inbounds [65 x i64]* @set_mask, i64 0, i64 %179 %189 = load i64* %188, align 8, !tbaa !2 %190 = and i64 %120, %189 %191 = icmp eq i64 %190, 0 br i1 %191, label %192, label %.thread207 ; :192 ; preds = %178 %193 = icmp ugt i32 %cast.i108, %cast.i107 br i1 %193, label %197, label %194 ; :194 ; preds = %192 %195 = icmp eq i32 %cast.i107, 0 %196 = icmp ult i32 %cast.i107, %cast.i108 %or.cond246 = or i1 %195, %196 %indvar.next433 = add i32 %indvar432, 1 br i1 %or.cond246, label %176, label %.thread218 ; :197 ; preds = %192 %198 = sub nsw i32 %cast.i108, %cast.i107 %199 = icmp eq i32 %cast.i108, %cast.i107 br i1 %199, label %.thread218, label %.thread207 "%195" seems like it is just "icmp ne i64 %182, 0" %196 seems like it is the same thing as %193. I wonder if we should always canonicalizing icmps to "lt" comparisons when both operands are non-constant. It seems that this would expose more CSEs. Here another case that might allow cleverness: %437 = tail call i64 @llvm.ctpop.i64(i64 %436) nounwind %cast.i138 = trunc i64 %437 to i32 ... %440 = tail call i64 @llvm.ctpop.i64(i64 %439) nounwind %cast.i139 = trunc i64 %440 to i32 %441 = sub nsw i32 %cast.i139, %cast.i138 %442 = icmp eq i32 %441, 2 Here's another obvious case: %592 = tail call i64 @llvm.ctpop.i64(i64 %591) nounwind %cast.i176 = trunc i64 %592 to i32 %593 = icmp eq i32 %cast.i176, 0 In this case, 592 has multiple uses. It seems that we should be able to eliminate the trunc though since we know the top bits are zero. Here's another interesting pattern: %778 = tail call i64 @llvm.ctpop.i64(i64 %777) nounwind %cast.i195 = trunc i64 %778 to i32 ... %781 = tail call i64 @llvm.ctpop.i64(i64 %780) nounwind %cast.i196 = trunc i64 %781 to i32 %782 = sub nsw i32 %cast.i195, %cast.i196 %783 = icmp eq i32 %782, 2 br i1 %783, label %.loopexit.thread, label %784 ; :784 ; preds = %775 switch i32 %cast.i195, label %.thread245 [ i32 1, label %785 i32 0, label %786 ] ; :785 ; preds = %784 %.not9 = icmp ne i32 %cast.i196, 0 ... And: %1037 = tail call i64 @llvm.ctpop.i64(i64 %1036) nounwind %1038 = icmp eq i64 %1037, 3 These "popcount = 3" and "popcount < 2" sorts of cases seems that they could use a couple iterations of the unrolled "a &= a-1" checks or something, instead of computing the full computation. For example, the top of GenerateCheckEvasions has "popcount(x) == 1" which seems that it could be something like "x != 0 && (x & (x-1) == 0)" cheaper than expanding the popcount. This sort of thing is a bad idea of ctpop expands to a single cycle instruction though, so this is probably best to do in dag combine instead of instcombine. Anyway, if you're interested in poking at it, here's the hacked bc file: -------------- next part -------------- A non-text attachment was scrubbed... Name: 186.crafty.llvm.bc Type: application/octet-stream Size: 401552 bytes Desc: not available Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/dc9747b7/attachment-0001.obj -------------- next part -------------- -Chris From sabre at nondot.org Sat Jan 15 19:04:49 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 01:04:49 -0000 Subject: [llvm-commits] [llvm] r123555 - /llvm/trunk/test/Linker/link-archive.ll Message-ID: <20110116010449.DA3352A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 19:04:49 2011 New Revision: 123555 URL: http://llvm.org/viewvc/llvm-project?rev=123555&view=rev Log: one of michael's recent patches broke this, temporarily disable it so the bots go green Modified: llvm/trunk/test/Linker/link-archive.ll Modified: llvm/trunk/test/Linker/link-archive.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Linker/link-archive.ll?rev=123555&r1=123554&r2=123555&view=diff ============================================================================== --- llvm/trunk/test/Linker/link-archive.ll (original) +++ llvm/trunk/test/Linker/link-archive.ll Sat Jan 15 19:04:49 2011 @@ -1,4 +1,5 @@ ; Test linking of a bc file to an archive via llvm-ld. +; XFAIL: * ; PR1434 ; RUN: llvm-as %s -o %t.bar.bc ; RUN: echo {define i32* @foo(i32 %x) \{ ret i32* @baz \} \ From clattner at apple.com Sat Jan 15 19:21:28 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 17:21:28 -0800 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp In-Reply-To: References: <20110115203031.19E282A6C12C@llvm.org> Message-ID: <555AB2F7-CAF1-40AC-ACD5-06F8DD506370@apple.com> On Jan 15, 2011, at 5:02 PM, Chris Lattner wrote: >> I hope this is fast enough to replace Kernighan-style counting loops even when >> the input is rather sparse. > > Wow, very nice! This is quite a bit faster than the old one. When I replace the manual ctpop loops in crafty with builtin_popcount, the performance still slows down from 5.5s to 5.62s though. This is very close to making me want to turn it on, particularly given that this is almost certainly a win for all targets with popcount and 32-bit platforms without them. Oh, also, when I build with -msse42, the runtime is 5.0s - a 10% speedup! -Chris From chandlerc at gmail.com Sat Jan 15 19:40:23 2011 From: chandlerc at gmail.com (Chandler Carruth) Date: Sun, 16 Jan 2011 01:40:23 -0000 Subject: [llvm-commits] [llvm] r123556 - /llvm/trunk/lib/Target/README.txt Message-ID: <20110116014023.678B42A6C12C@llvm.org> Author: chandlerc Date: Sat Jan 15 19:40:23 2011 New Revision: 123556 URL: http://llvm.org/viewvc/llvm-project?rev=123556&view=rev Log: Simplify a README.txt entry significantly to expose the core issue. 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=123556&r1=123555&r2=123556&view=diff ============================================================================== --- llvm/trunk/lib/Target/README.txt (original) +++ llvm/trunk/lib/Target/README.txt Sat Jan 15 19:40:23 2011 @@ -2162,37 +2162,35 @@ clang -O3 -fno-exceptions currently compiles this code: -void f(int N) { - std::vector v(N); - for (int k = 0; k < N; ++k) - v[k] = 0; - - extern void sink(void*); sink(&v); +void f(char* a, int n) { + __builtin_memset(a, 0, n); + for (int i = 0; i < n; ++i) + a[i] = 0; } -into almost the same as the previous note, but replace its final BB with: +into: -for.body.lr.ph: ; preds = %cond.true.i.i.i.i - %mul.i.i.i.i.i = shl i64 %conv, 2 - %call3.i.i.i.i.i = call noalias i8* @_Znwm(i64 %mul.i.i.i.i.i) nounwind - %0 = bitcast i8* %call3.i.i.i.i.i to i32* - store i32* %0, i32** %v8.sub, align 8, !tbaa !0 - %add.ptr.i.i.i = getelementptr inbounds i32* %0, i64 %conv - store i32* %add.ptr.i.i.i, i32** %tmp4.i.i.i.i.i, align 8, !tbaa !0 - call void @llvm.memset.p0i8.i64(i8* %call3.i.i.i.i.i, i8 0, i64 %mul.i.i.i.i.i, i32 4, i1 false) - store i32* %add.ptr.i.i.i, i32** %tmp3.i.i.i.i.i, align 8, !tbaa !0 - %tmp18 = add i32 %N, -1 - %tmp19 = zext i32 %tmp18 to i64 - %tmp20 = shl i64 %tmp19, 2 - %tmp21 = add i64 %tmp20, 4 - call void @llvm.memset.p0i8.i64(i8* %call3.i.i.i.i.i, i8 0, i64 %tmp21, i32 4, i1 false) - br label %for.end +define void @_Z1fPci(i8* nocapture %a, i32 %n) nounwind { +entry: + %conv = sext i32 %n to i64 + tail call void @llvm.memset.p0i8.i64(i8* %a, i8 0, i64 %conv, i32 1, i1 false) + %cmp8 = icmp sgt i32 %n, 0 + br i1 %cmp8, label %for.body.lr.ph, label %for.end + +for.body.lr.ph: ; preds = %entry + %tmp10 = add i32 %n, -1 + %tmp11 = zext i32 %tmp10 to i64 + %tmp12 = add i64 %tmp11, 1 + call void @llvm.memset.p0i8.i64(i8* %a, i8 0, i64 %tmp12, i32 1, i1 false) + ret void -First off, why (((zext %N - 1) << 2) + 4) instead of the ((sext %N) << 2) done -previously? (or better yet, re-use that one?) +for.end: ; preds = %entry + ret void +} -Then, the really painful one is the second memset, of the same memory, to the -same value. +This shouldn't need the ((zext (%n - 1)) + 1) game, and it should ideally fold +the two memset's together. The issue with %n seems to stem from poor handling +of the original loop. //===---------------------------------------------------------------------===// From bigcheesegs at gmail.com Sat Jan 15 19:43:22 2011 From: bigcheesegs at gmail.com (Michael J. Spencer) Date: Sun, 16 Jan 2011 01:43:22 -0000 Subject: [llvm-commits] [llvm] r123557 - in /llvm/trunk: include/llvm/Bitcode/Archive.h lib/Archive/ArchiveWriter.cpp test/Linker/link-archive.ll Message-ID: <20110116014322.92EBE2A6C12C@llvm.org> Author: mspencer Date: Sat Jan 15 19:43:22 2011 New Revision: 123557 URL: http://llvm.org/viewvc/llvm-project?rev=123557&view=rev Log: Revert "Archive: Replace all internal uses of PathV1 with PathV2. The external API still uses PathV1." Modified: llvm/trunk/include/llvm/Bitcode/Archive.h llvm/trunk/lib/Archive/ArchiveWriter.cpp llvm/trunk/test/Linker/link-archive.ll Modified: llvm/trunk/include/llvm/Bitcode/Archive.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Bitcode/Archive.h?rev=123557&r1=123556&r2=123557&view=diff ============================================================================== --- llvm/trunk/include/llvm/Bitcode/Archive.h (original) +++ llvm/trunk/include/llvm/Bitcode/Archive.h Sat Jan 15 19:43:22 2011 @@ -25,7 +25,6 @@ namespace llvm { class MemoryBuffer; - class raw_ostream; // Forward declare classes class Module; // From VMCore @@ -483,7 +482,7 @@ bool loadSymbolTable(std::string* ErrMessage); /// @brief Write the symbol table to an ofstream. - void writeSymbolTable(raw_ostream& ARFile); + void writeSymbolTable(std::ofstream& ARFile); /// Writes one ArchiveMember to an ofstream. If an error occurs, returns /// false, otherwise true. If an error occurs and error is non-null then @@ -492,7 +491,7 @@ /// @returns true Writing member failed, \p error set to error message bool writeMember( const ArchiveMember& member, ///< The member to be written - raw_ostream& ARFile, ///< The file to write member onto + std::ofstream& ARFile, ///< The file to write member onto bool CreateSymbolTable, ///< Should symbol table be created? bool TruncateNames, ///< Should names be truncated to 11 chars? bool ShouldCompress, ///< Should the member be compressed? Modified: llvm/trunk/lib/Archive/ArchiveWriter.