#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
+#include "llvm/IR/GetElementPtrTypeIterator.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/Operator.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/FEnv.h"
-#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetLibraryInfo.h"
#include <cerrno>
APInt &Offset, const DataLayout &TD) {
// Trivial case, constant is the global.
if ((GV = dyn_cast<GlobalValue>(C))) {
- Offset.clearAllBits();
+ unsigned BitWidth = TD.getPointerTypeSizeInBits(GV->getType());
+ Offset = APInt(BitWidth, 0);
return true;
}
return IsConstantOffsetFromGlobal(CE->getOperand(0), GV, Offset, TD);
// i32* getelementptr ([5 x i32]* @a, i32 0, i32 5)
- if (GEPOperator *GEP = dyn_cast<GEPOperator>(CE)) {
- // If the base isn't a global+constant, we aren't either.
- if (!IsConstantOffsetFromGlobal(CE->getOperand(0), GV, Offset, TD))
- return false;
+ GEPOperator *GEP = dyn_cast<GEPOperator>(CE);
+ if (!GEP)
+ return false;
- // Otherwise, add any offset that our operands provide.
- return GEP->accumulateConstantOffset(TD, Offset);
- }
+ unsigned BitWidth = TD.getPointerTypeSizeInBits(GEP->getType());
+ APInt TmpOffset(BitWidth, 0);
- return false;
+ // If the base isn't a global+constant, we aren't either.
+ if (!IsConstantOffsetFromGlobal(CE->getOperand(0), GV, TmpOffset, TD))
+ return false;
+
+ // Otherwise, add any offset that our operands provide.
+ if (!GEP->accumulateConstantOffset(TD, TmpOffset))
+ return false;
+
+ Offset = TmpOffset;
+ return true;
}
/// ReadDataFromGlobal - Recursive helper to read bits out of global. C is the
// If we read all of the bytes we needed from this element we're done.
uint64_t NextEltOffset = SL->getElementOffset(Index);
- if (BytesLeft <= NextEltOffset-CurEltOffset-ByteOffset)
+ if (BytesLeft <= NextEltOffset - CurEltOffset - ByteOffset)
return true;
// Move to the next element of the struct.
- CurPtr += NextEltOffset-CurEltOffset-ByteOffset;
- BytesLeft -= NextEltOffset-CurEltOffset-ByteOffset;
+ CurPtr += NextEltOffset - CurEltOffset - ByteOffset;
+ BytesLeft -= NextEltOffset - CurEltOffset - ByteOffset;
ByteOffset = 0;
CurEltOffset = NextEltOffset;
}
if (isa<ConstantArray>(C) || isa<ConstantVector>(C) ||
isa<ConstantDataSequential>(C)) {
- Type *EltTy = cast<SequentialType>(C->getType())->getElementType();
+ Type *EltTy = C->getType()->getSequentialElementType();
uint64_t EltSize = TD.getTypeAllocSize(EltTy);
uint64_t Index = ByteOffset / EltSize;
uint64_t Offset = ByteOffset - Index * EltSize;
if (ArrayType *AT = dyn_cast<ArrayType>(C->getType()))
NumElts = AT->getNumElements();
else
- NumElts = cast<VectorType>(C->getType())->getNumElements();
+ NumElts = C->getType()->getVectorNumElements();
for (; Index != NumElts; ++Index) {
if (!ReadDataFromGlobal(C->getAggregateElement(Index), Offset, CurPtr,
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
if (CE->getOpcode() == Instruction::IntToPtr &&
- CE->getOperand(0)->getType() == TD.getIntPtrType(CE->getContext())) {
+ CE->getOperand(0)->getType() == TD.getIntPtrType(CE->getType())) {
return ReadDataFromGlobal(CE->getOperand(0), ByteOffset, CurPtr,
BytesLeft, TD);
}
static Constant *FoldReinterpretLoadFromConstPtr(Constant *C,
const DataLayout &TD) {
- Type *LoadTy = cast<PointerType>(C->getType())->getElementType();
+ PointerType *PTy = cast<PointerType>(C->getType());
+ Type *LoadTy = PTy->getElementType();
IntegerType *IntType = dyn_cast<IntegerType>(LoadTy);
// If this isn't an integer load we can't fold it directly.
if (!IntType) {
+ unsigned AS = PTy->getAddressSpace();
+
// If this is a float/double load, we can try folding it as an int32/64 load
// and then bitcast the result. This can be useful for union cases. Note
// that address spaces don't matter here since we're not going to result in
// an actual new load.
