#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/ADT/SmallVector.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/MathExtras.h"
+#include "llvm/Support/FEnv.h"
#include <cerrno>
#include <cmath>
using namespace llvm;
// 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, 1), DestTy, TD);
+ return FoldBitCast(ConstantVector::get(Ops), DestTy, TD);
}
// If this is a bitcast from constant vector -> vector, fold it.
// 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->isFloatingPoint()) {
+ if (DstEltTy->isFloatingPointTy()) {
// Fold to an vector of integers with same size as our FP type.
unsigned FPWidth = DstEltTy->getPrimitiveSizeInBits();
const Type *DestIVTy =
// Okay, we know the destination is integer, if the input is FP, convert
// it to integer first.
- if (SrcEltTy->isFloatingPoint()) {
+ if (SrcEltTy->isFloatingPointTy()) {
unsigned FPWidth = SrcEltTy->getPrimitiveSizeInBits();
const Type *SrcIVTy =
VectorType::get(IntegerType::get(C->getContext(), FPWidth), NumSrcElt);
}
}
- return ConstantVector::get(Result.data(), Result.size());
+ return ConstantVector::get(Result);
}
i != e; ++i, ++GTI) {
ConstantInt *CI = dyn_cast<ConstantInt>(*i);
if (!CI) return false; // Index isn't a simple constant?
- if (CI->getZExtValue() == 0) continue; // Not adding anything.
+ if (CI->isZero()) continue; // Not adding anything.
if (const StructType *ST = dyn_cast<StructType>(*GTI)) {
// N = N + Offset
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);
+ }
+
// Otherwise, unknown initializer type.
return false;
}
MapTy = Type::getInt32PtrTy(C->getContext());
else if (LoadTy->isDoubleTy())
MapTy = Type::getInt64PtrTy(C->getContext());
- else if (isa<VectorType>(LoadTy)) {
+ else if (LoadTy->isVectorTy()) {
MapTy = IntegerType::get(C->getContext(),
TD.getTypeAllocSizeInBits(LoadTy));
MapTy = PointerType::getUnqual(MapTy);
APInt ResultVal = APInt(IntType->getBitWidth(), RawBytes[BytesLoaded-1]);
for (unsigned i = 1; i != BytesLoaded; ++i) {
ResultVal <<= 8;
- ResultVal |= APInt(IntType->getBitWidth(), RawBytes[BytesLoaded-1-i]);
+ ResultVal |= RawBytes[BytesLoaded-1-i];
}
return ConstantInt::get(IntType->getContext(), ResultVal);
unsigned StrLen = Str.length();
const Type *Ty = cast<PointerType>(CE->getType())->getElementType();
unsigned NumBits = Ty->getPrimitiveSizeInBits();
- // Replace LI with immediate integer store.
- if ((NumBits >> 3) == StrLen + 1) {
+ // Replace load with immediate integer if the result is an integer or fp
+ // value.
+ if ((NumBits >> 3) == StrLen + 1 && (NumBits & 7) == 0 &&
+ (isa<IntegerType>(Ty) || Ty->isFloatingPointTy())) {
APInt StrVal(NumBits, 0);
APInt SingleChar(NumBits, 0);
if (TD->isLittleEndian()) {
SingleChar = 0;
StrVal = (StrVal << 8) | SingleChar;
}
- return ConstantInt::get(CE->getContext(), StrVal);
+
+ 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<GlobalVariable>(CE->getUnderlyingObject())){
+ if (GlobalVariable *GV =
+ dyn_cast<GlobalVariable>(GetUnderlyingObject(CE, TD))) {
if (GV->isConstant() && GV->hasDefinitiveInitializer()) {
const Type *ResTy = cast<PointerType>(C->getType())->getElementType();
if (GV->getInitializer()->isNullValue())
for (unsigned i = 1; i != NumOps; ++i) {
if ((i == 1 ||
!isa<StructType>(GetElementPtrInst::getIndexedType(Ops[0]->getType(),
- reinterpret_cast<Value *const *>(Ops+1),
+ reinterpret_cast<Value *const *>(Ops+1),
i-1))) &&
Ops[i]->getType() != IntPtrTy) {
Any = true;
Constant *Ptr = Ops[0];
if (!TD || !cast<PointerType>(Ptr->getType())->getElementType()->isSized())
return 0;
-
- unsigned BitWidth =
- TD->getTypeSizeInBits(TD->getIntPtrType(Ptr->getContext()));
- APInt BasePtr(BitWidth, 0);
- bool BaseIsInt = true;
- if (!Ptr->isNullValue()) {
- // If this is a inttoptr from a constant int, we can fold this as the base,
- // otherwise we can't.
