//
// 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/ValueTracking.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/Operator.h"
#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/FEnv.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/MathExtras.h"
-#include "llvm/Support/FEnv.h"
+#include "llvm/Target/TargetLibraryInfo.h"
#include <cerrno>
#include <cmath>
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
+/// FoldBitCast - 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) {
+ const DataLayout &TD) {
// Catch the obvious splat cases.
if (C->isNullValue() && !DestTy->isX86_MMXTy())
return Constant::getNullValue(DestTy);
if (C->isAllOnesValue() && !DestTy->isX86_MMXTy())
return Constant::getAllOnesValue(DestTy);
+ // Handle a vector->integer cast.
+ if (IntegerType *IT = dyn_cast<IntegerType>(DestTy)) {
+ VectorType *VTy = dyn_cast<VectorType>(C->getType());
+ if (VTy == 0)
+ 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<ConstantDataVector>(C);
+ if (CDV == 0)
+ 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 = TD.getTypeAllocSizeInBits(SrcEltTy);
+ APInt Result(IT->getBitWidth(), 0);
+ for (unsigned i = 0; i != NumSrcElts; ++i) {
+ Result <<= BitShift;
+ if (TD.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<VectorType>(DestTy);
if (DestVTy == 0)
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<ConstantFP>(C) || isa<ConstantInt>(C)) {
Constant *Ops = C; // don't take the address of C!
return FoldBitCast(ConstantVector::get(Ops), DestTy, TD);
}
-
+
// If this is a bitcast from constant vector -> vector, fold it.
- ConstantVector *CV = dyn_cast<ConstantVector>(C);
- if (CV == 0)
+ if (!isa<ConstantDataVector>(C) && !isa<ConstantVector>(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> <i64 0, i64 1> to <4 x i32>)
// folds to (little endian):
// <4 x i32> <i32 0, i32 0, i32 1, i32 0>
// and to (big endian):
// <4 x i32> <i32 0, i32 0, i32 0, i32 1>
-
+
// 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()) {
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.
+
+ // 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<ConstantVector>(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<ConstantVector>(C) && // FIXME: Remove ConstantVector.
+ !isa<ConstantDataVector>(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();
-
+
SmallVector<Constant*, 32> Result;
if (NumDstElt < NumSrcElt) {
// Handle: bitcast (<4 x i32> <i32 0, i32 1, i32 2, i32 3> to <2 x i64>)
Constant *Elt = Zero;
unsigned ShiftAmt = isLittleEndian ? 0 : SrcBitSize*(Ratio-1);
for (unsigned j = 0; j != Ratio; ++j) {
- Constant *Src = dyn_cast<ConstantInt>(CV->getOperand(SrcElt++));
+ Constant *Src =dyn_cast<ConstantInt>(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> <i64 0, i64 1> 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<ConstantInt>(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> <i64 0, i64 1> 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<ConstantInt>(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 and remember this piece.
+ Result.push_back(ConstantExpr::getTrunc(Elt, DstEltTy));
}
}
-
+
return ConstantVector::get(Result);
}
/// 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 &TD) {
// Trivial case, constant is the global.
if ((GV = dyn_cast<GlobalValue>(C))) {
- Offset = 0;
+ unsigned BitWidth = TD.getPointerTypeSizeInBits(GV->getType());
+ Offset = APInt(BitWidth, 0);
return true;
}
-
+
// Otherwise, if this isn't a constant expr, bail out.
ConstantExpr *CE = dyn_cast<ConstantExpr>(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<PointerType>(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<ConstantInt>(*i);
- if (!CI) return false; // Index isn't a simple constant?
- if (CI->isZero()) continue; // Not adding anything.
-
- if (StructType *ST = dyn_cast<StructType>(*GTI)) {
- // N = N + Offset
- Offset += TD.getStructLayout(ST)->getElementOffset(CI->getZExtValue());
- } else {
- SequentialType *SQT = cast<SequentialType>(*GTI);
- Offset += TD.getTypeAllocSize(SQT->getElementType())*CI->getSExtValue();
- }
- }
- return true;
- }
-
- return false;
+
+ // i32* getelementptr ([5 x i32]* @a, i32 0, i32 5)
+ GEPOperator *GEP = dyn_cast<GEPOperator>(CE);
+ if (!GEP)
+ return false;
+
+ unsigned BitWidth = TD.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))
+ return false;
+
+ // Otherwise, add any offset that our operands provide.
+ if (!GEP->accumulateConstantOffset(TD, TmpOffset))
+ return false;
+
+ Offset = TmpOffset;
+ return true;
}
/// ReadDataFromGlobal - Recursive helper to read bits out of global. C is the
/// the CurPtr buffer. TD is the target data.
static bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset,
unsigned char *CurPtr, unsigned BytesLeft,
- const TargetData &TD) {
+ const DataLayout &TD) {
assert(ByteOffset <= TD.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<ConstantAggregateZero>(C) || isa<UndefValue>(C))
return true;
-
+
if (ConstantInt *CI = dyn_cast<ConstantInt>(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 (!TD.isLittleEndian())
+ n = IntBytes - n - 1;
+ CurPtr[i] = (unsigned char)(Val >> (n * 8));
++ByteOffset;
}
return true;
}
-
+
if (ConstantFP *CFP = dyn_cast<ConstantFP>(C)) {
if (CFP->getType()->isDoubleTy()) {
C = FoldBitCast(C, Type::getInt64Ty(C->getContext()), TD);
C = FoldBitCast(C, Type::getInt32Ty(C->getContext()), TD);
return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, TD);
}
+ if (CFP->getType()->isHalfTy()){
+ C = FoldBitCast(C, Type::getInt16Ty(C->getContext()), TD);
+ return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, TD);
+ }
return false;
}
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.
