aboutsummaryrefslogtreecommitdiff
path: root/lib/Transforms/InstCombine
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Transforms/InstCombine')
-rw-r--r--lib/Transforms/InstCombine/CMakeLists.txt2
-rw-r--r--lib/Transforms/InstCombine/InstCombine.h4
-rw-r--r--lib/Transforms/InstCombine/InstCombineAddSub.cpp88
-rw-r--r--lib/Transforms/InstCombine/InstCombineAndOrXor.cpp17
-rw-r--r--lib/Transforms/InstCombine/InstCombineCalls.cpp164
-rw-r--r--lib/Transforms/InstCombine/InstCombineCasts.cpp22
-rw-r--r--lib/Transforms/InstCombine/InstCombineCompares.cpp21
-rw-r--r--lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp121
-rw-r--r--lib/Transforms/InstCombine/InstCombineMulDivRem.cpp5
-rw-r--r--lib/Transforms/InstCombine/InstCombineSelect.cpp23
-rw-r--r--lib/Transforms/InstCombine/InstCombineShifts.cpp76
-rw-r--r--lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp29
-rw-r--r--lib/Transforms/InstCombine/InstructionCombining.cpp250
13 files changed, 420 insertions, 402 deletions
diff --git a/lib/Transforms/InstCombine/CMakeLists.txt b/lib/Transforms/InstCombine/CMakeLists.txt
index d070ccc0d63f..72cfe2c985bc 100644
--- a/lib/Transforms/InstCombine/CMakeLists.txt
+++ b/lib/Transforms/InstCombine/CMakeLists.txt
@@ -13,3 +13,5 @@ add_llvm_library(LLVMInstCombine
InstCombineSimplifyDemanded.cpp
InstCombineVectorOps.cpp
)
+
+add_dependencies(LLVMInstCombine intrinsics_gen)
diff --git a/lib/Transforms/InstCombine/InstCombine.h b/lib/Transforms/InstCombine/InstCombine.h
index 199df519ce07..0d5ef904ee47 100644
--- a/lib/Transforms/InstCombine/InstCombine.h
+++ b/lib/Transforms/InstCombine/InstCombine.h
@@ -11,11 +11,11 @@
#define INSTCOMBINE_INSTCOMBINE_H
#include "InstCombineWorklist.h"
+#include "llvm/IRBuilder.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/Operator.h"
#include "llvm/Pass.h"
#include "llvm/Analysis/ValueTracking.h"
-#include "llvm/Support/IRBuilder.h"
#include "llvm/Support/InstVisitor.h"
#include "llvm/Support/TargetFolder.h"
@@ -187,7 +187,7 @@ public:
Instruction *visitPHINode(PHINode &PN);
Instruction *visitGetElementPtrInst(GetElementPtrInst &GEP);
Instruction *visitAllocaInst(AllocaInst &AI);
- Instruction *visitMalloc(Instruction &FI);
+ Instruction *visitAllocSite(Instruction &FI);
Instruction *visitFree(CallInst &FI);
Instruction *visitLoadInst(LoadInst &LI);
Instruction *visitStoreInst(StoreInst &SI);
diff --git a/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index 05e702fa43b5..99b62f8d05a7 100644
--- a/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -170,10 +170,11 @@ Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
// -A + B --> B - A
// -A + -B --> -(A + B)
if (Value *LHSV = dyn_castNegVal(LHS)) {
- if (Value *RHSV = dyn_castNegVal(RHS)) {
- Value *NewAdd = Builder->CreateAdd(LHSV, RHSV, "sum");
- return BinaryOperator::CreateNeg(NewAdd);
- }
+ if (!isa<Constant>(RHS))
+ if (Value *RHSV = dyn_castNegVal(RHS)) {
+ Value *NewAdd = Builder->CreateAdd(LHSV, RHSV, "sum");
+ return BinaryOperator::CreateNeg(NewAdd);
+ }
return BinaryOperator::CreateSub(RHS, LHSV);
}
@@ -329,6 +330,20 @@ Instruction *InstCombiner::visitAdd(BinaryOperator &I) {
}
}
+ // Check for (x & y) + (x ^ y)
+ {
+ Value *A = 0, *B = 0;
+ if (match(RHS, m_Xor(m_Value(A), m_Value(B))) &&
+ (match(LHS, m_And(m_Specific(A), m_Specific(B))) ||
+ match(LHS, m_And(m_Specific(B), m_Specific(A)))))
+ return BinaryOperator::CreateOr(A, B);
+
+ if (match(LHS, m_Xor(m_Value(A), m_Value(B))) &&
+ (match(RHS, m_And(m_Specific(A), m_Specific(B))) ||
+ match(RHS, m_And(m_Specific(B), m_Specific(A)))))
+ return BinaryOperator::CreateOr(A, B);
+ }
+
return Changed ? &I : 0;
}
@@ -406,66 +421,6 @@ Instruction *InstCombiner::visitFAdd(BinaryOperator &I) {
}
-/// EmitGEPOffset - Given a getelementptr instruction/constantexpr, emit the
-/// code necessary to compute the offset from the base pointer (without adding
-/// in the base pointer). Return the result as a signed integer of intptr size.
-Value *InstCombiner::EmitGEPOffset(User *GEP) {
- TargetData &TD = *getTargetData();
- gep_type_iterator GTI = gep_type_begin(GEP);
- Type *IntPtrTy = TD.getIntPtrType(GEP->getContext());
- Value *Result = Constant::getNullValue(IntPtrTy);
-
- // If the GEP is inbounds, we know that none of the addressing operations will
- // overflow in an unsigned sense.
- bool isInBounds = cast<GEPOperator>(GEP)->isInBounds();
-
- // Build a mask for high order bits.
- unsigned IntPtrWidth = TD.getPointerSizeInBits();
- uint64_t PtrSizeMask = ~0ULL >> (64-IntPtrWidth);
-
- for (User::op_iterator i = GEP->op_begin() + 1, e = GEP->op_end(); i != e;
- ++i, ++GTI) {
- Value *Op = *i;
- uint64_t Size = TD.getTypeAllocSize(GTI.getIndexedType()) & PtrSizeMask;
- if (ConstantInt *OpC = dyn_cast<ConstantInt>(Op)) {
- if (OpC->isZero()) continue;
-
- // Handle a struct index, which adds its field offset to the pointer.
- if (StructType *STy = dyn_cast<StructType>(*GTI)) {
- Size = TD.getStructLayout(STy)->getElementOffset(OpC->getZExtValue());
-
- if (Size)
- Result = Builder->CreateAdd(Result, ConstantInt::get(IntPtrTy, Size),
- GEP->getName()+".offs");
- continue;
- }
-
- Constant *Scale = ConstantInt::get(IntPtrTy, Size);
- Constant *OC =
- ConstantExpr::getIntegerCast(OpC, IntPtrTy, true /*SExt*/);
- Scale = ConstantExpr::getMul(OC, Scale, isInBounds/*NUW*/);
- // Emit an add instruction.
- Result = Builder->CreateAdd(Result, Scale, GEP->getName()+".offs");
- continue;
- }
- // Convert to correct type.
- if (Op->getType() != IntPtrTy)
- Op = Builder->CreateIntCast(Op, IntPtrTy, true, Op->getName()+".c");
- if (Size != 1) {
- // We'll let instcombine(mul) convert this to a shl if possible.
- Op = Builder->CreateMul(Op, ConstantInt::get(IntPtrTy, Size),
- GEP->getName()+".idx", isInBounds /*NUW*/);
- }
-
- // Emit an add instruction.
- Result = Builder->CreateAdd(Op, Result, GEP->getName()+".offs");
- }
- return Result;
-}
-
-
-
-
/// Optimize pointer differences into the same array into a size. Consider:
/// &A[10] - &A[0]: we should compile this to "10". LHS/RHS are the pointer
/// operands to the ptrtoint instructions for the LHS/RHS of the subtract.