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ptr))
- if (CE->getOpcode() == Instruction::IntToPtr)
- if (ConstantInt *Base = dyn_cast<ConstantInt>(CE->getOperand(0))) {
- BasePtr = Base->getValue();
- BasePtr.zextOrTrunc(BitWidth);
- }
-
- if (BasePtr == 0)
- BaseIsInt = false;
- }
+
+ const Type *IntPtrTy = TD->getIntPtrType(Ptr->getContext());
// 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; i != NumOps; ++i)
- if (!isa<ConstantInt>(Ops[i]))
+ if (!isa<ConstantInt>(Ops[i])) {
+
+ // If this is "gep i8* Ptr, (sub 0, V)", fold this as:
+ // "inttoptr (sub (ptrtoint Ptr), V)"
+ if (NumOps == 2 &&
+ cast<PointerType>(ResultTy)->getElementType()->isIntegerTy(8)) {
+ ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[1]);
+ assert((CE == 0 || CE->getType() == IntPtrTy) &&
+ "CastGEPIndices didn't canonicalize index types!");
+ if (CE && CE->getOpcode() == Instruction::Sub &&
+ CE->getOperand(0)->isNullValue()) {
+ Constant *Res = ConstantExpr::getPtrToInt(Ptr, CE->getType());
+ Res = ConstantExpr::getSub(Res, CE->getOperand(1));
+ Res = ConstantExpr::getIntToPtr(Res, ResultTy);
+ if (ConstantExpr *ResCE = dyn_cast<ConstantExpr>(Res))
+ Res = ConstantFoldConstantExpression(ResCE, TD);
+ return Res;
+ }
+ }
return 0;
+ }
+ unsigned BitWidth = TD->getTypeSizeInBits(IntPtrTy);
APInt Offset = APInt(BitWidth,
TD->getIndexedOffset(Ptr->getType(),
(Value**)Ops+1, NumOps-1));
+ Ptr = cast<Constant>(Ptr->stripPointerCasts());
+
+ // 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());
+
+ // Do not try the incorporate the sub-GEP if some index is not a number.
+ bool AllConstantInt = true;
+ for (unsigned i = 0, e = NestedOps.size(); i != e; ++i)
+ if (!isa<ConstantInt>(NestedOps[i])) {
+ AllConstantInt = false;
+ break;
+ }
+ if (!AllConstantInt)
+ break;
+
+ Ptr = cast<Constant>(GEP->getOperand(0));
+ Offset += APInt(BitWidth,
+ TD->getIndexedOffset(Ptr->getType(),
+ (Value**)NestedOps.data(),
+ NestedOps.size()));
+ Ptr = cast<Constant>(Ptr->stripPointerCasts());
+ }
+
// 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.
- if (BaseIsInt) {
+ APInt BasePtr(BitWidth, 0);
+ 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);
return ConstantExpr::getIntToPtr(C, ResultTy);
}
// we eliminate over-indexing of the notional static type array bounds.
// This makes it easy to determine if the getelementptr is "inbounds".
// Also, this helps GlobalOpt do SROA on GlobalVariables.
- Ptr = cast<Constant>(Ptr->stripPointerCasts());
const Type *Ty = Ptr->getType();
SmallVector<Constant*, 32> NewIdxs;
do {
if (const SequentialType *ATy = dyn_cast<SequentialType>(Ty)) {
- if (isa<PointerType>(ATy)) {
+ if (ATy->isPointerTy()) {
// The only pointer indexing we'll do is on the first index of the GEP.
if (!NewIdxs.empty())
break;
// Determine which element of the array the offset points into.
