X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FAnalysis%2FConstantFolding.cpp;h=6a37f95194253a779a306147a6d0750a32c8e598;hb=a94bd7a780a437f52913841d50e2d5a59417bbe1;hp=2a9630e234bcba8675c185cf4318fcee6eb43c51;hpb=618c1dbd293d15ee19f61b1156ab8086ad28311a;p=oota-llvm.git diff --git a/lib/Analysis/ConstantFolding.cpp b/lib/Analysis/ConstantFolding.cpp index 2a9630e234b..6a37f951942 100644 --- a/lib/Analysis/ConstantFolding.cpp +++ b/lib/Analysis/ConstantFolding.cpp @@ -9,83 +9,126 @@ // // This file defines routines for folding instructions into constants. // -// Also, to supplement the basic VMCore ConstantExpr simplifications, +// Also, to supplement the basic IR ConstantExpr simplifications, // this file defines some additional folding routines that can make use of -// TargetData information. These functions cannot go in VMCore due to library +// DataLayout information. These functions cannot go in IR due to library // dependency issues. // //===----------------------------------------------------------------------===// #include "llvm/Analysis/ConstantFolding.h" -#include "llvm/Constants.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Function.h" -#include "llvm/GlobalVariable.h" -#include "llvm/Instructions.h" -#include "llvm/Intrinsics.h" -#include "llvm/Operator.h" -#include "llvm/Analysis/ValueTracking.h" -#include "llvm/Target/TargetData.h" -#include "llvm/Target/TargetLibraryInfo.h" +#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" +#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/GetElementPtrTypeIterator.h" #include "llvm/Support/MathExtras.h" -#include "llvm/Support/FEnv.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 -/// TargetData. 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 TargetData &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. + if (IntegerType *IT = dyn_cast(DestTy)) { + VectorType *VTy = dyn_cast(C->getType()); + if (!VTy) + return ConstantExpr::getBitCast(C, DestTy); + + unsigned NumSrcElts = VTy->getNumElements(); + Type *SrcEltTy = VTy->getElementType(); + + // If the vector is a vector of floating point, convert it to vector of int + // to simplify things. + if (SrcEltTy->isFloatingPointTy()) { + unsigned FPWidth = SrcEltTy->getPrimitiveSizeInBits(); + Type *SrcIVTy = + VectorType::get(IntegerType::get(C->getContext(), FPWidth), NumSrcElts); + // Ask IR to do the conversion now that #elts line up. + C = ConstantExpr::getBitCast(C, SrcIVTy); + } + + ConstantDataVector *CDV = dyn_cast(C); + if (!CDV) + return ConstantExpr::getBitCast(C, DestTy); + + // Now that we know that the input value is a vector of integers, just shift + // and insert them into our result. + unsigned BitShift = DL.getTypeAllocSizeInBits(SrcEltTy); + APInt Result(IT->getBitWidth(), 0); + for (unsigned i = 0; i != NumSrcElts; ++i) { + Result <<= BitShift; + if (DL.isLittleEndian()) + Result |= CDV->getElementAsInteger(NumSrcElts-i-1); + else + Result |= CDV->getElementAsInteger(i); + } + + return ConstantInt::get(IT, Result); + } + // The code below only handles casts to vectors currently. VectorType *DestVTy = dyn_cast(DestTy); - if (DestVTy == 0) + if (!DestVTy) return ConstantExpr::getBitCast(C, DestTy); - + // If this is a scalar -> vector cast, convert the input into a <1 x scalar> // 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. - ConstantVector *CV = dyn_cast(C); - if (CV == 0) + if (!isa(C) && !isa(C)) return ConstantExpr::getBitCast(C, DestTy); - - // If the element types match, VMCore can fold it. + + // If the element types match, IR can fold it. unsigned NumDstElt = DestVTy->getNumElements(); - unsigned NumSrcElt = CV->getNumOperands(); + unsigned NumSrcElt = C->getType()->getVectorNumElements(); if (NumDstElt == NumSrcElt) return ConstantExpr::getBitCast(C, DestTy); - - Type *SrcEltTy = CV->getType()->getElementType(); + + Type *SrcEltTy = C->getType()->getVectorElementType(); Type *DstEltTy = DestVTy->getElementType(); - - // Otherwise, we're changing the number of elements in a vector, which + + // Otherwise, we're changing the number of elements in a vector, which // requires endianness information to do the right thing. For example, // bitcast (<2 x i64> to <4 x i32>) // folds to (little endian): // <4 x i32> // and to (big endian): // <4 x i32> - + // First thing is first. We only want to think about integer here, so if // we have something in FP form, recast it as integer. if (DstEltTy->isFloatingPointTy()) { @@ -94,32 +137,32 @@ 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); - if (!C) return ConstantExpr::getBitCast(C, DestTy); - - // Finally, VMCore can handle this now that #elts line up. + C = FoldBitCast(C, DestIVTy, DL); + + // Finally, IR can handle this now that #elts line up. return ConstantExpr::getBitCast(C, DestTy); } - + // Okay, we know the destination is integer, if the input is FP, convert // it to integer first. if (SrcEltTy->isFloatingPointTy()) { unsigned FPWidth = SrcEltTy->getPrimitiveSizeInBits(); Type *SrcIVTy = VectorType::get(IntegerType::get(C->getContext(), FPWidth), NumSrcElt); - // Ask VMCore to do the conversion now that #elts line up. + // Ask IR to do the conversion now that #elts line up. C = ConstantExpr::getBitCast(C, SrcIVTy); - CV = dyn_cast(C); - if (!CV) // If VMCore wasn't able to fold it, bail out. + // If IR wasn't able to fold it, bail out. + if (!isa(C) && // FIXME: Remove ConstantVector. + !isa(C)) return C; } - + // Now we know that the input and output vectors are both integer vectors // 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) { // Handle: bitcast (<4 x i32> to <2 x i64>) @@ -132,166 +175,170 @@ static Constant *FoldBitCast(Constant *C, Type *DestTy, Constant *Elt = Zero; unsigned ShiftAmt = isLittleEndian ? 