//
//===----------------------------------------------------------------------===//
-#include "InstCombine.h"
+#include "InstCombineInternal.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/IR/DataLayout.h"
-#include "llvm/Support/PatternMatch.h"
-#include "llvm/Target/TargetLibraryInfo.h"
+#include "llvm/IR/PatternMatch.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
using namespace llvm;
using namespace PatternMatch;
+#define DEBUG_TYPE "instcombine"
+
/// DecomposeSimpleLinearExpr - Analyze 'Val', seeing if it is a simple linear
/// expression. If so, decompose it, returning some value X, such that Val is
/// X*Scale+Offset.
Instruction *InstCombiner::PromoteCastOfAllocation(BitCastInst &CI,
AllocaInst &AI) {
// This requires DataLayout to get the alloca alignment and size information.
- if (!DL) return 0;
+ if (!DL) return nullptr;
PointerType *PTy = cast<PointerType>(CI.getType());
// Get the type really allocated and the type casted to.
Type *AllocElTy = AI.getAllocatedType();
Type *CastElTy = PTy->getElementType();
- if (!AllocElTy->isSized() || !CastElTy->isSized()) return 0;
+ if (!AllocElTy->isSized() || !CastElTy->isSized()) return nullptr;
unsigned AllocElTyAlign = DL->getABITypeAlignment(AllocElTy);
unsigned CastElTyAlign = DL->getABITypeAlignment(CastElTy);
- if (CastElTyAlign < AllocElTyAlign) return 0;
+ if (CastElTyAlign < AllocElTyAlign) return nullptr;
// If the allocation has multiple uses, only promote it if we are strictly
// increasing the alignment of the resultant allocation. If we keep it the
// same, we open the door to infinite loops of various kinds.
- if (!AI.hasOneUse() && CastElTyAlign == AllocElTyAlign) return 0;
+ if (!AI.hasOneUse() && CastElTyAlign == AllocElTyAlign) return nullptr;
uint64_t AllocElTySize = DL->getTypeAllocSize(AllocElTy);
uint64_t CastElTySize = DL->getTypeAllocSize(CastElTy);
- if (CastElTySize == 0 || AllocElTySize == 0) return 0;
+ if (CastElTySize == 0 || AllocElTySize == 0) return nullptr;
// If the allocation has multiple uses, only promote it if we're not
// shrinking the amount of memory being allocated.
uint64_t AllocElTyStoreSize = DL->getTypeStoreSize(AllocElTy);
uint64_t CastElTyStoreSize = DL->getTypeStoreSize(CastElTy);
- if (!AI.hasOneUse() && CastElTyStoreSize < AllocElTyStoreSize) return 0;
+ if (!AI.hasOneUse() && CastElTyStoreSize < AllocElTyStoreSize) return nullptr;
// See if we can satisfy the modulus by pulling a scale out of the array
// size argument.
// If we can now satisfy the modulus, by using a non-1 scale, we really can
// do the xform.
if ((AllocElTySize*ArraySizeScale) % CastElTySize != 0 ||
- (AllocElTySize*ArrayOffset ) % CastElTySize != 0) return 0;
+ (AllocElTySize*ArrayOffset ) % CastElTySize != 0) return nullptr;
unsigned Scale = (AllocElTySize*ArraySizeScale)/CastElTySize;
- Value *Amt = 0;
+ Value *Amt = nullptr;
if (Scale == 1) {
Amt = NumElements;
} else {
AllocaInst *New = AllocaBuilder.CreateAlloca(CastElTy, Amt);
New->setAlignment(AI.getAlignment());
New->takeName(&AI);
+ New->setUsedWithInAlloca(AI.isUsedWithInAlloca());
// If the allocation has multiple real uses, insert a cast and change all
// things that used it to use the new cast. This will also hack on CI, but it
// Otherwise, it must be an instruction.
Instruction *I = cast<Instruction>(V);
- Instruction *Res = 0;
+ Instruction *Res = nullptr;
unsigned Opc = I->getOpcode();
switch (Opc) {
case Instruction::Add:
Instruction::CastOps firstOp = Instruction::CastOps(CI->getOpcode());
Instruction::CastOps secondOp = Instruction::CastOps(opcode);
Type *SrcIntPtrTy = DL && SrcTy->isPtrOrPtrVectorTy() ?
