1;; Test for memprof metadata propagation, ensuring metadata is moved to 2;; inlined callsites. 3;; Also check that callsite metadata was updated with inlined stack ids. 4;; 5;; The following code was used to generate the following IR and its memprof 6;; profile: 7;; 8;; #include <stdlib.h> 9;; #include <string.h> 10;; #include <unistd.h> 11;; char *foo() { 12;; return new char[10]; 13;; } 14;; char *foo2() __attribute((noinline)) { 15;; return foo(); 16;; } 17;; char *bar() { 18;; return foo2(); 19;; } 20;; char *baz() { 21;; return foo2(); 22;; } 23;; int main(int argc, char **argv) { 24;; char *c = foo(); 25;; char *d = foo(); 26;; char *e = bar(); 27;; char *f = baz(); 28;; memset(c, 0, 10); 29;; memset(d, 0, 10); 30;; memset(e, 0, 10); 31;; memset(f, 0, 10); 32;; delete[] c; 33;; sleep(200); 34;; delete[] d; 35;; delete[] e; 36;; delete[] f; 37;; return 0; 38;; } 39 40; RUN: opt -passes=inline %s -S | FileCheck %s 41 42; ModuleID = 'memprof_inline2.cc' 43source_filename = "memprof_inline2.cc" 44target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" 45target triple = "x86_64-unknown-linux-gnu" 46 47; Function Attrs: mustprogress uwtable 48; CHECK-LABEL: define dso_local noundef ptr @_Z3foov 49define dso_local noundef ptr @_Z3foov() #0 !dbg !39 { 50entry: 51 ;; We should keep the original memprof metadata intact. 52 ; CHECK: call {{.*}} @_Znam{{.*}} !memprof ![[ORIGMEMPROF:[0-9]+]] 53 %call = call noalias noundef nonnull ptr @_Znam(i64 noundef 10) #7, !dbg !42, !memprof !43, !callsite !52 54 ret ptr %call, !dbg !53 55} 56 57; Function Attrs: nobuiltin allocsize(0) 58declare noundef nonnull ptr @_Znam(i64 noundef) #1 59 60;; Mark noinline so we don't inline into calls from bar and baz. We should end 61;; up with a memprof metadata on the call to foo below. 62; Function Attrs: mustprogress noinline uwtable 63; CHECK-LABEL: define dso_local noundef ptr @_Z4foo2v 64define dso_local noundef ptr @_Z4foo2v() #2 !dbg !54 { 65entry: 66 ;; We should have memprof metadata for the call stacks from bar and baz, 67 ;; and the callsite metadata should be the concatentation of the id from the 68 ;; inlined call to new and the original callsite. 69 ; CHECK: call {{.*}} @_Znam{{.*}} !memprof ![[NEWMEMPROF:[0-9]+]], !callsite ![[NEWCALLSITE:[0-9]+]] 70 %call = call noundef ptr @_Z3foov(), !dbg !55, !callsite !56 71 ret ptr %call, !dbg !57 72} 73 74; Function Attrs: mustprogress uwtable 75define dso_local noundef ptr @_Z3barv() #0 !dbg !58 { 76entry: 77 %call = call noundef ptr @_Z4foo2v(), !dbg !59, !callsite !60 78 ret ptr %call, !dbg !61 79} 80 81; Function Attrs: mustprogress uwtable 82define dso_local noundef ptr @_Z3bazv() #0 !dbg !62 { 83entry: 84 %call = call noundef ptr @_Z4foo2v(), !dbg !63, !callsite !64 85 ret ptr %call, !dbg !65 86} 87 88;; Make sure we don't propagate any memprof/callsite metadata 89; Function Attrs: mustprogress uwtable 90; CHECK-LABEL: define dso_local noundef ptr @notprofiled 91define dso_local noundef ptr @notprofiled() #0 !dbg !66 { 92entry: 93 ;; When foo is inlined, both the memprof and callsite metadata should be 94 ;; stripped from the inlined call to new, as there is no callsite metadata on 95 ;; the call. 96 ; CHECK: call {{.*}} @_Znam 97 ; CHECK-NOT: !memprof 98 ; CHECK-NOT: !callsite 99 %call = call noundef ptr @_Z3foov(), !dbg !67 100 ;; When baz is inlined, the callsite metadata should be stripped from the 101 ;; inlined call to foo2, as there is no callsite metadata on the call. 102 ; CHECK: call {{.