#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"
// 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.
}
}
- return ConstantVector::get(Result.data(), Result.size());
+ return ConstantVector::get(Result);
}
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;
}
// 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())
Constant *Ptr = Ops[0];
if (!TD || !cast<PointerType>(Ptr->getType())->getElementType()->isSized())
return 0;
-
- unsigned BitWidth =
- TD->getTypeSizeInBits(TD->getIntPtrType(Ptr->getContext()));
+
+ 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));
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();
- BasePtr.zextOrTrunc(BitWidth);
- }
+ 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);
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':
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 (Name == "llvm.convert.to.fp16") {
+ if (F->getIntrinsicID() == Intrinsic::convert_to_fp16) {
APFloat Val(Op->getValueAPF());
bool lost = false;
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());
- else if (Name == "llvm.convert.from.fp16") {
+ case Intrinsic::convert_from_fp16: {
APFloat Val(Op->getValue());
bool lost = false;
Val.convert(APFloat::IEEEsingle, APFloat::rmNearestTiesToEven, &lost);
// Conversion is always precise.
- status = status;
+ (void)status;
assert(status == APFloat::opOK && !lost &&
"Precision lost during fp16 constfolding");
return ConstantFP::get(F->getContext(), Val);
}
- return 0;
+ 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 (Name.startswith("llvm.bswap"))
+ 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())));
case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
case Intrinsic::usub_with_overflow:
- case Intrinsic::smul_with_overflow: {
+ case Intrinsic::smul_with_overflow:
+ case Intrinsic::umul_with_overflow: {
APInt Res;
bool Overflow;
switch (F->getIntrinsicID()) {
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[] = {
ConstantInt::get(F->getContext(), Res),
}
return 0;
}
-