APInt ElemSize(BitWidth, TD->getTypeAllocSize(ATy->getElementType()));
+ const IntegerType *IntPtrTy = TD->getIntPtrType(Ty->getContext());
if (ElemSize == 0)
- return 0;
- APInt NewIdx = Offset.udiv(ElemSize);
- Offset -= NewIdx * ElemSize;
- NewIdxs.push_back(ConstantInt::get(TD->getIntPtrType(Ty->getContext()),
- NewIdx));
+ // 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
+ // accommodate the offset.
+ NewIdxs.push_back(ConstantInt::get(IntPtrTy, 0));
+ else {
+ // The element size is non-zero divide the offset by the element
+ // size (rounding down), to compute the index at this level.
+ APInt NewIdx = Offset.udiv(ElemSize);
+ Offset -= NewIdx * ElemSize;
+ NewIdxs.push_back(ConstantInt::get(IntPtrTy, NewIdx));
+ }
Ty = ATy->getElementType();
} else if (const StructType *STy = dyn_cast<StructType>(Ty)) {
// Determine which field of the struct the offset points into. The
// Constant Folding public APIs
//===----------------------------------------------------------------------===//
-
-/// ConstantFoldInstruction - Attempt to constant fold the specified
-/// instruction. If successful, the constant result is returned, if not, null
-/// is returned. Note that this function can only fail when attempting to fold
-/// instructions like loads and stores, which have no constant expression form.
-///
+/// ConstantFoldInstruction - 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) {
+ // Handle PHI nodes quickly here...
if (PHINode *PN = dyn_cast<PHINode>(I)) {
- if (PN->getNumIncomingValues() == 0)
- return UndefValue::get(PN->getType());
+ Constant *CommonValue = 0;
- Constant *Result = dyn_cast<Constant>(PN->getIncomingValue(0));
- if (Result == 0) return 0;
-
- // Handle PHI nodes specially here...
- for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i)
- if (PN->getIncomingValue(i) != Result && PN->getIncomingValue(i) != PN)
- return 0; // Not all the same incoming constants...
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
+ Value *Incoming = PN->getIncomingValue(i);
+ // If the incoming value is undef then skip it. Note that while we could
+ // skip the value if it is equal to the phi node itself we choose not to
+ // because that would break the rule that constant folding only applies if
+ // 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.
+ Constant *C = dyn_cast<Constant>(Incoming);
+ if (!C || (CommonValue && C != CommonValue))
+ return 0;
+ CommonValue = C;
+ }
- // If we reach here, all incoming values are the same constant.
- return Result;
+ // 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,
if (const LoadInst *LI = dyn_cast<LoadInst>(I))
return ConstantFoldLoadInst(LI, TD);
-
+
+ if (InsertValueInst *IVI = dyn_cast<InsertValueInst>(I))
+ return ConstantExpr::getInsertValue(
+ cast<Constant>(IVI->getAggregateOperand()),
+ cast<Constant>(IVI->getInsertedValueOperand()),
+ IVI->idx_begin(), IVI->getNumIndices());
+
+ if (ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(I))
+ return ConstantExpr::getExtractValue(
+ cast<Constant>(EVI->getAggregateOperand()),
+ EVI->idx_begin(), EVI->getNumIndices());
+
return ConstantFoldInstOperands(I->getOpcode(), I->getType(),
Ops.data(), Ops.size(), TD);
}
/// 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(ConstantExpr *CE,
+Constant *llvm::ConstantFoldConstantExpression(const ConstantExpr *CE,
const TargetData *TD) {
SmallVector<Constant*, 8> Ops;
- for (User::op_iterator i = CE->op_begin(), e = CE->op_end(); i != e; ++i) {
+ 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))
case Instruction::ICmp:
case Instruction::FCmp: assert(0 && "Invalid for compares");
case Instruction::Call:
- if (Function *F = dyn_cast<Function>(Ops[0]))
+ if (Function *F = dyn_cast<Function>(Ops[NumOps - 1]))
if (canConstantFoldCallTo(F))
- return ConstantFoldCall(F, Ops+1, NumOps-1);
+ return ConstantFoldCall(F, Ops, NumOps - 1);
return 0;
case Instruction::PtrToInt:
// If the input is a inttoptr, eliminate the pair. This requires knowing
// 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 (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[0]))
if (TD &&
- TD->getPointerSizeInBits() <=
- CE->getType()->getScalarSizeInBits()) {
- if (CE->getOpcode() == Instruction::PtrToInt)
- return FoldBitCast(CE->getOperand(0), DestTy, *TD);
-
- // If there's a constant offset added to the integer value before
- // it is casted back to a pointer, see if the expression can be
- // converted into a GEP.
