Skip to content

[InstCombine][X86] Fold blendv(x,y,shuffle(bitcast(sext(m)))) -> select(shuffle(m),x,y) #96882

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 1 commit into from
Jul 3, 2024
Merged

[InstCombine][X86] Fold blendv(x,y,shuffle(bitcast(sext(m)))) -> select(shuffle(m),x,y) #96882

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
36 changes: 34 additions & 2 deletions llvm/lib/Target/X86/X86InstCombineIntrinsic.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2800,6 +2800,23 @@ X86TTIImpl::instCombineIntrinsic(InstCombiner &IC, IntrinsicInst &II) const {
return SelectInst::Create(NewSelector, Op1, Op0, "blendv");
}

// Peek through a one-use shuffle - VectorCombine should have simplified
// this for cases where we're splitting wider vectors to use blendv
// intrinsics.
Value *MaskSrc = nullptr;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we peekThroughBitcast on Mask here?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'll investigate this as a followup, but I've already tried very hard on VectorCombine to make sure we simplify bitcast(shuffle(bitcast(shuffle()))) chains that we would have to peek through

ArrayRef<int> ShuffleMask;
if (match(Mask, PatternMatch::m_OneUse(PatternMatch::m_Shuffle(
PatternMatch::m_Value(MaskSrc), PatternMatch::m_Undef(),
PatternMatch::m_Mask(ShuffleMask))))) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: We should probably just start using PatternMatch namespace at the top of the function. That can be a later NFC though.

// Bail if the shuffle was irregular or contains undefs.
int NumElts = cast<FixedVectorType>(MaskSrc->getType())->getNumElements();
if (NumElts < ShuffleMask.size() || !isPowerOf2_32(NumElts) ||
any_of(ShuffleMask,
[NumElts](int M) { return M < 0 || M >= NumElts; }))
break;
Mask = MaskSrc;
}

// Convert to a vector select if we can bypass casts and find a boolean
// vector condition value.
Value *BoolVec;
Expand All @@ -2809,11 +2826,26 @@ X86TTIImpl::instCombineIntrinsic(InstCombiner &IC, IntrinsicInst &II) const {
BoolVec->getType()->getScalarSizeInBits() == 1) {
auto *MaskTy = cast<FixedVectorType>(Mask->getType());
auto *OpTy = cast<FixedVectorType>(II.getType());
unsigned NumMaskElts = MaskTy->getNumElements();
unsigned NumOperandElts = OpTy->getNumElements();

// If we peeked through a shuffle, reapply the shuffle to the bool vector.
if (MaskSrc) {
unsigned NumMaskSrcElts =
cast<FixedVectorType>(MaskSrc->getType())->getNumElements();
NumMaskElts = (ShuffleMask.size() * NumMaskElts) / NumMaskSrcElts;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Isn't ShuffleMask.size() equal to NumMaskSrcElts?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

IR can have different result and argument vector widths (this patch is mainly to help extract subvector style shuffles after all) - only DAG is fixed

// Multiple mask bits maps to the same operand element - bail out.
if (NumMaskElts > NumOperandElts)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

isn't the scaleShuffleMaskElts check sufficient? If the shuffle is only smaller type but keep all OpTy elements together isn't that fine?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

we don't handle NumMaskElts > NumOperandElts (IIRC due to endianess issues) - this is an early out so we don't bother creating an unused shuffle instruction

break;
SmallVector<int> ScaledMask;
if (!llvm::scaleShuffleMaskElts(NumMaskElts, ShuffleMask, ScaledMask))
break;
BoolVec = IC.Builder.CreateShuffleVector(BoolVec, ScaledMask);
MaskTy = FixedVectorType::get(MaskTy->getElementType(), NumMaskElts);
}
assert(MaskTy->getPrimitiveSizeInBits() ==
OpTy->getPrimitiveSizeInBits() &&
"Not expecting mask and operands with different sizes");
unsigned NumMaskElts = MaskTy->getNumElements();
unsigned NumOperandElts = OpTy->getNumElements();

if (NumMaskElts == NumOperandElts) {
return SelectInst::Create(BoolVec, Op1, Op0);
Expand Down
Loading
Loading