compiler-rt/test/profile/Linux/instrprof-vtable-value-prof.cpp

   1 // REQUIRES: lld, lld-available
   2
   3 // Building the instrumented binary will fail because lld doesn't support
   4 // big-endian ELF for PPC (aka ABI 1).
   5 // ld.lld: error: /lib/../lib64/Scrt1.o: ABI version 1 is not supported
   6 // UNSUPPORTED: ppc && host-byteorder-big-endian
   7
   8 // RUN: rm -rf %t && mkdir %t && split-file %s %t && cd %t
   9
  10 // RUN: %clangxx_pgogen -fuse-ld=lld -O2 -fprofile-generate=. -mllvm -enable-vtable-value-profiling lib.cpp main.cpp -o test
  11 // RUN: env LLVM_PROFILE_FILE=test.profraw ./test
  12
  13 // Show vtable profiles from raw profile.
  14 // RUN: llvm-profdata show --function=main --ic-targets --show-vtables test.profraw | FileCheck %s --check-prefixes=COMMON,RAW
  15
  16 // Generate indexed profile from raw profile and show the data.
  17 // RUN: llvm-profdata merge --keep-vtable-symbols test.profraw -o test.profdata
  18 // RUN: llvm-profdata show --function=main --ic-targets --show-vtables test.profdata | FileCheck %s --check-prefixes=COMMON,INDEXED
  19
  20 // Generate text profile from raw and indexed profiles respectively and show the data.
  21 // RUN: llvm-profdata merge --keep-vtable-symbols --text test.profraw -o raw.proftext
  22 // RUN: llvm-profdata show --function=main --ic-targets --show-vtables --text raw.proftext | FileCheck %s --check-prefix=ICTEXT
  23 // RUN: llvm-profdata merge --keep-vtable-symbols --text test.profdata -o indexed.proftext
  24 // RUN: llvm-profdata show --function=main --ic-targets --show-vtables --text indexed.proftext | FileCheck %s --check-prefix=ICTEXT
  25
  26 // Generate indexed profile from text profiles and show the data
  27 // RUN: llvm-profdata merge --keep-vtable-symbols --binary raw.proftext -o text.profraw
  28 // RUN: llvm-profdata show --function=main --ic-targets --show-vtables text.profraw | FileCheck %s --check-prefixes=COMMON,INDEXED
  29 // RUN: llvm-profdata merge --keep-vtable-symbols --binary indexed.proftext -o text.profdata
  30 // RUN: llvm-profdata show --function=main --ic-targets --show-vtables text.profdata | FileCheck %s --check-prefixes=COMMON,INDEXED
  31
  32 // COMMON: Counters:
  33 // COMMON-NEXT:  main:
  34 // COMMON-NEXT:  Hash: 0x068617320ec408a0
  35 // COMMON-NEXT:  Counters: 4
  36 // COMMON-NEXT:  Indirect Call Site Count: 2
  37 // COMMON-NEXT:  Number of instrumented vtables: 2
  38 // RAW:  Indirect Target Results:
  39 // RAW-NEXT:       [  0, _ZN8Derived14funcEii,        50 ] (25.00%)
  40 // RAW-NEXT:       [  0, {{.*}}lib.cpp;_ZN12_GLOBAL__N_18Derived24funcEii,        150 ] (75.00%)
  41 // RAW-NEXT:       [  1, _ZN8Derived1D0Ev,        250 ] (25.00%)
  42 // RAW-NEXT:       [  1, {{.*}}lib.cpp;_ZN12_GLOBAL__N_18Derived2D0Ev,        750 ] (75.00%)
  43 // RAW-NEXT:  VTable Results:
  44 // RAW-NEXT:       [  0, _ZTV8Derived1,        50 ] (25.00%)
  45 // RAW-NEXT:       [  0, {{.*}}lib.cpp;_ZTVN12_GLOBAL__N_18Derived2E,        150 ] (75.00%)
  46 // RAW-NEXT:       [  1, _ZTV8Derived1,        250 ] (25.00%)
