1 ; RUN: opt < %s -sample-profile -sample-profile-file=%S/Inputs/gcc-simple.afdo -S | FileCheck %s
2 ; XFAIL: powerpc64-, s390x, mips-, mips64-, sparc
7 ; long long int foo(long i) {
8 ; if (rand() < 500) return 2; else if (rand() > 5000) return 10; else return 90;
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;
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
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
27 define i64 @_Z3fool(i64 %i) #0 !dbg !4 {
28 ; CHECK: !prof ![[EC1:[0-9]+]]
30 %retval = alloca i64, align 8
31 %i.addr = alloca i64, align 8
32 store i64 %i, i64* %i.addr, align 8
33 call void @llvm.dbg.declare(metadata i64* %i.addr, metadata !16, metadata !17), !dbg !18
34 %call = call i32 @rand() #3, !dbg !19
35 %cmp = icmp slt i32 %call, 500, !dbg !21
36 br i1 %cmp, label %if.then, label %if.else, !dbg !22
37 ; CHECK: !prof ![[PROF1:[0-9]+]]
39 if.then: ; preds = %entry
40 store i64 2, i64* %retval, align 8, !dbg !23
41 br label %return, !dbg !23
43 if.else: ; preds = %entry
44 %call1 = call i32 @rand() #3, !dbg !25
45 %cmp2 = icmp sgt i32 %call1, 5000, !dbg !28
46 br i1 %cmp2, label %if.then.3, label %if.else.4, !dbg !29
47 ; CHECK: !prof ![[PROF2:[0-9]+]]
49 if.then.3: ; preds = %if.else
50 store i64 10, i64* %retval, align 8, !dbg !30
51 br label %return, !dbg !30
53 if.else.4: ; preds = %if.else
54 store i64 90, i64* %retval, align 8, !dbg !32
55 br label %return, !dbg !32
57 return: ; preds = %if.else.4, %if.then.3, %if.then
58 %0 = load i64, i64* %retval, align 8, !dbg !34
62 ; Function Attrs: nounwind readnone
63 declare void @llvm.dbg.declare(metadata, metadata, metadata) #1
65 ; Function Attrs: nounwind
66 declare i32 @rand() #2
68 ; Function Attrs: nounwind uwtable
69 define i32 @main() #0 !dbg !9 {
70 ; CHECK: !prof ![[EC2:[0-9]+]]
72 %retval = alloca i32, align 4
73 %sum = alloca i64, align 8
74 %k = alloca i32, align 4
75 %i = alloca i32, align 4
76 store i32 0, i32* %retval, align 4
77 call void @llvm.dbg.declare(metadata i64* %sum, metadata !35, metadata !17), !dbg !36
78 store i64 0, i64* %sum, align 8, !dbg !36
79 call void @llvm.dbg.declare(metadata i32* %k, metadata !37, metadata !17), !dbg !39
80 store i32 0, i32* %k, align 4, !dbg !39
81 br label %for.cond, !dbg !40
83 for.cond: ; preds = %for.inc.4, %entry
84 %0 = load i32, i32* %k, align 4, !dbg !41
85 %cmp = icmp slt i32 %0, 3000, !dbg !45
86 br i1 %cmp, label %for.body, label %for.end.6, !dbg !46
87 ; CHECK: !prof ![[PROF3:[0-9]+]]
89 for.body: ; preds = %for.cond
90 call void @llvm.dbg.declare(metadata i32* %i, metadata !47, metadata !17), !dbg !49
91 store i32 0, i32* %i, align 4, !dbg !49
92 br label %for.cond.1, !dbg !50
94 for.cond.1: ; preds = %for.inc, %for.body
95 %1 = load i32, i32* %i, align 4, !dbg !51
96 %cmp2 = icmp slt i32 %1, 200000, !dbg !55
97 br i1 %cmp2, label %for.body.3, label %for.end, !dbg !56
98 ; CHECK: !prof ![[PROF4:[0-9]+]]
100 for.body.3: ; preds = %for.cond.1
101 %2 = load i32, i32* %i, align 4, !dbg !57
102 %conv = sext i32 %2 to i64, !dbg !57
