X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FAnalysis%2FConstantFolding.cpp;h=07f53bedc7766610c1463397a8ed28726e1dfcd2;hp=4bf172758c9eef78cfe4a4f59852302b6d9013fe;hb=609d95290a4d913139ba0b46f5b5b982dc4b1d96;hpb=6b543713a25c20c028cc0bbca0dd8b052c61e000 diff --git a/lib/Analysis/ConstantFolding.cpp b/lib/Analysis/ConstantFolding.cpp index 4bf172758c9..07f53bedc77 100644 --- a/lib/Analysis/ConstantFolding.cpp +++ b/lib/Analysis/ConstantFolding.cpp @@ -20,7 +20,9 @@ #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringMap.h" +#include "llvm/Analysis/TargetLibraryInfo.h" #include "llvm/Analysis/ValueTracking.h" +#include "llvm/Config/config.h" #include "llvm/IR/Constants.h" #include "llvm/IR/DataLayout.h" #include "llvm/IR/DerivedTypes.h" @@ -31,26 +33,29 @@ #include "llvm/IR/Intrinsics.h" #include "llvm/IR/Operator.h" #include "llvm/Support/ErrorHandling.h" -#include "llvm/Support/FEnv.h" #include "llvm/Support/MathExtras.h" -#include "llvm/Target/TargetLibraryInfo.h" #include #include + +#ifdef HAVE_FENV_H +#include +#endif + using namespace llvm; //===----------------------------------------------------------------------===// // Constant Folding internal helper functions //===----------------------------------------------------------------------===// -/// FoldBitCast - Constant fold bitcast, symbolically evaluating it with -/// DataLayout. This always returns a non-null constant, but it may be a +/// Constant fold bitcast, symbolically evaluating it with DataLayout. +/// This always returns a non-null constant, but it may be a /// ConstantExpr if unfoldable. -static Constant *FoldBitCast(Constant *C, Type *DestTy, - const DataLayout &TD) { +static Constant *FoldBitCast(Constant *C, Type *DestTy, const DataLayout &DL) { // Catch the obvious splat cases. if (C->isNullValue() && !DestTy->isX86_MMXTy()) return Constant::getNullValue(DestTy); - if (C->isAllOnesValue() && !DestTy->isX86_MMXTy()) + if (C->isAllOnesValue() && !DestTy->isX86_MMXTy() && + !DestTy->isPtrOrPtrVectorTy()) // Don't get ones for ptr types! return Constant::getAllOnesValue(DestTy); // Handle a vector->integer cast. @@ -78,11 +83,11 @@ static Constant *FoldBitCast(Constant *C, Type *DestTy, // Now that we know that the input value is a vector of integers, just shift // and insert them into our result. - unsigned BitShift = TD.getTypeAllocSizeInBits(SrcEltTy); + unsigned BitShift = DL.getTypeAllocSizeInBits(SrcEltTy); APInt Result(IT->getBitWidth(), 0); for (unsigned i = 0; i != NumSrcElts; ++i) { Result <<= BitShift; - if (TD.isLittleEndian()) + if (DL.isLittleEndian()) Result |= CDV->getElementAsInteger(NumSrcElts-i-1); else Result |= CDV->getElementAsInteger(i); @@ -100,7 +105,7 @@ static Constant *FoldBitCast(Constant *C, Type *DestTy, // vector so the code below can handle it uniformly. if (isa(C) || isa(C)) { Constant *Ops = C; // don't take the address of C! - return FoldBitCast(ConstantVector::get(Ops), DestTy, TD); + return FoldBitCast(ConstantVector::get(Ops), DestTy, DL); } // If this is a bitcast from constant vector -> vector, fold it. @@ -132,7 +137,7 @@ static Constant *FoldBitCast(Constant *C, Type *DestTy, Type *DestIVTy = VectorType::get(IntegerType::get(C->getContext(), FPWidth), NumDstElt); // Recursively handle this integer conversion, if possible. - C = FoldBitCast(C, DestIVTy, TD); + C = FoldBitCast(C, DestIVTy, DL); // Finally, IR can handle this now that #elts line up. return ConstantExpr::getBitCast(C, DestTy); @@ -156,7 +161,7 @@ static Constant *FoldBitCast(Constant *C, Type *DestTy, // of the same size, and that their #elements is not the same. Do the // conversion here, which depends on whether the input or output has // more elements. - bool isLittleEndian = TD.isLittleEndian(); + bool isLittleEndian = DL.isLittleEndian(); SmallVector Result; if (NumDstElt < NumSrcElt) { @@ -192,7 +197,7 @@ static Constant *FoldBitCast(Constant *C, Type *DestTy, // Handle: bitcast (<2 x i64> to <4 x i32>) unsigned Ratio = NumDstElt/NumSrcElt; - unsigned DstBitSize = DstEltTy->getPrimitiveSizeInBits(); + unsigned DstBitSize = DL.getTypeSizeInBits(DstEltTy); // Loop over each source value, expanding into multiple results. for (unsigned i = 0; i != NumSrcElt; ++i) { @@ -208,6 +213,15 @@ static Constant *FoldBitCast(Constant *C, Type *DestTy, ConstantInt::get(Src->getType(), ShiftAmt)); ShiftAmt += isLittleEndian ? DstBitSize : -DstBitSize; + // Truncate the element to an integer with the same pointer size and + // convert the element back to a pointer using a inttoptr. + if (DstEltTy->isPointerTy()) { + IntegerType *DstIntTy = Type::getIntNTy(C->getContext(), DstBitSize); + Constant *CE = ConstantExpr::getTrunc(Elt, DstIntTy); + Result.push_back(ConstantExpr::getIntToPtr(CE, DstEltTy)); + continue; + } + // Truncate and remember this piece. Result.push_back(ConstantExpr::getTrunc(Elt, DstEltTy)); } @@ -217,14 +231,13 @@ static Constant *FoldBitCast(Constant *C, Type *DestTy, } -/// IsConstantOffsetFromGlobal - If this constant is actually a constant offset -/// from a global, return the global and the constant. Because of -/// constantexprs, this function is recursive. +/// If this constant is a constant offset from a global, return the global and +/// the constant. Because of constantexprs, this function is recursive. static bool IsConstantOffsetFromGlobal(Constant *C, GlobalValue *&GV, - APInt &Offset, const DataLayout &TD) { + APInt &Offset, const DataLayout &DL) { // Trivial case, constant is the global. if ((GV = dyn_cast(C))) { - unsigned BitWidth = TD.getPointerTypeSizeInBits(GV->getType()); + unsigned BitWidth = DL.getPointerTypeSizeInBits(GV->getType()); Offset = APInt(BitWidth, 0); return true; } @@ -236,36 +249,36 @@ static bool IsConstantOffsetFromGlobal(Constant *C, GlobalValue *&GV, // Look through ptr->int and ptr->ptr casts. if (CE->getOpcode() == Instruction::PtrToInt || CE->getOpcode() == Instruction::BitCast) - return IsConstantOffsetFromGlobal(CE->getOperand(0), GV, Offset, TD); + return IsConstantOffsetFromGlobal(CE->getOperand(0), GV, Offset, DL); // i32* getelementptr ([5 x i32]* @a, i32 0, i32 5) GEPOperator *GEP = dyn_cast(CE); if (!GEP) return false; - unsigned BitWidth = TD.getPointerTypeSizeInBits(GEP->getType()); + unsigned BitWidth = DL.getPointerTypeSizeInBits(GEP->getType()); APInt TmpOffset(BitWidth, 0); // If the base isn't a global+constant, we aren't either. - if (!IsConstantOffsetFromGlobal(CE->getOperand(0), GV, TmpOffset, TD)) + if (!IsConstantOffsetFromGlobal(CE->getOperand(0), GV, TmpOffset, DL)) return false; // Otherwise, add any offset that our operands provide. - if (!GEP->accumulateConstantOffset(TD, TmpOffset)) + if (!GEP->accumulateConstantOffset(DL, TmpOffset)) return false; Offset = TmpOffset; return true; } -/// ReadDataFromGlobal - Recursive helper to read bits out of global. C is the -/// constant being copied out of. ByteOffset is an offset into C. CurPtr is the -/// pointer to copy results into and BytesLeft is the number of bytes left in -/// the CurPtr buffer. TD is the target data. +/// Recursive helper to read bits out of global. C is the constant being copied +/// out of. ByteOffset is an offset into C. CurPtr is the pointer to copy +/// results into and BytesLeft is the number of bytes left in +/// the CurPtr buffer. DL is the DataLayout. static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, unsigned char *CurPtr, unsigned BytesLeft, - const DataLayout &TD) { - assert(ByteOffset <= TD.getTypeAllocSize(C->getType()) && + const DataLayout &DL) { + assert(ByteOffset <= DL.getTypeAllocSize(C->getType()) && "Out of range access"); // If this element is zero or undefined, we can just return since *CurPtr is @@ -283,7 +296,7 @@ static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, for (unsigned i = 0; i != BytesLeft && ByteOffset != IntBytes; ++i) { int n = ByteOffset; - if (!TD.isLittleEndian()) + if (!DL.isLittleEndian()) n = IntBytes - n - 1; CurPtr[i] = (unsigned char)(Val >> (n * 8)); ++ByteOffset; @@ -293,22 +306,22 @@ static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, if (ConstantFP *CFP = dyn_cast(C)) { if (CFP->getType()->isDoubleTy()) { - C = FoldBitCast(C, Type::getInt64Ty(C->getContext()), TD); - return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, TD); + C = FoldBitCast(C, Type::getInt64Ty(C->getContext()), DL); + return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, DL); } if (CFP->getType()->isFloatTy()){ - C = FoldBitCast(C, Type::getInt32Ty(C->getContext()), TD); - return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, TD); + C = FoldBitCast(C, Type::getInt32Ty(C->getContext()), DL); + return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, DL); } if (CFP->getType()->isHalfTy()){ - C = FoldBitCast(C, Type::getInt16Ty(C->getContext()), TD); - return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, TD); + C = FoldBitCast(C, Type::getInt16Ty(C->getContext()), DL); + return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, DL); } return false; } if (ConstantStruct *CS = dyn_cast(C)) { - const StructLayout *SL = TD.getStructLayout(CS->getType()); + const StructLayout *SL = DL.getStructLayout(CS->getType()); unsigned Index = SL->getElementContainingOffset(ByteOffset); uint64_t CurEltOffset = SL->getElementOffset(Index); ByteOffset -= CurEltOffset; @@ -316,11 +329,11 @@ static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, while (1) { // If the element access is to the element itself and not to tail padding, // read the bytes from the element. - uint64_t EltSize = TD.getTypeAllocSize(CS->getOperand(Index)->getType()); + uint64_t EltSize = DL.getTypeAllocSize(CS->getOperand(Index)->getType()); if (ByteOffset < EltSize && !ReadDataFromGlobal(CS->getOperand(Index), ByteOffset, CurPtr, - BytesLeft, TD)) + BytesLeft, DL)) return false; ++Index; @@ -347,7 +360,7 @@ static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, if (isa(C) || isa(C) || isa(C)) { Type *EltTy = C->getType()->getSequentialElementType(); - uint64_t EltSize = TD.getTypeAllocSize(EltTy); + uint64_t EltSize = DL.getTypeAllocSize(EltTy); uint64_t Index = ByteOffset / EltSize; uint64_t Offset = ByteOffset - Index * EltSize; uint64_t NumElts; @@ -358,7 +371,7 @@ static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, for (; Index != NumElts; ++Index) { if (!ReadDataFromGlobal(C->getAggregateElement(Index), Offset, CurPtr, - BytesLeft, TD)) + BytesLeft, DL)) return false; uint64_t BytesWritten = EltSize - Offset; @@ -375,9 +388,9 @@ static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, if (ConstantExpr *CE = dyn_cast(C)) { if (CE->getOpcode() == Instruction::IntToPtr && - CE->getOperand(0)->getType() == TD.getIntPtrType(CE->getType())) { + CE->getOperand(0)->getType() == DL.getIntPtrType(CE->getType())) { return ReadDataFromGlobal(CE->getOperand(0), ByteOffset, CurPtr, - BytesLeft, TD); + BytesLeft, DL); } } @@ -386,7 +399,7 @@ static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, } static Constant *FoldReinterpretLoadFromConstPtr(Constant *C, - const DataLayout &TD) { + const DataLayout &DL) { PointerType *PTy = cast(C->getType()); Type *LoadTy = PTy->getElementType(); IntegerType *IntType = dyn_cast(LoadTy); @@ -408,14 +421,13 @@ static Constant *FoldReinterpretLoadFromConstPtr(Constant *C, MapTy = Type::getInt64PtrTy(C->getContext(), AS); else if (LoadTy->isVectorTy()) { MapTy = PointerType::getIntNPtrTy(C->getContext(), - TD.getTypeAllocSizeInBits(LoadTy), - AS); + DL.getTypeAllocSizeInBits(LoadTy), AS); } else return nullptr; - C = FoldBitCast(C, MapTy, TD); - if (Constant *Res = FoldReinterpretLoadFromConstPtr(C, TD)) - return FoldBitCast(Res, LoadTy, TD); + C = FoldBitCast(C, MapTy, DL); + if (Constant *Res = FoldReinterpretLoadFromConstPtr(C, DL)) + return FoldBitCast(Res, LoadTy, DL); return nullptr; } @@ -425,7 +437,7 @@ static Constant *FoldReinterpretLoadFromConstPtr(Constant *C, GlobalValue *GVal; APInt Offset; - if (!IsConstantOffsetFromGlobal(C, GVal, Offset, TD)) + if (!IsConstantOffsetFromGlobal(C, GVal, Offset, DL)) return nullptr; GlobalVariable *GV = dyn_cast(GVal); @@ -440,16 +452,16 @@ static Constant *FoldReinterpretLoadFromConstPtr(Constant *C, // If we're not accessing anything in this constant, the result is undefined. if (Offset.getZExtValue() >= - TD.getTypeAllocSize(GV->getInitializer()->getType())) + DL.getTypeAllocSize(GV->getInitializer()->getType())) return UndefValue::get(IntType); unsigned char RawBytes[32] = {0}; if (!ReadDataFromGlobal(GV->getInitializer(), Offset.getZExtValue(), RawBytes, - BytesLoaded, TD)) + BytesLoaded, DL)) return nullptr; APInt ResultVal = APInt(IntType->getBitWidth(), 0); - if (TD.isLittleEndian()) { + if (DL.isLittleEndian()) { ResultVal = RawBytes[BytesLoaded - 1]; for (unsigned i = 1; i != BytesLoaded; ++i) { ResultVal <<= 8; @@ -466,16 +478,63 @@ static Constant *FoldReinterpretLoadFromConstPtr(Constant *C, return ConstantInt::get(IntType->getContext(), ResultVal); } -/// ConstantFoldLoadFromConstPtr - Return the value that a load from C would -/// produce if it is constant and determinable. If this is not determinable, -/// return null. +static Constant *ConstantFoldLoadThroughBitcast(ConstantExpr *CE, + const DataLayout &DL) { + auto *DestPtrTy = dyn_cast(CE->getType()); + if (!DestPtrTy) + return nullptr; + Type *DestTy = DestPtrTy->getElementType(); + + Constant *C = ConstantFoldLoadFromConstPtr(CE->getOperand(0), DL); + if (!C) + return nullptr; + + do { + Type *SrcTy = C->getType(); + + // If the type sizes are the same and a cast is legal, just directly + // cast the constant. + if (DL.getTypeSizeInBits(DestTy) == DL.getTypeSizeInBits(SrcTy)) { + Instruction::CastOps Cast = Instruction::BitCast; + // If we are going from a pointer to int or vice versa, we spell the cast + // differently. + if (SrcTy->isIntegerTy() && DestTy->isPointerTy()) + Cast = Instruction::IntToPtr; + else if (SrcTy->isPointerTy() && DestTy->isIntegerTy()) + Cast = Instruction::PtrToInt; + + if (CastInst::castIsValid(Cast, C, DestTy)) + return ConstantExpr::getCast(Cast, C, DestTy); + } + + // If this isn't an aggregate type, there is nothing we can do to drill down + // and find a bitcastable constant. + if (!SrcTy->isAggregateType()) + return nullptr; + + // We're simulating a load through a pointer that was bitcast to point to + // a different type, so we can try to walk down through the initial + // elements of an aggregate to see if some part of th e aggregate is + // castable to implement the "load" semantic model. + C = C->getAggregateElement(0u); + } while (C); + + return nullptr; +} + +/// Return the value that a load from C would produce if it is constant and +/// determinable. If this is not determinable, return null. Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C, - const DataLayout *TD) { + const DataLayout &DL) { // First, try the easy cases: if (GlobalVariable *GV = dyn_cast(C)) if (GV->isConstant() && GV->hasDefinitiveInitializer()) return GV->getInitializer(); + if (auto *GA = dyn_cast(C)) + if (GA->getAliasee() && !GA->mayBeOverridden()) + return ConstantFoldLoadFromConstPtr(GA->getAliasee(), DL); + // If the loaded value isn't a constant expr, we can't handle it. ConstantExpr *CE = dyn_cast(C); if (!CE) @@ -491,10 +550,14 @@ Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C, } } + if (CE->getOpcode() == Instruction::BitCast) + if (Constant *LoadedC = ConstantFoldLoadThroughBitcast(CE, DL)) + return LoadedC; + // Instead of loading constant c string, use corresponding integer value // directly if string length is small enough. StringRef Str; - if (TD && getConstantStringInfo(CE, Str) && !Str.empty()) { + if (getConstantStringInfo(CE, Str) && !Str.empty()) { unsigned StrLen = Str.size(); Type *Ty = cast(CE->getType())->getElementType(); unsigned NumBits = Ty->getPrimitiveSizeInBits(); @@ -504,7 +567,7 @@ Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C, (isa(Ty) || Ty->isFloatingPointTy())) { APInt StrVal(NumBits, 0); APInt SingleChar(NumBits, 0); - if (TD->isLittleEndian()) { + if (DL.isLittleEndian()) { for (signed i = StrLen-1; i >= 0; i--) { SingleChar = (uint64_t) Str[i] & UCHAR_MAX; StrVal = (StrVal << 8) | SingleChar; @@ -529,7 +592,7 @@ Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C, // If this load comes from anywhere in a constant global, and if the global // is all undef or zero, we know what it loads. if (GlobalVariable *GV = - dyn_cast(GetUnderlyingObject(CE, TD))) { + dyn_cast(GetUnderlyingObject(CE, DL))) { if (GV->isConstant() && GV->hasDefinitiveInitializer()) { Type *ResTy = cast(C->getType())->getElementType(); if (GV->getInitializer()->isNullValue()) @@ -540,35 +603,34 @@ Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C, } // Try hard to fold loads from bitcasted strange and non-type-safe things. - if (TD) - return FoldReinterpretLoadFromConstPtr(CE, *TD); - return nullptr; + return FoldReinterpretLoadFromConstPtr(CE, DL); } -static Constant *ConstantFoldLoadInst(const LoadInst *LI, const DataLayout *TD){ +static Constant *ConstantFoldLoadInst(const LoadInst *LI, + const DataLayout &DL) { if (LI->isVolatile()) return nullptr; if (Constant *C = dyn_cast(LI->getOperand(0))) - return ConstantFoldLoadFromConstPtr(C, TD); + return ConstantFoldLoadFromConstPtr(C, DL); return nullptr; } -/// SymbolicallyEvaluateBinop - One of Op0/Op1 is a constant expression. +/// One of Op0/Op1 is a constant expression. /// Attempt to symbolically evaluate the result of a binary operator merging /// these together. If target data info is available, it is provided as DL, /// otherwise DL is null. static Constant *SymbolicallyEvaluateBinop(unsigned Opc, Constant *Op0, - Constant *Op1, const DataLayout *DL){ + Constant *Op1, + const DataLayout &DL) { // SROA // Fold (and 0xffffffff00000000, (shl x, 32)) -> shl. // Fold (lshr (or X, Y), 32) -> (lshr [X/Y], 32) if one doesn't contribute // bits. - - if (Opc == Instruction::And && DL) { - unsigned BitWidth = DL->getTypeSizeInBits(Op0->getType()->getScalarType()); + if (Opc == Instruction::And) { + unsigned BitWidth = DL.getTypeSizeInBits(Op0->getType()->getScalarType()); APInt KnownZero0(BitWidth, 0), KnownOne0(BitWidth, 0); APInt KnownZero1(BitWidth, 0), KnownOne1(BitWidth, 0); computeKnownBits(Op0, KnownZero0, KnownOne0, DL); @@ -591,14 +653,13 @@ static Constant *SymbolicallyEvaluateBinop(unsigned Opc, Constant *Op0, // If the constant expr is something like &A[123] - &A[4].f, fold this into a // constant. This happens frequently when iterating over a global array. - if (Opc == Instruction::Sub && DL) { + if (Opc == Instruction::Sub) { GlobalValue *GV1, *GV2; APInt Offs1, Offs2; - if (IsConstantOffsetFromGlobal(Op0, GV1, Offs1, *DL)) - if (IsConstantOffsetFromGlobal(Op1, GV2, Offs2, *DL) && - GV1 == GV2) { - unsigned OpSize = DL->getTypeSizeInBits(Op0->getType()); + 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 @@ -611,24 +672,21 @@ static Constant *SymbolicallyEvaluateBinop(unsigned Opc, Constant *Op0, return nullptr; } -/// CastGEPIndices - If array indices are not pointer-sized integers, -/// explicitly cast them so that they aren't implicitly casted by the -/// getelementptr. -static Constant *CastGEPIndices(ArrayRef Ops, - Type *ResultTy, const DataLayout *TD, +/// If array indices are not pointer-sized integers, explicitly cast them so +/// that they aren't implicitly casted by the getelementptr. +static Constant *CastGEPIndices(Type *SrcTy, ArrayRef Ops, + Type *ResultTy, const DataLayout &DL, const TargetLibraryInfo *TLI) { - if (!TD) - return nullptr; - - Type *IntPtrTy = TD->getIntPtrType(ResultTy); + Type *IntPtrTy = DL.getIntPtrType(ResultTy); bool Any = false; SmallVector NewIdxs; for (unsigned i = 1, e = Ops.size(); i != e; ++i) { if ((i == 1 || !isa(GetElementPtrInst::getIndexedType( - Ops[0]->getType(), - Ops.slice(1, i - 1)))) && + cast(Ops[0]->getType()->getScalarType()) + ->getElementType(), + Ops.slice(1, i - 1)))) && Ops[i]->getType() != IntPtrTy) { Any = true; NewIdxs.push_back(ConstantExpr::getCast(CastInst::getCastOpcode(Ops[i], @@ -643,9 +701,9 @@ static Constant *CastGEPIndices(ArrayRef Ops, if (!Any) return nullptr; - Constant *C = ConstantExpr::getGetElementPtr(Ops[0], NewIdxs); + Constant *C = ConstantExpr::getGetElementPtr(SrcTy, Ops[0], NewIdxs); if (ConstantExpr *CE = dyn_cast(C)) { - if (Constant *Folded = ConstantFoldConstantExpression(CE, TD, TLI)) + if (Constant *Folded = ConstantFoldConstantExpression(CE, DL, TLI)) C = Folded; } @@ -656,7 +714,7 @@ static Constant *CastGEPIndices(ArrayRef Ops, static Constant* StripPtrCastKeepAS(Constant* Ptr) { assert(Ptr->getType()->isPointerTy() && "Not a pointer type"); PointerType *OldPtrTy = cast(Ptr->getType()); - Ptr = cast(Ptr->stripPointerCasts()); + Ptr = Ptr->stripPointerCasts(); PointerType *NewPtrTy = cast(Ptr->getType()); // Preserve the address space number of the pointer. @@ -668,17 +726,16 @@ static Constant* StripPtrCastKeepAS(Constant* Ptr) { return Ptr; } -/// SymbolicallyEvaluateGEP - If we can symbolically evaluate the specified GEP -/// constant expression, do so. -static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, - Type *ResultTy, const DataLayout *TD, +/// If we can symbolically evaluate the GEP constant expression, do so. +static Constant *SymbolicallyEvaluateGEP(Type *SrcTy, ArrayRef Ops, + Type *ResultTy, const DataLayout &DL, const TargetLibraryInfo *TLI) { Constant *Ptr = Ops[0]; - if (!TD || !Ptr->getType()->getPointerElementType()->isSized() || + if (!Ptr->getType()->getPointerElementType()->isSized() || !Ptr->getType()->isPointerTy()) return nullptr; - Type *IntPtrTy = TD->getIntPtrType(Ptr->getType()); + Type *IntPtrTy = DL.getIntPtrType(Ptr->getType()); Type *ResultElementTy = ResultTy->getPointerElementType(); // If this is a constant expr gep that is effectively computing an @@ -698,19 +755,19 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, Res = ConstantExpr::getSub(Res, CE->getOperand(1)); Res = ConstantExpr::getIntToPtr(Res, ResultTy); if (ConstantExpr *ResCE = dyn_cast(Res)) - Res = ConstantFoldConstantExpression(ResCE, TD, TLI); + Res = ConstantFoldConstantExpression(ResCE, DL, TLI); return Res; } } return