From ashukla at cs.uiuc.edu  Mon Feb 17 13:46:01 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Mon Feb 17 13:46:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Reoptimizer/Trigger/GetTraceTime.cpp
Message-ID: <200302171945.NAA25341@tank.cs.uiuc.edu>


Changes in directory llvm/lib/Reoptimizer/Trigger:

GetTraceTime.cpp updated: 1.1 -> 1.2

---
Log message:

Fixed compilation with linux

---
Diffs of the changes:

Index: llvm/lib/Reoptimizer/Trigger/GetTraceTime.cpp
diff -u llvm/lib/Reoptimizer/Trigger/GetTraceTime.cpp:1.1 llvm/lib/Reoptimizer/Trigger/GetTraceTime.cpp:1.2
--- llvm/lib/Reoptimizer/Trigger/GetTraceTime.cpp:1.1	Fri Feb 14 14:48:18 2003
+++ llvm/lib/Reoptimizer/Trigger/GetTraceTime.cpp	Mon Feb 17 13:45:22 2003
@@ -7,8 +7,11 @@
 //
 //===----------------------------------------------------------------------===//
 
+#ifdef __sparc__
 #include "llvm/Reoptimizer/GetTraceTime.h"
 
+#include <libcpc.h>
+
 #ifdef GET_TRACE_TIME
 
 extern "C" void llvm_time_start(){
@@ -29,5 +32,7 @@
 void llvm_trigger_time_end(){
   cpc_count_usr_events(0);
 }
+
+#endif
 
 #endif




From ashukla at cs.uiuc.edu  Mon Feb 17 13:46:03 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Mon Feb 17 13:46:03 2003
Subject: [llvm-commits] CVS: llvm/lib/Reoptimizer/Trigger/TriggerAuxillary.cpp
Message-ID: <200302171945.NAA25356@tank.cs.uiuc.edu>


Changes in directory llvm/lib/Reoptimizer/Trigger:

TriggerAuxillary.cpp updated: 1.1 -> 1.2

---
Log message:

Fixed compilation with linux

---
Diffs of the changes:

Index: llvm/lib/Reoptimizer/Trigger/TriggerAuxillary.cpp
diff -u llvm/lib/Reoptimizer/Trigger/TriggerAuxillary.cpp:1.1 llvm/lib/Reoptimizer/Trigger/TriggerAuxillary.cpp:1.2
--- llvm/lib/Reoptimizer/Trigger/TriggerAuxillary.cpp:1.1	Fri Feb 14 14:47:28 2003
+++ llvm/lib/Reoptimizer/Trigger/TriggerAuxillary.cpp	Mon Feb 17 13:45:34 2003
@@ -132,6 +132,7 @@
 
 
 //get a path vector from a BB*
+/*
 unsigned int getPathVector(vector<BasicBlock *> &vBB){
   int i = 0, path = 0;
   for(vector<BasicBlock *>::iterator BI = vBB.begin(), BE = vBB.end(); BI != BE;
@@ -152,3 +153,4 @@
     
   }
 }
+*/




From ashukla at cs.uiuc.edu  Mon Feb 17 13:47:02 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Mon Feb 17 13:47:02 2003
Subject: [llvm-commits] CVS: llvm/include/llvm/Reoptimizer/GetTraceTime.h
Message-ID: <200302171946.NAA25376@tank.cs.uiuc.edu>


Changes in directory llvm/include/llvm/Reoptimizer:

GetTraceTime.h updated: 1.2 -> 1.3

---
Log message:

Fixed compilation with linux

---
Diffs of the changes:

Index: llvm/include/llvm/Reoptimizer/GetTraceTime.h
diff -u llvm/include/llvm/Reoptimizer/GetTraceTime.h:1.2 llvm/include/llvm/Reoptimizer/GetTraceTime.h:1.3
--- llvm/include/llvm/Reoptimizer/GetTraceTime.h:1.2	Fri Feb 14 14:42:59 2003
+++ llvm/include/llvm/Reoptimizer/GetTraceTime.h	Mon Feb 17 13:46:01 2003
@@ -9,7 +9,8 @@
 
 #ifndef LLVM_REOPTIMIZER_TRACECACHE_GETTRACETIME_H
 #define LLVM_REOPTIMIZER_TRACECACHE_GETTRACETIME_H
-#include <libcpc.h>
+
+#ifdef __sparc
 
 //#define GET_TRACE_TIME
 //#define MAP_TIME
@@ -28,5 +29,6 @@
 void llvm_trigger_time_end();
 #endif
 
+#endif
 
 #endif




From ashukla at cs.uiuc.edu  Mon Feb 17 13:51:01 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Mon Feb 17 13:51:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Reoptimizer/TraceCache/TraceCache2.cpp
Message-ID: <200302171950.NAA14012@trinity.cs.uiuc.edu>


Changes in directory llvm/lib/Reoptimizer/TraceCache:

TraceCache2.cpp added (r1.1)

---
Log message:

Alternate tracecache and memory manager

---
Diffs of the changes:



From ashukla at cs.uiuc.edu  Mon Feb 17 13:53:01 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Mon Feb 17 13:53:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Reoptimizer/TraceCache/MemoryManager2.cpp
Message-ID: <200302171952.NAA14026@trinity.cs.uiuc.edu>


Changes in directory llvm/lib/Reoptimizer/TraceCache:

MemoryManager2.cpp added (r1.1)

---
Log message:

Alternate tracecache and memory manager

---
Diffs of the changes:



From ashukla at cs.uiuc.edu  Mon Feb 17 13:53:02 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Mon Feb 17 13:53:02 2003
Subject: [llvm-commits] CVS: llvm/lib/Reoptimizer/TraceCache/MemoryManager2.h
Message-ID: <200302171952.NAA14038@trinity.cs.uiuc.edu>


Changes in directory llvm/lib/Reoptimizer/TraceCache:

MemoryManager2.h added (r1.1)

---
Log message:

Alternate tracecache and memory manager

---
Diffs of the changes:



From ashukla at cs.uiuc.edu  Mon Feb 17 13:54:00 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Mon Feb 17 13:54:00 2003
Subject: [llvm-commits] CVS: llvm/include/llvm/Reoptimizer/TraceCache2.h
Message-ID: <200302171953.NAA14053@trinity.cs.uiuc.edu>


Changes in directory llvm/include/llvm/Reoptimizer:

TraceCache2.h added (r1.1)

---
Log message:

Alternate tracecache and memory manager

---
Diffs of the changes:



From lattner at cs.uiuc.edu  Tue Feb 18 12:06:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 12:06:01 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/InstCombine/or.ll
Message-ID: <200302181805.MAA01142@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/InstCombine:

or.ll updated: 1.6 -> 1.7

---
Log message:

Modernize testcase


---
Diffs of the changes:

Index: llvm/test/Regression/Transforms/InstCombine/or.ll
diff -u llvm/test/Regression/Transforms/InstCombine/or.ll:1.6 llvm/test/Regression/Transforms/InstCombine/or.ll:1.7
--- llvm/test/Regression/Transforms/InstCombine/or.ll:1.6	Tue Oct  8 11:09:47 2002
+++ llvm/test/Regression/Transforms/InstCombine/or.ll	Tue Feb 18 12:05:19 2003
@@ -8,52 +8,52 @@
 
 implementation
 
-int "test1"(int %A) {
+int %test1(int %A) {
 	%B = or int %A, 0
 	ret int %B
 }
 
-int "test2"(int %A) {
+int %test2(int %A) {
 	%B = or int %A, -1
 	ret int %B
 }
 
-bool "test3"(bool %A) {
+bool %test3(bool %A) {
 	%B = or bool %A, false
 	ret bool %B
 }
 
-bool "test4"(bool %A) {
+bool %test4(bool %A) {
 	%B = or bool %A, true
 	ret bool %B
 }
 
-bool "test5"(bool %A) {
+bool %test5(bool %A) {
 	%B = xor bool %A, false
 	ret bool %B
 }
 
-int "test6"(int %A) {
+int %test6(int %A) {
 	%B = xor int %A, 0
 	ret int %B
 }
 
-bool "test7"(bool %A) {
+bool %test7(bool %A) {
 	%B = xor bool %A, %A
 	ret bool %B
 }
 
-int "test8"(int %A) {
+int %test8(int %A) {
 	%B = xor int %A, %A
 	ret int %B
 }
 
-bool "test9"(bool %A) {
+bool %test9(bool %A) {
 	%B = or bool %A, %A
 	ret bool %B
 }
 
-int "test10"(int %A) {
+int %test10(int %A) {
 	%B = or int %A, %A
 	ret int %B
 }




From lattner at cs.uiuc.edu  Tue Feb 18 12:07:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 12:07:01 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/InstCombine/and.ll
Message-ID: <200302181806.MAA01197@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/InstCombine:

and.ll updated: 1.3 -> 1.4

---
Log message:

Modernize testcase


---
Diffs of the changes:

Index: llvm/test/Regression/Transforms/InstCombine/and.ll
diff -u llvm/test/Regression/Transforms/InstCombine/and.ll:1.3 llvm/test/Regression/Transforms/InstCombine/and.ll:1.4
--- llvm/test/Regression/Transforms/InstCombine/and.ll:1.3	Fri Aug  2 14:27:57 2002
+++ llvm/test/Regression/Transforms/InstCombine/and.ll	Tue Feb 18 12:06:44 2003
@@ -8,32 +8,32 @@
 
 implementation
 
-int "test1"(int %A) {
+int %test1(int %A) {
 	%B = and int %A, 0     ; zero result
 	ret int %B
 }
 
-int "test2"(int %A) {
+int %test2(int %A) {
 	%B = and int %A, -1    ; noop
 	ret int %B
 }
 
-bool "test3"(bool %A) {
+bool %test3(bool %A) {
 	%B = and bool %A, false  ; always = false
 	ret bool %B
 }
 
-bool "test4"(bool %A) {
+bool %test4(bool %A) {
 	%B = and bool %A, true  ; noop
 	ret bool %B
 }
 
-int "test5"(int %A) {
+int %test5(int %A) {
 	%B = and int %A, %A
 	ret int %B
 }
 
-bool "test6"(bool %A) {
+bool %test6(bool %A) {
 	%B = and bool %A, %A
 	ret bool %B
 }




From lattner at cs.uiuc.edu  Tue Feb 18 13:17:02 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 13:17:02 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/InstCombine/div.ll
Message-ID: <200302181916.NAA02270@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/InstCombine:

div.ll updated: 1.2 -> 1.3

---
Log message:

modernize testcase


---
Diffs of the changes:

Index: llvm/test/Regression/Transforms/InstCombine/div.ll
diff -u llvm/test/Regression/Transforms/InstCombine/div.ll:1.2 llvm/test/Regression/Transforms/InstCombine/div.ll:1.3
--- llvm/test/Regression/Transforms/InstCombine/div.ll:1.2	Fri Aug  2 14:27:57 2002
+++ llvm/test/Regression/Transforms/InstCombine/div.ll	Tue Feb 18 13:16:45 2003
@@ -8,9 +8,7 @@
 
 implementation
 
-int "test1"(int %A)
-begin
+int %test1(int %A) {
 	%B = div int %A, 1
 	ret int %B
-end
-
+}




From lattner at cs.uiuc.edu  Tue Feb 18 13:29:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 13:29:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Transforms/Scalar/InstructionCombining.cpp
Message-ID: <200302181928.NAA02712@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Transforms/Scalar:

InstructionCombining.cpp updated: 1.67 -> 1.68

---
Log message:

Add a variety of new transformations:
  * A & ~A == 0
  * A / (2^c) == A >> c  if unsigned
  * 0 / A == 0
  * 1.0 * A == A
  * A * (2^c) == A << c
  * A ^ ~A == -1
  * A | ~A == -1
  * 0 % X = 0
  * A % (2^c) == A & (c-1) if unsigned
  * A - (A & B) == A & ~B
  * -1 - A == ~A



---
Diffs of the changes:

Index: llvm/lib/Transforms/Scalar/InstructionCombining.cpp
diff -u llvm/lib/Transforms/Scalar/InstructionCombining.cpp:1.67 llvm/lib/Transforms/Scalar/InstructionCombining.cpp:1.68
--- llvm/lib/Transforms/Scalar/InstructionCombining.cpp:1.67	Thu Dec  5 16:41:53 2002
+++ llvm/lib/Transforms/Scalar/InstructionCombining.cpp	Tue Feb 18 13:28:33 2003
@@ -123,22 +123,27 @@
 // instruction if the LHS is a constant zero (which is the 'negate' form).
 //
 static inline Value *dyn_castNegInst(Value *V) {
-  Instruction *I = dyn_cast<Instruction>(V);
-  if (!I || I->getOpcode() != Instruction::Sub) return 0;
-
-  if (I->getOperand(0) == Constant::getNullValue(I->getType()))
-    return I->getOperand(1);
-  return 0;
+  return BinaryOperator::isNeg(V) ?
+    BinaryOperator::getNegArgument(cast<BinaryOperator>(V)) : 0;
 }
 
 static inline Value *dyn_castNotInst(Value *V) {
-  Instruction *I = dyn_cast<Instruction>(V);
-  if (!I || I->getOpcode() != Instruction::Xor) return 0;
+  return BinaryOperator::isNot(V) ?
+    BinaryOperator::getNotArgument(cast<BinaryOperator>(V)) : 0;
+}
 
-  if (ConstantIntegral *CI = dyn_cast<ConstantIntegral>(I->getOperand(1)))
-    if (CI->isAllOnesValue())
-      return I->getOperand(0);
-  return 0;
+
+// Log2 - Calculate the log base 2 for the specified value if it is exactly a
+// power of 2.
+static unsigned Log2(uint64_t Val) {
+  assert(Val > 1 && "Values 0 and 1 should be handled elsewhere!");
+  unsigned Count = 0;
+  while (Val != 1) {
+    if (Val & 1) return 0;    // Multiple bits set?
+    Val >>= 1;
+    ++Count;
+  }
+  return Count;
 }
 
 Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
@@ -197,56 +202,119 @@
   if (Value *V = dyn_castNegInst(Op1))
     return BinaryOperator::create(Instruction::Add, Op0, V);
 
-  // Replace (x - (y - z)) with (x + (z - y)) if the (y - z) subexpression is
-  // not used by anyone else...
-  //
+  // Replace (-1 - A) with (~A)...
+  if (ConstantInt *C = dyn_cast<ConstantInt>(Op0))
+    if (C->isAllOnesValue())
+      return BinaryOperator::createNot(Op1);
+
   if (BinaryOperator *Op1I = dyn_cast<BinaryOperator>(Op1))
-    if (Op1I->use_size() == 1 && Op1I->getOpcode() == Instruction::Sub) {
-      // Swap the two operands of the subexpr...
-      Value *IIOp0 = Op1I->getOperand(0), *IIOp1 = Op1I->getOperand(1);
-      Op1I->setOperand(0, IIOp1);
-      Op1I->setOperand(1, IIOp0);
+    if (Op1I->use_size() == 1) {
+      // Replace (x - (y - z)) with (x + (z - y)) if the (y - z) subexpression
+      // is not used by anyone else...
+      //
+      if (Op1I->getOpcode() == Instruction::Sub) {
+        // Swap the two operands of the subexpr...
+        Value *IIOp0 = Op1I->getOperand(0), *IIOp1 = Op1I->getOperand(1);
+        Op1I->setOperand(0, IIOp1);
+        Op1I->setOperand(1, IIOp0);
+        
+        // Create the new top level add instruction...
+        return BinaryOperator::create(Instruction::Add, Op0, Op1);
+      }
+
+      // Replace (A - (A & B)) with (A & ~B) if this is the only use of (A&B)...
+      //
+      if (Op1I->getOpcode() == Instruction::And &&
+          (Op1I->getOperand(0) == Op0 || Op1I->getOperand(1) == Op0)) {
+        Value *OtherOp = Op1I->getOperand(Op1I->getOperand(0) == Op0);
 
