|
150
|
1 ; RUN: opt < %s -codegenprepare -S | FileCheck %s
|
|
173
|
2 ; RUN: llc < %s | FileCheck --check-prefix=ASM1 %s
|
|
|
3 ; RUN: llc < %s -function-sections | FileCheck --check-prefix=ASM2 %s
|
|
150
|
4
|
|
|
5 target triple = "x86_64-pc-linux-gnu"
|
|
|
6
|
|
|
7 ; This tests that hot/cold functions get correct section prefix assigned
|
|
|
8
|
|
|
9 ; CHECK: hot_func1{{.*}}!section_prefix ![[HOT_ID:[0-9]+]]
|
|
173
|
10 ; ASM1: .section .text.hot.,"ax",@progbits
|
|
|
11 ; ASM2: .section .text.hot.hot_func1,"ax",@progbits
|
|
150
|
12 ; The entry is hot
|
|
|
13 define void @hot_func1() !prof !15 {
|
|
|
14 ret void
|
|
|
15 }
|
|
|
16
|
|
|
17 ; CHECK: hot_func2{{.*}}!section_prefix ![[HOT_ID:[0-9]+]]
|
|
|
18 ; Entry is cold but inner block is hot
|
|
|
19 define void @hot_func2(i32 %n) !prof !16 {
|
|
|
20 entry:
|
|
|
21 %n.addr = alloca i32, align 4
|
|
|
22 %i = alloca i32, align 4
|
|
|
23 store i32 %n, i32* %n.addr, align 4
|
|
|
24 store i32 0, i32* %i, align 4
|
|
|
25 br label %for.cond
|
|
|
26
|
|
|
27 for.cond:
|
|
|
28 %0 = load i32, i32* %i, align 4
|
|
|
29 %1 = load i32, i32* %n.addr, align 4
|
|
|
30 %cmp = icmp slt i32 %0, %1
|
|
|
31 br i1 %cmp, label %for.body, label %for.end, !prof !19
|
|
|
32
|
|
|
33 for.body:
|
|
|
34 %2 = load i32, i32* %i, align 4
|
|
|
35 %inc = add nsw i32 %2, 1
|
|
|
36 store i32 %inc, i32* %i, align 4
|
|
|
37 br label %for.cond
|
|
|
38
|
|
|
39 for.end:
|
|
|
40 ret void
|
|
|
41 }
|
|
|
42
|
|
|
43 ; For instrumentation based PGO, we should only look at block counts,
|
|
|
44 ; not call site VP metadata (which can exist on value profiled memcpy,
|
|
|
45 ; or possibly left behind after static analysis based devirtualization).
|
|
|
46 ; CHECK: cold_func1{{.*}}!section_prefix ![[COLD_ID:[0-9]+]]
|
|
173
|
47 ; ASM1: .section .text.unlikely.,"ax",@progbits
|
|
|
48 ; ASM2: .section .text.unlikely.cold_func1,"ax",@progbits
|
|
150
|
49 define void @cold_func1() !prof !16 {
|
|
|
50 call void @hot_func1(), !prof !17
|
|
|
51 call void @hot_func1(), !prof !17
|
|
|
52 ret void
|
|
|
53 }
|
|
|
54
|
|
|
55 ; CHECK: cold_func2{{.*}}!section_prefix ![[COLD_ID]]
|
|
|
56 define void @cold_func2() !prof !16 {
|
|
|
57 call void @hot_func1(), !prof !17
|
|
|
58 call void @hot_func1(), !prof !18
|
|
|
59 call void @hot_func1(), !prof !18
|
|
|
60 ret void
|
|
|
61 }
|
|
|
62
|
|
|
63 ; CHECK: cold_func3{{.*}}!section_prefix ![[COLD_ID]]
|
|
|
64 define void @cold_func3() !prof !16 {
|
|
|
65 call void @hot_func1(), !prof !18
|
|
|
66 ret void
|
|
|
67 }
|
|
|
68
|
|
|
69 ; CHECK: ![[HOT_ID]] = !{!"function_section_prefix", !".hot"}
|
|
|
70 ; CHECK: ![[COLD_ID]] = !{!"function_section_prefix", !".unlikely"}
|
|
|
71 !llvm.module.flags = !{!1}
|
|
|
72 !1 = !{i32 1, !"ProfileSummary", !2}
|
|
|
73 !2 = !{!3, !4, !5, !6, !7, !8, !9, !10}
|
|
|
74 !3 = !{!"ProfileFormat", !"InstrProf"}
|
|
|
75 !4 = !{!"TotalCount", i64 10000}
|
|
|
76 !5 = !{!"MaxCount", i64 1000}
|
|
|
77 !6 = !{!"MaxInternalCount", i64 1}
|
|
|
78 !7 = !{!"MaxFunctionCount", i64 1000}
|
|
|
79 !8 = !{!"NumCounts", i64 3}
|
|
|
80 !9 = !{!"NumFunctions", i64 3}
|
|
|
81 !10 = !{!"DetailedSummary", !11}
|
|
|
82 !11 = !{!12, !13, !14}
|
|
|
83 !12 = !{i32 10000, i64 100, i32 1}
|
|
|
84 !13 = !{i32 999000, i64 100, i32 1}
|
|
|
85 !14 = !{i32 999999, i64 1, i32 2}
|
|
|
86 !15 = !{!"function_entry_count", i64 1000}
|
|
|
87 !16 = !{!"function_entry_count", i64 1}
|
|
|
88 !17 = !{!"branch_weights", i32 80}
|
|
|
89 !18 = !{!"branch_weights", i32 1}
|
|
|
90 !19 = !{!"branch_weights", i32 1000, i32 1}
|