nullptr; } - unsigned BitWidth = TD->getTypeSizeInBits(IntPtrTy); + unsigned BitWidth = DL.getTypeSizeInBits(IntPtrTy); APInt Offset = - APInt(BitWidth, TD->getIndexedOffset(Ptr->getType(), - makeArrayRef((Value *const*) - Ops.data() + 1, - Ops.size() - 1))); + APInt(BitWidth, + DL.getIndexedOffset( + Ptr->getType(), + makeArrayRef((Value * const *)Ops.data() + 1, Ops.size() - 1))); Ptr = StripPtrCastKeepAS(Ptr); // If this is a GEP of a GEP, fold it all into a single GEP. @@ -728,8 +785,7 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, break; Ptr = cast(GEP->getOperand(0)); - Offset += APInt(BitWidth, - TD->getIndexedOffset(Ptr->getType(), NestedOps)); + Offset += APInt(BitWidth, DL.getIndexedOffset(Ptr->getType(), NestedOps)); Ptr = StripPtrCastKeepAS(Ptr); } @@ -769,7 +825,7 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, } // Determine which element of the array the offset points into. - APInt ElemSize(BitWidth, TD->getTypeAllocSize(ATy->getElementType())); + APInt ElemSize(BitWidth, DL.getTypeAllocSize(ATy->getElementType())); 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 @@ -788,7 +844,7 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, // can't re-form this GEP in a regular form, so bail out. The pointer // operand likely went through casts that are necessary to make the GEP // sensible. - const StructLayout &SL = *TD->getStructLayout(STy); + const StructLayout &SL = *DL.getStructLayout(STy); if (Offset.uge(SL.getSizeInBytes())) break; @@ -813,14 +869,14 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, return nullptr; // Create a GEP. - Constant *C = ConstantExpr::getGetElementPtr(Ptr, NewIdxs); + Constant *C = ConstantExpr::getGetElementPtr(SrcTy, Ptr, NewIdxs); 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 != ResultElementTy) - C = FoldBitCast(C, ResultTy, *TD); + C = FoldBitCast(C, ResultTy, DL); return C; } @@ -831,20 +887,18 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, // Constant Folding public APIs //===----------------------------------------------------------------------===// -/// ConstantFoldInstruction - Try to constant fold the specified instruction. +/// Try to constant fold the specified instruction. /// If successful, the constant result is returned, if not, null is returned. /// Note that this fails if not all of the operands are constant. Otherwise, /// this function can only fail when attempting to fold instructions like loads /// and stores, which have no constant expression form. -Constant *llvm::ConstantFoldInstruction(Instruction *I, - const DataLayout *TD, +Constant *llvm::ConstantFoldInstruction(Instruction *I, const DataLayout &DL, const TargetLibraryInfo *TLI) { // Handle PHI nodes quickly here... if (PHINode *PN = dyn_cast(I)) { Constant *CommonValue = nullptr; - for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { - Value *Incoming = PN->getIncomingValue(i); + for (Value *Incoming : PN->incoming_values()) { // If the incoming value is undef then skip it. Note that while we could // skip the value if it is equal to the phi node itself we choose not to // because that would break the rule that constant folding only applies if @@ -857,7 +911,7 @@ Constant *llvm::ConstantFoldInstruction(Instruction *I, return nullptr; // Fold the PHI's operands. if (ConstantExpr *NewC = dyn_cast(C)) - C = ConstantFoldConstantExpression(NewC, TD, TLI); + C = ConstantFoldConstantExpression(NewC, DL, TLI); // If the incoming value is a different constant to // the one we saw previously, then give up. if (CommonValue && C != CommonValue) @@ -880,17 +934,17 @@ Constant *llvm::ConstantFoldInstruction(Instruction *I, // Fold the Instruction's operands. if (ConstantExpr *NewCE = dyn_cast(Op)) - Op = ConstantFoldConstantExpression(NewCE, TD, TLI); + Op = ConstantFoldConstantExpression(NewCE, DL, TLI); Ops.push_back(Op); } if (const CmpInst *CI = dyn_cast(I)) return ConstantFoldCompareInstOperands(CI->getPredicate(), Ops[0], Ops[1], - TD, TLI); + DL, TLI); if (const LoadInst *LI = dyn_cast(I)) - return ConstantFoldLoadInst(LI, TD); + return ConstantFoldLoadInst(LI, DL); if (InsertValueInst *IVI = dyn_cast(I)) { return ConstantExpr::getInsertValue( @@ -905,13 +959,13 @@ Constant *llvm::ConstantFoldInstruction(Instruction *I, EVI->getIndices()); } - return ConstantFoldInstOperands(I->getOpcode(), I->getType(), Ops, TD, TLI); + return ConstantFoldInstOperands(I->getOpcode(), I->getType(), Ops, DL, TLI); } static Constant * -ConstantFoldConstantExpressionImpl(const ConstantExpr *CE, const DataLayout *TD, +ConstantFoldConstantExpressionImpl(const ConstantExpr *CE, const DataLayout &DL, const TargetLibraryInfo *TLI, - SmallPtrSet &FoldedOps) { + SmallPtrSetImpl &FoldedOps) { SmallVector Ops; for (User::const_op_iterator i = CE->op_begin(), e = CE->op_end(); i != e; ++i) { @@ -919,29 +973,29 @@ ConstantFoldConstantExpressionImpl(const ConstantExpr *CE, const DataLayout *TD, // Recursively fold the ConstantExpr's operands. If we have already folded // a ConstantExpr, we don't have to process it again. if (ConstantExpr *NewCE = dyn_cast(NewC)) { - if (FoldedOps.insert(NewCE)) - NewC = ConstantFoldConstantExpressionImpl(NewCE, TD, TLI, FoldedOps); + if (FoldedOps.insert(NewCE).second) + NewC = ConstantFoldConstantExpressionImpl(NewCE, DL, TLI, FoldedOps); } Ops.push_back(NewC); } if (CE->isCompare()) return ConstantFoldCompareInstOperands(CE->getPredicate(), Ops[0], Ops[1], - TD, TLI); - return ConstantFoldInstOperands(CE->getOpcode(), CE->getType(), Ops, TD, TLI); + DL, TLI); + return