Type *MapTy;
if (LoadTy->isHalfTy())
- MapTy = Type::getInt16PtrTy(C->getContext());
+ MapTy = Type::getInt16PtrTy(C->getContext(), AS);
else if (LoadTy->isFloatTy())
- MapTy = Type::getInt32PtrTy(C->getContext());
+ MapTy = Type::getInt32PtrTy(C->getContext(), AS);
else if (LoadTy->isDoubleTy())
- MapTy = Type::getInt64PtrTy(C->getContext());
+ MapTy = Type::getInt64PtrTy(C->getContext(), AS);
else if (LoadTy->isVectorTy()) {
- MapTy = IntegerType::get(C->getContext(),
- TD.getTypeAllocSizeInBits(LoadTy));
- MapTy = PointerType::getUnqual(MapTy);
+ MapTy = PointerType::getIntNPtrTy(C->getContext(),
+ TD.getTypeAllocSizeInBits(LoadTy),
+ AS);
} else
return 0;
}
unsigned BytesLoaded = (IntType->getBitWidth() + 7) / 8;
- if (BytesLoaded > 32 || BytesLoaded == 0) return 0;
+ if (BytesLoaded > 32 || BytesLoaded == 0)
+ return 0;
GlobalValue *GVal;
- APInt Offset(TD.getPointerSizeInBits(), 0);
+ APInt Offset;
if (!IsConstantOffsetFromGlobal(C, GVal, Offset, TD))
return 0;
// If we're loading off the beginning of the global, some bytes may be valid,
// but we don't try to handle this.
- if (Offset.isNegative()) return 0;
+ if (Offset.isNegative())
+ return 0;
// If we're not accessing anything in this constant, the result is undefined.
if (Offset.getZExtValue() >=
// constant. This happens frequently when iterating over a global array.
if (Opc == Instruction::Sub && DL) {
GlobalValue *GV1, *GV2;
- unsigned PtrSize = DL->getPointerSizeInBits();
- unsigned OpSize = DL->getTypeSizeInBits(Op0->getType());
- APInt Offs1(PtrSize, 0), Offs2(PtrSize, 0);
+ APInt Offs1, Offs2;
if (IsConstantOffsetFromGlobal(Op0, GV1, Offs1, *DL))
if (IsConstantOffsetFromGlobal(Op1, GV2, Offs2, *DL) &&
GV1 == GV2) {
+ unsigned OpSize = DL->getTypeSizeInBits(Op0->getType());
+
// (&GV+C1) - (&GV+C2) -> C1-C2, pointer arithmetic cannot overflow.
// PtrToInt may change the bitwidth so we have convert to the right size
// first.