*}} @_Z4foo2v 103 ; CHECK-NOT: !callsite 104 %call2 = call noundef ptr @_Z3bazv() 105 ; CHECK-NEXT: ret 106 ret ptr %call, !dbg !68 107} 108 109; Function Attrs: mustprogress noinline norecurse optnone uwtable 110define dso_local noundef i32 @main(i32 noundef %argc, ptr noundef %argv) #3 !dbg !69 { 111entry: 112 %retval = alloca i32, align 4 113 %argc.addr = alloca i32, align 4 114 %argv.addr = alloca ptr, align 8 115 %c = alloca ptr, align 8 116 %d = alloca ptr, align 8 117 %e = alloca ptr, align 8 118 %f = alloca ptr, align 8 119 store i32 0, ptr %retval, align 4 120 store i32 %argc, ptr %argc.addr, align 4 121 store ptr %argv, ptr %argv.addr, align 8 122 ;; The below 4 callsites are all annotated as noinline 123 %call = call noundef ptr @_Z3foov() #8, !dbg !70, !callsite !71 124 store ptr %call, ptr %c, align 8, !dbg !72 125 %call1 = call noundef ptr @_Z3foov() #8, !dbg !73, !callsite !74 126 store ptr %call1, ptr %d, align 8, !dbg !75 127 %call2 = call noundef ptr @_Z3barv() #8, !dbg !76, !callsite !77 128 store ptr %call2, ptr %e, align 8, !dbg !78 129 %call3 = call noundef ptr @_Z3bazv() #8, !dbg !79, !callsite !80 130 store ptr %call3, ptr %f, align 8, !dbg !81 131 %0 = load ptr, ptr %c, align 8, !dbg !82 132 call void @llvm.memset.p0.i64(ptr align 1 %0, i8 0, i64 10, i1 false), !dbg !83 133 %1 = load ptr, ptr %d, align 8, !dbg !84 134 call void @llvm.memset.p0.i64(ptr align 1 %1, i8 0, i64 10, i1 false), !dbg !85 135 %2 = load ptr, ptr %e, align 8, !dbg !86 136 call void @llvm.memset.p0.i64(ptr align 1 %2, i8 0, i64 10, i1 false), !dbg !87 137 %3 = load ptr, ptr %f, align 8, !dbg !88 138 call void @llvm.memset.p0.i64(ptr align 1 %3, i8 0, i64 10, i1 false), !dbg !89 139 %4 = load ptr, ptr %c, align 8, !dbg !90 140 %isnull = icmp eq ptr %4, null, !dbg !91 141 br i1 %isnull, label %delete.end, label %delete.notnull, !dbg !91 142 143delete.notnull: ; preds = %entry 144 call void @_ZdaPv(ptr noundef %4) #9, !dbg !92 145 br label %delete.end, !dbg !92 146 147delete.end: ; preds = %delete.notnull, %entry 148 %call4 = call i32 @sleep(i32 noundef 200), !dbg !94 149 %5 = load ptr, ptr %d, align 8, !dbg !95 150 %isnull5 = icmp eq ptr %5, null, !dbg !96 151 br i1 %isnull5, label %delete.end7, label %delete.notnull6, !dbg !96 152 153delete.notnull6: ; preds = %delete.end 154 call void @_ZdaPv(ptr noundef %5) #9, !dbg !97 155 br label %delete.end7, !dbg !97 156 157delete.end7: ; preds = %delete.notnull6, %delete.end 158 %6 = load ptr, ptr %e, align 8, !dbg !98 159 %isnull8 = icmp eq ptr %6, null, !dbg !99 160 br i1 %isnull8, label %delete.end10, label %delete.notnull9, !dbg !99 161 162delete.notnull9: ; preds = %delete.end7 163 call void @_ZdaPv(ptr noundef %6) #9, !dbg !100 164 br label %delete.end10, !dbg !100 165 166delete.end10: ; preds = %delete.notnull9, %delete.end7 167 %7 = load ptr, ptr %f, align 8, !dbg !101 168 %isnull11 = icmp eq ptr %7, null, !dbg !102 169 br i1 %isnull11, label %delete.end13, label %delete.notnull12, !dbg !102 170 171delete.notnull12: ; preds = %delete.end10 172 call void @_ZdaPv(ptr noundef %7) #9, !dbg !103 173 br label %delete.end13, !dbg !103 174 175delete.end13: ; preds = %delete.notnull12, %delete.end10 176 ret i32 0, !dbg !104 177} 178 179; Function Attrs: argmemonly nofree nounwind willreturn writeonly 180declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1 immarg) #4 181 182; Function Attrs: nobuiltin nounwind 183declare void @_ZdaPv(ptr noundef) #5 184 185declare i32 @sleep(i32 noundef) #6 186 187attributes #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" } 188attributes #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" } 189attributes #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" } 190attributes #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" } 191attributes #4 = { argmemonly nofree nounwind willreturn writeonly } 192attributes #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" } 193attributes #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" } 194attributes #7 = { builtin allocsize(0) } 195attributes #8 = { noinline } 196attributes #9 = { builtin nounwind } 197 198!llvm.dbg.cu = !