-      // Create the new top level add instruction...
-      return BinaryOperator::create(Instruction::Add, Op0, Op1);
+        Instruction *NewNot = BinaryOperator::createNot(OtherOp, "B.not", &I);
+        return BinaryOperator::create(Instruction::And, Op0, NewNot);
+      }
     }
+
   return 0;
 }
 
 Instruction *InstCombiner::visitMul(BinaryOperator &I) {
   bool Changed = SimplifyBinOp(I);
-  Value *Op1 = I.getOperand(0);
+  Value *Op0 = I.getOperand(0);
 
   // Simplify mul instructions with a constant RHS...
-  if (Constant *Op2 = dyn_cast<Constant>(I.getOperand(1))) {
-    if (I.getType()->isInteger() && cast<ConstantInt>(Op2)->equalsInt(1))
-      return ReplaceInstUsesWith(I, Op1);  // Eliminate 'mul int %X, 1'
-
-    if (I.getType()->isInteger() && cast<ConstantInt>(Op2)->equalsInt(2))
-      // Convert 'mul int %X, 2' to 'add int %X, %X'
-      return BinaryOperator::create(Instruction::Add, Op1, Op1, I.getName());
+  if (Constant *Op1 = dyn_cast<Constant>(I.getOperand(1))) {
+    if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
+      const Type *Ty = CI->getType();
+      uint64_t Val = Ty->isSigned() ?
+                          (uint64_t)cast<ConstantSInt>(CI)->getValue() : 
+                                    cast<ConstantUInt>(CI)->getValue();
+      switch (Val) {
+      case 0:
+        return ReplaceInstUsesWith(I, Op1);  // Eliminate 'mul double %X, 0'
+      case 1:
+        return ReplaceInstUsesWith(I, Op0);  // Eliminate 'mul int %X, 1'
+      case 2:                     // Convert 'mul int %X, 2' to 'add int %X, %X'
+        return BinaryOperator::create(Instruction::Add, Op0, Op0, I.getName());
+      }
 
-    if (Op2->isNullValue())
-      return ReplaceInstUsesWith(I, Op2);  // Eliminate 'mul int %X, 0'
+      if (uint64_t C = Log2(Val))            // Replace X*(2^C) with X << C
+        return new ShiftInst(Instruction::Shl, Op0,
+                             ConstantUInt::get(Type::UByteTy, C));
+    } else {
+      ConstantFP *Op1F = cast<ConstantFP>(Op1);
+      if (Op1F->isNullValue())
+        return ReplaceInstUsesWith(I, Op1);
+
+      // "In IEEE floating point, x*1 is not equivalent to x for nans.  However,
+      // ANSI says we can drop signals, so we can do this anyway." (from GCC)
+      if (Op1F->getValue() == 1.0)
+        return ReplaceInstUsesWith(I, Op0);  // Eliminate 'mul double %X, 1.0'
+    }
   }
 
   return Changed ? &I : 0;
 }
 
-
 Instruction *InstCombiner::visitDiv(BinaryOperator &I) {
   // div X, 1 == X
-  if (ConstantInt *RHS = dyn_cast<ConstantInt>(I.getOperand(1)))
+  if (ConstantInt *RHS = dyn_cast<ConstantInt>(I.getOperand(1))) {
     if (RHS->equalsInt(1))
       return ReplaceInstUsesWith(I, I.getOperand(0));
+
+    // Check to see if this is an unsigned division with an exact power of 2,
+    // if so, convert to a right shift.
+    if (ConstantUInt *C = dyn_cast<ConstantUInt>(RHS))
+      if (uint64_t Val = C->getValue())    // Don't break X / 0
+        if (uint64_t C = Log2(Val))
+          return new ShiftInst(Instruction::Shr, I.getOperand(0),
+                               ConstantUInt::get(Type::UByteTy, C));
+  }
+
+  // 0 / X == 0, we don't need to preserve faults!
+  if (ConstantInt *LHS = dyn_cast<ConstantInt>(I.getOperand(0)))
+    if (LHS->equalsInt(0))
+      return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
+
   return 0;
 }
 
 
 Instruction *InstCombiner::visitRem(BinaryOperator &I) {
-  // rem X, 1 == 0
-  if (ConstantInt *RHS = dyn_cast<ConstantInt>(I.getOperand(1)))
-    if (RHS->equalsInt(1))
+  if (ConstantInt *RHS = dyn_cast<ConstantInt>(I.getOperand(1))) {
+    if (RHS->equalsInt(1))  // X % 1 == 0
+      return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
+
+    // Check to see if this is an unsigned remainder with an exact power of 2,
+    // if so, convert to a bitwise and.
+    if (ConstantUInt *C = dyn_cast<ConstantUInt>(RHS))
+      if (uint64_t Val = C->getValue())    // Don't break X % 0 (divide by zero)
+        if (Log2(Val))
+          return BinaryOperator::create(Instruction::And, I.getOperand(0),
+                                        ConstantUInt::get(I.getType(), Val-1));
+  }
+
+  // 0 % X == 0, we don't need to preserve faults!
+  if (ConstantInt *LHS = dyn_cast<ConstantInt>(I.getOperand(0)))
+    if (LHS->equalsInt(0))
       return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
 
   return 0;
@@ -299,15 +367,19 @@
     if (RHS->isAllOnesValue())
       return ReplaceInstUsesWith(I, Op0);
 
-  // and (not A), (not B) == not (or A, B)
-  if (Op0->use_size() == 1 && Op1->use_size() == 1)
-    if (Value *A = dyn_castNotInst(Op0))
-      if (Value *B = dyn_castNotInst(Op1)) {
-        Instruction *Or = BinaryOperator::create(Instruction::Or, A, B,
-                                                 I.getName()+".demorgan");
-        InsertNewInstBefore(Or, I);
-        return BinaryOperator::createNot(Or, I.getName());
-      }
+  Value *Op0NotVal = dyn_castNotInst(Op0);
+  Value *Op1NotVal = dyn_castNotInst(Op1);
+
+  // (~A & ~B) == (~(A | B)) - Demorgan's Law
+  if (Op0->use_size() == 1 && Op1->use_size() == 1 && Op0NotVal && Op1NotVal) {
+    Instruction *Or = BinaryOperator::create(Instruction::Or, Op0NotVal,
+                                             Op1NotVal,I.getName()+".demorgan",
+                                             &I);
+    return BinaryOperator::createNot(Or);
+  }
+
+  if (Op0NotVal == Op1 || Op1NotVal == Op0)  // A & ~A  == ~A & A == 0
+    return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
 
   return Changed ? &I : 0;
 }
@@ -327,6 +399,16 @@
     if (RHS->isAllOnesValue())
       return ReplaceInstUsesWith(I, Op1);
 
+  if (Value *X = dyn_castNotInst(Op0))   // ~A | A == -1
+    if (X == Op1)
+      return ReplaceInstUsesWith(I, 
+                            ConstantIntegral::getAllOnesValue(I.getType()));
+
+  if (Value *X = dyn_castNotInst(Op1))   // A | ~A == -1
+    if (X == Op0)
+      return ReplaceInstUsesWith(I, 
+                            ConstantIntegral::getAllOnesValue(I.getType()));
+
   return Changed ? &I : 0;
 }
 
@@ -359,6 +441,16 @@
     }
   }
 
+  if (Value *X = dyn_castNotInst(Op0))   // ~A ^ A == -1
+    if (X == Op1)
+      return ReplaceInstUsesWith(I,
+                                ConstantIntegral::getAllOnesValue(I.getType()));
+
+  if (Value *X = dyn_castNotInst(Op1))   // A ^ ~A == -1
+    if (X == Op0)
+      return ReplaceInstUsesWith(I,
+                                ConstantIntegral::getAllOnesValue(I.getType()));
+
   return Changed ? &I : 0;
 }
 
@@ -501,7 +593,7 @@
     // Check to see if we are shifting left by 1.  If so, turn it into an add
     // instruction.
     if (I.getOpcode() == Instruction::Shl && CUI->equalsInt(1))
-      // Convert 'shl int %X, 2' to 'add int %X, %X'
+      // Convert 'shl int %X, 1' to 'add int %X, %X'
       return BinaryOperator::create(Instruction::Add, Op0, Op0, I.getName());
 
   }




From lattner at cs.uiuc.edu  Tue Feb 18 13:29:02 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 13:29:02 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/InstCombine/and.ll div.ll mul.ll or.ll rem.ll sub.ll
Message-ID: <200302181928.NAA02731@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/InstCombine:

and.ll updated: 1.4 -> 1.5
div.ll updated: 1.3 -> 1.4
mul.ll updated: 1.2 -> 1.3
or.ll updated: 1.7 -> 1.8
rem.ll updated: 1.2 -> 1.3
sub.ll updated: 1.5 -> 1.6

---
Log message:

test for a variety of new transformations:
  * A & ~A == 0
  * A / (2^c) == A >> c  if unsigned
  * 0 / A == 0
  * 1.0 * A == A
  * A * (2^c) == A << c
  * A ^ ~A == -1
  * A | ~A == -1
  * 0 % X = 0
  * A % (2^c) == A & (c-1) if unsigned
  * A - (A & B) == A & ~B
  * -1 - A == ~A


---
Diffs of the changes:

Index: llvm/test/Regression/Transforms/InstCombine/and.ll
diff -u llvm/test/Regression/Transforms/InstCombine/and.ll:1.4 llvm/test/Regression/Transforms/InstCombine/and.ll:1.5
--- llvm/test/Regression/Transforms/InstCombine/and.ll:1.4	Tue Feb 18 12:06:44 2003
+++ llvm/test/Regression/Transforms/InstCombine/and.ll	Tue Feb 18 13:28:47 2003
@@ -38,3 +38,8 @@
 	ret bool %B
 }
 
+int %test7(int %A) {         ; A & ~A == 0
+        %NotA = xor int %A, -1
+        %B = and int %A, %NotA
+        ret int %B
+}
\ No newline at end of file


Index: llvm/test/Regression/Transforms/InstCombine/div.ll
diff -u llvm/test/Regression/Transforms/InstCombine/div.ll:1.3 llvm/test/Regression/Transforms/InstCombine/div.ll:1.4
--- llvm/test/Regression/Transforms/InstCombine/div.ll:1.3	Tue Feb 18 13:16:45 2003
+++ llvm/test/Regression/Transforms/InstCombine/div.ll	Tue Feb 18 13:28:47 2003
@@ -12,3 +12,13 @@
 	%B = div int %A, 1
 	ret int %B
 }
+
+uint %test2(uint %A) {
+	%B = div uint %A, 8   ; => Shift
+	ret int %B
+}
+
+int %test3(int %A) {
+	%B = div int 0, %A    ; => 0, don't need to keep traps
+	ret int %B
+}


Index: llvm/test/Regression/Transforms/InstCombine/mul.ll
diff -u llvm/test/Regression/Transforms/InstCombine/mul.ll:1.2 llvm/test/Regression/Transforms/InstCombine/mul.ll:1.3
--- llvm/test/Regression/Transforms/InstCombine/mul.ll:1.2	Fri Aug  2 14:27:57 2002
+++ llvm/test/Regression/Transforms/InstCombine/mul.ll	Tue Feb 18 13:28:47 2003
@@ -26,3 +26,12 @@
 	ret int %B
 end
 
+double %test4(double %A) {
+	%B = mul double 1.0, %A   ; This is safe for FP
+	ret double %B
+}
+
+int %test5(int %A) {
+	%B = mul int %A, 8
+	ret int %B
+}


Index: llvm/test/Regression/Transforms/InstCombine/or.ll
diff -u llvm/test/Regression/Transforms/InstCombine/or.ll:1.7 llvm/test/Regression/Transforms/InstCombine/or.ll:1.8
--- llvm/test/Regression/Transforms/InstCombine/or.ll:1.7	Tue Feb 18 12:05:19 2003
+++ llvm/test/Regression/Transforms/InstCombine/or.ll	Tue Feb 18 13:28:47 2003
@@ -58,3 +58,14 @@
 	ret int %B
 }
 
+int %test11(int %A) {    ; A ^ ~A == -1
+        %NotA = xor int -1, %A
+        %B = xor int %A, %NotA
+        ret int %B
+}
+
+int %test12(int %A) {    ; A | ~A == -1
+        %NotA = xor int -1, %A
+        %B = or int %A, %NotA
+        ret int %B
+}


Index: llvm/test/Regression/Transforms/InstCombine/rem.ll
diff -u llvm/test/Regression/Transforms/InstCombine/rem.ll:1.2 llvm/test/Regression/Transforms/InstCombine/rem.ll:1.3
--- llvm/test/Regression/Transforms/InstCombine/rem.ll:1.2	Fri Aug  2 14:27:58 2002
+++ llvm/test/Regression/Transforms/InstCombine/rem.ll	Tue Feb 18 13:28:47 2003
@@ -8,8 +8,17 @@
 
 implementation
 
-int "test1"(int %A) {
+int %test1(int %A) {
 	%B = rem int %A, 1    ; ISA constant 0
 	ret int %B
 }
 
+int %test2(int %A) {          ; 0 % X = 0, we don't need ot preserve traps
+	%B = rem int 0, %A
+	ret int %B
+}
+
+uint %test3(uint %A) {
+	%B = rem uint %A, 8   ; & 7
+	ret uint %B
+}


Index: llvm/test/Regression/Transforms/InstCombine/sub.ll
diff -u llvm/test/Regression/Transforms/InstCombine/sub.ll:1.5 llvm/test/Regression/Transforms/InstCombine/sub.ll:1.6
--- llvm/test/Regression/Transforms/InstCombine/sub.ll:1.5	Fri Aug  2 14:27:58 2002
+++ llvm/test/Regression/Transforms/InstCombine/sub.ll	Tue Feb 18 13:28:47 2003
@@ -35,3 +35,15 @@
 	%E = sub int %A, %D
 	ret int %E
 }
+
+int %test6(int %A, int %B) {
+	%C = and int %A, %B   ; A - (A & B) => A & ~B
+	%D = sub int %A, %C
+	ret int %D
+}
+
+int %test7(int %A) {
+	%B = sub int -1, %A   ; B = ~A
+	ret int %B
+}
+




From lattner at cs.uiuc.edu  Tue Feb 18 13:44:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 13:44:01 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/InstCombine/add.ll sub.ll
Message-ID: <200302181943.NAA02790@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/InstCombine:

add.ll updated: 1.3 -> 1.4
sub.ll updated: 1.6 -> 1.7

---
Log message:

Modernize testcases


---
Diffs of the changes:

Index: llvm/test/Regression/Transforms/InstCombine/add.ll
diff -u llvm/test/Regression/Transforms/InstCombine/add.ll:1.3 llvm/test/Regression/Transforms/InstCombine/add.ll:1.4
--- llvm/test/Regression/Transforms/InstCombine/add.ll:1.3	Fri Aug  2 14:27:57 2002
+++ llvm/test/Regression/Transforms/InstCombine/add.ll	Tue Feb 18 13:43:21 2003
@@ -10,33 +10,33 @@
 
 implementation
 
-int "test1"(int %A)
+int %test1(int %A)
 begin
 	%B = add int %A, 0
 	ret int %B
 end
 
-int "test2"(int %A)
+int %test2(int %A)
 begin
 	%B = add int %A, 5
 	%C = add int %B, -5
 	ret int %C
 end
 
-int "test3"(int %A)
+int %test3(int %A)
 begin
 	%B = add int %A, 5
 	%C = sub int %B, 5   ;; This should get converted to an add
 	ret int %C
 end
 