0 : SrcBitSize*(Ratio-1); for (unsigned j = 0; j != Ratio; ++j) { - Constant *Src = dyn_cast(CV->getOperand(SrcElt++)); + Constant *Src =dyn_cast(C->getAggregateElement(SrcElt++)); if (!Src) // Reject constantexpr elements. return ConstantExpr::getBitCast(C, DestTy); - + // Zero extend the element to the right size. Src = ConstantExpr::getZExt(Src, Elt->getType()); - + // Shift it to the right place, depending on endianness. - Src = ConstantExpr::getShl(Src, + Src = ConstantExpr::getShl(Src, ConstantInt::get(Src->getType(), ShiftAmt)); ShiftAmt += isLittleEndian ? SrcBitSize : -SrcBitSize; - + // Mix it in. Elt = ConstantExpr::getOr(Elt, Src); } Result.push_back(Elt); } - } else { - // Handle: bitcast (<2 x i64> to <4 x i32>) - unsigned Ratio = NumDstElt/NumSrcElt; - unsigned DstBitSize = DstEltTy->getPrimitiveSizeInBits(); - - // Loop over each source value, expanding into multiple results. - for (unsigned i = 0; i != NumSrcElt; ++i) { - Constant *Src = dyn_cast(CV->getOperand(i)); - if (!Src) // Reject constantexpr elements. - return ConstantExpr::getBitCast(C, DestTy); - - unsigned ShiftAmt = isLittleEndian ? 0 : DstBitSize*(Ratio-1); - for (unsigned j = 0; j != Ratio; ++j) { - // Shift the piece of the value into the right place, depending on - // endianness. - Constant *Elt = ConstantExpr::getLShr(Src, - ConstantInt::get(Src->getType(), ShiftAmt)); - ShiftAmt += isLittleEndian ? DstBitSize : -DstBitSize; - - // Truncate and remember this piece. - Result.push_back(ConstantExpr::getTrunc(Elt, DstEltTy)); + return ConstantVector::get(Result); + } + + // Handle: bitcast (<2 x i64> to <4 x i32>) + unsigned Ratio = NumDstElt/NumSrcElt; + unsigned DstBitSize = DL.getTypeSizeInBits(DstEltTy); + + // Loop over each source value, expanding into multiple results. + for (unsigned i = 0; i != NumSrcElt; ++i) { + Constant *Src = dyn_cast(C->getAggregateElement(i)); + if (!Src) // Reject constantexpr elements. + return ConstantExpr::getBitCast(C, DestTy); + + unsigned ShiftAmt = isLittleEndian ? 0 : DstBitSize*(Ratio-1); + for (unsigned j = 0; j != Ratio; ++j) { + // Shift the piece of the value into the right place, depending on + // endianness. + Constant *Elt = ConstantExpr::getLShr(Src, + 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)); } } - + return ConstantVector::get(Result); } -/// 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, - int64_t &Offset, const TargetData &TD) { + APInt &Offset, const DataLayout &DL) { // Trivial case, constant is the global. if ((GV = dyn_cast(C))) { - Offset = 0; + unsigned BitWidth = DL.getPointerTypeSizeInBits(GV->getType()); + Offset = APInt(BitWidth, 0); return true; } - + // Otherwise, if this isn't a constant expr, bail out. ConstantExpr *CE = dyn_cast(C); if (!CE) return false; - + // 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); - - // i32* getelementptr ([5 x i32]* @a, i32 0, i32 5) - if (CE->getOpcode() == Instruction::GetElementPtr) { - // Cannot compute this if the element type of the pointer is missing size - // info. - if (!cast(CE->getOperand(0)->getType()) - ->getElementType()->isSized()) - return false; - - // If the base isn't a global+constant, we aren't either. - if (!IsConstantOffsetFromGlobal(CE->getOperand(0), GV, Offset, TD)) - return false; - - // Otherwise, add any offset that our operands provide. - gep_type_iterator GTI = gep_type_begin(CE); - for (User::const_op_iterator i = CE->op_begin() + 1, e = CE->op_end(); - i != e; ++i, ++GTI) { - ConstantInt *CI = dyn_cast(*i); - if (!CI) return false; // Index isn't a simple constant? - if (CI->isZero()) continue; // Not adding anything. - - if (StructType *ST = dyn_cast(*GTI)) { - // N = N + Offset - Offset += TD.getStructLayout(ST)->getElementOffset(CI->getZExtValue()); - } else { - SequentialType *SQT = cast(*GTI); - Offset += TD.getTypeAllocSize(SQT->getElementType())*CI->getSExtValue(); - } - } - return true; - } - - return false; + CE->getOpcode() == Instruction::BitCast || + CE->getOpcode() == Instruction::AddrSpaceCast) + 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 = 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, DL)) + return false; + + // Otherwise, add any offset that our operands provide. + 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 TargetData &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 // zero initialized. if (isa(C) || isa(C)) return true; - + if (ConstantInt *CI = dyn_cast(C)) { if (CI->getBitWidth() > 64 || (CI->getBitWidth() & 7) != 0) return false; - + uint64_t Val = CI->getZExtValue(); unsigned IntBytes = unsigned(CI->getBitWidth()/8); - + for (unsigned i = 0; i != BytesLeft && ByteOffset != IntBytes; ++i) { - CurPtr[i] = (unsigned char)(Val >> (ByteOffset * 8)); + int n = ByteOffset; + if (!DL.isLittleEndian()) + n = IntBytes - n - 1; + CurPtr[i] = (unsigned char)(Val >> (n * 8)); ++ByteOffset; } return true; } - + 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()), 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; - + 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; - + // Check to see if we read from the last struct element, if so we're done. if (Index == CS->getType()->getNumElements()) return true; @@ -299,59 +346,53 @@ static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, // 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; } // not reached. } - if (ConstantArray *CA = dyn_cast(C)) { - uint64_t EltSize = TD.getTypeAllocSize(CA->getType()->getElementType()); + if (isa(C) || isa(C) || + isa(C)) { + Type *EltTy = C->getType()->getSequentialElementType(); + uint64_t EltSize = DL.getTypeAllocSize(EltTy); uint64_t Index = ByteOffset / EltSize; uint64_t Offset = ByteOffset - Index * EltSize; - for (; Index != CA->getType()->getNumElements(); ++Index) { - if (!ReadDataFromGlobal(CA->getOperand(Index), Offset, CurPtr, - BytesLeft, TD)) - return false; - if (EltSize >= BytesLeft) - return true; - - Offset = 0; - BytesLeft -= EltSize; - CurPtr += EltSize; - } - return true; - } - - if (ConstantVector *CV = dyn_cast(C)) { - uint64_t EltSize = TD.getTypeAllocSize(CV->getType()->getElementType()); - uint64_t Index = ByteOffset / EltSize; - uint64_t Offset = ByteOffset - Index * EltSize; - for (; Index != CV->getType()->getNumElements(); ++Index) { - if (!ReadDataFromGlobal(CV->getOperand(Index), Offset, CurPtr, - BytesLeft, TD)) + uint64_t NumElts; + if (ArrayType *AT = dyn_cast(C->getType())) + NumElts = AT->getNumElements(); + else + NumElts = C->getType()->getVectorNumElements(); + + for (; Index != NumElts; ++Index) { + if (!ReadDataFromGlobal(C->getAggregateElement(Index), Offset, CurPtr, + BytesLeft, DL)) return false; - if (EltSize >= BytesLeft) + + uint64_t BytesWritten = EltSize - Offset; + assert(BytesWritten <= EltSize && "Not indexing into this element?"); + if (BytesWritten >= BytesLeft) return true; - + Offset = 0; - BytesLeft -= EltSize; - CurPtr += EltSize; + BytesLeft -= BytesWritten; + CurPtr += BytesWritten; } return true; } - + if (ConstantExpr *CE = dyn_cast(C)) { if (CE->getOpcode() == Instruction::IntToPtr && - CE->getOperand(0)->getType() == TD.getIntPtrType(CE->getContext())) - return ReadDataFromGlobal(CE->getOperand(0), ByteOffset, CurPtr, - BytesLeft, TD); + CE->getOperand(0)->getType() == DL.getIntPtrType(CE->getType())) { + return ReadDataFromGlobal(CE->getOperand(0), ByteOffset, CurPtr, + BytesLeft, DL); + } } // Otherwise, unknown initializer type. @@ -359,74 +400,133 @@ static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, } static Constant *FoldReinterpretLoadFromConstPtr(Constant *C, - const TargetData &TD) { - Type *LoadTy = cast(C->getType())->getElementType(); + const DataLayout &DL) { + PointerType *PTy = cast(C->getType()); + Type *LoadTy = PTy->getElementType(); IntegerType *IntType = dyn_cast(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->isFloatTy()) - MapTy = Type::getInt32PtrTy(C->getContext()); + if (LoadTy->isHalfTy()) + MapTy = Type::getInt16PtrTy(C->getContext(), AS); + else if (LoadTy->isFloatTy()) + 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(), + DL.getTypeAllocSizeInBits(LoadTy), AS); } else - return 0; + return nullptr; - C = FoldBitCast(C, MapTy, TD); - if (Constant *Res = FoldReinterpretLoadFromConstPtr(C, TD)) - return FoldBitCast(Res, LoadTy, TD); - return 0; + C = FoldBitCast(C, MapTy, DL); + if (Constant *Res = FoldReinterpretLoadFromConstPtr(C, DL)) + return FoldBitCast(Res, LoadTy, DL); + return nullptr; } - + unsigned BytesLoaded = (IntType->getBitWidth() + 7) / 8; - if (BytesLoaded > 32 || BytesLoaded == 0) return 0; - + if (BytesLoaded > 32 || BytesLoaded == 0) + return nullptr; + GlobalValue *GVal; - int64_t Offset; - if (!IsConstantOffsetFromGlobal(C, GVal, Offset, TD)) - return 0; - + APInt Offset; + if (!IsConstantOffsetFromGlobal(C, GVal, Offset, DL)) + return nullptr; + GlobalVariable *GV = dyn_cast(GVal); if (!GV || !GV->isConstant() || !GV->hasDefinitiveInitializer() || !GV->getInitializer()->getType()->isSized()) - return 0; + return nullptr; // 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 < 0) return 0; - + if (Offset.isNegative()) + return nullptr; + // If we're not accessing anything in this constant, the result is undefined. - if (uint64_t(Offset) >= TD.getTypeAllocSize(GV->getInitializer()->getType())) + if (Offset.getZExtValue() >= + DL.getTypeAllocSize(GV->getInitializer()->getType())) return UndefValue::get(IntType); - + unsigned char RawBytes[32] = {0}; - if (!ReadDataFromGlobal(GV->getInitializer(), Offset, RawBytes, - BytesLoaded, TD)) - return 0; - - APInt ResultVal = APInt(IntType->getBitWidth(), RawBytes[BytesLoaded-1]); - for (unsigned i = 1; i != BytesLoaded; ++i) { - ResultVal <<= 8; - ResultVal |= RawBytes[BytesLoaded-1-i]; + if (!ReadDataFromGlobal(GV->getInitializer(), Offset.getZExtValue(), RawBytes, + BytesLoaded, DL)) + return nullptr; + + APInt ResultVal = APInt(IntType->getBitWidth(), 0); + if (DL.isLittleEndian()) { + ResultVal = RawBytes[BytesLoaded - 1]; + for (unsigned i = 1; i != BytesLoaded; ++i) { + ResultVal <<= 8; + ResultVal |= RawBytes[BytesLoaded - 1 - i]; + } + } else { + ResultVal = RawBytes[0]; + for (unsigned i = 1; i != BytesLoaded; ++i) { + ResultVal <<= 8; + ResultVal |= RawBytes[i]; + } } 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 TargetData *TD) { + const DataLayout &DL) { // First, try the easy cases: if (GlobalVariable *GV = dyn_cast(C)) if (GV->isConstant() && GV->hasDefinitiveInitializer()) @@ -434,21 +534,28 @@ Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C, // If the loaded value isn't a constant expr, we can't handle it. ConstantExpr *CE = dyn_cast(C); - if (!CE) return 0; - + if (!CE) + return nullptr; + if (CE->getOpcode() == Instruction::GetElementPtr) { - if (GlobalVariable *GV = dyn_cast(CE->getOperand(0))) - if (GV->isConstant() && GV->hasDefinitiveInitializer()) - if (Constant *V = + if (GlobalVariable *GV = dyn_cast(CE->getOperand(0))) { + if (GV->isConstant() && GV->hasDefinitiveInitializer()) { + if (Constant *V = ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE)) return V; + } + } } - + + 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. - std::string Str; - if (TD && GetConstantStringInfo(CE, Str) && !Str.empty()) { - unsigned StrLen = Str.length(); + StringRef Str; + if (getConstantStringInfo(CE, Str) && !Str.empty()) { + unsigned StrLen = Str.size(); Type *Ty = cast(CE->getType())->getElementType(); unsigned NumBits = Ty->getPrimitiveSizeInBits(); // Replace load with immediate integer if the result is an integer or fp @@ -457,7 +564,7 @@ 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; @@ -471,18 +578,18 @@ Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C, SingleChar = 0; StrVal = (StrVal << 8) | SingleChar; } - + Constant *Res = ConstantInt::get(CE->getContext(), StrVal); if (Ty->isFloatingPointTy()) Res = ConstantExpr::getBitCast(Res, Ty); return Res; } } - + // 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()) @@ -491,69 +598,92 @@ Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C, return UndefValue::get(ResTy); } } - - // Try hard to fold loads from bitcasted strange and non-type-safe things. We - // currently don't do any of this for big endian systems. It can be - // generalized in the future if someone is interested. - if (TD && TD->isLittleEndian()) - return FoldReinterpretLoadFromConstPtr(CE, *TD); - return 0; + + // Try hard to fold loads from bitcasted strange and non-type-safe things. + return FoldReinterpretLoadFromConstPtr(CE, DL); } -static Constant *ConstantFoldLoadInst(const LoadInst *LI, const TargetData *TD){ - if (LI->isVolatile()) return 