1;; Test callsite context graph generation for simple call graph with 2;; two memprof contexts and no inlining. 3;; 4;; Original code looks like: 5;; 6;; char *bar() { 7;; return new char[10]; 8;; } 9;; 10;; char *baz() { 11;; return bar(); 12;; } 13;; 14;; char *foo() { 15;; return baz(); 16;; } 17;; 18;; int main(int argc, char **argv) { 19;; char *x = foo(); 20;; char *y = foo(); 21;; memset(x, 0, 10); 22;; memset(y, 0, 10); 23;; delete[] x; 24;; sleep(10); 25;; delete[] y; 26;; return 0; 27;; } 28;; 29;; Code compiled with -mllvm -memprof-ave-lifetime-cold-threshold=5 so that the 30;; memory freed after sleep(10) results in cold lifetimes. 31;; 32;; The IR was then reduced using llvm-reduce with the expected FileCheck input. 33 34; RUN: opt -thinlto-bc %s >%t.o 35; RUN: llvm-lto2 run %t.o -enable-memprof-context-disambiguation \ 36; RUN: -r=%t.o,main,plx \ 37; RUN: -r=%t.o,_ZdaPv, \ 38; RUN: -r=%t.o,sleep, \ 39; RUN: -r=%t.o,_Znam, \ 40; RUN: -memprof-verify-ccg -memprof-verify-nodes -memprof-dump-ccg \ 41; RUN: -memprof-export-to-dot -memprof-dot-file-path-prefix=%t. \ 42; RUN: -stats -pass-remarks=memprof-context-disambiguation -save-temps \ 43; RUN: -o %t.out 2>&1 | FileCheck %s --check-prefix=DUMP \ 44; RUN: --check-prefix=STATS 45 46; RUN: cat %t.ccg.postbuild.dot | FileCheck %s --check-prefix=DOT 47;; We should have cloned bar, baz, and foo, for the cold memory allocation. 48; RUN: cat %t.ccg.cloned.dot | FileCheck %s --check-prefix=DOTCLONED 49 50 51;; Try again but with distributed ThinLTO 52; RUN: llvm-lto2 run %t.o -enable-memprof-context-disambiguation \ 53; RUN: -thinlto-distributed-indexes \ 54; RUN: -r=%t.o,main,plx \ 55; RUN: -r=%t.o,_ZdaPv, \ 56; RUN: -r=%t.o,sleep, \ 57; RUN: -r=%t.o,_Znam, \ 58; RUN: -memprof-verify-ccg -memprof-verify-nodes -memprof-dump-ccg \ 59; RUN: -memprof-export-to-dot -memprof-dot-file-path-prefix=%t2. \ 60; RUN: -stats -pass-remarks=memprof-context-disambiguation \ 61; RUN: -o %t2.out 2>&1 | FileCheck %s --check-prefix=DUMP \ 62; RUN: --check-prefix=STATS 63 64; RUN: cat %t2.ccg.postbuild.dot | FileCheck %s --check-prefix=DOT 65;; We should have cloned bar, baz, and foo, for the cold memory allocation. 66; RUN: cat %t2.ccg.cloned.dot | FileCheck %s --check-prefix=DOTCLONED 67 68;; Check distributed index 69; RUN: llvm-dis %t.o.thinlto.bc -o - | FileCheck %s --check-prefix=DISTRIB 70 71source_filename = "memprof-basic.ll" 72target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" 73target triple = "x86_64-unknown-linux-gnu" 74 75define i32 @main() { 76entry: 77 %call = call ptr @_Z3foov(), !callsite !0 78 %call1 = call ptr @_Z3foov(), !callsite !1 