-int "test4"(int %A, int %B) {
+int %test4(int %A, int %B) {
         %C = sub int 0, %A
         %D = add int %B, %C      ; D = B + -A = B - A
         ret int %D
 }
 
-int "test5"(int %A, int %B) {
+int %test5(int %A, int %B) {
         %C = sub int 0, %A
         %D = add int %C, %B      ; D = -A + B = B - A
         ret int %D


Index: llvm/test/Regression/Transforms/InstCombine/sub.ll
diff -u llvm/test/Regression/Transforms/InstCombine/sub.ll:1.6 llvm/test/Regression/Transforms/InstCombine/sub.ll:1.7
--- llvm/test/Regression/Transforms/InstCombine/sub.ll:1.6	Tue Feb 18 13:28:47 2003
+++ llvm/test/Regression/Transforms/InstCombine/sub.ll	Tue Feb 18 13:43:21 2003
@@ -8,29 +8,29 @@
 
 implementation
 
-int "test1"(int %A) {
+int %test1(int %A) {
 	%B = sub int %A, %A    ; ISA constant 0
 	ret int %B
 }
 
-int "test2"(int %A) {
+int %test2(int %A) {
 	%B = sub int %A, 0
 	ret int %B
 }
 
-int "test3"(int %A) {
+int %test3(int %A) {
 	%B = sub int 0, %A       ; B = -A
 	%C = sub int 0, %B       ; C = -B = A
 	ret int %C
 }
 
-int "test4"(int %A, int %x) {
+int %test4(int %A, int %x) {
 	%B = sub int 0, %A
 	%C = sub int %x, %B
 	ret int %C
 }
 
-int "test5"(int %A, int %Bok, int %Cok) {
+int %test5(int %A, int %Bok, int %Cok) {
 	%D = sub int %Bok, %Cok
 	%E = sub int %A, %D
 	ret int %E




From lattner at cs.uiuc.edu  Tue Feb 18 13:45:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 13:45:01 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/InstCombine/add.ll
Message-ID: <200302181944.NAA02801@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/InstCombine:

add.ll updated: 1.4 -> 1.5

---
Log message:

Modernize testcase


---
Diffs of the changes:

Index: llvm/test/Regression/Transforms/InstCombine/add.ll
diff -u llvm/test/Regression/Transforms/InstCombine/add.ll:1.4 llvm/test/Regression/Transforms/InstCombine/add.ll:1.5
--- llvm/test/Regression/Transforms/InstCombine/add.ll:1.4	Tue Feb 18 13:43:21 2003
+++ llvm/test/Regression/Transforms/InstCombine/add.ll	Tue Feb 18 13:43:53 2003
@@ -10,25 +10,22 @@
 
 implementation
 
-int %test1(int %A)
-begin
+int %test1(int %A) {
 	%B = add int %A, 0
 	ret int %B
-end
+}
 
-int %test2(int %A)
-begin
+int %test2(int %A) {
 	%B = add int %A, 5
 	%C = add int %B, -5
 	ret int %C
-end
+}
 
-int %test3(int %A)
-begin
+int %test3(int %A) {
 	%B = add int %A, 5
 	%C = sub int %B, 5   ;; This should get converted to an add
 	ret int %C
-end
+}
 
 int %test4(int %A, int %B) {
         %C = sub int 0, %A




From lattner at cs.uiuc.edu  Tue Feb 18 13:56:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 13:56:01 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/InstCombine/add.ll sub.ll
Message-ID: <200302181955.NAA03352@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/InstCombine:

add.ll updated: 1.5 -> 1.6
sub.ll updated: 1.7 -> 1.8

---
Log message:

Add new testcases


---
Diffs of the changes:

Index: llvm/test/Regression/Transforms/InstCombine/add.ll
diff -u llvm/test/Regression/Transforms/InstCombine/add.ll:1.5 llvm/test/Regression/Transforms/InstCombine/add.ll:1.6
--- llvm/test/Regression/Transforms/InstCombine/add.ll:1.5	Tue Feb 18 13:43:53 2003
+++ llvm/test/Regression/Transforms/InstCombine/add.ll	Tue Feb 18 13:55:31 2003
@@ -39,3 +39,15 @@
         ret int %D
 }
 
+int %test6(int %A) {
+        %B = mul int 7, %A
+        %C = add int %B, %A      ; C = 7*A+A == 8*A == A << 3
+        ret int %C
+}
+
+int %test7(int %A) {
+        %B = mul int 7, %A
+        %C = add int %A, %B      ; C = A+7*A == 8*A == A << 3
+        ret int %C
+}
+


Index: llvm/test/Regression/Transforms/InstCombine/sub.ll
diff -u llvm/test/Regression/Transforms/InstCombine/sub.ll:1.7 llvm/test/Regression/Transforms/InstCombine/sub.ll:1.8
--- llvm/test/Regression/Transforms/InstCombine/sub.ll:1.7	Tue Feb 18 13:43:21 2003
+++ llvm/test/Regression/Transforms/InstCombine/sub.ll	Tue Feb 18 13:55:31 2003
@@ -47,3 +47,15 @@
 	ret int %B
 }
 
+int %test8(int %A) {
+        %B = mul int 9, %A
+        %C = sub int %B, %A      ; C = 9*A-A == A*8 == A << 3
+        ret int %C
+}
+
+int %test9(int %A) {
+        %B = mul int 3, %A
+        %C = sub int %A, %B      ; C = A-3*A == A*-2
+        ret int %C
+}
+




From lattner at cs.uiuc.edu  Tue Feb 18 13:58:00 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 13:58:00 2003
Subject: [llvm-commits] CVS: llvm/lib/Transforms/Scalar/InstructionCombining.cpp
Message-ID: <200302181957.NAA03364@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Transforms/Scalar:

InstructionCombining.cpp updated: 1.68 -> 1.69

---
Log message:

4 new transformations:
  * X*C + X --> X * (C+1)
  * X + X*C --> X * (C+1)
  * X - X*C --> X * (1-C)
  * X*C - X --> X * (C-1)



---
Diffs of the changes:

Index: llvm/lib/Transforms/Scalar/InstructionCombining.cpp
diff -u llvm/lib/Transforms/Scalar/InstructionCombining.cpp:1.68 llvm/lib/Transforms/Scalar/InstructionCombining.cpp:1.69
--- llvm/lib/Transforms/Scalar/InstructionCombining.cpp:1.68	Tue Feb 18 13:28:33 2003
+++ llvm/lib/Transforms/Scalar/InstructionCombining.cpp	Tue Feb 18 13:57:07 2003
@@ -146,6 +146,16 @@
   return Count;
 }
 
+static inline Value *dyn_castFoldableMul(Value *V) {
+  if (V->use_size() == 1 && V->getType()->isInteger())
+    if (Instruction *I = dyn_cast<Instruction>(V))
+      if (I->getOpcode() == Instruction::Mul)
+        if (isa<Constant>(I->getOperand(1)))
+          return I->getOperand(0);
+  return 0;
+}
+
+
 Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
   bool Changed = SimplifyBinOp(I);
   Value *LHS = I.getOperand(0), *RHS = I.getOperand(1);
@@ -182,6 +192,22 @@
     }
   }
 
+  // X*C + X --> X * (C+1)
+  if (dyn_castFoldableMul(LHS) == RHS) {
+    Constant *CP1 = *cast<Constant>(cast<Instruction>(LHS)->getOperand(1)) +
+                    *ConstantInt::get(I.getType(), 1);
+    assert(CP1 && "Couldn't constant fold C + 1?");
+    return BinaryOperator::create(Instruction::Mul, RHS, CP1);
+  }
+
+  // X + X*C --> X * (C+1)
+  if (dyn_castFoldableMul(RHS) == LHS) {
+    Constant *CP1 = *cast<Constant>(cast<Instruction>(RHS)->getOperand(1)) +
+                    *ConstantInt::get(I.getType(), 1);
+    assert(CP1 && "Couldn't constant fold C + 1?");
+    return BinaryOperator::create(Instruction::Mul, LHS, CP1);
+  }
+
   return Changed ? &I : 0;
 }
 
@@ -231,7 +257,23 @@
         Instruction *NewNot = BinaryOperator::createNot(OtherOp, "B.not", &I);
         return BinaryOperator::create(Instruction::And, Op0, NewNot);
       }
+
+      // X - X*C --> X * (1-C)
+      if (dyn_castFoldableMul(Op1I) == Op0) {
+        Constant *CP1 = *ConstantInt::get(I.getType(), 1) -
+                        *cast<Constant>(cast<Instruction>(Op1)->getOperand(1));
+        assert(CP1 && "Couldn't constant fold 1-C?");
+        return BinaryOperator::create(Instruction::Mul, Op0, CP1);
+      }
     }
+
+  // X*C - X --> X * (C-1)
+  if (dyn_castFoldableMul(Op0) == Op1) {
+    Constant *CP1 = *cast<Constant>(cast<Instruction>(Op0)->getOperand(1)) -
+                    *ConstantInt::get(I.getType(), 1);
+    assert(CP1 && "Couldn't constant fold C - 1?");
+    return BinaryOperator::create(Instruction::Mul, Op1, CP1);
+  }
 
   return 0;
 }




From lattner at cs.uiuc.edu  Tue Feb 18 14:41:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 14:41:01 2003
Subject: [llvm-commits] CVS: llvm/test/Programs/TEST.dsgraph.report
Message-ID: <200302182040.OAA03515@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Programs:

TEST.dsgraph.report updated: 1.4 -> 1.5

---
Log message:

many changes


---
Diffs of the changes:

Index: llvm/test/Programs/TEST.dsgraph.report
diff -u llvm/test/Programs/TEST.dsgraph.report:1.4 llvm/test/Programs/TEST.dsgraph.report:1.5
--- llvm/test/Programs/TEST.dsgraph.report:1.4	Fri Feb 14 18:50:40 2003
+++ llvm/test/Programs/TEST.dsgraph.report	Tue Feb 18 14:40:04 2003
@@ -15,36 +15,84 @@
   }
 }
 
+sub FormatSize {
+  my $Size = shift;
+  if ($Size > 10*1024*1024) {
+    return (int $Size*10 / (1024*1024))/10 . "MB";
+  } elsif ($Size > 10*1024) {
+    return (int $Size / 1024);
+  } else {
+    return $Size . "B";
+  }
+}
+
+sub NoStar {
+  return "0" if ($_[0] eq '*');
+  return $_[0];
+}
+
+sub NoCallNodes {
+  $_[0] =~ m/([0-9]+)\+/;
+  return $1;
+}
+
 # Sort by total analyze time
 $SortCol = 7;
 
 # Sort in descending order
 $SortReverse = 1;
 
-%NameMap = (
-            treeadd   => 'Olden-treeadd',
-            bisort    => 'Olden-bisort',
-            mst       => 'Olden-mst',
-            perimeter => 'Olden-perimeter',
-            health    => 'Olden-health',
-            tsp       => 'Olden-tsp',
-            power     => 'Olden-power',
-            em3d      => 'Olden-em3d',
-            voronoi   => 'Olden-voronoi',
-            bh        => 'Olden-bh',
-
-            ks        => 'ptrdist-ks',
-            anagram   => 'ptrdist-anagram',
-            yacr2     => 'ptrdist-yacr2',
-            ft        => 'ptrdist-ft',
-            bc        => 'ptrdist-bc');
+# For latex output, limit benchmarks and rename as appropriate
+ at LatexRowMapOrder = (
+            'treeadd'    => 'Olden-treeadd',
+            'bisort'     => 'Olden-bisort',
+            'mst'        => 'Olden-mst',
+            'perimeter'  => 'Olden-perimeter',
+            'health'     => 'Olden-health',
+            'tsp'        => 'Olden-tsp',
+            'power'      => 'Olden-power',
+            'em3d'       => 'Olden-em3d',
+            'voronoi'    => 'Olden-voronoi',
+            'bh'         => 'Olden-bh',
+            '-' => '-',
+            'anagram'    => 'ptrdist-anagram',
+            'ks'         => 'ptrdist-ks',
+            'ft'         => 'ptrdist-ft',
+            'yacr2'      => 'ptrdist-yacr2',
+            'bc'         => 'ptrdist-bc',
+            '-' => '-',
+            '181.mcf'    => '181.mcf',
+            '256.bzip2'  => '256.bzip2',
+            '164.gzip'   => '164.gzip',
+            '197.parser' => '197.parser',
+            '300.twolf'  => '300.twolf',
+            '255.vortex' => '255.vortex',
+            '-' => '-',
+            'sgefa'      => 'sgefa',
+            'sim'        => 'sim',
+            'burg'       => 'burg',
+            'gnuchess'   => 'gnuchess',
+            'larn'       => 'larn',
+            'flex'       => 'flex',
+            'moria'      => 'moria',
+            'povray31'   => 'povray31'
+           );
 
- at LatexColumns = (0, 1, 26, 19, # Program, LOC, #Instrs, MaxSCC
+ at LatexColumns = (1, 27, 19,    # LOC, #Instrs, MaxSCC
                  4, 5, 6, 7,   # Execution times
-                 13,           # Memory BU+TD
+                 11,           # Memory BU
+                 12,           # Memory TD
                  16, 17,       # Total, Max Nodes
                  15);          # NumFolded
 
+# Specify how to format columns nicely for printing...
+%LatexColumnFormat = (
+                      11 => \&FormatSize,
+                      12 => \&FormatSize,
+                      15 => \&NoStar,
+                      16 => \&NoCallNodes
+                     );
+
 # These are the columns for the report.  The first entry is the header for the
 # column, the second is the regex to use to match the value.  Empty list create
 # seperators, and closures may be put in for custom processing.
@@ -65,7 +113,7 @@
             ["LcSize:" , '([0-9]+)  Local'],
             ["BUSize:" , '([0-9]+)  Bottom-up'],
             ["TDSize:" , '([0-9]+)  Top-down'],
-            ["BUTDSz:" , sub { return SumCols(@_, 3); }],
+            ["BUTDSz:" , sub { return SumCols(@_, 2); }],
             [],
 # Misc stuff
             ["NumFold" , '([0-9]+).*Number of folded nodes '],
@@ -79,7 +127,8 @@
             ["Calls"   , '([0-9]+).*Number of Call insts'],
             ["Allca"   , '([0-9]+).*Number of Alloca insts'],
             ["Mallc"   , '([0-9]+).*Number of Malloc insts'],
-            ["Sum"   , sub { return SumCols(@_, 5); }],
+            ["GEP"     , '([0-9]+).*Number of GetElementPtr insts'],
+            ["Sum"   , sub { return SumCols(@_, 6); }],
             [],
             ["num/ind" , '([0-9]+).*number of indirect call sites'],
             ["indcallee",'([0-9]+).*number of callee functions at'],




From lattner at cs.uiuc.edu  Tue Feb 18 14:41:02 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 14:41:02 2003
Subject: [llvm-commits] CVS: llvm/test/Programs/TEST.dsgraph.Makefile
Message-ID: <200302182040.OAA03528@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Programs:

TEST.dsgraph.Makefile updated: 1.9 -> 1.10

---
Log message:

Track memory usage


---
Diffs of the changes:

Index: llvm/test/Programs/TEST.dsgraph.Makefile
diff -u llvm/test/Programs/TEST.dsgraph.Makefile:1.9 llvm/test/Programs/TEST.dsgraph.Makefile:1.10
--- llvm/test/Programs/TEST.dsgraph.Makefile:1.9	Fri Feb 14 18:31:28 2003
+++ llvm/test/Programs/TEST.dsgraph.Makefile	Tue Feb 18 14:40:20 2003
@@ -13,12 +13,14 @@
 