@@ -589,11 +544,6 @@ Instruction *InstCombiner::visitSub(BinaryOperator &I) {
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
- // C - zext(bool) -> bool ? C - 1 : C
- if (ZExtInst *ZI = dyn_cast<ZExtInst>(Op1))
- if (ZI->getSrcTy()->isIntegerTy(1))
- return SelectInst::Create(ZI->getOperand(0), SubOne(C), C);
-
// C-(X+C2) --> (C-C2)-X
ConstantInt *C2;
if (match(Op1, m_Add(m_Value(X), m_ConstantInt(C2))))
diff --git a/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index 0dbe11d2f01f..7d0af0d80226 100644
--- a/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -986,19 +986,23 @@ Value *InstCombiner::FoldAndOfFCmps(FCmpInst *LHS, FCmpInst *RHS) {
bool Op1Ordered;
unsigned Op0Pred = getFCmpCode(Op0CC, Op0Ordered);
unsigned Op1Pred = getFCmpCode(Op1CC, Op1Ordered);
+ // uno && ord -> false
+ if (Op0Pred == 0 && Op1Pred == 0 && Op0Ordered != Op1Ordered)
+ return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
if (Op1Pred == 0) {
std::swap(LHS, RHS);
std::swap(Op0Pred, Op1Pred);
std::swap(Op0Ordered, Op1Ordered);
}
if (Op0Pred == 0) {
- // uno && ueq -> uno && (uno || eq) -> ueq
+ // uno && ueq -> uno && (uno || eq) -> uno
// ord && olt -> ord && (ord && lt) -> olt
- if (Op0Ordered == Op1Ordered)
+ if (!Op0Ordered && (Op0Ordered == Op1Ordered))
+ return LHS;
+ if (Op0Ordered && (Op0Ordered == Op1Ordered))
return RHS;
// uno && oeq -> uno && (ord && eq) -> false
- // uno && ord -> false
if (!Op0Ordered)
return ConstantInt::get(CmpInst::makeCmpResultType(LHS->getType()), 0);
// ord && ueq -> ord && (uno || eq) -> oeq
@@ -1932,10 +1936,15 @@ Instruction *InstCombiner::visitOr(BinaryOperator &I) {
// A | ( A ^ B) -> A | B
// A | (~A ^ B) -> A | ~B
+ // (A & B) | (A ^ B)
if (match(Op1, m_Xor(m_Value(A), m_Value(B)))) {
if (Op0 == A || Op0 == B)
return BinaryOperator::CreateOr(A, B);
+ if (match(Op0, m_And(m_Specific(A), m_Specific(B))) ||
+ match(Op0, m_And(m_Specific(B), m_Specific(A))))
+ return BinaryOperator::CreateOr(A, B);
+
if (Op1->hasOneUse() && match(A, m_Not(m_Specific(Op0)))) {
Value *Not = Builder->CreateNot(B, B->getName()+".not");
return BinaryOperator::CreateOr(Not, Op0);
@@ -2212,7 +2221,7 @@ Instruction *InstCombiner::visitXor(BinaryOperator &I) {
if (Op0I && Op1I && Op0I->isShift() &&
Op0I->getOpcode() == Op1I->getOpcode() &&
Op0I->getOperand(1) == Op1I->getOperand(1) &&
- (Op1I->hasOneUse() || Op1I->hasOneUse())) {
+ (Op0I->hasOneUse() || Op1I->hasOneUse())) {
Value *NewOp =
Builder->CreateXor(Op0I->getOperand(0), Op1I->getOperand(0),
Op0I->getName());
diff --git a/lib/Transforms/InstCombine/InstCombineCalls.cpp b/lib/Transforms/InstCombine/InstCombineCalls.cpp
index 77e47271008c..d34fab103fa3 100644
--- a/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -172,8 +172,6 @@ Instruction *InstCombiner::SimplifyMemSet(MemSetInst *MI) {
Instruction *InstCombiner::visitCallInst(CallInst &CI) {
if (isFreeCall(&CI))
return visitFree(CI);
- if (isMalloc(&CI))
- return visitMalloc(CI);
// If the caller function is nounwind, mark the call as nounwind, even if the
// callee isn't.
@@ -246,78 +244,10 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
switch (II->getIntrinsicID()) {
default: break;
case Intrinsic::objectsize: {
- // We need target data for just about everything so depend on it.
- if (!TD) break;
-
- Type *ReturnTy = CI.getType();
- uint64_t DontKnow = II->getArgOperand(1) == Builder->getTrue() ? 0 : -1ULL;
-
- // Get to the real allocated thing and offset as fast as possible.
- Value *Op1 = II->getArgOperand(0)->stripPointerCasts();
-
- uint64_t Offset = 0;
- uint64_t Size = -1ULL;
-
- // Try to look through constant GEPs.
- if (GEPOperator *GEP = dyn_cast<GEPOperator>(Op1)) {
- if (!GEP->hasAllConstantIndices()) break;
-
- // Get the current byte offset into the thing. Use the original
- // operand in case we're looking through a bitcast.
- SmallVector<Value*, 8> Ops(GEP->idx_begin(), GEP->idx_end());
- if (!GEP->getPointerOperandType()->isPointerTy())
- return 0;
- Offset = TD->getIndexedOffset(GEP->getPointerOperandType(), Ops);
-
- Op1 = GEP->getPointerOperand()->stripPointerCasts();
-
- // Make sure we're not a constant offset from an external
- // global.
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Op1))
- if (!GV->hasDefinitiveInitializer()) break;
- }
-
- // If we've stripped down to a single global variable that we
- // can know the size of then just return that.
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Op1)) {
- if (GV->hasDefinitiveInitializer()) {
- Constant *C = GV->getInitializer();
- Size = TD->getTypeAllocSize(C->getType());
- } else {
- // Can't determine size of the GV.
- Constant *RetVal = ConstantInt::get(ReturnTy, DontKnow);
- return ReplaceInstUsesWith(CI, RetVal);
- }
- } else if (AllocaInst *AI = dyn_cast<AllocaInst>(Op1)) {
- // Get alloca size.
- if (AI->getAllocatedType()->isSized()) {
- Size = TD->getTypeAllocSize(AI->getAllocatedType());
- if (AI->isArrayAllocation()) {
- const ConstantInt *C = dyn_cast<ConstantInt>(AI->getArraySize());
- if (!C) break;
- Size *= C->getZExtValue();
- }
- }
- } else if (CallInst *MI = extractMallocCall(Op1)) {
- // Get allocation size.
- Type* MallocType = getMallocAllocatedType(MI);
- if (MallocType && MallocType->isSized())
- if (Value *NElems = getMallocArraySize(MI, TD, true))
- if (ConstantInt *NElements = dyn_cast<ConstantInt>(NElems))
- Size = NElements->getZExtValue() * TD->getTypeAllocSize(MallocType);
- }
-
- // Do not return "I don't know" here. Later optimization passes could
- // make it possible to evaluate objectsize to a constant.
- if (Size == -1ULL)
- break;
-
- if (Size < Offset) {
- // Out of bound reference? Negative index normalized to large
- // index? Just return "I don't know".
- return ReplaceInstUsesWith(CI, ConstantInt::get(ReturnTy, DontKnow));
- }
- return ReplaceInstUsesWith(CI, ConstantInt::get(ReturnTy, Size-Offset));
+ uint64_t Size;
+ if (getObjectSize(II->getArgOperand(0), Size, TD))
+ return ReplaceInstUsesWith(CI, ConstantInt::get(CI.getType(), Size));
+ return 0;
}
case Intrinsic::bswap:
// bswap(bswap(x)) -> x
@@ -694,6 +624,57 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
break;
}
+ case Intrinsic::arm_neon_vmulls:
+ case Intrinsic::arm_neon_vmullu: {
+ Value *Arg0 = II->getArgOperand(0);
+ Value *Arg1 = II->getArgOperand(1);
+
+ // Handle mul by zero first:
+ if (isa<ConstantAggregateZero>(Arg0) || isa<ConstantAggregateZero>(Arg1)) {
+ return ReplaceInstUsesWith(CI, ConstantAggregateZero::get(II->getType()));
+ }
+
+ // Check for constant LHS & RHS - in this case we just simplify.
+ bool Zext = (II->getIntrinsicID() == Intrinsic::arm_neon_vmullu);
+ VectorType *NewVT = cast<VectorType>(II->getType());
+ unsigned NewWidth = NewVT->getElementType()->getIntegerBitWidth();
+ if (ConstantDataVector *CV0 = dyn_cast<ConstantDataVector>(Arg0)) {
+ if (ConstantDataVector *CV1 = dyn_cast<ConstantDataVector>(Arg1)) {
+ VectorType* VT = cast<VectorType>(CV0->getType());
+ SmallVector<Constant*, 4> NewElems;
+ for (unsigned i = 0; i < VT->getNumElements(); ++i) {
+ APInt CV0E =
+ (cast<ConstantInt>(CV0->getAggregateElement(i)))->getValue();
+ CV0E = Zext ? CV0E.zext(NewWidth) : CV0E.sext(NewWidth);
+ APInt CV1E =
+ (cast<ConstantInt>(CV1->getAggregateElement(i)))->getValue();
+ CV1E = Zext ? CV1E.zext(NewWidth) : CV1E.sext(NewWidth);
+ NewElems.push_back(
+ ConstantInt::get(NewVT->getElementType(), CV0E * CV1E));
+ }
+ return ReplaceInstUsesWith(CI, ConstantVector::get(NewElems));
+ }
+
+ // Couldn't simplify - cannonicalize constant to the RHS.
+ std::swap(Arg0, Arg1);
+ }
+
+ // Handle mul by one:
+ if (ConstantDataVector *CV1 = dyn_cast<ConstantDataVector>(Arg1)) {
+ if (ConstantInt *Splat =
+ dyn_cast_or_null<ConstantInt>(CV1->getSplatValue())) {
+ if (Splat->isOne()) {
+ if (Zext)
+ return CastInst::CreateZExtOrBitCast(Arg0, II->getType());
+ // else
+ return CastInst::CreateSExtOrBitCast(Arg0, II->getType());
+ }
+ }
+ }
+
+ break;
+ }
+
case Intrinsic::stackrestore: {
// If the save is right next to the restore, remove the restore. This can
// happen when variable allocas are DCE'd.