- DL->getIntPtrType(SrcTy) : 0;
+ DL->getIntPtrType(SrcTy) : nullptr;
Type *MidIntPtrTy = DL && MidTy->isPtrOrPtrVectorTy() ?
- DL->getIntPtrType(MidTy) : 0;
+ DL->getIntPtrType(MidTy) : nullptr;
Type *DstIntPtrTy = DL && DstTy->isPtrOrPtrVectorTy() ?
- DL->getIntPtrType(DstTy) : 0;
+ DL->getIntPtrType(DstTy) : nullptr;
unsigned Res = CastInst::isEliminableCastPair(firstOp, secondOp, SrcTy, MidTy,
DstTy, SrcIntPtrTy, MidIntPtrTy,
DstIntPtrTy);
return NV;
}
- return 0;
+ return nullptr;
}
/// CanEvaluateTruncated - Return true if we can evaluate the specified
///
/// This function works on both vectors and scalars.
///
-static bool CanEvaluateTruncated(Value *V, Type *Ty) {
+static bool CanEvaluateTruncated(Value *V, Type *Ty, InstCombiner &IC,
+ Instruction *CxtI) {
// We can always evaluate constants in another type.
if (isa<Constant>(V))
return true;
case Instruction::Or:
case Instruction::Xor:
// These operators can all arbitrarily be extended or truncated.
- return CanEvaluateTruncated(I->getOperand(0), Ty) &&
- CanEvaluateTruncated(I->getOperand(1), Ty);
+ return CanEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI) &&
+ CanEvaluateTruncated(I->getOperand(1), Ty, IC, CxtI);
case Instruction::UDiv:
case Instruction::URem: {
uint32_t BitWidth = Ty->getScalarSizeInBits();
if (BitWidth < OrigBitWidth) {
APInt Mask = APInt::getHighBitsSet(OrigBitWidth, OrigBitWidth-BitWidth);
- if (MaskedValueIsZero(I->getOperand(0), Mask) &&
- MaskedValueIsZero(I->getOperand(1), Mask)) {
- return CanEvaluateTruncated(I->getOperand(0), Ty) &&
- CanEvaluateTruncated(I->getOperand(1), Ty);
+ if (IC.MaskedValueIsZero(I->getOperand(0), Mask, 0, CxtI) &&
+ IC.MaskedValueIsZero(I->getOperand(1), Mask, 0, CxtI)) {
+ return CanEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI) &&
+ CanEvaluateTruncated(I->getOperand(1), Ty, IC, CxtI);
}
}
break;
if (ConstantInt *CI = dyn_cast<ConstantInt>(I->getOperand(1))) {
uint32_t BitWidth = Ty->getScalarSizeInBits();
if (CI->getLimitedValue(BitWidth) < BitWidth)
- return CanEvaluateTruncated(I->getOperand(0), Ty);
+ return CanEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI);
}
break;
case Instruction::LShr:
if (ConstantInt *CI = dyn_cast<ConstantInt>(I->getOperand(1))) {
uint32_t OrigBitWidth = OrigTy->getScalarSizeInBits();
uint32_t BitWidth = Ty->getScalarSizeInBits();
- if (MaskedValueIsZero(I->getOperand(0),
- APInt::getHighBitsSet(OrigBitWidth, OrigBitWidth-BitWidth)) &&
+ if (IC.MaskedValueIsZero(I->getOperand(0),
+ APInt::getHighBitsSet(OrigBitWidth, OrigBitWidth-BitWidth), 0, CxtI) &&
CI->getLimitedValue(BitWidth) < BitWidth) {
- return CanEvaluateTruncated(I->getOperand(0), Ty);
+ return CanEvaluateTruncated(I->getOperand(0), Ty, IC, CxtI);
}
}
break;
return true;
case Instruction::Select: {
SelectInst *SI = cast<SelectInst>(I);
- return CanEvaluateTruncated(SI->getTrueValue(), Ty) &&
- CanEvaluateTruncated(SI->getFalseValue(), Ty);
+ return CanEvaluateTruncated(SI->getTrueValue(), Ty, IC, CxtI) &&
+ CanEvaluateTruncated(SI->getFalseValue(), Ty, IC, CxtI);
}
case Instruction::PHI: {
// We can change a phi if we can change all operands. Note that we never
// instructions with a single use.
