1 ; RUN: opt < %s -sample-profile -sample-profile-file=%S/Inputs/remarks.prof -S -pass-remarks=sample-profile 2>&1 | FileCheck %s 2 ; RUN: opt < %s -passes=sample-profile -sample-profile-file=%S/Inputs/remarks.prof -S -pass-remarks=sample-profile 2>&1 | FileCheck %s 3 ; 4 ; Original test case. 5 ; 6 ; 1 #include <stdlib.h> 7 ; 2 8 ; 3 long long foo() { 9 ; 4 long long int sum = 0; 10 ; 5 for (int i = 0; i < 500000000; i++) 11 ; 6 if (i < 1000) 12 ; 7 sum -= i; 13 ; 8 else 14 ; 9 sum += -i * rand(); 15 ; 10 return sum; 16 ; 11 } 17 ; 12 18 ; 13 int main() { return foo() > 0; } 19 20 ; We are expecting foo() to be inlined in main() (almost all the cycles are 21 ; spent inside foo). 22 ; CHECK: remark: remarks.cc:13:21: inlined hot callee '_Z3foov' with 623868 samples into 'main' 23 24 ; The back edge for the loop is the hottest edge in the loop subgraph. 25 ; CHECK: remark: remarks.cc:6:9: most popular destination for conditional branches at remarks.cc:5:3 26 27 ; The predicate almost always chooses the 'else' branch. 28 ; CHECK: remark: remarks.cc:9:15: most popular destination for conditional branches at remarks.cc:6:9 29 30 ; Function Attrs: nounwind uwtable 31 define i64 @_Z3foov() #0 !dbg !4 { 32 entry: 33 %sum = alloca i64, align 8 34 %i = alloca i32, align 4 35 %0 = bitcast i64* %sum to i8*, !dbg !19 36 call void @llvm.lifetime.start(i64 8, i8* %0) #4, !dbg !19 37 call void @llvm.dbg.declare(metadata i64* %sum, metadata !9, metadata !20), !dbg !21 38 store i64 0, i64* %sum, align 8, !dbg !21, !tbaa !22 39 %1 = bitcast i32* %i to i8*, !dbg !26 40 call void @llvm.lifetime.start(i64 4, i8* %1) #4, !dbg !26 41 call void @llvm.dbg.declare(metadata i32* %i, metadata !10, metadata !20), !dbg !27 42 store i32 0, i32* %i, align 4, !dbg !27, !tbaa !28 43 br label %for.cond, !dbg !26 44 45 for.cond: ; preds = %for.inc, %entry 46 %2 = load i32, i32* %i, align 4, !dbg !30, !tbaa !28 47 %cmp = icmp slt i32 %2, 500000000, !dbg !34 48 br i1 %cmp, label %for.body, label %for.cond.cleanup, !dbg !35 49 50 for.cond.cleanup: ; preds = %for.cond 51 %3 = bitcast i32* %i to i8*, !dbg !36 52 call void @llvm.lifetime.end(i64 4, i8* %3) #4, !dbg !36 53 br label %for.end 54 55 for.body: ; preds = %for.cond 56 %4 = load i32, i32* %i, align 4, !dbg !38, !tbaa !28 57 %cmp1 = icmp slt i32 %4, 1000, !dbg !40 58 br i1 %cmp1, label %if.then, label %if.else, !dbg !41 59 60 if.then: ; preds = %for.body 61 %5 = load i32, i32* %i, align 4, !dbg !42, !tbaa !28 62 %conv = sext i32 %5 to i64, !dbg !42 63 %6 = load i64, i64* %sum, align 8, !dbg !43, !tbaa !22 64 %sub = sub nsw i64 %6, %conv, !dbg !43 65 store i64 %sub, i64* %sum, align 8, !dbg !43, !tbaa !22 66 br label %if.end, !dbg !44 67 68 if.else: ; preds = %for.body 69 %7 = load i32, i32* %i, align 4, !dbg !45, !tbaa !28 70 %sub2 = sub nsw i32 0, %7, !dbg !46 71 %call = call i32 @rand() #4, !dbg !47 72 %mul = mul nsw i32 %sub2, %call, !dbg !48 73 %conv3 = sext i32 %mul to i64, !dbg !46 74 %8 = load i64, i64* %sum, align 8, !dbg !49, !tbaa !22 75 %add = add nsw i64 %8, %conv3, !dbg !49 76 store i64 %add, i64* %sum, align 8, !dbg !49, !tbaa !22 77 br label %if.end 78 79 if.end: ; preds = %if.else, %if.then 80 br label %for.inc, !dbg !50 81 82 for.inc: ; preds = %if.end 83 %9 = load i32, i32* %i, align 4, !dbg !51, !tbaa !28 84 %inc = add nsw i32 %9, 1, !dbg !51 85 store i32 %inc, i32* %i, align 4, !dbg !51, !tbaa !28 86 br label %for.cond, !dbg !52 87 88 for.end: ; preds = %for.cond.cleanup 89 %10 = load i64, i64* %sum, align 8, !dbg !53, !tbaa !22 90 %11 = bitcast i64* %sum to i8*, !dbg !54 91 call void @llvm.lifetime.end(i64 8, i8* %11) #4, !dbg !54 92 ret i64 %10, !dbg !55 93 } 94 95 ; Function Attrs: nounwind argmemonly 96 declare void @llvm.lifetime.start(i64, i8* nocapture) #1 97 98 ; Function Attrs: nounwind readnone 99 declare void @llvm.dbg.declare(metadata, metadata, metadata) #2 100 101 ; Function Attrs: nounwind 102 declare i32 @rand() #3 103 104 ; Function Attrs: nounwind argmemonly 105 declare void @llvm.lifetime.end(i64, i8* nocapture) #1 106 107 ; Function Attrs: nounwind uwtable 108 define i32 @main() #0 !dbg !13 { 109 entry: 110 %retval = alloca i32, align 4 111 store i32 0, i32* %retval, align 4 112 %call = call i64 @_Z3foov(), !dbg !56 113 %cmp = icmp sgt i64 %call, 0, !dbg !57 114 %conv = zext i1 %cmp to i32, !dbg !56 115 ret i32 %conv, !dbg !58 116 } 117 118 attributes #0 = { nounwind uwtable "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2" "unsafe-fp-math"="false" "use-soft-float"="false" } 119 attributes #1 = { nounwind argmemonly } 120 attributes #2 = { nounwind readnone } 121 attributes #3 = { nounwind "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2" "unsafe-fp-math"="false" "use-soft-float"="false" } 122 attributes #4 = { nounwind } 123 124 !llvm.dbg.cu = !