  47 // RAW-NEXT:       [  1, {{.*}}lib.cpp;_ZTVN12_GLOBAL__N_18Derived2E,        750 ] (75.00%)
  48 // INDEXED:     Indirect Target Results:
  49 // INDEXED-NEXT:         [  0, {{.*}}lib.cpp;_ZN12_GLOBAL__N_18Derived24funcEii,        150 ] (75.00%)
  50 // INDEXED-NEXT:         [  0, _ZN8Derived14funcEii,        50 ] (25.00%)
  51 // INDEXED-NEXT:         [  1, {{.*}}lib.cpp;_ZN12_GLOBAL__N_18Derived2D0Ev,        750 ] (75.00%)
  52 // INDEXED-NEXT:         [  1, _ZN8Derived1D0Ev,        250 ] (25.00%)
  53 // INDEXED-NEXT:     VTable Results:
  54 // INDEXED-NEXT:         [  0, {{.*}}lib.cpp;_ZTVN12_GLOBAL__N_18Derived2E,        150 ] (75.00%)
  55 // INDEXED-NEXT:         [  0, _ZTV8Derived1,        50 ] (25.00%)
  56 // INDEXED-NEXT:         [  1, {{.*}}lib.cpp;_ZTVN12_GLOBAL__N_18Derived2E,        750 ] (75.00%)
  57 // INDEXED-NEXT:         [  1, _ZTV8Derived1,        250 ] (25.00%)
  58 // COMMON: Instrumentation level: IR  entry_first = 0
  59 // COMMON-NEXT: Functions shown: 1
  60 // COMMON-NEXT: Total functions: 7
  61 // COMMON-NEXT: Maximum function count: 1000
  62 // COMMON-NEXT: Maximum internal block count: 1000
  63 // COMMON-NEXT: Statistics for indirect call sites profile:
  64 // COMMON-NEXT:   Total number of sites: 2
  65 // COMMON-NEXT:   Total number of sites with values: 2
  66 // COMMON-NEXT:   Total number of profiled values: 4
  67 // COMMON-NEXT:   Value sites histogram:
  68 // COMMON-NEXT:         NumTargets, SiteCount
  69 // COMMON-NEXT:         2, 2
  70 // COMMON-NEXT: Statistics for vtable profile:
  71 // COMMON-NEXT:   Total number of sites: 2
  72 // COMMON-NEXT:   Total number of sites with values: 2
  73 // COMMON-NEXT:   Total number of profiled values: 4
  74 // COMMON-NEXT:   Value sites histogram:
  75 // COMMON-NEXT:         NumTargets, SiteCount
  76 // COMMON-NEXT:         2, 2
  77
  78 // ICTEXT: :ir
  79 // ICTEXT: main
  80 // ICTEXT: # Func Hash:
  81 // ICTEXT: 470088714870327456
  82 // ICTEXT: # Num Counters:
  83 // ICTEXT: 4
  84 // ICTEXT: # Counter Values:
  85 // ICTEXT: 1000
  86 // ICTEXT: 1000
  87 // ICTEXT: 200
  88 // ICTEXT: 1
  89 // ICTEXT: # Num Value Kinds:
  90 // ICTEXT: 2
  91 // ICTEXT: # ValueKind = IPVK_IndirectCallTarget:
  92 // ICTEXT: 0
  93 // ICTEXT: # NumValueSites:
  94 // ICTEXT: 2
  95 // ICTEXT: 2
  96 // ICTEXT: {{.*}}lib.cpp;_ZN12_GLOBAL__N_18Derived24funcEii:150
  97 // ICTEXT: _ZN8Derived14funcEii:50
  98 // ICTEXT: 2
  99 // ICTEXT: {{.*}}lib.cpp;_ZN12_GLOBAL__N_18Derived2D0Ev:750
 100 // ICTEXT: _ZN8Derived1D0Ev:250
 101 // ICTEXT: # ValueKind = IPVK_VTableTarget:
 102 // ICTEXT: 2
 103 // ICTEXT: # NumValueSites:
 104 // ICTEXT: 2
 105 // ICTEXT: 2
 106 // ICTEXT: {{.*}}lib.cpp;_ZTVN12_GLOBAL__N_18Derived2E:150
 107 // ICTEXT: _ZTV8Derived1:50
 108 // ICTEXT: 2
 109 // ICTEXT: {{.*}}lib.cpp;_ZTVN12_GLOBAL__N_18Derived2E:750
 110 // ICTEXT: _ZTV8Derived1:250
 111
 112 // When vtable value profiles exist, pgo-instr-use pass should annotate them
 113 // even if `-enable-vtable-value-profiling` is not explicitly on.