 ANALYZE_OPTS := -stats -time-passes -only-print-main-ds -dsstats -instcount
 
+MEM := -track-memory
+
 $(PROGRAMS_TO_TEST:%=Output/%.$(TEST).report.txt): \
 Output/%.$(TEST).report.txt: Output/%.lib.bc Output/%.LOC.txt $(LANALYZE) $(LOPT)
 	@echo -n "LOC: " > $@
 	@cat Output/$*.LOC.txt >> $@
-	-(time -p $(LANALYZE) -$(PASS)datastructure $(ANALYZE_OPTS) $<)>> $@ 2>&1
-	-($(LOPT) -steens-aa -time-passes > /dev/null < $<) >> $@ 2>&1
+	-(time -p $(LANALYZE) $(MEM) -$(PASS)datastructure $(ANALYZE_OPTS) $<)>> $@ 2>&1
+	-($(LOPT) -steens-aa $(MEM) -time-passes > /dev/null < $<) >> $@ 2>&1
 
 $(PROGRAMS_TO_TEST:%=test.$(TEST).%): \
 test.$(TEST).%: Output/%.$(TEST).report.txt




From lattner at cs.uiuc.edu  Tue Feb 18 14:41:03 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 14:41:03 2003
Subject: [llvm-commits] CVS: llvm/test/Programs/GenerateReport.pl
Message-ID: <200302182040.OAA03541@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Programs:

GenerateReport.pl updated: 1.8 -> 1.9

---
Log message:

Add support for LatexColumnFormat


---
Diffs of the changes:

Index: llvm/test/Programs/GenerateReport.pl
diff -u llvm/test/Programs/GenerateReport.pl:1.8 llvm/test/Programs/GenerateReport.pl:1.9
--- llvm/test/Programs/GenerateReport.pl:1.8	Fri Feb 14 18:31:49 2003
+++ llvm/test/Programs/GenerateReport.pl	Tue Feb 18 14:40:45 2003
@@ -73,6 +73,7 @@
   die "Couldn't open report description '$ReportFN'!";
 
 my @LatexColumns;  # Filled in by report if it supports Latex mode
+my %LatexColumnFormat;  # Filled in by report if supports latex mode
 my @Fields = eval <REPORTDESC>;
 
 
@@ -184,16 +185,41 @@
     # Print out the latexified table...
     #
     shift @Values;  # Don't print the header...
-    foreach $Row (@Values) {
-      my $First = 1;
-      
-      for $ColNum (@LatexColumns) {
-        # Print a seperator...
-        print " & " if ($First == 0);
-        $First = 0;
-        printf "%-$FieldWidths[$ColNum]s", $Row->[$ColNum];
+
+    # Make sure the benchmark name field is wide enough for any aliases.
+    foreach $Name (@LatexRowMapOrder) {
+      $FieldWidths[0] = length $Name if (length($Name) > $FieldWidths[0]);
+    }
+
+    # Print out benchmarks listed in the LatexRowMapOrder
+    for ($i = 0; $i < @LatexRowMapOrder; $i += 2) {
+      my $Name = $LatexRowMapOrder[$i];
+      if ($Name eq '-') {
+        print "\\hline\n";
+      } else {
+        # Output benchmark name...
+        printf "%-$FieldWidths[0]s", $LatexRowMapOrder[$i+1];
+
+        # Find the row that this benchmark name corresponds to...
+        foreach $Row (@Values) {
+          if ($Row->[0] eq $Name) {
+            for $ColNum (@LatexColumns) {
+              # Print a seperator...
+              my $Val = $Row->[$ColNum];
+              if (exists $LatexColumnFormat{$ColNum}) {
+                # If a column format routine has been specified, run it now...
+                $Val = &{$LatexColumnFormat{$ColNum}}($Val);
+              }
+
+              printf " & %-$FieldWidths[$ColNum]s", $Val;
+            }
+            goto Done;
+          }
+        }
+        print "UNKNOWN Benchmark name: " . $Name;
+      Done:
+        print "\\\\\n";
       }
-      print "\\\\\n";
     }
   } else {
     #




From lattner at cs.uiuc.edu  Tue Feb 18 14:42:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 14:42:01 2003
Subject: [llvm-commits] CVS: llvm/test/Programs/Makefile
Message-ID: <200302182041.OAA03552@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Programs:

Makefile updated: 1.12 -> 1.13

---
Log message:

Add support for report.tex pseudo target


---
Diffs of the changes:

Index: llvm/test/Programs/Makefile
diff -u llvm/test/Programs/Makefile:1.12 llvm/test/Programs/Makefile:1.13
--- llvm/test/Programs/Makefile:1.12	Fri Feb 14 17:25:44 2003
+++ llvm/test/Programs/Makefile	Tue Feb 18 14:41:10 2003
@@ -35,8 +35,14 @@
 report.$(TEST).html: report.$(TEST).raw.out $(TestReport)
 	./GenerateReport.pl -html TEST.$(TEST).report < $< > $@
 
+report.$(TEST).tex: report.$(TEST).raw.out $(TestReport)
+	./GenerateReport.pl -latex TEST.$(TEST).report < $< > $@
+
 report: report.$(TEST).txt
-	@cat report.$(TEST).txt
+	@cat $<
+
+report.tex: report.$(TEST).tex
+	@cat $<
 
 clean::
 	rm -f report.*.raw.out report.*.txt




From lattner at cs.uiuc.edu  Tue Feb 18 14:42:02 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 14:42:02 2003
Subject: [llvm-commits] CVS: llvm/test/Programs/TEST.aa.report
Message-ID: <200302182041.OAA03566@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Programs:

TEST.aa.report updated: 1.1 -> 1.2

---
Log message:

Add latex output support


---
Diffs of the changes:

Index: llvm/test/Programs/TEST.aa.report
diff -u llvm/test/Programs/TEST.aa.report:1.1 llvm/test/Programs/TEST.aa.report:1.2
--- llvm/test/Programs/TEST.aa.report:1.1	Wed Feb 12 13:34:23 2003
+++ llvm/test/Programs/TEST.aa.report	Tue Feb 18 14:41:27 2003
@@ -16,11 +16,63 @@
   }
 }
 
-# Sort by program name...
-$SortCol = 1;
+sub OnlyMayInfo {
+  $_[0] =~ m|/([0-9]+)%/|;
+  return $1."\\%";
+}
+
+# Sort by ds time...
+$SortCol = 15;
 
 # Sort in ascending order
 $SortReverse = 0;
+
+# For latex output, limit benchmarks and rename as appropriate
+ at LatexRowMapOrder = (
+            'treeadd'    => 'Olden-treeadd',
+            'bisort'     => 'Olden-bisort',
+            'mst'        => 'Olden-mst',
+            'perimeter'  => 'Olden-perimeter',
+            'health'     => 'Olden-health',
+            'tsp'        => 'Olden-tsp',
+            'power'      => 'Olden-power',
+            'em3d'       => 'Olden-em3d',
+            'voronoi'    => 'Olden-voronoi',
+            'bh'         => 'Olden-bh',
+            '-' => '-',
+            'anagram'    => 'ptrdist-anagram',
+            'ks'         => 'ptrdist-ks',
+            'ft'         => 'ptrdist-ft',
+            'yacr2'      => 'ptrdist-yacr2',
+            'bc'         => 'ptrdist-bc',
+            '-' => '-',
+            '181.mcf'    => '181.mcf',
+            '256.bzip2'  => '256.bzip2',
+            '164.gzip'   => '164.gzip',
+            '197.parser' => '197.parser',
+            '300.twolf'  => '300.twolf',
+            '255.vortex' => '255.vortex',
+            '-' => '-',
+            'sgefa'      => 'sgefa',
+            'sim'        => 'sim',
+            'burg'       => 'burg',
+            'gnuchess'   => 'gnuchess',
+            'larn'       => 'larn',
+            'flex'       => 'flex',
+            'moria'      => 'moria',
+            'povray31'   => 'povray31'
+           );
+
+ at LatexColumns = (2, 3, 4, 6, 7, 5);
+
+# Specify how to reformat columns to be presentable...
+%LatexColumnFormat = (
+                      2 => \&OnlyMayInfo,
+                      3 => \&OnlyMayInfo,
+                      4 => \&OnlyMayInfo,
+                      5 => \&OnlyMayInfo,
+                      6 => \&OnlyMayInfo,
+                      7 => \&OnlyMayInfo);
 
 # These are the columns for the report.  The first entry is the header for the
 # column, the second is the regex to use to match the value.  Empty list create




From lattner at cs.uiuc.edu  Tue Feb 18 14:43:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 14:43:01 2003
Subject: [llvm-commits] CVS: llvm/test/Libraries/libdummy/dummylib.c
Message-ID: <200302182042.OAA03581@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Libraries/libdummy:

dummylib.c updated: 1.4 -> 1.5

---
Log message:

Add lots of functions


---
Diffs of the changes:

Index: llvm/test/Libraries/libdummy/dummylib.c
diff -u llvm/test/Libraries/libdummy/dummylib.c:1.4 llvm/test/Libraries/libdummy/dummylib.c:1.5
--- llvm/test/Libraries/libdummy/dummylib.c:1.4	Tue Feb 11 17:12:23 2003
+++ llvm/test/Libraries/libdummy/dummylib.c	Tue Feb 18 14:42:07 2003
@@ -3,10 +3,19 @@
 #include <time.h>
 #include <stdarg.h>
 #include <sys/time.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+int stat(const char *file_name, struct stat *buf) { return 0; }
+int fstat(int filedes, struct stat *buf) { return 0; }
+int lstat(const char *file_name, struct stat *buf) { return 0; }
 
 // Varargs function definitions
-int ioctl(int d, int request) {return 0; }
+int ioctl(int d, int request, ...) {return 0; }
 int printf(const char *X) {return 0; }
+int sscanf(const char *X, const char *Y, ...) { return 0; }
+int fprintf(FILE *stream, const char *format, ...) { return 0; }
 
 
 int gettimeofday(struct timeval *tv, void *tz) { return 0; }
@@ -21,6 +30,7 @@
 void __main() {}
 int atoi(const char*x) { return 1; }
 char *fgets(char*Ptr, int x, FILE*F) { return Ptr; }
+char *gets(char *C) { return C; }
 int fclose(FILE*F) { return 0; }
 FILE *fopen(const char *n, const char*x) { return malloc(sizeof(FILE)); }
 FILE *freopen(const char *path, const char *mode, FILE *stream) { return 0; }
@@ -45,6 +55,9 @@
 int fputs(const char *s, char *stream) { return 0; }
 int ferror(FILE *F) { return 0; }
 FILE *fdopen(int fildes, const char *mode) { return 0;}
+FILE *popen(const char *command, const char *type) { return 0; }
+int pclose(FILE *stream) { return 0; }
+ 
 int ungetc(int c, FILE *stream) { return 0; }
 int setvbuf(FILE *stream, char *buf, int mode , size_t size) { return 0; }
 void rewind(FILE*F) { }




From lattner at cs.uiuc.edu  Tue Feb 18 14:43:02 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Tue Feb 18 14:43:02 2003
Subject: [llvm-commits] CVS: llvm/test/Libraries/libc/string.c
Message-ID: <200302182042.OAA03588@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Libraries/libc:

string.c updated: 1.3 -> 1.4

---
Log message:

Remove support for memset


---
Diffs of the changes:

Index: llvm/test/Libraries/libc/string.c
diff -u llvm/test/Libraries/libc/string.c:1.3 llvm/test/Libraries/libc/string.c:1.4
--- llvm/test/Libraries/libc/string.c:1.3	Wed Jul 17 19:15:29 2002
+++ llvm/test/Libraries/libc/string.c	Tue Feb 18 14:42:15 2003
@@ -53,7 +53,7 @@
 }
 
 // http://sources.redhat.com/cgi-bin/cvsweb.cgi/libc/sysdeps/generic/?cvsroot=glibc
-
+#if 0
 typedef unsigned int op_t;
 #define OPSIZ 4
 
@@ -119,6 +119,7 @@
 
   return dstpp;
 }
+#endif
 
 void *memcpy(void *dstpp, const void *srcpp, size_t len) {
   char *dstp = (char*)dstpp;




From ashukla at cs.uiuc.edu  Wed Feb 19 13:20:01 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Wed Feb 19 13:20:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Reoptimizer/TraceCache/MemoryManager2.cpp
Message-ID: <200302191919.NAA18784@trinity.cs.uiuc.edu>


Changes in directory llvm/lib/Reoptimizer/TraceCache:

MemoryManager2.cpp updated: 1.1 -> 1.2

---
Log message:

corrected for compilation warnings

---
Diffs of the changes:

Index: llvm/lib/Reoptimizer/TraceCache/MemoryManager2.cpp
diff -u llvm/lib/Reoptimizer/TraceCache/MemoryManager2.cpp:1.1 llvm/lib/Reoptimizer/TraceCache/MemoryManager2.cpp:1.2
--- llvm/lib/Reoptimizer/TraceCache/MemoryManager2.cpp:1.1	Mon Feb 17 13:51:00 2003
+++ llvm/lib/Reoptimizer/TraceCache/MemoryManager2.cpp	Wed Feb 19 13:19:17 2003
@@ -32,8 +32,8 @@
 int dummyFunction2(int i); //the function that provides memory!
 
 MemoryManager2::MemoryManager2(){
-  memStart = (uint64_t)&dummyFunction2;
-  memEnd = (uint64_t)&dummyFunction2+globalMemSize2*4;
+  memStart = (uint64_t)(intptr_t)&dummyFunction2;
+  memEnd = (uint64_t)(intptr_t)&dummyFunction2+globalMemSize2*4;
   //std::cerr<<"Dummy address :"<<(void *)memStart<<"\n";
   freeMemList.push_back(std::make_pair(memStart, memEnd));
 }
@@ -49,11 +49,11 @@
   for(std::list<std::pair<uint64_t, uint64_t> >::iterator LI =
         freeMemList.begin(), LE = freeMemList.end(); LI!=LE; ++LI){
 
-    if(LI->second - LI->first >= sz-1){
+    if(LI->second - LI->first >= (uint64_t)(sz-1)){
       uint64_t temp = LI->first;
 
       //reassign the pair
-      if(LI->second - LI->first == sz-1)
+      if(LI->second - LI->first == (uint64_t)(sz-1))
         freeMemList.erase(LI);
       else
         LI->first += sz;




From ashukla at cs.uiuc.edu  Wed Feb 19 13:20:03 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Wed Feb 19 13:20:03 2003
Subject: [llvm-commits] CVS: llvm/lib/Reoptimizer/TraceCache/MemoryManager.cpp
Message-ID: <200302191919.NAA18792@trinity.cs.uiuc.edu>


Changes in directory llvm/lib/Reoptimizer/TraceCache:

MemoryManager.cpp updated: 1.3 -> 1.4

---
Log message:

corrected for compilation warnings

---
Diffs of the changes:

Index: llvm/lib/Reoptimizer/TraceCache/MemoryManager.cpp
diff -u llvm/lib/Reoptimizer/TraceCache/MemoryManager.cpp:1.3 llvm/lib/Reoptimizer/TraceCache/MemoryManager.cpp:1.4
--- llvm/lib/Reoptimizer/TraceCache/MemoryManager.cpp:1.3	Fri Feb 14 14:45:53 2003
+++ llvm/lib/Reoptimizer/TraceCache/MemoryManager.cpp	Wed Feb 19 13:19:11 2003
@@ -32,8 +32,8 @@
 int dummyFunction(int i); //the function that provides memory!
 
