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