0; - +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 0; + 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 TD, -/// otherwise TD is null. +/// 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 TargetData *TD){ + 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) { + 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); + computeKnownBits(Op1, KnownZero1, KnownOne1, DL); + if ((KnownOne1 | KnownZero0).isAllOnesValue()) { + // All the bits of Op0 that the 'and' could be masking are already zero. + return Op0; + } + if ((KnownOne0 | KnownZero1).isAllOnesValue()) { + // All the bits of Op1 that the 'and' could be masking are already zero. + return Op1; + } + + APInt KnownZero = KnownZero0 | KnownZero1; + APInt KnownOne = KnownOne0 & KnownOne1; + if ((KnownZero | KnownOne).isAllOnesValue()) { + return ConstantInt::get(Op0->getType(), KnownOne); + } + } + // 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 && TD) { + if (Opc == Instruction::Sub) { GlobalValue *GV1, *GV2; - int64_t Offs1, Offs2; - - if (IsConstantOffsetFromGlobal(Op0, GV1, Offs1, *TD)) - if (IsConstantOffsetFromGlobal(Op1, GV2, Offs2, *TD) && - 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()); + // (&GV+C1) - (&GV+C2) -> C1-C2, pointer arithmetic cannot overflow. - return ConstantInt::get(Op0->getType(), Offs1-Offs2); + // PtrToInt may change the bitwidth so we have convert to the right size + // first. + return ConstantInt::get(Op0->getType(), Offs1.zextOrTrunc(OpSize) - + Offs2.zextOrTrunc(OpSize)); } } - - return 0; + + 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 TargetData *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 0; - Type *IntPtrTy = TD->getIntPtrType(ResultTy->getContext()); + 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)))) && + !isa(GetElementPtrInst::getIndexedType( + 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], @@ -564,38 +694,57 @@ static Constant *CastGEPIndices(ArrayRef Ops, } else NewIdxs.push_back(Ops[i]); } - if (!Any) return 0; - Constant *C = - ConstantExpr::getGetElementPtr(Ops[0], NewIdxs); - if (ConstantExpr *CE = dyn_cast(C)) - if (Constant *Folded = ConstantFoldConstantExpression(CE, TD, TLI)) + if (!Any) + return nullptr; + + Constant *C = ConstantExpr::getGetElementPtr(SrcTy, Ops[0], NewIdxs); + if (ConstantExpr *CE = dyn_cast(C)) { + if (Constant *Folded = ConstantFoldConstantExpression(CE, DL, TLI)) C = Folded; + } + return C; } -/// SymbolicallyEvaluateGEP - If we can symbolically evaluate the specified GEP -/// constant expression, do so. -static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, - Type *ResultTy, const TargetData *TD, +/// Strip the pointer casts, but preserve the address space information. +static Constant* StripPtrCastKeepAS(Constant* Ptr) { + assert(Ptr->getType()->isPointerTy() && "Not a pointer type"); + PointerType *OldPtrTy = cast(Ptr->getType()); + Ptr = Ptr->stripPointerCasts(); + PointerType *NewPtrTy = cast(Ptr->getType()); + + // Preserve the address space number of the pointer. + if (NewPtrTy->getAddressSpace() != OldPtrTy->getAddressSpace()) { + NewPtrTy = NewPtrTy->getElementType()->getPointerTo( + OldPtrTy->getAddressSpace()); + Ptr = ConstantExpr::getPointerCast(Ptr, NewPtrTy); + } + return Ptr; +} + +/// 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 || !cast(Ptr->getType())->getElementType()->isSized()) - return 0; - - Type *IntPtrTy = TD->getIntPtrType(Ptr->getContext()); + if (!Ptr->getType()->getPointerElementType()->isSized() || + !Ptr->getType()->isPointerTy()) + return nullptr; + + Type *IntPtrTy = DL.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' for (unsigned i = 1, e = Ops.size(); i != e; ++i) if (!isa(Ops[i])) { - + // If this is "gep i8* Ptr, (sub 0, V)", fold this as: // "inttoptr (sub (ptrtoint Ptr), V)" - if (Ops.size() == 2 && - cast(ResultTy)->getElementType()->isIntegerTy(8)) { + if (Ops.size() == 2 && ResultElementTy->isIntegerTy(8)) { ConstantExpr *CE = dyn_cast(Ops[1]); - assert((CE == 0 || CE->getType() == IntPtrTy) && + assert((!CE || CE->getType() == IntPtrTy) && "CastGEPIndices didn't canonicalize index types!"); if (CE && CE->getOpcode() == Instruction::Sub && CE->getOperand(0)->isNullValue()) { @@ -603,23 +752,24 @@ 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 0; + return nullptr; } - - unsigned BitWidth = TD->getTypeSizeInBits(IntPtrTy); + + unsigned BitWidth = DL.getTypeSizeInBits(IntPtrTy); APInt Offset = - APInt(BitWidth, TD->getIndexedOffset(Ptr->getType(), - makeArrayRef((Value **)Ops.data() + 1, - Ops.size() - 1))); - Ptr = cast(Ptr->stripPointerCasts()); + 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. while (GEPOperator *GEP = dyn_cast(Ptr)) { - SmallVector NestedOps(GEP->op_begin()+1, GEP->op_end()); + SmallVector NestedOps(GEP->op_begin() + 1, GEP->op_end()); // Do not try the incorporate the sub-GEP if some index is not a number. bool AllConstantInt = true; @@ -632,20 +782,22 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, break; Ptr = cast(GEP->getOperand(0)); - Offset += APInt(BitWidth, - TD->getIndexedOffset(Ptr->getType(), NestedOps)); - Ptr = cast(Ptr->stripPointerCasts()); + Offset += APInt(BitWidth, DL.getIndexedOffset(Ptr->getType(), NestedOps)); + Ptr = StripPtrCastKeepAS(Ptr); } // If the base value for this address is a literal integer value, fold the // getelementptr to the resulting integer value casted to the pointer type. APInt BasePtr(BitWidth, 0); - if (ConstantExpr *CE = dyn_cast(Ptr)) - if (CE->getOpcode() == Instruction::IntToPtr) + if (ConstantExpr *CE = dyn_cast(Ptr)) { + if (CE->getOpcode() == Instruction::IntToPtr) { if (ConstantInt *Base = dyn_cast(CE->getOperand(0))) BasePtr = Base->getValue().zextOrTrunc(BitWidth); + } + } + if (Ptr->isNullValue() || BasePtr != 0) { - Constant *C = ConstantInt::get(Ptr->getContext(), Offset+BasePtr); + Constant *C = ConstantInt::get(Ptr->getContext(), Offset + BasePtr); return ConstantExpr::getIntToPtr(C, ResultTy); } @@ -654,22 +806,23 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, // This makes it easy to determine if the getelementptr is "inbounds". // Also, this helps GlobalOpt do SROA on