When GFNI is present, custom bit reversal lowerings for scalar integers become active. They work by swapping the bytes in the scalar value and then reversing bits in a vector of bytes. However, the custom bit reversal lowering for a vector of bytes is disabled if GFNI is present in isolation, resulting messed up code.

Copy link

github-actionsbot commentedJul 11, 2025

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using@ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by theLLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on theLLVM Discord or on theforums.

llvmbot added the backend:X86 label

Jul 11, 2025

Copy link

Member

llvmbot commentedJul 11, 2025

@llvm/pr-subscribers-backend-x86

Author: Tobias Decking (TDecking)

Changes

Full diff:https://github.com/llvm/llvm-project/pull/148304.diff

2 Files Affected:

(modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+2-1)
(modified) llvm/test/CodeGen/X86/vector-bitreverse.ll (+53-47)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cppindex 5e35d5630d667..81bdf1d79f881 100644--- a/llvm/lib/Target/X86/X86ISelLowering.cpp+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp@@ -1328,6 +1328,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,     setOperationAction(ISD::BITREVERSE, MVT::i16, Custom);     setOperationAction(ISD::BITREVERSE, MVT::i32, Custom);     setOperationAction(ISD::BITREVERSE, MVT::i64, Custom);+    setOperationAction(ISD::BITREVERSE, MVT::v16i8, Custom);   }    if (!Subtarget.useSoftFloat() && Subtarget.hasSSSE3()) {@@ -32694,7 +32695,7 @@ static SDValue LowerBITREVERSE(SDValue Op, const X86Subtarget &Subtarget,   if (Subtarget.hasXOP() && !VT.is512BitVector())     return LowerBITREVERSE_XOP(Op, DAG);-  assert(Subtarget.hasSSSE3() && "SSSE3 required for BITREVERSE");+  assert((Subtarget.hasSSSE3() || Subtarget.hasGFNI()) && "SSSE3 or GFNI required for BITREVERSE");    SDValue In = Op.getOperand(0);   SDLoc DL(Op);diff --git a/llvm/test/CodeGen/X86/vector-bitreverse.ll b/llvm/test/CodeGen/X86/vector-bitreverse.llindex 5dcf19013f0b7..4ba2ec393d5d4 100644--- a/llvm/test/CodeGen/X86/vector-bitreverse.ll+++ b/llvm/test/CodeGen/X86/vector-bitreverse.ll@@ -8,6 +8,7 @@ ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=ALL,AVX,AVX512,AVX512BW ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefixes=ALL,XOP,XOPAVX1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=ALL,XOP,XOPAVX2+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+gfni | FileCheck %s --check-prefixes=ALL,GFNIPURE ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3,+gfni | FileCheck %s --check-prefixes=ALL,GFNISSE ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+gfni | FileCheck %s --check-prefixes=ALL,GFNIAVX,GFNIAVX1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+gfni | FileCheck %s --check-prefixes=ALL,GFNIAVX,GFNIAVX2@@ -60,13 +61,13 @@ define i8 @test_bitreverse_i8(i8 %a) nounwind { ; XOP-NEXT:    # kill: def $al killed $al killed $eax ; XOP-NEXT:    retq ;-; GFNISSE-LABEL: test_bitreverse_i8:-; GFNISSE:       # %bb.0:-; GFNISSE-NEXT:    movd %edi, %xmm0-; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0-; GFNISSE-NEXT:    movd %xmm0, %eax-; GFNISSE-NEXT:    # kill: def $al killed $al killed $eax-; GFNISSE-NEXT:    retq+; GFNIPURE-LABEL: test_bitreverse_i8:+; GFNIPURE:       # %bb.0:+; GFNIPURE-NEXT:    movd %edi, %xmm0+; GFNIPURE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0+; GFNIPURE-NEXT:    movd %xmm0, %eax+; GFNIPURE-NEXT:    # kill: def $al killed $al killed $eax+; GFNIPURE-NEXT:    retq ; ; GFNIAVX-LABEL: test_bitreverse_i8: ; GFNIAVX:       # %bb.0:@@ -134,14 +135,14 @@ define i16 @test_bitreverse_i16(i16 %a) nounwind { ; XOP-NEXT:    # kill: def $ax killed $ax killed $eax ; XOP-NEXT:    retq ;-; GFNISSE-LABEL: test_bitreverse_i16:-; GFNISSE:       # %bb.0:-; GFNISSE-NEXT:    movd %edi, %xmm0-; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0-; GFNISSE-NEXT:    movd %xmm0, %eax-; GFNISSE-NEXT:    rolw $8, %ax-; GFNISSE-NEXT:    # kill: def $ax killed $ax killed $eax-; GFNISSE-NEXT:    retq+; GFNIPURE-LABEL: test_bitreverse_i16:+; GFNIPURE:       # %bb.0:+; GFNIPURE-NEXT:    movd %edi, %xmm0+; GFNIPURE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0+; GFNIPURE-NEXT:    movd %xmm0, %eax+; GFNIPURE-NEXT:    rolw $8, %ax+; GFNIPURE-NEXT:    # kill: def $ax killed $ax killed $eax+; GFNIPURE-NEXT:    retq ; ; GFNIAVX-LABEL: test_bitreverse_i16: ; GFNIAVX:       # %bb.0:@@ -207,13 +208,13 @@ define i32 @test_bitreverse_i32(i32 %a) nounwind { ; XOP-NEXT:    vmovd %xmm0, %eax ; XOP-NEXT:    retq ;-; GFNISSE-LABEL: test_bitreverse_i32:-; GFNISSE:       # %bb.0:-; GFNISSE-NEXT:    movd %edi, %xmm0-; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0-; GFNISSE-NEXT:    movd %xmm0, %eax-; GFNISSE-NEXT:    bswapl %eax-; GFNISSE-NEXT:    retq+; GFNIPURE-LABEL: test_bitreverse_i32:+; GFNIPURE:       # %bb.0:+; GFNIPURE-NEXT:    movd %edi, %xmm0+; GFNIPURE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0+; GFNIPURE-NEXT:    movd %xmm0, %eax+; GFNIPURE-NEXT:    bswapl %eax+; GFNIPURE-NEXT:    retq ; ; GFNIAVX-LABEL: test_bitreverse_i32: ; GFNIAVX:       # %bb.0:@@ -282,13 +283,13 @@ define i64 @test_bitreverse_i64(i64 %a) nounwind { ; XOP-NEXT:    vmovq %xmm0, %rax ; XOP-NEXT:    retq ;-; GFNISSE-LABEL: test_bitreverse_i64:-; GFNISSE:       # %bb.0:-; GFNISSE-NEXT:    movq %rdi, %xmm0-; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0-; GFNISSE-NEXT:    movq %xmm0, %rax-; GFNISSE-NEXT:    bswapq %rax-; GFNISSE-NEXT:    retq+; GFNIPURE-LABEL: test_bitreverse_i64:+; GFNIPURE:       # %bb.0:+; GFNIPURE-NEXT:    movq %rdi, %xmm0+; GFNIPURE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0+; GFNIPURE-NEXT:    movq %xmm0, %rax+; GFNIPURE-NEXT:    bswapq %rax+; GFNIPURE-NEXT:    retq ; ; GFNIAVX-LABEL: test_bitreverse_i64: ; GFNIAVX:       # %bb.0:@@ -386,10 +387,10 @@ define <16 x i8> @test_bitreverse_v16i8(<16 x i8> %a) nounwind { ; XOP-NEXT:    vpperm {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0, %xmm0 ; XOP-NEXT:    retq ;-; GFNISSE-LABEL: test_bitreverse_v16i8:-; GFNISSE:       # %bb.0:-; GFNISSE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0-; GFNISSE-NEXT:    retq+; GFNIPURE-LABEL: test_bitreverse_v16i8:+; GFNIPURE:       # %bb.0:+; GFNIPURE-NEXT:    gf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0+; GFNIPURE-NEXT:    retq ; ; GFNIAVX-LABEL: test_bitreverse_v16i8: ; GFNIAVX:       # %bb.0:@@ -873,12 +874,12 @@ define <32 x i8> @test_bitreverse_v32i8(<32 x i8> %a) nounwind { ; XOPAVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0 ; XOPAVX2-NEXT:    retq ;-; GFNISSE-LABEL: test_bitreverse_v32i8:-; GFNISSE:       # %bb.0:-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1-; GFNISSE-NEXT:    retq+; GFNIPURE-LABEL: test_bitreverse_v32i8:+; GFNIPURE:       # %bb.0:+; GFNIPURE-NEXT:    movdqa {{.*#+}} xmm2 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]+; GFNIPURE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0+; GFNIPURE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1+; GFNIPURE-NEXT:    retq ; ; GFNIAVX-LABEL: test_bitreverse_v32i8: ; GFNIAVX:       # %bb.0:@@ -1724,14 +1725,14 @@ define <64 x i8> @test_bitreverse_v64i8(<64 x i8> %a) nounwind { ; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1 ; XOPAVX2-NEXT:    retq ;-; GFNISSE-LABEL: test_bitreverse_v64i8:-; GFNISSE:       # %bb.0:-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm4 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2-; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3-; GFNISSE-NEXT:    retq+; GFNIPURE-LABEL: test_bitreverse_v64i8:+; GFNIPURE:       # %bb.0:+; GFNIPURE-NEXT:    movdqa {{.*#+}} xmm4 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]+; GFNIPURE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm0+; GFNIPURE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm1+; GFNIPURE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm2+; GFNIPURE-NEXT:    gf2p8affineqb $0, %xmm4, %xmm3+; GFNIPURE-NEXT:    retq ; ; GFNIAVX1-LABEL: test_bitreverse_v64i8: ; GFNIAVX1:       # %bb.0:@@ -2849,6 +2850,11 @@ define <16 x i8> @fold_bitreverse_v16i8() nounwind { ; XOP-NEXT:    vmovaps {{.*#+}} xmm0 = [0,255,64,191,32,223,96,159,16,239,80,175,48,207,112,143] ; XOP-NEXT:    retq ;+; GFNIPURE-LABEL: fold_bitreverse_v16i8:+; GFNIPURE:       # %bb.0:+; GFNIPURE-NEXT:    movaps {{.*#+}} xmm0 = [0,255,64,191,32,223,96,159,16,239,80,175,48,207,112,143]+; GFNIPURE-NEXT:    retq+; ; GFNISSE-LABEL: fold_bitreverse_v16i8: ; GFNISSE:       # %bb.0: ; GFNISSE-NEXT:    movaps {{.*#+}} xmm0 = [0,255,64,191,32,223,96,159,16,239,80,175,48,207,112,143]