- if (CE->getOpcode() == Instruction::Add)
- if (ConstantInt *L = dyn_cast<ConstantInt>(CE->getOperand(0)))
- if (ConstantExpr *R = dyn_cast<ConstantExpr>(CE->getOperand(1)))
- if (R->getOpcode() == Instruction::PtrToInt)
- if (GlobalVariable *GV =
- dyn_cast<GlobalVariable>(R->getOperand(0))) {
- const PointerType *GVTy = cast<PointerType>(GV->getType());
- if (const ArrayType *AT =
- dyn_cast<ArrayType>(GVTy->getElementType())) {
- const Type *ElTy = AT->getElementType();
- uint64_t AllocSize = TD->getTypeAllocSize(ElTy);
- APInt PSA(L->getValue().getBitWidth(), AllocSize);
- if (ElTy == cast<PointerType>(DestTy)->getElementType() &&
- L->getValue().urem(PSA) == 0) {
- APInt ElemIdx = L->getValue().udiv(PSA);
- if (ElemIdx.ult(APInt(ElemIdx.getBitWidth(),
- AT->getNumElements()))) {
- Constant *Index[] = {
- Constant::getNullValue(CE->getType()),
- ConstantInt::get(ElTy->getContext(), ElemIdx)
- };
- return
- ConstantExpr::getGetElementPtr(GV, &Index[0], 2);
- }
- }
- }
- }
- }
- }
+ TD->getPointerSizeInBits() <= CE->getType()->getScalarSizeInBits() &&
+ CE->getOpcode() == Instruction::PtrToInt)
+ return FoldBitCast(CE->getOperand(0), DestTy, *TD);
+
return ConstantExpr::getCast(Opcode, Ops[0], DestTy);
case Instruction::Trunc:
case Instruction::ZExt:
case Intrinsic::ctpop:
case Intrinsic::ctlz:
case Intrinsic::cttz:
- case Intrinsic::uadd_with_overflow:
- case Intrinsic::usub_with_overflow:
case Intrinsic::sadd_with_overflow:
+ case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
+ case Intrinsic::usub_with_overflow:
+ case Intrinsic::smul_with_overflow:
+ case Intrinsic::umul_with_overflow:
+ case Intrinsic::convert_from_fp16:
+ case Intrinsic::convert_to_fp16:
+ case Intrinsic::x86_sse_cvtss2si:
+ case Intrinsic::x86_sse_cvtss2si64:
+ case Intrinsic::x86_sse_cvttss2si:
+ case Intrinsic::x86_sse_cvttss2si64:
+ case Intrinsic::x86_sse2_cvtsd2si:
+ case Intrinsic::x86_sse2_cvtsd2si64:
+ case Intrinsic::x86_sse2_cvttsd2si:
+ case Intrinsic::x86_sse2_cvttsd2si64:
return true;
default:
return false;
case 'c':
return Name == "cos" || Name == "ceil" || Name == "cosf" || Name == "cosh";
case 'e':
- return Name == "exp";
+ return Name == "exp" || Name == "exp2";
case 'f':
return Name == "fabs" || Name == "fmod" || Name == "floor";
case 'l':
static Constant *ConstantFoldFP(double (*NativeFP)(double), double V,
const Type *Ty) {
- errno = 0;
+ sys::llvm_fenv_clearexcept();
V = NativeFP(V);
- if (errno != 0) {
- errno = 0;
+ if (sys::llvm_fenv_testexcept()) {
+ sys::llvm_fenv_clearexcept();
return 0;
}
static Constant *ConstantFoldBinaryFP(double (*NativeFP)(double, double),
double V, double W, const Type *Ty) {
- errno = 0;
+ sys::llvm_fenv_clearexcept();
V = NativeFP(V, W);
- if (errno != 0) {
- errno = 0;
+ if (sys::llvm_fenv_testexcept()) {
+ sys::llvm_fenv_clearexcept();
return 0;
}
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,
+ const Type *Ty) {
+ assert(Op && "Called with NULL operand");
+ APFloat Val(Op->getValueAPF());
+
+ // All of these conversion intrinsics form an integer of at most 64bits.
