This is an archive of the discontinued LLVM Phabricator instance.

Differential D152530

[SLP] Update X86/schedule_budget.ll test case to show budget impact
ClosedPublic

Authored by bjope on Jun 9 2023, 5:54 AM.

Details

Reviewers
ABataev
Commits
rG8813c1087e5e: [SLP] Update X86/schedule_budget.ll test case to show budget impact
Summary

The comments and the checks in

test/Transforms/SLPVectorizer/X86/schedule_budget.ll

did not match. After commit 352c46e70716061e99 the vectorization
has happened also with the reduced budget.

This patch is supposed to restore the original intention with the
test case (the one described in the comments). We want to see
that a restricted budget may reduce the amount of vectorization (i.e.
verifying that the -slp-schedule-budget option makes a difference),
while a higher budget still result in vectorization.

Diff Detail

Event Timeline

bjope created this revision.Jun 9 2023, 5:54 AM
Herald added a project: Restricted Project. · View Herald TranscriptJun 9 2023, 5:54 AM
Herald added subscribers: vporpo, pengfei. · View Herald Transcript
bjope requested review of this revision.Jun 9 2023, 5:54 AM
Herald added a project: Restricted Project. · View Herald TranscriptJun 9 2023, 5:54 AM
Herald added a subscriber: • pcwang-thead. · View Herald Transcript
ABataev accepted this revision.Jun 9 2023, 5:58 AM

LG

This revision is now accepted and ready to land.Jun 9 2023, 5:58 AM
This revision was landed with ongoing or failed builds.Jun 9 2023, 6:34 AM
Closed by commit rG8813c1087e5e: [SLP] Update X86/schedule_budget.ll test case to show budget impact (authored by bjope). · Explain Why
This revision was automatically updated to reflect the committed changes.
bjope added a commit: rG8813c1087e5e: [SLP] Update X86/schedule_budget.ll test case to show budget impact.
Harbormaster completed remote builds in B237745: Diff 529934.Jun 9 2023, 7:10 AM