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Archive/ArchiveWriter.cpp?rev=123557&r1=123556&r2=123557&view=diff ============================================================================== --- llvm/trunk/lib/Archive/ArchiveWriter.cpp (original) +++ llvm/trunk/lib/Archive/ArchiveWriter.cpp Sat Jan 15 19:43:22 2011 @@ -18,7 +18,6 @@ #include "llvm/Support/FileSystem.h" #include "llvm/Support/MemoryBuffer.h" #include "llvm/Support/Process.h" -#include "llvm/Support/raw_ostream.h" #include "llvm/Support/Signals.h" #include "llvm/Support/system_error.h" #include @@ -28,7 +27,7 @@ // Write an integer using variable bit rate encoding. This saves a few bytes // per entry in the symbol table. -static inline void writeInteger(unsigned num, raw_ostream& ARFile) { +static inline void writeInteger(unsigned num, std::ofstream& ARFile) { while (1) { if (num < 0x80) { // done? ARFile << (unsigned char)num; @@ -202,14 +201,14 @@ bool Archive::writeMember( const ArchiveMember& member, - raw_ostream& ARFile, + std::ofstream& ARFile, bool CreateSymbolTable, bool TruncateNames, bool ShouldCompress, std::string* ErrMsg ) { - unsigned filepos = ARFile.tell(); + unsigned filepos = ARFile.tellp(); filepos -= 8; // Get the data and its size either from the @@ -282,7 +281,7 @@ ARFile.write(data,fSize); // Make sure the member is an even length - if ((ARFile.tell() & 1) == 1) + if ((ARFile.tellp() & 1) == 1) ARFile << ARFILE_PAD; // Close the mapped file if it was opened @@ -292,7 +291,7 @@ // Write out the LLVM symbol table as an archive member to the file. void -Archive::writeSymbolTable(raw_ostream& ARFile) { +Archive::writeSymbolTable(std::ofstream& ARFile) { // Construct the symbol table's header ArchiveMemberHeader Hdr; @@ -316,7 +315,7 @@ #ifndef NDEBUG // Save the starting position of the symbol tables data content. - unsigned startpos = ARFile.tell(); + unsigned startpos = ARFile.tellp(); #endif // Write out the symbols sequentially @@ -333,7 +332,7 @@ #ifndef NDEBUG // Now that we're done with the symbol table, get the ending file position - unsigned endpos = ARFile.tell(); + unsigned endpos = ARFile.tellp(); #endif // Make sure that the amount we wrote is what we pre-computed. This is @@ -362,19 +361,25 @@ } // Create a temporary file to store the archive in - SmallString<128> TempArchivePath; - int ArchFD; - if (error_code ec = sys::fs::unique_file("%%-%%-%%-%%" + archPath.str(), - ArchFD, TempArchivePath)) { - if (ErrMsg) *ErrMsg = ec.message(); + sys::Path TmpArchive = archPath; + if (TmpArchive.createTemporaryFileOnDisk(ErrMsg)) return true; - } // Make sure the temporary gets removed if we crash - sys::RemoveFileOnSignal(sys::Path(TempArchivePath.str())); + sys::RemoveFileOnSignal(TmpArchive); // Create archive file for output. - raw_fd_ostream ArchiveFile(ArchFD, true); + std::ios::openmode io_mode = std::ios::out | std::ios::trunc | + std::ios::binary; + std::ofstream ArchiveFile(TmpArchive.c_str(), io_mode); + + // Check for errors opening or creating archive file. + if (!ArchiveFile.is_open() || ArchiveFile.bad()) { + TmpArchive.eraseFromDisk(); + if (ErrMsg) + *ErrMsg = "Error opening archive file: " + archPath.str(); + return true; + } // If we're creating a symbol table, reset it now if (CreateSymbolTable) { @@ -390,9 +395,8 @@ for (MembersList::iterator I = begin(), E = end(); I != E; ++I) { if (writeMember(*I, ArchiveFile, CreateSymbolTable, TruncateNames, Compress, ErrMsg)) { + TmpArchive.eraseFromDisk(); ArchiveFile.close(); - bool existed; - sys::fs::remove(TempArchivePath.str(), existed); return true; } } @@ -407,12 +411,12 @@ // ensure compatibility with other archivers we need to put the symbol // table first in the file. Unfortunately, this means mapping the file // we just wrote back in and copying it to the destination file. - SmallString<128> TempArchiveWithSymbolTablePath; + sys::Path FinalFilePath = archPath; // Map in the archive we just wrote. { OwningPtr arch; - if (error_code ec = MemoryBuffer::getFile(TempArchivePath.c_str(), arch)) { + if (error_code ec = MemoryBuffer::getFile(TmpArchive.c_str(), arch)) { if (ErrMsg) *ErrMsg = ec.message(); return true; @@ -421,14 +425,17 @@ // Open another temporary file in order to avoid invalidating the // mmapped data - if (error_code ec = sys::fs::unique_file("%%-%%-%%-%%" + archPath.str(), - ArchFD, TempArchiveWithSymbolTablePath)) { - if (ErrMsg) *ErrMsg = ec.message(); + if (FinalFilePath.createTemporaryFileOnDisk(ErrMsg)) return true; - } - sys::RemoveFileOnSignal(sys::Path(TempArchiveWithSymbolTablePath.str())); + sys::RemoveFileOnSignal(FinalFilePath); - raw_fd_ostream FinalFile(ArchFD, true); + std::ofstream FinalFile(FinalFilePath.c_str(), io_mode); + if (!FinalFile.is_open() || FinalFile.bad()) { + TmpArchive.eraseFromDisk(); + if (ErrMsg) + *ErrMsg = "Error opening archive file: " + FinalFilePath.str(); + return true; + } // Write the file magic number FinalFile << ARFILE_MAGIC; @@ -441,8 +448,7 @@ if (foreignST) { if (writeMember(*foreignST, FinalFile, false, false, false, ErrMsg)) { FinalFile.close(); - bool existed; - sys::fs::remove(TempArchiveWithSymbolTablePath.str(), existed); + TmpArchive.eraseFromDisk(); return true; } } @@ -460,11 +466,8 @@ } // free arch. // Move the final file over top of TmpArchive - if (error_code ec = sys::fs::rename(TempArchiveWithSymbolTablePath.str(), - TempArchivePath.str())) { - if (ErrMsg) *ErrMsg = ec.message(); + if (FinalFilePath.renamePathOnDisk(TmpArchive, ErrMsg)) return true; - } } // Before we replace the actual archive, we need to forget all the @@ -472,11 +475,8 @@ // this because we cannot replace an open file on Windows. cleanUpMemory(); - if (error_code ec = sys::fs::rename(TempArchivePath.str(), - archPath.str())) { - if (ErrMsg) *ErrMsg = ec.message(); + if (TmpArchive.renamePathOnDisk(archPath, ErrMsg)) return true; - } // Set correct read and write permissions after temporary file is moved // to final destination path. Modified: llvm/trunk/test/Linker/link-archive.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Linker/link-archive.ll?rev=123557&r1=123556&r2=123557&view=diff ============================================================================== --- llvm/trunk/test/Linker/link-archive.ll (original) +++ llvm/trunk/test/Linker/link-archive.ll Sat Jan 15 19:43:22 2011 @@ -1,5 +1,4 @@ ; Test linking of a bc file to an archive via llvm-ld. -; XFAIL: * ; PR1434 ; RUN: llvm-as %s -o %t.bar.bc ; RUN: echo {define i32* @foo(i32 %x) \{ ret i32* @baz \} \ From clattner at apple.com Sat Jan 15 19:50:58 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 17:50:58 -0800 Subject: [llvm-commits] [PATCH] new source file User.cpp In-Reply-To: References: Message-ID: <0E086E0F-1701-43EC-B327-F51D20158926@apple.com> Makes sense, please do. -Chris On Jan 15, 2011, at 6:38 AM, Jay Foad wrote: > This patch moves the implementation of User methods out of Use.cpp and > Value.cpp, and into a new source file User.cpp. > > Tested with "make check". Comments? Or, OK to commit? > > Thanks, > Jay. > _______________________________________________ > llvm-commits mailing list > llvm-commits at cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits From clattner at apple.com Sat Jan 15 19:51:26 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 17:51:26 -0800 Subject: [llvm-commits] [PATCH] remove unnecessary specialization of OperandTraits<> In-Reply-To: References: Message-ID: <7747297D-8370-45ED-8EA1-E2F145FCE8E4@apple.com> On Jan 15, 2011, at 5:43 AM, Jay Foad wrote: > OperandTraits<> is always used with a concrete subclass of User. There > is no need to specialize it for User itself. > > Tested with "make check". I also tested a patch that only removed the > out-of-line definitions of the methods declared in > OperandTraits, just to double-check that they weren't being used > for anything. Ok, please feel free to commit "obvious" patches like this. Thanks for all the great work Jay! -Chris From sabre at nondot.org Sat Jan 15 20:05:10 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 02:05:10 -0000 Subject: [llvm-commits] [llvm] r123558 - /llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Message-ID: <20110116020510.89FAA2A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 20:05:10 2011 New Revision: 123558 URL: http://llvm.org/viewvc/llvm-project?rev=123558&view=rev Log: simplify this code, it is still broken but will follow up on llvm-commits. Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp?rev=123558&r1=123557&r2=123558&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Sat Jan 15 20:05:10 2011 @@ -2169,16 +2169,10 @@ // and we know how to evaluate it by moving the bitcast from the pointer // operand to the value operand. } else if (CE->getOpcode() == Instruction::BitCast && - isa(CE->getOperand(0)) && - CE->getType()->isPointerTy() && - CE->getOperand(0)->getType()->isPointerTy()) { - GlobalVariable *GV = cast(CE->getOperand(0)); + isa(CE->getOperand(0))) { // Do not allow weak/*_odr/linkonce/dllimport/dllexport linkage or // external globals. - if (!GV->hasUniqueInitializer()) - return false; - - return true; + return cast(CE->getOperand(0))->hasUniqueInitializer(); } } @@ -2360,12 +2354,9 @@ // If we're evaluating a store through a bitcast, then we need // to pull the bitcast off the pointer type and push it onto the // stored value. - Ptr = dyn_cast(Ptr->getOperand(0)); - if (!Ptr) return false; + Ptr = CE->getOperand(0); - const PointerType *Ty = dyn_cast(Ptr->getType()); - if (!Ty) return false; - const Type *NewTy = Ty->getElementType(); + const Type *NewTy=cast(Ptr->getType())->getElementType(); // A bitcast'd pointer implicitly points to the first field of a // struct. Insert implicity "gep @x, 0, 0, ..." until we get down @@ -2374,8 +2365,7 @@ while (const StructType *STy = dyn_cast(NewTy)) { NewTy = STy->getTypeAtIndex(0U); - const IntegerType *IdxTy = - IntegerType::get(NewTy->getContext(), 32); + const IntegerType *IdxTy =IntegerType::get(NewTy->getContext(), 32); Constant *IdxZero = ConstantInt::get(IdxTy, 0, false); Constant * const IdxList[] = {IdxZero, IdxZero}; From clattner at apple.com Sat Jan 15 20:09:41 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 18:09:41 -0800 Subject: [llvm-commits] [llvm] r123477 - /llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp In-Reply-To: <20110114221920.6E9742A6C12C@llvm.org> References: <20110114221920.6E9742A6C12C@llvm.org> Message-ID: <46BAC2A9-8A86-4088-8F07-E2C32134EA4A@apple.com> On Jan 14, 2011, at 2:19 PM, Owen Anderson wrote: > Author: resistor > Date: Fri Jan 14 16:19:20 2011 > New Revision: 123477 > > URL: http://llvm.org/viewvc/llvm-project?rev=123477&view=rev > Log: > Enhance GlobalOpt to be able evaluate initializers that involve stores through > bitcasts, at least in simple cases. This fixes clang's CodeGenCXX/virtual-base-dtor.cpp Hi Owen. Thanks for fixing the regression, but you should definitely add a testcase to the globalopt testsuite to catch this. In particular, the code you've added isn't safe (besides breaking the testsuite, which has already been fixed on mainline): > + // A bitcast'd pointer implicitly points to the first field of a > + // struct. Insert implicity "gep @x, 0, 0, ..." until we get down > + // to the first concrete member. > + // FIXME: This could be extended to work for arrays as well. > + while (const StructType *STy = dyn_cast(NewTy)) { > + NewTy = STy->getTypeAtIndex(0U); > + > + const IntegerType *IdxTy = > + IntegerType::get(NewTy->getContext(), 32); > + Constant *IdxZero = ConstantInt::get(IdxTy, 0, false); > + Constant * const IdxList[] = {IdxZero, IdxZero}; > + > + Ptr = ConstantExpr::getGetElementPtr(Ptr, IdxList, 2); > + } > + > + if (!isa(NewTy)) return false; > + Val = ConstantExpr::getBitCast(Val, NewTy); > + } > + This code seems highly confused. If the thing being stored isn't a pointer, the final getBitCast call will abort. -Chris From clattner at apple.com Sat Jan 15 20:13:36 