ConstantFoldInstOperands(CE->getOpcode(), CE->getType(), Ops, DL, TLI); } -/// ConstantFoldConstantExpression - Attempt to fold the constant expression +/// Attempt to fold the constant expression /// using the specified DataLayout. If successful, the constant result is /// result is returned, if not, null is returned. Constant *llvm::ConstantFoldConstantExpression(const ConstantExpr *CE, - const DataLayout *TD, + const DataLayout &DL, const TargetLibraryInfo *TLI) { SmallPtrSet FoldedOps; - return ConstantFoldConstantExpressionImpl(CE, TD, TLI, FoldedOps); + return ConstantFoldConstantExpressionImpl(CE, DL, TLI, FoldedOps); } -/// ConstantFoldInstOperands - Attempt to constant fold an instruction with the +/// Attempt to constant fold an instruction with the /// specified opcode and operands. If successful, the constant result is /// returned, if not, null is returned. Note that this function can fail when /// attempting to fold instructions like loads and stores, which have no @@ -953,12 +1007,12 @@ Constant *llvm::ConstantFoldConstantExpression(const ConstantExpr *CE, /// Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy, ArrayRef Ops, - const DataLayout *TD, + const DataLayout &DL, const TargetLibraryInfo *TLI) { // Handle easy binops first. if (Instruction::isBinaryOp(Opcode)) { if (isa(Ops[0]) || isa(Ops[1])) { - if (Constant *C = SymbolicallyEvaluateBinop(Opcode, Ops[0], Ops[1], TD)) + if (Constant *C = SymbolicallyEvaluateBinop(Opcode, Ops[0], Ops[1], DL)) return C; } @@ -978,10 +1032,10 @@ Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy, // If the input is a inttoptr, eliminate the pair. This requires knowing // the width of a pointer, so it can't be done in ConstantExpr::getCast. if (ConstantExpr *CE = dyn_cast(Ops[0])) { - if (TD && CE->getOpcode() == Instruction::IntToPtr) { + if (CE->getOpcode() == Instruction::IntToPtr) { Constant *Input = CE->getOperand(0); unsigned InWidth = Input->getType()->getScalarSizeInBits(); - unsigned PtrWidth = TD->getPointerTypeSizeInBits(CE->getType()); + unsigned PtrWidth = DL.getPointerTypeSizeInBits(CE->getType()); if (PtrWidth < InWidth) { Constant *Mask = ConstantInt::get(CE->getContext(), @@ -999,15 +1053,15 @@ Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy, // This requires knowing the width of a pointer, so it can't be done in // ConstantExpr::getCast. if (ConstantExpr *CE = dyn_cast(Ops[0])) { - if (TD && CE->getOpcode() == Instruction::PtrToInt) { + if (CE->getOpcode() == Instruction::PtrToInt) { Constant *SrcPtr = CE->getOperand(0); - unsigned SrcPtrSize = TD->getPointerTypeSizeInBits(SrcPtr->getType()); + unsigned SrcPtrSize = DL.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 FoldBitCast(CE->getOperand(0), DestTy, DL); } } } @@ -1025,9 +1079,7 @@ Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy, case Instruction::AddrSpaceCast: return ConstantExpr::getCast(Opcode, Ops[0], DestTy); case Instruction::BitCast: - if (TD) - return FoldBitCast(Ops[0], DestTy, *TD); - return ConstantExpr::getBitCast(Ops[0], DestTy); + return FoldBitCast(Ops[0], DestTy, DL); case Instruction::Select: return ConstantExpr::getSelect(Ops[0], Ops[1], Ops[2]); case Instruction::ExtractElement: @@ -1036,59 +1088,61 @@ Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy, return ConstantExpr::getInsertElement(Ops[0], Ops[1], Ops[2]); case Instruction::ShuffleVector: return ConstantExpr::getShuffleVector(Ops[0], Ops[1], Ops[2]); - case Instruction::GetElementPtr: - if (Constant *C = CastGEPIndices(Ops, DestTy, TD, TLI)) + case Instruction::GetElementPtr: { + Type *SrcTy = nullptr; + if (Constant *C = CastGEPIndices(SrcTy, Ops, DestTy, DL, TLI)) return C; - if (Constant *C = SymbolicallyEvaluateGEP(Ops, DestTy, TD, TLI)) + if (Constant *C = SymbolicallyEvaluateGEP(SrcTy, Ops, DestTy, DL, TLI)) return C; - return ConstantExpr::getGetElementPtr(Ops[0], Ops.slice(1)); + return ConstantExpr::getGetElementPtr(SrcTy, Ops[0], Ops.slice(1)); + } } } -/// ConstantFoldCompareInstOperands - Attempt to constant fold a compare +/// Attempt to constant fold a compare /// instruction (icmp/fcmp) with the specified operands. If it fails, it /// returns a constant expression of the specified operands. -/// Constant *llvm::ConstantFoldCompareInstOperands(unsigned Predicate, Constant *Ops0, Constant *Ops1, - const DataLayout *TD, + const DataLayout &DL, const TargetLibraryInfo *TLI) { // fold: icmp (inttoptr x), null -> icmp x, 0 // fold: icmp (ptrtoint x), 0 -> icmp x, null // fold: icmp (inttoptr x), (inttoptr y) -> icmp trunc/zext x, trunc/zext y // fold: icmp (ptrtoint x), (ptrtoint y) -> icmp x, y // - // ConstantExpr::getCompare cannot do this, because it doesn't have TD + // FIXME: The following comment is out of data and the DataLayout is here now. + // ConstantExpr::getCompare cannot do this, because it doesn't have DL // around to know if bit truncation is happening. if (ConstantExpr *CE0 = dyn_cast(Ops0)) { - if (TD && Ops1->isNullValue()) { + if (Ops1->isNullValue()) { if (CE0->getOpcode() == Instruction::IntToPtr) { - Type *IntPtrTy = TD->getIntPtrType(CE0->getType()); + Type *IntPtrTy = DL.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), IntPtrTy, false); Constant *Null = Constant::getNullValue(C->getType()); - return ConstantFoldCompareInstOperands(Predicate, C, Null, TD, TLI); + return ConstantFoldCompareInstOperands(Predicate, C, Null, DL, TLI); } // 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) { - Type *IntPtrTy = TD->getIntPtrType(CE0->getOperand(0)->getType()); + Type *IntPtrTy = DL.