1; RUN: opt < %s -sample-profile -sample-profile-file=%S/Inputs/gcc-simple.afdo -S | FileCheck %s
2; RUN: opt < %s -passes=sample-profile -sample-profile-file=%S/Inputs/gcc-simple.afdo -S | FileCheck %s
3;
4; Original code:
5;
6; #include <stdlib.h>
7;
8; long long int foo(long i) {
9;   if (rand() < 500) return 2; else if (rand() > 5000) return 10; else return 90;
10; }
11;
12; int main() {
13;   long long int sum = 0;
14;   for (int k = 0; k < 3000; k++)
15;     for (int i = 0; i < 200000; i++) sum += foo(i);
16;   return sum > 0 ? 0 : 1;
17; }
18;
19; This test was compiled down to bytecode at -O0 to avoid inlining foo() into
20; main(). The profile was generated using a GCC-generated binary (also compiled
21; at -O0). The conversion from the Linux Perf profile to the GCC autofdo
22; profile used the converter at https://github.com/google/autofdo
23;
24; $ gcc -g -O0 gcc-simple.cc -o gcc-simple
25; $ perf record -b ./gcc-simple
26; $ create_gcov --binary=gcc-simple --gcov=gcc-simple.afdo
27
28define i64 @_Z3fool(i64 %i) #0 !dbg !4 {
29; CHECK: !prof ![[EC1:[0-9]+]]
30entry:
31  %retval = alloca i64, align 8
32  %i.addr = alloca i64, align 8
33  store i64 %i, i64* %i.addr, align 8
34  call void @llvm.dbg.declare(metadata i64* %i.addr, metadata !16, metadata !17), !dbg !18
35  %call = call i32 @rand() #3, !dbg !19
36; CHECK: !prof ![[PROF1:[0-9]+]]
37  %cmp = icmp slt i32 %call, 500, !dbg !21
38  br i1 %cmp, label %if.then, label %if.else, !dbg !22
39; CHECK: !prof ![[PROF2:[0-9]+]]
40
41if.then:                                          ; preds = %entry
42  store i64 2, i64* %retval, align 8, !dbg !23
43  br label %return, !dbg !23
44
45if.else:                                          ; preds = %entry
46  %call1 = call i32 @rand() #3, !dbg !25
47; CHECK: !prof ![[PROF3:[0-9]+]]
48  %cmp2 = icmp sgt i32 %call1, 5000, !dbg !28
49  br i1 %cmp2, label %if.then.3, label %if.else.4, !dbg !29
50; CHECK: !prof ![[PROF4:[0-9]+]]
51
52if.then.3:                                        ; preds = %if.else
53  store i64 10, i64* %retval, align 8, !dbg !30
54  br label %return, !dbg !30
55
56if.else.4:                                        ; preds = %if.else
57  store i64 90, i64* %retval, align 8, !dbg !32
58  br label %return, !dbg !32
59
60return:                                           ; preds = %if.else.4, %if.then.3, %if.then
61  %0 = load i64, i64* %retval, align 8, !dbg !34
62  ret i64 %0, !dbg !34
63}
64
65; Function Attrs: nounwind readnone
66declare void @llvm.dbg.declare(metadata, metadata, metadata) #1
67
68; Function Attrs: nounwind
69declare i32 @rand() #2
70
71; Function Attrs: nounwind uwtable
72define i32 @main() #0 !dbg !9 {
73; CHECK: !prof ![[EC2:[0-9]+]]
74entry:
75  %retval = alloca i32, align 4
76  %sum = alloca i64, align 8
77  %k = alloca i32, align 4
78  %i = alloca i32, align 4
79  store i32 0, i32* %retval, align 4
80  call void @llvm.dbg.declare(metadata i64* %sum, metadata !35, metadata !17), !dbg !36
81  store i64 0, i64* %sum, align 8, !dbg !36
82  call void @llvm.dbg.declare(metadata i32* %k, metadata !37, metadata !17), !dbg !39
83  store i32 0, i32* %k, align 4, !dbg !39
84  br label %for.cond, !dbg !40
85
86for.cond:                                         ; preds = %for.inc.4, %entry
87  %0 = load i32, i32* %k, align 4, !dbg !41
88  %cmp = icmp slt i32 %0, 3000, !dbg !45
89  br i1 %cmp, label %for.body, label %for.end.6, !dbg !46
90; CHECK: !prof ![[PROF6:[0-9]+]]