!ReadDataFromGlobal(CS->getOperand(Index), ByteOffset, CurPtr,
BytesLeft, TD))
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;
// 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<ConstantArray>(C)) {
- uint64_t EltSize = TD.getTypeAllocSize(CA->getType()->getElementType());
+ if (isa<ConstantArray>(C) || isa<ConstantVector>(C) ||
+ isa<ConstantDataSequential>(C)) {
+ Type *EltTy = C->getType()->getSequentialElementType();
+ uint64_t EltSize = TD.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<ConstantVector>(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,
+ uint64_t NumElts;
+ if (ArrayType *AT = dyn_cast<ArrayType>(C->getType()))
+ NumElts = AT->getNumElements();
+ else
+ NumElts = C->getType()->getVectorNumElements();
+
+ for (; Index != NumElts; ++Index) {
+ if (!ReadDataFromGlobal(C->getAggregateElement(Index), Offset, CurPtr,
BytesLeft, TD))
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<ConstantExpr>(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() == TD.getIntPtrType(CE->getType())) {
+ return ReadDataFromGlobal(CE->getOperand(0), ByteOffset, CurPtr,
+ BytesLeft, TD);
+ }
}
// Otherwise, unknown initializer type.
}
static Constant *FoldReinterpretLoadFromConstPtr(Constant *C,
- const TargetData &TD) {
- Type *LoadTy = cast<PointerType>(C->getType())->getElementType();
+ const DataLayout &TD) {
+ PointerType *PTy = cast<PointerType>(C->getType());
+ Type *LoadTy = PTy->getElementType();
IntegerType *IntType = dyn_cast<IntegerType>(LoadTy);
-
+
// If this isn't an integer load we can't fold it directly.
if (!IntType) {
+ unsigned AS = PTy->getAddressSpace();
+
// If this is a float/double load, we can try folding it as an int32/64 load
// and then bitcast the result. This can be useful for union cases. Note
// that address spaces don't matter here since we're not going to result in
// an actual new load.
Type *MapTy;
- if (LoadTy->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(),
+ TD.getTypeAllocSizeInBits(LoadTy),
+ AS);
} else
return 0;
return FoldBitCast(Res, LoadTy, TD);
return 0;
}
-
+
unsigned BytesLoaded = (IntType->getBitWidth() + 7) / 8;
- if (BytesLoaded > 32 || BytesLoaded == 0) return 0;
-
+ if (BytesLoaded > 32 || BytesLoaded == 0)
+ return 0;
+
GlobalValue *GVal;
- int64_t Offset;
+ APInt Offset;
if (!IsConstantOffsetFromGlobal(C, GVal, Offset, TD))
return 0;
-
+
GlobalVariable *GV = dyn_cast<GlobalVariable>(GVal);
if (!GV || !GV->isConstant() || !GV->hasDefinitiveInitializer() ||
!GV->getInitializer()->getType()->isSized())
// 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 0;
+
// 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() >=
+ TD.getTypeAllocSize(GV->getInitializer()->getType()))
return UndefValue::get(IntType);
-
+
unsigned char RawBytes[32] = {0};
- if (!ReadDataFromGlobal(GV->getInitializer(), Offset, RawBytes,
+ if (!ReadDataFromGlobal(GV->getInitializer(), Offset.getZExtValue(), 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];
+ APInt ResultVal = APInt(IntType->getBitWidth(), 0);
+ if (TD.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);
/// produce if it is constant and determinable. If this is not determinable,
/// return null.
Constant *llvm::ConstantFoldLoadFromConstPtr(Constant *C,
- const TargetData *TD) {
+ const DataLayout *TD) {
// First, try the easy cases:
if (GlobalVariable *GV = dyn_cast<GlobalVariable>(C))
if (GV->isConstant() && GV->hasDefinitiveInitializer())
// If the loaded value isn't a constant expr, we can't handle it.
ConstantExpr *CE = dyn_cast<ConstantExpr>(C);
- if (!CE) return 0;
-
+ if (!CE)
+ return 0;
+
if (CE->getOpcode() == Instruction::GetElementPtr) {
- if (GlobalVariable *GV = dyn_cast<GlobalVariable>(CE->getOperand(0)))
- if (GV->isConstant() && GV->hasDefinitiveInitializer())
- if (Constant *V =
+ if (GlobalVariable *GV = dyn_cast<GlobalVariable>(CE->getOperand(0))) {
+ if (GV->isConstant() && GV->hasDefinitiveInitializer()) {
+ if (Constant *V =
ConstantFoldLoadThroughGEPConstantExpr(GV->getInitializer(), CE))
return V;
+ }
+ }
}
-
+
// 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 (TD && getConstantStringInfo(CE, Str) && !Str.empty()) {
+ unsigned StrLen = Str.size();
Type *Ty = cast<PointerType>(CE->getType())->getElementType();
unsigned NumBits = Ty->getPrimitiveSizeInBits();
// Replace load with immediate integer if the result is an integer or fp
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 =
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())
+
+ // Try hard to fold loads from bitcasted strange and non-type-safe things.