GlobalVariables. Type *Ty = Ptr->getType(); - SmallVector NewIdxs; + assert(Ty->isPointerTy() && "Forming regular GEP of non-pointer type"); + SmallVector NewIdxs; + do { if (SequentialType *ATy = dyn_cast(Ty)) { if (ATy->isPointerTy()) { // The only pointer indexing we'll do is on the first index of the GEP. if (!NewIdxs.empty()) break; - + // Only handle pointers to sized types, not pointers to functions. if (!ATy->getElementType()->isSized()) - return 0; + return nullptr; } - + // Determine which element of the array the offset points into. - APInt ElemSize(BitWidth, TD->getTypeAllocSize(ATy->getElementType())); - IntegerType *IntPtrTy = TD->getIntPtrType(Ty->getContext()); + 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 @@ -684,10 +837,17 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, } Ty = ATy->getElementType(); } else if (StructType *STy = dyn_cast(Ty)) { + // If we end up with an offset that isn't valid for this struct type, we + // 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 = *DL.getStructLayout(STy); + if (Offset.uge(SL.getSizeInBytes())) + break; + // Determine which field of the struct the offset points into. The - // getZExtValue is at least as safe as the StructLayout API because we - // know the offset is within the struct at this point. - const StructLayout &SL = *TD->getStructLayout(STy); + // getZExtValue is fine as we've already ensured that the offset is + // within the range representable by the StructLayout API. unsigned ElIdx = SL.getElementContainingOffset(Offset.getZExtValue()); NewIdxs.push_back(ConstantInt::get(Type::getInt32Ty(Ty->getContext()), ElIdx)); @@ -697,24 +857,23 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef Ops, // We've reached some non-indexable type. break; } - } while (Ty != cast(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 // member, so we can't simplify it. if (Offset != 0) - return 0; + return nullptr; // Create a GEP. - Constant *C = - ConstantExpr::getGetElementPtr(Ptr, NewIdxs); - assert(cast(C->getType())->getElementType() == Ty && + 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 != cast(ResultTy)->getElementType()) - C = FoldBitCast(C, ResultTy, *TD); + if (Ty != ResultElementTy) + C = FoldBitCast(C, ResultTy, DL); return C; } @@ -725,17 +884,16 @@ 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 TargetData *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 = 0; + Constant *CommonValue = nullptr; for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { Value *Incoming = PN->getIncomingValue(i); @@ -745,14 +903,21 @@ Constant *llvm::ConstantFoldInstruction(Instruction *I, // all operands are constants. if (isa(Incoming)) continue; - // If the incoming value is not a constant, or is a different constant to - // the one we saw previously, then give up. + // If the incoming value is not a constant, then give up. Constant *C = dyn_cast(Incoming); - if (!C || (CommonValue && C != CommonValue)) - return 0; + if (!C) + return nullptr; + // Fold the PHI's operands. + if (ConstantExpr *NewC = dyn_cast(C)) + 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) + return nullptr; CommonValue = C; } + // If we reach here, all incoming values are the same constant or undef. return CommonValue ? CommonValue : UndefValue::get(PN->getType()); } @@ -760,56 +925,75 @@ Constant *llvm::ConstantFoldInstruction(Instruction *I, // Scan the operand list, checking to see if they are all constants, if so, // hand off to ConstantFoldInstOperands. SmallVector Ops; - for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i) - if (Constant *Op = dyn_cast(*i)) - Ops.push_back(Op); - else - return 0; // All operands not constant! + for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i) { + Constant *Op = dyn_cast(*i); + if (!Op) + return nullptr; // All operands not constant! + + // Fold the Instruction's operands. + if (ConstantExpr *NewCE = dyn_cast(Op)) + 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)) + if (InsertValueInst *IVI = dyn_cast(I)) { return ConstantExpr::getInsertValue( cast(IVI->getAggregateOperand()), cast(IVI->getInsertedValueOperand()), IVI->getIndices()); + } - if (ExtractValueInst *EVI = dyn_cast(I)) + if (ExtractValueInst *EVI = dyn_cast(I)) { return ConstantExpr::getExtractValue( cast(EVI->getAggregateOperand()), EVI->getIndices()); + } - return ConstantFoldInstOperands(I->getOpcode(), I->getType(), Ops, TD, TLI); + return ConstantFoldInstOperands(I->getOpcode(), I->getType(), Ops, DL, TLI); } -/// ConstantFoldConstantExpression - Attempt to fold the constant expression -/// using the specified TargetData. If successful, the constant result is -/// result is returned, if not, null is returned. -Constant *llvm::ConstantFoldConstantExpression(const ConstantExpr *CE, - const TargetData *TD, - const TargetLibraryInfo *TLI) { - SmallVector Ops; - for (User::const_op_iterator i = CE->op_begin(), e = CE->op_end(); - i != e; ++i) { +static Constant * +ConstantFoldConstantExpressionImpl(const ConstantExpr *CE, const DataLayout &DL, + const TargetLibraryInfo *TLI, + SmallPtrSetImpl &FoldedOps) { + SmallVector Ops; + for (User::const_op_iterator i = CE->op_begin(), e = CE->op_end(); i != e; + ++i) { Constant *NewC = cast(*i); - // Recursively fold the ConstantExpr's operands. - if (ConstantExpr *NewCE = dyn_cast(NewC)) - NewC = ConstantFoldConstantExpression(NewCE, TD, TLI); + // 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).