+ unsigned ResultWidth = cast<IntegerType>(Ty)->getBitWidth();
+ assert(ResultWidth <= 64 &&
+ "Can only constant fold conversions to 64 and 32 bit ints");
+
+ uint64_t UIntVal;
+ bool isExact = false;
+ APFloat::roundingMode mode = roundTowardZero? APFloat::rmTowardZero
+ : APFloat::rmNearestTiesToEven;
+ APFloat::opStatus status = Val.convertToInteger(&UIntVal, ResultWidth,
+ /*isSigned=*/true, mode,
+ &isExact);
+ if (status != APFloat::opOK && status != APFloat::opInexact)
+ return 0;
+ 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 *
const Type *Ty = F->getReturnType();
if (NumOperands == 1) {
if (ConstantFP *Op = dyn_cast<ConstantFP>(Operands[0])) {
+ if (F->getIntrinsicID() == 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());
+ }
+
if (!Ty->isFloatTy() && !Ty->isDoubleTy())
return 0;
+
+ /// 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;
+
/// 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)) ==
case 'e':
if (Name == "exp")
return ConstantFoldFP(exp, V, Ty);
+
+ if (Name == "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")
return ConstantFoldFP(log, V, Ty);
else if (Name == "log10" && V > 0)
return ConstantFoldFP(log10, V, Ty);
- else if (Name == "llvm.sqrt.f32" ||
- Name == "llvm.sqrt.f64") {
+ else if (F->getIntrinsicID() == Intrinsic::sqrt &&
+ (Ty->isFloatTy() || Ty->isDoubleTy())) {
if (V >= -0.0)
return ConstantFoldFP(sqrt, V, Ty);
else // Undefined
}
return 0;
}
-
-
+
if (ConstantInt *Op = dyn_cast<ConstantInt>(Operands[0])) {
- if (Name.startswith("llvm.bswap"))
+ switch (F->getIntrinsicID()) {
+ case Intrinsic::bswap:
return ConstantInt::get(F->getContext(), Op->getValue().byteSwap());
- else if (Name.startswith("llvm.ctpop"))
+ case Intrinsic::ctpop:
return ConstantInt::get(Ty, Op->getValue().countPopulation());
- else if (Name.startswith("llvm.cttz"))
+ case Intrinsic::cttz:
return ConstantInt::get(Ty, Op->getValue().countTrailingZeros());
- else if (Name.startswith("llvm.ctlz"))
+ case Intrinsic::ctlz:
return ConstantInt::get(Ty, Op->getValue().countLeadingZeros());
+ case Intrinsic::convert_from_fp16: {
+ APFloat Val(Op->getValue());
+
+ bool lost = false;
+ APFloat::opStatus status =
+ Val.convert(APFloat::IEEEsingle, APFloat::rmNearestTiesToEven, &lost);
+
+ // Conversion is always precise.