91
92for.body:                                         ; preds = %for.cond
93  call void @llvm.dbg.declare(metadata i32* %i, metadata !47, metadata !17), !dbg !49
94  store i32 0, i32* %i, align 4, !dbg !49
95  br label %for.cond.1, !dbg !50
96
97for.cond.1:                                       ; preds = %for.inc, %for.body
98  %1 = load i32, i32* %i, align 4, !dbg !51
99  %cmp2 = icmp slt i32 %1, 200000, !dbg !55
100  br i1 %cmp2, label %for.body.3, label %for.end, !dbg !56
101; CHECK: !prof ![[PROF7:[0-9]+]]
102
103for.body.3:                                       ; preds = %for.cond.1
104  %2 = load i32, i32* %i, align 4, !dbg !57
105  %conv = sext i32 %2 to i64, !dbg !57
106  %call = call i64 @_Z3fool(i64 %conv), !dbg !59
107; CHECK: !prof ![[PROF8:[0-9]+]]
108  %3 = load i64, i64* %sum, align 8, !dbg !60
109  %add = add nsw i64 %3, %call, !dbg !60
110  store i64 %add, i64* %sum, align 8, !dbg !60
111  br label %for.inc, !dbg !61
112
113for.inc:                                          ; preds = %for.body.3
114  %4 = load i32, i32* %i, align 4, !dbg !62
115  %inc = add nsw i32 %4, 1, !dbg !62
116  store i32 %inc, i32* %i, align 4, !dbg !62
117  br label %for.cond.1, !dbg !64
118
119for.end:                                          ; preds = %for.cond.1
120  br label %for.inc.4, !dbg !65
121
122for.inc.4:                                        ; preds = %for.end
123  %5 = load i32, i32* %k, align 4, !dbg !67
124  %inc5 = add nsw i32 %5, 1, !dbg !67
125  store i32 %inc5, i32* %k, align 4, !dbg !67
126  br label %for.cond, !dbg !68
127
128for.end.6:                                        ; preds = %for.cond
129  %6 = load i64, i64* %sum, align 8, !dbg !69
130  %cmp7 = icmp sgt i64 %6, 0, !dbg !70
131  %cond = select i1 %cmp7, i32 0, i32 1, !dbg !69
132  ret i32 %cond, !dbg !71
133}
134
135; CHECK: ![[EC1]] = !{!"function_entry_count", i64 24109}
136; CHECK: ![[PROF1]] = !{!"branch_weights", i32 24109}
137; CHECK: ![[PROF2]] = !{!"branch_weights", i32 30125, i32 30125}
138; CHECK: ![[PROF3]] = !{!"branch_weights", i32 30124}
139; CHECK: ![[PROF4]] = !{!"branch_weights", i32 30125, i32 29580}
140; CHECK: ![[EC2]] = !{!"function_entry_count", i64 1}
141; CHECK: ![[PROF6]] = !{!"branch_weights", i32 1, i32 2}
142; CHECK: ![[PROF7]] = !{!"branch_weights", i32 18943, i32 1}
143; CHECK: ![[PROF8]] = !{!"branch_weights", i32 18942}
144
145attributes #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" }
146attributes #1 = { nounwind readnone }
147attributes #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" }
148attributes #3 = { nounwind }
149
150!llvm.dbg.cu = !{!0}
151!llvm.module.flags = !{!13, !14}
152!llvm.ident = !{!15}
153
154!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)
155!1 = !DIFile(filename: "discriminator.cc", directory: "/usr/local/google/home/dnovillo/llvm/test/autofdo")
156!2 = !{}
157!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)
158!5 = !DISubroutineType(types: !6)
159!6 = !{!7, !8}
160!7 = !DIBasicType(name: "long long int", size: 64, align: 64, encoding: DW_ATE_signed)
161!8 = !DIBasicType(name: "long int", size: 64, align: 64, encoding: DW_ATE_signed)
162!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)
163!10 = !DISubroutineType(types: !11)
164!11 = !{!12}
165!12 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
166!13 = !{i32 2, !"Dwarf Version", i32 4}
167!14 = !{i32 2, !"Debug Info Version", i32 3}