PHINode *PN = cast<PHINode>(I);
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
- if (!CanEvaluateTruncated(PN->getIncomingValue(i), Ty))
+ if (!CanEvaluateTruncated(PN->getIncomingValue(i), Ty, IC, CxtI))
return false;
return true;
}
// expression tree to something weird like i93 unless the source is also
// strange.
if ((DestTy->isVectorTy() || ShouldChangeType(SrcTy, DestTy)) &&
- CanEvaluateTruncated(Src, DestTy)) {
+ CanEvaluateTruncated(Src, DestTy, *this, &CI)) {
// If this cast is a truncate, evaluting in a different type always
// eliminates the cast, so it is always a win.
}
// Transform trunc(lshr (zext A), Cst) to eliminate one type conversion.
- Value *A = 0; ConstantInt *Cst = 0;
+ Value *A = nullptr; ConstantInt *Cst = nullptr;
if (Src->hasOneUse() &&
match(Src, m_LShr(m_ZExt(m_Value(A)), m_ConstantInt(Cst)))) {
// We have three types to worry about here, the type of A, the source of
ConstantExpr::getTrunc(Cst, CI.getType()));
}
- return 0;
+ return nullptr;
}
/// transformZExtICmp - Transform (zext icmp) to bitwise / integer operations
// If Op1C some other power of two, convert:
uint32_t BitWidth = Op1C->getType()->getBitWidth();
APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
- ComputeMaskedBits(ICI->getOperand(0), KnownZero, KnownOne);
+ computeKnownBits(ICI->getOperand(0), KnownZero, KnownOne, 0, &CI);
APInt KnownZeroMask(~KnownZero);
if (KnownZeroMask.isPowerOf2()) { // Exactly 1 possible 1?
APInt KnownZeroLHS(BitWidth, 0), KnownOneLHS(BitWidth, 0);
APInt KnownZeroRHS(BitWidth, 0), KnownOneRHS(BitWidth, 0);
- ComputeMaskedBits(LHS, KnownZeroLHS, KnownOneLHS);
- ComputeMaskedBits(RHS, KnownZeroRHS, KnownOneRHS);
+ computeKnownBits(LHS, KnownZeroLHS, KnownOneLHS, 0, &CI);
+ computeKnownBits(RHS, KnownZeroRHS, KnownOneRHS, 0, &CI);
if (KnownZeroLHS == KnownZeroRHS && KnownOneLHS == KnownOneRHS) {
APInt KnownBits = KnownZeroLHS | KnownOneLHS;
}
}
- return 0;
+ return nullptr;
}
/// CanEvaluateZExtd - Determine if the specified value can be computed in the
/// clear the top bits anyway, doing this has no extra cost.
///
/// This function works on both vectors and scalars.
-static bool CanEvaluateZExtd(Value *V, Type *Ty, unsigned &BitsToClear) {
+static bool CanEvaluateZExtd(Value *V, Type *Ty, unsigned &BitsToClear,
+ InstCombiner &IC, Instruction *CxtI) {
BitsToClear = 0;
if (isa<Constant>(V))
return true;
case Instruction::Add:
case Instruction::Sub:
case Instruction::Mul:
- if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear) ||
- !CanEvaluateZExtd(I->getOperand(1), Ty, Tmp))
+ if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear, IC, CxtI) ||
+ !CanEvaluateZExtd(I->getOperand(1), Ty, Tmp, IC, CxtI))
return false;
// These can all be promoted if neither operand has 'bits to clear'.
if (BitsToClear == 0 && Tmp == 0)
// We use MaskedValueIsZero here for generality, but the case we care
// about the most is constant RHS.
unsigned VSize = V->getType()->getScalarSizeInBits();
- if (MaskedValueIsZero(I->getOperand(1),
- APInt::getHighBitsSet(VSize, BitsToClear)))
+ if (IC.MaskedValueIsZero(I->getOperand(1),
+ APInt::getHighBitsSet(VSize, BitsToClear),
+ 0, CxtI))
return true;
}
// We can promote shl(x, cst) if we can promote x. Since shl overwrites the
// upper bits we can reduce BitsToClear by the shift amount.