{!0} 199!llvm.module.flags = !{!2, !3, !4, !5, !6, !7, !8, !9} 200!llvm.ident = !{!38} 201 202!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) 203!1 = !DIFile(filename: "memprof_inline.cc", directory: "/usr/local/google/home/tejohnson/llvm/tmp", checksumkind: CSK_MD5, checksum: "8711f6fd269e6cb5611fef48bc906eab") 204!2 = !{i32 7, !"Dwarf Version", i32 5} 205!3 = !{i32 2, !"Debug Info Version", i32 3} 206!4 = !{i32 1, !"wchar_size", i32 4} 207!5 = !{i32 7, !"PIC Level", i32 2} 208!6 = !{i32 7, !"PIE Level", i32 2} 209!7 = !{i32 7, !"uwtable", i32 2} 210!8 = !{i32 7, !"frame-pointer", i32 2} 211!9 = !{i32 1, !"ProfileSummary", !10} 212!10 = !{!11, !12, !13, !14, !15, !16, !17, !18, !19, !20} 213!11 = !{!"ProfileFormat", !"InstrProf"} 214!12 = !{!"TotalCount", i64 0} 215!13 = !{!"MaxCount", i64 0} 216!14 = !{!"MaxInternalCount", i64 0} 217!15 = !{!"MaxFunctionCount", i64 0} 218!16 = !{!"NumCounts", i64 0} 219!17 = !{!"NumFunctions", i64 0} 220!18 = !{!"IsPartialProfile", i64 0} 221!19 = !{!"PartialProfileRatio", double 0.000000e+00} 222!20 = !{!"DetailedSummary", !21} 223!21 = !{!22, !23, !24, !25, !26, !27, !28, !29, !30, !31, !32, !33, !34, !35, !36, !37} 224!22 = !{i32 10000, i64 0, i32 0} 225!23 = !{i32 100000, i64 0, i32 0} 226!24 = !{i32 200000, i64 0, i32 0} 227!25 = !{i32 300000, i64 0, i32 0} 228!26 = !{i32 400000, i64 0, i32 0} 229!27 = !{i32 500000, i64 0, i32 0} 230!28 = !{i32 600000, i64 0, i32 0} 231!29 = !{i32 700000, i64 0, i32 0} 232!30 = !{i32 800000, i64 0, i32 0} 233!31 = !{i32 900000, i64 0, i32 0} 234!32 = !{i32 950000, i64 0, i32 0} 235!33 = !{i32 990000, i64 0, i32 0} 236!34 = !{i32 999000, i64 0, i32 0} 237!35 = !{i32 999900, i64 0, i32 0} 238!36 = !{i32 999990, i64 0, i32 0} 239!37 = !{i32 999999, i64 0, i32 0} 240!38 = !{!"clang version 15.0.0 (https://github.com/llvm/llvm-project.git e09c924f98ec157adeaa74819b0aec9a07a1b552)"} 241!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) 242!40 = !DISubroutineType(types: !41) 243!41 = !{} 244!42 = !DILocation(line: 5, column: 10, scope: !39) 245;; The first 2 are from the direct calls to foo from main. Those stay on the 246;; callsite in foo, which isn't inlined into main due to the callsites in main 247;; being annotated as noinline. 248;; The second 2 are from the calls from foo2, which inlines its callsite to foo 249;; but is not itself inlined into its callers. Therefore they get moved to a 250;; new memprof metadata within foo2. 251!43 = !{!44, !46, !48, !50} 252!44 = !{!45, !"cold"} 253!45 = !{i64 -2458008693472584243, i64 7394638144382192936} 254!46 = !{!47, !"notcold"} 255!47 = !{i64 -2458008693472584243, i64 -8908997186479157179} 256!48 = !{!49, !"notcold"} 257!49 = !{i64 -2458008693472584243, i64 -8079659623765193173, i64 -4805294506621015872} 258!50 = !{!51, !"cold"} 259!51 = !{i64 -2458008693472584243, i64 -8079659623765193173, i64 -972865200055133905} 260; CHECK: ![[ORIGMEMPROF]] = !{![[ORIGMIB1:[0-9]+]], ![[ORIGMIB2:[0-9]+]], ![[ORIGMIB3:[0-9]+]], ![[ORIGMIB4:[0-9]+]]} 261; CHECK: ![[ORIGMIB1]] = !{![[ORIGMIBSTACK1:[0-9]+]], !"cold"} 262; CHECK: ![[ORIGMIBSTACK1]] = !