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); +} + +/// 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 &DL, + const TargetLibraryInfo *TLI) { + SmallPtrSet 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 @@ -819,39 +1003,41 @@ Constant *llvm::ConstantFoldConstantExpression(const ConstantExpr *CE, /// information, due to only being passed an opcode and operands. Constant /// folding using this function strips this information. /// -Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy, +Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy, ArrayRef Ops, - const TargetData *TD, - const TargetLibraryInfo *TLI) { + 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 (isa(Ops[0]) || isa(Ops[1])) { + if (Constant *C = SymbolicallyEvaluateBinop(Opcode, Ops[0], Ops[1], DL)) return C; - + } + return ConstantExpr::get(Opcode, Ops[0], Ops[1]); } - + switch (Opcode) { - default: return 0; + default: return nullptr; case Instruction::ICmp: - case Instruction::FCmp: assert(0 && "Invalid for compares"); + case Instruction::FCmp: llvm_unreachable("Invalid for compares"); case Instruction::Call: if (Function *F = dyn_cast(Ops.back())) if (canConstantFoldCallTo(F)) return ConstantFoldCall(F, Ops.slice(0, Ops.size() - 1), TLI); - return 0; + return nullptr; case Instruction::PtrToInt: // 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(); - if (TD->getPointerSizeInBits() < InWidth) { - Constant *Mask = - ConstantInt::get(CE->getContext(), APInt::getLowBitsSet(InWidth, - TD->getPointerSizeInBits())); + unsigned PtrWidth = DL.getPointerTypeSizeInBits(CE->getType()); + if (PtrWidth < InWidth) { + Constant *Mask = + ConstantInt::get(CE->getContext(), + APInt::getLowBitsSet(InWidth, PtrWidth)); Input = ConstantExpr::getAnd(Input, Mask); } // Do a zext or trunc to get to the dest size. @@ -861,13 +1047,22 @@ Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy, 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(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(Ops[0])) { + if (CE->getOpcode() == Instruction::PtrToInt) { + Constant *SrcPtr = CE->getOperand(0); + 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, DL); + } + } + } return ConstantExpr::getCast(Opcode, Ops[0], DestTy); case Instruction::Trunc: @@ -879,11 +1074,10 @@ Constant *llvm::ConstantFoldInstOperands(unsigned Opcode, Type *DestTy, case Instruction::SIToFP: case Instruction::FPToUI: case Instruction::FPToSI: + 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: @@ -892,153 +1086,133 @@ 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 TargetData *TD, + Constant *Ops0, Constant *Ops1, + 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()) { - Type *IntPtrTy = TD->getIntPtrType(CE0->getContext()); + if (Ops1->isNullValue()) { if (CE0->getOpcode() == Instruction::IntToPtr) { + 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 && - 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 = 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, DL, TLI); + } } } - - if (ConstantExpr *CE1 = dyn_cast(Ops1)) { - if (TD && CE0->getOpcode() == CE1->getOpcode()) { - Type *IntPtrTy = TD->getIntPtrType(CE0->getContext()); + if (ConstantExpr *CE1 = dyn_cast(Ops1)) { + if (CE0->getOpcode() == CE1->getOpcode()) { if (CE0->getOpcode() == Instruction::IntToPtr) { + Type *IntPtrTy = DL.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), 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 && - 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 = 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), DL, TLI); + } + } } } - + // icmp eq (or x, y), 0 -> (icmp eq x, 0) & (icmp eq y, 0) // 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); - unsigned OpC = + 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); } } - + return ConstantExpr::getCompare(Predicate, Ops0, Ops1); } -/// 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. -Constant *llvm::ConstantFoldLoadThroughGEPConstantExpr(Constant *C, +/// 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) != Constant::getNullValue(CE->getOperand(1)->getType())) - return 0; // Do not allow stepping over the value! - + if (!CE->getOperand(1)->isNullValue()) + return nullptr; // Do not allow stepping over the value! + // Loop over all of the operands, tracking down which value we are - // addressing... - gep_type_iterator I = gep_type_begin(CE), E = gep_type_end(CE); - for (++I; I != E; ++I) - if (StructType *STy = dyn_cast(*I)) { - ConstantInt *CU = cast(I.getOperand()); - assert(CU->getZExtValue() < STy->getNumElements() && - "Struct index out of range!"); - unsigned El = (unsigned)CU->getZExtValue(); - if (ConstantStruct *CS = dyn_cast(C)) { - C = CS->getOperand(El); - } else if (isa(C)) { - C = Constant::getNullValue(STy->getElementType(El)); - } else if (isa(C)) { - C = UndefValue::get(STy->getElementType(El)); - } else { - return 0; - } - } else if (ConstantInt *CI = dyn_cast(I.getOperand())) { - if (ArrayType *ATy = dyn_cast(*I)) { - if (CI->getZExtValue() >= ATy->getNumElements()) - return 0; - if (ConstantArray *CA = dyn_cast(C)) - C = CA->getOperand(CI->getZExtValue()); - else if (isa(C)) - C = Constant::getNullValue(ATy->getElementType()); - else if (isa(C)) - C = UndefValue::get(ATy->getElementType()); - else - return 0; - } else if (VectorType *VTy = dyn_cast(*I)) { - if (CI->getZExtValue() >= VTy->getNumElements()) - return 0; - if (ConstantVector *CP = dyn_cast(C)) - C = CP->getOperand(CI->getZExtValue()); - else if (isa(C)) - C = Constant::getNullValue(VTy->getElementType()); - else if (isa(C)) - C = UndefValue::get(VTy->getElementType()); - else - return 0; - } else { - return 0; - } - } else { - return 0; - } + // addressing. + for (unsigned i = 2, e = CE->getNumOperands(); i != e; ++i) { + C = C->getAggregateElement(CE->getOperand(i)); + if (!C) + return nullptr; + } + return C; +} + +/// 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 + // addressing. + for (unsigned i = 0, e = Indices.size(); i != e; ++i) { + C = C->getAggregateElement(Indices[i]); + if (!C) + return nullptr; + } return C; } @@ -1047,17 +1221,30 @@ Constant *llvm::ConstantFoldLoadThroughGEPConstantExpr(Constant *C, // Constant Folding for Calls // -/// canConstantFoldCallTo - Return true if its even possible to fold a call to -/// the specified function. -bool -llvm::canConstantFoldCallTo(const Function *F) { +/// 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: + case Intrinsic::exp: + case Intrinsic::exp2: + case Intrinsic::floor: + case Intrinsic::ceil: case Intrinsic::sqrt: + case Intrinsic::pow: case Intrinsic::powi: case Intrinsic::bswap: 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: @@ -1080,17 +1267,17 @@ llvm::canConstantFoldCallTo(const Function *F) { case 0: break; } - if (!F->hasName()) return false; + if (!F->hasName()) + return false; StringRef