Skip to content

[Arch64][SVE] Lower svrev_* to llvm.vector.reverse and fold svrev(svrev(x)) -> x #116422

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

Open
jf-botto wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[Arch64][SVE] Lower svrev_* to llvm.vector.reverse and fold svrev(svrev(x)) -> x #116422

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9d70718
Pre-committing tests
jf-botto Apr 24, 2025
3569d1c
Adding rev(rev(x)) -> x optimisation
jf-botto Apr 25, 2025
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
56 changes: 56 additions & 0 deletions llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2368,6 +2368,7 @@ static std::optional<Instruction *> instCombineSVEUzp1(InstCombiner &IC,

// uzp1(to_svbool(A), to_svbool(B)) --> <A, B>
// uzp1(from_svbool(to_svbool(A)), from_svbool(to_svbool(B))) --> <A, B>

if ((match(II.getArgOperand(0),
m_Intrinsic<FromSVB>(m_Intrinsic<ToSVB>(m_Value(A)))) &&
match(II.getArgOperand(1),
Expand Down Expand Up @@ -2674,6 +2675,52 @@ static std::optional<Instruction *> instCombinePTrue(InstCombiner &IC,
return std::nullopt;
}

static std::optional<Instruction *> instCombineSVERev(InstCombiner &IC,
IntrinsicInst &II) {
// rev(rev(x)) -> x
switch (II.getIntrinsicID()) {
default:
return std::nullopt;

case Intrinsic::aarch64_sve_rev:
case Intrinsic::aarch64_sve_rev_b16:
case Intrinsic::aarch64_sve_rev_b32:
case Intrinsic::aarch64_sve_rev_b64: {
Value *InnerArg = II.getArgOperand(0);
IntrinsicInst *InnerRev = dyn_cast<IntrinsicInst>(InnerArg);
// Fold rev(rev(x)) -> x, if intrinsic IDs match and InnerRev has one use
if (InnerRev && InnerRev->getIntrinsicID() == II.getIntrinsicID() &&
InnerRev->hasOneUse())
return IC.replaceInstUsesWith(II, InnerRev->getArgOperand(0));

return std::nullopt;
}

case Intrinsic::aarch64_sve_revb:
case Intrinsic::aarch64_sve_revd:
case Intrinsic::aarch64_sve_revh:
case Intrinsic::aarch64_sve_revw: {
Value *InnerArg = II.getArgOperand(2);
IntrinsicInst *InnerRev = dyn_cast<IntrinsicInst>(InnerArg);
Copy link
Collaborator

@sdesmalen-arm sdesmalen-arm May 2, 2025

Choose a reason for hiding this comment

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

This is still missing checks for the governing predicate.


// Early exit if InnerRev != outerId and doesn't only have one use
if (!InnerRev || InnerRev->getIntrinsicID() != II.getIntrinsicID() ||
!InnerRev->hasOneUse())
return std::nullopt;

Value *OuterPred = II.getArgOperand(0);
Value *OuterPassThru = II.getArgOperand(1);
Value *InnerPred = InnerRev->getArgOperand(0);
Value *InnerPassThru = InnerRev->getArgOperand(1);
Comment on lines +2711 to +2714
Copy link
Collaborator

@sdesmalen-arm sdesmalen-arm May 2, 2025

Choose a reason for hiding this comment

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

These variables are all unused.


// Fold rev(rev(x)) -> x, if predicates and pass-thrus match
return IC.replaceInstUsesWith(II, InnerRev->getArgOperand(2));

return std::nullopt;
Comment on lines +2717 to +2719
Copy link
Collaborator

@sdesmalen-arm sdesmalen-arm May 2, 2025

Choose a reason for hiding this comment

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

a return after a return?

}
}
}

std::optional<Instruction *>
AArch64TTIImpl::instCombineIntrinsic(InstCombiner &IC,
IntrinsicInst &II) const {
Expand Down Expand Up @@ -2773,6 +2820,15 @@ AArch64TTIImpl::instCombineIntrinsic(InstCombiner &IC,
return instCombineSVEInsr(IC, II);
case Intrinsic::aarch64_sve_ptrue:
return instCombinePTrue(IC, II);
case Intrinsic::aarch64_sve_rev:
case Intrinsic::aarch64_sve_rev_b16:
case Intrinsic::aarch64_sve_rev_b32:
case Intrinsic::aarch64_sve_rev_b64:
case Intrinsic::aarch64_sve_revb:
case Intrinsic::aarch64_sve_revd:
case Intrinsic::aarch64_sve_revh:
case Intrinsic::aarch64_sve_revw:
return instCombineSVERev(IC, II);
}

return std::nullopt;
Expand Down
Loading
Loading