@@ -711,7 +692,7 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
TerminatorInst *TI = II->getParent()->getTerminator();
bool CannotRemove = false;
for (++BI; &*BI != TI; ++BI) {
- if (isa<AllocaInst>(BI) || isMalloc(BI)) {
+ if (isa<AllocaInst>(BI)) {
CannotRemove = true;
break;
}
@@ -814,7 +795,7 @@ Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const TargetData *TD) {
if (CI->getCalledFunction() == 0) return 0;
InstCombineFortifiedLibCalls Simplifier(this);
- Simplifier.fold(CI, TD);
+ Simplifier.fold(CI, TD, TLI);
return Simplifier.NewInstruction;
}
@@ -898,6 +879,9 @@ static IntrinsicInst *FindInitTrampoline(Value *Callee) {
// visitCallSite - Improvements for call and invoke instructions.
//
Instruction *InstCombiner::visitCallSite(CallSite CS) {
+ if (isAllocLikeFn(CS.getInstruction()))
+ return visitAllocSite(*CS.getInstruction());
+
bool Changed = false;
// If the callee is a pointer to a function, attempt to move any casts to the
@@ -933,24 +917,24 @@ Instruction *InstCombiner::visitCallSite(CallSite CS) {
}
if (isa<ConstantPointerNull>(Callee) || isa<UndefValue>(Callee)) {
- // This instruction is not reachable, just remove it. We insert a store to
- // undef so that we know that this code is not reachable, despite the fact
- // that we can't modify the CFG here.
- new StoreInst(ConstantInt::getTrue(Callee->getContext()),
- UndefValue::get(Type::getInt1PtrTy(Callee->getContext())),
- CS.getInstruction());
-
// If CS does not return void then replaceAllUsesWith undef.
// This allows ValueHandlers and custom metadata to adjust itself.
if (!CS.getInstruction()->getType()->isVoidTy())
ReplaceInstUsesWith(*CS.getInstruction(),
UndefValue::get(CS.getInstruction()->getType()));
- if (InvokeInst *II = dyn_cast<InvokeInst>(CS.getInstruction())) {
- // Don't break the CFG, insert a dummy cond branch.
- BranchInst::Create(II->getNormalDest(), II->getUnwindDest(),
- ConstantInt::getTrue(Callee->getContext()), II);
+ if (isa<InvokeInst>(CS.getInstruction())) {
+ // Can't remove an invoke because we cannot change the CFG.
+ return 0;
}
+
+ // This instruction is not reachable, just remove it. We insert a store to
+ // undef so that we know that this code is not reachable, despite the fact
+ // that we can't modify the CFG here.
+ new StoreInst(ConstantInt::getTrue(Callee->getContext()),
+ UndefValue::get(Type::getInt1PtrTy(Callee->getContext())),
+ CS.getInstruction());
+
return EraseInstFromFunction(*CS.getInstruction());
}
@@ -1194,8 +1178,7 @@ bool InstCombiner::transformConstExprCastCall(CallSite CS) {
if (NewRetTy->isVoidTy())
Caller->setName(""); // Void type should not have a name.
- const AttrListPtr &NewCallerPAL = AttrListPtr::get(attrVec.begin(),
- attrVec.end());
+ const AttrListPtr &NewCallerPAL = AttrListPtr::get(attrVec);
Instruction *NC;
if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
@@ -1367,8 +1350,7 @@ InstCombiner::transformCallThroughTrampoline(CallSite CS,
NestF->getType() == PointerType::getUnqual(NewFTy) ?
NestF : ConstantExpr::getBitCast(NestF,
PointerType::getUnqual(NewFTy));
- const AttrListPtr &NewPAL = AttrListPtr::get(NewAttrs.begin(),
- NewAttrs.end());
+ const AttrListPtr &NewPAL = AttrListPtr::get(NewAttrs);
Instruction *NewCaller;
if (InvokeInst *II = dyn_cast<InvokeInst>(Caller)) {
diff --git a/lib/Transforms/InstCombine/InstCombineCasts.cpp b/lib/Transforms/InstCombine/InstCombineCasts.cpp
index 39279f437205..555b4428d2e8 100644
--- a/lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -34,7 +34,7 @@ static Value *DecomposeSimpleLinearExpr(Value *Val, unsigned &Scale,
if (BinaryOperator *I = dyn_cast<BinaryOperator>(Val)) {
// Cannot look past anything that might overflow.
OverflowingBinaryOperator *OBI = dyn_cast<OverflowingBinaryOperator>(Val);
- if (OBI && !OBI->hasNoUnsignedWrap()) {
+ if (OBI && !OBI->hasNoUnsignedWrap() && !OBI->hasNoSignedWrap()) {
Scale = 1;
Offset = 0;
return Val;
@@ -648,10 +648,8 @@ static bool CanEvaluateZExtd(Value *V, Type *Ty, unsigned &BitsToClear) {
if (!I) return false;
// If the input is a truncate from the destination type, we can trivially
- // eliminate it, even if it has multiple uses.
- // FIXME: This is currently disabled until codegen can handle this without
- // pessimizing code, PR5997.
- if (0 && isa<TruncInst>(I) && I->getOperand(0)->getType() == Ty)
+ // eliminate it.
+ if (isa<TruncInst>(I) && I->getOperand(0)->getType() == Ty)
return true;
// We can't extend or shrink something that has multiple uses: doing so would
@@ -992,11 +990,8 @@ static bool CanEvaluateSExtd(Value *V, Type *Ty) {
Instruction *I = dyn_cast<Instruction>(V);
if (!I) return false;
- // If this is a truncate from the dest type, we can trivially eliminate it,
- // even if it has multiple uses.
- // FIXME: This is currently disabled until codegen can handle this without
- // pessimizing code, PR5997.
- if (0 && isa<TruncInst>(I) && I->getOperand(0)->getType() == Ty)
+ // If this is a truncate from the dest type, we can trivially eliminate it.
+ if (isa<TruncInst>(I) && I->getOperand(0)->getType() == Ty)
return true;
// We can't extend or shrink something that has multiple uses: doing so would
@@ -1341,10 +1336,9 @@ Instruction *InstCombiner::commonPointerCastTransforms(CastInst &CI) {
// non-type-safe code.
if (TD && GEP->hasOneUse() && isa<BitCastInst>(GEP->getOperand(0)) &&
GEP->hasAllConstantIndices()) {
- // We are guaranteed to get a constant from EmitGEPOffset.
- ConstantInt *OffsetV = cast<ConstantInt>(EmitGEPOffset(GEP));
- int64_t Offset = OffsetV->getSExtValue();
-
+ SmallVector<Value*, 8> Ops(GEP->idx_begin(), GEP->idx_end());
+ int64_t Offset = TD->getIndexedOffset(GEP->getPointerOperandType(), Ops);
+
// Get the base pointer input of the bitcast, and the type it points to.
Value *OrigBase = cast<BitCastInst>(GEP->getOperand(0))->getOperand(0);
Type *GEPIdxTy =
diff --git a/lib/Transforms/InstCombine/InstCombineCompares.cpp b/lib/Transforms/InstCombine/InstCombineCompares.cpp
index ab2987ff24cd..bdd310e97f6c 100644
--- a/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -1035,7 +1035,7 @@ Instruction *InstCombiner::visitICmpInstWithInstAndIntCst(ICmpInst &ICI,
if ((KnownZero|KnownOne).countLeadingOnes() >= SrcBits-DstBits) {
// Pull in the high bits from known-ones set.