79 ret i32 0 80} 81 82declare void @_ZdaPv() 83 84declare i32 @sleep() 85 86define internal ptr @_Z3barv() { 87entry: 88 %call = call ptr @_Znam(i64 0), !memprof !2, !callsite !7 89 ret ptr null 90} 91 92declare ptr @_Znam(i64) 93 94define internal ptr @_Z3bazv() { 95entry: 96 %call = call ptr @_Z3barv(), !callsite !8 97 ret ptr null 98} 99 100define internal ptr @_Z3foov() { 101entry: 102 %call = call ptr @_Z3bazv(), !callsite !9 103 ret ptr null 104} 105 106; uselistorder directives 107uselistorder ptr @_Z3foov, { 1, 0 } 108 109!0 = !{i64 8632435727821051414} 110!1 = !{i64 -3421689549917153178} 111!2 = !{!3, !5} 112!3 = !{!4, !"notcold"} 113!4 = !{i64 9086428284934609951, i64 -5964873800580613432, i64 2732490490862098848, i64 8632435727821051414} 114!5 = !{!6, !"cold"} 115!6 = !{i64 9086428284934609951, i64 -5964873800580613432, i64 2732490490862098848, i64 -3421689549917153178} 116!7 = !{i64 9086428284934609951} 117!8 = !{i64 -5964873800580613432} 118!9 = !{i64 2732490490862098848} 119 120 121; DUMP: CCG before cloning: 122; DUMP: Callsite Context Graph: 123; DUMP: Node [[BAR:0x[a-z0-9]+]] 124; DUMP: Versions: 1 MIB: 125; DUMP: AllocType 1 StackIds: 2, 3, 0 126; DUMP: AllocType 2 StackIds: 2, 3, 1 127; DUMP: (clone 0) 128; DUMP: AllocTypes: NotColdCold 129; DUMP: ContextIds: 1 2 130; DUMP: CalleeEdges: 131; DUMP: CallerEdges: 132; DUMP: Edge from Callee [[BAR]] to Caller: [[BAZ:0x[a-z0-9]+]] AllocTypes: NotColdCold ContextIds: 1 2 133 134; DUMP: Node [[BAZ]] 135; DUMP: Callee: 9832687305761716512 (_Z3barv) Clones: 0 StackIds: 2 (clone 0) 136; DUMP: AllocTypes: NotColdCold 137; DUMP: ContextIds: 1 2 138; DUMP: CalleeEdges: 139; DUMP: Edge from Callee [[BAR]] to Caller: [[BAZ]] AllocTypes: NotColdCold ContextIds: 1 2 140; DUMP: CallerEdges: 141; DUMP: Edge from Callee [[BAZ]] to Caller: [[FOO:0x[a-z0-9]+]] AllocTypes: NotColdCold ContextIds: 1 2 142 143; DUMP: Node [[FOO]] 144; DUMP: Callee: 5878270615442837395 (_Z3bazv) Clones: 0 StackIds: 3 (clone 0) 145; DUMP: AllocTypes: NotColdCold 146; DUMP: ContextIds: 1 2 147; DUMP: CalleeEdges: 148; DUMP: Edge from Callee [[BAZ]] to Caller: [[FOO]] AllocTypes: NotColdCold ContextIds: 1 2 149; DUMP: CallerEdges: 150; DUMP: Edge from Callee [[FOO]] to Caller: [[MAIN1:0x[a-z0-9]+]] AllocTypes: NotCold ContextIds: 1 151; DUMP: Edge from Callee [[FOO]] to Caller: [[MAIN2:0x[a-z0-9]+]] AllocTypes: Cold ContextIds: 2 152 153; DUMP: Node [[MAIN1]] 154; DUMP: Callee: 6731117468105397038 (_Z3foov) Clones: 0 StackIds: 0 (clone 0) 155; DUMP: AllocTypes: NotCold 156; DUMP: ContextIds: 1 157; DUMP: CalleeEdges: 158; DUMP: Edge from Callee [[FOO]] to Caller: [[MAIN1]] AllocTypes: NotCold ContextIds: 1 159; DUMP: CallerEdges: 160 161; DUMP: Node [[MAIN2]] 162; DUMP: Callee: 6731117468105397038 (_Z3foov) Clones: 0 StackIds: 1 (clone 0) 163; DUMP: AllocTypes: Cold 164; DUMP: ContextIds: 2 165; DUMP: CalleeEdges: 166; DUMP: Edge from Callee [[FOO]] to Caller: [[MAIN2]] AllocTypes: Cold ContextIds: 2 167; DUMP: CallerEdges: 168 169; DUMP: CCG after cloning: 170; DUMP: Callsite Context Graph: 171; DUMP: Node [[BAR]] 172; DUMP: Versions: 1 MIB: 173; DUMP: AllocType 1 StackIds: 2, 3, 0 174; DUMP: AllocType 2 StackIds: 2, 3, 1 175; DUMP: (clone 0) 176; DUMP: AllocTypes: NotCold 177; DUMP: ContextIds: 1 178; DUMP: CalleeEdges: 179; DUMP: CallerEdges: 180; DUMP: Edge from Callee [[BAR]] to Caller: [[BAZ]] AllocTypes: NotCold ContextIds: 1 181; DUMP: Clones: [[BAR2:0x[a-z0-9]+]] 182 183; DUMP: Node [[BAZ]] 184; DUMP: Callee: 9832687305761716512 (_Z3barv) Clones: 0 StackIds: 2 (clone 0) 185; DUMP: AllocTypes: NotCold 186; DUMP: ContextIds: 1 187; DUMP: CalleeEdges: 188; DUMP: Edge from Callee [[BAR]] to Caller: [[BAZ]] AllocTypes: NotCold ContextIds: 1 189; DUMP: CallerEdges: 190; DUMP: Edge from Callee [[BAZ]] to Caller: [[FOO]] AllocTypes: NotCold ContextIds: 1 191; DUMP: Clones: [[BAZ2:0x[a-z0-9]+]] 192 193; DUMP: Node [[FOO]] 194; DUMP: Callee: 5878270615442837395 (_Z3bazv) Clones: 0 StackIds: 3 (clone 0) 195; DUMP: AllocTypes: NotCold 196; DUMP: ContextIds: 1 197; DUMP: CalleeEdges: 198; DUMP: Edge from Callee [[BAZ]] to Caller: [[FOO]] AllocTypes: NotCold ContextIds: 1 199; DUMP: CallerEdges: 200; DUMP: Edge from Callee [[FOO]] to Caller: [[MAIN1]] AllocTypes: NotCold ContextIds: 1 201; DUMP: Clones: [[FOO2:0x[a-z0-9]+]] 202 203; DUMP: Node [[MAIN1]] 204; DUMP: Callee: 6731117468105397038 (_Z3foov) Clones: 0 StackIds: 0 (clone 0) 205; DUMP: AllocTypes: NotCold 206; DUMP: ContextIds: 1 207; DUMP: CalleeEdges: 208; DUMP: Edge from Callee [[FOO]] to Caller: [[MAIN1]] AllocTypes: NotCold ContextIds: 1 209; DUMP: CallerEdges: 210 211; DUMP: Node [[MAIN2]] 212; DUMP: Callee: 6731117468105397038 (_Z3foov) Clones: 0 StackIds: 1 (clone 0) 213; DUMP: AllocTypes: Cold 214; DUMP: ContextIds: 2 215; DUMP: CalleeEdges: 216; DUMP: Edge from Callee [[FOO2]] to Caller: [[MAIN2]] AllocTypes: Cold ContextIds: 2 217; DUMP: CallerEdges: 218 219; DUMP: Node [[FOO2]] 220; DUMP: Callee: 5878270615442837395 (_Z3bazv) Clones: 0 StackIds: 3 (clone 0) 221; DUMP: AllocTypes: Cold 222; DUMP: ContextIds: 2 223; DUMP: CalleeEdges: 224; DUMP: Edge from Callee [[BAZ2]] to Caller: [[FOO2]] AllocTypes: Cold ContextIds: 2 225; DUMP: CallerEdges: 226; DUMP: Edge from Callee [[FOO2]] to Caller: [[MAIN2]] AllocTypes: Cold ContextIds: 2 227; DUMP: Clone of [[FOO]] 228 229; DUMP: Node [[BAZ2]] 230; DUMP: Callee: 9832687305761716512 (_Z3barv) Clones: 0 StackIds: 2 (clone 0) 231; DUMP: AllocTypes: Cold 232; DUMP: ContextIds: 2 233; DUMP: CalleeEdges: 234; DUMP: Edge from Callee [[BAR2]] to Caller: [[BAZ2]] AllocTypes: Cold ContextIds: 2 235; DUMP: CallerEdges: 236; DUMP: Edge from Callee [[BAZ2]] to Caller: [[FOO2]] AllocTypes: Cold ContextIds: 2 237; DUMP: Clone of [[BAZ]] 238 239; DUMP: Node [[BAR2]] 240; DUMP: Versions: 1 MIB: 241; DUMP: AllocType 1 StackIds: 2, 3, 0 242; DUMP: AllocType 2 StackIds: 2, 3, 1 243; DUMP: (clone 0) 244; DUMP: AllocTypes: Cold 245; DUMP: ContextIds: 2 246; DUMP: CalleeEdges: 247; DUMP: CallerEdges: 248; DUMP: Edge from Callee [[BAR2]] to Caller: [[BAZ2]] AllocTypes: Cold ContextIds: 2 249; DUMP: Clone of [[BAR]] 250 251 252; STATS: 1 memprof-context-disambiguation - Number of cold static allocations (possibly cloned) 253; STATS: 1 memprof-context-disambiguation - Number of not cold static allocations (possibly cloned) 254; STATS: 3 memprof-context-disambiguation - Number of function clones created during whole program analysis 255 256 257; DOT: digraph "postbuild" { 258; DOT: label="postbuild"; 259; DOT: Node[[BAR:0x[a-z0-9]+]] [shape=record,tooltip="N[[BAR]] ContextIds: 1 2",fillcolor="mediumorchid1",style="filled",style="filled",label="{OrigId: Alloc0\n_Z3barv -\> alloc}"]; 260; DOT: Node[[BAZ:0x[a-z0-9]+]] [shape=record,tooltip="N[[BAZ]] ContextIds: 1 2",fillcolor="mediumorchid1",style="filled",style="filled",label="{OrigId: 12481870273128938184\n_Z3bazv -\> _Z3barv}"]; 261; DOT: Node[[BAZ]] -> Node[[BAR]][tooltip="ContextIds: 1 2",fillcolor="mediumorchid1"]; 262; DOT: Node[[FOO:0x[a-z0-9]+]] [shape=record,tooltip="N[[FOO]] ContextIds: 1 2",fillcolor="mediumorchid1",style="filled",style="filled",label="{OrigId: 2732490490862098848\n_Z3foov -\> _Z3bazv}"]; 263; DOT: Node[[FOO]] -> Node[[BAZ]][tooltip="ContextIds: 1 2",fillcolor="mediumorchid1"]; 264; DOT: Node[[MAIN1:0x[a-z0-9]+]] [shape=record,tooltip="N[[MAIN1]] ContextIds: 1",fillcolor="brown1",style="filled",style="filled",label="{OrigId: 8632435727821051414\nmain -\> _Z3foov}"]; 265; DOT: Node[[MAIN1]] -> Node[[FOO]][tooltip="ContextIds: 1",fillcolor="brown1"]; 266; DOT: Node[[MAIN2:0x[a-z0-9]+]] [shape=record,tooltip="N[[MAIN2]] ContextIds: 2",fillcolor="cyan",style="filled",style="filled",label="{OrigId: 15025054523792398438\nmain -\> _Z3foov}"]; 267; DOT: Node[[MAIN2]] -> Node[[FOO]][tooltip="ContextIds: 2",fillcolor="cyan"]; 268; DOT: } 269 270 271; DOTCLONED: digraph "cloned" { 272; DOTCLONED: label="cloned"; 