if (ConstantInt *Amt = dyn_cast<ConstantInt>(I->getOperand(1))) {
- if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear))
+ if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear, IC, CxtI))
return false;
uint64_t ShiftAmt = Amt->getZExtValue();
BitsToClear = ShiftAmt < BitsToClear ? BitsToClear - ShiftAmt : 0;
// We can promote lshr(x, cst) if we can promote x. This requires the
// ultimate 'and' to clear out the high zero bits we're clearing out though.
if (ConstantInt *Amt = dyn_cast<ConstantInt>(I->getOperand(1))) {
- if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear))
+ if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear, IC, CxtI))
return false;
BitsToClear += Amt->getZExtValue();
if (BitsToClear > V->getType()->getScalarSizeInBits())
// Cannot promote variable LSHR.
return false;
case Instruction::Select:
- if (!CanEvaluateZExtd(I->getOperand(1), Ty, Tmp) ||
- !CanEvaluateZExtd(I->getOperand(2), Ty, BitsToClear) ||
+ if (!CanEvaluateZExtd(I->getOperand(1), Ty, Tmp, IC, CxtI) ||
+ !CanEvaluateZExtd(I->getOperand(2), Ty, BitsToClear, IC, CxtI) ||
// TODO: If important, we could handle the case when the BitsToClear are
// known zero in the disagreeing side.
Tmp != BitsToClear)
// get into trouble with cyclic PHIs here because we only consider
// instructions with a single use.
PHINode *PN = cast<PHINode>(I);
- if (!CanEvaluateZExtd(PN->getIncomingValue(0), Ty, BitsToClear))
+ if (!CanEvaluateZExtd(PN->getIncomingValue(0), Ty, BitsToClear, IC, CxtI))
return false;
for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i)
- if (!CanEvaluateZExtd(PN->getIncomingValue(i), Ty, Tmp) ||
+ if (!CanEvaluateZExtd(PN->getIncomingValue(i), Ty, Tmp, IC, CxtI) ||
// TODO: If important, we could handle the case when the BitsToClear
// are known zero in the disagreeing input.
Tmp != BitsToClear)
Instruction *InstCombiner::visitZExt(ZExtInst &CI) {
// If this zero extend is only used by a truncate, let the truncate be
// eliminated before we try to optimize this zext.
- if (CI.hasOneUse() && isa<TruncInst>(CI.use_back()))
- return 0;
+ if (CI.hasOneUse() && isa<TruncInst>(CI.user_back()))
+ return nullptr;
// If one of the common conversion will work, do it.
if (Instruction *Result = commonCastTransforms(CI))
// strange.
unsigned BitsToClear;
if ((DestTy->isVectorTy() || ShouldChangeType(SrcTy, DestTy)) &&
- CanEvaluateZExtd(Src, DestTy, BitsToClear)) {
+ CanEvaluateZExtd(Src, DestTy, BitsToClear, *this, &CI)) {
assert(BitsToClear < SrcTy->getScalarSizeInBits() &&
"Unreasonable BitsToClear");
// If the high bits are already filled with zeros, just replace this
// cast with the result.
- if (MaskedValueIsZero(Res, APInt::getHighBitsSet(DestBitSize,
- DestBitSize-SrcBitsKept)))
+ if (MaskedValueIsZero(Res,
+ APInt::getHighBitsSet(DestBitSize,
+ DestBitSize-SrcBitsKept),
+ 0, &CI))
return ReplaceInstUsesWith(CI, Res);
// We need to emit an AND to clear the high bits.
return BinaryOperator::CreateXor(New, ConstantInt::get(CI.getType(), 1));
}
- return 0;
+ return nullptr;
}
/// transformSExtICmp - Transform (sext icmp) to bitwise / integer operations
Value *Op0 = ICI->getOperand(0), *Op1 = ICI->getOperand(1);
ICmpInst::Predicate Pred = ICI->getPredicate();
+ // Don't bother if Op1 isn't of vector or integer type.