static Constant *CastGEPIndices(ArrayRef<Constant *> Ops,
Type *ResultTy, const DataLayout *TD,
const TargetLibraryInfo *TLI) {
- if (!TD) return 0;
- Type *IntPtrTy = TD->getIntPtrType(ResultTy->getContext());
+ if (!TD)
+ return 0;
+
+ Type *IntPtrTy = TD->getIntPtrType(ResultTy);
bool Any = false;
SmallVector<Constant*, 32> NewIdxs;
if (NewPtrTy->getAddressSpace() != OldPtrTy->getAddressSpace()) {
NewPtrTy = NewPtrTy->getElementType()->getPointerTo(
OldPtrTy->getAddressSpace());
- Ptr = ConstantExpr::getBitCast(Ptr, NewPtrTy);
+ Ptr = ConstantExpr::getPointerCast(Ptr, NewPtrTy);
}
return Ptr;
}
Type *ResultTy, const DataLayout *TD,
const TargetLibraryInfo *TLI) {
Constant *Ptr = Ops[0];
- if (!TD || !cast<PointerType>(Ptr->getType())->getElementType()->isSized() ||
+ if (!TD || !Ptr->getType()->getPointerElementType()->isSized() ||
!Ptr->getType()->isPointerTy())
return 0;
- Type *IntPtrTy = TD->getIntPtrType(Ptr->getContext());
+ Type *IntPtrTy = TD->getIntPtrType(Ptr->getType());
+ Type *ResultElementTy = ResultTy->getPointerElementType();
// If this is a constant expr gep that is effectively computing an
// "offsetof", fold it into 'cast int Size to T*' instead of 'gep 0, 0, 12'
// If this is "gep i8* Ptr, (sub 0, V)", fold this as:
// "inttoptr (sub (ptrtoint Ptr), V)"
- if (Ops.size() == 2 &&
- cast<PointerType>(ResultTy)->getElementType()->isIntegerTy(8)) {
+ if (Ops.size() == 2 && ResultElementTy->isIntegerTy(8)) {
ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[1]);
assert((CE == 0 || CE->getType() == IntPtrTy) &&
"CastGEPIndices didn't canonicalize index types!");
// Also, this helps GlobalOpt do SROA on GlobalVariables.
Type *Ty = Ptr->getType();
assert(Ty->isPointerTy() && "Forming regular GEP of non-pointer type");
- SmallVector<Constant*, 32> NewIdxs;
+ SmallVector<Constant *, 32> NewIdxs;
+
do {
if (SequentialType *ATy = dyn_cast<SequentialType>(Ty)) {
if (ATy->isPointerTy()) {
// Determine which element of the array the offset points into.
APInt ElemSize(BitWidth, TD->getTypeAllocSize(ATy->getElementType()));
- IntegerType *IntPtrTy = TD->getIntPtrType(Ty->getContext());
if (ElemSize == 0)
// The element size is 0. This may be [0 x Ty]*, so just use a zero
// index for this level and proceed to the next level to see if it can
// We've reached some non-indexable type.
break;
}
- } while (Ty != cast<PointerType>(ResultTy)->getElementType());
+ } while (Ty != ResultElementTy);
// If we haven't used up the entire offset by descending the static
// type, then the offset is pointing into the middle of an indivisible
// Create a GEP.
Constant *C = ConstantExpr::getGetElementPtr(Ptr, NewIdxs);
- assert(cast<PointerType>(C->getType())->getElementType() == Ty &&
+ assert(C->getType()->getPointerElementType() == Ty &&
"Computed GetElementPtr has unexpected type!");
// If we ended up indexing a member with a type that doesn't match
// the type of what the original indices indexed, add a cast.
- if (Ty != cast<PointerType>(ResultTy)->getElementType())
+ if (Ty != ResultElementTy)
C = FoldBitCast(C, ResultTy, *TD);
return C;
if (TD && CE->getOpcode() == Instruction::IntToPtr) {
Constant *Input = CE->getOperand(0);
unsigned InWidth = Input->getType()->getScalarSizeInBits();
- if (TD->getPointerSizeInBits() < InWidth) {
+ unsigned PtrWidth = TD->getPointerTypeSizeInBits(CE->getType());
+ if (PtrWidth < InWidth) {
Constant *Mask =
- ConstantInt::get(CE->getContext(), APInt::getLowBitsSet(InWidth,
- TD->getPointerSizeInBits()));
+ ConstantInt::get(CE->getContext(),
+ APInt::getLowBitsSet(InWidth, PtrWidth));
Input = ConstantExpr::getAnd(Input, Mask);
}
// Do a zext or trunc to get to the dest size.
return ConstantExpr::getCast(Opcode, Ops[0], DestTy);
case Instruction::IntToPtr:
// If the input is a ptrtoint, turn the pair into a ptr to ptr bitcast if
- // the int size is >= the ptr size. This requires knowing the width of a
- // pointer, so it can't be done in ConstantExpr::getCast.