{!0} 125 !llvm.module.flags = !{!16, !17} 126 !llvm.ident = !{!18} 127 128 !0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !1, producer: "clang version 3.8.0 (trunk 251041) (llvm/trunk 251053)", isOptimized: true, runtimeVersion: 0, emissionKind: NoDebug, enums: !2) 129 !1 = !DIFile(filename: "remarks.cc", directory: ".") 130 !2 = !{} 131 !4 = distinct !DISubprogram(name: "foo", linkageName: "_Z3foov", scope: !1, file: !1, line: 3, type: !5, isLocal: false, isDefinition: true, scopeLine: 3, flags: DIFlagPrototyped, isOptimized: true, unit: !0, variables: !8) 132 !5 = !DISubroutineType(types: !6) 133 !6 = !{!7} 134 !7 = !DIBasicType(name: "long long int", size: 64, align: 64, encoding: DW_ATE_signed) 135 !8 = !{!9, !10} 136 !9 = !DILocalVariable(name: "sum", scope: !4, file: !1, line: 4, type: !7) 137 !10 = !DILocalVariable(name: "i", scope: !11, file: !1, line: 5, type: !12) 138 !11 = distinct !DILexicalBlock(scope: !4, file: !1, line: 5, column: 3) 139 !12 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed) 140 !13 = distinct !DISubprogram(name: "main", scope: !1, file: !1, line: 13, type: !14, isLocal: false, isDefinition: true, scopeLine: 13, flags: DIFlagPrototyped, isOptimized: true, unit: !0, variables: !2) 141 !14 = !DISubroutineType(types: !15) 142 !15 = !{!12} 143 !16 = !{i32 2, !"Dwarf Version", i32 4} 144 !17 = !{i32 2, !"Debug Info Version", i32 3} 145 !18 = !{!"clang version 3.8.0 (trunk 251041) (llvm/trunk 251053)"} 146 !19 = !DILocation(line: 4, column: 3, scope: !4) 147 !20 = !DIExpression() 148 !21 = !DILocation(line: 4, column: 17, scope: !4) 149 !22 = !{!23, !23, i64 0} 150 !23 = !{!"long long", !24, i64 0} 151 !24 = !{!"omnipotent char", !25, i64 0} 152 !25 = !{!"Simple C/C++ TBAA"} 153 !26 = !DILocation(line: 5, column: 8, scope: !11) 154 !27 = !DILocation(line: 5, column: 12, scope: !11) 155 !28 = !{!29, !29, i64 0} 156 !29 = !{!"int", !24, i64 0} 157 !30 = !DILocation(line: 5, column: 19, scope: !31) 158 !31 = !DILexicalBlockFile(scope: !32, file: !1, discriminator: 3) 159 !32 = !DILexicalBlockFile(scope: !33, file: !1, discriminator: 1) 160 !33 = distinct !DILexicalBlock(scope: !11, file: !1, line: 5, column: 3) 161 !34 = !DILocation(line: 5, column: 21, scope: !33) 162 !35 = !DILocation(line: 5, column: 3, scope: !11) 163 !36 = !DILocation(line: 5, column: 3, scope: !37) 164 !37 = !DILexicalBlockFile(scope: !33, file: !1, discriminator: 2) 165 !38 = !DILocation(line: 6, column: 9, scope: !39) 166 !39 = distinct !DILexicalBlock(scope: !33, file: !1, line: 6, column: 9) 167 !40 = !DILocation(line: 6, column: 11, scope: !39) 168 !41 = !DILocation(line: 6, column: 9, scope: !33) 169 !42 = !DILocation(line: 7, column: 14, scope: !39) 170 !43 = !DILocation(line: 7, column: 11, scope: !39) 171 !44 = !DILocation(line: 7, column: 7, scope: !39) 172 !45 = !DILocation(line: 9, column: 15, scope: !39) 173 !46 = !DILocation(line: 9, column: 14, scope: !39) 174 !47 = !DILocation(line: 9, column: 19, scope: !39) 175 !48 = !DILocation(line: 9, column: 17, scope: !39) 176 !49 = !DILocation(line: 9, column: 11, scope: !39) 177 !50 = !DILocation(line: 6, column: 13, scope: !39) 178 !51 = !DILocation(line: 5, column: 35, scope: !33) 179 !52 = !DILocation(line: 5, column: 3, scope: !33) 180 !53 = !DILocation(line: 10, column: 10, scope: !4) 181 !54 = !DILocation(line: 11, column: 1, scope: !4) 182 !55 = !DILocation(line: 10, column: 3, scope: !4) 183 !56 = !DILocation(line: 13, column: 21, scope: !13) 184 !57 = !DILocation(line: 13, column: 27, scope: !13) 185 !58 = !DILocation(line: 13, column: 14, scope: !13) 186