+ if (!Op1->getType()->isIntOrIntVectorTy())
+ return nullptr;
+
if (Constant *Op1C = dyn_cast<Constant>(Op1)) {
// (x <s 0) ? -1 : 0 -> ashr x, 31 -> all ones if negative
// (x >s -1) ? -1 : 0 -> not (ashr x, 31) -> all ones if positive
ICI->isEquality() && (Op1C->isZero() || Op1C->getValue().isPowerOf2())){
unsigned BitWidth = Op1C->getType()->getBitWidth();
APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
- ComputeMaskedBits(Op0, KnownZero, KnownOne);
+ computeKnownBits(Op0, KnownZero, KnownOne, 0, &CI);
APInt KnownZeroMask(~KnownZero);
if (KnownZeroMask.isPowerOf2()) {
}
}
- return 0;
+ return nullptr;
}
/// CanEvaluateSExtd - Return true if we can take the specified value
Instruction *InstCombiner::visitSExt(SExtInst &CI) {
// If this sign extend is only used by a truncate, let the truncate be
// eliminated before we try to optimize this sext.
- if (CI.hasOneUse() && isa<TruncInst>(CI.use_back()))
- return 0;
+ if (CI.hasOneUse() && isa<TruncInst>(CI.user_back()))
+ return nullptr;
if (Instruction *I = commonCastTransforms(CI))
return I;
Value *Src = CI.getOperand(0);
Type *SrcTy = Src->getType(), *DestTy = CI.getType();
+ // If we know that the value being extended is positive, we can use a zext
+ // instead.
+ bool KnownZero, KnownOne;
+ ComputeSignBit(Src, KnownZero, KnownOne, 0, &CI);
+ if (KnownZero) {
+ Value *ZExt = Builder->CreateZExt(Src, DestTy);
+ return ReplaceInstUsesWith(CI, ZExt);
+ }
+
// Attempt to extend the entire input expression tree to the destination
// type. Only do this if the dest type is a simple type, don't convert the
// expression tree to something weird like i93 unless the source is also
// If the high bits are already filled with sign bit, just replace this
// cast with the result.
- if (ComputeNumSignBits(Res) > DestBitSize - SrcBitSize)
+ if (ComputeNumSignBits(Res, 0, &CI) > DestBitSize - SrcBitSize)
return ReplaceInstUsesWith(CI, Res);
// We need to emit a shl + ashr to do the sign extend.
// into:
// %a = shl i32 %i, 30
// %d = ashr i32 %a, 30
- Value *A = 0;
+ Value *A = nullptr;
// TODO: Eventually this could be subsumed by EvaluateInDifferentType.
- ConstantInt *BA = 0, *CA = 0;
+ ConstantInt *BA = nullptr, *CA = nullptr;
if (match(Src, m_AShr(m_Shl(m_Trunc(m_Value(A)), m_ConstantInt(BA)),
m_ConstantInt(CA))) &&
BA == CA && A->getType() == CI.getType()) {
return BinaryOperator::CreateAShr(A, ShAmtV);
}
- return 0;
+ return nullptr;
}
(void)F.convert(Sem, APFloat::rmNearestTiesToEven, &losesInfo);
if (!losesInfo)
return ConstantFP::get(CFP->getContext(), F);
- return 0;
+ return nullptr;
}
/// LookThroughFPExtensions - If this is an fp extension instruction, look
// type of OpI doesn't enter into things at all. We simply evaluate
// in whichever source type is larger, then convert to the
// destination type.
+ if (SrcWidth == OpWidth)
+ break;
if (LHSWidth < SrcWidth)
LHSOrig = Builder->CreateFPExt(LHSOrig, RHSOrig->getType());
else if (RHSWidth <= SrcWidth)
RHSOrig = Builder->CreateFPExt(RHSOrig, LHSOrig->getType());
- Value *ExactResult = Builder->CreateFRem(LHSOrig, RHSOrig);
- if (Instruction *RI = dyn_cast<Instruction>(ExactResult))
- RI->copyFastMathFlags(OpI);
- return CastInst::CreateFPCast(ExactResult, CI.getType());
+ if (LHSOrig != OpI->getOperand(0) || RHSOrig != OpI->getOperand(1)) {
+ Value *ExactResult = Builder->CreateFRem(LHSOrig, RHSOrig);
+ if (Instruction *RI = dyn_cast<Instruction>(ExactResult))
+ RI->copyFastMathFlags(OpI);
+ return CastInst::CreateFPCast(ExactResult, CI.getType());
+ }
}
// (fptrunc (fneg x)) -> (fneg (fptrunc x))
}
}
- // Fold (fptrunc (sqrt (fpext x))) -> (sqrtf x)
- // Note that we restrict this transformation based on
- // TLI->has(LibFunc::sqrtf), even for the sqrt intrinsic, because
- // TLI->has(LibFunc::sqrtf) is sufficient to guarantee that the
- // single-precision intrinsic can be expanded in the backend.