+ (void)status;
+ assert(status == APFloat::opOK && !lost &&
+ "Precision lost during fp16 constfolding");
+
+ return ConstantFP::get(F->getContext(), Val);
+ }
+ default:
+ return 0;
+ }
+ }
+
+ if (ConstantVector *Op = dyn_cast<ConstantVector>(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);
+ 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 (isa<UndefValue>(Operands[0])) {
+ if (F->getIntrinsicID() == Intrinsic::bswap)
+ return Operands[0];
return 0;
}
-
+
return 0;
}
-
+
if (NumOperands == 2) {
if (ConstantFP *Op1 = dyn_cast<ConstantFP>(Operands[0])) {
if (!Ty->isFloatTy() && !Ty->isDoubleTy())
if (Name == "atan2")
return ConstantFoldBinaryFP(atan2, Op1V, Op2V, Ty);
} else if (ConstantInt *Op2C = dyn_cast<ConstantInt>(Operands[1])) {
- if (Name == "llvm.powi.f32")
+ if (F->getIntrinsicID() == Intrinsic::powi && Ty->isFloatTy())
return ConstantFP::get(F->getContext(),
APFloat((float)std::pow((float)Op1V,
(int)Op2C->getZExtValue())));
- if (Name == "llvm.powi.f64")
+ if (F->getIntrinsicID() == Intrinsic::powi && Ty->isDoubleTy())
return ConstantFP::get(F->getContext(),
APFloat((double)std::pow((double)Op1V,
(int)Op2C->getZExtValue())));
if (ConstantInt *Op2 = dyn_cast<ConstantInt>(Operands[1])) {
switch (F->getIntrinsicID()) {
default: break;
- case Intrinsic::uadd_with_overflow: {
- Constant *Res = ConstantExpr::getAdd(Op1, Op2); // result.
- Constant *Ops[] = {
- Res, ConstantExpr::getICmp(CmpInst::ICMP_ULT, Res, Op1) // overflow.
- };
- return ConstantStruct::get(F->getContext(), Ops, 2, false);
- }
- case Intrinsic::usub_with_overflow: {
- Constant *Res = ConstantExpr::getSub(Op1, Op2); // result.
+ case Intrinsic::sadd_with_overflow:
+ case Intrinsic::uadd_with_overflow:
+ case Intrinsic::ssub_with_overflow:
+ case Intrinsic::usub_with_overflow:
+ case Intrinsic::smul_with_overflow:
+ case Intrinsic::umul_with_overflow: {
+ APInt Res;
+ bool Overflow;
+ switch (F->getIntrinsicID()) {
+ default: assert(0 && "Invalid case");
+ case Intrinsic::sadd_with_overflow:
+ Res = Op1->getValue().sadd_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::uadd_with_overflow:
+ Res = Op1->getValue().uadd_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::ssub_with_overflow:
+ Res = Op1->getValue().ssub_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::usub_with_overflow:
+ Res = Op1->getValue().usub_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::smul_with_overflow:
+ Res = Op1->getValue().smul_ov(Op2->getValue(), Overflow);
+ break;
+ case Intrinsic::umul_with_overflow:
+ Res = Op1->getValue().umul_ov(Op2->getValue(), Overflow);
+ break;
+ }
Constant *Ops[] = {
- Res, ConstantExpr::getICmp(CmpInst::ICMP_UGT, Res, Op1) // overflow.
+ ConstantInt::get(F->getContext(), Res),
+ ConstantInt::get(Type::getInt1Ty(F->getContext()), Overflow)
};
return ConstantStruct::get(F->getContext(), Ops, 2, false);
}
- case Intrinsic::sadd_with_overflow: {
- Constant *Res = ConstantExpr::getAdd(Op1, Op2); // result.
- Constant *Overflow = ConstantExpr::getSelect(
- ConstantExpr::getICmp(CmpInst::ICMP_SGT,
- ConstantInt::get(Op1->getType(), 0), Op1),
- ConstantExpr::getICmp(CmpInst::ICMP_SGT, Res, Op2),
- ConstantExpr::getICmp(CmpInst::ICMP_SLT, Res, Op2)); // overflow.
-
- Constant *Ops[] = { Res, Overflow };
- return ConstantStruct::get(F->getContext(), Ops, 2, false);
- }
- case Intrinsic::ssub_with_overflow: {
- Constant *Res = ConstantExpr::getSub(Op1, Op2); // result.
- Constant *Overflow = ConstantExpr::getSelect(
- ConstantExpr::getICmp(CmpInst::ICMP_SGT,
- ConstantInt::get(Op2->getType(), 0), Op2),
- ConstantExpr::getICmp(CmpInst::ICMP_SLT, Res, Op1),
- ConstantExpr::getICmp(CmpInst::ICMP_SGT, Res, Op1)); // overflow.
-
- Constant *Ops[] = { Res, Overflow };
- return ConstantStruct::get(F->getContext(), Ops, 2, false);
- }
}
}
}
return 0;
}
-
projects / oota-llvm.git / blobdiff