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 18:13:36 -0800 Subject: [llvm-commits] [llvm] r123533 - in /llvm/trunk: include/llvm/IntrinsicInst.h lib/Analysis/LazyValueInfo.cpp In-Reply-To: <20110115091612.75AB02A6C12D@llvm.org> References: <20110115091612.75AB02A6C12D@llvm.org> Message-ID: On Jan 15, 2011, at 1:16 AM, Nick Lewycky wrote: > Author: nicholas > Date: Sat Jan 15 03:16:12 2011 > New Revision: 123533 > > URL: http://llvm.org/viewvc/llvm-project?rev=123533&view=rev > Log: > Teach LazyValueInfo that allocas aren't NULL. Over all of llvm-test, this saves > half a million non-local queries, each of which would otherwise have triggered a > linear scan over a basic block. > > Also fix a fixme for memory intrinsics which dereference pointers. With this, > we prove that a pointer is non-null because it was dereferenced by an intrinsic > 112 times in llvm-test. Cool. Please add a testcase for this, one other thing: > + if (MI->getRawDest() == Ptr || MI->getDest() == Ptr) > + return true; > + if (MemTransferInst *MTI = dyn_cast(MI)) > + return MTI->getRawSource() == Ptr || MTI->getSource() == Ptr; Would it be better to do something like: Ptr = Ptr->stripPointerCasts(); and then use MI->getDest()? -Chris From sabre at nondot.org Sat Jan 15 20:27:39 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 02:27:39 -0000 Subject: [llvm-commits] [llvm] r123559 - in /llvm/trunk: lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp lib/Target/X86/X86FastISel.cpp test/CodeGen/X86/fast-isel-gep.ll Message-ID: <20110116022739.121562A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 20:27:38 2011 New Revision: 123559 URL: http://llvm.org/viewvc/llvm-project?rev=123559&view=rev Log: reapply my fix for PR8961 with a tweak to properly handle multi-instruction sequences like calls. Many thanks to Jakob for finding a testcase. Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp llvm/trunk/lib/Target/X86/X86FastISel.cpp llvm/trunk/test/CodeGen/X86/fast-isel-gep.ll Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp?rev=123559&r1=123558&r2=123559&view=diff ============================================================================== --- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp (original) +++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp Sat Jan 15 20:27:38 2011 @@ -771,8 +771,16 @@ assert(RI.getOperand().isUse() && "The only use of the vreg must be a use, we haven't emitted the def!"); + MachineInstr *User = &*RI; + + // Set the insertion point properly. Folding the load can cause generation of + // other random instructions (like sign extends) for addressing modes, make + // sure they get inserted in a logical place before the new instruction. + FuncInfo->InsertPt = User; + FuncInfo->MBB = User->getParent(); + // Ask the target to try folding the load. - return FastIS->TryToFoldLoad(&*RI, RI.getOperandNo(), LI); + return FastIS->TryToFoldLoad(User, RI.getOperandNo(), LI); } #ifndef NDEBUG @@ -890,11 +898,9 @@ if (Inst != Begin) BeforeInst = llvm::prior(llvm::prior(BI)); if (BeforeInst && isa(BeforeInst) && - BeforeInst->hasOneUse() && *BeforeInst->use_begin() == Inst) { - FastIS->recomputeInsertPt(); - if (TryToFoldFastISelLoad(cast(BeforeInst), FastIS)) - --BI; // If we succeeded, don't re-select the load. - } + BeforeInst->hasOneUse() && *BeforeInst->use_begin() == Inst && + TryToFoldFastISelLoad(cast(BeforeInst), FastIS)) + --BI; // If we succeeded, don't re-select the load. continue; } Modified: llvm/trunk/lib/Target/X86/X86FastISel.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86FastISel.cpp?rev=123559&r1=123558&r2=123559&view=diff ============================================================================== --- llvm/trunk/lib/Target/X86/X86FastISel.cpp (original) +++ llvm/trunk/lib/Target/X86/X86FastISel.cpp Sat Jan 15 20:27:38 2011 @@ -1933,7 +1933,7 @@ XII.foldMemoryOperandImpl(*FuncInfo.MF, MI, OpNo, AddrOps, Size, Alignment); if (Result == 0) return false; - MI->getParent()->insert(MI, Result); + FuncInfo.MBB->insert(FuncInfo.InsertPt, Result); MI->eraseFromParent(); return true; } Modified: llvm/trunk/test/CodeGen/X86/fast-isel-gep.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/fast-isel-gep.ll?rev=123559&r1=123558&r2=123559&view=diff ============================================================================== --- llvm/trunk/test/CodeGen/X86/fast-isel-gep.ll (original) +++ llvm/trunk/test/CodeGen/X86/fast-isel-gep.ll Sat Jan 15 20:27:38 2011 @@ -70,3 +70,20 @@ ; X64: test4: ; X64: 128(%r{{.*}},%r{{.*}},8) } + +; PR8961 - Make sure the sext for the GEP addressing comes before the load that +; is folded. +define i64 @test5(i8* %A, i32 %I, i64 %B) nounwind { + %v8 = getelementptr i8* %A, i32 %I + %v9 = bitcast i8* %v8 to i64* + %v10 = load i64* %v9 + %v11 = add i64 %B, %v10 + ret i64 %v11 +; X64: test5: +; X64: movslq %esi, %rax +; X64-NEXT: movq (%rdi,%rax), %rax +; X64-NEXT: addq %rdx, %rax +; X64-NEXT: ret +} + + From clattner at apple.com Sat Jan 15 20:32:31 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 18:32:31 -0800 Subject: [llvm-commits] [llvm] r123372 - in /llvm/trunk: lib/Analysis/InstructionSimplify.cpp test/Transforms/InstSimplify/2010-12-20-I1Arithmetic.ll In-Reply-To: <4D2F892C.2060206@free.fr> References: <20110113085629.B95662A6C12D@llvm.org> <0E8DEDA5-D30C-4832-98A9-EE9E3420B4CD@apple.com> <4D2F892C.2060206@free.fr> Message-ID: On Jan 13, 2011, at 3:22 PM, Duncan Sands wrote: > Hi Chris, > >>> The most common simplification missed by instsimplify in unoptimized bitcode >>> is "X != 0 -> X" when X is a boolean. This occurs a lot because of the way >>> llvm-gcc converts gcc's conditional expressions. Add this, and a few other >>> similar transforms for completeness. >> >> Hi Duncan, >> >> I don't really understand the motivation here: unoptimized code has lots of missed stuff, and instsimplify doesn't run for it, no? > > by unoptimized I mean just before the first instcombine run. Ah ok! > Since instcombine first sees if SimplifyInstruction can mulch each instruction before working on > it itself, catching missing simplifies on "unoptimized code" basically means > finding stuff that could have been moved from instcombine to SimplifyInstruction > but wasn't. In the case in question I didn't remove stuff from instcombine but > it is still a win: if instcombine sees an ICmp with i1 operands it turns it into > a logical operation, for example > > case ICmpInst::ICMP_NE: // icmp eq i1 A, B -> A^B > return BinaryOperator::CreateXor(Op0, Op1); > > By catching cases in SimplifyInstruction you avoid creating a new instruction, > revisiting it etc. This change reduced the number of instructions combined by > 1% without any impact on final bitcode. Makes a lot of sense. I'm just hopeful that these simple cases can *move* from instcombine to instsimplify, since instcombine calls into instsimplify. > >> In any case, this looks very familiar to code already in instcombine. Can the instcombine code be zapped now? > > Unfortunately not because the instcombine logic is too general, see above. Ok. > PS: My program that automatically finds simplifications finds all kinds of > interesting ones [*] in optimized code, this unoptimized code one was just a > warm up you might say :) > > [*] Interesting = occurs a lot. The program first harvests IR sequences from > the test-suite, and orders them by how often they occur, before looking for > simplifications in them. Nice, I'm looking forward to hearing more about your tool! -Chris From sabre at nondot.org Sat Jan 15 20:44:02 2011 From: sabre at nondot.org (Chris Lattner) Date: Sat, 15 Jan 2011 18:44:02 -0800 Subject: [llvm-commits] [llvm] r123131 - in /llvm/trunk: include/llvm/Analysis/ScalarEvolution.h lib/Analysis/ScalarEvolution.cpp In-Reply-To: <27D0D229-AF31-4314-B451-40B81E43D2DA@apple.com> References: <20110109222635.527042A6C12C@llvm.org> <9F4529E4-DEDD-44EE-8EE8-16DE3C7E5113@apple.com> <34473955-0F55-4763-9702-2AF0212E3534@nondot.org> <27D0D229-AF31-4314-B451-40B81E43D2DA@apple.com> Message-ID: On Jan 11, 2011, at 11:36 AM, Dan Gohman wrote: > > On Jan 11, 2011, at 9:16 AM, Chris Lattner wrote: > >> On Jan 10, 2011, at 2:07 PM, Dan Gohman wrote: >>> On Jan 9, 2011, at 2:26 PM, Chris Lattner wrote: >>>> /// getMinusSCEV - Return a SCEV corresponding to LHS - RHS. >>>> /// >>>> -const SCEV *ScalarEvolution::getMinusSCEV(const SCEV *LHS, >>>> - const SCEV *RHS) { >>>> +const SCEV *ScalarEvolution::getMinusSCEV(const SCEV *LHS, const SCEV *RHS, >>>> + bool HasNUW, bool HasNSW) { >>>> // Fast path: X - X --> 0. >>>> if (LHS == RHS) >>>> return getConstant(LHS->getType(), 0); >>>> >>>> // X - Y --> X + -Y >>>> - return getAddExpr(LHS, getNegativeSCEV(RHS)); >>>> + return getAddExpr(LHS, getNegativeSCEV(RHS), HasNUW, HasNSW); >>> >>> This isn't safe. For example, suppose X and Y are bot INT_MIN (dynamically). >>> Then X - Y does not overflow, but X + -Y does. >> >> Ok, the only current client of this is reasoning about the generated add, not the hypothetical subtract. You're right that this is confusing, I added a clarifying comment in r123243. > > For the unsigned case, this is still really confusing. Take the most innocent- > looking expression like "X - 1" for example. This turns into X + UINT_MAX, > which has unsigned overflow. > > For the signed case, setting HasNSW requires proving that the RHS is not > UINT_MIN. > > A side note here is that the nsw and nuw concepts don't cover every > interesting kind of overflow. There are many situations where one operand > of an add is unsigned and the other operand is signed, but these are hard > to represent since they're asymmetric. Hrm, I think I see what you mean. The code in getMinusSCEVForExitTest ends up producing an AddRec from the getMinusExpr (it *knows* it will be an addrec) that counts down to (but doesn't pass) zero. Do you think a better way to handle this is to just call getMinusSCEV, get the addrec, then ask for the same addrec in non-overflowing form? >>>> +static const SCEVAddRecExpr * >>>> +isSimpleUnwrappingAddRec(const SCEV *S, const Loop *L) { >>>> + const SCEVAddRecExpr *SA = dyn_cast(S); >>>> + >>>> + // The SCEV must be an addrec of this loop. >>>> + if (!SA || SA->getLoop() != L || !SA->isAffine()) >>>> + return 0; >>>> + >>>> + // The SCEV must be known to not wrap in some way to be interesting. >>>> + if (!SA->hasNoUnsignedWrap() && !SA->hasNoSignedWrap()) >>>> + return 0; >>> >>> It's suspicious that this code doesn't care which of nsw or nuw >>> the instruction has. >> >> NSW is enough to know that the bounds is limited to (e.g.) 2^32, just in a different coordinate space. It means you can't "stride past" a value. > > Unsigned overflow happens in the middle of the signed range. > Signed overflow happens at two points along the unsigned range. > It's quite possible to have a well-behaved stride pattern in > one range that doesn't "stride past" its destination, but does > happen to cross