 MemoryManager::MemoryManager(){
-  memStart = (uint64_t)&dummyFunction;
-  memEnd = (uint64_t)&dummyFunction+globalMemSize*4;
+  memStart = (uint64_t)(intptr_t)&dummyFunction;
+  memEnd = (uint64_t)(intptr_t)&dummyFunction+globalMemSize*4;
   //std::cerr<<"Dummy address :"<<(void *)memStart<<"\n";
   freeMemList.push_back(std::make_pair(memStart, memEnd));
 }
@@ -49,11 +49,11 @@
   for(std::list<std::pair<uint64_t, uint64_t> >::iterator LI =
         freeMemList.begin(), LE = freeMemList.end(); LI!=LE; ++LI){
 
-    if(int64_t(LI->second - LI->first) >= sz-1){
+    if(LI->second - LI->first >= (uint64_t)(sz-1)){
       uint64_t temp = LI->first;
 
       //reassign the pair
-      if(int64_t(LI->second - LI->first) == sz-1)
+      if(LI->second - LI->first == (uint64_t)(sz-1))
         freeMemList.erase(LI);
       else
         LI->first += sz;




From ashukla at cs.uiuc.edu  Wed Feb 19 13:27:01 2003
From: ashukla at cs.uiuc.edu (Anand Shukla)
Date: Wed Feb 19 13:27:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Reoptimizer/Trigger/Trigger.cpp
Message-ID: <200302191926.NAA18805@trinity.cs.uiuc.edu>


Changes in directory llvm/lib/Reoptimizer/Trigger:

Trigger.cpp updated: 1.15 -> 1.16

---
Log message:

corrected for compilation warnings

---
Diffs of the changes:

Index: llvm/lib/Reoptimizer/Trigger/Trigger.cpp
diff -u llvm/lib/Reoptimizer/Trigger/Trigger.cpp:1.15 llvm/lib/Reoptimizer/Trigger/Trigger.cpp:1.16
--- llvm/lib/Reoptimizer/Trigger/Trigger.cpp:1.15	Fri Feb 14 14:46:29 2003
+++ llvm/lib/Reoptimizer/Trigger/Trigger.cpp	Wed Feb 19 13:26:38 2003
@@ -271,8 +271,10 @@
                    && "Incorrect trace");
 
             //insert a BA going to the top
-            trace.push_back(getBranchInst(0x10800000, (uint64_t)(&trace[0]),
-                                          (uint64_t)(&trace[instIndex])));
+            trace.push_back(getBranchInst
+                            (0x10800000, 
+                             (uint64_t)(intptr_t)&trace[0],
+                             (uint64_t)(intptr_t)&trace[instIndex]));
             //insert delay slot
             if(addrToRm.find(addr+4) == addrToRm.end())
               trace.push_back(vm->readInstrFrmVm(addr+4, tr));
@@ -351,8 +353,10 @@
                    && "Incorrect trace");
             
             //insert a BA going to the top
-            trace.push_back(getBranchInst(0x10800000, (uint64_t)(&trace[0]),
-                                          (uint64_t)(&trace[instIndex])));
+            trace.push_back(getBranchInst
+                            (0x10800000, 
+                             (uint64_t)(intptr_t)&trace[0],
+                             (uint64_t)(intptr_t)&trace[instIndex]));
             
             
             
@@ -506,7 +510,9 @@
     instVec.push_back(jmpl);
     instVec.push_back(load1); //delay slot!
 
-    uint64_t newInst = getBranchInst(instr, (uint64_t)(&instVec[instIndex]), (uint64_t)(&instVec[MI->first]));
+    uint64_t newInst = getBranchInst(instr, 
+                                     (uint64_t)(intptr_t)&instVec[instIndex], 
+                                     (uint64_t)(intptr_t)&instVec[MI->first]);
     instVec[MI->first] = newInst;
 
     //change original call map to point to new PC




From lattner at cs.uiuc.edu  Wed Feb 19 16:13:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Wed Feb 19 16:13:01 2003
Subject: [llvm-commits] CVS: llvm/Makefile.common
Message-ID: <200302192212.QAA15915@apoc.cs.uiuc.edu>


Changes in directory llvm:

Makefile.common updated: 1.78 -> 1.79

---
Log message:

Fix warnings compiling C files


---
Diffs of the changes:

Index: llvm/Makefile.common
diff -u llvm/Makefile.common:1.78 llvm/Makefile.common:1.79
--- llvm/Makefile.common:1.78	Thu Feb 13 10:56:30 2003
+++ llvm/Makefile.common	Wed Feb 19 16:12:20 2003
@@ -164,8 +164,8 @@
 # Compile a c file, don't link...
 CompileC  := $(CC) -c $(CPPFLAGS) $(CCFLAGS) $(CompileCommonOpts)
 CompileCG := $(CompileC) -g  -D_DEBUG
-CompileCO := $(CompileC) $(CompileOptimizeOpts) -felide-constructors -fomit-frame-pointer
-CompileCP := $(CompileC) $(CompileOptimizeOpts) -felide-constructors $(PROFILE)
+CompileCO := $(CompileC) $(CompileOptimizeOpts) -fomit-frame-pointer
+CompileCP := $(CompileC) $(CompileOptimizeOpts) $(PROFILE)
 
 
 # Link final executable




From lattner at cs.uiuc.edu  Wed Feb 19 18:10:02 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Wed Feb 19 18:10:02 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Other/Makefile
Message-ID: <200302200009.SAA19313@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Other:

Makefile updated: 1.1 -> 1.2

---
Log message:

Make tests rerun if analyze is rebuilt


---
Diffs of the changes:

Index: llvm/test/Regression/Other/Makefile
diff -u llvm/test/Regression/Other/Makefile:1.1 llvm/test/Regression/Other/Makefile:1.2
--- llvm/test/Regression/Other/Makefile:1.1	Thu Jan 31 22:24:20 2002
+++ llvm/test/Regression/Other/Makefile	Wed Feb 19 18:09:19 2003
@@ -6,5 +6,5 @@
 
 all:: $(addprefix Output/, $(TESTS:%.ll=%.ll.out))
 
-Output/%.ll.out: %.ll Output/.dir
+Output/%.ll.out: %.ll Output/.dir $(LANALYZE)
 	-$(TESTRUNR) $<




From lattner at cs.uiuc.edu  Wed Feb 19 18:10:06 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Wed Feb 19 18:10:06 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Other/2003-02-19-LoopInfoNestingBug.ll
Message-ID: <200302200009.SAA19345@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Other:

2003-02-19-LoopInfoNestingBug.ll added (r1.1)

---
Log message:

Add a testcase for broken loopinfo


---
Diffs of the changes:



From lattner at cs.uiuc.edu  Wed Feb 19 18:18:00 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Wed Feb 19 18:18:00 2003
Subject: [llvm-commits] CVS: llvm/lib/Analysis/LoopInfo.cpp
Message-ID: <200302200017.SAA19709@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Analysis:

LoopInfo.cpp updated: 1.26 -> 1.27

---
Log message:

Fix bug: 2003-02-19-LoopInfoNestingBug.ll


---
Diffs of the changes:

Index: llvm/lib/Analysis/LoopInfo.cpp
diff -u llvm/lib/Analysis/LoopInfo.cpp:1.26 llvm/lib/Analysis/LoopInfo.cpp:1.27
--- llvm/lib/Analysis/LoopInfo.cpp:1.26	Fri Oct 11 00:30:13 2002
+++ llvm/lib/Analysis/LoopInfo.cpp	Wed Feb 19 18:17:17 2003
@@ -37,13 +37,12 @@
   unsigned numBackEdges = 0;
   BasicBlock *header = Blocks.front();
 
-  for (std::vector<BasicBlock*>::const_iterator i = Blocks.begin(), e = Blocks.end();
-       i != e; ++i) {
-    for (BasicBlock::succ_iterator Successor = succ_begin(*i), SEnd = succ_end(*i);
-	 Successor != SEnd; ++Successor) {
-      if (header == *Successor)
+  for (std::vector<BasicBlock*>::const_iterator I = Blocks.begin(),
+         E = Blocks.end(); I != E; ++I) {
+    for (BasicBlock::succ_iterator SI = succ_begin(*I), SE = succ_end(*I);
+         SI != SE; ++SI)
+      if (header == *SI)
 	++numBackEdges;
-    }
   }
   return numBackEdges;
 }
@@ -138,17 +137,38 @@
   // they start subloops of their own.
   //
   for (std::vector<BasicBlock*>::reverse_iterator I = L->Blocks.rbegin(),
-	 E = L->Blocks.rend(); I != E; ++I) {
+	 E = L->Blocks.rend(); I != E; ++I)
 
     // Check to see if this block starts a new loop
-    if (Loop *NewLoop = ConsiderForLoop(*I, DS)) {
-      L->SubLoops.push_back(NewLoop);
-      NewLoop->ParentLoop = L;
-    }
-  
-    if (BBMap.find(*I) == BBMap.end())
-      BBMap.insert(std::make_pair(*I, L));
-  }
+    if (*I != BB)
+      if (Loop *NewLoop = ConsiderForLoop(*I, DS)) {
+        L->SubLoops.push_back(NewLoop);
+        NewLoop->ParentLoop = L;
+      } else {
+        std::map<BasicBlock*, Loop*>::iterator BBMI = BBMap.lower_bound(*I);
+        if (BBMI == BBMap.end() || BBMI->first != *I) {  // Not in map yet...
+          BBMap.insert(BBMI, std::make_pair(*I, L));
+        } else {
+          // If this is already in the BBMap then this means that we already added
+          // a loop for it, but incorrectly added the loop to a higher level loop
+          // instead of the current loop we are creating.  Fix this now by moving
+          // the loop into the correct subloop.
+          //
+          Loop *SubLoop = BBMI->second;
+          Loop *OldSubLoopParent = SubLoop->getParentLoop();
+          if (OldSubLoopParent != L) {
+            // Remove SubLoop from OldSubLoopParent's list of subloops...
+            std::vector<Loop*>::iterator I =
+              std::find(OldSubLoopParent->SubLoops.begin(),
+                        OldSubLoopParent->SubLoops.end(), SubLoop);
+            assert(I != OldSubLoopParent->SubLoops.end()
+                   && "Loop parent doesn't contain loop?");
+            OldSubLoopParent->SubLoops.erase(I);
+            SubLoop->ParentLoop = L;
+            L->SubLoops.push_back(SubLoop);
+          }
+        }
+      }
 
   return L;
 }




From lattner at cs.uiuc.edu  Wed Feb 19 18:19:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Wed Feb 19 18:19:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Analysis/LoopInfo.cpp
Message-ID: <200302200018.SAA19753@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Analysis:

LoopInfo.cpp updated: 1.27 -> 1.28

---
Log message:

Fix 80 character formatting


---
Diffs of the changes:

Index: llvm/lib/Analysis/LoopInfo.cpp
diff -u llvm/lib/Analysis/LoopInfo.cpp:1.27 llvm/lib/Analysis/LoopInfo.cpp:1.28
--- llvm/lib/Analysis/LoopInfo.cpp:1.27	Wed Feb 19 18:17:17 2003
+++ llvm/lib/Analysis/LoopInfo.cpp	Wed Feb 19 18:18:07 2003
@@ -149,10 +149,10 @@
         if (BBMI == BBMap.end() || BBMI->first != *I) {  // Not in map yet...
           BBMap.insert(BBMI, std::make_pair(*I, L));
         } else {
-          // If this is already in the BBMap then this means that we already added
-          // a loop for it, but incorrectly added the loop to a higher level loop
-          // instead of the current loop we are creating.  Fix this now by moving
-          // the loop into the correct subloop.
+          // If this is already in the BBMap then this means that we already
+          // added a loop for it, but incorrectly added the loop to a higher
+          // level loop instead of the current loop we are creating.  Fix this
+          // now by moving the loop into the correct subloop.
           //
           Loop *SubLoop = BBMI->second;
           Loop *OldSubLoopParent = SubLoop->getParentLoop();




From lattner at cs.uiuc.edu  Wed Feb 19 18:29:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Wed Feb 19 18:29:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Analysis/LoopInfo.cpp
Message-ID: <200302200028.SAA19952@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Analysis:

LoopInfo.cpp updated: 1.28 -> 1.29

---
Log message:

Fix the requisite bug that I introduced


---
Diffs of the changes:

Index: llvm/lib/Analysis/LoopInfo.cpp
diff -u llvm/lib/Analysis/LoopInfo.cpp:1.28 llvm/lib/Analysis/LoopInfo.cpp:1.29
--- llvm/lib/Analysis/LoopInfo.cpp:1.28	Wed Feb 19 18:18:07 2003
+++ llvm/lib/Analysis/LoopInfo.cpp	Wed Feb 19 18:28:00 2003
@@ -155,17 +155,19 @@
           // now by moving the loop into the correct subloop.
           //
           Loop *SubLoop = BBMI->second;
-          Loop *OldSubLoopParent = SubLoop->getParentLoop();
-          if (OldSubLoopParent != L) {
-            // Remove SubLoop from OldSubLoopParent's list of subloops...
-            std::vector<Loop*>::iterator I =
-              std::find(OldSubLoopParent->SubLoops.begin(),
-                        OldSubLoopParent->SubLoops.end(), SubLoop);
-            assert(I != OldSubLoopParent->SubLoops.end()
-                   && "Loop parent doesn't contain loop?");
-            OldSubLoopParent->SubLoops.erase(I);
-            SubLoop->ParentLoop = L;
-            L->SubLoops.push_back(SubLoop);
+          if (SubLoop->getHeader() == *I) { // Only do this once for the loop...
+            Loop *OldSubLoopParent = SubLoop->getParentLoop();
+            if (OldSubLoopParent != L) {
+              // Remove SubLoop from OldSubLoopParent's list of subloops...
+              std::vector<Loop*>::iterator I =
+                std::find(OldSubLoopParent->SubLoops.begin(),
+                          OldSubLoopParent->SubLoops.end(), SubLoop);
+              assert(I != OldSubLoopParent->SubLoops.end()
+                     && "Loop parent doesn't contain loop?");
+              OldSubLoopParent->SubLoops.erase(I);
+              SubLoop->ParentLoop = L;
+              L->SubLoops.push_back(SubLoop);
+            }
           }
         }
       }




From lattner at cs.uiuc.edu  Sat Feb 22 15:34:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sat Feb 22 15:34:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Analysis/LoopInfo.cpp
Message-ID: <200302222133.PAA28190@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Analysis:

LoopInfo.cpp updated: 1.29 -> 1.30

---
Log message:

Dramatically simplify building of natural loops and fix a bug where the BBMap 
was not correctly computed.