APInt NewRHS = RHS->getValue().zext(SrcBits);
- NewRHS |= KnownOne;
+ NewRHS |= KnownOne & APInt::getHighBitsSet(SrcBits, SrcBits-DstBits);
return new ICmpInst(ICI.getPredicate(), LHSI->getOperand(0),
ConstantInt::get(ICI.getContext(), NewRHS));
}
@@ -2580,10 +2580,25 @@ Instruction *InstCombiner::visitICmpInst(ICmpInst &I) {
}
}
+ // Transform (zext A) == (B & (1<<X)-1) --> A == (trunc B)
+ // and (B & (1<<X)-1) == (zext A) --> A == (trunc B)
+ ConstantInt *Cst1;
+ if ((Op0->hasOneUse() &&
+ match(Op0, m_ZExt(m_Value(A))) &&
+ match(Op1, m_And(m_Value(B), m_ConstantInt(Cst1)))) ||
+ (Op1->hasOneUse() &&
+ match(Op0, m_And(m_Value(B), m_ConstantInt(Cst1))) &&
+ match(Op1, m_ZExt(m_Value(A))))) {
+ APInt Pow2 = Cst1->getValue() + 1;
+ if (Pow2.isPowerOf2() && isa<IntegerType>(A->getType()) &&
+ Pow2.logBase2() == cast<IntegerType>(A->getType())->getBitWidth())
+ return new ICmpInst(I.getPredicate(), A,
+ Builder->CreateTrunc(B, A->getType()));
+ }
+
// Transform "icmp eq (trunc (lshr(X, cst1)), cst" to
// "icmp (and X, mask), cst"
uint64_t ShAmt = 0;
- ConstantInt *Cst1;
if (Op0->hasOneUse() &&
match(Op0, m_Trunc(m_OneUse(m_LShr(m_Value(A),
m_ConstantInt(ShAmt))))) &&
@@ -2809,7 +2824,7 @@ Instruction *InstCombiner::FoldFCmp_IntToFP_Cst(FCmpInst &I,
case ICmpInst::ICMP_UGE:
// (float)int >= -4.4 --> true
// (float)int >= 4.4 --> int > 4
- if (!RHS.isNegative())
+ if (RHS.isNegative())
return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getContext()));
Pred = ICmpInst::ICMP_UGT;
break;
diff --git a/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp b/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
index b2f2e248e417..c485844aaeb4 100644
--- a/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
+++ b/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
@@ -22,72 +22,6 @@ using namespace llvm;
STATISTIC(NumDeadStore, "Number of dead stores eliminated");
-// Try to kill dead allocas by walking through its uses until we see some use
-// that could escape. This is a conservative analysis which tries to handle
-// GEPs, bitcasts, stores, and no-op intrinsics. These tend to be the things
-// left after inlining and SROA finish chewing on an alloca.
-static Instruction *removeDeadAlloca(InstCombiner &IC, AllocaInst &AI) {
- SmallVector<Instruction *, 4> Worklist, DeadStores;
- Worklist.push_back(&AI);
- do {
- Instruction *PI = Worklist.pop_back_val();
- for (Value::use_iterator UI = PI->use_begin(), UE = PI->use_end();
- UI != UE; ++UI) {
- Instruction *I = cast<Instruction>(*UI);
- switch (I->getOpcode()) {
- default:
- // Give up the moment we see something we can't handle.
- return 0;
-
- case Instruction::GetElementPtr:
- case Instruction::BitCast:
- Worklist.push_back(I);
- continue;
-
- case Instruction::Call:
- // We can handle a limited subset of calls to no-op intrinsics.
- if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
- switch (II->getIntrinsicID()) {
- case Intrinsic::dbg_declare:
- case Intrinsic::dbg_value:
- case Intrinsic::invariant_start:
- case Intrinsic::invariant_end:
- case Intrinsic::lifetime_start:
- case Intrinsic::lifetime_end:
- continue;
- default:
- return 0;
- }
- }
- // Reject everything else.
- return 0;
-
- case Instruction::Store: {
- // Stores into the alloca are only live if the alloca is live.
- StoreInst *SI = cast<StoreInst>(I);
- // We can eliminate atomic stores, but not volatile.
- if (SI->isVolatile())
- return 0;
- // The store is only trivially safe if the poniter is the destination
- // as opposed to the value. We're conservative here and don't check for
- // the case where we store the address of a dead alloca into a dead
- // alloca.
- if (SI->getPointerOperand() != PI)
- return 0;
- DeadStores.push_back(I);
- continue;
- }
- }
- }
- } while (!Worklist.empty());
-
- // The alloca is dead. Kill off all the stores to it, and then replace it
- // with undef.
- while (!DeadStores.empty())
- IC.EraseInstFromFunction(*DeadStores.pop_back_val());
- return IC.ReplaceInstUsesWith(AI, UndefValue::get(AI.getType()));
-}
-
Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
// Ensure that the alloca array size argument has type intptr_t, so that
// any casting is exposed early.
@@ -106,7 +40,6 @@ Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
if (const ConstantInt *C = dyn_cast<ConstantInt>(AI.getArraySize())) {
Type *NewTy =
ArrayType::get(AI.getAllocatedType(), C->getZExtValue());
- assert(isa<AllocaInst>(AI) && "Unknown type of allocation inst!");
AllocaInst *New = Builder->CreateAlloca(NewTy, 0, AI.getName());
New->setAlignment(AI.getAlignment());
@@ -135,22 +68,54 @@ Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
}
}
- if (TD && isa<AllocaInst>(AI) && AI.getAllocatedType()->isSized()) {
- // If alloca'ing a zero byte object, replace the alloca with a null pointer.
- // Note that we only do this for alloca's, because malloc should allocate
- // and return a unique pointer, even for a zero byte allocation.
- if (TD->getTypeAllocSize(AI.getAllocatedType()) == 0)
- return ReplaceInstUsesWith(AI, Constant::getNullValue(AI.getType()));
-
+ if (TD && AI.getAllocatedType()->isSized()) {
// If the alignment is 0 (unspecified), assign it the preferred alignment.
if (AI.getAlignment() == 0)
AI.setAlignment(TD->getPrefTypeAlignment(AI.getAllocatedType()));
+
+ // Move all alloca's of zero byte objects to the entry block and merge them
+ // together. Note that we only do this for alloca's, because malloc should
+ // allocate and return a unique pointer, even for a zero byte allocation.
+ if (TD->getTypeAllocSize(AI.getAllocatedType()) == 0) {
+ // For a zero sized alloca there is no point in doing an array allocation.
+ // This is helpful if the array size is a complicated expression not used
+ // elsewhere.
+ if (AI.isArrayAllocation()) {
+ AI.setOperand(0, ConstantInt::get(AI.getArraySize()->getType(), 1));
+ return &AI;
+ }
+
+ // Get the first instruction in the entry block.
+ BasicBlock &EntryBlock = AI.getParent()->getParent()->getEntryBlock();
+ Instruction *FirstInst = EntryBlock.getFirstNonPHIOrDbg();
+ if (FirstInst != &AI) {
+ // If the entry block doesn't start with a zero-size alloca then move
+ // this one to the start of the entry block. There is no problem with
+ // dominance as the array size was forced to a constant earlier already.
+ AllocaInst *EntryAI = dyn_cast<AllocaInst>(FirstInst);
+ if (!EntryAI || !EntryAI->getAllocatedType()->isSized() ||
+ TD->getTypeAllocSize(EntryAI->getAllocatedType()) != 0) {
+ AI.moveBefore(FirstInst);
+ return &AI;
+ }
+
+ // Replace this zero-sized alloca with the one at the start of the entry
+ // block after ensuring that the address will be aligned enough for both
+ // types.
+ unsigned MaxAlign =
+ std::max(TD->getPrefTypeAlignment(EntryAI->getAllocatedType()),
+ TD->getPrefTypeAlignment(AI.getAllocatedType()));
+ EntryAI->setAlignment(MaxAlign);
+ if (AI.getType() != EntryAI->getType())
+ return new BitCastInst(EntryAI, AI.getType());
+ return ReplaceInstUsesWith(AI, EntryAI);
+ }
+ }
}
- // Try to aggressively remove allocas which are only used for GEPs, lifetime
- // markers, and stores. This happens when SROA iteratively promotes stores
- // out of the alloca, and we need to cleanup after it.
- return removeDeadAlloca(*this, AI);
+ // At last, use the generic allocation site handler to aggressively remove
+ // unused allocas.
+ return visitAllocSite(AI);
}
diff --git a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
index 5168e2a113ca..35a0bbb76146 100644
--- a/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
+++ b/lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
@@ -464,9 +464,12 @@ Instruction *InstCombiner::visitUDiv(BinaryOperator &I) {
// X udiv (C1 << N), where C1 is "1<<C2" --> X >> (N+C2)
{ const APInt *CI; Value *N;
- if (match(Op1, m_Shl(m_Power2(CI), m_Value(N)))) {
+ if (match(Op1, m_Shl(m_Power2(CI), m_Value(N))) ||
+ match(Op1, m_ZExt(m_Shl(m_Power2(CI), m_Value(N))))) {
if (*CI != 1)
N = Builder->CreateAdd(N, ConstantInt::get(I.getType(),CI->logBase2()));
+ if (ZExtInst *Z = dyn_cast<ZExtInst>(Op1))
+ N = Builder->CreateZExt(N, Z->getDestTy());
if (I.isExact())
return BinaryOperator::CreateExactLShr(Op0, N);
return BinaryOperator::CreateLShr(Op0, N);
diff --git a/lib/Transforms/InstCombine/InstCombineSelect.cpp b/lib/Transforms/InstCombine/InstCombineSelect.cpp
index e727b2c592db..291e80019e8d 100644
--- a/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -129,6 +129,12 @@ Instruction *InstCombiner::FoldSelectOpOp(SelectInst &SI, Instruction *TI,
if (TI->isCast()) {
if (TI->getOperand(0)->getType() != FI->getOperand(0)->getType())
return 0;
+ // The select condition may be a vector. We may only change the operand
+ // type if the vector width remains the same (and matches the condition).