Name = F->getName(); - + // In these cases, the check of the length is required. We don't want to // return true for a name like "cos\0blah" which strcmp would return equal to // "cos", but has length 8. switch (Name[0]) { default: return false; case 'a': - return Name == "acos" || Name == "asin" || - Name == "atan" || Name == "atan2"; + return Name == "acos" || Name == "asin" || Name == "atan" || Name =="atan2"; case 'c': return Name == "cos" || Name == "ceil" || Name == "cosf" || Name == "cosh"; case 'e': @@ -1109,55 +1296,78 @@ llvm::canConstantFoldCallTo(const Function *F) { } } -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; + 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 float/double"); - return 0; // dummy return to suppress warning + llvm_unreachable("Can only constant fold half/float/double"); + +} + +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) { + llvm_fenv_clearexcept(); + V = NativeFP(V); + if (llvm_fenv_testexcept()) { + llvm_fenv_clearexcept(); + return nullptr; + } + + return GetConstantFoldFPValue(V, Ty); } 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(); - return 0; + if (llvm_fenv_testexcept()) { + llvm_fenv_clearexcept(); + return nullptr; } - - 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 float/double"); - return 0; // dummy return to suppress warning -} -/// 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. -static Constant *ConstantFoldConvertToInt(ConstantFP *Op, bool roundTowardZero, - Type *Ty) { - assert(Op && "Called with NULL operand"); - APFloat Val(Op->getValueAPF()); + return GetConstantFoldFPValue(V, Ty); +} +/// 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. - unsigned ResultWidth = cast(Ty)->getBitWidth(); + unsigned ResultWidth = Ty->getIntegerBitWidth(); assert(ResultWidth <= 64 && "Can only constant fold conversions to 64 and 32 bit ints"); @@ -1169,129 +1379,169 @@ static Constant *ConstantFoldConvertToInt(ConstantFP *Op, bool roundTowardZero, /*isSigned=*/true, mode, &isExact); if (status != APFloat::opOK && status != APFloat::opInexact) - return 0; + return nullptr; 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 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 Operands, + const TargetLibraryInfo *TLI) { if (Operands.size() == 1) { if (ConstantFP *Op = dyn_cast(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 (!Ty->isFloatTy() && !Ty->isDoubleTy()) - return 0; + if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy()) + return nullptr; + + 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. if (Op->getValueAPF().isNaN() || Op->getValueAPF().isInfinity()) - return 0; + return nullptr; /// Currently APFloat versions of these functions do not exist, so we use /// 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 = Ty->isFloatTy() ? (double)Op->getValueAPF().convertToFloat() : - Op->getValueAPF().convertToDouble(); + double V = getValueAsDouble(Op); + + switch (IntrinsicID) { + default: break; + case Intrinsic::fabs: + return ConstantFoldFP(fabs, V, Ty); + case Intrinsic::log2: + return ConstantFoldFP(Log2, V, Ty); + case Intrinsic::log: + return ConstantFoldFP(log, V, Ty); + case Intrinsic::log10: + return ConstantFoldFP(log10, V, Ty); + case Intrinsic::exp: + return ConstantFoldFP(exp, V, Ty); + case Intrinsic::exp2: + return ConstantFoldFP(exp2, V, Ty); + case Intrinsic::floor: + return ConstantFoldFP(floor, V, Ty); + case Intrinsic::ceil: + return ConstantFoldFP(ceil, V, Ty); + } + + if (!TLI) + return nullptr; + switch (Name[0]) { case 'a': - if (Name == "acos") + if (Name == "acos" && TLI->has(LibFunc::acos)) return ConstantFoldFP(acos, V, Ty); - else if (Name == "asin") + else if (Name == "asin" && TLI->has(LibFunc::asin)) return ConstantFoldFP(asin, V, Ty); - else if (Name == "atan") + else if (Name == "atan" && TLI->has(LibFunc::atan)) return ConstantFoldFP(atan, V, Ty); break; case 'c': - if (Name == "ceil") + if (Name == "ceil" && TLI->has(LibFunc::ceil)) return ConstantFoldFP(ceil, V, Ty); - else if (Name == "cos") + else if (Name == "cos" && TLI->has(LibFunc::cos)) return ConstantFoldFP(cos, V, Ty); - else if (Name == "cosh") + else if (Name == "cosh" && TLI->has(LibFunc::cosh)) return ConstantFoldFP(cosh, V, Ty); - else if (Name == "cosf") + else if (Name == "cosf" && TLI->has(LibFunc::cosf)) return ConstantFoldFP(cos, V, Ty); break; case 'e': - if (Name == "exp") + if (Name == "exp" && TLI->has(LibFunc::exp)) return ConstantFoldFP(exp, V, Ty); - - if (Name == "exp2") { + + if (Name == "exp2" && TLI->has(LibFunc::exp2)) { // Constant fold exp2(x) as pow(2,x) in case the host doesn't have a // C99 library. return ConstantFoldBinaryFP(pow, 2.0, V, Ty); } break; case 'f': - if (Name == "fabs") + if (Name == "fabs" && TLI->has(LibFunc::fabs)) return ConstantFoldFP(fabs, V, Ty); - else if (Name == "floor") + else if (Name == "floor" && TLI->has(LibFunc::floor)) return ConstantFoldFP(floor, V, Ty); break; case 'l': - if (Name == "log" && V > 0) + if (Name == "log" && V > 0 && TLI->has(LibFunc::log)) return ConstantFoldFP(log, V, Ty); - else if (Name == "log10" && V > 0) + else if (Name == "log10" && V > 0 && TLI->has(LibFunc::log10)) return ConstantFoldFP(log10, V, Ty); - else if (F->getIntrinsicID() == Intrinsic::sqrt && - (Ty->isFloatTy() || Ty->isDoubleTy())) { + else if (IntrinsicID == Intrinsic::sqrt && + (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': - if (Name == "sin") + if (Name == "sin" && TLI->has(LibFunc::sin)) return ConstantFoldFP(sin, V, Ty); - else if (Name == "sinh") + else if (Name == "sinh" && TLI->has(LibFunc::sinh)) return ConstantFoldFP(sinh, V, Ty); - else if (Name == "sqrt" && V >= 0) + else if (Name == "sqrt" && V >= 0 && TLI->has(LibFunc::sqrt)) return ConstantFoldFP(sqrt, V, Ty); - else if (Name == "sqrtf" && V >= 0) + else if (Name == "sqrtf" && V >= 0 && TLI->has(LibFunc::sqrtf)) return ConstantFoldFP(sqrt, V, Ty); - else if (Name == "sinf") + else if (Name == "sinf" && TLI->has(LibFunc::sinf)) return ConstantFoldFP(sin, V, Ty); break; case 't': - if (Name == "tan") + if (Name == "tan" && TLI->has(LibFunc::tan)) return ConstantFoldFP(tan, V, Ty); - else