---
Diffs of the changes:

Index: llvm/lib/Analysis/LoopInfo.cpp
diff -u llvm/lib/Analysis/LoopInfo.cpp:1.29 llvm/lib/Analysis/LoopInfo.cpp:1.30
--- llvm/lib/Analysis/LoopInfo.cpp:1.29	Wed Feb 19 18:28:00 2003
+++ llvm/lib/Analysis/LoopInfo.cpp	Sat Feb 22 15:33:11 2003
@@ -60,6 +60,7 @@
     getSubLoops()[i]->print(OS);
 }
 
+
 //===----------------------------------------------------------------------===//
 // LoopInfo implementation
 //
@@ -101,6 +102,12 @@
 void LoopInfo::print(std::ostream &OS) const {
   for (unsigned i = 0; i < TopLevelLoops.size(); ++i)
     TopLevelLoops[i]->print(OS);
+#if 0
+  for (std::map<BasicBlock*, Loop*>::const_iterator I = BBMap.begin(),
+         E = BBMap.end(); I != E; ++I)
+    OS << "BB '" << I->first->getName() << "' level = "
+       << I->second->LoopDepth << "\n";
+#endif
 }
 
 Loop *LoopInfo::ConsiderForLoop(BasicBlock *BB, const DominatorSet &DS) {
@@ -132,45 +139,24 @@
     }
   }
 
+  // If there are any loops nested within this loop, create them now!
+  for (std::vector<BasicBlock*>::iterator I = L->Blocks.begin(),
+	 E = L->Blocks.end(); I != E; ++I)
+    if (Loop *NewLoop = ConsiderForLoop(*I, DS)) {
+      L->SubLoops.push_back(NewLoop);
+      NewLoop->ParentLoop = L;
+    }
+
+
   // Add the basic blocks that comprise this loop to the BBMap so that this
-  // loop can be found for them.  Also check subsidary basic blocks to see if
-  // they start subloops of their own.
+  // loop can be found for them.
   //
-  for (std::vector<BasicBlock*>::reverse_iterator I = L->Blocks.rbegin(),
-	 E = L->Blocks.rend(); I != E; ++I)
-
-    // Check to see if this block starts a new loop
-    if (*I != BB)
-      if (Loop *NewLoop = ConsiderForLoop(*I, DS)) {
-        L->SubLoops.push_back(NewLoop);
-        NewLoop->ParentLoop = L;
-      } else {
-        std::map<BasicBlock*, Loop*>::iterator BBMI = BBMap.lower_bound(*I);
-        if (BBMI == BBMap.end() || BBMI->first != *I) {  // Not in map yet...
-          BBMap.insert(BBMI, std::make_pair(*I, L));
-        } else {
-          // If this is already in the BBMap then this means that we already
-          // added a loop for it, but incorrectly added the loop to a higher
-          // level loop instead of the current loop we are creating.  Fix this
-          // now by moving the loop into the correct subloop.
-          //
-          Loop *SubLoop = BBMI->second;
-          if (SubLoop->getHeader() == *I) { // Only do this once for the loop...
-            Loop *OldSubLoopParent = SubLoop->getParentLoop();
-            if (OldSubLoopParent != L) {
-              // Remove SubLoop from OldSubLoopParent's list of subloops...
-              std::vector<Loop*>::iterator I =
-                std::find(OldSubLoopParent->SubLoops.begin(),
-                          OldSubLoopParent->SubLoops.end(), SubLoop);
-              assert(I != OldSubLoopParent->SubLoops.end()
-                     && "Loop parent doesn't contain loop?");
-              OldSubLoopParent->SubLoops.erase(I);
-              SubLoop->ParentLoop = L;
-              L->SubLoops.push_back(SubLoop);
-            }
-          }
-        }
-      }
+  for (std::vector<BasicBlock*>::iterator I = L->Blocks.begin(),
+	 E = L->Blocks.end(); I != E; ++I) {
+    std::map<BasicBlock*, Loop*>::iterator BBMI = BBMap.lower_bound(*I);
+    if (BBMI == BBMap.end() || BBMI->first != *I)  // Not in map yet...
+      BBMap.insert(BBMI, std::make_pair(*I, L));   // Must be at this level
+  }
 
   return L;
 }




From lattner at cs.uiuc.edu  Sat Feb 22 16:21:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sat Feb 22 16:21:01 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/LICM/scalar_promote.ll
Message-ID: <200302222220.QAA29061@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/LICM:

scalar_promote.ll added (r1.1)

---
Log message:

New testcase to check to see if LICM is performing scalar promotion


---
Diffs of the changes:



From lattner at cs.uiuc.edu  Sat Feb 22 16:26:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sat Feb 22 16:26:01 2003
Subject: [llvm-commits] CVS: llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp
Message-ID: <200302222225.QAA29124@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Transforms/Utils:

PromoteMemoryToRegister.cpp updated: 1.33 -> 1.34

---
Log message:

Clean up std namespace references


---
Diffs of the changes:

Index: llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp
diff -u llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp:1.33 llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp:1.34
--- llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp:1.33	Mon Oct 21 15:00:26 2002
+++ llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp	Sat Feb 22 16:25:17 2003
@@ -13,7 +13,9 @@
 // Currently this just loops over all alloca instructions, looking for
 // instructions that are only used in simple load and stores.
 //
-// After this, the code is transformed by...something magical :)
+// After this, the code is transformed by looping over all of the alloca
+// instruction, calculating dominator frontiers, then inserting phi-nodes
+// following the usual SSA construction algorithm.
 //
 //===----------------------------------------------------------------------===//
 
@@ -27,23 +29,20 @@
 #include "llvm/Type.h"
 #include "Support/Statistic.h"
 
-using std::vector;
-using std::map;
-using std::set;
-
 namespace {
   Statistic<> NumPromoted("mem2reg", "Number of alloca's promoted");
 
   struct PromotePass : public FunctionPass {
-    vector<AllocaInst*>          Allocas;      // the alloca instruction..
-    map<Instruction*, unsigned>  AllocaLookup; // reverse mapping of above
+    std::vector<AllocaInst*>          Allocas;      // the alloca instruction..
+    std::map<Instruction*, unsigned>  AllocaLookup; // reverse mapping of above
     
-    vector<vector<BasicBlock*> > PhiNodes;     // index corresponds to Allocas
+    std::vector<std::vector<BasicBlock*> > PhiNodes;// Idx corresponds 2 Allocas
     
     // List of instructions to remove at end of pass
-    vector<Instruction *>        KillList;
+    std::vector<Instruction *>        KillList;
     
-    map<BasicBlock*,vector<PHINode*> > NewPhiNodes; // the PhiNodes we're adding
+    std::map<BasicBlock*,
+             std::vector<PHINode*> >  NewPhiNodes; // the PhiNodes we're adding
 
   public:
     // runOnFunction - To run this pass, first we calculate the alloca
@@ -59,8 +58,9 @@
     }
 
   private:
-    void Traverse(BasicBlock *BB, BasicBlock *Pred, vector<Value*> &IncVals,
-                  set<BasicBlock*> &Visited);
+    void RenamePass(BasicBlock *BB, BasicBlock *Pred,
+                    std::vector<Value*> &IncVals,
+                    std::set<BasicBlock*> &Visited);
     bool QueuePhiNode(BasicBlock *BB, unsigned AllocaIdx);
     void FindSafeAllocas(Function &F);
   };
@@ -95,7 +95,7 @@
   BasicBlock &BB = F.getEntryNode();  // Get the entry node for the function
 
   // Look at all instructions in the entry node
-  for (BasicBlock::iterator I = BB.begin(), E = BB.end(); I != E; ++I)
+  for (BasicBlock::iterator I = BB.begin(), E = --BB.end(); I != E; ++I)
     if (AllocaInst *AI = dyn_cast<AllocaInst>(&*I))       // Is it an alloca?
       if (isSafeAlloca(AI)) {   // If safe alloca, add alloca to safe list
         AllocaLookup[AI] = Allocas.size();  // Keep reverse mapping
@@ -118,7 +118,7 @@
 
   // Calculate the set of write-locations for each alloca.  This is analogous to
   // counting the number of 'redefinitions' of each variable.
-  vector<vector<BasicBlock*> > WriteSets;    // index corresponds to Allocas
+  std::vector<std::vector<BasicBlock*> > WriteSets;// Idx corresponds to Allocas
   WriteSets.resize(Allocas.size());
   for (unsigned i = 0; i != Allocas.size(); ++i) {
     AllocaInst *AI = Allocas[i];
@@ -159,15 +159,15 @@
   // the alloca's.  We do this in case there is a load of a value that has not
   // been stored yet.  In this case, it will get this null value.
   //
-  vector<Value *> Values(Allocas.size());
+  std::vector<Value *> Values(Allocas.size());
   for (unsigned i = 0, e = Allocas.size(); i != e; ++i)
     Values[i] = Constant::getNullValue(Allocas[i]->getAllocatedType());
 
   // Walks all basic blocks in the function performing the SSA rename algorithm
   // and inserting the phi nodes we marked as necessary
   //
-  set<BasicBlock*> Visited;         // The basic blocks we've already visited
-  Traverse(F.begin(), 0, Values, Visited);
+  std::set<BasicBlock*> Visited;      // The basic blocks we've already visited
+  RenamePass(F.begin(), 0, Values, Visited);
 
   // Remove all instructions marked by being placed in the KillList...
   //
@@ -194,7 +194,7 @@
 //
 bool PromotePass::QueuePhiNode(BasicBlock *BB, unsigned AllocaNo) {
   // Look up the basic-block in question
-  vector<PHINode*> &BBPNs = NewPhiNodes[BB];
+  std::vector<PHINode*> &BBPNs = NewPhiNodes[BB];
   if (BBPNs.empty()) BBPNs.resize(Allocas.size());
 
   // If the BB already has a phi node added for the i'th alloca then we're done!
@@ -210,12 +210,12 @@
   return true;
 }
 
-void PromotePass::Traverse(BasicBlock *BB, BasicBlock *Pred,
-                           vector<Value*> &IncomingVals,
-                           set<BasicBlock*> &Visited) {
+void PromotePass::RenamePass(BasicBlock *BB, BasicBlock *Pred,
+                             std::vector<Value*> &IncomingVals,
+                             std::set<BasicBlock*> &Visited) {
   // If this is a BB needing a phi node, lookup/create the phinode for each
   // variable we need phinodes for.
-  vector<PHINode *> &BBPNs = NewPhiNodes[BB];
+  std::vector<PHINode *> &BBPNs = NewPhiNodes[BB];
   for (unsigned k = 0; k != BBPNs.size(); ++k)
     if (PHINode *PN = BBPNs[k]) {
       // at this point we can assume that the array has phi nodes.. let's add
@@ -237,10 +237,8 @@
     Instruction *I = II; // get the instruction
 
     if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
-      Value *Ptr = LI->getPointerOperand();
-
-      if (AllocaInst *Src = dyn_cast<AllocaInst>(Ptr)) {
-        map<Instruction*, unsigned>::iterator AI = AllocaLookup.find(Src);
+      if (AllocaInst *Src = dyn_cast<AllocaInst>(LI->getPointerOperand())) {
+        std::map<Instruction*, unsigned>::iterator AI = AllocaLookup.find(Src);
         if (AI != AllocaLookup.end()) {
           Value *V = IncomingVals[AI->second];
 
@@ -250,11 +248,10 @@
         }
       }
     } else if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
-      // delete this instruction and mark the name as the current holder of the
+      // Delete this instruction and mark the name as the current holder of the
       // value
-      Value *Ptr = SI->getPointerOperand();
-      if (AllocaInst *Dest = dyn_cast<AllocaInst>(Ptr)) {
-        map<Instruction *, unsigned>::iterator ai = AllocaLookup.find(Dest);
+      if (AllocaInst *Dest = dyn_cast<AllocaInst>(SI->getPointerOperand())) {
+        std::map<Instruction *, unsigned>::iterator ai =AllocaLookup.find(Dest);
         if (ai != AllocaLookup.end()) {
           // what value were we writing?
           IncomingVals[ai->second] = SI->getOperand(0);
@@ -265,8 +262,8 @@
     } else if (TerminatorInst *TI = dyn_cast<TerminatorInst>(I)) {
       // Recurse across our successors
       for (unsigned i = 0; i != TI->getNumSuccessors(); i++) {
-        vector<Value*> OutgoingVals(IncomingVals);
-        Traverse(TI->getSuccessor(i), BB, OutgoingVals, Visited);
+        std::vector<Value*> OutgoingVals(IncomingVals);
+        RenamePass(TI->getSuccessor(i), BB, OutgoingVals, Visited);
       }
     }
   }




From lattner at cs.uiuc.edu  Sat Feb 22 17:06:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sat Feb 22 17:06:01 2003
Subject: [llvm-commits] CVS: llvm/include/llvm/Transforms/Utils/PromoteMemToReg.h
Message-ID: <200302222305.RAA29381@apoc.cs.uiuc.edu>


Changes in directory llvm/include/llvm/Transforms/Utils:

PromoteMemToReg.h added (r1.1)

---
Log message:

Add a new interface file for the PromoteMemToReg interface in the Utils library


---
Diffs of the changes:



From lattner at cs.uiuc.edu  Sat Feb 22 17:58:00 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sat Feb 22 17:58:00 2003
Subject: [llvm-commits] CVS: llvm/lib/Transforms/Scalar/Mem2Reg.cpp
Message-ID: <200302222357.RAA30650@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Transforms/Scalar:

Mem2Reg.cpp added (r1.1)

---
Log message:

Split mem2reg promotion into two parts: a function which does the work, and
a pass which wraps the function.  This allows other passes to use the functionality


---
Diffs of the changes:



From lattner at cs.uiuc.edu  Sat Feb 22 17:58:02 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sat Feb 22 17:58:02 2003
Subject: [llvm-commits] CVS: llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp
Message-ID: <200302222357.RAA30659@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Transforms/Utils:

PromoteMemoryToRegister.cpp updated: 1.34 -> 1.35

---
Log message:

Split mem2reg promotion into two parts: a function which does the work, and
a pass which wraps the function.  This allows other passes to use the functionality



---
Diffs of the changes:

Index: llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp
diff -u llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp:1.34 llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp:1.35
--- llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp:1.34	Sat Feb 22 16:25:17 2003
+++ llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp	Sat Feb 22 17:57:48 2003
@@ -1,7 +1,7 @@
 //===- PromoteMemoryToRegister.cpp - Convert memory refs to regs ----------===//
 //
-// This pass is used to promote memory references to be register references.  A
-// simple example of the transformation performed by this pass is:
+// This file is used to promote memory references to be register references.  A
+// simple example of the transformation performed by this function is:
 //
 //        FROM CODE                           TO CODE
 //   %X = alloca int, uint 1                 ret int 42
@@ -9,17 +9,14 @@
 //   %Y = load int* %X
 //   ret int %Y
 //
-// To do this transformation, a simple analysis is done to ensure it is safe.
-// Currently this just loops over all alloca instructions, looking for
-// instructions that are only used in simple load and stores.
-//
-// After this, the code is transformed by looping over all of the alloca
-// instruction, calculating dominator frontiers, then inserting phi-nodes
-// following the usual SSA construction algorithm.
+// The code is transformed by looping over all of the alloca instruction,
+// calculating dominator frontiers, then inserting phi-nodes following the usual
+// SSA construction algorithm.  This code does not modify the CFG of the
+// function.
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/Transforms/Scalar.h"
+#include "llvm/Transforms/Utils/PromoteMemToReg.h"
 #include "llvm/Analysis/Dominators.h"
 #include "llvm/iMemory.h"
 #include "llvm/iPHINode.h"
@@ -27,13 +24,31 @@
 #include "llvm/Function.h"
 #include "llvm/Constant.h"
 #include "llvm/Type.h"
-#include "Support/Statistic.h"
+
+/// isAllocaPromotable - Return true if this alloca is legal for promotion.
+/// This is true if there are only loads and stores to the alloca...
+///
+bool isAllocaPromotable(const AllocaInst *AI) {
+  // Only allow direct loads and stores...
+  for (Value::use_const_iterator UI = AI->use_begin(), UE = AI->use_end();
+       UI != UE; ++UI)     // Loop over all of the uses of the alloca
+    if (!isa<LoadInst>(*UI))
+      if (const StoreInst *SI = dyn_cast<StoreInst>(*UI)) {
+        if (SI->getOperand(0) == AI)
+          return false;   // Don't allow a store of the AI, only INTO the AI.
+      } else {
+        return false;   // Not a load or store?
+      }
+  
+  return true;
+}
+
 
 namespace {
-  Statistic<> NumPromoted("mem2reg", "Number of alloca's promoted");
+  struct PromoteMem2Reg {
+    const std::vector<AllocaInst*>   &Allocas;      // the alloca instructions..
+    DominanceFrontier &DF;
 