+ if (TD)
return FoldReinterpretLoadFromConstPtr(CE, *TD);
return 0;
}
-static Constant *ConstantFoldLoadInst(const LoadInst *LI, const TargetData *TD){
+static Constant *ConstantFoldLoadInst(const LoadInst *LI, const DataLayout *TD){
if (LI->isVolatile()) return 0;
-
+
if (Constant *C = dyn_cast<Constant>(LI->getOperand(0)))
return ConstantFoldLoadFromConstPtr(C, TD);
/// SymbolicallyEvaluateBinop - 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 && DL) {
+ unsigned BitWidth = DL->getTypeSizeInBits(Op0->getType()->getScalarType());
+ APInt KnownZero0(BitWidth, 0), KnownOne0(BitWidth, 0);
+ APInt KnownZero1(BitWidth, 0), KnownOne1(BitWidth, 0);
+ ComputeMaskedBits(Op0, KnownZero0, KnownOne0, DL);
+ ComputeMaskedBits(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 && DL) {
GlobalValue *GV1, *GV2;
- int64_t Offs1, Offs2;
-
- if (IsConstantOffsetFromGlobal(Op0, GV1, Offs1, *TD))
- if (IsConstantOffsetFromGlobal(Op1, GV2, Offs2, *TD) &&
+ 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;
}
/// explicitly cast them so that they aren't implicitly casted by the
/// getelementptr.
static Constant *CastGEPIndices(ArrayRef<Constant *> Ops,
- Type *ResultTy, const TargetData *TD,
+ Type *ResultTy, const DataLayout *TD,
const TargetLibraryInfo *TLI) {
- if (!TD) return 0;
- Type *IntPtrTy = TD->getIntPtrType(ResultTy->getContext());
+ if (!TD)
+ return 0;
+
+ Type *IntPtrTy = TD->getIntPtrType(ResultTy);
bool Any = false;
SmallVector<Constant*, 32> NewIdxs;
for (unsigned i = 1, e = Ops.size(); i != e; ++i) {
if ((i == 1 ||
- !isa<StructType>(GetElementPtrInst::getIndexedType(Ops[0]->getType(),
- Ops.slice(1, i-1)))) &&
+ !isa<StructType>(GetElementPtrInst::getIndexedType(
+ Ops[0]->getType(),
+ Ops.slice(1, i - 1)))) &&
Ops[i]->getType() != IntPtrTy) {
Any = true;
NewIdxs.push_back(ConstantExpr::getCast(CastInst::getCastOpcode(Ops[i],
} else
NewIdxs.push_back(Ops[i]);
}
- if (!Any) return 0;
- Constant *C =
- ConstantExpr::getGetElementPtr(Ops[0], NewIdxs);
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
+ if (!Any)
+ return 0;
+
+ Constant *C = ConstantExpr::getGetElementPtr(Ops[0], NewIdxs);
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
if (Constant *Folded = ConstantFoldConstantExpression(CE, TD, TLI))
C = Folded;
+ }
+
return C;
}
+/// 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<PointerType>(Ptr->getType());
+ Ptr = cast<Constant>(Ptr->stripPointerCasts());
+ PointerType *NewPtrTy = cast<PointerType>(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;
+}
+
/// SymbolicallyEvaluateGEP - If we can symbolically evaluate the specified GEP
/// constant expression, do so.
static Constant *SymbolicallyEvaluateGEP(ArrayRef<Constant *> Ops,
- Type *ResultTy, const TargetData *TD,
+ Type *ResultTy, const DataLayout *TD,
const TargetLibraryInfo *TLI) {
Constant *Ptr = Ops[0];
- if (!TD || !cast<PointerType>(Ptr->getType())->getElementType()->isSized() ||
+ if (!TD || !Ptr->getType()->getPointerElementType()->isSized() ||
!Ptr->getType()->isPointerTy())
return 0;
-
- Type *IntPtrTy = TD->getIntPtrType(Ptr->getContext());
+
+ Type *IntPtrTy = TD->getIntPtrType(Ptr->getType());
+ Type *ResultElementTy = ResultTy->getPointerElementType();
// If this is a constant expr gep that is effectively computing an
// "offsetof", fold it into 'cast int Size to T*' instead of 'gep 0, 0, 12'
for (unsigned i = 1, e = Ops.size(); i != e; ++i)
if (!isa<ConstantInt>(Ops[i])) {
-
+
// If this is "gep i8* Ptr, (sub 0, V)", fold this as:
// "inttoptr (sub (ptrtoint Ptr), V)"
- if (Ops.size() == 2 &&
- cast<PointerType>(ResultTy)->getElementType()->isIntegerTy(8)) {
+ if (Ops.size() == 2 && ResultElementTy->isIntegerTy(8)) {
ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[1]);
assert((CE == 0 || CE->getType() == IntPtrTy) &&
"CastGEPIndices didn't canonicalize index types!");
}
return 0;
}
-
+
unsigned BitWidth = TD->getTypeSizeInBits(IntPtrTy);
APInt Offset =
APInt(BitWidth, TD->getIndexedOffset(Ptr->getType(),
- makeArrayRef((Value **)Ops.data() + 1,
+ makeArrayRef((Value *const*)
+ Ops.data() + 1,
Ops.size() - 1)));
- Ptr = cast<Constant>(Ptr->stripPointerCasts());
+ Ptr = StripPtrCastKeepAS(Ptr);
// If this is a GEP of a GEP, fold it all into a single GEP.
