297 lines · plain
1;; This test is the same code as memprof_inline2.ll, except that it has2;; manually synthesized context size information. This test ensures that we3;; don't attempt to apply -memprof-callsite-cold-threshold again when4;; rebuilding the metadata after inlining.5;6; RUN: opt -passes=inline %s -S | FileCheck %s7;; We should not perform additional discarding of non-cold contexts when8;; rebuilding the tries after inlining, even with a very low threshold.9; RUN: opt -passes=inline -memprof-callsite-cold-threshold=0 %s -S | FileCheck %s10 11; ModuleID = 'memprof_inline2.cc'12source_filename = "memprof_inline2.cc"13target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"14target triple = "x86_64-unknown-linux-gnu"15 16; Function Attrs: mustprogress uwtable17; CHECK-LABEL: define dso_local noundef ptr @_Z3foov18define dso_local noundef ptr @_Z3foov() #0 !dbg !39 {19entry:20 ;; We should keep the original memprof metadata intact.21 ; CHECK: call {{.*}} @_Znam{{.*}} !memprof ![[ORIGMEMPROF:[0-9]+]]22 %call = call noalias noundef nonnull ptr @_Znam(i64 noundef 10) #7, !dbg !42, !memprof !43, !callsite !5223 ret ptr %call, !dbg !5324}25 26; Function Attrs: nobuiltin allocsize(0)27declare noundef nonnull ptr @_Znam(i64 noundef) #128 29;; Mark noinline so we don't inline into calls from bar and baz. We should end30;; up with a memprof metadata on the call to foo below.31; Function Attrs: mustprogress noinline uwtable32; CHECK-LABEL: define dso_local noundef ptr @_Z4foo2v33define dso_local noundef ptr @_Z4foo2v() #2 !dbg !54 {34entry:35 ;; We should have memprof metadata for the call stacks from bar and baz,36 ;; and the callsite metadata should be the concatentation of the id from the37 ;; inlined call to new and the original callsite.38 ; CHECK: call {{.*}} @_Znam{{.*}} !memprof ![[NEWMEMPROF:[0-9]+]], !callsite ![[NEWCALLSITE:[0-9]+]]39 %call = call noundef ptr @_Z3foov(), !dbg !55, !callsite !5640 ret ptr %call, !dbg !5741}42 43; Function Attrs: mustprogress uwtable44define dso_local noundef ptr @_Z3barv() #0 !dbg !58 {45entry:46 %call = call noundef ptr @_Z4foo2v(), !dbg !59, !callsite !6047 ret ptr %call, !dbg !6148}49 50; Function Attrs: mustprogress uwtable51define dso_local noundef ptr @_Z3bazv() #0 !dbg !62 {52entry:53 %call = call noundef ptr @_Z4foo2v(), !dbg !63, !callsite !6454 ret ptr %call, !dbg !6555}56 57;; Make sure we don't propagate any memprof/callsite metadata58; Function Attrs: mustprogress uwtable59; CHECK-LABEL: define dso_local noundef ptr @notprofiled60define dso_local noundef ptr @notprofiled() #0 !dbg !66 {61entry:62 ;; When foo is inlined, both the memprof and callsite metadata should be63 ;; stripped from the inlined call to new, as there is no callsite metadata on64 ;; the call.65 ; CHECK: call {{.*}} @_Znam66 ; CHECK-NOT: !memprof67 ; CHECK-NOT: !callsite68 %call = call noundef ptr @_Z3foov(), !dbg !6769 ;; When baz is inlined, the callsite metadata should be stripped from the70 ;; inlined call to foo2, as there is no callsite metadata on the call.71 ; CHECK: call {{.*}} @_Z4foo2v72 ; CHECK-NOT: !callsite73 %call2 = call noundef ptr @_Z3bazv()74 ; CHECK-NEXT: ret75 ret ptr %call, !dbg !6876}77 78; Function Attrs: mustprogress noinline norecurse optnone uwtable79define dso_local noundef i32 @main(i32 noundef %argc, ptr noundef %argv) #3 !dbg !69 {80entry:81 %retval = alloca i32, align 482 %argc.addr = alloca i32, align 483 %argv.addr = alloca ptr, align 884 %c = alloca ptr, align 885 %d = alloca ptr, align 886 %e = alloca ptr, align 887 %f = alloca ptr, align 888 store i32 0, ptr %retval, align 489 store i32 %argc, ptr %argc.addr, align 490 store ptr %argv, ptr %argv.addr, align 891 ;; The below 4 callsites are all annotated as noinline92 %call = call noundef