-  struct PromotePass : public FunctionPass {
-    std::vector<AllocaInst*>          Allocas;      // the alloca instruction..
     std::map<Instruction*, unsigned>  AllocaLookup; // reverse mapping of above
     
     std::vector<std::vector<BasicBlock*> > PhiNodes;// Idx corresponds 2 Allocas
@@ -45,72 +60,34 @@
              std::vector<PHINode*> >  NewPhiNodes; // the PhiNodes we're adding
 
   public:
-    // runOnFunction - To run this pass, first we calculate the alloca
-    // instructions that are safe for promotion, then we promote each one.
-    //
-    virtual bool runOnFunction(Function &F);
-
-    // getAnalysisUsage - We need dominance frontiers
-    //
-    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
-      AU.addRequired<DominanceFrontier>();
-      AU.setPreservesCFG();
-    }
+    PromoteMem2Reg(const std::vector<AllocaInst*> &A, DominanceFrontier &df)
+      :Allocas(A), DF(df) {}
+
+    void run();
 
   private:
     void RenamePass(BasicBlock *BB, BasicBlock *Pred,
                     std::vector<Value*> &IncVals,
                     std::set<BasicBlock*> &Visited);
     bool QueuePhiNode(BasicBlock *BB, unsigned AllocaIdx);
-    void FindSafeAllocas(Function &F);
   };
-
-  RegisterOpt<PromotePass> X("mem2reg", "Promote Memory to Register");
 }  // end of anonymous namespace
 
 
-// isSafeAlloca - This predicate controls what types of alloca instructions are
-// allowed to be promoted...
-//
-static inline bool isSafeAlloca(const AllocaInst *AI) {
-  if (AI->isArrayAllocation()) return false;
-
-  // Only allow direct loads and stores...
-  for (Value::use_const_iterator UI = AI->use_begin(), UE = AI->use_end();
-       UI != UE; ++UI)     // Loop over all of the uses of the alloca
-    if (!isa<LoadInst>(*UI))
-      if (const StoreInst *SI = dyn_cast<StoreInst>(*UI)) {
-        if (SI->getOperand(0) == AI)
-          return false;   // Don't allow a store of the AI, only INTO the AI.
-      } else {
-        return false;   // Not a load or store?
-      }
-  
-  return true;
-}
-
-// FindSafeAllocas - Find allocas that are safe to promote
-//
-void PromotePass::FindSafeAllocas(Function &F) {
-  BasicBlock &BB = F.getEntryNode();  // Get the entry node for the function
-
-  // Look at all instructions in the entry node
-  for (BasicBlock::iterator I = BB.begin(), E = --BB.end(); I != E; ++I)
-    if (AllocaInst *AI = dyn_cast<AllocaInst>(&*I))       // Is it an alloca?
-      if (isSafeAlloca(AI)) {   // If safe alloca, add alloca to safe list
-        AllocaLookup[AI] = Allocas.size();  // Keep reverse mapping
-        Allocas.push_back(AI);
-      }
-}
-
+void PromoteMem2Reg::run() {
+  // If there is nothing to do, bail out...
+  if (Allocas.empty()) return;
 
+  Function &F = *DF.getRoot()->getParent();
 
-bool PromotePass::runOnFunction(Function &F) {
-  // Calculate the set of safe allocas
-  FindSafeAllocas(F);
+  for (unsigned i = 0, e = Allocas.size(); i != e; ++i) {
+    assert(isAllocaPromotable(Allocas[i]) &&
+           "Cannot promote non-promotable alloca!");
+    assert(Allocas[i]->getParent()->getParent() == &F &&
+           "All allocas should be in the same function, which is same as DF!");
+    AllocaLookup[Allocas[i]] = i;
+  }
 
-  // If there is nothing to do, bail out...
-  if (Allocas.empty()) return false;
 
   // Add each alloca to the KillList.  Note: KillList is destroyed MOST recently
   // added to least recently.
@@ -128,9 +105,6 @@
         WriteSets[i].push_back(SI->getParent());
   }
 
-  // Get dominance frontier information...
-  DominanceFrontier &DF = getAnalysis<DominanceFrontier>();
-
   // Compute the locations where PhiNodes need to be inserted.  Look at the
   // dominance frontier of EACH basic-block we have a write in
   //
@@ -177,22 +151,13 @@
 
     I->getParent()->getInstList().erase(I);
   }
-
-  NumPromoted += Allocas.size();
-
-  // Purge data structurse so they are available the next iteration...
-  Allocas.clear();
-  AllocaLookup.clear();
-  PhiNodes.clear();
-  NewPhiNodes.clear();
-  return true;
 }
 
 
 // QueuePhiNode - queues a phi-node to be added to a basic-block for a specific
 // Alloca returns true if there wasn't already a phi-node for that variable
 //
-bool PromotePass::QueuePhiNode(BasicBlock *BB, unsigned AllocaNo) {
+bool PromoteMem2Reg::QueuePhiNode(BasicBlock *BB, unsigned AllocaNo) {
   // Look up the basic-block in question
   std::vector<PHINode*> &BBPNs = NewPhiNodes[BB];
   if (BBPNs.empty()) BBPNs.resize(Allocas.size());
@@ -210,7 +175,7 @@
   return true;
 }
 
-void PromotePass::RenamePass(BasicBlock *BB, BasicBlock *Pred,
+void PromoteMem2Reg::RenamePass(BasicBlock *BB, BasicBlock *Pred,
                              std::vector<Value*> &IncomingVals,
                              std::set<BasicBlock*> &Visited) {
   // If this is a BB needing a phi node, lookup/create the phinode for each
@@ -269,9 +234,12 @@
   }
 }
 
-
-// createPromoteMemoryToRegister - Provide an entry point to create this pass.
-//
-Pass *createPromoteMemoryToRegister() {
-  return new PromotePass();
+/// PromoteMemToReg - Promote the specified list of alloca instructions into
+/// scalar registers, inserting PHI nodes as appropriate.  This function makes
+/// use of DominanceFrontier information.  This function does not modify the CFG
+/// of the function at all.  All allocas must be from the same function.
+///
+void PromoteMemToReg(const std::vector<AllocaInst*> &Allocas,
+                     DominanceFrontier &DF) {
+  PromoteMem2Reg(Allocas, DF).run();
 }




From lattner at cs.uiuc.edu  Sun Feb 23 21:53:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sun Feb 23 21:53:01 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/LICM/scalar_promote.ll
Message-ID: <200302240352.VAA02643@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/LICM:

scalar_promote.ll updated: 1.1 -> 1.2

---
Log message:

New testcase for scalar promotion


---
Diffs of the changes:

Index: llvm/test/Regression/Transforms/LICM/scalar_promote.ll
diff -u llvm/test/Regression/Transforms/LICM/scalar_promote.ll:1.1 llvm/test/Regression/Transforms/LICM/scalar_promote.ll:1.2
--- llvm/test/Regression/Transforms/LICM/scalar_promote.ll:1.1	Sat Feb 22 16:20:19 2003
+++ llvm/test/Regression/Transforms/LICM/scalar_promote.ll	Sun Feb 23 21:52:04 2003
@@ -19,3 +19,18 @@
 Out:
 	ret void
 }
+
+void %testhard(int %i) {
+	br label %Loop
+Loop:
+	%X1 = getelementptr int* %X, long 0
+	%A = load int* %X1 ; Aliases X, needs to be rewritten
+	%V = add int %A, 1
+	%X2 = getelementptr int* %X, long 0
+	store int %V, int* %X2
+	br bool false, label %Loop, label %Exit
+
+Exit:
+	ret void
+
+}




From lattner at cs.uiuc.edu  Sun Feb 23 21:53:03 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sun Feb 23 21:53:03 2003
Subject: [llvm-commits] CVS: llvm/test/Regression/Transforms/BasicAA/licmtest.ll
Message-ID: <200302240352.VAA02654@apoc.cs.uiuc.edu>


Changes in directory llvm/test/Regression/Transforms/BasicAA:

licmtest.ll updated: 1.2 -> 1.3

---
Log message:

Tweak testcase


---
Diffs of the changes:

Index: llvm/test/Regression/Transforms/BasicAA/licmtest.ll
diff -u llvm/test/Regression/Transforms/BasicAA/licmtest.ll:1.2 llvm/test/Regression/Transforms/BasicAA/licmtest.ll:1.3
--- llvm/test/Regression/Transforms/BasicAA/licmtest.ll:1.2	Sat Sep  7 17:48:30 2002
+++ llvm/test/Regression/Transforms/BasicAA/licmtest.ll	Sun Feb 23 21:52:13 2003
@@ -22,7 +22,8 @@
 
 	br bool %c, label %Out, label %Loop
 Out:
-	ret int 7
+	%X = sub int %ToRemove, %Atmp
+	ret int %X
 
 Dummy:
 	store int 7, int* %A




From lattner at cs.uiuc.edu  Sun Feb 23 21:53:04 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sun Feb 23 21:53:04 2003
Subject: [llvm-commits] CVS: llvm/lib/Transforms/Scalar/LICM.cpp
Message-ID: <200302240352.VAA02663@apoc.cs.uiuc.edu>


Changes in directory llvm/lib/Transforms/Scalar:

LICM.cpp updated: 1.20 -> 1.21

---
Log message:

Initial implementation of Loop invariant memory->scalar promotion


---
Diffs of the changes:

Index: llvm/lib/Transforms/Scalar/LICM.cpp
diff -u llvm/lib/Transforms/Scalar/LICM.cpp:1.20 llvm/lib/Transforms/Scalar/LICM.cpp:1.21
--- llvm/lib/Transforms/Scalar/LICM.cpp:1.20	Mon Oct 21 15:00:26 2002
+++ llvm/lib/Transforms/Scalar/LICM.cpp	Sun Feb 23 21:52:32 2003
@@ -1,27 +1,147 @@
 //===-- LICM.cpp - Loop Invariant Code Motion Pass ------------------------===//
 //
-// This pass is a simple loop invariant code motion pass.
+// This pass is a simple loop invariant code motion pass.  An interesting aspect
+// of this pass is that it uses alias analysis for two purposes:
+//
+//  1. Moving loop invariant loads out of loops.  If we can determine that a
+//     load inside of a loop never aliases anything stored to, we can hoist it
+//     like any other instruction.
+//  2. Scalar Promotion of Memory - If there is a store instruction inside of
+//     the loop, we try to move the store to happen AFTER the loop instead of
+//     inside of the loop.  This can only happen if a few conditions are true:
+//       A. The pointer stored through is loop invariant
+//       B. There are no stores or loads in the loop which _may_ alias the
+//          pointer.  There are no calls in the loop which mod/ref the pointer.
+//     If these conditions are true, we can promote the loads and stores in the
+//     loop of the pointer to use a temporary alloca'd variable.  We then use
+//     the mem2reg functionality to construct the appropriate SSA form for the
+//     variable.
 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Transforms/Scalar.h"
+#include "llvm/Transforms/Utils/PromoteMemToReg.h"
 #include "llvm/Transforms/Utils/Local.h"
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Analysis/Dominators.h"
-#include "llvm/iOperators.h"
-#include "llvm/iMemory.h"
+#include "llvm/Instructions.h"
+#include "llvm/DerivedTypes.h"
 #include "llvm/Support/InstVisitor.h"
-#include "Support/STLExtras.h"
+#include "llvm/Support/CFG.h"
 #include "Support/Statistic.h"
+#include "Support/CommandLine.h"
 #include "llvm/Assembly/Writer.h"
 #include <algorithm>
-using std::string;
 
 namespace {
+  cl::opt<bool> DisablePromotion("disable-licm-promotion", cl::Hidden,
+                             cl::desc("Disable memory promotion in LICM pass"));
+
   Statistic<> NumHoisted("licm", "Number of instructions hoisted out of loop");
   Statistic<> NumHoistedLoads("licm", "Number of load insts hoisted");
+  Statistic<> NumPromoted("licm", "Number of memory locations promoted to registers");
+
+  /// LoopBodyInfo - We recursively traverse loops from most-deeply-nested to
+  /// least-deeply-nested.  For all of the loops nested within the current one,
+  /// we keep track of information so that we don't have to repeat queries.
+  ///
+  struct LoopBodyInfo {
+    std::vector<CallInst*> Calls;          // Call instructions in loop
+    std::vector<InvokeInst*> Invokes;      // Invoke instructions in loop
+
+    // StoredPointers - Targets of store instructions...
+    std::set<Value*> StoredPointers;
+
+    // LoadedPointers - Source pointers for load instructions...
+    std::set<Value*> LoadedPointers;
+
+    enum PointerClass {
+      PointerUnknown = 0, // Nothing is known about this pointer yet
+      PointerMustStore,   // Memory is stored to ONLY through this pointer
+      PointerMayStore,    // Memory is stored to through this or other pointers
+      PointerNoStore      // Memory is not modified in this loop
+    };
+
+    // PointerIsModified - Keep track of information as we find out about it in
+    // the loop body...
+    //
+    std::map<Value*, enum PointerClass> PointerIsModified;
+
+    /// CantModifyAnyPointers - Return true if no memory modifying instructions
+    /// occur in this loop.  This is just a conservative approximation, because
+    /// a call may not actually store anything.
+    bool CantModifyAnyPointers() const {
+      return Calls.empty() && Invokes.empty() && StoredPointers.empty();
+    }
 
+    /// incorporate - Incorporate information about a subloop into the current
+    /// loop.
+    void incorporate(const LoopBodyInfo &OtherLBI);
+    void incorporate(BasicBlock &BB);  // do the same for a basic block
+
+    PointerClass getPointerInfo(Value *V, AliasAnalysis &AA) {
+      PointerClass &VInfo = PointerIsModified[V];
+      if (VInfo == PointerUnknown)
+        VInfo = calculatePointerInfo(V, AA);
+      return VInfo;
+    }
+  private:
+    /// calculatePointerInfo - Calculate information about the specified
+    /// pointer.
+    PointerClass calculatePointerInfo(Value *V, AliasAnalysis &AA) const;
+  };
+}
+
+/// incorporate - Incorporate information about a subloop into the current loop.
+void LoopBodyInfo::incorporate(const LoopBodyInfo &OtherLBI) {
+  // Do not incorporate NonModifiedPointers (which is just a cache) because it
+  // is too much trouble to make sure it's still valid.
+  Calls.insert  (Calls.end(),  OtherLBI.Calls.begin(),  OtherLBI.Calls.end());
+  Invokes.insert(Invokes.end(),OtherLBI.Invokes.begin(),OtherLBI.Invokes.end());
+  StoredPointers.insert(OtherLBI.StoredPointers.begin(),
+                        OtherLBI.StoredPointers.end());
+  LoadedPointers.insert(OtherLBI.LoadedPointers.begin(),
+                        OtherLBI.LoadedPointers.end());
+}
+
+void LoopBodyInfo::incorporate(BasicBlock &BB) {
+  for (BasicBlock::iterator I = BB.begin(), E = --BB.end(); I != E; ++I)
+    if (CallInst *CI = dyn_cast<CallInst>(&*I))
+      Calls.push_back(CI);
+    else if (StoreInst *SI = dyn_cast<StoreInst>(&*I))
+      StoredPointers.insert(SI->getOperand(1));
+    else if (LoadInst *LI = dyn_cast<LoadInst>(&*I))
+      LoadedPointers.insert(LI->getOperand(0));
+
+  if (InvokeInst *II = dyn_cast<InvokeInst>(BB.getTerminator()))
+    Invokes.push_back(II);
+}
+
+
+// calculatePointerInfo - Calculate information about the specified pointer.
+LoopBodyInfo::PointerClass LoopBodyInfo::calculatePointerInfo(Value *V,
+                                                      AliasAnalysis &AA) const {
+  for (unsigned i = 0, e = Calls.size(); i != e; ++i)
+    if (AA.canCallModify(*Calls[i], V))
+      return PointerMayStore;
+
+  for (unsigned i = 0, e = Invokes.size(); i != e; ++i)
+    if (AA.canInvokeModify(*Invokes[i], V))
+      return PointerMayStore;
+
+  PointerClass Result = PointerNoStore;
+  for (std::set<Value*>::const_iterator I = StoredPointers.begin(),
+         E = StoredPointers.end(); I != E; ++I)
+    if (AA.alias(V, *I))
+      if (V == *I)
+        Result = PointerMustStore;   // If this is the only alias, return must
+      else
+        return PointerMayStore;      // We have to return may
+  return Result;
+}
+
+namespace {
   struct LICM : public FunctionPass, public InstVisitor<LICM> {
     virtual bool runOnFunction(Function &F);
 
@@ -33,18 +153,21 @@
       AU.addRequiredID(LoopPreheadersID);
       AU.addRequired<LoopInfo>();
       AU.addRequired<DominatorTree>();
+      AU.addRequired<DominanceFrontier>();
       AU.addRequired<AliasAnalysis>();
     }
 
   private:
-    Loop *CurLoop;         // The current loop we are working on...
-    BasicBlock *Preheader; // The preheader block of the current loop...
-    bool Changed;          // Set to true when we change anything.
-    AliasAnalysis *AA;     // Currently AliasAnalysis information
+    LoopInfo      *LI;       // Current LoopInfo
+    AliasAnalysis *AA;       // Current AliasAnalysis information
+    bool Changed;            // Set to true when we change anything.
+    BasicBlock *Preheader;   // The preheader block of the current loop...
+    Loop *CurLoop;           // The current loop we are working on...
+    LoopBodyInfo *CurLBI;    // Information about the current loop...
 
