This is an archive of the discontinued LLVM Phabricator instance.

Differential D152868

[SLP][NFC] Precommit test that exposes a bug in ShuffleBuilder.
ClosedPublic

Authored by vporpo on Jun 13 2023, 4:44 PM.

Details

Reviewers
ABataev
Commits
rG9d5466849a77: [SLP][NFC] Precommit test that exposes a bug in ShuffleBuilder.
Summary

ShuffleBuilder generates a zero mask here:
[[TMP6:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> poison, <4 x i32> zeroinitializer
But the correct mask is 0,0,1,1, or we should have reused TMP4.

Diff Detail

Unit TestsFailed

TimeTest
60,030 msx64 debian > MLIR.Examples/standalone::test.toy

Event Timeline

vporpo created this revision.Jun 13 2023, 4:44 PM
Herald added a project: Restricted Project. · View Herald TranscriptJun 13 2023, 4:44 PM
vporpo requested review of this revision.Jun 13 2023, 4:44 PM
Herald added a project: Restricted Project. · View Herald TranscriptJun 13 2023, 4:44 PM
Herald added subscribers: llvm-commits, • pcwang-thead. · View Herald Transcript
ABataev accepted this revision.Jun 13 2023, 4:50 PM

LG

This revision is now accepted and ready to land.Jun 13 2023, 4:50 PM
This revision was landed with ongoing or failed builds.Jun 13 2023, 4:53 PM
Closed by commit rG9d5466849a77: [SLP][NFC] Precommit test that exposes a bug in ShuffleBuilder. (authored by vporpo). · Explain Why
This revision was automatically updated to reflect the committed changes.
vporpo added a commit: rG9d5466849a77: [SLP][NFC] Precommit test that exposes a bug in ShuffleBuilder..
Harbormaster completed remote builds in B238651: Diff 531114.Jun 13 2023, 6:31 PM

Revision Contents

PathSize
llvm/
test/
Transforms/
SLPVectorizer/
X86/
45 lines

Diff 531114

llvm/test/Transforms/SLPVectorizer/X86/shufflebuilder-bug.ll

  • This file was added.
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
; RUN: opt -S -p slp-vectorizer -mtriple=x86_64-- %s | FileCheck %s
define void @foo(<4 x float> %vec, float %val, ptr %ptr) {
; CHECK-LABEL: define void @foo
; CHECK-SAME: (<4 x float> [[VEC:%.*]], float [[VAL:%.*]], ptr [[PTR:%.*]]) {
; CHECK-NEXT: [[GEP0:%.*]] = getelementptr inbounds float, ptr [[PTR]], i64 0
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr [[GEP0]], align 8
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[VEC]], <4 x float> poison, <2 x i32> <i32 3, i32 poison>
; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x float> [[TMP2]], float [[VAL]], i32 1
; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> poison, <4 x i32> <i32 0, i32 0, i32 1, i32 1>
; CHECK-NEXT: [[TMP5:%.*]] = fadd <4 x float> [[TMP1]], [[TMP4]]
; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> poison, <4 x i32> zeroinitializer
; CHECK-NEXT: [[TMP7:%.*]] = fmul <4 x float> [[TMP5]], [[TMP6]]
; CHECK-NEXT: store <4 x float> [[TMP7]], ptr [[GEP0]], align 4
; CHECK-NEXT: ret void
;
%vec_3 = extractelement <4 x float> %vec, i32 3
%gep0 = getelementptr inbounds float, ptr %ptr, i64 0
%gep1 = getelementptr inbounds float, ptr %ptr, i64 1
%gep2 = getelementptr inbounds float, ptr %ptr, i64 2
%gep3 = getelementptr inbounds float, ptr %ptr, i64 3
%l0 = load float, ptr %gep0, align 8
%l1 = load float, ptr %gep1, align 8
%l2 = load float, ptr %gep2, align 8
%l3 = load float, ptr %gep3, align 8
%fadd0 = fadd float %l0, %vec_3
%fadd1 = fadd float %l1, %vec_3
%fadd2 = fadd float %l2, %val
%fadd3 = fadd float %l3, %val
%fmul0 = fmul float %fadd0, %vec_3
%fmul1 = fmul float %fadd1, %vec_3
%fmul2 = fmul float %fadd2, %val
%fmul3 = fmul float %fadd3, %val
store float %fmul0, ptr %gep0, align 4
store float %fmul1, ptr %gep1, align 4
store float %fmul2, ptr %gep2, align 4
store float %fmul3, ptr %gep3, align 4
ret void
}