ptr @_Z3foov() #8, !dbg !70, !callsite !7193 store ptr %call, ptr %c, align 8, !dbg !7294 %call1 = call noundef ptr @_Z3foov() #8, !dbg !73, !callsite !7495 store ptr %call1, ptr %d, align 8, !dbg !7596 %call2 = call noundef ptr @_Z3barv() #8, !dbg !76, !callsite !7797 store ptr %call2, ptr %e, align 8, !dbg !7898 %call3 = call noundef ptr @_Z3bazv() #8, !dbg !79, !callsite !8099 store ptr %call3, ptr %f, align 8, !dbg !81100 %0 = load ptr, ptr %c, align 8, !dbg !82101 call void @llvm.memset.p0.i64(ptr align 1 %0, i8 0, i64 10, i1 false), !dbg !83102 %1 = load ptr, ptr %d, align 8, !dbg !84103 call void @llvm.memset.p0.i64(ptr align 1 %1, i8 0, i64 10, i1 false), !dbg !85104 %2 = load ptr, ptr %e, align 8, !dbg !86105 call void @llvm.memset.p0.i64(ptr align 1 %2, i8 0, i64 10, i1 false), !dbg !87106 %3 = load ptr, ptr %f, align 8, !dbg !88107 call void @llvm.memset.p0.i64(ptr align 1 %3, i8 0, i64 10, i1 false), !dbg !89108 %4 = load ptr, ptr %c, align 8, !dbg !90109 %isnull = icmp eq ptr %4, null, !dbg !91110 br i1 %isnull, label %delete.end, label %delete.notnull, !dbg !91111 112delete.notnull: ; preds = %entry113 call void @_ZdaPv(ptr noundef %4) #9, !dbg !92114 br label %delete.end, !dbg !92115 116delete.end: ; preds = %delete.notnull, %entry117 %call4 = call i32 @sleep(i32 noundef 200), !dbg !94118 %5 = load ptr, ptr %d, align 8, !dbg !95119 %isnull5 = icmp eq ptr %5, null, !dbg !96120 br i1 %isnull5, label %delete.end7, label %delete.notnull6, !dbg !96121 122delete.notnull6: ; preds = %delete.end123 call void @_ZdaPv(ptr noundef %5) #9, !dbg !97124 br label %delete.end7, !dbg !97125 126delete.end7: ; preds = %delete.notnull6, %delete.end127 %6 = load ptr, ptr %e, align 8, !dbg !98128 %isnull8 = icmp eq ptr %6, null, !dbg !99129 br i1 %isnull8, label %delete.end10, label %delete.notnull9, !dbg !99130 131delete.notnull9: ; preds = %delete.end7132 call void @_ZdaPv(ptr noundef %6) #9, !dbg !100133 br label %delete.end10, !dbg !100134 135delete.end10: ; preds = %delete.notnull9, %delete.end7136 %7 = load ptr, ptr %f, align 8, !dbg !101137 %isnull11 = icmp eq ptr %7, null, !dbg !102138 br i1 %isnull11, label %delete.end13, label %delete.notnull12, !dbg !102139 140delete.notnull12: ; preds = %delete.end10141 call void @_ZdaPv(ptr noundef %7) #9, !dbg !103142 br label %delete.end13, !dbg !103143 144delete.end13: ; preds = %delete.notnull12, %delete.end10145 ret i32 0, !dbg !104146}147 148; Function Attrs: argmemonly nofree nounwind willreturn writeonly149declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1 immarg) #4150 151; Function Attrs: nobuiltin nounwind152declare void @_ZdaPv(ptr noundef) #5153 154declare i32 @sleep(i32 noundef) #6155 156attributes #0 = { mustprogress uwtable "disable-tail-calls"="true" "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }157attributes #1 = { nobuiltin allocsize(0) "disable-tail-calls"="true" "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }158attributes #2 = { mustprogress noinline uwtable "disable-tail-calls"="true" "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }159attributes #3 = { mustprogress noinline norecurse optnone uwtable "disable-tail-calls"="true" "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }160attributes #4 = { argmemonly nofree nounwind willreturn writeonly }161attributes #5 = { nobuiltin nounwind "disable-tail-calls"="true" "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }162attributes #6 = { "disable-tail-calls"="true" "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }163attributes #7 = { builtin allocsize(0) }164attributes #8 = { noinline }165attributes #9 = { builtin nounwind }166 167!llvm.dbg.cu = !{!0}168!llvm.module.flags = !{!2, !3, !4, !5, !6, !7, !8, !9}169!llvm.ident = !{!38}170 171!