103 %call = call i64 @_Z3fool(i64 %conv), !dbg !59
104 %3 = load i64, i64* %sum, align 8, !dbg !60
105 %add = add nsw i64 %3, %call, !dbg !60
106 store i64 %add, i64* %sum, align 8, !dbg !60
107 br label %for.inc, !dbg !61
109 for.inc: ; preds = %for.body.3
110 %4 = load i32, i32* %i, align 4, !dbg !62
111 %inc = add nsw i32 %4, 1, !dbg !62
112 store i32 %inc, i32* %i, align 4, !dbg !62
113 br label %for.cond.1, !dbg !64
115 for.end: ; preds = %for.cond.1
116 br label %for.inc.4, !dbg !65
118 for.inc.4: ; preds = %for.end
119 %5 = load i32, i32* %k, align 4, !dbg !67
120 %inc5 = add nsw i32 %5, 1, !dbg !67
121 store i32 %inc5, i32* %k, align 4, !dbg !67
122 br label %for.cond, !dbg !68
124 for.end.6: ; preds = %for.cond
125 %6 = load i64, i64* %sum, align 8, !dbg !69
126 %cmp7 = icmp sgt i64 %6, 0, !dbg !70
127 %cond = select i1 %cmp7, i32 0, i32 1, !dbg !69
128 ret i32 %cond, !dbg !71
131 ; CHECK ![[EC1]] = !{!"function_entry_count", i64 24108}
132 ; CHECK ![[PROF1]] = !{!"branch_weights", i32 1, i32 30124}
133 ; CHECK ![[PROF2]] = !{!"branch_weights", i32 30177, i32 29579}
134 ; CHECK ![[EC2]] = !{!"function_entry_count", i64 0}
135 ; CHECK ![[PROF3]] = !{!"branch_weights", i32 1, i32 1}
136 ; CHECK ![[PROF4]] = !{!"branch_weights", i32 1, i32 20238}
138 attributes #0 = { nounwind uwtable "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "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" }
139 attributes #1 = { nounwind readnone }
140 attributes #2 = { nounwind "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "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" }
141 attributes #3 = { nounwind }
144 !llvm.module.flags = !{!13, !14}
147 !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: 1, enums: !2, subprograms: !3)
148 !1 = !DIFile(filename: "discriminator.cc", directory: "/usr/local/google/home/dnovillo/llvm/test/autofdo")
151 !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, variables: !2)
152 !5 = !DISubroutineType(types: !6)
154 !7 = !DIBasicType(name: "long long int", size: 64, align: 64, encoding: DW_ATE_signed)
155 !8 = !DIBasicType(name: "long int", size: 64, align: 64, encoding: DW_ATE_signed)
156 !9 = distinct !DISubprogram(name: "main", scope: !1, file: !1, line: 7, type: !10, isLocal: false, isDefinition: true, scopeLine: 7, flags: DIFlagPrototyped, isOptimized: false, variables: !2)
157 !10 = !DISubroutineType(types: !11)
159 !12 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
160 !13 = !{i32 2, !"Dwarf Version", i32 4}
161 !14 = !{i32 2, !"Debug Info Version", i32 3}
162 !15 = !{!"clang version 3.8.0 (trunk 247554) (llvm/trunk 247557)"}
163 !16 = !DILocalVariable(name: "i", arg: 1, scope: !4, file: !1, line: 3, type: !8)
164 !17 = !DIExpression()
165 !18 = !DILocation(line: 3, column: 24, scope: !4)
166 !19 = !DILocation(line: 4, column: 7, scope: !20)
167 !20 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 7)
168 !21 = !DILocation(line: 4, column: 14, scope: !20)
169 !22 = !DILocation(line: 4, column: 7, scope: !4)