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); + return ConstantFoldCompareInstOperands(Predicate, C, Null, DL, TLI); } } } if (ConstantExpr *CE1 = dyn_cast(Ops1)) { - if (TD && CE0->getOpcode() == CE1->getOpcode()) { + if (CE0->getOpcode() == CE1->getOpcode()) { if (CE0->getOpcode() == Instruction::IntToPtr) { - Type *IntPtrTy = TD->getIntPtrType(CE0->getType()); + Type *IntPtrTy = DL.getIntPtrType(CE0->getType()); // Convert the integer value to the right size to ensure we get the // proper extension or truncation. @@ -1096,20 +1150,17 @@ Constant *llvm::ConstantFoldCompareInstOperands(unsigned Predicate, IntPtrTy, false); Constant *C1 = ConstantExpr::getIntegerCast(CE1->getOperand(0), IntPtrTy, false); - return ConstantFoldCompareInstOperands(Predicate, C0, C1, TD, TLI); + return ConstantFoldCompareInstOperands(Predicate, C0, C1, DL, TLI); } // 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) { - Type *IntPtrTy = TD->getIntPtrType(CE0->getOperand(0)->getType()); + Type *IntPtrTy = DL.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); + return ConstantFoldCompareInstOperands( + Predicate, CE0->getOperand(0), CE1->getOperand(0), DL, TLI); } } } @@ -1119,16 +1170,14 @@ Constant *llvm::ConstantFoldCompareInstOperands(unsigned Predicate, // icmp ne (or x, y), 0 -> (icmp ne x, 0) | (icmp ne y, 0) if ((Predicate == ICmpInst::ICMP_EQ || Predicate == ICmpInst::ICMP_NE) && CE0->getOpcode() == Instruction::Or && Ops1->isNullValue()) { - Constant *LHS = - ConstantFoldCompareInstOperands(Predicate, CE0->getOperand(0), Ops1, - TD, TLI); - Constant *RHS = - ConstantFoldCompareInstOperands(Predicate, CE0->getOperand(1), Ops1, - TD, TLI); + Constant *LHS = ConstantFoldCompareInstOperands( + Predicate, CE0->getOperand(0), Ops1, DL, TLI); + Constant *RHS = ConstantFoldCompareInstOperands( + Predicate, CE0->getOperand(1), Ops1, DL, TLI); unsigned OpC = Predicate == ICmpInst::ICMP_EQ ? Instruction::And : Instruction::Or; Constant *Ops[] = { LHS, RHS }; - return ConstantFoldInstOperands(OpC, LHS->getType(), Ops, TD, TLI); + return ConstantFoldInstOperands(OpC, LHS->getType(), Ops, DL, TLI); } } @@ -1136,9 +1185,9 @@ Constant *llvm::ConstantFoldCompareInstOperands(unsigned Predicate, } -/// ConstantFoldLoadThroughGEPConstantExpr - Given a constant and a -/// getelementptr constantexpr, return the constant value being addressed by the -/// constant expression, or null if something is funny and we can't decide. +/// Given a constant and a getelementptr constantexpr, return the constant value +/// being addressed by the constant expression, or null if something is funny +/// and we can't decide. Constant *llvm::ConstantFoldLoadThroughGEPConstantExpr(Constant *C, ConstantExpr *CE) { if (!CE->getOperand(1)->isNullValue()) @@ -1154,10 +1203,9 @@ Constant *llvm::ConstantFoldLoadThroughGEPConstantExpr(Constant *C, return C; } -/// ConstantFoldLoadThroughGEPIndices - Given a constant and getelementptr -/// indices (with an *implied* zero pointer index that is not in the list), -/// return the constant value being addressed by a virtual load, or null if -/// something is funny and we can't decide. +/// Given a constant and getelementptr indices (with an *implied* zero pointer +/// index that is not in the list), return the constant value being addressed by +/// a virtual load, or null if something is funny and we can't decide. Constant *llvm::ConstantFoldLoadThroughGEPIndices(Constant *C, ArrayRef Indices) { // Loop over all of the operands, tracking down which value we are @@ -1175,11 +1223,12 @@ Constant *llvm::ConstantFoldLoadThroughGEPIndices(Constant *C, // Constant Folding for Calls // -/// canConstantFoldCallTo - Return true if its even possible to fold a call to -/// the specified function. +/// Return true if it's even possible to fold a call to the specified function. bool llvm::canConstantFoldCallTo(const Function *F) { switch (F->getIntrinsicID()) { case Intrinsic::fabs: + case Intrinsic::minnum: + case Intrinsic::maxnum: case Intrinsic::log: case Intrinsic::log2: case Intrinsic::log10: @@ -1188,6 +1237,11 @@ bool llvm::canConstantFoldCallTo(const Function *F) { case Intrinsic::floor: case Intrinsic::ceil: case Intrinsic::sqrt: + case Intrinsic::sin: + case Intrinsic::cos: + case Intrinsic::trunc: + case Intrinsic::rint: + case Intrinsic::nearbyint: case Intrinsic::pow: case Intrinsic::powi: case Intrinsic::bswap: @@ -1264,12 +1318,34 @@ static Constant *GetConstantFoldFPValue(double V, Type *Ty) { } +namespace { +/// Clear the floating-point exception state. +static inline void llvm_fenv_clearexcept() { +#if defined(HAVE_FENV_H) && HAVE_DECL_FE_ALL_EXCEPT + feclearexcept(FE_ALL_EXCEPT); +#endif + errno = 0; +} + +/// Test if a floating-point exception was raised. +static inline bool llvm_fenv_testexcept() { + int errno_val = errno; + if (errno_val == ERANGE || errno_val == EDOM) + return true; +#if defined(HAVE_FENV_H) && HAVE_DECL_FE_ALL_EXCEPT && HAVE_DECL_FE_INEXACT + if (fetestexcept(FE_ALL_EXCEPT & ~FE_INEXACT)) + return true; +#endif + return false; +} +} // End namespace + static Constant *ConstantFoldFP(double (*NativeFP)(double), double V, Type *Ty) { - sys::llvm_fenv_clearexcept(); + llvm_fenv_clearexcept(); V = NativeFP(V); - if (sys::llvm_fenv_testexcept()) { - sys::llvm_fenv_clearexcept(); + if (llvm_fenv_testexcept()) { + llvm_fenv_clearexcept(); return nullptr; } @@ -1278,24 +1354,23 @@ static Constant *ConstantFoldFP(double (*NativeFP)(double), double V, static Constant *ConstantFoldBinaryFP(double (*NativeFP)(double, double), double V, double