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[0]))
- if (TD &&
- TD->getPointerSizeInBits() <= CE->getType()->getScalarSizeInBits() &&
- CE->getOpcode() == Instruction::PtrToInt)
- return FoldBitCast(CE->getOperand(0), DestTy, *TD);
+ // the int size is >= the ptr size and the address spaces are the same.
+ // This requires knowing the width of a pointer, so it can't be done in
+ // ConstantExpr::getCast.
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[0])) {
+ if (TD && CE->getOpcode() == Instruction::PtrToInt) {
+ Constant *SrcPtr = CE->getOperand(0);
+ unsigned SrcPtrSize = TD->getPointerTypeSizeInBits(SrcPtr->getType());
+ unsigned MidIntSize = CE->getType()->getScalarSizeInBits();
+
+ if (MidIntSize >= SrcPtrSize) {
+ unsigned SrcAS = SrcPtr->getType()->getPointerAddressSpace();
+ if (SrcAS == DestTy->getPointerAddressSpace())
+ return FoldBitCast(CE->getOperand(0), DestTy, *TD);
+ }
+ }
+ }
return ConstantExpr::getCast(Opcode, Ops[0], DestTy);
case Instruction::Trunc:
case Instruction::SIToFP:
case Instruction::FPToUI:
case Instruction::FPToSI:
+ case Instruction::AddrSpaceCast:
return ConstantExpr::getCast(Opcode, Ops[0], DestTy);
case Instruction::BitCast:
if (TD)
// around to know if bit truncation is happening.
if (ConstantExpr *CE0 = dyn_cast<ConstantExpr>(Ops0)) {
if (TD && Ops1->isNullValue()) {
- Type *IntPtrTy = TD->getIntPtrType(CE0->getContext());
if (CE0->getOpcode() == Instruction::IntToPtr) {
+ Type *IntPtrTy = TD->getIntPtrType(CE0->getType());
// Convert the integer value to the right size to ensure we get the
// proper extension or truncation.
Constant *C = ConstantExpr::getIntegerCast(CE0->getOperand(0),
// Only do this transformation if the int is intptrty in size, otherwise
// there is a truncation or extension that we aren't modeling.
- if (CE0->getOpcode() == Instruction::PtrToInt &&
- CE0->getType() == IntPtrTy) {
- Constant *C = CE0->getOperand(0);
- Constant *Null = Constant::getNullValue(C->getType());
- return ConstantFoldCompareInstOperands(Predicate, C, Null, TD, TLI);
+ if (CE0->getOpcode() == Instruction::PtrToInt) {
+ Type *IntPtrTy = TD->getIntPtrType(CE0->getOperand(0)->getType());
+ if (CE0->getType() == IntPtrTy) {
+ Constant *C = CE0->getOperand(0);
+ Constant *Null = Constant::getNullValue(C->getType());
+ return ConstantFoldCompareInstOperands(Predicate, C, Null, TD, TLI);
+ }
}
}
if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(Ops1)) {
if (TD && CE0->getOpcode() == CE1->getOpcode()) {
- Type *IntPtrTy = TD->getIntPtrType(CE0->getContext());
-
if (CE0->getOpcode() == Instruction::IntToPtr) {
+ Type *IntPtrTy = TD->getIntPtrType(CE0->getType());
+
// Convert the integer value to the right size to ensure we get the
// proper extension or truncation.
Constant *C0 = ConstantExpr::getIntegerCast(CE0->getOperand(0),
// Only do this transformation if the int is intptrty in size, otherwise
// there is a truncation or extension that we aren't modeling.
