Skip to content

[AMDGPU] Simplify API of matchFPExtFromF16. NFC. #108223

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Sep 11, 2024
Merged

[AMDGPU] Simplify API of matchFPExtFromF16. NFC. #108223

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
a344ef5
[AMDGPU] Use std::optional in InstCombine of amdgcn_fmed3. NFC.
jayfoad Sep 11, 2024
eebbb4c
address review comments
jayfoad Sep 11, 2024
488bde8
drop std::optional
jayfoad Sep 11, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
38 changes: 18 additions & 20 deletions llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -353,23 +353,20 @@ bool GCNTTIImpl::canSimplifyLegacyMulToMul(const Instruction &I,
}

/// Match an fpext from half to float, or a constant we can convert.
static bool matchFPExtFromF16(Value *Arg, Value *&FPExtSrc) {
if (match(Arg, m_OneUse(m_FPExt(m_Value(FPExtSrc)))))
return FPExtSrc->getType()->isHalfTy();

ConstantFP *CFP;
if (match(Arg, m_ConstantFP(CFP))) {
static Value *matchFPExtFromF16(Value *Arg) {
Value *Src = nullptr;
ConstantFP *CFP = nullptr;
if (match(Arg, m_OneUse(m_FPExt(m_Value(Src))))) {
if (Src->getType()->isHalfTy())
return Src;
} else if (match(Arg, m_ConstantFP(CFP))) {
bool LosesInfo;
APFloat Val(CFP->getValueAPF());
Val.convert(APFloat::IEEEhalf(), APFloat::rmNearestTiesToEven, &LosesInfo);
if (LosesInfo)
return false;

FPExtSrc = ConstantFP::get(Type::getHalfTy(Arg->getContext()), Val);
return true;
if (!LosesInfo)
return ConstantFP::get(Type::getHalfTy(Arg->getContext()), Val);
}

return false;
return nullptr;
}

// Trim all zero components from the end of the vector \p UseV and return
Expand Down Expand Up @@ -839,15 +836,16 @@ GCNTTIImpl::instCombineIntrinsic(InstCombiner &IC, IntrinsicInst &II) const {
if (!ST->hasMed3_16())
break;

Value *X, *Y, *Z;

// Repeat floating-point width reduction done for minnum/maxnum.
// fmed3((fpext X), (fpext Y), (fpext Z)) -> fpext (fmed3(X, Y, Z))
if (matchFPExtFromF16(Src0, X) && matchFPExtFromF16(Src1, Y) &&
matchFPExtFromF16(Src2, Z)) {
Value *NewCall = IC.Builder.CreateIntrinsic(IID, {X->getType()},
{X, Y, Z}, &II, II.getName());
return new FPExtInst(NewCall, II.getType());
if (Value *X = matchFPExtFromF16(Src0)) {
if (Value *Y = matchFPExtFromF16(Src1)) {
if (Value *Z = matchFPExtFromF16(Src2)) {
Value *NewCall = IC.Builder.CreateIntrinsic(
IID, {X->getType()}, {X, Y, Z}, &II, II.getName());
return new FPExtInst(NewCall, II.getType());
}
}
}

break;
Expand Down
Loading