{i64 -2458008693472584243, i64 7394638144382192936} 263; CHECK: ![[ORIGMIB2]] = !{![[ORIGMIBSTACK2:[0-9]+]], !"notcold"} 264; CHECK: ![[ORIGMIBSTACK2]] = !{i64 -2458008693472584243, i64 -8908997186479157179} 265; CHECK: ![[ORIGMIB3]] = !{![[ORIGMIBSTACK3:[0-9]+]], !"notcold"} 266; CHECK: ![[ORIGMIBSTACK3]] = !{i64 -2458008693472584243, i64 -8079659623765193173, i64 -4805294506621015872} 267; CHECK: ![[ORIGMIB4]] = !{![[ORIGMIBSTACK4:[0-9]+]], !"cold"} 268; CHECK: ![[ORIGMIBSTACK4]] = !{i64 -2458008693472584243, i64 -8079659623765193173, i64 -972865200055133905} 269; CHECK: ![[NEWMEMPROF]] = !{![[ORIGMIB3:[0-9]+]], ![[ORIGMIB4:[0-9]+]]} 270; CHECK: ![[NEWCALLSITE]] = !{i64 -2458008693472584243, i64 -8079659623765193173} 271!52 = !{i64 -2458008693472584243} 272!53 = !DILocation(line: 5, column: 3, scope: !39) 273!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) 274!55 = !DILocation(line: 8, column: 10, scope: !54) 275!56 = !{i64 -8079659623765193173} 276!57 = !DILocation(line: 8, column: 3, scope: !54) 277!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) 278!59 = !DILocation(line: 11, column: 10, scope: !58) 279!60 = !{i64 -972865200055133905} 280!61 = !DILocation(line: 11, column: 3, scope: !58) 281!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) 282!63 = !DILocation(line: 14, column: 10, scope: !62) 283!64 = !{i64 -4805294506621015872} 284!65 = !DILocation(line: 14, column: 3, scope: !62) 285!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) 286!67 = !DILocation(line: 401, column: 10, scope: !66) 287!68 = !DILocation(line: 401, column: 3, scope: !66) 288!69 = distinct !DISubprogram(name: "main", scope: !1, file: !1, line: 16, type: !40, scopeLine: 16, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !41) 289!70 = !DILocation(line: 17, column: 13, scope: !69) 290!71 = !{i64 -8908997186479157179} 291!72 = !DILocation(line: 17, column: 9, scope: !69) 292!73 = !DILocation(line: 18, column: 13, scope: !69) 293!74 = !{i64 7394638144382192936} 294!75 = !DILocation(line: 18, column: 9, scope: !69) 295!76 = !DILocation(line: 19, column: 13, scope: !69) 296!77 = !{i64 -5510257407004945023} 297!78 = !DILocation(line: 19, column: 9, scope: !69) 298!79 = !DILocation(line: 20, column: 13, scope: !69) 299!80 = !{i64 8771588133652501463} 300!81 = !DILocation(line: 20, column: 9, scope: !69) 301!82 = !DILocation(line: 21, column: 10, scope: !69) 302!83 = !DILocation(line: 21, column: 3, scope: !69) 303!84 = !DILocation(line: 22, column: 10, scope: !69) 304!85 = !DILocation(line: 22, column: 3, scope: !69) 305!86 = !DILocation(line: 23, column: 10, scope: !69) 306!87 = !DILocation(line: 23, column: 3, scope: !69) 307!88 = !DILocation(line: 24, column: 10, scope: !69) 308!89 = !DILocation(line: 24, column: 3, scope: !69) 309!90 = !DILocation(line: 25, column: 12, scope: !69) 310!91 = !DILocation(line: 25, column: 3, scope: !69) 311!92 = !DILocation(line: 25, column: 3, scope: !93) 312!93 = !DILexicalBlockFile(scope: !69, file: !1, discriminator: 2) 313!94 = !DILocation(line: 26, column: 3, scope: !69) 314!95 = !DILocation(line: 27, column: 12, scope: !69) 315!96 = !DILocation(line: 27, column: 3, scope: !69) 316!97 = !DILocation(line: 27, column: 3, scope: !93) 317!98 = !DILocation(line: 28, column: 12, scope: !69) 318!99 = !DILocation(line: 28, column: 3, scope: !69) 319!100 = !DILocation(line: 28, column: 3, scope: !93) 320!101 = !DILocation(line: 29, column: 12, scope: !69) 321!102 = !DILocation(line: 29, column: 3, scope: !69) 322!103 = !DILocation(line: 29, column: 3, scope: !93) 323!104 = !DILocation(line: 30, column: 3, scope: !69) 324