 114 // RUN: %clangxx -m64 -fprofile-use=test.profdata -fuse-ld=lld -O2 \
 115 // RUN:   -mllvm -print-after=pgo-instr-use -mllvm -filter-print-funcs=main \
 116 // RUN:   -mllvm -print-module-scope lib.cpp main.cpp 2>&1 | FileCheck %s --check-prefix=ANNOTATE
 117
 118 // ANNOTATE-NOT: Inconsistent number of value sites
 119 // ANNOTATE: !{!"VP", i32 2
 120
 121 // When vtable value profiles exist, pgo-instr-use pass will not annotate them
 122 // if `-icp-max-num-vtables` is set to zero.
 123 // RUN: %clangxx -m64 -fprofile-use=test.profdata -fuse-ld=lld -O2 \
 124 // RUN:   -mllvm -icp-max-num-vtables=0 -mllvm -print-after=pgo-instr-use \
 125 // RUN:   -mllvm -filter-print-funcs=main -mllvm -print-module-scope lib.cpp main.cpp 2>&1 | \
 126 // RUN:   FileCheck %s --check-prefix=OMIT
 127
 128 // OMIT: Inconsistent number of value sites
 129 // OMIT-NOT: !{!"VP", i32 2
 130
 131 // Test indirect call promotion transformation using vtable profiles.
 132 // - Build with `-g` to enable debug information.
 133 // - In real world settings, ICP pass is disabled in prelink pipeline. In
 134 //   the postlink pipeline, ICP is enabled after whole-program-devirtualization
 135 //   pass. Do the same thing in this test.
 136 // - Enable `-fwhole-program-vtables` generate type metadata and intrinsics.
 137 // - Enable `-fno-split-lto-unit` and `-Wl,-lto-whole-program-visibility` to
 138 //   preserve type intrinsics for ICP pass.
 139 // RUN: %clangxx -m64  -fprofile-use=test.profdata -Wl,--lto-whole-program-visibility \
 140 // RUN:    -mllvm -disable-icp=true -Wl,-mllvm,-disable-icp=false -fuse-ld=lld \
 141 // RUN:    -g -flto=thin -fwhole-program-vtables -fno-split-lto-unit -O2 \
 142 // RUN:    -mllvm -enable-vtable-value-profiling -Wl,-mllvm,-enable-vtable-value-profiling \
 143 // RUN:    -mllvm -enable-vtable-profile-use \
 144 // RUN:    -Wl,-mllvm,-enable-vtable-profile-use -Rpass=pgo-icall-prom \
 145 // RUN:    -Wl,-mllvm,-print-after=pgo-icall-prom \
 146 // RUN:    -Wl,-mllvm,-filter-print-funcs=main lib.cpp main.cpp 2>&1 \
 147 // RUN:    | FileCheck %s --check-prefixes=REMARK,IR --implicit-check-not="!VP"
 148
 149 // For the indirect call site `ptr->func`
 150 // REMARK: main.cpp:10:19: Promote indirect call to _ZN12_GLOBAL__N_18Derived24funcEii.llvm.{{.*}} with count 150 out of 200, sink 1 instruction(s) and compare 1 vtable(s): {_ZTVN12_GLOBAL__N_18Derived2E.llvm.{{.*}}}
 151 // REMARK: main.cpp:10:19: Promote indirect call to _ZN8Derived14funcEii with count 50 out of 50, sink 1 instruction(s) and compare 1 vtable(s): {_ZTV8Derived1}
 152 //
 153 // For the indirect call site `delete ptr`
 154 // REMARK: main.cpp:12:5: Promote indirect call to _ZN12_GLOBAL__N_18Derived2D0Ev.llvm.{{.*}} with count 750 out of 1000, sink 2 instruction(s) and compare 1 vtable(s): {_ZTVN12_GLOBAL__N_18Derived2E.llvm.{{.*}}}
 155 // REMARK: main.cpp:12:5: Promote indirect call to _ZN8Derived1D0Ev with count 250 out of 250, sink 2 instruction(s) and compare 1 vtable(s): {_ZTV8Derived1}
 156
 157 // The IR matchers for indirect callsite `ptr->func`.
