This is an archive of the discontinued LLVM Phabricator instance.

Differential D147056

[SLP][AArch64] Add test to check for the vectorization of fshl
ClosedPublic

Authored by zjaffal on Mar 28 2023, 8:49 AM.

Details

Reviewers
fhahn
Commits
rG4d7d4543344b: [SLP][AArch64] Add test to check for the vectorization of fshl
Summary

Currently the cost for fshl is an overestimate causing SLP to vectorize when it is not necessary.

Diff Detail

Unit TestsFailed

TimeTest
90 msx64 debian > Flang.Examples::feature-list-class.f90
60 msx64 debian > Flang.Examples::feature-list-functions.f90

Event Timeline

zjaffal created this revision.Mar 28 2023, 8:49 AM
Herald added a project: Restricted Project. · View Herald TranscriptMar 28 2023, 8:49 AM
Herald added subscribers: vporpo, StephenFan, kristof.beyls. · View Herald Transcript
zjaffal requested review of this revision.Mar 28 2023, 8:49 AM
Herald added a project: Restricted Project. · View Herald TranscriptMar 28 2023, 8:49 AM
Herald added subscribers: llvm-commits, • pcwang-thead. · View Herald Transcript
fhahn accepted this revision.Mar 28 2023, 9:34 AM

LGTM, thanks! Maybe drop the -vectorize part from the name :)

llvm/test/Transforms/SLPVectorizer/AArch64/fshl-vectorize.ll
3

Can you add a newline before and remove the newline before the function definition?

This revision is now accepted and ready to land.Mar 28 2023, 9:34 AM
Harbormaster completed remote builds in B222269: Diff 509037.Mar 28 2023, 9:38 AM
This revision was landed with ongoing or failed builds.Mar 28 2023, 9:46 AM
Closed by commit rG4d7d4543344b: [SLP][AArch64] Add test to check for the vectorization of fshl (authored by zjaffal). · Explain Why
This revision was automatically updated to reflect the committed changes.
zjaffal added a commit: rG4d7d4543344b: [SLP][AArch64] Add test to check for the vectorization of fshl.

Revision Contents

PathSize
llvm/
test/
Transforms/
SLPVectorizer/
AArch64/
42 lines

Diff 509037

llvm/test/Transforms/SLPVectorizer/AArch64/fshl-vectorize.ll

  • This file was added.
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
; RUN: opt -mtriple=arm64-apple-ios -S -passes=slp-vectorizer < %s | FileCheck %s
; fshl instruction cost model is an overestimate causing this test to vectorize when it is not beneficial to do so.
fhahnUnsubmitted
Not Done
ReplyInline Actions

Can you add a newline before and remove the newline before the function definition?

fhahn: Can you add a newline before and remove the newline before the function definition?
define i64 @fshl(i64 %or1, i64 %or2, i64 %or3 ) {
; CHECK-LABEL: define i64 @fshl
; CHECK-SAME: (i64 [[OR1:%.*]], i64 [[OR2:%.*]], i64 [[OR3:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x i64> poison, i64 [[OR2]], i32 0
; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i64> [[TMP0]], i64 [[OR3]], i32 1
; CHECK-NEXT: [[TMP2:%.*]] = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> [[TMP1]], <2 x i64> zeroinitializer, <2 x i64> <i64 1, i64 2>)
; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x i64> <i64 poison, i64 0>, i64 [[OR1]], i32 0
; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <2 x i64> [[TMP1]], <2 x i64> <i64 poison, i64 0>, <2 x i32> <i32 0, i32 3>
; CHECK-NEXT: [[TMP5:%.*]] = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> [[TMP3]], <2 x i64> [[TMP4]], <2 x i64> <i64 17, i64 21>)
; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <2 x i64> [[TMP3]], <2 x i64> <i64 0, i64 poison>, <2 x i32> <i32 2, i32 0>
; CHECK-NEXT: [[TMP7:%.*]] = xor <2 x i64> [[TMP2]], [[TMP6]]
; CHECK-NEXT: [[TMP8:%.*]] = add <2 x i64> [[TMP7]], [[TMP3]]
; CHECK-NEXT: [[TMP9:%.*]] = xor <2 x i64> [[TMP5]], [[TMP8]]
; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i64> [[TMP9]], i32 0
; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i64> [[TMP8]], i32 1
; CHECK-NEXT: [[ADD3:%.*]] = or i64 [[TMP10]], [[TMP11]]
; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i64> [[TMP9]], i32 1
; CHECK-NEXT: [[XOR5:%.*]] = xor i64 [[ADD3]], [[TMP12]]
; CHECK-NEXT: ret i64 [[XOR5]]
;
entry:
%or4 = tail call i64 @llvm.fshl.i64(i64 %or2, i64 0, i64 1)
%xor1 = xor i64 %or4, 0
%or5 = tail call i64 @llvm.fshl.i64(i64 %or3, i64 0, i64 2)
%xor2 = xor i64 %or5, %or1
%add1 = add i64 %xor1, %or1
%add2 = add i64 0, %xor2
%or6 = tail call i64 @llvm.fshl.i64(i64 %or1, i64 %or2, i64 17)
%xor3 = xor i64 %or6, %add1
%or7 = tail call i64 @llvm.fshl.i64(i64 0, i64 0, i64 21)
%xor4 = xor i64 %or7, %add2
%add3 = or i64 %xor3, %add2
%xor5 = xor i64 %add3, %xor4
ret i64 %xor5
}
declare i64 @llvm.fshl.i64(i64, i64, i64)