1; RUN: opt < %s -passes=sample-profile -sample-profile-file=%S/Inputs/gcc-simple.afdo -S | FileCheck %s 2; 3; Original code: 4; 5; #include <stdlib.h> 6; 7; long long int foo(long i) { 8; if (rand() < 500) return 2; else if (rand() > 5000) return 10; else return 90; 9; } 10; 11; int main() { 12; long long int sum = 0; 13; for (int k = 0; k < 3000; k++) 14; for (int i = 0; i < 200000; i++) sum += foo(i); 15; return sum > 0 ? 0 : 1; 16; } 17; 18; This test was compiled down to bytecode at -O0 to avoid inlining foo() into 19; main(). The profile was generated using a GCC-generated binary (also compiled 20; at -O0). The conversion from the Linux Perf profile to the GCC autofdo 21; profile used the converter at https://github.com/google/autofdo 22; 23; $ gcc -g -O0 gcc-simple.cc -o gcc-simple 24; $ perf record -b ./gcc-simple 25; $ create_gcov --binary=gcc-simple --gcov=gcc-simple.afdo 26 27define i64 @_Z3fool(i64 %i) #0 !dbg !4 { 28; CHECK: !prof ![[EC1:[0-9]+]] 29entry: 30 %retval = alloca i64, align 8 31 %i.addr = alloca i64, align 8 32 store i64 %i, ptr %i.addr, align 8 33 call void @llvm.dbg.declare(metadata ptr %i.addr, metadata !16, metadata !17), !dbg !18 34 %call = call i32 @rand() #3, !dbg !19 35; CHECK: !prof ![[PROF1:[0-9]+]] 36 %cmp = icmp slt i32 %call, 500, !dbg !21 37 br i1 %cmp, label %if.then, label %if.else, !dbg !22 38; CHECK: !prof ![[PROF2:[0-9]+]] 39 40if.then: ; preds = %entry 41 store i64 2, ptr %retval, align 8, !dbg !23 42 br label %return, !dbg !23 43 44if.else: ; preds = %entry 45 %call1 = call i32 @rand() #3, !dbg !25 46; CHECK: !prof ![[PROF3:[0-9]+]] 47 %cmp2 = icmp sgt i32 %call1, 5000, !dbg !28 48 br i1 %cmp2, label %if.then.3, label %if.else.4, !dbg !29 49; CHECK: !prof ![[PROF4:[0-9]+]] 50 51if.then.3: ; preds = %if.else 52 store i64 10, ptr %retval, align 8, !dbg !30 53 br label %return, !dbg !30 54 55if.else.4: ; preds = %if.else 56 store i64 90, ptr %retval, align 8, !dbg !32 57 br label %return, !dbg !32 58 59return: ; preds = %if.else.4, %if.then.3, %if.then 60 %0 = load i64, ptr %retval, align 8, !dbg !34 61 ret i64 %0, !dbg !34 62} 63 64; Function Attrs: nounwind readnone 65declare void @llvm.dbg.declare(metadata, metadata, metadata) #1 66 67; Function Attrs: nounwind 68declare i32 @rand() #2 69 70; Function Attrs: nounwind uwtable 71define i32 @main() #0 !dbg !9 { 72; CHECK: !prof ![[EC2:[0-9]+]] 73entry: 74 %retval = alloca i32, align 4 75 %sum = alloca i64, align 8 76 %k = alloca i32, align 4 77 %i = alloca i32, align 4 78 store i32 0, ptr %retval, align 4 79 call void @llvm.dbg.declare(metadata ptr %sum, metadata !35, metadata !17), !dbg !36 80 store i64 0, ptr %sum, align 8, !dbg !36 81 call void @llvm.dbg.declare(metadata ptr %k, metadata !37, metadata !17), !dbg !39 82 store i32 0, ptr %k, align 4, !dbg !39 83 br label %for.cond, !dbg !40 84 85for.cond: ; preds = %for.inc.4, %entry 86 %0 = load i32, ptr %k, align 4, !dbg !41 87 %cmp = icmp slt i32 %0, 3000, !dbg !45 88 br i1 %cmp, label %for.body, label %for.end.6, !dbg !46 89; CHECK: !prof ![[PROF6:[0-9]+]] 90 91for.body: ; preds = %for.cond 92 call void @llvm.dbg.declare(metadata ptr %i, metadata !47, metadata !17), !dbg !49 93 store i32 0, ptr %i, align 4, !dbg !49 94 br label %for.cond.1, !dbg !50 95 96for.cond.1: ; preds = %for.inc, %for.body 97 %1 = load i32, ptr %i, align 4, !dbg !51 98 %cmp2 = icmp slt i32 %1, 200000, !dbg !55 99 br i1 %cmp2, label %for.body.3, label %for.end, !dbg !56 100; CHECK: !prof ![[PROF7:[0-9]+]] 101 102for.body.3: ; preds = %for.cond.1 103 %2 = load i32, ptr %i, align 4, !dbg !57 104 %conv = sext i32 %2 to i64, !dbg !57 105 %call = call i64 @_Z3fool(i64 %conv), !dbg !59 106; CHECK: !prof ![[PROF8:[0-9]+]] 107 %3 = load i64, ptr %sum, align 8, !dbg !60 108 %add = add nsw i64 %3, %call, !dbg !60 109 store i64 %add, ptr %sum, align 8, !dbg !60 110 br label %for.inc, !dbg !61 111 112for.inc: ; preds = %for.body.3 113 %4 = load i32, ptr %i, align 4, !dbg !62 114 %inc = add nsw i32 %4, 1, !dbg !62 115 store i32 %inc, ptr %i, align 4, !dbg !62 116 br label %for.cond.1, !dbg !64 117 118for.end: ; preds = %for.cond.1 119 br label %for.inc.4, !dbg !65 120 121for.inc.4: ; preds = %for.end 122 %5 = load i32, ptr %k, align 4, !dbg !67 123 %inc5 = add nsw i32 %5, 1, !dbg !67 124 store i32 %inc5, ptr %k, align 4, !dbg !67 125 br label %for.cond, !dbg !68 126 127for.end.6: ; preds = %for.cond 128 %6 = load i64, ptr %sum, align 8, !dbg !69 129 %cmp7 = icmp sgt i64 %6, 0, !dbg !70 130 %cond = select i1 %cmp7, i32 0, i32 1, !dbg !69 131 ret i32 %cond, !dbg !71 132} 133 134; CHECK: ![[EC1]] = !{!"function_entry_count", i64 24109} 135; CHECK: ![[PROF1]] = !{!"branch_weights", i32 24109} 136; CHECK: ![[PROF2]] = !{!"branch_weights", i32 30125, i32 30125} 137; CHECK: ![[PROF3]] = !{!"branch_weights", i32 30124} 138; CHECK: ![[PROF4]] = !{!"branch_weights", i32 30125, i32 29580} 139; CHECK: ![[EC2]] = !{!"function_entry_count", i64 1} 140; CHECK: ![[PROF6]] = !{!"branch_weights", i32 1, i32 2} 141; CHECK: ![[PROF7]] = !{!"branch_weights", i32 18943, i32 1} 142; CHECK: ![[PROF8]] = !{!"branch_weights", i32 18942} 143 144attributes #0 = { nounwind uwtable "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+sse,+sse2" "unsafe-fp-math"="false" "use-soft-float"="false" "use-sample-profile" } 145attributes #1 = { nounwind readnone } 146attributes #2 = { nounwind "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+sse,+sse2" "unsafe-fp-math"="false" "use-soft-float"="false" } 147attributes #3 = { nounwind } 148 149!llvm.dbg.cu = !{!0} 150!llvm.module.flags = !{!13, !14} 151!llvm.ident = !{!15} 152 153!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !1, producer: "clang version 3.8.0 (trunk 247554) (llvm/trunk 247557)", isOptimized: false, runtimeVersion: 0, emissionKind: NoDebug, enums: !2) 154!1 = !DIFile(filename: "discriminator.cc", directory: "/usr/local/google/home/dnovillo/llvm/test/autofdo") 155!2 = !{} 156!4 = distinct !DISubprogram(name: "foo", linkageName: "_Z3fool", scope: !1, file: !1, line: 3, type: !5, isLocal: false, isDefinition: true, scopeLine: 3, flags: DIFlagPrototyped, isOptimized: false, unit: !0, retainedNodes: !2) 157!5 = !DISubroutineType(types: !6) 158!6 = !{!7, !8} 159!7 = !DIBasicType(name: "long long int", size: 64, align: 64, encoding: DW_ATE_signed) 160!8 = !DIBasicType(name: "long int", size: 64, align: 64, encoding: DW_ATE_signed) 161!9 = distinct !DISubprogram(name: "main", scope: !1, file: !1, line: 7, type: !10, isLocal: false, isDefinition: true, scopeLine: 7, flags: DIFlagPrototyped, isOptimized: false, unit: !0, retainedNodes: !2) 162!10 = !DISubroutineType(types: !11) 163!11 = !{!12} 164!12 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed) 165!13 = !{i32 2, !"Dwarf Version", i32 4} 166!14 = !{i32 2, !"Debug Info Version", i32 3} 167!15 = !