if (Name == "tanh") + else if (Name == "tanh" && TLI->has(LibFunc::tanh)) return ConstantFoldFP(tanh, V, Ty); break; default: break; } - return 0; + return nullptr; } if (ConstantInt *Op = dyn_cast(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::cttz: - return ConstantInt::get(Ty, Op->getValue().countTrailingZeros()); - case Intrinsic::ctlz: - return ConstantInt::get(Ty, Op->getValue().countLeadingZeros()); case Intrinsic::convert_from_fp16: { - APFloat Val(Op->getValue()); + APFloat Val(APFloat::IEEEhalf, Op->getValue()); bool lost = false; APFloat::opStatus status = @@ -1302,78 +1552,108 @@ llvm::ConstantFoldCall(Function *F, ArrayRef Operands, 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; + return nullptr; } } - if (ConstantVector *Op = dyn_cast(Operands[0])) { - switch (F->getIntrinsicID()) { + // Support ConstantVector in case we have an Undef in the top. + if (isa(Operands[0]) || + isa(Operands[0])) { + Constant *Op = cast(Operands[0]); + switch (IntrinsicID) { default: break; case Intrinsic::x86_sse_cvtss2si: case Intrinsic::x86_sse_cvtss2si64: case Intrinsic::x86_sse2_cvtsd2si: case Intrinsic::x86_sse2_cvtsd2si64: - if (ConstantFP *FPOp = dyn_cast(Op->getOperand(0))) - return ConstantFoldConvertToInt(FPOp, /*roundTowardZero=*/false, Ty); + if (ConstantFP *FPOp = + dyn_cast_or_null(Op->getAggregateElement(0U))) + return ConstantFoldConvertToInt(FPOp->getValueAPF(), + /*roundTowardZero=*/false, Ty); case Intrinsic::x86_sse_cvttss2si: case Intrinsic::x86_sse_cvttss2si64: case Intrinsic::x86_sse2_cvttsd2si: case Intrinsic::x86_sse2_cvttsd2si64: - if (ConstantFP *FPOp = dyn_cast(Op->getOperand(0))) - return ConstantFoldConvertToInt(FPOp, /*roundTowardZero=*/true, Ty); + if (ConstantFP *FPOp = + dyn_cast_or_null(Op->getAggregateElement(0U))) + return ConstantFoldConvertToInt(FPOp->getValueAPF(), + /*roundTowardZero=*/true, Ty); } } if (isa(Operands[0])) { - if (F->getIntrinsicID() == Intrinsic::bswap) + if (IntrinsicID == Intrinsic::bswap) return Operands[0]; - return 0; + return nullptr; } - return 0; + return nullptr; } if (Operands.size() == 2) { if (ConstantFP *Op1 = dyn_cast(Operands[0])) { - if (!Ty->isFloatTy() && !Ty->isDoubleTy()) - return 0; - double Op1V = Ty->isFloatTy() ? - (double)Op1->getValueAPF().convertToFloat() : - Op1->getValueAPF().convertToDouble(); + if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy()) + return nullptr; + double Op1V = getValueAsDouble(Op1); + if (ConstantFP *Op2 = dyn_cast(Operands[1])) { if (Op2->getType() != Op1->getType()) - return 0; - - double Op2V = Ty->isFloatTy() ? - (double)Op2->getValueAPF().convertToFloat(): - Op2->getValueAPF().convertToDouble(); + return nullptr; - if (Name == "pow") + double Op2V = getValueAsDouble(Op2); + if (IntrinsicID == Intrinsic::pow) { return ConstantFoldBinaryFP(pow, Op1V, Op2V, Ty); - if (Name == "fmod") + } + if (IntrinsicID == Intrinsic::copysign) { + APFloat V1 = Op1->getValueAPF(); + APFloat V2 = Op2->getValueAPF(); + 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)) + return ConstantFoldBinaryFP(pow, Op1V, Op2V, Ty); + if (Name == "fmod" && TLI->has(LibFunc::fmod)) return ConstantFoldBinaryFP(fmod, Op1V, Op2V, Ty); - if (Name == "atan2") + if (Name == "atan2" && TLI->has(LibFunc::atan2)) return ConstantFoldBinaryFP(atan2, Op1V, Op2V, Ty); } else if (ConstantInt *Op2C = dyn_cast(Operands[1])) { - if (F->getIntrinsicID() == Intrinsic::powi && Ty->isFloatTy()) - 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->isDoubleTy()) - 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 (IntrinsicID == Intrinsic::powi && Ty->isDoubleTy()) + return ConstantFP::get(Ty->getContext(), APFloat((double)std::pow((double)Op1V, (int)Op2C->getZExtValue()))); } - return 0; + return nullptr; } - - + if (ConstantInt *Op1 = dyn_cast(Operands[0])) { if (ConstantInt *Op2 = dyn_cast(Operands[1])) { - switch (F->getIntrinsicID()) { + switch (IntrinsicID) { default: break; case Intrinsic::sadd_with_overflow: case Intrinsic::uadd_with_overflow: @@ -1383,8 +1663,8 @@ llvm::ConstantFoldCall(Function *F, ArrayRef Operands, case Intrinsic::umul_with_overflow: { APInt Res; bool Overflow; - switch (F->getIntrinsicID()) { - default: assert(0 && "Invalid case"); + switch (IntrinsicID) { + default: llvm_unreachable("Invalid case"); case Intrinsic::sadd_with_overflow: Res = Op1->getValue().sadd_ov(Op2->getValue(), Overflow); break; @@ -1405,17 +1685,95 @@ llvm::ConstantFoldCall(Function *F, ArrayRef Operands, 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(F->getReturnType()), Ops); + return ConstantStruct::get(cast(Ty), Ops); } + case Intrinsic::cttz: + if (Op2->isOne() && Op1->isZero()) // cttz(0, 1) is undef. + return UndefValue::get(Ty); + return ConstantInt::get(Ty, Op1->getValue().countTrailingZeros()); + case Intrinsic::ctlz: + if (Op2->isOne() && Op1->isZero()) // ctlz(0, 1) is undef. + return UndefValue::get(Ty); + return ConstantInt::get(Ty, Op1->getValue().countLeadingZeros()); } } - - return 0; + + return nullptr; + } + return nullptr; + } + + if (Operands.size() != 3) + return nullptr; + + if (const ConstantFP *Op1 = dyn_cast(Operands[0])) { + if (const ConstantFP *Op2 = dyn_cast(Operands[1])) { + if (const ConstantFP *Op3 = dyn_cast(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 nullptr; + } + } + } + } + } + + return nullptr; +} + +static Constant *ConstantFoldVectorCall(StringRef Name, unsigned IntrinsicID, + VectorType *VTy, + ArrayRef Operands, + const TargetLibraryInfo *TLI) { + SmallVector Result(VTy->getNumElements()); + SmallVector 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; } - return 0; + + // 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 0; + + return ConstantVector::get(Result); +} + +/// 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, + const TargetLibraryInfo *TLI) { + if (!F->hasName()) + return nullptr; + StringRef Name = F->getName(); + + Type *Ty = F->getReturnType(); + + if (VectorType *VTy = dyn_cast(Ty)) + return ConstantFoldVectorCall(Name, F->getIntrinsicID(), VTy, Operands, TLI); + + return ConstantFoldScalarCall(Name, F->getIntrinsicID(), Ty, Operands, TLI); }