over the overflow point of the other range. I don't think that matters for this code. The code is carefully set up so that it has an AddRec that doesn't overflow "somehow", and it knows which direction it is stepping, and it knows that it gets to a specific integer value. If it "strides past" this value, the only way for the program to be defined is if it wraps all the way back around to the value. Just knowing that the addrec has cannot wrap *somewhere* should be enough, we're not doing an analysis in its original domain. >>>> + // If the signs of the strides differ, then the negative stride is counting >>>> + // down to the positive stride. >>>> + if (LHSStride->getValue().isNegative() != RHSStride->getValue().isNegative()){ >>>> + if (RHSStride->getValue().isNegative()) >>>> + std::swap(LHS, RHS); >>>> + } else { >>>> + // If LHS's stride is smaller than RHS's stride, then "b" must be less than >>>> + // "a" and "b" is RHS is counting up (catching up) to LHS. This is true >>>> + // whether the strides are positive or negative. >>>> + if (RHSStride->getValue().slt(LHSStride->getValue())) >>>> + std::swap(LHS, RHS); >>>> + } >>>> + >>>> + return SE.getMinusSCEV(LHS, RHS, true /*HasNUW*/); >>> >>> This isn't safe if the addrecs are nsw but not nuw (following the suspicion >>> above). >> >> How so? > > > Suppose LHS is {X,+,-1} and RHS is {-4,+,1}. Both could legitimately have > NSW and not NUW. In the sixth iteration of the loop, LHS is X-6 and RHS is 1. > > X-6 - 1 => X-6 + -1 => X-6 + UINT_MAX, which has unsigned overflow if X != 6. Maybe (probably) I'm seriously misunderstanding something, but in this case, we'll end up getting a new SCEV {X+4,+,-2}. Setting HasNUW on is resultant addrec tells the clients that the *result of the addrec* (not all the subcomputations that happen) will not wrap past zero. This allows the client to know that the crossover point happens at x/2+2. -Chris From clattner at apple.com Sat Jan 15 20:48:29 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 18:48:29 -0800 Subject: [llvm-commits] [llvm] r122959 - in /llvm/trunk: lib/Target/README.txt lib/Transforms/InstCombine/InstCombineCalls.cpp test/Transforms/InstCombine/objsize.ll In-Reply-To: References: <20110106131105.59E222A6C12C@llvm.org> <36B13217-CC55-4CF9-9570-1ECCD6389A62@apple.com> <96404383-9073-4D74-8FE3-65CF220AC676@apple.com> Message-ID: <9916E82C-35D4-49E5-88FF-553342F2E6BF@apple.com> On Jan 8, 2011, at 1:12 AM, Frits van Bommel wrote: > On Sat, Jan 8, 2011 at 7:52 AM, Chris Lattner wrote: >> >> On Jan 7, 2011, at 12:49 PM, Benjamin Kramer wrote: >>> I reverted the change for now, there is another issue with this as it changes the semantics of __builtin_objectsize. >>> Unless we want to extend this GCC extension even more my patch is not safe in it's current form :( >> >> What exactly is the issue here? While I'm sympathetic to the "cdecl.c/ds" regression, I still think that your patch is a good route forward. Optimizing to "know" the strlen size seems like a nice second step. >> >> However, if your patch wasn't safe, then it clearly isn't good. What is the issue? > > The safety concern he refers to is that it changes the semantics of > llvm.objectsize. LangRef.html states: > "The llvm.objectsize intrinsic is lowered to either a constant > representing the size of the object concerned, or i32/i64 -1 or 0, > depending on the type argument, if the size cannot be determined at > compile time." > This patch would instead replace it with a non-constant value. Ok, I think that it is just over-specified. The result of this gets passed into memset_chk etc as a dynamic parameter. Passing in a variable is just fine, and is actually good for safety (an abort will happen if the dynamic size is less than the dynamic # bytes to copy). > It might therefore not be unreasonable to replace it with a > non-constant value if this is *known* to significantly simplify all > uses, for instance by requiring that they all simplify to constants > (or in the case of terminators, to an unconditional branch or > unreachable). > This would still require adjusting the semantics first, though, as > it's currently not legal to do any of that. > > P.S. Is it safe to return a constant malloc argument anyway? malloc() > might return null, in which case the only safe-to-use size would be > 0... Good point. I'd say "no, it's not" unless there is something else in the code that tells you that the result is assumed to not be null. For example a store through the returned pointer or a dynamic null check. -Chris From stoklund at 2pi.dk Sat Jan 15 20:51:57 2011 From: stoklund at 2pi.dk (Jakob Stoklund Olesen) Date: Sat, 15 Jan 2011 18:51:57 -0800 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp In-Reply-To: References: <20110115203031.19E282A6C12C@llvm.org> Message-ID: On Jan 15, 2011, at 5:02 PM, Chris Lattner wrote: > One major advantage of recognizing popcount though is that the optimizer has a chance of hacking on it, and there are a lot of instcombine xforms that we could do. Here are some that I note looking at the bc file for crafy (when hacked to use builtin_popcount). I attached the bc file below if you're interested. I noticed that we have intrinsics and ISD nodes for ctlz, cttz, and ctpop, but none for parity. Is that on purpose? Currently, clang compiles __builtin_parity to: define i32 @f(i32 %x) nounwind readnone ssp { entry: %tmp1 = tail call i32 @llvm.ctpop.i32(i32 %x) %tmp2 = and i32 %tmp1, 1 ret i32 %tmp2 } With corresponding horrible codegen on x86. The IR would be fine for a target with a POPCNT instruction and no parity, but it is much cheaper to compute just the parity when you don't have a POPCNT. You want something like: x ^= x >> 16; x ^= x >> 8; return parity_bit(eflags); Or portably in compiler-rt: si_int __paritysi2(si_int a) { su_int x = (su_int)a; x ^= x >> 16; x ^= x >> 8; x ^= x >> 4; return (0x6996 >> (x & 0xF)) & 1; } /jakob -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/394310a5/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: smime.p7s Type: application/pkcs7-signature Size: 1929 bytes Desc: not available Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/394310a5/attachment.bin From sabre at nondot.org Sat Jan 15 20:56:53 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 02:56:53 -0000 Subject: [llvm-commits] [llvm] r123560 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/crash.ll Message-ID: <20110116025653.D312E2A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 20:56:53 2011 New Revision: 123560 URL: http://llvm.org/viewvc/llvm-project?rev=123560&view=rev Log: fix PR8981, a crash trying to form a conditional inc with a floating point compare. Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp llvm/trunk/test/CodeGen/X86/crash.ll Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=123560&r1=123559&r2=123560&view=diff ============================================================================== --- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original) +++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Sat Jan 15 20:56:53 2011 @@ -11560,7 +11560,8 @@ SDValue Cmp = SetCC.getOperand(1); if (Cmp.getOpcode() != X86ISD::CMP || !Cmp.hasOneUse() || - !X86::isZeroNode(Cmp.getOperand(1))) + !X86::isZeroNode(Cmp.getOperand(1)) || + !Cmp.getOperand(0).getValueType().isInteger()) return SDValue(); SDValue CmpOp0 = Cmp.getOperand(0); Modified: llvm/trunk/test/CodeGen/X86/crash.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/crash.ll?rev=123560&r1=123559&r2=123560&view=diff ============================================================================== --- llvm/trunk/test/CodeGen/X86/crash.ll (original) +++ llvm/trunk/test/CodeGen/X86/crash.ll Sat Jan 15 20:56:53 2011 @@ -141,3 +141,13 @@ call void asm sideeffect "outb $0, ${1:w}", "{ax},N{dx},~{dirflag},~{fpsr},~{flags}"(i8 %conv4.i, i32 1017) nounwind unreachable } + +; Crash trying to form conditional increment with fp value. +; PR8981 +define i32 @test9(double %X) ssp align 2 { +entry: + %0 = fcmp one double %X, 0.000000e+00 + %cond = select i1 %0, i32 1, i32 2 + ret i32 %cond +} + From venkatra at cs.wisc.edu Sat Jan 15 21:15:11 2011 From: venkatra at cs.wisc.edu (Venkatraman Govindaraju) Date: Sun, 16 Jan 2011 03:15:11 -0000 Subject: [llvm-commits] [llvm] r123561 - in /llvm/trunk: lib/Target/Sparc/SparcInstrInfo.cpp lib/Target/Sparc/SparcInstrInfo.h test/CodeGen/SPARC/2011-01-11-CC.ll Message-ID: <20110116031511.3ECB72A6C12C@llvm.org> Author: venkatra Date: Sat Jan 15 21:15:11 2011 New Revision: 123561 URL: http://llvm.org/viewvc/llvm-project?rev=123561&view=rev Log: Implement AnalyzeBranch in Sparc Backend. Modified: llvm/trunk/lib/Target/Sparc/SparcInstrInfo.cpp llvm/trunk/lib/Target/Sparc/SparcInstrInfo.h llvm/trunk/test/CodeGen/SPARC/2011-01-11-CC.ll Modified: llvm/trunk/lib/Target/Sparc/SparcInstrInfo.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/Sparc/SparcInstrInfo.cpp?rev=123561&r1=123560&r2=123561&view=diff ============================================================================== --- llvm/trunk/lib/Target/Sparc/SparcInstrInfo.cpp (original) +++ llvm/trunk/lib/Target/Sparc/SparcInstrInfo.cpp Sat Jan 15 21:15:11 2011 @@ -66,15 +66,200 @@ return 0; } +static bool IsIntegerCC(unsigned CC) +{ + return (CC <= SPCC::ICC_VC); +} + + +static SPCC::CondCodes GetOppositeBranchCondition(SPCC::CondCodes CC) +{ + switch(CC) { + default: llvm_unreachable("Unknown condition code"); + case SPCC::ICC_NE: return SPCC::ICC_E; + case SPCC::ICC_E: return SPCC::ICC_NE; + case SPCC::ICC_G: return SPCC::ICC_LE; + case SPCC::ICC_LE: return SPCC::ICC_G; + case SPCC::ICC_GE: return SPCC::ICC_L; + case SPCC::ICC_L: return SPCC::ICC_GE; + case SPCC::ICC_GU: return SPCC::ICC_LEU; + case SPCC::ICC_LEU: return SPCC::ICC_GU; + case SPCC::ICC_CC: return SPCC::ICC_CS; + case SPCC::ICC_CS: return SPCC::ICC_CC; + case SPCC::ICC_POS: return SPCC::ICC_NEG; + case SPCC::ICC_NEG: return SPCC::ICC_POS; + case SPCC::ICC_VC: return SPCC::ICC_VS; + case SPCC::ICC_VS: return SPCC::ICC_VC; + + case SPCC::FCC_U: return SPCC::FCC_O; + case SPCC::FCC_O: return SPCC::FCC_U; + case SPCC::FCC_G: return SPCC::FCC_LE; + case SPCC::FCC_LE: return SPCC::FCC_G; + case SPCC::FCC_UG: return SPCC::FCC_ULE; + case SPCC::FCC_ULE: return SPCC::FCC_UG; + case SPCC::FCC_L: return SPCC::FCC_GE; + case SPCC::FCC_GE: return SPCC::FCC_L; + case SPCC::FCC_UL: return SPCC::FCC_UGE; + case SPCC::FCC_UGE: return SPCC::FCC_UL; + case SPCC::FCC_LG: return SPCC::FCC_UE; + case SPCC::FCC_UE: return SPCC::FCC_LG; + case SPCC::FCC_NE: return SPCC::FCC_E; + case SPCC::FCC_E: return SPCC::FCC_NE; + } +} + + +bool SparcInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB, + MachineBasicBlock *&TBB, + MachineBasicBlock *&FBB, + SmallVectorImpl &Cond, + bool AllowModify) const +{ + + MachineBasicBlock::iterator I = MBB.end(); + MachineBasicBlock::iterator UnCondBrIter = MBB.end(); + while (I != MBB.begin()) { + --I; + + if (I->isDebugValue()) + continue; + + //When we see a non-terminator, we are done + if (!isUnpredicatedTerminator(I)) + break; + + //Terminator is not a branch + if (!I->getDesc().isBranch()) + return true; + + //Handle Unconditional branches + if (I->getOpcode() == SP::BA) { + UnCondBrIter = I; + + if (!AllowModify) { + TBB = I->getOperand(0).getMBB(); + continue; + } + + while (llvm::next(I) != MBB.end()) + llvm::next(I)->eraseFromParent(); + + Cond.clear(); + FBB = 0; + + if (MBB.isLayoutSuccessor(I->getOperand(0).getMBB())) { + TBB = 0; + I->eraseFromParent(); + I = MBB.end(); + UnCondBrIter = MBB.end(); + continue; + } + + TBB = I->getOperand(0).getMBB(); + continue; + } + + unsigned Opcode = I->getOpcode(); + if (Opcode != SP::BCOND && Opcode != SP::FBCOND) + return true; //Unknown Opcode + + SPCC::CondCodes BranchCode = (SPCC::CondCodes)I->getOperand(1).getImm(); + + if (Cond.empty()) { + MachineBasicBlock *TargetBB = I->getOperand(0).getMBB(); + if (AllowModify && UnCondBrIter != MBB.end() && + MBB.isLayoutSuccessor(TargetBB)) { + + //Transform the code + // + // brCC L1 + // ba L2 + // L1: + // .. + // L2: + // + // into + // + // brnCC L2 + // L1: + // ... + // L2: + // + BranchCode = GetOppositeBranchCondition(BranchCode); + MachineBasicBlock::iterator OldInst = I; + BuildMI(MBB, UnCondBrIter, MBB.findDebugLoc(I), get(Opcode)) + .addMBB(UnCondBrIter->getOperand(0).getMBB()).addImm(BranchCode); + BuildMI(MBB, UnCondBrIter, MBB.findDebugLoc(I), get(SP::BA)) + .addMBB(TargetBB); + MBB.addSuccessor(TargetBB); + OldInst->eraseFromParent(); + UnCondBrIter->eraseFromParent(); + + UnCondBrIter = MBB.end(); + I = MBB.end(); + continue; + } + FBB = TBB; + TBB = I->getOperand(0).getMBB(); + Cond.push_back(MachineOperand::CreateImm(BranchCode)); + continue; + } + //FIXME: Handle subsequent conditional branches + //For now, we can't handle multiple conditional branches + return true; + } + return false; +} + unsigned SparcInstrInfo::InsertBranch(MachineBasicBlock &MBB,MachineBasicBlock *TBB, MachineBasicBlock *FBB, const SmallVectorImpl &Cond, - DebugLoc DL)const{ - // Can only insert uncond branches so far. - assert(Cond.empty() && !FBB && TBB && "Can only handle uncond branches!"); - BuildMI(&MBB, DL, get(SP::BA)).addMBB(TBB); - return 1; + DebugLoc DL) const { + assert(TBB && "InsertBranch must not be told to insert a fallthrough"); + assert((Cond.size() == 1 || Cond.size() == 0) && + "Sparc branch conditions should have one component!"); + + if (Cond.empty()) { + assert(!FBB && "Unconditional branch with multiple successors!"); + BuildMI(&MBB, DL, get(SP::BA)).addMBB(TBB); + return 1; + } + + //Conditional branch + unsigned CC = Cond[0].getImm(); + + if (IsIntegerCC(CC)) + BuildMI(&MBB, DL, get(SP::BCOND)).addMBB(TBB).addImm(CC); + else + BuildMI(&MBB, DL, get(SP::FBCOND)).addMBB(TBB).addImm(CC); + if (!FBB) + return 1; + + BuildMI(&MBB, DL, get(SP::BA)).addMBB(FBB); + return 2; +} + +unsigned SparcInstrInfo::RemoveBranch(MachineBasicBlock &MBB) const +{ + MachineBasicBlock::iterator I = MBB.end(); + unsigned Count = 0; + while (I != MBB.begin()) { + --I; + + if (I->isDebugValue()) + continue; + + if (I->getOpcode() != SP::BA + && I->getOpcode() != SP::BCOND + && I->getOpcode() != SP::FBCOND) + break; // Not a branch + + I->eraseFromParent(); + I = MBB.end(); + ++Count; + } + return Count; } void SparcInstrInfo::copyPhysReg(MachineBasicBlock &MBB, Modified: llvm/trunk/lib/Target/Sparc/SparcInstrInfo.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/Sparc/SparcInstrInfo.h?rev=123561&r1=123560&r2=123561&view=diff ============================================================================== --- llvm/trunk/lib/Target/Sparc/SparcInstrInfo.h (original) +++ llvm/trunk/lib/Target/Sparc/SparcInstrInfo.h Sat Jan 15 21:15:11 2011 @@ -58,8 +58,15 @@ /// any side effects other than storing to the stack slot. virtual unsigned isStoreToStackSlot(const MachineInstr *MI, int &FrameIndex) const; - - + + + virtual bool AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB, + MachineBasicBlock *&FBB, + SmallVectorImpl &Cond, + bool AllowModify = false) const ; + + virtual unsigned RemoveBranch(MachineBasicBlock &MBB) const; + virtual unsigned InsertBranch(MachineBasicBlock &MBB, MachineBasicBlock *TBB, MachineBasicBlock *FBB, const SmallVectorImpl &Cond, Modified: llvm/trunk/test/CodeGen/SPARC/2011-01-11-CC.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/SPARC/2011-01-11-CC.ll?rev=123561&r1=123560&r2=123561&view=diff ============================================================================== --- llvm/trunk/test/CodeGen/SPARC/2011-01-11-CC.ll (original) +++ llvm/trunk/test/CodeGen/SPARC/2011-01-11-CC.ll Sat Jan 15 21:15:11 2011 @@ -17,7 +17,7 @@ entry: ; CHECK: test_select_int_icc ; CHECK: subcc -; CHECK: be +; CHECK: {{be|bne}} %0 = icmp eq i32 %a, 0 %1 = select i1 %0, i32 %b, i32 %c ret i32 %1 @@ -28,7 +28,7 @@ entry: ; CHECK: test_select_fp_icc ; CHECK: subcc -; CHECK: be +; CHECK: {{be|bne}} %0 = icmp eq i32 %a, 0 %1 = select i1 %0, float %f1, float %f2 ret float %1 @@ -36,9 +36,9 @@ define double @test_select_dfp_icc(i32 %a, double %f1, double %f2) nounwind readnone noinline { entry: -; CHECK: test_select_fp_icc +; CHECK: test_select_dfp_icc ; CHECK: subcc -; CHECK: be +; CHECK: {{be|bne}} %0 = icmp eq i32 %a, 0 %1 = select i1 %0, double %f1, double %f2 ret double %1 @@ -48,7 +48,7 @@ entry: ;CHECK: test_select_int_fcc ;CHECK: fcmps -;CHECK: fbne +;CHECK: {{fbe|fbne}} %0 = fcmp une float %f, 0.000000e+00 %a.b = select i1 %0, i32 %a, i32 %b ret i32 %a.b @@ -59,7 +59,7 @@ entry: ;CHECK: test_select_fp_fcc ;CHECK: fcmps -;CHECK: fbne +;CHECK: {{fbe|fbne}} %0 = fcmp une float %f, 0.000000e+00 %1 = select i1 %0, float %f1, float %f2 ret float %1 @@ -69,7 +69,7 @@ entry: ;CHECK: test_select_dfp_fcc ;CHECK: fcmpd -;CHECK: fbne +;CHECK: {{fbne|fbe}} %0 = fcmp une double %f, 0.000000e+00 %1 = select i1 %0, double %f1, double %f2 ret double %1 From rafael.espindola at gmail.com Sat Jan 15 21:37:37 2011 From: rafael.espindola at gmail.com (=?ISO-8859-1?Q?Rafael_=C1vila_de_Esp=EDndola?=) Date: Sat, 15 Jan 2011 22:37:37 -0500 Subject: [llvm-commits] [patch][pr8927] Don't merge two constants if both addresses are significant Message-ID: <4D326801.6040504@gmail.com> This patch fixes the original testcase in PR8927. It also causes a clang binary built with a patched clang to increase in size by 0.21%. We can probably get some of the size back by writing a pass that detects that a global never has its pointer compared and adds unnamed_addr to it. It is also possible that there are some other cases clang could add unnamed_addr to. Is the patch OK? Cheers, Rafael -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: PR8927.patch Url: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/b5f90ee6/attachment-0001.pl From sabre at nondot.org Sat Jan 15 21:43:53 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 03:43:53 -0000 Subject: [llvm-commits] [llvm] r123562 - in /llvm/trunk: lib/Analysis/ConstantFolding.cpp test/Transforms/InstCombine/crash.ll Message-ID: <20110116034354.084E12A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 21:43:53 2011 New Revision: 123562 URL: http://llvm.org/viewvc/llvm-project?rev=123562&view=rev Log: fix PR8983, a broken assertion. Modified: llvm/trunk/lib/Analysis/ConstantFolding.cpp llvm/trunk/test/Transforms/InstCombine/crash.ll Modified: llvm/trunk/lib/Analysis/ConstantFolding.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ConstantFolding.cpp?rev=123562&r1=123561&r2=123562&view=diff ============================================================================== --- llvm/trunk/lib/Analysis/ConstantFolding.cpp (original) +++ llvm/trunk/lib/Analysis/ConstantFolding.cpp Sat Jan 15 21:43:53 2011 @@ -581,7 +581,7 @@ if (NumOps == 2 && cast(ResultTy)->getElementType()->isIntegerTy(8)) { ConstantExpr *CE = dyn_cast(Ops[1]); - assert(CE->getType() == IntPtrTy && + assert((CE == 0 || CE->getType() == IntPtrTy) && "CastGEPIndices didn't canonicalize index types!"); if (CE && CE->getOpcode() == Instruction::Sub && CE->getOperand(0)->isNullValue()) { Modified: llvm/trunk/test/Transforms/InstCombine/crash.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/crash.ll?rev=123562&r1=123561&r2=123562&view=diff ============================================================================== --- llvm/trunk/test/Transforms/InstCombine/crash.ll (original) +++ llvm/trunk/test/Transforms/InstCombine/crash.ll Sat Jan 15 21:43:53 2011 @@ -341,3 +341,15 @@ %select = select i1 %cmp, double 2.000000e+00, double 3.141592e+00 ret double %select } + + +; PR8983 +%struct.basic_ios = type { i8 } + +define %struct.basic_ios *@test17() ssp { +entry: + %add.ptr.i = getelementptr i8* null, i64 undef + %0 = bitcast i8* %add.ptr.i to %struct.basic_ios* + ret %struct.basic_ios* %0 +} + From rafael.espindola at gmail.com Sat Jan 15 22:13:43 2011 From: rafael.espindola at gmail.com (=?ISO-8859-1?Q?Rafael_=C1vila_de_Esp=EDndola?=) Date: Sat, 15 Jan 2011 23:13:43 -0500 Subject: [llvm-commits] [patch][pr8297] Don't use mergeable sections for constants with significant address Message-ID: <4D327077.1020803@gmail.com> I think the attached patch is the last necessary step in fixing PR8297. Building clang with a patched clang (on top of the constmerge patch) *reduces* the final clang binary by 472 bytes. Not sure what is going on, maybe we avoid creating a merge section that was not useful in the end? Cheers, Rafael -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: merge.patch Url: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/3aad296d/attachment.pl From resistor at mac.com Sat Jan 15 22:33:33 2011 From: resistor at mac.com (Owen Anderson) Date: Sun, 16 Jan 2011 04:33:33 -0000 Subject: [llvm-commits] [llvm] r123563 - in /llvm/trunk: lib/Transforms/IPO/GlobalOpt.cpp test/Transforms/GlobalOpt/static-ctor-bitcast.ll Message-ID: <20110116043333.82FA32A6C12C@llvm.org> Author: resistor Date: Sat Jan 15 22:33:33 2011 New Revision: 123563 URL: http://llvm.org/viewvc/llvm-project?rev=123563&view=rev Log: Improve the safety of my globalopt enhancement by ensuring that the bitcast of the stored value to the new store type is always. Also, add a testcase. Added: llvm/trunk/test/Transforms/GlobalOpt/static-ctor-bitcast.ll Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Modified: llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp?rev=123563&r1=123562&r2=123563&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp (original) +++ llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp Sat Jan 15 22:33:33 2011 @@ -2358,21 +2358,31 @@ const Type *NewTy=cast(Ptr->getType())->getElementType(); - // A bitcast'd pointer implicitly points to the first field of a - // struct. Insert implicity "gep @x, 0, 0, ..." until we get down - // to the first concrete member. - // FIXME: This could be extended to work for arrays as well. - while (const StructType *STy = dyn_cast(NewTy)) { - NewTy = STy->getTypeAtIndex(0U); - - const IntegerType *IdxTy =IntegerType::get(NewTy->getContext(), 32); - Constant *IdxZero = ConstantInt::get(IdxTy, 0, false); - Constant * const IdxList[] = {IdxZero, IdxZero}; + // In order to push the bitcast onto the stored value, a bitcast + // from NewTy to Val's type must be legal. If it's not, we can try + // introspecting NewTy to find a legal conversion. + while (!Val->getType()->canLosslesslyBitCastTo(NewTy)) { + // If NewTy is a struct, we can convert the pointer to the struct + // into a pointer to its first member. + // FIXME: This could be extended to support arrays as well. + if (const StructType *STy = dyn_cast(NewTy)) { + NewTy = STy->getTypeAtIndex(0U); + + const IntegerType *IdxTy =IntegerType::get(NewTy->getContext(), 32); + Constant *IdxZero = ConstantInt::get(IdxTy, 0, false); + Constant * const IdxList[] = {IdxZero, IdxZero}; + + Ptr = ConstantExpr::getGetElementPtr(Ptr, IdxList, 2); - Ptr = ConstantExpr::getGetElementPtr(Ptr, IdxList, 2); + // If we can't improve the situation by introspecting NewTy, + // we have to give up. + } else { + return 0; + } } - if (!isa(NewTy)) return false; + // If we found compatible types, go ahead and push the bitcast + // onto the stored value. Val = ConstantExpr::getBitCast(Val, NewTy); } Added: llvm/trunk/test/Transforms/GlobalOpt/static-ctor-bitcast.