1;; Test to make sure that missing tail call frames in memprof profiles are 2;; identified and cloned as needed for regular LTO. 3 4;; -stats requires asserts 5; REQUIRES: asserts 6 7; RUN: opt -thinlto-bc %s >%t.o 8; RUN: llvm-lto2 run %t.o -enable-memprof-context-disambiguation \ 9; RUN: -supports-hot-cold-new \ 10; RUN: -r=%t.o,_Z3barv,plx \ 11; RUN: -r=%t.o,_Z3bazv,plx \ 12; RUN: -r=%t.o,_Z3foov,plx \ 13; RUN: -r=%t.o,main,plx \ 14; RUN: -r=%t.o,_Znam, \ 15; RUN: -stats -save-temps \ 16; RUN: -o %t.out 2>&1 | FileCheck %s --check-prefix=STATS 17 18; RUN: llvm-dis %t.out.1.4.opt.bc -o - | FileCheck %s --check-prefix=IR 19 20;; Try again but with distributed ThinLTO 21; RUN: llvm-lto2 run %t.o -enable-memprof-context-disambiguation \ 22; RUN: -supports-hot-cold-new \ 23; RUN: -thinlto-distributed-indexes \ 24; RUN: -r=%t.o,_Z3barv,plx \ 25; RUN: -r=%t.o,_Z3bazv,plx \ 26; RUN: -r=%t.o,_Z3foov,plx \ 27; RUN: -r=%t.o,main,plx \ 28; RUN: -r=%t.o,_Znam, \ 29; RUN: -stats \ 30; RUN: -o %t2.out 2>&1 | FileCheck %s --check-prefix=STATS 31 32;; Run ThinLTO backend 33; RUN: opt -passes=memprof-context-disambiguation \ 34; RUN: -memprof-import-summary=%t.o.thinlto.bc \ 35; RUN: -stats %t.o -S 2>&1 | FileCheck %s --check-prefix=IR 36 37; STATS: 2 memprof-context-disambiguation - Number of profiled callees found via tail calls 38; STATS: 4 memprof-context-disambiguation - Aggregate depth of profiled callees found via tail calls 39; STATS: 2 memprof-context-disambiguation - Maximum depth of profiled callees found via tail calls 40 41source_filename = "memprof-tailcall.cc" 42target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128" 43target triple = "x86_64-unknown-linux-gnu" 44 45; Function Attrs: noinline 46; IR-LABEL: @_Z3barv() 47define ptr @_Z3barv() local_unnamed_addr #0 { 48entry: 49 ; IR: call {{.*}} @_Znam(i64 10) #[[NOTCOLD:[0-9]+]] 50 %call = tail call ptr @_Znam(i64 10) #2, !memprof !0, !callsite !5 51 ret ptr %call 52} 53 54; Function Attrs: nobuiltin allocsize(0) 55declare ptr @_Znam(i64) #1 56 57; Function Attrs: noinline 58; IR-LABEL: @_Z3bazv() 59define ptr @_Z3bazv() #0 { 60entry: 61 ; IR: call ptr @_Z3barv() 62 %call = tail call ptr @_Z3barv() 63 ret ptr %call 64} 65 66; Function Attrs: noinline 67; IR-LABEL: @_Z3foov() 68define ptr @_Z3foov() #0 { 69entry: 70 ; IR: call ptr @_Z3bazv() 71 %call = tail call ptr @_Z3bazv() 72 ret ptr %call 73} 74 75; Function Attrs: noinline 76; IR-LABEL: @main() 77define i32 @main() #0 { 78 ;; The first call to foo is part of a cold context, and should use the 79 ;; original functions. 80 ; IR: call ptr @_Z3foov() 81 %call = tail call ptr @_Z3foov(), !callsite !6 82 ;; The second call to foo is part of a cold context, and should call the 83 ;; cloned functions. 84 ; IR: call ptr @_Z3foov.memprof.1() 85 %call1 = tail call ptr @_Z3foov(), !callsite !7 86 ret i32 0 87} 88 89; IR-LABEL: @_Z3barv.memprof.1() 90; IR: call {{.*}} @_Znam(i64 10) #[[COLD:[0-9]+]] 91; IR-LABEL: @_Z3bazv.memprof.1() 92; IR: call ptr @_Z3barv.memprof.1() 93; IR-LABEL: @_Z3foov.memprof.1() 94; IR: call ptr @_Z3bazv.memprof.1() 95 96; IR: attributes #[[NOTCOLD]] = { builtin allocsize(0) "memprof"="notcold" } 97; IR: attributes #[[COLD]] = { builtin allocsize(0) "memprof"="cold" } 98 99attributes #0 = { noinline } 100attributes #1 = { nobuiltin allocsize(0) } 101attributes #2 = { builtin allocsize(0) } 102 103!0 = !{!1, !3} 104!1 = !{!2, !"notcold"} 105!2 = !{i64 3186456655321080972, i64 8632435727821051414} 106!3 = !{!4, !"cold"} 107!4 = !{i64 3186456655321080972, i64 -3421689549917153178} 108!5 = !{i64 3186456655321080972} 109!6 = !{i64 8632435727821051414} 110!7 = !{i64 -3421689549917153178} 111