- if ((CE0->getOpcode() == Instruction::PtrToInt &&
- CE0->getType() == IntPtrTy &&
- CE0->getOperand(0)->getType() == CE1->getOperand(0)->getType()))
- return ConstantFoldCompareInstOperands(Predicate, CE0->getOperand(0),
- CE1->getOperand(0), TD, TLI);
+ if (CE0->getOpcode() == Instruction::PtrToInt) {
+ Type *IntPtrTy = TD->getIntPtrType(CE0->getOperand(0)->getType());
+ if (CE0->getType() == IntPtrTy &&
+ CE0->getOperand(0)->getType() == CE1->getOperand(0)->getType()) {
+ return ConstantFoldCompareInstOperands(Predicate,
+ CE0->getOperand(0),
+ CE1->getOperand(0),
+ TD,
+ TLI);
+ }
+ }
}
}
case Intrinsic::ctpop:
case Intrinsic::ctlz:
case Intrinsic::cttz:
+ case Intrinsic::fma:
+ case Intrinsic::fmuladd:
+ case Intrinsic::copysign:
+ case Intrinsic::round:
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
}
}
-static Constant *ConstantFoldFP(double (*NativeFP)(double), double V,
- Type *Ty) {
- sys::llvm_fenv_clearexcept();
- V = NativeFP(V);
- if (sys::llvm_fenv_testexcept()) {
- sys::llvm_fenv_clearexcept();
- return 0;
- }
-
+static Constant *GetConstantFoldFPValue(double V, Type *Ty) {
if (Ty->isHalfTy()) {
APFloat APF(V);
bool unused;
if (Ty->isDoubleTy())
return ConstantFP::get(Ty->getContext(), APFloat(V));
llvm_unreachable("Can only constant fold half/float/double");
+
+}
+
+static Constant *ConstantFoldFP(double (*NativeFP)(double), double V,
+ Type *Ty) {
+ sys::llvm_fenv_clearexcept();
+ V = NativeFP(V);
+ if (sys::llvm_fenv_testexcept()) {
+ sys::llvm_fenv_clearexcept();
+ return 0;
+ }
+
+ return GetConstantFoldFPValue(V, Ty);
}
static Constant *ConstantFoldBinaryFP(double (*NativeFP)(double, double),
return 0;
}
- if (Ty->isHalfTy()) {
- APFloat APF(V);
- bool unused;
- APF.convert(APFloat::IEEEhalf, APFloat::rmNearestTiesToEven, &unused);
- return ConstantFP::get(Ty->getContext(), APF);
- }
- if (Ty->isFloatTy())
- return ConstantFP::get(Ty->getContext(), APFloat((float)V));
- if (Ty->isDoubleTy())
- return ConstantFP::get(Ty->getContext(), APFloat(V));
- llvm_unreachable("Can only constant fold half/float/double");
+ return GetConstantFoldFPValue(V, Ty);
}
/// ConstantFoldConvertToInt - Attempt to an SSE floating point to integer
static Constant *ConstantFoldConvertToInt(const APFloat &Val,
bool roundTowardZero, Type *Ty) {
// All of these conversion intrinsics form an integer of at most 64bits.
- unsigned ResultWidth = cast<IntegerType>(Ty)->getBitWidth();
+ unsigned ResultWidth = Ty->getIntegerBitWidth();
assert(ResultWidth <= 64 &&
"Can only constant fold conversions to 64 and 32 bit ints");
return ConstantInt::get(Ty, UIntVal, /*isSigned=*/true);
}
-/// ConstantFoldCall - Attempt to constant fold a call to the specified function
-/// with the specified arguments, returning null if unsuccessful.
