1;; Test callsite context graph generation for call graph with two memprof 2;; contexts and multiple levels of inlining, requiring generation of new 3;; fused nodes to represent the inlined sequence while matching callsite 4;; nodes onto the graph. In particular this tests the case where a function 5;; has inlined a callee containing an inlined callee. 6;; 7;; Original code looks like: 8;; 9;; char *bar() __attribute__((noinline)) { 10;; return new char[10]; 11;; } 12;; 13;; char *baz() { 14;; return bar(); 15;; } 16;; 17;; char *foo() { 18;; return baz(); 19;; } 20;; 21;; int main(int argc, char **argv) { 22;; char *x = foo(); 23;; char *y = foo(); 24;; memset(x, 0, 10); 25;; memset(y, 0, 10); 26;; delete[] x; 27;; sleep(10); 28;; delete[] y; 29;; return 0; 30;; } 31;; 32;; Code compiled with -mllvm -memprof-ave-lifetime-cold-threshold=5 so that the 33;; memory freed after sleep(10) results in cold lifetimes. 34;; 35;; Both foo and baz are inlined into main, at both foo callsites. 36;; We should update the graph for new fused nodes for both of those inlined 37;; callsites to bar. 38;; 39;; Note that baz and bar are both dead due to the inlining, but have been left 40;; in the input IR to ensure that the MIB call chain is matched to the longer 41;; inline sequences from main. 42;; 43;; The IR was then reduced using llvm-reduce with the expected FileCheck input. 44 45; RUN: opt -passes=memprof-context-disambiguation -supports-hot-cold-new \ 46; RUN: -memprof-verify-ccg -memprof-verify-nodes -memprof-dump-ccg \ 47; RUN: %s -S 2>&1 | FileCheck %s --check-prefix=DUMP 48 49 50target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" 51target triple = "x86_64-unknown-linux-gnu" 52 53define ptr @_Z3barv() #0 { 54entry: 55 %call = call noalias noundef nonnull dereferenceable(10) ptr @_Znam(i64 noundef 10) #7, !memprof !7, !callsite !12, !heapallocsite !13 56 ret ptr null 57} 58 59; Function Attrs: nobuiltin 60declare ptr @_Znam(i64) #1 61 62; Function Attrs: mustprogress 63declare ptr @_Z3bazv() #2 64 65define i32 @main() #3 { 66delete.end5: 67 %call.i.i = call noundef ptr @_Z3barv(), !callsite !14 68 %call.i.i8 = call noundef ptr @_Z3barv(), !callsite !15 69 ret i32 0 70} 71 72; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: write) 73declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1 immarg) #4 74 75declare void @_ZdaPv() #5 76 77declare i32 @sleep() #6 78 79attributes #0 = { "stack-protector-buffer-size"="8" } 80attributes #1 = { nobuiltin } 81attributes #2 = { mustprogress } 82attributes #3 = { "tune-cpu"="generic" } 83attributes #4 = { nocallback nofree nounwind willreturn memory(argmem: write) } 84attributes #5 = { "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" } 85attributes #6 = { "disable-tail-calls"="true" } 86attributes #7 = { builtin } 87 88!llvm.module.flags = !{!0, !1, !2, !3, !4, !5, !6} 89 90!0 = !{i32 7, !"Dwarf Version", i32 5} 91!1 = !{i32 2, !"Debug Info Version", i32 3} 92!2 = !{i32 1, !"wchar_size", i32 4} 93!3 = !{i32 8, !"PIC Level", i32 2} 94!4 = !{i32 7, !"PIE Level", i32 2} 95!5 = !{i32 7, !"uwtable", i32 2} 96!6 = !{i32 7, !"frame-pointer", i32 2} 97!7 = !{!8, !10} 98!8 = !{!9, !"notcold"} 99!9 = !{i64 9086428284934609951, i64 -5964873800580613432, i64 2732490490862098848, i64 8632435727821051414} 100!10 = !{!11, !"cold"} 101!11 = !{i64 9086428284934609951, i64 -5964873800580613432, i64 2732490490862098848, i64 -3421689549917153178} 102!12 = !{i64 9086428284934609951} 103!13 = !DIBasicType(name: "char", size: 8, encoding: DW_ATE_signed_char) 104!14 = !{i64 -5964873800580613432, i64 2732490490862098848, i64 8632435727821051414} 105!15 = !{i64 -5964873800580613432, i64 2732490490862098848, i64 -3421689549917153178} 106 107 108; DUMP: CCG before cloning: 109; DUMP: Callsite Context Graph: 110; DUMP: Node [[BAR:0x[a-z0-9]+]] 111; DUMP: %call = call noalias noundef nonnull dereferenceable(10) ptr @_Znam(i64 noundef 10) #7, !heapallocsite !7 (clone 0) 112; DUMP: AllocTypes: NotColdCold 113; DUMP: ContextIds: 1 2 114; DUMP: CalleeEdges: 115; DUMP: CallerEdges: 116; DUMP: Edge from Callee [[BAR]] to Caller: [[MAIN1:0x[a-z0-9]+]] AllocTypes: NotCold ContextIds: 1 117; DUMP: Edge from Callee [[BAR]] to Caller: [[MAIN2:0x[a-z0-9]+]] AllocTypes: Cold ContextIds: 2 118 119;; This is the node synthesized for the first inlined call chain of main->foo->baz 120; DUMP: Node [[MAIN1]] 121; DUMP: %call.i.i = call noundef ptr @_Z3barv() (clone 0) 122; DUMP: AllocTypes: NotCold 123; DUMP: ContextIds: 1 124; DUMP: CalleeEdges: 125; DUMP: Edge from Callee [[BAR]] to Caller: [[MAIN1]] AllocTypes: NotCold ContextIds: 1 126; DUMP: CallerEdges: 127 128;; This is the node synthesized for the second inlined call chain of main->foo->baz 129; DUMP: Node [[MAIN2]] 130; DUMP: %call.i.i8 = call noundef ptr @_Z3barv() (clone 0) 131; DUMP: AllocTypes: Cold 132; DUMP: ContextIds: 2 133; DUMP: CalleeEdges: 134; DUMP: Edge from Callee [[BAR]] to Caller: [[MAIN2]] AllocTypes: Cold ContextIds: 2 135; DUMP: CallerEdges: 136