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/GlobalOpt/static-ctor-bitcast.ll?rev=123563&view=auto ============================================================================== --- llvm/trunk/test/Transforms/GlobalOpt/static-ctor-bitcast.ll (added) +++ llvm/trunk/test/Transforms/GlobalOpt/static-ctor-bitcast.ll Sat Jan 15 22:33:33 2011 @@ -0,0 +1,59 @@ +; RUN: opt -S -globalopt < %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-n8:16:32:64" +target triple = "x86_64-apple-darwin10.0.0" + +%0 = type { i8*, i8* } +%1 = type { i8*, i8*, i32, i32, i8*, i64 } +%2 = type { i32, void ()* } +%struct.A = type { i32 } +%struct.B = type { i32 (...)**, i8*, [4 x i8] } + + at y = global i8 0, align 1 + at x = global %struct.B zeroinitializer, align 8 + at _ZTV1B = weak_odr unnamed_addr constant [3 x i8*] [i8* inttoptr (i64 16 to i8*), i8* null, i8* bitcast (%1* @_ZTI1B to i8*)] + at _ZTVN10__cxxabiv121__vmi_class_type_infoE = external global i8* + at _ZTS1B = weak_odr constant [3 x i8] c"1B\00" + at _ZTVN10__cxxabiv117__class_type_infoE = external global i8* + at _ZTS1A = weak_odr constant [3 x i8] c"1A\00" + at _ZTI1A = weak_odr unnamed_addr constant %0 { i8* bitcast (i8** getelementptr inbounds (i8** @_ZTVN10__cxxabiv117__class_type_infoE, i64 2) to i8*), i8* getelementptr inbounds ([3 x i8]* @_ZTS1A, i32 0, i32 0) } + at _ZTI1B = weak_odr unnamed_addr constant %1 { i8* bitcast (i8** getelementptr inbounds (i8** @_ZTVN10__cxxabiv121__vmi_class_type_infoE, i64 2) to i8*), i8* getelementptr inbounds ([3 x i8]* @_ZTS1B, i32 0, i32 0), i32 0, i32 1, i8* bitcast (%0* @_ZTI1A to i8*), i64 -6141 } + at _ZTT1B = weak_odr unnamed_addr constant [1 x i8*] [i8* bitcast (i8** getelementptr inbounds ([3 x i8*]* @_ZTV1B, i64 1, i64 0) to i8*)] + at llvm.global_ctors = appending global [1 x %2] [%2 { i32 65535, void ()* @_GLOBAL__I_a }] + +; CHECK-NOT: __cxx_global_var_init +define internal void @__cxx_global_var_init() section "__TEXT,__StaticInit,regular,pure_instructions" { +entry: + call void @_ZN1BC1Ev(%struct.B* @x) + ret void +} + +; CHECK-NOT: _ZN1BC1Ev +define linkonce_odr unnamed_addr void @_ZN1BC1Ev(%struct.B* %this) inlinehint ssp align 2 { +entry: + %0 = bitcast %struct.B* %this to i8* + %1 = getelementptr inbounds i8* %0, i64 16 + %2 = bitcast i8* %1 to %struct.A* + call void @_ZN1AC2Ev(%struct.A* %2) + %3 = bitcast %struct.B* %this to i8*** + store i8** getelementptr inbounds ([3 x i8*]* @_ZTV1B, i64 1, i64 0), i8*** %3 + ret void +} + +; CHECK-NOT: _ZN1AC2Ev +define linkonce_odr unnamed_addr void @_ZN1AC2Ev(%struct.A* %this) nounwind ssp align 2 { +entry: + %0 = ptrtoint %struct.A* %this to i64 + %sub = sub i64 %0, ptrtoint (%struct.B* @x to i64) + %div = udiv i64 %sub, 8 + %conv = trunc i64 %div to i8 + store i8 %conv, i8* @y, align 1 + ret void +} + +; CHECK-NOT: _GLOBAL__I_a +define internal void @_GLOBAL__I_a() section "__TEXT,__StaticInit,regular,pure_instructions" { +entry: + call void @__cxx_global_var_init() + ret void +} From resistor at mac.com Sat Jan 15 22:36:59 2011 From: resistor at mac.com (Owen Anderson) Date: Sat, 15 Jan 2011 20:36:59 -0800 Subject: [llvm-commits] [llvm] r123477 - /llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp In-Reply-To: <46BAC2A9-8A86-4088-8F07-E2C32134EA4A@apple.com> References: <20110114221920.6E9742A6C12C@llvm.org> <46BAC2A9-8A86-4088-8F07-E2C32134EA4A@apple.com> Message-ID: <81EC8E53-108A-4C84-A423-5C23F88DAAA4@mac.com> On Jan 15, 2011, at 6:09 PM, Chris Lattner wrote: >> + // A bitcast'd pointer implicitly points to the first field of a >> + // struct. Insert implicity "gep @x, 0, 0, ..." until we get down >> + // to the first concrete member. >> + // FIXME: This could be extended to work for arrays as well. >> + while (const StructType *STy = dyn_cast(NewTy)) { >> + NewTy = STy->getTypeAtIndex(0U); >> + >> + const IntegerType *IdxTy = >> + IntegerType::get(NewTy->getContext(), 32); >> + Constant *IdxZero = ConstantInt::get(IdxTy, 0, false); >> + Constant * const IdxList[] = {IdxZero, IdxZero}; >> + >> + Ptr = ConstantExpr::getGetElementPtr(Ptr, IdxList, 2); >> + } >> + >> + if (!isa(NewTy)) return false; >> + Val = ConstantExpr::getBitCast(Val, NewTy); >> + } >> + > > This code seems highly confused. If the thing being stored isn't a pointer, the final getBitCast call will abort. Fixed in r123563. --Owen From sabre at nondot.org Sat Jan 15 22:37:29 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 04:37:29 -0000 Subject: [llvm-commits] [llvm] r123564 - /llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Message-ID: <20110116043729.7B40D2A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 22:37:29 2011 New Revision: 123564 URL: http://llvm.org/viewvc/llvm-project?rev=123564&view=rev Log: tidy up code. Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp?rev=123564&r1=123563&r2=123564&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Sat Jan 15 22:37:29 2011 @@ -520,12 +520,13 @@ AllowAggressive = false; PHINode *PN = cast(I.getOperand(0)); unsigned NumPHIValues = PN->getNumIncomingValues(); - if (NumPHIValues == 0 || - // We normally only transform phis with a single use, unless we're trying - // hard to make jump threading happen. - (!PN->hasOneUse() && !AllowAggressive)) + if (NumPHIValues == 0) return 0; + // We normally only transform phis with a single use, unless we're trying + // hard to make jump threading happen. + if (!PN->hasOneUse() && !AllowAggressive) + return 0; // Check to see if all of the operands of the PHI are simple constants // (constantint/constantfp/undef). If there is one non-constant value, @@ -533,18 +534,21 @@ // bail out. We don't do arbitrary constant expressions here because moving // their computation can be expensive without a cost model. BasicBlock *NonConstBB = 0; - for (unsigned i = 0; i != NumPHIValues; ++i) - if (!isa(PN->getIncomingValue(i)) || - isa(PN->getIncomingValue(i))) { - if (NonConstBB) return 0; // More than one non-const value. - if (isa(PN->getIncomingValue(i))) return 0; // Itself a phi. - NonConstBB = PN->getIncomingBlock(i); - - // If the incoming non-constant value is in I's block, we have an infinite - // loop. - if (NonConstBB == I.getParent()) - return 0; - } + for (unsigned i = 0; i != NumPHIValues; ++i) { + Value *InVal = PN->getIncomingValue(i); + if (isa(InVal) && !isa(InVal)) + continue; + + if (isa(InVal)) return 0; // Itself a phi. + if (NonConstBB) return 0; // More than one non-const value. + + NonConstBB = PN->getIncomingBlock(i); + + // If the incoming non-constant value is in I's block, we have an infinite + // loop. + if (NonConstBB == I.getParent()) + return 0; + } // If there is exactly one non-constant value, we can insert a copy of the // operation in that block. However, if this is a critical edge, we would be From stoklund at 2pi.dk Sat Jan 15 22:55:20 2011 From: stoklund at 2pi.dk (Jakob Stoklund Olesen) Date: Sat, 15 Jan 2011 20:55:20 -0800 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp In-Reply-To: References: <20110115203031.19E282A6C12C@llvm.org> Message-ID: <56D7F26B-3638-423A-950A-CD16B6A359B2@2pi.dk> On Jan 15, 2011, at 6:51 PM, Jakob Stoklund Olesen wrote: > > On Jan 15, 2011, at 5:02 PM, Chris Lattner wrote: > >> One major advantage of recognizing popcount though is that the optimizer has a chance of hacking on it, and there are a lot of instcombine xforms that we could do. Here are some that I note looking at the bc file for crafy (when hacked to use builtin_popcount). I attached the bc file below if you're interested. > > I noticed that we have intrinsics and ISD nodes for ctlz, cttz, and ctpop, but none for parity. Is that on purpose? > > Currently, clang compiles __builtin_parity to: > > define i32 @f(i32 %x) nounwind readnone ssp { > entry: > %tmp1 = tail call i32 @llvm.ctpop.i32(i32 %x) > %tmp2 = and i32 %tmp1, 1 > ret i32 %tmp2 > } I found an old PR1983 covering this. I don't know if there would be any advantage to a parity intrinsic over the current ctpop(x)&1 representation. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/5de7e773/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: smime.p7s Type: application/pkcs7-signature Size: 1929 bytes Desc: not available Url : http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/5de7e773/attachment.bin From sabre at nondot.org Sat Jan 15 23:08:00 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 05:08:00 -0000 Subject: [llvm-commits] [llvm] r123565 - /llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Message-ID: <20110116050801.0221E2A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 23:08:00 2011 New Revision: 123565 URL: http://llvm.org/viewvc/llvm-project?rev=123565&view=rev Log: more cleanups: use the IR builder. Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp?rev=123565&r1=123564&r2=123565&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Sat Jan 15 23:08:00 2011 @@ -543,6 +543,12 @@ if (NonConstBB) return 0; // More than one non-const value. NonConstBB = PN->getIncomingBlock(i); + + // If the InVal is an invoke at the end of the pred block, then we can't + // insert a computation after it without breaking the edge. + if (InvokeInst *II = dyn_cast(InVal)) + if (II->getParent() == NonConstBB) + return 0; // If the incoming non-constant value is in I's block, we have an infinite // loop. @@ -564,7 +570,12 @@ NewPN->reserveOperandSpace(PN->getNumOperands()/2); InsertNewInstBefore(NewPN, *PN); NewPN->takeName(PN); - + + // If we are going to have to insert a new computation, do so right before the + // predecessors terminator. + if (NonConstBB) + Builder->SetInsertPoint(NonConstBB->getTerminator()); + // Next, add all of the operands to the PHI. if (SelectInst *SI = dyn_cast(&I)) { // We only currently try to fold the condition of a select when it is a phi, @@ -577,42 +588,36 @@ Value *TrueVInPred = TrueV->DoPHITranslation(PhiTransBB, ThisBB); Value *FalseVInPred = FalseV->DoPHITranslation(PhiTransBB, ThisBB); Value *InV = 0; - if (Constant *InC = dyn_cast(PN->getIncomingValue(i))) { + if (Constant *InC = dyn_cast(PN->getIncomingValue(i))) InV = InC->isNullValue() ? FalseVInPred : TrueVInPred; - } else { - assert(PN->getIncomingBlock(i) == NonConstBB); - InV = SelectInst::Create(PN->getIncomingValue(i), TrueVInPred, - FalseVInPred, - "phitmp", NonConstBB->getTerminator()); - Worklist.Add(cast(InV)); - } + else + InV = Builder->CreateSelect(PN->getIncomingValue(i), + TrueVInPred, FalseVInPred, "phitmp"); NewPN->addIncoming(InV, ThisBB); } + } else if (CmpInst *CI = dyn_cast(&I)) { + Constant *C = cast(I.getOperand(1)); + for (unsigned i = 0; i != NumPHIValues; ++i) { + Value *InV = 0; + if (Constant *InC = dyn_cast(PN->getIncomingValue(i))) + InV = ConstantExpr::getCompare(CI->getPredicate(), InC, C); + else if (isa(CI)) + InV = Builder->CreateICmp(CI->getPredicate(), PN->getIncomingValue(i), + C, "phitmp"); + else + InV = Builder->CreateFCmp(CI->getPredicate(), PN->getIncomingValue(i), + C, "phitmp"); + NewPN->addIncoming(InV, PN->getIncomingBlock(i)); + } } else if (I.getNumOperands() == 2) { Constant *C = cast(I.getOperand(1)); for (unsigned i = 0; i != NumPHIValues; ++i) { Value *InV = 0; - if (Constant *InC = dyn_cast(PN->getIncomingValue(i))) { - if (CmpInst *CI = dyn_cast(&I)) - InV = ConstantExpr::getCompare(CI->getPredicate(), InC, C); - else - InV = ConstantExpr::get(I.getOpcode(), InC, C); - } else { - assert(PN->getIncomingBlock(i) == NonConstBB); - if (BinaryOperator *BO = dyn_cast(&I)) - InV = BinaryOperator::Create(BO->getOpcode(), - PN->getIncomingValue(i), C, "phitmp", - NonConstBB->getTerminator()); - else if (CmpInst *CI = dyn_cast(&I)) - InV = CmpInst::Create(CI->getOpcode(), - CI->getPredicate(), - PN->getIncomingValue(i), C, "phitmp", - NonConstBB->getTerminator()); - else - llvm_unreachable("Unknown binop!"); - - Worklist.Add(cast(InV)); - } + if (Constant *InC = dyn_cast(PN->getIncomingValue(i))) + InV = ConstantExpr::get(I.getOpcode(), InC, C); + else + InV = Builder->CreateBinOp(cast(I).getOpcode(), + PN->getIncomingValue(i), C, "phitmp"); NewPN->addIncoming(InV, PN->getIncomingBlock(i)); } } else { @@ -620,15 +625,11 @@ const Type *RetTy = CI->getType(); for (unsigned i = 0; i != NumPHIValues; ++i) { Value *InV; - if (Constant *InC = dyn_cast(PN->getIncomingValue(i))) { + if (Constant *InC = dyn_cast(PN->getIncomingValue(i))) InV = ConstantExpr::getCast(CI->getOpcode(), InC, RetTy); - } else { - assert(PN->getIncomingBlock(i) == NonConstBB); - InV = CastInst::Create(CI->getOpcode(), PN->getIncomingValue(i), - I.getType(), "phitmp", - NonConstBB->getTerminator()); - Worklist.Add(cast(InV)); - } + else + InV = Builder->CreateCast(CI->getOpcode(), + PN->getIncomingValue(i), I.getType(), "phitmp"); NewPN->addIncoming(InV, PN->getIncomingBlock(i)); } } From clattner at apple.com Sat Jan 15 23:15:17 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 21:15:17 -0800 Subject: [llvm-commits] [llvm] r123547 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp In-Reply-To: <56D7F26B-3638-423A-950A-CD16B6A359B2@2pi.dk> References: <20110115203031.19E282A6C12C@llvm.org> <56D7F26B-3638-423A-950A-CD16B6A359B2@2pi.dk> Message-ID: <1AF61B7B-05E7-46D7-AB9D-D488EB93851F@apple.com> On Jan 15, 2011, at 8:55 PM, Jakob Stoklund Olesen wrote: > > On Jan 15, 2011, at 6:51 PM, Jakob Stoklund Olesen wrote: > >> >> On Jan 15, 2011, at 5:02 PM, Chris Lattner wrote: >> >>> One major advantage of recognizing popcount though is that the optimizer has a chance of hacking on it, and there are a lot of instcombine xforms that we could do. Here are some that I note looking at the bc file for crafy (when hacked to use builtin_popcount). I attached the bc file below if you're interested. >> >> I noticed that we have intrinsics and ISD nodes for ctlz, cttz, and ctpop, but none for parity. Is that on purpose? >> >> Currently, clang compiles __builtin_parity to: >> >> define i32 @f(i32 %x) nounwind readnone ssp { >> entry: >> %tmp1 = tail call i32 @llvm.ctpop.i32(i32 %x) >> %tmp2 = and i32 %tmp1, 1 >> ret i32 %tmp2 >> } > > I found an old PR1983 covering this. > > I don't know if there would be any advantage to a parity intrinsic over the current ctpop(x)&1 representation. I think that representing it as ctpop(x)&1 in IR is reasonable, but you're right that codegen should turn that into an ISD::PARITY node if it can lower it to something better. -Chris -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.cs.uiuc.edu/pipermail/llvm-commits/attachments/20110115/8592abee/attachment.html From sabre at nondot.org Sat Jan 15 23:14:26 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 05:14:26 -0000 Subject: [llvm-commits] [llvm] r123566 - in /llvm/trunk/lib/Transforms/InstCombine: InstCombine.h InstCombineCompares.cpp InstructionCombining.cpp Message-ID: <20110116051426.444462A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 23:14:26 2011 New Revision: 123566 URL: http://llvm.org/viewvc/llvm-project?rev=123566&view=rev Log: remove the AllowAggressive argument to FoldOpIntoPhi. It is forced to false in the first line of the function because it isn't a good idea, even for compares. Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombine.h llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombine.h URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombine.h?rev=123566&r1=123565&r2=123566&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstCombine.h (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstCombine.h Sat Jan 15 23:14:26 2011 @@ -319,10 +319,7 @@ // into the PHI (which is only possible if all operands to the PHI are // constants). // - // If AllowAggressive is true, FoldOpIntoPhi will allow certain transforms - // that would normally be unprofitable because they strongly encourage jump - // threading. - Instruction *FoldOpIntoPhi(Instruction &I, bool AllowAggressive = false); + Instruction *FoldOpIntoPhi(Instruction &I); // FoldPHIArgOpIntoPHI - If all operands to a PHI node are the same "unary" // operator and they all are only used by the PHI, PHI together their Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp?rev=123566&r1=123565&r2=123566&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstCombineCompares.cpp Sat Jan 15 23:14:26 2011 @@ -2125,7 +2125,7 @@ // block. If in the same block, we're encouraging jump threading. If // not, we are just pessimizing the code by making an i1 phi. if (LHSI->getParent() == I.getParent()) - if (Instruction *NV = FoldOpIntoPhi(I, true)) + if (Instruction *NV = FoldOpIntoPhi(I)) return NV; break; case Instruction::Select: { @@ -2648,7 +2648,7 @@ // block. If in the same block, we're encouraging jump threading. If // not, we are just pessimizing the code by making an i1 phi. if (LHSI->getParent() == I.getParent()) - if (Instruction *NV = FoldOpIntoPhi(I, true)) + if (Instruction *NV = FoldOpIntoPhi(I)) return NV; break; case Instruction::SIToFP: Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp?rev=123566&r1=123565&r2=123566&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Sat Jan 15 23:14:26 2011 @@ -512,12 +512,7 @@ /// has a PHI node as operand #0, see if we can fold the instruction into the /// PHI (which is only possible if all operands to the PHI are constants). /// -/// If AllowAggressive is true, FoldOpIntoPhi will allow certain transforms -/// that would normally be unprofitable because they strongly encourage jump -/// threading. -Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I, - bool AllowAggressive) { - AllowAggressive = false; +Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I) { PHINode *PN = cast(I.getOperand(0)); unsigned NumPHIValues = PN->getNumIncomingValues(); if (NumPHIValues == 0) @@ -525,7 +520,7 @@ // We normally only transform phis with a single use, unless we're trying // hard to make jump threading happen. - if (!PN->hasOneUse() && !AllowAggressive) + if (!PN->hasOneUse()) return 0; // Check to see if all of the operands of the PHI are simple constants @@ -560,7 +555,7 @@ // operation in that block. However, if this is a critical edge, we would be // inserting the computation one some other paths (e.g. inside a loop). Only // do this if the pred block is unconditionally branching into the phi block. - if (NonConstBB != 0 && !AllowAggressive) { + if (NonConstBB != 0) { BranchInst *BI = dyn_cast(NonConstBB->getTerminator()); if (!BI || !BI->isUnconditional()) return 0; } From evan.cheng at apple.com Sat Jan 15 23:14:33 2011 From: evan.cheng at apple.com (Evan Cheng) Date: Sun, 16 Jan 2011 05:14:33 -0000 Subject: [llvm-commits] [llvm] r123567 - in /llvm/trunk: lib/Target/ARM/ARMFrameLowering.cpp test/CodeGen/Thumb/2011-EpilogueBug.ll Message-ID: <20110116051433.E89C62A6C12C@llvm.org> Author: evancheng Date: Sat Jan 15 23:14:33 2011 New Revision: 123567 URL: http://llvm.org/viewvc/llvm-project?rev=123567&view=rev Log: Spill R4 if it's going to be used to restore SP from FP. Added: llvm/trunk/test/CodeGen/Thumb/2011-EpilogueBug.ll Modified: llvm/trunk/lib/Target/ARM/ARMFrameLowering.cpp Modified: llvm/trunk/lib/Target/ARM/ARMFrameLowering.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMFrameLowering.cpp?rev=123567&r1=123566&r2=123567&view=diff ============================================================================== --- llvm/trunk/lib/Target/ARM/ARMFrameLowering.cpp (original) +++ llvm/trunk/lib/Target/ARM/ARMFrameLowering.cpp Sat Jan 15 23:14:33 2011 @@ -801,15 +801,23 @@ // Spill R4 if Thumb2 function requires stack realignment - it will be used as // scratch register. Also spill R4 if Thumb2 function has varsized objects, - // since it's always posible to restore sp from fp in a single instruction. + // since it's not always possible to restore sp from fp in a single + // instruction. // FIXME: It will be better just to find spare register here. if (AFI->isThumb2Function() && (MFI->hasVarSizedObjects() || RegInfo->needsStackRealignment(MF))) MF.getRegInfo().setPhysRegUsed(ARM::R4); - // Spill LR if Thumb1 function uses variable length argument lists. - if (AFI->isThumb1OnlyFunction() && AFI->getVarArgsRegSaveSize() > 0) - MF.getRegInfo().setPhysRegUsed(ARM::LR); + if (AFI->isThumb1OnlyFunction()) { + // Spill LR if Thumb1 function uses variable length argument lists. + if (AFI->getVarArgsRegSaveSize() > 0) + MF.getRegInfo().setPhysRegUsed(ARM::LR); + + // Spill R4 if Thumb1 epilogue has to restore SP from FP since + // FIXME: It will be better just to find spare register here. + if (MFI->hasVarSizedObjects()) + MF.getRegInfo().setPhysRegUsed(ARM::R4); + } // Spill the BasePtr if it's used. if (RegInfo->hasBasePointer(MF)) Added: llvm/trunk/test/CodeGen/Thumb/2011-EpilogueBug.