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !1, producer: "clang version 15.0.0 (https://github.com/llvm/llvm-project.git e09c924f98ec157adeaa74819b0aec9a07a1b552)", isOptimized: false, runtimeVersion: 0, emissionKind: LineTablesOnly, splitDebugInlining: false, debugInfoForProfiling: true, nameTableKind: None)172!1 = !DIFile(filename: "memprof_inline.cc", directory: "/usr/local/google/home/tejohnson/llvm/tmp", checksumkind: CSK_MD5, checksum: "8711f6fd269e6cb5611fef48bc906eab")173!2 = !{i32 7, !"Dwarf Version", i32 5}174!3 = !{i32 2, !"Debug Info Version", i32 3}175!4 = !{i32 1, !"wchar_size", i32 4}176!5 = !{i32 7, !"PIC Level", i32 2}177!6 = !{i32 7, !"PIE Level", i32 2}178!7 = !{i32 7, !"uwtable", i32 2}179!8 = !{i32 7, !"frame-pointer", i32 2}180!9 = !{i32 1, !"ProfileSummary", !10}181!10 = !{!11, !12, !13, !14, !15, !16, !17, !18, !19, !20}182!11 = !{!"ProfileFormat", !"InstrProf"}183!12 = !{!"TotalCount", i64 0}184!13 = !{!"MaxCount", i64 0}185!14 = !{!"MaxInternalCount", i64 0}186!15 = !{!"MaxFunctionCount", i64 0}187!16 = !{!"NumCounts", i64 0}188!17 = !{!"NumFunctions", i64 0}189!18 = !{!"IsPartialProfile", i64 0}190!19 = !{!"PartialProfileRatio", double 0.000000e+00}191!20 = !{!"DetailedSummary", !21}192!21 = !{!22, !23, !24, !25, !26, !27, !28, !29, !30, !31, !32, !33, !34, !35, !36, !37}193!22 = !{i32 10000, i64 0, i32 0}194!23 = !{i32 100000, i64 0, i32 0}195!24 = !{i32 200000, i64 0, i32 0}196!25 = !{i32 300000, i64 0, i32 0}197!26 = !{i32 400000, i64 0, i32 0}198!27 = !{i32 500000, i64 0, i32 0}199!28 = !{i32 600000, i64 0, i32 0}200!29 = !{i32 700000, i64 0, i32 0}201!30 = !{i32 800000, i64 0, i32 0}202!31 = !{i32 900000, i64 0, i32 0}203!32 = !{i32 950000, i64 0, i32 0}204!33 = !{i32 990000, i64 0, i32 0}205!34 = !{i32 999000, i64 0, i32 0}206!35 = !{i32 999900, i64 0, i32 0}207!36 = !{i32 999990, i64 0, i32 0}208!37 = !{i32 999999, i64 0, i32 0}209!38 = !{!"clang version 15.0.0 (https://github.com/llvm/llvm-project.git e09c924f98ec157adeaa74819b0aec9a07a1b552)"}210!39 = distinct !DISubprogram(name: "foo", linkageName: "_Z3foov", scope: !1, file: !1, line: 4, type: !40, scopeLine: 4, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !41)211!40 = !DISubroutineType(types: !41)212!41 = !{}213!42 = !DILocation(line: 5, column: 10, scope: !39)214;; The first 2 are from the direct calls to foo from main. Those stay on the215;; callsite in foo, which isn't inlined into main due to the callsites in main216;; being annotated as noinline.217;; The second 2 are from the calls from foo2, which inlines its callsite to foo218;; but is not itself inlined into its callers. Therefore they get moved to a219;; new memprof metadata within foo2.220!43 = !{!44, !46, !48, !50}221!44 = !{!45, !"cold", !105}222!105 = !{i64 123, i64 5000}223!45 = !{i64 -2458008693472584243, i64 7394638144382192936}224!46 = !{!47, !"notcold", !106}225!47 = !{i64 -2458008693472584243, i64 -8908997186479157179}226!106 = !{i64 345, i64 1}227!48 = !{!49, !"notcold", !107}228!49 = !{i64 -2458008693472584243, i64 -8079659623765193173, i64 -4805294506621015872}229!107 = !{i64 678, i64 1}230!50 = !{!51, !"cold", !108}231!51 = !{i64 -2458008693472584243, i64 -8079659623765193173, i64 -972865200055133905}232!108 = !{i64 234, i64 5000}233; CHECK: ![[ORIGMEMPROF]] = !{![[ORIGMIB1:[0-9]+]], ![[ORIGMIB2:[0-9]+]], ![[ORIGMIB3:[0-9]+]], ![[ORIGMIB4:[0-9]+]]}234; CHECK: ![[ORIGMIB1]] = !{![[ORIGMIBSTACK1:[0-9]+]], !"cold"235; CHECK: ![[ORIGMIBSTACK1]] = !{i64 -2458008693472584243, i64 7394638144382192936}236; CHECK: ![[ORIGMIB2]] = !{![[ORIGMIBSTACK2:[0-9]+]], !"notcold"237; CHECK: ![[ORIGMIBSTACK2]] = !{i64 -2458008693472584243, i64 -8908997186479157179}238; CHECK: ![[ORIGMIB3]] = !{![[ORIGMIBSTACK3:[0-9]+]], !"notcold"239; CHECK: ![[ORIGMIBSTACK3]] = !