return nullptr;
}
-//===----------------------------------------------------------------------===//
-// Double -> Float Shrinking Optimizations for Unary Functions like 'floor'
+/// Any floating-point library function that we're trying to simplify will have
+/// a signature of the form: fptype foo(fptype param1, fptype param2, ...).
+/// CheckDoubleTy indicates that 'fptype' must be 'double'.
+static bool matchesFPLibFunctionSignature(const Function *F, unsigned NumParams,
+ bool CheckDoubleTy) {
+ FunctionType *FT = F->getFunctionType();
+ if (FT->getNumParams() != NumParams)
+ return false;
+
+ // The return type must match what we're looking for.
+ Type *RetTy = FT->getReturnType();
+ if (CheckDoubleTy ? !RetTy->isDoubleTy() : !RetTy->isFloatingPointTy())
+ return false;
-Value *LibCallSimplifier::optimizeUnaryDoubleFP(CallInst *CI, IRBuilder<> &B,
- bool CheckRetType) {
+ // Each parameter must match the return type, and therefore, match every other
+ // parameter too.
+ for (const Type *ParamTy : FT->params())
+ if (ParamTy != RetTy)
+ return false;
+
+ return true;
+}
+
+/// Shrink double -> float for unary functions like 'floor'.
+static Value *optimizeUnaryDoubleFP(CallInst *CI, IRBuilder<> &B,
+ bool CheckRetType) {
Function *Callee = CI->getCalledFunction();
- FunctionType *FT = Callee->getFunctionType();
- if (FT->getNumParams() != 1 || !FT->getReturnType()->isDoubleTy() ||
- !FT->getParamType(0)->isDoubleTy())
+ if (!matchesFPLibFunctionSignature(Callee, 1, true))
return nullptr;
if (CheckRetType) {
// Propagate fast-math flags from the existing call to the new call.
IRBuilder<>::FastMathFlagGuard Guard(B);
- B.SetFastMathFlags(CI->getFastMathFlags());
+ B.setFastMathFlags(CI->getFastMathFlags());
// floor((double)floatval) -> (double)floorf(floatval)
if (Callee->isIntrinsic()) {
return B.CreateFPExt(V, B.getDoubleTy());
}
-// Double -> Float Shrinking Optimizations for Binary Functions like 'fmin/fmax'
-Value *LibCallSimplifier::optimizeBinaryDoubleFP(CallInst *CI, IRBuilder<> &B) {
+/// Shrink double -> float for binary functions like 'fmin/fmax'.
+static Value *optimizeBinaryDoubleFP(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
- FunctionType *FT = Callee->getFunctionType();
- // Just make sure this has 2 arguments of the same FP type, which match the
- // result type.
- if (FT->getNumParams() != 2 || FT->getReturnType() != FT->getParamType(0) ||
- FT->getParamType(0) != FT->getParamType(1) ||
- !FT->getParamType(0)->isFloatingPointTy())
+ if (!matchesFPLibFunctionSignature(Callee, 2, true))
return nullptr;
// If this is something like 'fmin((double)floatval1, (double)floatval2)',
// Propagate fast-math flags from the existing call to the new call.
IRBuilder<>::FastMathFlagGuard Guard(B);
- B.SetFastMathFlags(CI->getFastMathFlags());
+ B.setFastMathFlags(CI->getFastMathFlags());
// fmin((double)floatval1, (double)floatval2)
// -> (double)fminf(floatval1, floatval2)
Callee->getAttributes());
}
+ // FIXME: Use instruction-level FMF.
bool UnsafeFPMath = canUseUnsafeFPMath(CI->getParent()->getParent());
- // pow(exp(x), y) -> exp(x*y)
+ // pow(exp(x), y) -> exp(x * y)
// pow(exp2(x), y) -> exp2(x * y)
- // We enable these only under fast-math. Besides rounding
- // differences the transformation changes overflow and
- // underflow behavior quite dramatically.
+ // We enable these only with fast-math. Besides rounding differences, the
+ // transformation changes overflow and underflow behavior quite dramatically.
// Example: x = 1000, y = 0.001.
// pow(exp(x), y) = pow(inf, 0.001) = inf, whereas exp(x*y) = exp(1).