while (GEPOperator *GEP = dyn_cast<GEPOperator>(Ptr)) {
- SmallVector<Value *, 4> NestedOps(GEP->op_begin()+1, GEP->op_end());
+ SmallVector<Value *, 4> 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;
Ptr = cast<Constant>(GEP->getOperand(0));
Offset += APInt(BitWidth,
TD->getIndexedOffset(Ptr->getType(), NestedOps));
- Ptr = cast<Constant>(Ptr->stripPointerCasts());
+ 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<ConstantExpr>(Ptr))
- if (CE->getOpcode() == Instruction::IntToPtr)
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ptr)) {
+ if (CE->getOpcode() == Instruction::IntToPtr) {
if (ConstantInt *Base = dyn_cast<ConstantInt>(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);
}
// This makes it easy to determine if the getelementptr is "inbounds".
// Also, this helps GlobalOpt do SROA on GlobalVariables.
Type *Ty = Ptr->getType();
- SmallVector<Constant*, 32> NewIdxs;
+ assert(Ty->isPointerTy() && "Forming regular GEP of non-pointer type");
+ SmallVector<Constant *, 32> NewIdxs;
+
do {
if (SequentialType *ATy = dyn_cast<SequentialType>(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;
}
-
+
// Determine which element of the array the offset points into.
APInt ElemSize(BitWidth, TD->getTypeAllocSize(ATy->getElementType()));
- IntegerType *IntPtrTy = TD->getIntPtrType(Ty->getContext());
if (ElemSize == 0)
// The element size is 0. This may be [0 x Ty]*, so just use a zero
// index for this level and proceed to the next level to see if it can
}
Ty = ATy->getElementType();
} else if (StructType *STy = dyn_cast<StructType>(Ty)) {
- // 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.
+ // 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 = *TD->getStructLayout(STy);
+ if (Offset.uge(SL.getSizeInBytes()))
+ break;
+
+ // Determine which field of the struct the offset points into. The
+ // 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));
// We've reached some non-indexable type.
break;
}
- } while (Ty != cast<PointerType>(ResultTy)->getElementType());
+ } while (Ty != ResultElementTy);
// If we haven't used up the entire offset by descending the static
// type, then the offset is pointing into the middle of an indivisible
return 0;
// Create a GEP.
- Constant *C =
- ConstantExpr::getGetElementPtr(Ptr, NewIdxs);
- assert(cast<PointerType>(C->getType())->getElementType() == Ty &&
+ Constant *C = ConstantExpr::getGetElementPtr(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<PointerType>(ResultTy)->getElementType())
+ if (Ty != ResultElementTy)
C = FoldBitCast(C, ResultTy, *TD);
return C;
/// 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,
+ const DataLayout *TD,
const TargetLibraryInfo *TLI) {
// Handle PHI nodes quickly here...
if (PHINode *PN = dyn_cast<PHINode>(I)) {
// all operands are constants.
if (isa<UndefValue>(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<Constant>(Incoming);
- if (!C || (CommonValue && C != CommonValue))
+ if (!C)
+ return 0;
+ // Fold the PHI's operands.
+ if (ConstantExpr *NewC = dyn_cast<ConstantExpr>(C))
+ C = ConstantFoldConstantExpression(NewC, TD, TLI);
+ // If the incoming value is a different constant to
+ // the one we saw previously, then give up.
+ if (CommonValue && C != CommonValue)
return 0;
CommonValue = C;
}
+
// If we reach here, all incoming values are the same constant or undef.
return CommonValue ? CommonValue : UndefValue::get(PN->getType());
}
// Scan the operand list, checking to see if they are all constants, if so,
// hand off to ConstantFoldInstOperands.
SmallVector<Constant*, 8> Ops;
- for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i)
- if (Constant *Op = dyn_cast<Constant>(*i))
- Ops.push_back(Op);
- else
+ for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i) {
+ Constant *Op = dyn_cast<Constant>(*i);
+ if (!Op)
return 0; // All operands not constant!
+ // Fold the Instruction's operands.