170 !23 = !DILocation(line: 4, column: 21, scope: !24)
171 !24 = !DILexicalBlockFile(scope: !20, file: !1, discriminator: 1)
172 !25 = !DILocation(line: 4, column: 40, scope: !26)
173 !26 = !DILexicalBlockFile(scope: !27, file: !1, discriminator: 2)
174 !27 = distinct !DILexicalBlock(scope: !20, file: !1, line: 4, column: 40)
175 !28 = !DILocation(line: 4, column: 47, scope: !27)
176 !29 = !DILocation(line: 4, column: 40, scope: !20)
177 !30 = !DILocation(line: 4, column: 55, scope: !31)
178 !31 = !DILexicalBlockFile(scope: !27, file: !1, discriminator: 3)
179 !32 = !DILocation(line: 4, column: 71, scope: !33)
180 !33 = !DILexicalBlockFile(scope: !27, file: !1, discriminator: 4)
181 !34 = !DILocation(line: 5, column: 1, scope: !4)
182 !35 = !DILocalVariable(name: "sum", scope: !9, file: !1, line: 8, type: !7)
183 !36 = !DILocation(line: 8, column: 17, scope: !9)
184 !37 = !DILocalVariable(name: "k", scope: !38, file: !1, line: 9, type: !12)
185 !38 = distinct !DILexicalBlock(scope: !9, file: !1, line: 9, column: 3)
186 !39 = !DILocation(line: 9, column: 12, scope: !38)
187 !40 = !DILocation(line: 9, column: 8, scope: !38)
188 !41 = !DILocation(line: 9, column: 19, scope: !42)
189 !42 = !DILexicalBlockFile(scope: !43, file: !1, discriminator: 2)
190 !43 = !DILexicalBlockFile(scope: !44, file: !1, discriminator: 1)
191 !44 = distinct !DILexicalBlock(scope: !38, file: !1, line: 9, column: 3)
192 !45 = !DILocation(line: 9, column: 21, scope: !44)
193 !46 = !DILocation(line: 9, column: 3, scope: !38)
194 !47 = !DILocalVariable(name: "i", scope: !48, file: !1, line: 10, type: !12)
195 !48 = distinct !DILexicalBlock(scope: !44, file: !1, line: 10, column: 5)
196 !49 = !DILocation(line: 10, column: 14, scope: !48)
197 !50 = !DILocation(line: 10, column: 10, scope: !48)
198 !51 = !DILocation(line: 10, column: 21, scope: !52)
199 !52 = !DILexicalBlockFile(scope: !53, file: !1, discriminator: 5)
200 !53 = !DILexicalBlockFile(scope: !54, file: !1, discriminator: 1)
201 !54 = distinct !DILexicalBlock(scope: !48, file: !1, line: 10, column: 5)
202 !55 = !DILocation(line: 10, column: 23, scope: !54)
203 !56 = !DILocation(line: 10, column: 5, scope: !48)
204 !57 = !DILocation(line: 10, column: 49, scope: !58)
205 !58 = !DILexicalBlockFile(scope: !54, file: !1, discriminator: 2)
206 !59 = !DILocation(line: 10, column: 45, scope: !54)
207 !60 = !DILocation(line: 10, column: 42, scope: !54)
208 !61 = !DILocation(line: 10, column: 38, scope: !54)
209 !62 = !DILocation(line: 10, column: 34, scope: !63)
210 !63 = !DILexicalBlockFile(scope: !54, file: !1, discriminator: 4)
211 !64 = !DILocation(line: 10, column: 5, scope: !54)
212 !65 = !DILocation(line: 10, column: 50, scope: !66)
213 !66 = !DILexicalBlockFile(scope: !48, file: !1, discriminator: 3)
214 !67 = !DILocation(line: 9, column: 30, scope: !44)
215 !68 = !DILocation(line: 9, column: 3, scope: !44)
216 !69 = !DILocation(line: 11, column: 10, scope: !9)
217 !70 = !DILocation(line: 11, column: 14, scope: !9)
218 !71 = !DILocation(line: 11, column: 3, scope: !9)