-Constant *
-llvm::ConstantFoldCall(Function *F, ArrayRef<Constant *> Operands,
- const TargetLibraryInfo *TLI) {
- if (!F->hasName())
- return 0;
- StringRef Name = F->getName();
+static double getValueAsDouble(ConstantFP *Op) {
+ Type *Ty = Op->getType();
- Type *Ty = F->getReturnType();
+ if (Ty->isFloatTy())
+ return Op->getValueAPF().convertToFloat();
+
+ if (Ty->isDoubleTy())
+ return Op->getValueAPF().convertToDouble();
+
+ bool unused;
+ APFloat APF = Op->getValueAPF();
+ APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &unused);
+ return APF.convertToDouble();
+}
+
+static Constant *ConstantFoldScalarCall(StringRef Name, unsigned IntrinsicID,
+ Type *Ty, ArrayRef<Constant *> Operands,
+ const TargetLibraryInfo *TLI) {
if (Operands.size() == 1) {
if (ConstantFP *Op = dyn_cast<ConstantFP>(Operands[0])) {
- if (F->getIntrinsicID() == Intrinsic::convert_to_fp16) {
+ if (IntrinsicID == Intrinsic::convert_to_fp16) {
APFloat Val(Op->getValueAPF());
bool lost = false;
Val.convert(APFloat::IEEEhalf, APFloat::rmNearestTiesToEven, &lost);
- return ConstantInt::get(F->getContext(), Val.bitcastToAPInt());
+ return ConstantInt::get(Ty->getContext(), Val.bitcastToAPInt());
}
- if (!TLI)
- return 0;
if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy())
return 0;
+ if (IntrinsicID == Intrinsic::round) {
+ APFloat V = Op->getValueAPF();
+ V.roundToIntegral(APFloat::rmNearestTiesToAway);
+ return ConstantFP::get(Ty->getContext(), V);
+ }
+
/// We only fold functions with finite arguments. Folding NaN and inf is
/// likely to be aborted with an exception anyway, and some host libms
/// have known errors raising exceptions.
/// the host native double versions. Float versions are not called
/// directly but for all these it is true (float)(f((double)arg)) ==
/// f(arg). Long double not supported yet.
- double V;
- if (Ty->isFloatTy())
- V = Op->getValueAPF().convertToFloat();
- else if (Ty->isDoubleTy())
- V = Op->getValueAPF().convertToDouble();
- else {
- bool unused;
- APFloat APF = Op->getValueAPF();
- APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &unused);
- V = APF.convertToDouble();
- }
+ double V = getValueAsDouble(Op);
- switch (F->getIntrinsicID()) {
+ switch (IntrinsicID) {
default: break;
case Intrinsic::fabs:
return ConstantFoldFP(fabs, V, Ty);
return ConstantFoldFP(floor, V, Ty);
}
+ if (!TLI)
+ return 0;
+
switch (Name[0]) {
case 'a':
if (Name == "acos" && TLI->has(LibFunc::acos))
return ConstantFoldFP(log, V, Ty);
else if (Name == "log10" && V > 0 && TLI->has(LibFunc::log10))
return ConstantFoldFP(log10, V, Ty);
- else if (F->getIntrinsicID() == Intrinsic::sqrt &&
+ else if (IntrinsicID == Intrinsic::sqrt &&
(Ty->isHalfTy() || Ty->isFloatTy() || Ty->isDoubleTy())) {
if (V >= -0.0)
return ConstantFoldFP(sqrt, V, Ty);
}
if (ConstantInt *Op = dyn_cast<ConstantInt>(Operands[0])) {
- switch (F->getIntrinsicID()) {
+ switch (IntrinsicID) {
case Intrinsic::bswap:
- return ConstantInt::get(F->getContext(), Op->getValue().byteSwap());
+ return ConstantInt::get(Ty->getContext(), Op->getValue().byteSwap());
case Intrinsic::ctpop:
return ConstantInt::get(Ty, Op->getValue().countPopulation());
case Intrinsic::convert_from_fp16: {
assert(status == APFloat::opOK && !lost &&
"Precision lost during fp16 constfolding");
- return ConstantFP::get(F->getContext(), Val);
+ return ConstantFP::get(Ty->getContext(), Val);
}
default:
return 0;
if (isa<ConstantVector>(Operands[0]) ||
isa<ConstantDataVector>(Operands[0])) {
Constant *Op = cast<Constant>(Operands[0]);
- switch (F->getIntrinsicID()) {