Revision Contents

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

Diff 529946

llvm/test/Transforms/SLPVectorizer/X86/schedule_budget.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=slp-vectorizer -S -slp-schedule-budget=16 -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s ; RUN: opt < %s -passes=slp-vectorizer -S -slp-schedule-budget=16 -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck --check-prefix LOBUDGET %s
; RUN: opt < %s -passes=slp-vectorizer -S -slp-schedule-budget=32 -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck --check-prefix HIBUDGET %s
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.9.0" target triple = "x86_64-apple-macosx10.9.0"
; Test if the budget for the scheduling region size works. ; Test if the budget for the scheduling region size works.
; We test with a reduced budget of 16 which should prevent vectorizing the loads. ; We test with a reduced budget of 16 which should prevent vectorizing the loads.
declare void @unknown() declare void @unknown()
define void @test(ptr %a, ptr %b, ptr %c, ptr %d) { define void @test(ptr %a, ptr %b, ptr %c, ptr %d) {
; CHECK-LABEL: @test( ; LOBUDGET-LABEL: @test(
; CHECK-NEXT: entry: ; LOBUDGET-NEXT: entry:
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr [[A:%.*]], align 4 ; LOBUDGET-NEXT: [[L0:%.*]] = load float, ptr [[A:%.*]], align 4
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[A1:%.*]] = getelementptr inbounds float, ptr [[A]], i64 1
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[L1:%.*]] = load float, ptr [[A1]], align 4
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[A2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 2
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[L2:%.*]] = load float, ptr [[A2]], align 4
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[A3:%.*]] = getelementptr inbounds float, ptr [[A]], i64 3
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[L3:%.*]] = load float, ptr [[A3]], align 4
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[L00:%.*]] = fadd float [[L0]], [[L0]]
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[L10:%.*]] = fadd float [[L1]], [[L1]]
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[L20:%.*]] = fadd float [[L2]], [[L2]]
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: [[L30:%.*]] = fadd float [[L3]], [[L3]]
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: call void @unknown() ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: store <4 x float> [[TMP1]], ptr [[B:%.*]], align 4 ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: [[TMP4:%.*]] = load <4 x float>, ptr [[C:%.*]], align 4 ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: store <4 x float> [[TMP4]], ptr [[D:%.*]], align 4 ; LOBUDGET-NEXT: call void @unknown()
; CHECK-NEXT: ret void ; LOBUDGET-NEXT: call void @unknown()
; LOBUDGET-NEXT: call void @unknown()
; LOBUDGET-NEXT: call void @unknown()
; LOBUDGET-NEXT: call void @unknown()
; LOBUDGET-NEXT: call void @unknown()
; LOBUDGET-NEXT: call void @unknown()
; LOBUDGET-NEXT: call void @unknown()
; LOBUDGET-NEXT: store float [[L00]], ptr [[B:%.*]], align 4
; LOBUDGET-NEXT: [[B1:%.*]] = getelementptr inbounds float, ptr [[B]], i64 1
; LOBUDGET-NEXT: store float [[L10]], ptr [[B1]], align 4
; LOBUDGET-NEXT: [[B2:%.*]] = getelementptr inbounds float, ptr [[B]], i64 2
; LOBUDGET-NEXT: store float [[L20]], ptr [[B2]], align 4
; LOBUDGET-NEXT: [[B3:%.*]] = getelementptr inbounds float, ptr [[B]], i64 3
; LOBUDGET-NEXT: store float [[L30]], ptr [[B3]], align 4
; LOBUDGET-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[C:%.*]], align 4
; LOBUDGET-NEXT: store <4 x float> [[TMP0]], ptr [[D:%.*]], align 4
; LOBUDGET-NEXT: ret void
;
; HIBUDGET-LABEL: @test(
; HIBUDGET-NEXT: entry:
; HIBUDGET-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[A:%.*]], align 4
; HIBUDGET-NEXT: [[TMP1:%.*]] = fadd <4 x float> [[TMP0]], [[TMP0]]
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: call void @unknown()
; HIBUDGET-NEXT: store <4 x float> [[TMP1]], ptr [[B:%.*]], align 4
; HIBUDGET-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[C:%.*]], align 4
; HIBUDGET-NEXT: store <4 x float> [[TMP2]], ptr [[D:%.*]], align 4
; HIBUDGET-NEXT: ret void
; ;
entry: entry:
; Don't vectorize these loads. ; Don't vectorize these loads (with the reduced budget).
%l0 = load float, ptr %a %l0 = load float, ptr %a
%a1 = getelementptr inbounds float, ptr %a, i64 1 %a1 = getelementptr inbounds float, ptr %a, i64 1
%l1 = load float, ptr %a1 %l1 = load float, ptr %a1
%a2 = getelementptr inbounds float, ptr %a, i64 2 %a2 = getelementptr inbounds float, ptr %a, i64 2
%l2 = load float, ptr %a2 %l2 = load float, ptr %a2
%a3 = getelementptr inbounds float, ptr %a, i64 3 %a3 = getelementptr inbounds float, ptr %a, i64 3
%l3 = load float, ptr %a3 %l3 = load float, ptr %a3
%l00 = fadd float %l0, %l0
%l10 = fadd float %l1, %l1
%l20 = fadd float %l2, %l2
%l30 = fadd float %l3, %l3
; some unrelated instructions inbetween to enlarge the scheduling region ; some unrelated instructions inbetween to enlarge the scheduling region
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
Show All 15 Linesentry:
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
call void @unknown() call void @unknown()
; Don't vectorize these stores because their operands are too far away. ; Don't vectorize these stores because their operands are too far away (with
store float %l0, ptr %b ; the reduced budget).
store float %l00, ptr %b
%b1 = getelementptr inbounds float, ptr %b, i64 1 %b1 = getelementptr inbounds float, ptr %b, i64 1
store float %l1, ptr %b1 store float %l10, ptr %b1
%b2 = getelementptr inbounds float, ptr %b, i64 2 %b2 = getelementptr inbounds float, ptr %b, i64 2
store float %l2, ptr %b2 store float %l20, ptr %b2
%b3 = getelementptr inbounds float, ptr %b, i64 3 %b3 = getelementptr inbounds float, ptr %b, i64 3
store float %l3, ptr %b3 store float %l30, ptr %b3
; But still vectorize the following instructions, because even if the budget ; But still vectorize the following instructions, because even if the budget
; is exceeded there is a minimum region size. ; is exceeded there is a minimum region size.
%l4 = load float, ptr %c %l4 = load float, ptr %c
%c1 = getelementptr inbounds float, ptr %c, i64 1 %c1 = getelementptr inbounds float, ptr %c, i64 1
%l5 = load float, ptr %c1 %l5 = load float, ptr %c1
%c2 = getelementptr inbounds float, ptr %c, i64 2 %c2 = getelementptr inbounds float, ptr %c, i64 2
%l6 = load float, ptr %c2 %l6 = load float, ptr %c2
Show All 14 Lines