{i64 -2458008693472584243, i64 -8079659623765193173, i64 -4805294506621015872}240; CHECK: ![[ORIGMIB4]] = !{![[ORIGMIBSTACK4:[0-9]+]], !"cold"241; CHECK: ![[ORIGMIBSTACK4]] = !{i64 -2458008693472584243, i64 -8079659623765193173, i64 -972865200055133905}242; CHECK: ![[NEWMEMPROF]] = !{![[ORIGMIB3:[0-9]+]], ![[ORIGMIB4:[0-9]+]]}243; CHECK: ![[NEWCALLSITE]] = !{i64 -2458008693472584243, i64 -8079659623765193173}244!52 = !{i64 -2458008693472584243}245!53 = !DILocation(line: 5, column: 3, scope: !39)246!54 = distinct !DISubprogram(name: "foo2", linkageName: "_Z4foo2v", scope: !1, file: !1, line: 7, type: !40, scopeLine: 7, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !41)247!55 = !DILocation(line: 8, column: 10, scope: !54)248!56 = !{i64 -8079659623765193173}249!57 = !DILocation(line: 8, column: 3, scope: !54)250!58 = distinct !DISubprogram(name: "bar", linkageName: "_Z3barv", scope: !1, file: !1, line: 10, type: !40, scopeLine: 10, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !41)251!59 = !DILocation(line: 11, column: 10, scope: !58)252!60 = !{i64 -972865200055133905}253!61 = !DILocation(line: 11, column: 3, scope: !58)254!62 = distinct !DISubprogram(name: "baz", linkageName: "_Z3bazv", scope: !1, file: !1, line: 13, type: !40, scopeLine: 13, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !41)255!63 = !DILocation(line: 14, column: 10, scope: !62)256!64 = !{i64 -4805294506621015872}257!65 = !DILocation(line: 14, column: 3, scope: !62)258!66 = distinct !DISubprogram(name: "notprofiled", linkageName: "notprofiled", scope: !1, file: !1, line: 400, type: !40, scopeLine: 400, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !41)259!67 = !DILocation(line: 401, column: 10, scope: !66)260!68 = !DILocation(line: 401, column: 3, scope: !66)261!69 = distinct !DISubprogram(name: "main", scope: !1, file: !1, line: 16, type: !40, scopeLine: 16, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !41)262!70 = !DILocation(line: 17, column: 13, scope: !69)263!71 = !{i64 -8908997186479157179}264!72 = !DILocation(line: 17, column: 9, scope: !69)265!73 = !DILocation(line: 18, column: 13, scope: !69)266!74 = !{i64 7394638144382192936}267!75 = !DILocation(line: 18, column: 9, scope: !69)268!76 = !DILocation(line: 19, column: 13, scope: !69)269!77 = !{i64 -5510257407004945023}270!78 = !DILocation(line: 19, column: 9, scope: !69)271!79 = !DILocation(line: 20, column: 13, scope: !69)272!80 = !{i64 8771588133652501463}273!81 = !DILocation(line: 20, column: 9, scope: !69)274!82 = !DILocation(line: 21, column: 10, scope: !69)275!83 = !DILocation(line: 21, column: 3, scope: !69)276!84 = !DILocation(line: 22, column: 10, scope: !69)277!85 = !DILocation(line: 22, column: 3, scope: !69)278!86 = !DILocation(line: 23, column: 10, scope: !69)279!87 = !DILocation(line: 23, column: 3, scope: !69)280!88 = !DILocation(line: 24, column: 10, scope: !69)281!89 = !DILocation(line: 24, column: 3, scope: !69)282!90 = !DILocation(line: 25, column: 12, scope: !69)283!91 = !DILocation(line: 25, column: 3, scope: !69)284!92 = !DILocation(line: 25, column: 3, scope: !93)285!93 = !DILexicalBlockFile(scope: !69, file: !1, discriminator: 2)286!94 = !DILocation(line: 26, column: 3, scope: !69)287!95 = !DILocation(line: 27, column: 12, scope: !69)288!96 = !DILocation(line: 27, column: 3, scope: !69)289!97 = !DILocation(line: 27, column: 3, scope: !93)290!98 = !DILocation(line: 28, column: 12, scope: !69)291!99 = !DILocation(line: 28, column: 3, scope: !69)292!100 = !DILocation(line: 28, column: 3, scope: !93)293!101 = !DILocation(line: 29, column: 12, scope: !69)294!102 = !DILocation(line: 29, column: 3, scope: !69)295!103 = !DILocation(line: 29, column: 3, scope: !93)296!104 = !DILocation(line: 30, column: 3, scope: !69)297