- CallInst *Call = dyn_cast<CallInst>(CI.getOperand(0));
- if (Call && Call->getCalledFunction() && TLI->has(LibFunc::sqrtf) &&
- (Call->getCalledFunction()->getName() == TLI->getName(LibFunc::sqrt) ||
- Call->getCalledFunction()->getIntrinsicID() == Intrinsic::sqrt) &&
- Call->getNumArgOperands() == 1 &&
- Call->hasOneUse()) {
- CastInst *Arg = dyn_cast<CastInst>(Call->getArgOperand(0));
- if (Arg && Arg->getOpcode() == Instruction::FPExt &&
- CI.getType()->isFloatTy() &&
- Call->getType()->isDoubleTy() &&
- Arg->getType()->isDoubleTy() &&
- Arg->getOperand(0)->getType()->isFloatTy()) {
- Function *Callee = Call->getCalledFunction();
- Module *M = CI.getParent()->getParent()->getParent();
- Constant *SqrtfFunc = (Callee->getIntrinsicID() == Intrinsic::sqrt) ?
- Intrinsic::getDeclaration(M, Intrinsic::sqrt, Builder->getFloatTy()) :
- M->getOrInsertFunction("sqrtf", Callee->getAttributes(),
- Builder->getFloatTy(), Builder->getFloatTy(),
- NULL);
- CallInst *ret = CallInst::Create(SqrtfFunc, Arg->getOperand(0),
- "sqrtfcall");
- ret->setAttributes(Callee->getAttributes());
-
-
- // Remove the old Call. With -fmath-errno, it won't get marked readnone.
- ReplaceInstUsesWith(*Call, UndefValue::get(Call->getType()));
- EraseInstFromFunction(*Call);
- return ret;
- }
- }
-
- return 0;
+ return nullptr;
}
Instruction *InstCombiner::visitFPExt(CastInst &CI) {
return commonCastTransforms(CI);
}
+// fpto{s/u}i({u/s}itofp(X)) --> X or zext(X) or sext(X) or trunc(X)
+// This is safe if the intermediate type has enough bits in its mantissa to
+// accurately represent all values of X. For example, this won't work with
+// i64 -> float -> i64.
+Instruction *InstCombiner::FoldItoFPtoI(Instruction &FI) {
+ if (!isa<UIToFPInst>(FI.getOperand(0)) && !isa<SIToFPInst>(FI.getOperand(0)))
+ return nullptr;
+ Instruction *OpI = cast<Instruction>(FI.getOperand(0));
+
+ Value *SrcI = OpI->getOperand(0);
+ Type *FITy = FI.getType();
+ Type *OpITy = OpI->getType();
+ Type *SrcTy = SrcI->getType();
+ bool IsInputSigned = isa<SIToFPInst>(OpI);
+ bool IsOutputSigned = isa<FPToSIInst>(FI);
+
+ // We can safely assume the conversion won't overflow the output range,
+ // because (for example) (uint8_t)18293.f is undefined behavior.
+
+ // Since we can assume the conversion won't overflow, our decision as to
+ // whether the input will fit in the float should depend on the minimum
+ // of the input range and output range.
+
+ // This means this is also safe for a signed input and unsigned output, since
+ // a negative input would lead to undefined behavior.