+ Type *CondTy = SI.getCondition()->getType();
+ if (CondTy->isVectorTy() && CondTy->getVectorNumElements() !=
+ FI->getOperand(0)->getType()->getVectorNumElements())
+ return 0;
} else {
return 0; // unknown unary op.
}
@@ -498,7 +504,7 @@ Instruction *InstCombiner::visitSelectInstWithICmp(SelectInst &SI,
// NOTE: if we wanted to, this is where to detect integer MIN/MAX
- if (isa<Constant>(CmpRHS)) {
+ if (CmpRHS != CmpLHS && isa<Constant>(CmpRHS)) {
if (CmpLHS == TrueVal && Pred == ICmpInst::ICMP_EQ) {
// Transform (X == C) ? X : Y -> (X == C) ? C : Y
SI.setOperand(1, CmpRHS);
@@ -875,12 +881,16 @@ Instruction *InstCombiner::visitSelectInst(SelectInst &SI) {
if (SelectInst *TrueSI = dyn_cast<SelectInst>(TrueVal)) {
if (TrueSI->getCondition() == CondVal) {
+ if (SI.getTrueValue() == TrueSI->getTrueValue())
+ return 0;
SI.setOperand(1, TrueSI->getTrueValue());
return &SI;
}
}
if (SelectInst *FalseSI = dyn_cast<SelectInst>(FalseVal)) {
if (FalseSI->getCondition() == CondVal) {
+ if (SI.getFalseValue() == FalseSI->getFalseValue())
+ return 0;
SI.setOperand(2, FalseSI->getFalseValue());
return &SI;
}
@@ -893,5 +903,16 @@ Instruction *InstCombiner::visitSelectInst(SelectInst &SI) {
return &SI;
}
+ if (VectorType* VecTy = dyn_cast<VectorType>(SI.getType())) {
+ unsigned VWidth = VecTy->getNumElements();
+ APInt UndefElts(VWidth, 0);
+ APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
+ if (Value *V = SimplifyDemandedVectorElts(&SI, AllOnesEltMask, UndefElts)) {
+ if (V != &SI)
+ return ReplaceInstUsesWith(SI, V);
+ return &SI;
+ }
+ }
+
return 0;
}
diff --git a/lib/Transforms/InstCombine/InstCombineShifts.cpp b/lib/Transforms/InstCombine/InstCombineShifts.cpp
index b31049e59f18..4bb2403299ce 100644
--- a/lib/Transforms/InstCombine/InstCombineShifts.cpp
+++ b/lib/Transforms/InstCombine/InstCombineShifts.cpp
@@ -151,7 +151,7 @@ static bool CanEvaluateShifted(Value *V, unsigned NumBits, bool isLeftShift,
// We can always turn lshr(c1)+shl(c2) -> lshr(c3)+and(c4), but it isn't
// profitable unless we know the and'd out bits are already zero.
- if (CI->getZExtValue() > NumBits) {
+ if (CI->getValue().ult(TypeWidth) && CI->getZExtValue() > NumBits) {
unsigned LowBits = CI->getZExtValue() - NumBits;
if (MaskedValueIsZero(I->getOperand(0),
APInt::getLowBitsSet(TypeWidth, NumBits) << LowBits))
@@ -529,6 +529,19 @@ Instruction *InstCombiner::FoldShiftByConstant(Value *Op0, ConstantInt *Op1,
ShiftOp = 0;
if (ShiftOp && isa<ConstantInt>(ShiftOp->getOperand(1))) {
+
+ // This is a constant shift of a constant shift. Be careful about hiding
+ // shl instructions behind bit masks. They are used to represent multiplies
+ // by a constant, and it is important that simple arithmetic expressions
+ // are still recognizable by scalar evolution.
+ //
+ // The transforms applied to shl are very similar to the transforms applied
+ // to mul by constant. We can be more aggressive about optimizing right
+ // shifts.
+ //
+ // Combinations of right and left shifts will still be optimized in
+ // DAGCombine where scalar evolution no longer applies.
+
ConstantInt *ShiftAmt1C = cast<ConstantInt>(ShiftOp->getOperand(1));
uint32_t ShiftAmt1 = ShiftAmt1C->getLimitedValue(TypeBits);
uint32_t ShiftAmt2 = Op1->getLimitedValue(TypeBits);
@@ -554,13 +567,6 @@ Instruction *InstCombiner::FoldShiftByConstant(Value *Op0, ConstantInt *Op1,
}
if (ShiftAmt1 == ShiftAmt2) {
- // If we have ((X >>? C) << C), turn this into X & (-1 << C).
- if (I.getOpcode() == Instruction::Shl &&
- ShiftOp->getOpcode() != Instruction::Shl) {
- APInt Mask(APInt::getHighBitsSet(TypeBits, TypeBits - ShiftAmt1));
- return BinaryOperator::CreateAnd(X,
- ConstantInt::get(I.getContext(),Mask));
- }
// If we have ((X << C) >>u C), turn this into X & (-1 >>u C).
if (I.getOpcode() == Instruction::LShr &&
ShiftOp->getOpcode() == Instruction::Shl) {
@@ -570,28 +576,23 @@ Instruction *InstCombiner::FoldShiftByConstant(Value *Op0, ConstantInt *Op1,
}
} else if (ShiftAmt1 < ShiftAmt2) {
uint32_t ShiftDiff = ShiftAmt2-ShiftAmt1;
-
- // (X >>? C1) << C2 --> X << (C2-C1) & (-1 << C2)
+
+ // (X >>?,exact C1) << C2 --> X << (C2-C1)
+ // The inexact version is deferred to DAGCombine so we don't hide shl
+ // behind a bit mask.
if (I.getOpcode() == Instruction::Shl &&
- ShiftOp->getOpcode() != Instruction::Shl) {
+ ShiftOp->getOpcode() != Instruction::Shl &&
+ ShiftOp->isExact()) {
assert(ShiftOp->getOpcode() == Instruction::LShr ||
ShiftOp->getOpcode() == Instruction::AShr);
ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
- if (ShiftOp->isExact()) {
- // (X >>?,exact C1) << C2 --> X << (C2-C1)
- BinaryOperator *NewShl = BinaryOperator::Create(Instruction::Shl,
- X, ShiftDiffCst);
- NewShl->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
- NewShl->setHasNoSignedWrap(I.hasNoSignedWrap());
- return NewShl;
- }
- Value *Shift = Builder->CreateShl(X, ShiftDiffCst);
-
- APInt Mask(APInt::getHighBitsSet(TypeBits, TypeBits - ShiftAmt2));
- return BinaryOperator::CreateAnd(Shift,
- ConstantInt::get(I.getContext(),Mask));
+ BinaryOperator *NewShl = BinaryOperator::Create(Instruction::Shl,
+ X, ShiftDiffCst);
+ NewShl->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
+ NewShl->setHasNoSignedWrap(I.hasNoSignedWrap());
+ return NewShl;
}
-
+
// (X << C1) >>u C2 --> X >>u (C2-C1) & (-1 >> C2)
if (I.getOpcode() == Instruction::LShr &&
ShiftOp->getOpcode() == Instruction::Shl) {
@@ -627,24 +628,19 @@ Instruction *InstCombiner::FoldShiftByConstant(Value *Op0, ConstantInt *Op1,
assert(ShiftAmt2 < ShiftAmt1);
uint32_t ShiftDiff = ShiftAmt1-ShiftAmt2;
- // (X >>? C1) << C2 --> X >>? (C1-C2) & (-1 << C2)
+ // (X >>?exact C1) << C2 --> X >>?exact (C1-C2)
+ // The inexact version is deferred to DAGCombine so we don't hide shl
+ // behind a bit mask.
if (I.getOpcode() == Instruction::Shl &&
- ShiftOp->getOpcode() != Instruction::Shl) {
+ ShiftOp->getOpcode() != Instruction::Shl &&
+ ShiftOp->isExact()) {
ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
- if (ShiftOp->isExact()) {
- // (X >>?exact C1) << C2 --> X >>?exact (C1-C2)
- BinaryOperator *NewShr = BinaryOperator::Create(ShiftOp->getOpcode(),
- X, ShiftDiffCst);
- NewShr->setIsExact(true);
- return NewShr;
- }
- Value *Shift = Builder->CreateBinOp(ShiftOp->getOpcode(),
- X, ShiftDiffCst);
- APInt Mask(APInt::getHighBitsSet(TypeBits, TypeBits - ShiftAmt2));
- return BinaryOperator::CreateAnd(Shift,
- ConstantInt::get(I.getContext(),Mask));
+ BinaryOperator *NewShr = BinaryOperator::Create(ShiftOp->getOpcode(),
+ X, ShiftDiffCst);
+ NewShr->setIsExact(true);
+ return NewShr;
}
-
+
// (X << C1) >>u C2 --> X << (C1-C2) & (-1 >> C2)
if (I.getOpcode() == Instruction::LShr &&
ShiftOp->getOpcode() == Instruction::Shl) {
diff --git a/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
index 125c74a89a11..54be8ed3fa90 100644
--- a/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
+++ b/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
@@ -989,6 +989,29 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
}
break;
}
+ case Instruction::Select: {
+ APInt LeftDemanded(DemandedElts), RightDemanded(DemandedElts);
+ if (ConstantVector* CV = dyn_cast<ConstantVector>(I->getOperand(0))) {
+ for (unsigned i = 0; i < VWidth; i++) {
+ if (CV->getAggregateElement(i)->isNullValue())
+ LeftDemanded.clearBit(i);
+ else
+ RightDemanded.clearBit(i);
+ }
+ }
+
+ TmpV = SimplifyDemandedVectorElts(I->getOperand(1), LeftDemanded,
+ UndefElts, Depth+1);
+ if (TmpV) { I->setOperand(1, TmpV); MadeChange = true; }
+
+ TmpV = SimplifyDemandedVectorElts(I->getOperand(2), RightDemanded,
+ UndefElts2, Depth+1);
+ if (TmpV) { I->setOperand(2, TmpV); MadeChange = true; }
+
+ // Output elements are undefined if both are undefined.