- if (UnsafeFPMath) {
- if (auto *OpC = dyn_cast<CallInst>(Op1)) {
+ auto *OpC = dyn_cast<CallInst>(Op1);
+ if (OpC && OpC->hasUnsafeAlgebra() && CI->hasUnsafeAlgebra()) {
+ LibFunc::Func Func;
+ Function *OpCCallee = OpC->getCalledFunction();
+ if (OpCCallee && TLI->getLibFunc(OpCCallee->getName(), Func) &&
+ TLI->has(Func) && (Func == LibFunc::exp || Func == LibFunc::exp2)) {
IRBuilder<>::FastMathFlagGuard Guard(B);
- FastMathFlags FMF;
- FMF.setUnsafeAlgebra();
- B.SetFastMathFlags(FMF);
-
- LibFunc::Func Func;
- Function *OpCCallee = OpC->getCalledFunction();
- if (OpCCallee && TLI->getLibFunc(OpCCallee->getName(), Func) &&
- TLI->has(Func) && (Func == LibFunc::exp || Func == LibFunc::exp2))
- return EmitUnaryFloatFnCall(
- B.CreateFMul(OpC->getArgOperand(0), Op2, "mul"),
- OpCCallee->getName(), B, OpCCallee->getAttributes());
+ B.setFastMathFlags(CI->getFastMathFlags());
+ Value *FMul = B.CreateFMul(OpC->getArgOperand(0), Op2, "mul");
+ return EmitUnaryFloatFnCall(FMul, OpCCallee->getName(), B,
+ OpCCallee->getAttributes());
}
}
LibFunc::fabsl)) {
// In -ffast-math, pow(x, 0.5) -> sqrt(x).
- if (UnsafeFPMath)
+ if (CI->hasUnsafeAlgebra()) {
+ IRBuilder<>::FastMathFlagGuard Guard(B);
+ B.setFastMathFlags(CI->getFastMathFlags());
return EmitUnaryFloatFnCall(Op1, TLI->getName(LibFunc::sqrt), B,
Callee->getAttributes());
+ }
// Expand pow(x, 0.5) to (x == -infinity ? +infinity : fabs(sqrt(x))).
// This is faster than calling pow, and still handles negative zero
FMF.setNoSignedZeros();
FMF.setNoNaNs();
}
- B.SetFastMathFlags(FMF);
+ B.setFastMathFlags(FMF);
// We have a relaxed floating-point environment. We can ignore NaN-handling
// and transform to a compare and select. We do not have to consider errno or
!FT->getParamType(0)->isFloatingPointTy())
return Ret;
- if (!canUseUnsafeFPMath(CI->getParent()->getParent()))
+ if (!CI->hasUnsafeAlgebra())
return Ret;
Value *Op1 = CI->getArgOperand(0);
auto *OpC = dyn_cast<CallInst>(Op1);
- if (!OpC)
+
+ // The earlier call must also be unsafe in order to do these transforms.
+ if (!OpC || !OpC->hasUnsafeAlgebra())
return Ret;
// log(pow(x,y)) -> y*log(x)
IRBuilder<>::FastMathFlagGuard Guard(B);
FastMathFlags FMF;
FMF.setUnsafeAlgebra();
- B.SetFastMathFlags(FMF);
+ B.setFastMathFlags(FMF);
LibFunc::Func Func;
Function *F = OpC->getCalledFunction();
Value *LibCallSimplifier::optimizeSqrt(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
-
+
Value *Ret = nullptr;
if (TLI->has(LibFunc::sqrtf) && (Callee->getName() == "sqrt" ||
Callee->getIntrinsicID() == Intrinsic::sqrt))
Ret = optimizeUnaryDoubleFP(CI, B, true);
+ // FIXME: Refactor - this check is repeated all over this file and even in the
+ // preceding call to shrink double -> float.
+
+ // Make sure this has 1 argument of FP type, which matches the result type.
+ FunctionType *FT = Callee->getFunctionType();
+ if (FT->getNumParams() != 1 || FT->getReturnType() != FT->getParamType(0) ||
+ !FT->getParamType(0)->isFloatingPointTy())
+ return Ret;
+
if (!CI->hasUnsafeAlgebra())
return Ret;
// variations of this pattern because instcombine's visitFMUL and/or the
// reassociation pass should give us this form.
Value *OtherMul0, *OtherMul1;
- // FIXME: This multiply must be unsafe to allow this transform.
if (match(Op0, m_FMul(m_Value(OtherMul0), m_Value(OtherMul1)))) {
// Pattern: sqrt((x * y) * z)
- if (OtherMul0 == OtherMul1) {
+ if (OtherMul0 == OtherMul1 &&
+ cast<Instruction>(Op0)->hasUnsafeAlgebra()) {
// Matched: sqrt((x * x) * z)
RepeatOp = OtherMul0;
OtherOp = Op1;
// Fast math flags for any created instructions should match the sqrt
// and multiply.
IRBuilder<>::FastMathFlagGuard Guard(B);
- B.SetFastMathFlags(I->getFastMathFlags());
+ B.setFastMathFlags(I->getFastMathFlags());
+
// If we found a repeated factor, hoist it out of the square root and
// replace it with the fabs of that factor.
Module *M = Callee->getParent();