+ if (ConstantExpr *NewCE = dyn_cast<ConstantExpr>(Op))
+ Op = ConstantFoldConstantExpression(NewCE, TD, TLI);
+
+ Ops.push_back(Op);
+ }
+
if (const CmpInst *CI = dyn_cast<CmpInst>(I))
return ConstantFoldCompareInstOperands(CI->getPredicate(), Ops[0], Ops[1],
TD, TLI);
-
+
if (const LoadInst *LI = dyn_cast<LoadInst>(I))
return ConstantFoldLoadInst(LI, TD);
- if (InsertValueInst *IVI = dyn_cast<InsertValueInst>(I))
+ if (InsertValueInst *IVI = dyn_cast<InsertValueInst>(I)) {
return ConstantExpr::getInsertValue(
cast<Constant>(IVI->getAggregateOperand()),
cast<Constant>(IVI->getInsertedValueOperand()),
IVI->getIndices());
+ }
- if (ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(I))
+ if (ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(I)) {
return ConstantExpr::getExtractValue(
cast<Constant>(EVI->getAggregateOperand()),
EVI->getIndices());
+ }
return ConstantFoldInstOperands(I->getOpcode(), I->getType(), Ops, TD, 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<Constant*, 8> 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 *TD,
+ const TargetLibraryInfo *TLI,
+ SmallPtrSet<ConstantExpr *, 4> &FoldedOps) {
+ SmallVector<Constant *, 8> Ops;
+ for (User::const_op_iterator i = CE->op_begin(), e = CE->op_end(); i != e;
+ ++i) {
Constant *NewC = cast<Constant>(*i);
- // Recursively fold the ConstantExpr's operands.
- if (ConstantExpr *NewCE = dyn_cast<ConstantExpr>(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<ConstantExpr>(NewC)) {
+ if (FoldedOps.insert(NewCE))
+ NewC = ConstantFoldConstantExpressionImpl(NewCE, TD, TLI, FoldedOps);
+ }
Ops.push_back(NewC);
}
return ConstantFoldInstOperands(CE->getOpcode(), CE->getType(), Ops, TD, TLI);
}
+/// ConstantFoldConstantExpression - 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 TargetLibraryInfo *TLI) {
+ SmallPtrSet<ConstantExpr *, 4> FoldedOps;
+ return ConstantFoldConstantExpressionImpl(CE, TD, TLI, FoldedOps);
+}
+
/// ConstantFoldInstOperands - 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
/// 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<Constant *> Ops,
- const TargetData *TD,
- const TargetLibraryInfo *TLI) {
+ const DataLayout *TD,
+ const TargetLibraryInfo *TLI) {
// Handle easy binops first.
if (Instruction::isBinaryOp(Opcode)) {
- if (isa<ConstantExpr>(Ops[0]) || isa<ConstantExpr>(Ops[1]))
+ if (isa<ConstantExpr>(Ops[0]) || isa<ConstantExpr>(Ops[1])) {
if (Constant *C = SymbolicallyEvaluateBinop(Opcode, Ops[0], Ops[1], TD))
return C;
-
+ }
+
return ConstantExpr::get(Opcode, Ops[0], Ops[1]);
}
-
+
switch (Opcode) {
default: return 0;
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<Function>(Ops.back()))
if (canConstantFoldCallTo(F))
if (TD && 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 = TD->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.
return ConstantExpr::getCast(Opcode, Ops[0], DestTy);
case Instruction::IntToPtr:
// If the input is a ptrtoint, turn the pair into a ptr to ptr bitcast if
- // the int size is >= the ptr size. This requires knowing the width of a
- // pointer, so it can't be done in ConstantExpr::getCast.
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[0]))
- if (TD &&
- TD->getPointerSizeInBits() <= CE->getType()->getScalarSizeInBits() &&
- CE->getOpcode() == Instruction::PtrToInt)
- return FoldBitCast(CE->getOperand(0), DestTy, *TD);
+ // the int size is >= the ptr size and the address spaces are the same.
+ // This requires knowing the width of a pointer, so it can't be done in
+ // ConstantExpr::getCast.
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[0])) {
+ if (TD && CE->getOpcode() == Instruction::PtrToInt) {
+ Constant *SrcPtr = CE->getOperand(0);
+ unsigned SrcPtrSize = TD->getPointerTypeSizeInBits(SrcPtr->getType());
+ unsigned MidIntSize = CE->getType()->getScalarSizeInBits();
+
+ if (MidIntSize >= SrcPtrSize) {
+ unsigned SrcAS = SrcPtr->getType()->getPointerAddressSpace();
+ if (SrcAS == DestTy->getPointerAddressSpace())
+ return FoldBitCast(CE->getOperand(0), DestTy, *TD);
+ }
+ }
+ }
return ConstantExpr::getCast(Opcode, Ops[0], DestTy);
case Instruction::Trunc:
case Instruction::SIToFP:
case Instruction::FPToUI:
case Instruction::FPToSI:
+ case Instruction::AddrSpaceCast:
return ConstantExpr::getCast(Opcode, Ops[0], DestTy);
case Instruction::BitCast:
if (TD)
return C;
if (Constant *C = SymbolicallyEvaluateGEP(Ops, DestTy, TD, TLI))
return C;
-
+
return ConstantExpr::getGetElementPtr(Ops[0], Ops.slice(1));
}
}
/// 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 *TD,
const TargetLibraryInfo *TLI) {
// fold: icmp (inttoptr x), null -> icmp x, 0
// fold: icmp (ptrtoint x), 0 -> icmp x, null
// around to know if bit truncation is happening.
if (ConstantExpr *CE0 = dyn_cast<ConstantExpr>(Ops0)) {
if (TD && Ops1->isNullValue()) {
- Type *IntPtrTy = TD->getIntPtrType(CE0->getContext());
if (CE0->getOpcode() == Instruction::IntToPtr) {
+ Type *IntPtrTy = TD->getIntPtrType(CE0->getType());
// Convert the integer value to the right size to ensure we get the
// proper extension or truncation.