W, Type *Ty) { - sys::llvm_fenv_clearexcept(); + llvm_fenv_clearexcept(); V = NativeFP(V, W); - if (sys::llvm_fenv_testexcept()) { - sys::llvm_fenv_clearexcept(); + if (llvm_fenv_testexcept()) { + llvm_fenv_clearexcept(); return nullptr; } return GetConstantFoldFPValue(V, Ty); } -/// ConstantFoldConvertToInt - Attempt to an SSE floating point to integer -/// conversion of a constant floating point. If roundTowardZero is false, the -/// default IEEE rounding is used (toward nearest, ties to even). This matches -/// the behavior of the non-truncating SSE instructions in the default rounding -/// mode. The desired integer type Ty is used to select how many bits are -/// available for the result. Returns null if the conversion cannot be -/// performed, otherwise returns the Constant value resulting from the -/// conversion. +/// Attempt to fold an SSE floating point to integer conversion of a constant +/// floating point. If roundTowardZero is false, the default IEEE rounding is +/// used (toward nearest, ties to even). This matches the behavior of the +/// non-truncating SSE instructions in the default rounding mode. The desired +/// integer type Ty is used to select how many bits are available for the +/// result. Returns null if the conversion cannot be performed, otherwise +/// returns the Constant value resulting from the conversion. static Constant *ConstantFoldConvertToInt(const APFloat &Val, bool roundTowardZero, Type *Ty) { // All of these conversion intrinsics form an integer of at most 64bits. @@ -1353,6 +1428,36 @@ static Constant *ConstantFoldScalarCall(StringRef Name, unsigned IntrinsicID, return ConstantFP::get(Ty->getContext(), V); } + if (IntrinsicID == Intrinsic::floor) { + APFloat V = Op->getValueAPF(); + V.roundToIntegral(APFloat::rmTowardNegative); + return ConstantFP::get(Ty->getContext(), V); + } + + if (IntrinsicID == Intrinsic::ceil) { + APFloat V = Op->getValueAPF(); + V.roundToIntegral(APFloat::rmTowardPositive); + return ConstantFP::get(Ty->getContext(), V); + } + + if (IntrinsicID == Intrinsic::trunc) { + APFloat V = Op->getValueAPF(); + V.roundToIntegral(APFloat::rmTowardZero); + return ConstantFP::get(Ty->getContext(), V); + } + + if (IntrinsicID == Intrinsic::rint) { + APFloat V = Op->getValueAPF(); + V.roundToIntegral(APFloat::rmNearestTiesToEven); + return ConstantFP::get(Ty->getContext(), V); + } + + if (IntrinsicID == Intrinsic::nearbyint) { + APFloat V = Op->getValueAPF(); + V.roundToIntegral(APFloat::rmNearestTiesToEven); + 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. @@ -1369,30 +1474,20 @@ static Constant *ConstantFoldScalarCall(StringRef Name, unsigned IntrinsicID, default: break; case Intrinsic::fabs: return ConstantFoldFP(fabs, V, Ty); -#if HAVE_LOG2 case Intrinsic::log2: - return ConstantFoldFP(log2, V, Ty); -#endif -#if HAVE_LOG + return ConstantFoldFP(Log2, V, Ty); case Intrinsic::log: return ConstantFoldFP(log, V, Ty); -#endif -#if HAVE_LOG10 case Intrinsic::log10: return ConstantFoldFP(log10, V, Ty); -#endif -#if HAVE_EXP case Intrinsic::exp: return ConstantFoldFP(exp, V, Ty); -#endif -#if HAVE_EXP2 case Intrinsic::exp2: return ConstantFoldFP(exp2, V, Ty); -#endif - case Intrinsic::floor: - return ConstantFoldFP(floor, V, Ty); - case Intrinsic::ceil: - return ConstantFoldFP(ceil, V, Ty); + case Intrinsic::sin: + return ConstantFoldFP(sin, V, Ty); + case Intrinsic::cos: + return ConstantFoldFP(cos, V, Ty); } if (!TLI) @@ -1442,8 +1537,14 @@ static Constant *ConstantFoldScalarCall(StringRef Name, unsigned IntrinsicID, (Ty->isHalfTy() || Ty->isFloatTy() || Ty->isDoubleTy())) { if (V >= -0.0) return ConstantFoldFP(sqrt, V, Ty); - else // Undefined - return Constant::getNullValue(Ty); + else { + // Unlike the sqrt definitions in C/C++, POSIX, and IEEE-754 - which + // all guarantee or favor returning NaN - the square root of a + // negative number is not defined for the LLVM sqrt intrinsic. + // This is because the intrinsic should only be emitted in place of + // libm's sqrt function when using "no-nans-fp-math". + return UndefValue::get(Ty); + } } break; case 's': @@ -1480,8 +1581,8 @@ static Constant *ConstantFoldScalarCall(StringRef Name, unsigned IntrinsicID, APFloat Val(APFloat::IEEEhalf, Op->getValue()); bool lost = false; - APFloat::opStatus status = - Val.convert(APFloat::IEEEsingle, APFloat::rmNearestTiesToEven, &lost); + APFloat::opStatus status = Val.convert( + Ty->getFltSemantics(), APFloat::rmNearestTiesToEven, &lost); // Conversion is always precise. (void)status; @@ -1549,6 +1650,19 @@ static Constant *ConstantFoldScalarCall(StringRef Name, unsigned IntrinsicID, V1.copySign(V2); return ConstantFP::get(Ty->getContext(), V1); } + + if (IntrinsicID == Intrinsic::minnum) { + const APFloat &C1 = Op1->getValueAPF(); + const APFloat &C2 = Op2->getValueAPF(); + return ConstantFP::get(Ty->getContext(), minnum(C1, C2)); + } + + if (IntrinsicID == Intrinsic::maxnum) { + const APFloat &C1 = Op1->getValueAPF(); + const APFloat &C2 = Op2->getValueAPF(); + return ConstantFP::get(Ty->getContext(), maxnum(C1, C2)); + } + if (!TLI) return nullptr; if (Name == "pow" && TLI->has(LibFunc::pow)) @@ -1684,7 +1798,7 @@ static Constant *ConstantFoldVectorCall(StringRef Name, unsigned IntrinsicID, return ConstantVector::get(Result); } -/// ConstantFoldCall - Attempt to constant fold a call to the specified function +/// Attempt to constant fold a call to the specified function /// with the specified arguments, returning null if unsuccessful. Constant * llvm::ConstantFoldCall(Function *F, ArrayRef Operands,