168!15 = !{!"clang version 3.8.0 (trunk 247554) (llvm/trunk 247557)"}
169!16 = !DILocalVariable(name: "i", arg: 1, scope: !4, file: !1, line: 3, type: !8)
170!17 = !DIExpression()
171!18 = !DILocation(line: 3, column: 24, scope: !4)
172!19 = !DILocation(line: 4, column: 7, scope: !20)
173!20 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 7)
174!21 = !DILocation(line: 4, column: 14, scope: !20)
175!22 = !DILocation(line: 4, column: 7, scope: !4)
176!23 = !DILocation(line: 4, column: 21, scope: !24)
177!24 = !DILexicalBlockFile(scope: !20, file: !1, discriminator: 1)
178!25 = !DILocation(line: 4, column: 40, scope: !26)
179!26 = !DILexicalBlockFile(scope: !27, file: !1, discriminator: 2)
180!27 = distinct !DILexicalBlock(scope: !20, file: !1, line: 4, column: 40)
181!28 = !DILocation(line: 4, column: 47, scope: !27)
182!29 = !DILocation(line: 4, column: 40, scope: !20)
183!30 = !DILocation(line: 4, column: 55, scope: !31)
184!31 = !DILexicalBlockFile(scope: !27, file: !1, discriminator: 3)
185!32 = !DILocation(line: 4, column: 71, scope: !33)
186!33 = !DILexicalBlockFile(scope: !27, file: !1, discriminator: 4)
187!34 = !DILocation(line: 5, column: 1, scope: !4)
188!35 = !DILocalVariable(name: "sum", scope: !9, file: !1, line: 8, type: !7)
189!36 = !DILocation(line: 8, column: 17, scope: !9)
190!37 = !DILocalVariable(name: "k", scope: !38, file: !1, line: 9, type: !12)
191!38 = distinct !DILexicalBlock(scope: !9, file: !1, line: 9, column: 3)
192!39 = !DILocation(line: 9, column: 12, scope: !38)
193!40 = !DILocation(line: 9, column: 8, scope: !38)
194!41 = !DILocation(line: 9, column: 19, scope: !42)
195!42 = !DILexicalBlockFile(scope: !43, file: !1, discriminator: 2)
196!43 = !DILexicalBlockFile(scope: !44, file: !1, discriminator: 1)
197!44 = distinct !DILexicalBlock(scope: !38, file: !1, line: 9, column: 3)
198!45 = !DILocation(line: 9, column: 21, scope: !44)
199!46 = !DILocation(line: 9, column: 3, scope: !38)
200!47 = !DILocalVariable(name: "i", scope: !48, file: !1, line: 10, type: !12)
201!48 = distinct !DILexicalBlock(scope: !44, file: !1, line: 10, column: 5)
202!49 = !DILocation(line: 10, column: 14, scope: !48)
203!50 = !DILocation(line: 10, column: 10, scope: !48)
204!51 = !DILocation(line: 10, column: 21, scope: !52)
205!52 = !DILexicalBlockFile(scope: !53, file: !1, discriminator: 5)
206!53 = !DILexicalBlockFile(scope: !54, file: !1, discriminator: 1)
207!54 = distinct !DILexicalBlock(scope: !48, file: !1, line: 10, column: 5)
208!55 = !DILocation(line: 10, column: 23, scope: !54)
209!56 = !DILocation(line: 10, column: 5, scope: !48)
210!57 = !DILocation(line: 10, column: 49, scope: !58)
211!58 = !DILexicalBlockFile(scope: !54, file: !1, discriminator: 2)
212!59 = !DILocation(line: 10, column: 45, scope: !54)
213!60 = !DILocation(line: 10, column: 42, scope: !54)
214!61 = !DILocation(line: 10, column: 38, scope: !54)
215!62 = !DILocation(line: 10, column: 34, scope: !63)
216!63 = !DILexicalBlockFile(scope: !54, file: !1, discriminator: 4)
217!64 = !DILocation(line: 10, column: 5, scope: !54)
218!65 = !DILocation(line: 10, column: 50, scope: !66)
219!66 = !DILexicalBlockFile(scope: !48, file: !1, discriminator: 3)
220!67 = !DILocation(line: 9, column: 30, scope: !44)
221!68 = !DILocation(line: 9, column: 3, scope: !44)
222!69 = !DILocation(line: 11, column: 10, scope: !9)
223!70 = !DILocation(line: 11, column: 14, scope: !9)
224!71 = !DILocation(line: 11, column: 3, scope: !9)
225