Constant *C = ConstantExpr::getIntegerCast(CE0->getOperand(0),
Constant *Null = Constant::getNullValue(C->getType());
return ConstantFoldCompareInstOperands(Predicate, C, Null, TD, 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 = TD->getIntPtrType(CE0->getOperand(0)->getType());
+ if (CE0->getType() == IntPtrTy) {
+ Constant *C = CE0->getOperand(0);
+ Constant *Null = Constant::getNullValue(C->getType());
+ return ConstantFoldCompareInstOperands(Predicate, C, Null, TD, TLI);
+ }
}
}
-
+
if (ConstantExpr *CE1 = dyn_cast<ConstantExpr>(Ops1)) {
if (TD && CE0->getOpcode() == CE1->getOpcode()) {
- Type *IntPtrTy = TD->getIntPtrType(CE0->getContext());
-
if (CE0->getOpcode() == Instruction::IntToPtr) {
+ Type *IntPtrTy = TD->getIntPtrType(CE0->getType());
+
// Convert the integer value to the right size to ensure we get the
// proper extension or truncation.
Constant *C0 = ConstantExpr::getIntegerCast(CE0->getOperand(0),
// Only do this transformation if the int is intptrty in size, otherwise
// there is a truncation or extension that we aren't modeling.
- if ((CE0->getOpcode() == Instruction::PtrToInt &&
- CE0->getType() == IntPtrTy &&
- CE0->getOperand(0)->getType() == CE1->getOperand(0)->getType()))
- return ConstantFoldCompareInstOperands(Predicate, CE0->getOperand(0),
- CE1->getOperand(0), TD, TLI);
+ if (CE0->getOpcode() == Instruction::PtrToInt) {
+ Type *IntPtrTy = TD->getIntPtrType(CE0->getOperand(0)->getType());
+ if (CE0->getType() == IntPtrTy &&
+ CE0->getOperand(0)->getType() == CE1->getOperand(0)->getType()) {
+ return ConstantFoldCompareInstOperands(Predicate,
+ CE0->getOperand(0),
+ CE1->getOperand(0),
+ TD,
+ TLI);
+ }
+ }
}
}
-
+
// 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 =
+ Constant *LHS =
ConstantFoldCompareInstOperands(Predicate, CE0->getOperand(0), Ops1,
TD, TLI);
- Constant *RHS =
+ Constant *RHS =
ConstantFoldCompareInstOperands(Predicate, CE0->getOperand(1), Ops1,
TD, TLI);
- unsigned OpC =
+ unsigned OpC =
Predicate == ICmpInst::ICMP_EQ ? Instruction::And : Instruction::Or;
Constant *Ops[] = { LHS, RHS };
return ConstantFoldInstOperands(OpC, LHS->getType(), Ops, TD, 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,
+Constant *llvm::ConstantFoldLoadThroughGEPConstantExpr(Constant *C,
ConstantExpr *CE) {
- if (CE->getOperand(1) != Constant::getNullValue(CE->getOperand(1)->getType()))
+ if (!CE->getOperand(1)->isNullValue())
return 0; // 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<StructType>(*I)) {
- ConstantInt *CU = cast<ConstantInt>(I.getOperand());
- assert(CU->getZExtValue() < STy->getNumElements() &&
- "Struct index out of range!");
- unsigned El = (unsigned)CU->getZExtValue();
- if (ConstantStruct *CS = dyn_cast<ConstantStruct>(C)) {
- C = CS->getOperand(El);
- } else if (isa<ConstantAggregateZero>(C)) {
- C = Constant::getNullValue(STy->getElementType(El));
- } else if (isa<UndefValue>(C)) {
- C = UndefValue::get(STy->getElementType(El));
- } else {
- return 0;
- }
- } else if (ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand())) {
- if (ArrayType *ATy = dyn_cast<ArrayType>(*I)) {
- if (CI->getZExtValue() >= ATy->getNumElements())
- return 0;
- if (ConstantArray *CA = dyn_cast<ConstantArray>(C))
- C = CA->getOperand(CI->getZExtValue());
- else if (isa<ConstantAggregateZero>(C))
- C = Constant::getNullValue(ATy->getElementType());
- else if (isa<UndefValue>(C))
- C = UndefValue::get(ATy->getElementType());
- else
- return 0;
- } else if (VectorType *VTy = dyn_cast<VectorType>(*I)) {
- if (CI->getZExtValue() >= VTy->getNumElements())
- return 0;
- if (ConstantVector *CP = dyn_cast<ConstantVector>(C))
- C = CP->getOperand(CI->getZExtValue());
- else if (isa<ConstantAggregateZero>(C))
- C = Constant::getNullValue(VTy->getElementType());
- else if (isa<UndefValue>(C))
- C = UndefValue::get(VTy->getElementType());
- else
- return 0;
- } else {
- return 0;
- }
- } else {
+ // addressing.