+ switch (IntrinsicID) {
default: break;
case Intrinsic::x86_sse_cvtss2si:
case Intrinsic::x86_sse_cvtss2si64:
}
if (isa<UndefValue>(Operands[0])) {
- if (F->getIntrinsicID() == Intrinsic::bswap)
+ if (IntrinsicID == Intrinsic::bswap)
return Operands[0];
return 0;
}
if (ConstantFP *Op1 = dyn_cast<ConstantFP>(Operands[0])) {
if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy())
return 0;
- double Op1V;
- if (Ty->isFloatTy())
- Op1V = Op1->getValueAPF().convertToFloat();
- else if (Ty->isDoubleTy())
- Op1V = Op1->getValueAPF().convertToDouble();
- else {
- bool unused;
- APFloat APF = Op1->getValueAPF();
- APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &unused);
- Op1V = APF.convertToDouble();
- }
+ double Op1V = getValueAsDouble(Op1);
if (ConstantFP *Op2 = dyn_cast<ConstantFP>(Operands[1])) {
if (Op2->getType() != Op1->getType())
return 0;
- double Op2V;
- if (Ty->isFloatTy())
- Op2V = Op2->getValueAPF().convertToFloat();
- else if (Ty->isDoubleTy())
- Op2V = Op2->getValueAPF().convertToDouble();
- else {
- bool unused;
- APFloat APF = Op2->getValueAPF();
- APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &unused);
- Op2V = APF.convertToDouble();
- }
-
- if (F->getIntrinsicID() == Intrinsic::pow) {
+ double Op2V = getValueAsDouble(Op2);
+ if (IntrinsicID == Intrinsic::pow) {
return ConstantFoldBinaryFP(pow, Op1V, Op2V, Ty);
}
+ if (IntrinsicID == Intrinsic::copysign) {
+ APFloat V1 = Op1->getValueAPF();
+ APFloat V2 = Op2->getValueAPF();
+ V1.copySign(V2);
+ return ConstantFP::get(Ty->getContext(), V1);
+ }
if (!TLI)
return 0;
if (Name == "pow" && TLI->has(LibFunc::pow))
if (Name == "atan2" && TLI->has(LibFunc::atan2))
return ConstantFoldBinaryFP(atan2, Op1V, Op2V, Ty);
} else if (ConstantInt *Op2C = dyn_cast<ConstantInt>(Operands[1])) {
- if (F->getIntrinsicID() == Intrinsic::powi && Ty->isHalfTy())
- return ConstantFP::get(F->getContext(),
+ if (IntrinsicID == Intrinsic::powi && Ty->isHalfTy())
+ return ConstantFP::get(Ty->getContext(),
APFloat((float)std::pow((float)Op1V,
(int)Op2C->getZExtValue())));
- if (F->getIntrinsicID() == Intrinsic::powi && Ty->isFloatTy())
- return ConstantFP::get(F->getContext(),
+ if (IntrinsicID == Intrinsic::powi && Ty->isFloatTy())
+ return ConstantFP::get(Ty->getContext(),
APFloat((float)std::pow((float)Op1V,
(int)Op2C->getZExtValue())));
- if (F->getIntrinsicID() == Intrinsic::powi && Ty->isDoubleTy())
- return ConstantFP::get(F->getContext(),
+ if (IntrinsicID == Intrinsic::powi && Ty->isDoubleTy())
+ return ConstantFP::get(Ty->getContext(),
APFloat((double)std::pow((double)Op1V,
(int)Op2C->getZExtValue())));
}
if (ConstantInt *Op1 = dyn_cast<ConstantInt>(Operands[0])) {
if (ConstantInt *Op2 = dyn_cast<ConstantInt>(Operands[1])) {
- switch (F->getIntrinsicID()) {
+ switch (IntrinsicID) {
default: break;
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow:
case Intrinsic::umul_with_overflow: {
APInt Res;
bool Overflow;
- switch (F->getIntrinsicID()) {
+ switch (IntrinsicID) {
default: llvm_unreachable("Invalid case");
case Intrinsic::sadd_with_overflow:
Res = Op1->getValue().sadd_ov(Op2->getValue(), Overflow);
break;
}
Constant *Ops[] = {
- ConstantInt::get(F->getContext(), Res),
- ConstantInt::get(Type::getInt1Ty(F->getContext()), Overflow)
+ ConstantInt::get(Ty->getContext(), Res),
+ ConstantInt::get(Type::getInt1Ty(Ty->getContext()), Overflow)
};
- return ConstantStruct::get(cast<StructType>(F->getReturnType()), Ops);
+ return ConstantStruct::get(cast<StructType>(Ty), Ops);
}
case Intrinsic::cttz:
if (Op2->isOne() && Op1->isZero()) // cttz(0, 1) is undef.