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/Thumb/2011-EpilogueBug.ll?rev=123567&view=auto ============================================================================== --- llvm/trunk/test/CodeGen/Thumb/2011-EpilogueBug.ll (added) +++ llvm/trunk/test/CodeGen/Thumb/2011-EpilogueBug.ll Sat Jan 15 23:14:33 2011 @@ -0,0 +1,17 @@ +; RUN: llc -mtriple=thumbv6-apple-darwin < %s | FileCheck %s +; r8869722 + +%struct.state = type { i32, %struct.info*, float**, i32, i32, i32, i32, i32, i32, i32, i32, i32, i64, i64, i64, i64, i64, i64, i8* } +%struct.info = type { i32, i32, i32, i32, i32, i32, i32, i8* } + +define void @t1(%struct.state* %v) { +; CHECK: push {r4 + %tmp6 = load i32* null + %tmp8 = alloca float, i32 %tmp6 + store i32 1, i32* null + br label %return + +return: ; preds = %0 +; CHECK: mov sp, r4 + ret void +} From sabre at nondot.org Sat Jan 15 23:29:00 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 05:29:00 -0000 Subject: [llvm-commits] [llvm] r123568 - in /llvm/trunk: lib/Transforms/InstCombine/InstructionCombining.cpp lib/Transforms/Scalar/ScalarReplAggregates.cpp test/Transforms/InstCombine/phi.ll Message-ID: <20110116052900.1CA9F2A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 23:28:59 2011 New Revision: 123568 URL: http://llvm.org/viewvc/llvm-project?rev=123568&view=rev Log: enhance FoldOpIntoPhi in instcombine to try harder when a phi has multiple uses. In some cases, all the uses are the same operation, so instcombine can go ahead and promote the phi. In the testcase this pushes an add out of the loop. Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp llvm/trunk/test/Transforms/InstCombine/phi.ll Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp?rev=123568&r1=123567&r2=123568&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Sat Jan 15 23:28:59 2011 @@ -519,9 +519,17 @@ return 0; // We normally only transform phis with a single use, unless we're trying - // hard to make jump threading happen. - if (!PN->hasOneUse()) - return 0; + // hard to make jump threading happen. However, if a PHI has multiple uses + // and they are all the same operation, we can fold *all* of the uses into the + // PHI. + if (!PN->hasOneUse()) { + // Walk the use list for the instruction, comparing them to I. + for (Value::use_iterator UI = PN->use_begin(), E = PN->use_end(); + UI != E; ++UI) + if (!I.isIdenticalTo(cast(*UI))) + return 0; + // Otherwise, we can replace *all* users with the new PHI we form. + } // Check to see if all of the operands of the PHI are simple constants // (constantint/constantfp/undef). If there is one non-constant value, @@ -628,6 +636,14 @@ NewPN->addIncoming(InV, PN->getIncomingBlock(i)); } } + + for (Value::use_iterator UI = PN->use_begin(), E = PN->use_end(); + UI != E; ) { + Instruction *User = cast(*UI++); + if (User == &I) continue; + ReplaceInstUsesWith(*User, NewPN); + EraseInstFromFunction(*User); + } return ReplaceInstUsesWith(I, NewPN); } Modified: llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp?rev=123568&r1=123567&r2=123568&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp Sat Jan 15 23:28:59 2011 @@ -1279,6 +1279,7 @@ // address operand will be updated, so nothing else needs to be done. } else if (LoadInst *LI = dyn_cast(User)) { const Type *LIType = LI->getType(); + if (isCompatibleAggregate(LIType, AI->getAllocatedType())) { // Replace: // %res = load { i32, i32 }* %alloc Modified: llvm/trunk/test/Transforms/InstCombine/phi.ll URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/phi.ll?rev=123568&r1=123567&r2=123568&view=diff ============================================================================== --- llvm/trunk/test/Transforms/InstCombine/phi.ll (original) +++ llvm/trunk/test/Transforms/InstCombine/phi.ll Sat Jan 15 23:28:59 2011 @@ -503,3 +503,25 @@ ret: ret void } + +define i32 @test23(i32 %A, i1 %b, i32 * %P) { +BB0: + br label %Loop + +Loop: ; preds = %Loop, %BB0 + ; PHI has same value always. + %B = phi i32 [ %A, %BB0 ], [ 42, %Loop ] + %D = add i32 %B, 19 + store i32 %D, i32* %P + br i1 %b, label %Loop, label %Exit + +Exit: ; preds = %Loop + %E = add i32 %B, 19 + ret i32 %E +; CHECK: @test23 +; CHECK: %phitmp = add i32 %A, 19 +; CHECK: Loop: +; CHECK-NEXT: %B = phi i32 [ %phitmp, %BB0 ], [ 61, %Loop ] +; CHECK: Exit: +; CHECK-NEXT: ret i32 %B +} From sabre at nondot.org Sat Jan 15 23:37:55 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 05:37:55 -0000 Subject: [llvm-commits] [llvm] r123569 - /llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Message-ID: <20110116053755.2FCB52A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 23:37:55 2011 New Revision: 123569 URL: http://llvm.org/viewvc/llvm-project?rev=123569&view=rev Log: remove a dead check, this was needed before we had an explicit veto on uses of phis. Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp?rev=123569&r1=123568&r2=123569&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp (original) +++ llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Sat Jan 15 23:37:55 2011 @@ -552,11 +552,6 @@ if (InvokeInst *II = dyn_cast(InVal)) if (II->getParent() == NonConstBB) return 0; - - // If the incoming non-constant value is in I's block, we have an infinite - // loop. - if (NonConstBB == I.getParent()) - return 0; } // If there is exactly one non-constant value, we can insert a copy of the From clattner at apple.com Sat Jan 15 23:48:34 2011 From: clattner at apple.com (Chris Lattner) Date: Sat, 15 Jan 2011 21:48:34 -0800 Subject: [llvm-commits] [llvm] r123563 - in /llvm/trunk: lib/Transforms/IPO/GlobalOpt.cpp test/Transforms/GlobalOpt/static-ctor-bitcast.ll In-Reply-To: <20110116043333.82FA32A6C12C@llvm.org> References: <20110116043333.82FA32A6C12C@llvm.org> Message-ID: <6A5A4105-9D29-43F7-B76C-CC25FBC51358@apple.com> On Jan 15, 2011, at 8:33 PM, Owen Anderson wrote: > Author: resistor > Date: Sat Jan 15 22:33:33 2011 > New Revision: 123563 > > URL: http://llvm.org/viewvc/llvm-project?rev=123563&view=rev > Log: > Improve the safety of my globalopt enhancement by ensuring that the bitcast > of the stored value to the new store type is always. Also, add a testcase. Thanks Owen, this patch looks correct. Please hand reduce the testcase and add it as an entry in GlobalOpt/ctor-list-opt.ll. -Chris From sabre at nondot.org Sat Jan 15 23:58:24 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 05:58:24 -0000 Subject: [llvm-commits] [llvm] r123570 - /llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp Message-ID: <20110116055824.659BC2A6C12C@llvm.org> Author: lattner Date: Sat Jan 15 23:58:24 2011 New Revision: 123570 URL: http://llvm.org/viewvc/llvm-project?rev=123570&view=rev Log: Use an irbuilder to get some trivial constant folding when doing a store of a constant. Modified: llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp Modified: llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp?rev=123570&r1=123569&r2=123570&view=diff ============================================================================== --- llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/ScalarReplAggregates.cpp Sat Jan 15 23:58:24 2011 @@ -1601,11 +1601,12 @@ const Type *AllocaEltTy = AI->getAllocatedType(); uint64_t AllocaSizeBits = TD->getTypeAllocSizeInBits(AllocaEltTy); + IRBuilder<> Builder(SI); + // Handle tail padding by extending the operand if (TD->getTypeSizeInBits(SrcVal->getType()) != AllocaSizeBits) - SrcVal = new ZExtInst(SrcVal, - IntegerType::get(SI->getContext(), AllocaSizeBits), - "", SI); + SrcVal = Builder.CreateZExt(SrcVal, + IntegerType::get(SI->getContext(), AllocaSizeBits)); DEBUG(dbgs() << "PROMOTING STORE TO WHOLE ALLOCA: " << *AI << '\n' << *SI << '\n'); @@ -1626,8 +1627,7 @@ Value *EltVal = SrcVal; if (Shift) { Value *ShiftVal = ConstantInt::get(EltVal->getType(), Shift); - EltVal = BinaryOperator::CreateLShr(EltVal, ShiftVal, - "sroa.store.elt", SI); + EltVal = Builder.CreateLShr(EltVal, ShiftVal, "sroa.store.elt"); } // Truncate down to an integer of the right size. @@ -1637,20 +1637,18 @@ if (FieldSizeBits == 0) continue; if (FieldSizeBits != AllocaSizeBits) - EltVal = new TruncInst(EltVal, - IntegerType::get(SI->getContext(), FieldSizeBits), - "", SI); + EltVal = Builder.CreateTrunc(EltVal, + IntegerType::get(SI->getContext(), FieldSizeBits)); Value *DestField = NewElts[i]; if (EltVal->getType() == FieldTy) { // Storing to an integer field of this size, just do it. } else if (FieldTy->isFloatingPointTy() || FieldTy->isVectorTy()) { // Bitcast to the right element type (for fp/vector values). - EltVal = new BitCastInst(EltVal, FieldTy, "", SI); + EltVal = Builder.CreateBitCast(EltVal, FieldTy); } else { // Otherwise, bitcast the dest pointer (for aggregates). - DestField = new BitCastInst(DestField, - PointerType::getUnqual(EltVal->getType()), - "", SI); + DestField = Builder.CreateBitCast(DestField, + PointerType::getUnqual(EltVal->getType())); } new StoreInst(EltVal, DestField, SI); } @@ -1675,27 +1673,25 @@ Value *EltVal = SrcVal; if (Shift) { Value *ShiftVal = ConstantInt::get(EltVal->getType(), Shift); - EltVal = BinaryOperator::CreateLShr(EltVal, ShiftVal, - "sroa.store.elt", SI); + EltVal = Builder.CreateLShr(EltVal, ShiftVal, "sroa.store.elt"); } // Truncate down to an integer of the right size. if (ElementSizeBits != AllocaSizeBits) - EltVal = new TruncInst(EltVal, - IntegerType::get(SI->getContext(), - ElementSizeBits), "", SI); + EltVal = Builder.CreateTrunc(EltVal, + IntegerType::get(SI->getContext(), + ElementSizeBits)); Value *DestField = NewElts[i]; if (EltVal->getType() == ArrayEltTy) { // Storing to an integer field of this size, just do it. } else if (ArrayEltTy->isFloatingPointTy() || ArrayEltTy->isVectorTy()) { // Bitcast to the right element type (for fp/vector values). - EltVal = new BitCastInst(EltVal, ArrayEltTy, "", SI); + EltVal = Builder.CreateBitCast(EltVal, ArrayEltTy); } else { // Otherwise, bitcast the dest pointer (for aggregates). - DestField = new BitCastInst(DestField, - PointerType::getUnqual(EltVal->getType()), - "", SI); + DestField = Builder.CreateBitCast(DestField, + PointerType::getUnqual(EltVal->getType())); } new StoreInst(EltVal, DestField, SI); From sabre at nondot.org Sun Jan 16 00:18:29 2011 From: sabre at nondot.org (Chris Lattner) Date: Sun, 16 Jan 2011 06:18:29 -0000 Subject: [llvm-commits] [llvm] r123571 - in /llvm/trunk: lib/Transforms/Scalar/ScalarReplAggregates.cpp test/Transforms/ScalarRepl/basictest.ll Message-ID: <20110116061829.1E0CB2A6C12C@llvm.org> Author: lattner Date: Sun Jan 16 00:18:28 2011 New Revision: 123571 URL: http://llvm.org/viewvc/llvm-project?rev=123571&view=rev Log: if an alloca is only ever accessed as a unit, and is accessed with load/store instructions, then don't try to decimate it into its indi

Graduate Students

Undergraduate Students

Graduate Student Alumni

Undergraduate Student Alumni

Undergraduate Student Alumni