+ for (unsigned i = 2, e = CE->getNumOperands(); i != e; ++i) {
+ C = C->getAggregateElement(CE->getOperand(i));
+ if (C == 0)
return 0;
- }
+ }
+ 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.
+Constant *llvm::ConstantFoldLoadThroughGEPIndices(Constant *C,
+ ArrayRef<Constant*> 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 == 0)
+ return 0;
+ }
return C;
}
/// canConstantFoldCallTo - Return true if its even possible to fold a call to
/// the specified function.
-bool
-llvm::canConstantFoldCallTo(const Function *F) {
+bool llvm::canConstantFoldCallTo(const Function *F) {
switch (F->getIntrinsicID()) {
+ case Intrinsic::fabs:
+ case Intrinsic::log:
+ case Intrinsic::log2:
+ case Intrinsic::log10:
+ case Intrinsic::exp:
+ case Intrinsic::exp2:
+ case Intrinsic::floor:
case Intrinsic::sqrt:
case Intrinsic::pow:
case Intrinsic::powi:
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':
}
}
-static Constant *ConstantFoldFP(double (*NativeFP)(double), double V,
+static Constant *ConstantFoldFP(double (*NativeFP)(double), double V,
Type *Ty) {
sys::llvm_fenv_clearexcept();
V = NativeFP(V);
sys::llvm_fenv_clearexcept();
return 0;
}
-
+
+ if (Ty->isHalfTy()) {
+ APFloat APF(V);
+ bool unused;
+ APF.convert(APFloat::IEEEhalf, APFloat::rmNearestTiesToEven, &unused);
+ return ConstantFP::get(Ty->getContext(), APF);
+ }
if (Ty->isFloatTy())
return ConstantFP::get(Ty->getContext(), APFloat((float)V));
if (Ty->isDoubleTy())
return ConstantFP::get(Ty->getContext(), APFloat(V));
- llvm_unreachable("Can only constant fold float/double");
+ llvm_unreachable("Can only constant fold half/float/double");
}
static Constant *ConstantFoldBinaryFP(double (*NativeFP)(double, double),
sys::llvm_fenv_clearexcept();
return 0;
}
-
+
+ if (Ty->isHalfTy()) {
+ APFloat APF(V);
+ bool unused;
+ APF.convert(APFloat::IEEEhalf, APFloat::rmNearestTiesToEven, &unused);
+ return ConstantFP::get(Ty->getContext(), APF);
+ }
if (Ty->isFloatTy())
return ConstantFP::get(Ty->getContext(), APFloat((float)V));
if (Ty->isDoubleTy())
return ConstantFP::get(Ty->getContext(), APFloat(V));
- llvm_unreachable("Can only constant fold float/double");
+ llvm_unreachable("Can only constant fold half/float/double");
}
/// ConstantFoldConvertToInt - Attempt to an SSE floating point to integer
/// 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());
-
+static Constant *ConstantFoldConvertToInt(const APFloat &Val,
+ bool roundTowardZero, Type *Ty) {
// All of these conversion intrinsics form an integer of at most 64bits.
- unsigned ResultWidth = cast<IntegerType>(Ty)->getBitWidth();
+ unsigned ResultWidth = Ty->getIntegerBitWidth();
assert(ResultWidth <= 64 &&
"Can only constant fold conversions to 64 and 32 bit ints");
Constant *
llvm::ConstantFoldCall(Function *F, ArrayRef<Constant *> Operands,
const TargetLibraryInfo *TLI) {
- if (!F->hasName()) return 0;
+ if (!F->hasName())
+ return 0;
StringRef Name = F->getName();
Type *Ty = F->getReturnType();
if (!TLI)
return 0;
- if (!Ty->isFloatTy() && !Ty->isDoubleTy())
+ if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy())
return 0;
/// We only fold functions with finite arguments. Folding NaN and inf is
/// 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;
+ if (Ty->isFloatTy())
+ V = Op->getValueAPF().convertToFloat();
+ else if (Ty->isDoubleTy())
+ V = Op->getValueAPF().convertToDouble();
+ else {
+ bool unused;
+ APFloat APF = Op->getValueAPF();
+ APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &unused);
+ V = APF.convertToDouble();
+ }
+
+ switch (F->getIntrinsicID()) {
+ 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
+ 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);
+ }
+
switch (Name[0]) {
case 'a':
if (Name == "acos" && TLI->has(LibFunc::acos))
case 'e':
if (Name == "exp" && TLI->has(LibFunc::exp))
return ConstantFoldFP(exp, V, Ty);
-
+
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.