     /// visitLoop - Hoist expressions out of the specified loop...    
     ///
-    void visitLoop(Loop *L);
+    void visitLoop(Loop *L, LoopBodyInfo &LBI);
 
     /// HoistRegion - Walk the specified region of the CFG (defined by all
     /// blocks dominated by the specified block, and that are in the current
@@ -73,7 +196,10 @@
     /// pointerInvalidatedByLoop - Return true if the body of this loop may
     /// store into the memory location pointed to by V.
     /// 
-    bool pointerInvalidatedByLoop(Value *V);
+    bool pointerInvalidatedByLoop(Value *V) {
+      // Check to see if any of the basic blocks in CurLoop invalidate V.
+      return CurLBI->getPointerInfo(V, *AA) != LoopBodyInfo::PointerNoStore;
+    }
 
     /// isLoopInvariant - Return true if the specified value is loop invariant
     ///
@@ -83,6 +209,22 @@
       return true;  // All non-instructions are loop invariant
     }
 
+    /// PromoteValuesInLoop - Look at the stores in the loop and promote as many
+    /// to scalars as we can.
+    ///
+    void PromoteValuesInLoop();
+
+    /// findPromotableValuesInLoop - Check the current loop for stores to
+    /// definate pointers, which are not loaded and stored through may aliases.
+    /// If these are found, create an alloca for the value, add it to the
+    /// PromotedValues list, and keep track of the mapping from value to
+    /// alloca...
+    ///
+    void findPromotableValuesInLoop(
+                   std::vector<std::pair<AllocaInst*, Value*> > &PromotedValues,
+                                    std::map<Value*, AllocaInst*> &Val2AlMap);
+    
+
     /// Instruction visitation handlers... these basically control whether or
     /// not the specified instruction types are hoisted.
     ///
@@ -116,35 +258,52 @@
 /// function, hoisting expressions out of loops if possible.
 ///
 bool LICM::runOnFunction(Function &) {
-  // Get information about the top level loops in the function...
-  const std::vector<Loop*> &TopLevelLoops =
-    getAnalysis<LoopInfo>().getTopLevelLoops();
+  Changed = false;
 
-  // Get our alias analysis information...
+  // Get our Loop and Alias Analysis information...
+  LI = &getAnalysis<LoopInfo>();
   AA = &getAnalysis<AliasAnalysis>();
 
-  // Traverse loops in postorder, hoisting expressions out of the deepest loops
-  // first.
-  //
-  Changed = false;
-  std::for_each(TopLevelLoops.begin(), TopLevelLoops.end(),
-                bind_obj(this, &LICM::visitLoop));
+  // Hoist expressions out of all of the top-level loops.
+  const std::vector<Loop*> &TopLevelLoops = LI->getTopLevelLoops();
+  for (std::vector<Loop*>::const_iterator I = TopLevelLoops.begin(),
+         E = TopLevelLoops.end(); I != E; ++I) {
+    LoopBodyInfo LBI;
+    LICM::visitLoop(*I, LBI);
+  }
   return Changed;
 }
 
 
 /// visitLoop - Hoist expressions out of the specified loop...    
 ///
-void LICM::visitLoop(Loop *L) {
+void LICM::visitLoop(Loop *L, LoopBodyInfo &LBI) {
   // Recurse through all subloops before we process this loop...
-  std::for_each(L->getSubLoops().begin(), L->getSubLoops().end(),
-                bind_obj(this, &LICM::visitLoop));
+  for (std::vector<Loop*>::const_iterator I = L->getSubLoops().begin(),
+         E = L->getSubLoops().end(); I != E; ++I) {
+    LoopBodyInfo SubLBI;
+    LICM::visitLoop(*I, SubLBI);
+
+    // Incorporate information about the subloops into this loop...
+    LBI.incorporate(SubLBI);
+  }
   CurLoop = L;
+  CurLBI = &LBI;
 
   // Get the preheader block to move instructions into...
   Preheader = L->getLoopPreheader();
   assert(Preheader&&"Preheader insertion pass guarantees we have a preheader!");
 
+  // Loop over the body of this loop, looking for calls, invokes, and stores.
+  // Because subloops have already been incorporated into LBI, we skip blocks in
+  // subloops.
+  //
+  const std::vector<BasicBlock*> &LoopBBs = L->getBlocks();
+  for (std::vector<BasicBlock*>::const_iterator I = LoopBBs.begin(),
+         E = LoopBBs.end(); I != E; ++I)
+    if (LI->getLoopFor(*I) == L)        // Ignore blocks in subloops...
+      LBI.incorporate(**I);             // Incorporate the specified basic block
+
   // We want to visit all of the instructions in this loop... that are not parts
   // of our subloops (they have already had their invariants hoisted out of
   // their loop, into this loop, so there is no need to process the BODIES of
@@ -156,6 +315,11 @@
   //
   HoistRegion(getAnalysis<DominatorTree>()[L->getHeader()]);
 
+  // Now that all loop invariants have been removed from the loop, promote any
+  // memory references to scalars that we can...
+  if (!DisablePromotion)
+    PromoteValuesInLoop();
+
   // Clear out loops state information for the next iteration
   CurLoop = 0;
   Preheader = 0;
@@ -211,13 +375,144 @@
   }
 }
 
-/// pointerInvalidatedByLoop - Return true if the body of this loop may store
-/// into the memory location pointed to by V.
-/// 
-bool LICM::pointerInvalidatedByLoop(Value *V) {
-  // Check to see if any of the basic blocks in CurLoop invalidate V.
-  for (unsigned i = 0, e = CurLoop->getBlocks().size(); i != e; ++i)
-    if (AA->canBasicBlockModify(*CurLoop->getBlocks()[i], V))
-      return true;
-  return false;
+/// PromoteValuesInLoop - 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
+/// which are loop invariant.  We promote these memory locations to use allocas
+/// instead.  These allocas can easily be raised to register values by the
+/// PromoteMem2Reg functionality.
+///
+void LICM::PromoteValuesInLoop() {
+  // PromotedValues - List of values that are promoted out of the loop.  Each
+  // value has an alloca instruction for it, and a cannonical version of the
+  // pointer.
+  std::vector<std::pair<AllocaInst*, Value*> > PromotedValues;
+  std::map<Value*, AllocaInst*> ValueToAllocaMap; // Map of ptr to alloca
+
+  findPromotableValuesInLoop(PromotedValues, ValueToAllocaMap);
+  if (ValueToAllocaMap.empty()) return;   // If there are values to promote...
+
+  Changed = true;
+  NumPromoted += PromotedValues.size();
+
+  // Emit a copy from the value into the alloca'd value in the loop preheader
+  TerminatorInst *LoopPredInst = Preheader->getTerminator();
+  for (unsigned i = 0, e = PromotedValues.size(); i != e; ++i) {
+    // Load from the memory we are promoting...
+    LoadInst *LI = new LoadInst(PromotedValues[i].second, 
+                                PromotedValues[i].second->getName()+".promoted",
+                                LoopPredInst);
+    // Store into the temporary alloca...
+    new StoreInst(LI, PromotedValues[i].first, LoopPredInst);
+  }
+  
+  // Scan the basic blocks in the loop, replacing uses of our pointers with
+  // uses of the allocas in question.  If we find a branch that exits the
+  // loop, make sure to put reload code into all of the successors of the
+  // loop.
+  //
+  const std::vector<BasicBlock*> &LoopBBs = CurLoop->getBlocks();
+  for (std::vector<BasicBlock*>::const_iterator I = LoopBBs.begin(),
+         E = LoopBBs.end(); I != E; ++I) {
+    // Rewrite all loads and stores in the block of the pointer...
+    for (BasicBlock::iterator II = (*I)->begin(), E = (*I)->end();
+         II != E; ++II) {
+      if (LoadInst *L = dyn_cast<LoadInst>(&*II)) {
+        std::map<Value*, AllocaInst*>::iterator
+          I = ValueToAllocaMap.find(L->getOperand(0));
+        if (I != ValueToAllocaMap.end())
+          L->setOperand(0, I->second);    // Rewrite load instruction...
+      } else if (StoreInst *S = dyn_cast<StoreInst>(&*II)) {
+        std::map<Value*, AllocaInst*>::iterator
+          I = ValueToAllocaMap.find(S->getOperand(1));
+        if (I != ValueToAllocaMap.end())
+          S->setOperand(1, I->second);    // Rewrite store instruction...
+      }
+    }
+
+    // Check to see if any successors of this block are outside of the loop.
+    // If so, we need to copy the value from the alloca back into the memory
+    // location...
+    //
+    for (succ_iterator SI = succ_begin(*I), SE = succ_end(*I); SI != SE; ++SI)
+      if (!CurLoop->contains(*SI)) {
+        // Copy all of the allocas into their memory locations...
+        Instruction *InsertPos = (*SI)->begin();
+        for (unsigned i = 0, e = PromotedValues.size(); i != e; ++i) {
+          // Load from the alloca...
+          LoadInst *LI = new LoadInst(PromotedValues[i].first, "", InsertPos);
+          // Store into the memory we promoted...
+          new StoreInst(LI, PromotedValues[i].second, InsertPos);
+        }
+      }
+  }
+
+  // Now that we have done the deed, use the mem2reg functionality to promote
+  // all of the new allocas we just created into real SSA registers...
+  //
+  std::vector<AllocaInst*> PromotedAllocas;
+  PromotedAllocas.reserve(PromotedValues.size());
+  for (unsigned i = 0, e = PromotedValues.size(); i != e; ++i)
+    PromotedAllocas.push_back(PromotedValues[i].first);
+  PromoteMemToReg(PromotedAllocas, getAnalysis<DominanceFrontier>());
+}
+
+/// findPromotableValuesInLoop - Check the current loop for stores to definate
+/// pointers, which are not loaded and stored through may aliases.  If these are
+/// found, create an alloca for the value, add it to the PromotedValues list,
+/// and keep track of the mapping from value to alloca...
+///
+void LICM::findPromotableValuesInLoop(
+                   std::vector<std::pair<AllocaInst*, Value*> > &PromotedValues,
+                             std::map<Value*, AllocaInst*> &ValueToAllocaMap) {
+  Instruction *FnStart = CurLoop->getHeader()->getParent()->begin()->begin();
+
+  for (std::set<Value*>::iterator I = CurLBI->StoredPointers.begin(),
+         E = CurLBI->StoredPointers.end(); I != E; ++I) {
+    Value *V = *I;
+    if (isLoopInvariant(V) &&
+        CurLBI->getPointerInfo(V, *AA) == LoopBodyInfo::PointerMustStore) {
+
+      // Don't add a new entry for this stored pointer if it aliases something
+      // we have already processed.
+      std::map<Value*, AllocaInst*>::iterator V2AMI = 
+        ValueToAllocaMap.lower_bound(V);
+      if (V2AMI == ValueToAllocaMap.end() || V2AMI->first != V) {
+        // Check to make sure that any loads in the loop are either NO or MUST
+        // aliases.  We cannot rewrite loads that _might_ come from this memory
+        // location.
+
+        bool PointerOk = true;
+        for (std::set<Value*>::const_iterator I =CurLBI->LoadedPointers.begin(),
+               E = CurLBI->LoadedPointers.end(); I != E; ++I)
+          if (AA->alias(V, *I) == AliasAnalysis::MayAlias) {
+            PointerOk = false;
+            break;
+          }
+
+        if (PointerOk) {
+          const Type *Ty = cast<PointerType>(V->getType())->getElementType();
+          AllocaInst *AI = new AllocaInst(Ty, 0, V->getName()+".tmp", FnStart);
+          PromotedValues.push_back(std::make_pair(AI, V));
+          ValueToAllocaMap.insert(V2AMI, std::make_pair(V, AI));
+
+          DEBUG(std::cerr << "LICM: Promoting value: " << *V << "\n");
+
+          // Loop over all of the loads and stores that alias this pointer,
+          // adding them to the Value2AllocaMap as well...
+          for (std::set<Value*>::const_iterator
+                 I = CurLBI->LoadedPointers.begin(),
+                 E = CurLBI->LoadedPointers.end(); I != E; ++I)
+            if (AA->alias(V, *I) == AliasAnalysis::MustAlias)
+              ValueToAllocaMap[*I] = AI;
+
+          for (std::set<Value*>::const_iterator
+                 I = CurLBI->StoredPointers.begin(),
+                 E = CurLBI->StoredPointers.end(); I != E; ++I)
+            if (AA->alias(V, *I) == AliasAnalysis::MustAlias)
+              ValueToAllocaMap[*I] = AI;
+        }
+      }
+    }
+  }
 }




From lattner at cs.uiuc.edu  Sun Feb 23 22:32:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sun Feb 23 22:32:01 2003
Subject: [llvm-commits] CVS: llvm/include/Support/slist
Message-ID: <200302240431.WAA02833@apoc.cs.uiuc.edu>


Changes in directory llvm/include/Support:

slist added (r1.1)

---
Log message:

Add support for the slist extension


---
Diffs of the changes:



From lattner at cs.uiuc.edu  Sun Feb 23 22:41:01 2003
From: lattner at cs.uiuc.edu (Chris Lattner)
Date: Sun Feb 23 22:41:01 2003
Subject: [llvm-commits] CVS: llvm/include/Support/slist
Message-ID: <200302240440.WAA02965@apoc.cs.uiuc.edu>


Changes in directory llvm/include/Support:

slist updated: 1.1 -> 1.2

---
Log message:

Fix #endif


---
Diffs of the changes:

Index: llvm/include/Support/slist
diff -u llvm/include/Support/slist:1.1 llvm/include/Support/slist:1.2
--- llvm/include/Support/slist:1.1	Sun Feb 23 22:31:49 2003
+++ llvm/include/Support/slist	Sun Feb 23 22:40:35 2003
@@ -30,3 +30,5 @@
 // GCC 2.x
 #include <slist>
 #endif
+
+#endif