//===----------------------------------------------------------------------===//
#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/Analysis/ValueTracking.h"
// Handle a vector->integer cast.
if (IntegerType *IT = dyn_cast<IntegerType>(DestTy)) {
- ConstantDataVector *CDV = dyn_cast<ConstantDataVector>(C);
- if (CDV == 0)
+ VectorType *VTy = dyn_cast<VectorType>(C->getType());
+ if (VTy == 0)
return ConstantExpr::getBitCast(C, DestTy);
- unsigned NumSrcElts = CDV->getType()->getNumElements();
-
- Type *SrcEltTy = CDV->getType()->getElementType();
+ 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.
VectorType::get(IntegerType::get(C->getContext(), FPWidth), NumSrcElts);
// Ask IR to do the conversion now that #elts line up.
C = ConstantExpr::getBitCast(C, SrcIVTy);
- CDV = cast<ConstantDataVector>(C);
}
+ 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);
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;
}
// that address spaces don't matter here since we're not going to result in
// an actual new load.
Type *MapTy;
- if (LoadTy->isFloatTy())
+ if (LoadTy->isHalfTy())
+ MapTy = Type::getInt16PtrTy(C->getContext());
+ else if (LoadTy->isFloatTy())
MapTy = Type::getInt32PtrTy(C->getContext());
else if (LoadTy->isDoubleTy())
MapTy = Type::getInt64PtrTy(C->getContext());
/// 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 DataLayout *TD){
+ Constant *Op1, const DataLayout *DL){
// SROA
// Fold (and 0xffffffff00000000, (shl x, 32)) -> shl.
// 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;
- APInt Offs1(TD->getPointerSizeInBits(), 0),
- Offs2(TD->getPointerSizeInBits(), 0);
+ unsigned PtrSize = DL->getPointerSizeInBits();
+ unsigned OpSize = DL->getTypeSizeInBits(Op0->getType());
+ APInt Offs1(PtrSize, 0), Offs2(PtrSize, 0);
- if (IsConstantOffsetFromGlobal(Op0, GV1, Offs1, *TD))
- if (IsConstantOffsetFromGlobal(Op1, GV2, Offs2, *TD) &&
+ if (IsConstantOffsetFromGlobal(Op0, GV1, Offs1, *DL))
+ if (IsConstantOffsetFromGlobal(Op1, GV2, Offs2, *DL) &&
GV1 == GV2) {
// (&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 ConstantFoldInstOperands(I->getOpcode(), I->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) {
- 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
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:
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':
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),
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
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))
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
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,
projects / oota-llvm.git / blobdiff