273; DOTCLONED: Node[[BAR:0x[a-z0-9]+]] [shape=record,tooltip="N[[BAR]] ContextIds: 1",fillcolor="brown1",style="filled",style="filled",label="{OrigId: Alloc0\n_Z3barv -\> alloc}"]; 274; DOTCLONED: Node[[BAZ:0x[a-z0-9]+]] [shape=record,tooltip="N[[BAZ]] ContextIds: 1",fillcolor="brown1",style="filled",style="filled",label="{OrigId: 12481870273128938184\n_Z3bazv -\> _Z3barv}"]; 275; DOTCLONED: Node[[BAZ]] -> Node[[BAR]][tooltip="ContextIds: 1",fillcolor="brown1"]; 276; DOTCLONED: Node[[FOO:0x[a-z0-9]+]] [shape=record,tooltip="N[[FOO]] ContextIds: 1",fillcolor="brown1",style="filled",style="filled",label="{OrigId: 2732490490862098848\n_Z3foov -\> _Z3bazv}"]; 277; DOTCLONED: Node[[FOO]] -> Node[[BAZ]][tooltip="ContextIds: 1",fillcolor="brown1"]; 278; DOTCLONED: Node[[MAIN1:0x[a-z0-9]+]] [shape=record,tooltip="N[[MAIN1]] ContextIds: 1",fillcolor="brown1",style="filled",style="filled",label="{OrigId: 8632435727821051414\nmain -\> _Z3foov}"]; 279; DOTCLONED: Node[[MAIN1]] -> Node[[FOO]][tooltip="ContextIds: 1",fillcolor="brown1"]; 280; DOTCLONED: Node[[MAIN2:0x[a-z0-9]+]] [shape=record,tooltip="N[[MAIN2]] ContextIds: 2",fillcolor="cyan",style="filled",style="filled",label="{OrigId: 15025054523792398438\nmain -\> _Z3foov}"]; 281; DOTCLONED: Node[[MAIN2]] -> Node[[FOO2:0x[a-z0-9]+]][tooltip="ContextIds: 2",fillcolor="cyan"]; 282; DOTCLONED: Node[[FOO2]] [shape=record,tooltip="N[[FOO2]] ContextIds: 2",fillcolor="cyan",style="filled",color="blue",style="filled,bold,dashed",label="{OrigId: 0\n_Z3foov -\> _Z3bazv}"]; 283; DOTCLONED: Node[[FOO2]] -> Node[[BAZ2:0x[a-z0-9]+]][tooltip="ContextIds: 2",fillcolor="cyan"]; 284; DOTCLONED: Node[[BAZ2]] [shape=record,tooltip="N[[BAZ2]] ContextIds: 2",fillcolor="cyan",style="filled",color="blue",style="filled,bold,dashed",label="{OrigId: 0\n_Z3bazv -\> _Z3barv}"]; 285; DOTCLONED: Node[[BAZ2]] -> Node[[BAR2:0x[a-z0-9]+]][tooltip="ContextIds: 2",fillcolor="cyan"]; 286; DOTCLONED: Node[[BAR2]] [shape=record,tooltip="N[[BAR2]] ContextIds: 2",fillcolor="cyan",style="filled",color="blue",style="filled,bold,dashed",label="{OrigId: Alloc0\n_Z3barv -\> alloc}"]; 287; DOTCLONED: } 288 289 290; DISTRIB: ^[[BAZ:[0-9]+]] = gv: (guid: 5878270615442837395, {{.*}} callsites: ((callee: ^[[BAR:[0-9]+]], clones: (0, 1) 291; DISTRIB: ^[[FOO:[0-9]+]] = gv: (guid: 6731117468105397038, {{.*}} callsites: ((callee: ^[[BAZ]], clones: (0, 1) 292; DISTRIB: ^[[BAR]] = gv: (guid: 9832687305761716512, {{.*}} allocs: ((versions: (notcold, cold) 293; DISTRIB: ^[[MAIN:[0-9]+]] = gv: (guid: 15822663052811949562, {{.*}} callsites: ((callee: ^[[FOO]], clones: (0), {{.*}} (callee: ^[[FOO]], clones: (1) 294