Copy link

github-actionsbot commentedJul 11, 2025•
edited
Loading

✅ With the latest revision this PR passed the C/C++ code formatter.

Copy link

Collaborator

topperc commentedJul 11, 2025•
edited
Loading

Please write a more descriptive title including[X86] and put the bug number in the description.

Fixllvm#148238

3a15c6a

TDecking force-pushed thebitrev branch from64d08b3 to3a15c6aCompare

July 11, 2025 23:35

TDecking changed the title~~Fix #148238~~[X86] Ensure that bit reversals of byte vectors are properly lowered on pure GFNI targets

Jul 11, 2025

RKSimon self-requested a review

July 14, 2025 13:59

RKSimon reviewed

Jul 14, 2025

View reviewed changes

Copy link

Collaborator

RKSimon left a comment

Choose a reason for hiding this comment

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

You need to use the update_llc_test_checks.py script to regenerate the vector-bitreverse.ll file

llvm/test/CodeGen/X86/vector-bitreverse.ll Outdated

		@@ -8,6 +8,7 @@
		; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw \| FileCheck %s --check-prefixes=ALL,AVX,AVX512,AVX512BW
		; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx \| FileCheck %s --check-prefixes=ALL,XOP,XOPAVX1
		; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 \| FileCheck %s --check-prefixes=ALL,XOP,XOPAVX2
		; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+gfni \| FileCheck %s --check-prefixes=ALL,GFNIPURE
		; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3,+gfni \| FileCheck %s --check-prefixes=ALL,GFNISSE

Copy link

Collaborator

RKSimonJul 14, 2025

Choose a reason for hiding this comment

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

Change this (and the new line) to:

; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+gfni | FileCheck %s --check-prefixes=ALL,GFNISSE,,GFNISSE2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3,+gfni | FileCheck %s --check-prefixes=ALL,GFNISSE,GFNISSSE2

Copy link

ContributorAuthor

TDeckingJul 18, 2025

Choose a reason for hiding this comment

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

Thanks for the help.