+ UndefElts &= UndefElts2;
+ break;
+ }
case Instruction::BitCast: {
// Vector->vector casts only.
VectorType *VTy = dyn_cast<VectorType>(I->getOperand(0)->getType());
@@ -1074,6 +1097,12 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
// like undef&0. The result is known zero, not undef.
UndefElts &= UndefElts2;
break;
+ case Instruction::FPTrunc:
+ case Instruction::FPExt:
+ TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts,
+ UndefElts, Depth+1);
+ if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
+ break;
case Instruction::Call: {
IntrinsicInst *II = dyn_cast<IntrinsicInst>(I);
diff --git a/lib/Transforms/InstCombine/InstructionCombining.cpp b/lib/Transforms/InstCombine/InstructionCombining.cpp
index 066b2ec89c3e..68ecd516049d 100644
--- a/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ b/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -87,30 +87,34 @@ void InstCombiner::getAnalysisUsage(AnalysisUsage &AU) const {
}
+Value *InstCombiner::EmitGEPOffset(User *GEP) {
+ return llvm::EmitGEPOffset(Builder, *getTargetData(), GEP);
+}
+
/// ShouldChangeType - Return true if it is desirable to convert a computation
/// from 'From' to 'To'. We don't want to convert from a legal to an illegal
/// type for example, or from a smaller to a larger illegal type.
bool InstCombiner::ShouldChangeType(Type *From, Type *To) const {
assert(From->isIntegerTy() && To->isIntegerTy());
-
+
// If we don't have TD, we don't know if the source/dest are legal.
if (!TD) return false;
-
+
unsigned FromWidth = From->getPrimitiveSizeInBits();
unsigned ToWidth = To->getPrimitiveSizeInBits();
bool FromLegal = TD->isLegalInteger(FromWidth);
bool ToLegal = TD->isLegalInteger(ToWidth);
-
+
// If this is a legal integer from type, and the result would be an illegal
// type, don't do the transformation.
if (FromLegal && !ToLegal)
return false;
-
+
// Otherwise, if both are illegal, do not increase the size of the result. We
// do allow things like i160 -> i64, but not i64 -> i160.
if (!FromLegal && !ToLegal && ToWidth > FromWidth)
return false;
-
+
return true;
}
@@ -127,7 +131,7 @@ static bool MaintainNoSignedWrap(BinaryOperator &I, Value *B, Value *C) {
// We reason about Add and Sub Only.
Instruction::BinaryOps Opcode = I.getOpcode();
- if (Opcode != Instruction::Add &&
+ if (Opcode != Instruction::Add &&
Opcode != Instruction::Sub) {
return false;
}
@@ -203,7 +207,7 @@ bool InstCombiner::SimplifyAssociativeOrCommutative(BinaryOperator &I) {
// Conservatively clear the optional flags, since they may not be
// preserved by the reassociation.
if (MaintainNoSignedWrap(I, B, C) &&
- (!Op0 || (isa<BinaryOperator>(Op0) && Op0->hasNoSignedWrap()))) {
+ (!Op0 || (isa<BinaryOperator>(Op0) && Op0->hasNoSignedWrap()))) {
// Note: this is only valid because SimplifyBinOp doesn't look at
// the operands to Op0.
I.clearSubclassOptionalData();
@@ -211,7 +215,7 @@ bool InstCombiner::SimplifyAssociativeOrCommutative(BinaryOperator &I) {
} else {
I.clearSubclassOptionalData();
}
-
+
Changed = true;
++NumReassoc;
continue;
@@ -540,7 +544,7 @@ static Value *FoldOperationIntoSelectOperand(Instruction &I, Value *SO,
Value *Op0 = SO, *Op1 = ConstOperand;
if (!ConstIsRHS)
std::swap(Op0, Op1);
-
+
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(&I))
return IC->Builder->CreateBinOp(BO->getOpcode(), Op0, Op1,
SO->getName()+".op");
@@ -579,7 +583,7 @@ Instruction *InstCombiner::FoldOpIntoSelect(Instruction &Op, SelectInst *SI) {
if (SrcTy && SrcTy->getNumElements() != DestTy->getNumElements())
return 0;
}
-
+
Value *SelectTrueVal = FoldOperationIntoSelectOperand(Op, TV, this);
Value *SelectFalseVal = FoldOperationIntoSelectOperand(Op, FV, this);
@@ -599,7 +603,7 @@ Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I) {
unsigned NumPHIValues = PN->getNumIncomingValues();
if (NumPHIValues == 0)
return 0;
-
+
// We normally only transform phis with a single use. However, if a PHI has
// multiple uses and they are all the same operation, we can fold *all* of the
// uses into the PHI.
@@ -613,7 +617,7 @@ Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I) {
}
// Otherwise, we can replace *all* users with the new PHI we form.
}
-
+
// Check to see if all of the operands of the PHI are simple constants
// (constantint/constantfp/undef). If there is one non-constant value,
// remember the BB it is in. If there is more than one or if *it* is a PHI,
@@ -627,7 +631,7 @@ Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I) {
if (isa<PHINode>(InVal)) return 0; // Itself a phi.
if (NonConstBB) return 0; // More than one non-const value.
-
+
NonConstBB = PN->getIncomingBlock(i);
// If the InVal is an invoke at the end of the pred block, then we can't
@@ -635,14 +639,14 @@ Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I) {
if (InvokeInst *II = dyn_cast<InvokeInst>(InVal))
if (II->getParent() == NonConstBB)
return 0;
-
+
// If the incoming non-constant value is in I's block, we will remove one
// instruction, but insert another equivalent one, leading to infinite
// instcombine.
if (NonConstBB == I.getParent())
return 0;
}
-
+
// If there is exactly one non-constant value, we can insert a copy of the
// operation in that block. However, if this is a critical edge, we would be
// inserting the computation one some other paths (e.g. inside a loop). Only
@@ -656,12 +660,12 @@ Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I) {
PHINode *NewPN = PHINode::Create(I.getType(), PN->getNumIncomingValues());
InsertNewInstBefore(NewPN, *PN);
NewPN->takeName(PN);
-
+
// If we are going to have to insert a new computation, do so right before the
// predecessors terminator.
if (NonConstBB)
Builder->SetInsertPoint(NonConstBB->getTerminator());
-
+
// Next, add all of the operands to the PHI.
if (SelectInst *SI = dyn_cast<SelectInst>(&I)) {
// We only currently try to fold the condition of a select when it is a phi,
@@ -706,20 +710,20 @@ Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I) {
PN->getIncomingValue(i), C, "phitmp");
NewPN->addIncoming(InV, PN->getIncomingBlock(i));
}
- } else {
+ } else {
CastInst *CI = cast<CastInst>(&I);
Type *RetTy = CI->getType();
for (unsigned i = 0; i != NumPHIValues; ++i) {
Value *InV;
if (Constant *InC = dyn_cast<Constant>(PN->getIncomingValue(i)))
InV = ConstantExpr::getCast(CI->getOpcode(), InC, RetTy);
- else
+ else
InV = Builder->CreateCast(CI->getOpcode(),
PN->getIncomingValue(i), I.getType(), "phitmp");
NewPN->addIncoming(InV, PN->getIncomingBlock(i));
}
}
-
+
for (Value::use_iterator UI = PN->use_begin(), E = PN->use_end();
UI != E; ) {
Instruction *User = cast<Instruction>(*UI++);
@@ -734,11 +738,11 @@ Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I) {
/// or not there is a sequence of GEP indices into the type that will land us at
/// the specified offset. If so, fill them into NewIndices and return the
/// resultant element type, otherwise return null.