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())) {
+ (Ty->isHalfTy() || Ty->isFloatTy() || Ty->isDoubleTy())) {
if (V >= -0.0)
return ConstantFoldFP(sqrt, V, Ty);
else // Undefined
case Intrinsic::ctpop:
return ConstantInt::get(Ty, Op->getValue().countPopulation());
case Intrinsic::convert_from_fp16: {
- APFloat Val(Op->getValue());
+ APFloat Val(APFloat::IEEEhalf, Op->getValue());
bool lost = false;
APFloat::opStatus status =
}
}
- if (ConstantVector *Op = dyn_cast<ConstantVector>(Operands[0])) {
+ // Support ConstantVector in case we have an Undef in the top.
+ if (isa<ConstantVector>(Operands[0]) ||
+ isa<ConstantDataVector>(Operands[0])) {
+ Constant *Op = cast<Constant>(Operands[0]);
switch (F->getIntrinsicID()) {
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<ConstantFP>(Op->getOperand(0)))
- return ConstantFoldConvertToInt(FPOp, /*roundTowardZero=*/false, Ty);
+ if (ConstantFP *FPOp =
+ dyn_cast_or_null<ConstantFP>(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<ConstantFP>(Op->getOperand(0)))
- return ConstantFoldConvertToInt(FPOp, /*roundTowardZero=*/true, Ty);
+ if (ConstantFP *FPOp =
+ dyn_cast_or_null<ConstantFP>(Op->getAggregateElement(0U)))
+ return ConstantFoldConvertToInt(FPOp->getValueAPF(),
+ /*roundTowardZero=*/true, Ty);
}
}
if (Operands.size() == 2) {
if (ConstantFP *Op1 = dyn_cast<ConstantFP>(Operands[0])) {
- if (!Ty->isFloatTy() && !Ty->isDoubleTy())
+ if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy())
return 0;
- double Op1V = Ty->isFloatTy() ?
- (double)Op1->getValueAPF().convertToFloat() :
- Op1->getValueAPF().convertToDouble();
+ double Op1V;
+ if (Ty->isFloatTy())
+ Op1V = Op1->getValueAPF().convertToFloat();
+ else if (Ty->isDoubleTy())
+ Op1V = Op1->getValueAPF().convertToDouble();
+ else {
+ bool unused;
+ APFloat APF = Op1->getValueAPF();
+ APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &unused);
+ Op1V = APF.convertToDouble();
+ }
+
if (ConstantFP *Op2 = dyn_cast<ConstantFP>(Operands[1])) {
if (Op2->getType() != Op1->getType())
return 0;
- double Op2V = Ty->isFloatTy() ?
- (double)Op2->getValueAPF().convertToFloat():
- Op2->getValueAPF().convertToDouble();
+ double Op2V;
+ if (Ty->isFloatTy())
+ Op2V = Op2->getValueAPF().convertToFloat();
+ else if (Ty->isDoubleTy())
+ Op2V = Op2->getValueAPF().convertToDouble();
+ else {
+ bool unused;
+ APFloat APF = Op2->getValueAPF();
+ APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &unused);
+ Op2V = APF.convertToDouble();
+ }
if (F->getIntrinsicID() == Intrinsic::pow) {
return ConstantFoldBinaryFP(pow, Op1V, Op2V, Ty);
if (Name == "atan2" && TLI->has(LibFunc::atan2))
return ConstantFoldBinaryFP(atan2, Op1V, Op2V, Ty);
} else if (ConstantInt *Op2C = dyn_cast<ConstantInt>(Operands[1])) {
+ if (F->getIntrinsicID() == Intrinsic::powi && Ty->isHalfTy())
+ return ConstantFP::get(F->getContext(),
+ APFloat((float)std::pow((float)Op1V,
+ (int)Op2C->getZExtValue())));
if (F->getIntrinsicID() == Intrinsic::powi && Ty->isFloatTy())
return ConstantFP::get(F->getContext(),
APFloat((float)std::pow((float)Op1V,
}
return 0;
}
-
+
if (ConstantInt *Op1 = dyn_cast<ConstantInt>(Operands[0])) {
if (ConstantInt *Op2 = dyn_cast<ConstantInt>(Operands[1])) {
switch (F->getIntrinsicID()) {
APInt Res;
bool Overflow;
switch (F->getIntrinsicID()) {
- default: assert(0 && "Invalid case");
+ default: llvm_unreachable("Invalid case");
case Intrinsic::sadd_with_overflow:
Res = Op1->getValue().sadd_ov(Op2->getValue(), Overflow);
break;
return ConstantStruct::get(cast<StructType>(F->getReturnType()), Ops);
}
case Intrinsic::cttz:
- // FIXME: This should check for Op2 == 1, and become unreachable if
- // Op1 == 0.
+ if (Op2->isOne() && Op1->isZero()) // cttz(0, 1) is undef.
+ return UndefValue::get(Ty);
return ConstantInt::get(Ty, Op1->getValue().countTrailingZeros());
case Intrinsic::ctlz:
- // FIXME: This should check for Op2 == 1, and become unreachable if
- // Op1 == 0.
+ if (Op2->isOne() && Op1->isZero()) // ctlz(0, 1) is undef.
+ return UndefValue::get(Ty);
return ConstantInt::get(Ty, Op1->getValue().countLeadingZeros());
}
}
-
+
return 0;
}
return 0;