Should the second line be turned intoGFNISSSE3 instead ofGFNISSSE2?

RKSimon requested a review fromtopperc

July 14, 2025 14:02

topperc reviewed

Jul 15, 2025

View reviewed changes

llvm/lib/Target/X86/X86ISelLowering.cppShow resolvedHide resolved

TDecking added2 commits

July 18, 2025 13:37

Update the test file

d4194ec

Extend the optimization towards other vectors

c097e12

Also add a missing `Subtarget.useSoftFloat()`

RKSimon reviewed

Jul 18, 2025

View reviewed changes

llvm/test/CodeGen/X86/vector-bitreverse.ll Outdated

		; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+gfni \| FileCheck %s --check-prefixes=ALL,GFNIPURE
		; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3,+gfni \| FileCheck %s --check-prefixes=ALL,GFNISSE
		; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+gfni \| FileCheck %s --check-prefixes=ALL,GFNISSE,GFNISSE2
		; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3,+gfni \| FileCheck %s --check-prefixes=ALL,GFNISSE,GFNISSSE2

Copy link

Collaborator

RKSimonJul 18, 2025

Choose a reason for hiding this comment

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

GFNISSSE2 -> GFNISSSE3

Copy link

ContributorAuthor

TDeckingJul 18, 2025

Choose a reason for hiding this comment

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

Done.

Turn SSSE2 into SSSE3

718deea

RKSimon approved these changes

Jul 18, 2025

View reviewed changes

Copy link

Collaborator

RKSimon left a comment

Choose a reason for hiding this comment

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

LGTM

Formatting

4f27cd7

TDecking force-pushed thebitrev branch fromf975d75 to4f27cd7Compare

July 18, 2025 16:52

Copy link

ContributorAuthor

TDecking commentedJul 18, 2025

Sorry if I caused any inconvenience.

Merge branch 'main' into bitrev

aaf7f6d

RKSimon merged commit10b0dee intollvm:main

Jul 18, 2025

8 of 9 checks passed

Copy link

github-actionsbot commentedJul 18, 2025

@TDecking Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested by ourbuild bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detailhere.

If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part ofLLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!

TDecking deleted the bitrev branch

July 18, 2025 19:12

mdfazlay pushed a commit to mdfazlay/llvm-project that referenced this pull request

Jul 18, 2025

[X86] Ensure that bit reversals of byte vectors are properly lowered …

aa831f5

…on pure GFNI targets (llvm#148304)Fixesllvm#148238.When GFNI is present, custom bit reversal lowerings for scalar integersbecome active. They work by swapping the bytes in the scalar value andthen reversing bits in a vector of bytes. However, the custom bitreversal lowering for a vector of bytes is disabled if GFNI is presentin isolation, resulting messed up code.---------Co-authored-by: Simon Pilgrim <llvm-dev@redking.me.uk>

Labels

backend:X86

4 participants

Movatterモバイル変換

[X86] Ensure that bit reversals of byte vectors are properly lowered on pure GFNI targets#148304

[X86] Ensure that bit reversals of byte vectors are properly lowered on pure GFNI targets#148304

Conversation

TDecking commentedJul 11, 2025• editedLoading Uh oh!There was an error while loading.Please reload this page.

Uh oh!

Uh oh!

github-actionsbot commentedJul 11, 2025

Uh oh!

llvmbot commentedJul 11, 2025

Uh oh!

github-actionsbot commentedJul 11, 2025• editedLoading Uh oh!There was an error while loading.Please reload this page.

Uh oh!

Uh oh!

topperc commentedJul 11, 2025• editedLoading Uh oh!There was an error while loading.Please reload this page.

Uh oh!

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

RKSimonJul 14, 2025

Choose a reason for hiding this comment

Uh oh!

TDeckingJul 18, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

RKSimonJul 18, 2025

Choose a reason for hiding this comment

Uh oh!

TDeckingJul 18, 2025

Choose a reason for hiding this comment

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

TDecking commentedJul 18, 2025

Uh oh!

Uh oh!

github-actionsbot commentedJul 18, 2025

Uh oh!

Uh oh!

TDecking commentedJul 11, 2025•
edited
Loading

github-actionsbot commentedJul 11, 2025•
edited
Loading

topperc commentedJul 11, 2025•
edited
Loading