+ int InputSize = (int)SrcTy->getScalarSizeInBits() - IsInputSigned;
+ int OutputSize = (int)FITy->getScalarSizeInBits() - IsOutputSigned;
+ int ActualSize = std::min(InputSize, OutputSize);
+
+ if (ActualSize <= OpITy->getFPMantissaWidth()) {
+ if (FITy->getScalarSizeInBits() > SrcTy->getScalarSizeInBits()) {
+ if (IsInputSigned && IsOutputSigned)
+ return new SExtInst(SrcI, FITy);
+ return new ZExtInst(SrcI, FITy);
+ }
+ if (FITy->getScalarSizeInBits() < SrcTy->getScalarSizeInBits())
+ return new TruncInst(SrcI, FITy);
+ if (SrcTy == FITy)
+ return ReplaceInstUsesWith(FI, SrcI);
+ return new BitCastInst(SrcI, FITy);
+ }
+ return nullptr;
+}
+
Instruction *InstCombiner::visitFPToUI(FPToUIInst &FI) {
Instruction *OpI = dyn_cast<Instruction>(FI.getOperand(0));
- if (OpI == 0)
+ if (!OpI)
return commonCastTransforms(FI);
- // fptoui(uitofp(X)) --> X
- // fptoui(sitofp(X)) --> X
- // This is safe if the intermediate type has enough bits in its mantissa to
- // accurately represent all values of X. For example, do not do this with
- // i64->float->i64. This is also safe for sitofp case, because any negative
- // 'X' value would cause an undefined result for the fptoui.
- if ((isa<UIToFPInst>(OpI) || isa<SIToFPInst>(OpI)) &&
- OpI->getOperand(0)->getType() == FI.getType() &&
- (int)FI.getType()->getScalarSizeInBits() < /*extra bit for sign */
- OpI->getType()->getFPMantissaWidth())
- return ReplaceInstUsesWith(FI, OpI->getOperand(0));
+ if (Instruction *I = FoldItoFPtoI(FI))
+ return I;
return commonCastTransforms(FI);
}
Instruction *InstCombiner::visitFPToSI(FPToSIInst &FI) {
Instruction *OpI = dyn_cast<Instruction>(FI.getOperand(0));
- if (OpI == 0)
+ if (!OpI)
return commonCastTransforms(FI);
- // fptosi(sitofp(X)) --> X
- // fptosi(uitofp(X)) --> X
- // This is safe if the intermediate type has enough bits in its mantissa to
- // accurately represent all values of X. For example, do not do this with
- // i64->float->i64. This is also safe for sitofp case, because any negative
- // 'X' value would cause an undefined result for the fptoui.
- if ((isa<UIToFPInst>(OpI) || isa<SIToFPInst>(OpI)) &&
- OpI->getOperand(0)->getType() == FI.getType() &&
- (int)FI.getType()->getScalarSizeInBits() <=
- OpI->getType()->getFPMantissaWidth())
- return ReplaceInstUsesWith(FI, OpI->getOperand(0));
+ if (Instruction *I = FoldItoFPtoI(FI))
+ return I;
return commonCastTransforms(FI);
}
if (Instruction *I = commonCastTransforms(CI))
return I;
- return 0;
+ return nullptr;
}
/// @brief Implement the transforms for cast of pointer (bitcast/ptrtoint)
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Src)) {
// If casting the result of a getelementptr instruction with no offset, turn
// this into a cast of the original pointer!
- if (GEP->hasAllZeroIndices()) {
+ if (GEP->hasAllZeroIndices() &&
+ // If CI is an addrspacecast and GEP changes the poiner type, merging
+ // GEP into CI would undo canonicalizing addrspacecast with different
+ // pointer types, causing infinite loops.
+ (!isa<AddrSpaceCastInst>(CI) ||
+ GEP->getType() == GEP->getPointerOperand()->getType())) {
// Changing the cast operand is usually not a good idea but it is safe
// here because the pointer operand is being replaced with another
// pointer operand so the opcode doesn't need to change.
// there yet.
if (SrcTy->getElementType()->getPrimitiveSizeInBits() !=
DestTy->getElementType()->getPrimitiveSizeInBits())
- return 0;
+ return nullptr;
SrcTy = VectorType::get(DestTy->getElementType(), SrcTy->getNumElements());
InVal = IC.Builder->CreateBitCast(InVal, SrcTy);
ElementIndex = Elements.size() - ElementIndex - 1;
// Fail if multiple elements are inserted into this slot.