}
return 0;
}
+
+ if (Operands.size() != 3)
+ return 0;
+
+ if (const ConstantFP *Op1 = dyn_cast<ConstantFP>(Operands[0])) {
+ if (const ConstantFP *Op2 = dyn_cast<ConstantFP>(Operands[1])) {
+ if (const ConstantFP *Op3 = dyn_cast<ConstantFP>(Operands[2])) {
+ switch (IntrinsicID) {
+ default: break;
+ case Intrinsic::fma:
+ case Intrinsic::fmuladd: {
+ APFloat V = Op1->getValueAPF();
+ APFloat::opStatus s = V.fusedMultiplyAdd(Op2->getValueAPF(),
+ Op3->getValueAPF(),
+ APFloat::rmNearestTiesToEven);
+ if (s != APFloat::opInvalidOp)
+ return ConstantFP::get(Ty->getContext(), V);
+
+ return 0;
+ }
+ }
+ }
+ }
+ }
+
return 0;
}
+
+static Constant *ConstantFoldVectorCall(StringRef Name, unsigned IntrinsicID,
+ VectorType *VTy,
+ ArrayRef<Constant *> Operands,
+ const TargetLibraryInfo *TLI) {
+ SmallVector<Constant *, 4> Result(VTy->getNumElements());
+ SmallVector<Constant *, 4> Lane(Operands.size());
+ Type *Ty = VTy->getElementType();
+
+ for (unsigned I = 0, E = VTy->getNumElements(); I != E; ++I) {
+ // Gather a column of constants.
+ for (unsigned J = 0, JE = Operands.size(); J != JE; ++J) {
+ Constant *Agg = Operands[J]->getAggregateElement(I);
+ if (!Agg)
+ return nullptr;
+
+ Lane[J] = Agg;
+ }
+
+ // Use the regular scalar folding to simplify this column.
+ Constant *Folded = ConstantFoldScalarCall(Name, IntrinsicID, Ty, Lane, TLI);
+ if (!Folded)
+ return nullptr;
+ Result[I] = Folded;
+ }
+
+ return ConstantVector::get(Result);
+}
+
+/// ConstantFoldCall - Attempt to constant fold a call to the specified function
+/// with the specified arguments, returning null if unsuccessful.
+Constant *
+llvm::ConstantFoldCall(Function *F, ArrayRef<Constant *> Operands,
+ const TargetLibraryInfo *TLI) {
+ if (!F->hasName())
+ return 0;
+ StringRef Name = F->getName();
+
+ Type *Ty = F->getReturnType();
+
+ if (VectorType *VTy = dyn_cast<VectorType>(Ty))
+ return ConstantFoldVectorCall(Name, F->getIntrinsicID(), VTy, Operands, TLI);
+
+ return ConstantFoldScalarCall(Name, F->getIntrinsicID(), Ty, Operands, TLI);
+}