{!"clang version 3.8.0 (trunk 247554) (llvm/trunk 247557)"} 168!16 = !DILocalVariable(name: "i", arg: 1, scope: !4, file: !1, line: 3, type: !8) 169!17 = !DIExpression() 170!18 = !DILocation(line: 3, column: 24, scope: !4) 171!19 = !DILocation(line: 4, column: 7, scope: !20) 172!20 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 7) 173!21 = !DILocation(line: 4, column: 14, scope: !20) 174!22 = !DILocation(line: 4, column: 7, scope: !4) 175!23 = !DILocation(line: 4, column: 21, scope: !24) 176!24 = !DILexicalBlockFile(scope: !20, file: !1, discriminator: 1) 177!25 = !DILocation(line: 4, column: 40, scope: !26) 178!26 = !DILexicalBlockFile(scope: !27, file: !1, discriminator: 2) 179!27 = distinct !DILexicalBlock(scope: !20, file: !1, line: 4, column: 40) 180!28 = !DILocation(line: 4, column: 47, scope: !27) 181!29 = !DILocation(line: 4, column: 40, scope: !20) 182!30 = !DILocation(line: 4, column: 55, scope: !31) 183!31 = !DILexicalBlockFile(scope: !27, file: !1, discriminator: 3) 184!32 = !DILocation(line: 4, column: 71, scope: !33) 185!33 = !DILexicalBlockFile(scope: !27, file: !1, discriminator: 4) 186!34 = !DILocation(line: 5, column: 1, scope: !4) 187!35 = !DILocalVariable(name: "sum", scope: !9, file: !1, line: 8, type: !7) 188!36 = !DILocation(line: 8, column: 17, scope: !9) 189!37 = !DILocalVariable(name: "k", scope: !38, file: !1, line: 9, type: !12) 190!38 = distinct !DILexicalBlock(scope: !9, file: !1, line: 9, column: 3) 191!39 = !DILocation(line: 9, column: 12, scope: !38) 192!40 = !DILocation(line: 9, column: 8, scope: !38) 193!41 = !DILocation(line: 9, column: 19, scope: !42) 194!42 = !DILexicalBlockFile(scope: !43, file: !1, discriminator: 2) 195!43 = !DILexicalBlockFile(scope: !44, file: !1, discriminator: 1) 196!44 = distinct !DILexicalBlock(scope: !38, file: !1, line: 9, column: 3) 197!45 = !DILocation(line: 9, column: 21, scope: !44) 198!46 = !DILocation(line: 9, column: 3, scope: !38) 199!47 = !DILocalVariable(name: "i", scope: !48, file: !1, line: 10, type: !12) 200!48 = distinct !DILexicalBlock(scope: !44, file: !1, line: 10, column: 5) 201!49 = !DILocation(line: 10, column: 14, scope: !48) 202!50 = !DILocation(line: 10, column: 10, scope: !48) 203!51 = !DILocation(line: 10, column: 21, scope: !52) 204!52 = !DILexicalBlockFile(scope: !53, file: !1, discriminator: 5) 205!53 = !DILexicalBlockFile(scope: !54, file: !1, discriminator: 1) 206!54 = distinct !DILexicalBlock(scope: !48, file: !1, line: 10, column: 5) 207!55 = !DILocation(line: 10, column: 23, scope: !54) 208!56 = !DILocation(line: 10, column: 5, scope: !48) 209!57 = !DILocation(line: 10, column: 49, scope: !58) 210!58 = !DILexicalBlockFile(scope: !54, file: !1, discriminator: 2) 211!59 = !DILocation(line: 10, column: 45, scope: !54) 212!60 = !DILocation(line: 10, column: 42, scope: !54) 213!61 = !DILocation(line: 10, column: 38, scope: !54) 214!62 = !DILocation(line: 10, column: 34, scope: !63) 215!63 = !DILexicalBlockFile(scope: !54, file: !1, discriminator: 4) 216!64 = !DILocation(line: 10, column: 5, scope: !54) 217!65 = !DILocation(line: 10, column: 50, scope: !66) 218!66 = !DILexicalBlockFile(scope: !48, file: !1, discriminator: 3) 219!67 = !DILocation(line: 9, column: 30, scope: !44) 220!68 = !DILocation(line: 9, column: 3, scope: !44) 221!69 = !DILocation(line: 11, column: 10, scope: !9) 222!70 = !DILocation(line: 11, column: 14, scope: !9) 223!71 = !DILocation(line: 11, column: 3, scope: !9) 224