- if (Elements[ElementIndex] != 0)
+ if (Elements[ElementIndex])
return false;
Elements[ElementIndex] = V;
if (!V->hasOneUse()) return false;
Instruction *I = dyn_cast<Instruction>(V);
- if (I == 0) return false;
+ if (!I) return false;
switch (I->getOpcode()) {
default: return false; // Unhandled case.
case Instruction::BitCast:
case Instruction::Shl: {
// Must be shifting by a constant that is a multiple of the element size.
ConstantInt *CI = dyn_cast<ConstantInt>(I->getOperand(1));
- if (CI == 0) return false;
+ if (!CI) return false;
Shift += CI->getZExtValue();
if (!isMultipleOfTypeSize(Shift, VecEltTy)) return false;
return CollectInsertionElements(I->getOperand(0), Shift,
static Value *OptimizeIntegerToVectorInsertions(BitCastInst &CI,
InstCombiner &IC) {
// We need to know the target byte order to perform this optimization.
- if (!IC.getDataLayout()) return 0;
+ if (!IC.getDataLayout()) return nullptr;
VectorType *DestVecTy = cast<VectorType>(CI.getType());
Value *IntInput = CI.getOperand(0);
SmallVector<Value*, 8> Elements(DestVecTy->getNumElements());
if (!CollectInsertionElements(IntInput, 0, Elements,
DestVecTy->getElementType(), IC))
- return 0;
+ return nullptr;
// If we succeeded, we know that all of the element are specified by Elements
// or are zero if Elements has a null entry. Recast this as a set of
// insertions.
Value *Result = Constant::getNullValue(CI.getType());
for (unsigned i = 0, e = Elements.size(); i != e; ++i) {
- if (Elements[i] == 0) continue; // Unset element.
+ if (!Elements[i]) continue; // Unset element.
Result = IC.Builder->CreateInsertElement(Result, Elements[i],
IC.Builder->getInt32(i));
/// bitcast. The various long double bitcasts can't get in here.
static Instruction *OptimizeIntToFloatBitCast(BitCastInst &CI,InstCombiner &IC){
// We need to know the target byte order to perform this optimization.
- if (!IC.getDataLayout()) return 0;
+ if (!IC.getDataLayout()) return nullptr;
Value *Src = CI.getOperand(0);
Type *DestTy = CI.getType();
// If this is a bitcast from int to float, check to see if the int is an
// extraction from a vector.
- Value *VecInput = 0;
+ Value *VecInput = nullptr;
// bitcast(trunc(bitcast(somevector)))
if (match(Src, m_Trunc(m_BitCast(m_Value(VecInput)))) &&
isa<VectorType>(VecInput->getType())) {
}
// bitcast(trunc(lshr(bitcast(somevector), cst))
- ConstantInt *ShAmt = 0;
+ ConstantInt *ShAmt = nullptr;
if (match(Src, m_Trunc(m_LShr(m_BitCast(m_Value(VecInput)),
m_ConstantInt(ShAmt)))) &&
isa<VectorType>(VecInput->getType())) {
return ExtractElementInst::Create(VecInput, IC.Builder->getInt32(Elt));
}
}
- return 0;
+ return nullptr;
}
Instruction *InstCombiner::visitBitCast(BitCastInst &CI) {
}
Instruction *InstCombiner::visitAddrSpaceCast(AddrSpaceCastInst &CI) {
+ // If the destination pointer element type is not the same as the source's
+ // first do a bitcast to the destination type, and then the addrspacecast.
+ // This allows the cast to be exposed to other transforms.
+ Value *Src = CI.getOperand(0);
+ PointerType *SrcTy = cast<PointerType>(Src->getType()->getScalarType());
+ PointerType *DestTy = cast<PointerType>(CI.getType()->getScalarType());
+
+ Type *DestElemTy = DestTy->getElementType();
+ if (SrcTy->getElementType() != DestElemTy) {
+ Type *MidTy = PointerType::get(DestElemTy, SrcTy->getAddressSpace());
+ if (VectorType *VT = dyn_cast<VectorType>(CI.getType())) {
+ // Handle vectors of pointers.
+ MidTy = VectorType::get(MidTy, VT->getNumElements());
+ }
+
+ Value *NewBitCast = Builder->CreateBitCast(Src, MidTy);
+ return new AddrSpaceCastInst(NewBitCast, CI.getType());
+ }
+
return commonPointerCastTransforms(CI);
}