-Type *InstCombiner::FindElementAtOffset(Type *Ty, int64_t Offset,
+Type *InstCombiner::FindElementAtOffset(Type *Ty, int64_t Offset,
SmallVectorImpl<Value*> &NewIndices) {
if (!TD) return 0;
if (!Ty->isSized()) return 0;
-
+
// Start with the index over the outer type. Note that the type size
// might be zero (even if the offset isn't zero) if the indexed type
// is something like [0 x {int, int}]
@@ -747,7 +751,7 @@ Type *InstCombiner::FindElementAtOffset(Type *Ty, int64_t Offset,
if (int64_t TySize = TD->getTypeAllocSize(Ty)) {
FirstIdx = Offset/TySize;
Offset -= FirstIdx*TySize;
-
+
// Handle hosts where % returns negative instead of values [0..TySize).
if (Offset < 0) {
--FirstIdx;
@@ -756,24 +760,24 @@ Type *InstCombiner::FindElementAtOffset(Type *Ty, int64_t Offset,
}
assert((uint64_t)Offset < (uint64_t)TySize && "Out of range offset");
}
-
+
NewIndices.push_back(ConstantInt::get(IntPtrTy, FirstIdx));
-
+
// Index into the types. If we fail, set OrigBase to null.
while (Offset) {
// Indexing into tail padding between struct/array elements.
if (uint64_t(Offset*8) >= TD->getTypeSizeInBits(Ty))
return 0;
-
+
if (StructType *STy = dyn_cast<StructType>(Ty)) {
const StructLayout *SL = TD->getStructLayout(STy);
assert(Offset < (int64_t)SL->getSizeInBytes() &&
"Offset must stay within the indexed type");
-
+
unsigned Elt = SL->getElementContainingOffset(Offset);
NewIndices.push_back(ConstantInt::get(Type::getInt32Ty(Ty->getContext()),
Elt));
-
+
Offset -= SL->getElementOffset(Elt);
Ty = STy->getElementType(Elt);
} else if (ArrayType *AT = dyn_cast<ArrayType>(Ty)) {
@@ -787,7 +791,7 @@ Type *InstCombiner::FindElementAtOffset(Type *Ty, int64_t Offset,
return 0;
}
}
-
+
return Ty;
}
@@ -948,7 +952,7 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
Res->setIsInBounds(GEP.isInBounds());
return Res;
}
-
+
if (ArrayType *XATy =
dyn_cast<ArrayType>(StrippedPtrTy->getElementType())){
// GEP (bitcast [10 x i8]* X to [0 x i8]*), i32 0, ... ?
@@ -981,16 +985,16 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
// V and GEP are both pointer types --> BitCast
return new BitCastInst(NewGEP, GEP.getType());
}
-
+
// Transform things like:
// getelementptr i8* bitcast ([100 x double]* X to i8*), i32 %tmp
// (where tmp = 8*tmp2) into:
// getelementptr [100 x double]* %arr, i32 0, i32 %tmp2; bitcast
-
+
if (TD && SrcElTy->isArrayTy() && ResElTy->isIntegerTy(8)) {
uint64_t ArrayEltSize =
TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType());
-
+
// Check to see if "tmp" is a scale by a multiple of ArrayEltSize. We
// allow either a mul, shift, or constant here.
Value *NewIdx = 0;
@@ -1015,7 +1019,7 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
NewIdx = Inst->getOperand(0);
}
}
-
+
// If the index will be to exactly the right offset with the scale taken
// out, perform the transformation. Note, we don't know whether Scale is
// signed or not. We'll use unsigned version of division/modulo
@@ -1054,10 +1058,9 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
!isa<BitCastInst>(BCI->getOperand(0)) && GEP.hasAllConstantIndices() &&
StrippedPtrTy->getAddressSpace() == GEP.getPointerAddressSpace()) {
- // Determine how much the GEP moves the pointer. We are guaranteed to get
- // a constant back from EmitGEPOffset.
- ConstantInt *OffsetV = cast<ConstantInt>(EmitGEPOffset(&GEP));
- int64_t Offset = OffsetV->getSExtValue();
+ // Determine how much the GEP moves the pointer.
+ SmallVector<Value*, 8> Ops(GEP.idx_begin(), GEP.idx_end());
+ int64_t Offset = TD->getIndexedOffset(GEP.getPointerOperandType(), Ops);
// If this GEP instruction doesn't move the pointer, just replace the GEP
// with a bitcast of the real input to the dest type.
@@ -1065,7 +1068,7 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
// If the bitcast is of an allocation, and the allocation will be
// converted to match the type of the cast, don't touch this.
if (isa<AllocaInst>(BCI->getOperand(0)) ||
- isMalloc(BCI->getOperand(0))) {
+ isAllocationFn(BCI->getOperand(0))) {
// See if the bitcast simplifies, if so, don't nuke this GEP yet.
if (Instruction *I = visitBitCast(*BCI)) {
if (I != BCI) {
@@ -1078,7 +1081,7 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
}
return new BitCastInst(BCI->getOperand(0), GEP.getType());
}
-
+
// Otherwise, if the offset is non-zero, we need to find out if there is a
// field at Offset in 'A's type. If so, we can pull the cast through the
// GEP.
@@ -1089,68 +1092,103 @@ Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
Value *NGEP = GEP.isInBounds() ?
Builder->CreateInBoundsGEP(BCI->getOperand(0), NewIndices) :
Builder->CreateGEP(BCI->getOperand(0), NewIndices);
-
+
if (NGEP->getType() == GEP.getType())
return ReplaceInstUsesWith(GEP, NGEP);
NGEP->takeName(&GEP);
return new BitCastInst(NGEP, GEP.getType());
}
}
- }
-
+ }
+
return 0;
}
-static bool IsOnlyNullComparedAndFreed(Value *V, SmallVectorImpl<WeakVH> &Users,
- int Depth = 0) {
- if (Depth == 8)
- return false;
+static bool
+isAllocSiteRemovable(Instruction *AI, SmallVectorImpl<WeakVH> &Users) {
+ SmallVector<Instruction*, 4> Worklist;
+ Worklist.push_back(AI);
- for (Value::use_iterator UI = V->use_begin(), UE = V->use_end();
- UI != UE; ++UI) {
- User *U = *UI;
- if (isFreeCall(U)) {
- Users.push_back(U);
- continue;
- }
- if (ICmpInst *ICI = dyn_cast<ICmpInst>(U)) {
- if (ICI->isEquality() && isa<ConstantPointerNull>(ICI->getOperand(1))) {
- Users.push_back(ICI);
+ do {
+ Instruction *PI = Worklist.pop_back_val();
+ for (Value::use_iterator UI = PI->use_begin(), UE = PI->use_end(); UI != UE;
+ ++UI) {
+ Instruction *I = cast<Instruction>(*UI);
+ switch (I->getOpcode()) {
+ default:
+ // Give up the moment we see something we can't handle.
+ return false;
+
+ case Instruction::BitCast:
+ case Instruction::GetElementPtr:
+ Users.push_back(I);
+ Worklist.push_back(I);
continue;
- }
- }
- if (BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
- if (IsOnlyNullComparedAndFreed(BCI, Users, Depth+1)) {
- Users.push_back(BCI);
+
+ case Instruction::ICmp: {
+ ICmpInst *ICI = cast<ICmpInst>(I);
+ // We can fold eq/ne comparisons with null to false/true, respectively.
+ if (!ICI->isEquality() || !isa<ConstantPointerNull>(ICI->getOperand(1)))
+ return false;
+ Users.push_back(I);
continue;
}
- }
- if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(U)) {
- if (IsOnlyNullComparedAndFreed(GEPI, Users, Depth+1)) {
- Users.push_back(GEPI);
+
+ case Instruction::Call:
+ // Ignore no-op and store intrinsics.
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+ switch (II->getIntrinsicID()) {
+ default:
+ return false;
+
+ case Intrinsic::memmove:
+ case Intrinsic::memcpy:
+ case Intrinsic::memset: {
+ MemIntrinsic *MI = cast<MemIntrinsic>(II);
+ if (MI->isVolatile() || MI->getRawDest() != PI)
+ return false;
+ }
+ // fall through
+ case Intrinsic::dbg_declare:
+ case Intrinsic::dbg_value:
+ case Intrinsic::invariant_start:
+ case Intrinsic::invariant_end:
+ case Intrinsic::lifetime_start:
+ case Intrinsic::lifetime_end:
+ case Intrinsic::objectsize:
+ Users.push_back(I);
+ continue;
+ }
+ }
+
+ if (isFreeCall(I)) {
+ Users.push_back(I);
+ continue;
+ }
+ return false;
+
+ case Instruction::Store: {
+ StoreInst *SI = cast<StoreInst>(I);
+ if (SI->isVolatile() || SI->getPointerOperand() != PI)
+ return false;
+ Users.push_back(I);
continue;
}
- }
- if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
- if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
- II->getIntrinsicID() == Intrinsic::lifetime_end) {
- Users.push_back(II);
- continue;
}
+ llvm_unreachable("missing a return?");
}
- return false;
- }
+ } while (!Worklist.empty());
return true;
}
-Instruction *InstCombiner::visitMalloc(Instruction &MI) {
+Instruction *InstCombiner::visitAllocSite(Instruction &MI) {
// If we have a malloc call which is only used in any amount of comparisons
// to null and free calls, delete the calls and replace the comparisons with
// true or false as appropriate.