 158 // IR-LABEL: @main
 159 // IR:   [[OBJ:%.*]] = {{.*}}call {{.*}} @_Z10createTypei
 160 // IR:   [[VTABLE:%.*]] = load ptr, ptr [[OBJ]]
 161 // IR:   [[CMP1:%.*]] = icmp eq ptr [[VTABLE]], getelementptr inbounds (i8, ptr @_ZTVN12_GLOBAL__N_18Derived2E.llvm.{{.*}}, i32 16)
 162 // IR:   br i1 [[CMP1]], label %[[BB1:.*]], label %[[BB2:[a-zA-Z0-9_.]+]],
 163 //
 164 // IR: [[BB1]]:
 165 // IR:   [[RESBB1:%.*]] = {{.*}}call {{.*}} @_ZN12_GLOBAL__N_18Derived24funcEii.llvm.{{.*}}
 166 // IR:   br label %[[MERGE0:[a-zA-Z0-9_.]+]]
 167 //
 168 // IR: [[BB2]]:
 169 // IR:   [[CMP2:%.*]] = icmp eq ptr [[VTABLE]], getelementptr inbounds (i8, ptr @_ZTV8Derived1, i32 16)
 170 // IR:   br i1 [[CMP2]], label %[[BB3:.*]], label %[[BB4:[a-zA-Z0-9_.]+]],
 171 //
 172 // IR: [[BB3]]:
 173 // IR:   [[RESBB3:%.*]] = {{.*}}call {{.*}} @_ZN8Derived14funcEii
 174 // IR:   br label %[[MERGE1:[a-zA-Z0-9_.]+]],
 175 //
 176 // IR: [[BB4]]:
 177 // IR:   [[FUNCPTR:%.*]] = load ptr, ptr [[VTABLE]]
 178 // IR:   [[RESBB4:%.*]] = {{.*}}call {{.*}} [[FUNCPTR]]
 179 // IR:   br label %[[MERGE1]]
 180 //
 181 // IR: [[MERGE1]]:
 182 // IR:    [[RES1:%.*]] = phi i32 [ [[RESBB4]], %[[BB4]] ], [ [[RESBB3]], %[[BB3]] ]
 183 // IR:    br label %[[MERGE0]]
 184 //
 185 // IR: [[MERGE0]]:
 186 // IR:    [[RES2:%.*]] = phi i32 [ [[RES1]], %[[MERGE1]] ], [ [[RESBB1]], %[[BB1]] ]
 187
 188 //--- lib.h
 189 #include <stdio.h>
 190 #include <stdlib.h>
 191 class Base {
 192 public:
 193   virtual int func(int a, int b) = 0;
 194
 195   virtual ~Base() {};
 196 };
 197
 198 class Derived1 : public Base {
 199 public:
 200   int func(int a, int b) override;
 201
 202   ~Derived1() {}
 203 };
 204
 205 __attribute__((noinline)) Base *createType(int a);
 206
 207 //--- lib.cpp
 208 #include "lib.h"
 209
 210 namespace {
 211 class Derived2 : public Base {
 212 public:
 213   int func(int a, int b) override { return a * (a - b); }
 214
 215   ~Derived2() {}
 216 };
 217 } // namespace
 218
 219 int Derived1::func(int a, int b) { return a * b; }
 220
 221 Base *createType(int a) {
 222   Base *base = nullptr;
 223   if (a % 4 == 0)
 224     base = new Derived1();
 225   else
 226     base = new Derived2();
 227   return base;
 228 }
 229
 230 //--- main.cpp
 231 #include "lib.h"
 232
 233 int main(int argc, char **argv) {
 234   int sum = 0;
 235   for (int i = 0; i < 1000; i++) {
 236     int a = rand();
 237     int b = rand();
 238     Base *ptr = createType(i);
 239     if (i % 5 == 0)
 240       sum += ptr->func(b, a);
 241
 242     delete ptr;
 243   }
 244   printf("sum is %d\n", sum);
 245   return 0;
 246 }