SmallVector<WeakVH, 64> Users;
- if (IsOnlyNullComparedAndFreed(&MI, Users)) {
+ if (isAllocSiteRemovable(&MI, Users)) {
for (unsigned i = 0, e = Users.size(); i != e; ++i) {
Instruction *I = cast_or_null<Instruction>(&*Users[i]);
if (!I) continue;
@@ -1161,9 +1199,23 @@ Instruction *InstCombiner::visitMalloc(Instruction &MI) {
C->isFalseWhenEqual()));
} else if (isa<BitCastInst>(I) || isa<GetElementPtrInst>(I)) {
ReplaceInstUsesWith(*I, UndefValue::get(I->getType()));
+ } else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+ if (II->getIntrinsicID() == Intrinsic::objectsize) {
+ ConstantInt *CI = cast<ConstantInt>(II->getArgOperand(1));
+ uint64_t DontKnow = CI->isZero() ? -1ULL : 0;
+ ReplaceInstUsesWith(*I, ConstantInt::get(I->getType(), DontKnow));
+ }
}
EraseInstFromFunction(*I);
}
+
+ if (InvokeInst *II = dyn_cast<InvokeInst>(&MI)) {
+ // Replace invoke with a NOP intrinsic to maintain the original CFG
+ Module *M = II->getParent()->getParent()->getParent();
+ Function *F = Intrinsic::getDeclaration(M, Intrinsic::donothing);
+ InvokeInst::Create(F, II->getNormalDest(), II->getUnwindDest(),
+ ArrayRef<Value *>(), "", II->getParent());
+ }
return EraseInstFromFunction(MI);
}
return 0;
@@ -1181,7 +1233,7 @@ Instruction *InstCombiner::visitFree(CallInst &FI) {
UndefValue::get(Type::getInt1PtrTy(FI.getContext())));
return EraseInstFromFunction(FI);
}
-
+
// If we have 'free null' delete the instruction. This can happen in stl code
// when lots of inlining happens.
if (isa<ConstantPointerNull>(Op))
@@ -1207,14 +1259,14 @@ Instruction *InstCombiner::visitBranchInst(BranchInst &BI) {
// Cannonicalize fcmp_one -> fcmp_oeq
FCmpInst::Predicate FPred; Value *Y;
- if (match(&BI, m_Br(m_FCmp(FPred, m_Value(X), m_Value(Y)),
+ if (match(&BI, m_Br(m_FCmp(FPred, m_Value(X), m_Value(Y)),
TrueDest, FalseDest)) &&
BI.getCondition()->hasOneUse())
if (FPred == FCmpInst::FCMP_ONE || FPred == FCmpInst::FCMP_OLE ||
FPred == FCmpInst::FCMP_OGE) {
FCmpInst *Cond = cast<FCmpInst>(BI.getCondition());
Cond->setPredicate(FCmpInst::getInversePredicate(FPred));
-
+
// Swap Destinations and condition.
BI.swapSuccessors();
Worklist.Add(Cond);
@@ -1280,7 +1332,7 @@ Instruction *InstCombiner::visitExtractValueInst(ExtractValueInst &EV) {
}
return 0; // Can't handle other constants
}
-
+
if (InsertValueInst *IV = dyn_cast<InsertValueInst>(Agg)) {
// We're extracting from an insertvalue instruction, compare the indices
const unsigned *exti, *exte, *insi, *inse;
@@ -1329,7 +1381,7 @@ Instruction *InstCombiner::visitExtractValueInst(ExtractValueInst &EV) {
// %E = extractvalue { i32, { i32 } } %I, 1, 0
// with
// %E extractvalue { i32 } { i32 42 }, 0
- return ExtractValueInst::Create(IV->getInsertedValueOperand(),
+ return ExtractValueInst::Create(IV->getInsertedValueOperand(),
makeArrayRef(exti, exte));
}
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Agg)) {
@@ -1349,7 +1401,7 @@ Instruction *InstCombiner::visitExtractValueInst(ExtractValueInst &EV) {
EraseInstFromFunction(*II);
return BinaryOperator::CreateAdd(LHS, RHS);
}
-
+
// If the normal result of the add is dead, and the RHS is a constant,
// we can transform this into a range comparison.
// overflow = uadd a, -4 --> overflow = icmp ugt a, 3
@@ -1798,7 +1850,7 @@ static bool TryToSinkInstruction(Instruction *I, BasicBlock *DestBlock) {
/// many instructions are dead or constant). Additionally, if we find a branch
/// whose condition is a known constant, we only visit the reachable successors.
///
-static bool AddReachableCodeToWorklist(BasicBlock *BB,
+static bool AddReachableCodeToWorklist(BasicBlock *BB,
SmallPtrSet<BasicBlock*, 64> &Visited,
InstCombiner &IC,
const TargetData *TD,
@@ -1812,13 +1864,13 @@ static bool AddReachableCodeToWorklist(BasicBlock *BB,
do {
BB = Worklist.pop_back_val();
-
+
// We have now visited this block! If we've already been here, ignore it.
if (!Visited.insert(BB)) continue;
for (BasicBlock::iterator BBI = BB->begin(), E = BB->end(); BBI != E; ) {
Instruction *Inst = BBI++;
-
+
// DCE instruction if trivially dead.
if (isInstructionTriviallyDead(Inst)) {
++NumDeadInst;
@@ -1826,7 +1878,7 @@ static bool AddReachableCodeToWorklist(BasicBlock *BB,
Inst->eraseFromParent();
continue;
}
-
+
// ConstantProp instruction if trivially constant.
if (!Inst->use_empty() && isa<Constant>(Inst->getOperand(0)))
if (Constant *C = ConstantFoldInstruction(Inst, TD, TLI)) {
@@ -1837,7 +1889,7 @@ static bool AddReachableCodeToWorklist(BasicBlock *BB,
Inst->eraseFromParent();
continue;
}
-
+
if (TD) {
// See if we can constant fold its operands.
for (User::op_iterator i = Inst->op_begin(), e = Inst->op_end();
@@ -1881,17 +1933,17 @@ static bool AddReachableCodeToWorklist(BasicBlock *BB,
Worklist.push_back(ReachableBB);
continue;
}
-
+
// Otherwise it is the default destination.
Worklist.push_back(SI->getDefaultDest());
continue;
}
}
-
+
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
Worklist.push_back(TI->getSuccessor(i));
} while (!Worklist.empty());
-
+
// Once we've found all of the instructions to add to instcombine's worklist,
// add them in reverse order. This way instcombine will visit from the top
// of the function down. This jives well with the way that it adds all uses
@@ -1899,13 +1951,13 @@ static bool AddReachableCodeToWorklist(BasicBlock *BB,
// some N^2 behavior in pathological cases.
IC.Worklist.AddInitialGroup(&InstrsForInstCombineWorklist[0],
InstrsForInstCombineWorklist.size());
-
+
return MadeIRChange;
}
bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) {
MadeIRChange = false;
-
+
DEBUG(errs() << "\n\nINSTCOMBINE ITERATION #" << Iteration << " on "
<< F.getName() << "\n");
@@ -1976,13 +2028,13 @@ bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) {
BasicBlock *BB = I->getParent();
Instruction *UserInst = cast<Instruction>(I->use_back());
BasicBlock *UserParent;
-
+
// Get the block the use occurs in.
if (PHINode *PN = dyn_cast<PHINode>(UserInst))
UserParent = PN->getIncomingBlock(I->use_begin().getUse());
else
UserParent = UserInst->getParent();
-
+
if (UserParent != BB) {
bool UserIsSuccessor = false;
// See if the user is one of our successors.
@@ -2004,7 +2056,7 @@ bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) {
// Now that we have an instruction, try combining it to simplify it.
Builder->SetInsertPoint(I->getParent(), I);
Builder->SetCurrentDebugLocation(I->getDebugLoc());
-
+
#ifndef NDEBUG
std::string OrigI;
#endif
@@ -2069,14 +2121,14 @@ bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) {
bool InstCombiner::runOnFunction(Function &F) {
TD = getAnalysisIfAvailable<TargetData>();
TLI = &getAnalysis<TargetLibraryInfo>();
-
+
/// Builder - This is an IRBuilder that automatically inserts new
/// instructions into the worklist when they are created.
- IRBuilder<true, TargetFolder, InstCombineIRInserter>
+ IRBuilder<true, TargetFolder, InstCombineIRInserter>
TheBuilder(F.getContext(), TargetFolder(TD),
InstCombineIRInserter(Worklist));
Builder = &TheBuilder;
-
+
bool EverMadeChange = false;
// Lower dbg.declare intrinsics otherwise their value may be clobbered
@@ -2087,7 +2139,7 @@ bool InstCombiner::runOnFunction(Function &F) {
unsigned Iteration = 0;
while (DoOneIteration(F, Iteration++))
EverMadeChange = true;
-
+
Builder = 0;
return EverMadeChange;
}
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-18 04:47:49 +0000