Skip to content

Commit 603d58b

Browse files
petrhosekpetrhosek
committed
[InstrProfiling] Use !associated metadata for counters, data and values
The !associated metadata may be attached to a global object declaration with a single argument that references another global object. This metadata prevents discarding of the global object in linker GC unless the referenced object is also discarded. Furthermore, when a function symbol is discarded by the linker, setting up !associated metadata allows linker to discard counters, data and values associated with that function symbol. This is not possible today because there's metadata to guide the linker. This approach is also used by other instrumentations like sanitizers. Note that !associated metadata is only supported by ELF, it does not have any effect on non-ELF targets. Differential Revision: https://reviews.llvm.org/D76802
1 parent ba10bed commit 603d58b

File tree

7 files changed

+140
-22
lines changed

7 files changed

+140
-22
lines changed

‎compiler-rt/lib/profile/InstrProfilingPlatformLinux.c

Lines changed: 14 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -24,27 +24,21 @@
2424
#define PROF_VNODES_STOP INSTR_PROF_SECT_STOP(INSTR_PROF_VNODES_COMMON)
2525

2626
/* Declare section start and stop symbols for various sections
27-
* generated by compiler instrumentation.
27+
* generated by compiler instrumentation. These symbols are
28+
* declared as weak, which means that they'll end up as NULL if
29+
* those sections are empty, and any code that iterates over the
30+
* content using __llvm_profile_begin_* and __llvm_profile_end_*
31+
* functions defined below will be a no-op.
2832
*/
29-
extern __llvm_profile_data PROF_DATA_START COMPILER_RT_VISIBILITY;
30-
extern __llvm_profile_data PROF_DATA_STOP COMPILER_RT_VISIBILITY;
31-
extern uint64_t PROF_CNTS_START COMPILER_RT_VISIBILITY;
32-
extern uint64_t PROF_CNTS_STOP COMPILER_RT_VISIBILITY;
33-
extern uint32_t PROF_ORDERFILE_START COMPILER_RT_VISIBILITY;
34-
extern char PROF_NAME_START COMPILER_RT_VISIBILITY;
35-
extern char PROF_NAME_STOP COMPILER_RT_VISIBILITY;
36-
extern ValueProfNode PROF_VNODES_START COMPILER_RT_VISIBILITY;
37-
extern ValueProfNode PROF_VNODES_STOP COMPILER_RT_VISIBILITY;
38-
39-
/* Add dummy data to ensure the section is always created. */
40-
__llvm_profile_data
41-
__prof_data_sect_data[0] COMPILER_RT_SECTION(INSTR_PROF_DATA_SECT_NAME);
42-
uint64_t
43-
__prof_cnts_sect_data[0] COMPILER_RT_SECTION(INSTR_PROF_CNTS_SECT_NAME);
44-
uint32_t
45-
__prof_orderfile_sect_data[0] COMPILER_RT_SECTION(INSTR_PROF_ORDERFILE_SECT_NAME);
46-
char __prof_nms_sect_data[0] COMPILER_RT_SECTION(INSTR_PROF_NAME_SECT_NAME);
47-
ValueProfNode __prof_vnodes_sect_data[0] COMPILER_RT_SECTION(INSTR_PROF_VNODES_SECT_NAME);
33+
extern __llvm_profile_data PROF_DATA_START COMPILER_RT_VISIBILITY COMPILER_RT_WEAK;
34+
extern __llvm_profile_data PROF_DATA_STOP COMPILER_RT_VISIBILITY COMPILER_RT_WEAK;
35+
extern uint64_t PROF_CNTS_START COMPILER_RT_VISIBILITY COMPILER_RT_WEAK;
36+
extern uint64_t PROF_CNTS_STOP COMPILER_RT_VISIBILITY COMPILER_RT_WEAK;
37+
extern uint32_t PROF_ORDERFILE_START COMPILER_RT_VISIBILITY COMPILER_RT_WEAK;
38+
extern char PROF_NAME_START COMPILER_RT_VISIBILITY COMPILER_RT_WEAK;
39+
extern char PROF_NAME_STOP COMPILER_RT_VISIBILITY COMPILER_RT_WEAK;
40+
extern ValueProfNode PROF_VNODES_START COMPILER_RT_VISIBILITY COMPILER_RT_WEAK;
41+
extern ValueProfNode PROF_VNODES_STOP COMPILER_RT_VISIBILITY COMPILER_RT_WEAK;
4842

4943
COMPILER_RT_VISIBILITY const __llvm_profile_data *
5044
__llvm_profile_begin_data(void) {

‎compiler-rt/test/profile/instrprof-gc-sections.c

Lines changed: 91 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,91 @@
1+
// REQUIRES: linux, lld-available
2+
3+
// RUN: %clang_profgen=%t.profraw -fuse-ld=lld -fcoverage-mapping -mllvm -enable-name-compression=false -DCODE=1 -ffunction-sections -fdata-sections -Wl,--gc-sections -o %t %s
4+
// RUN: %run %t
5+
// RUN: llvm-profdata merge -o %t.profdata %t.profraw
6+
// RUN: llvm-profdata show --all-functions %t.profdata | FileCheck %s -check-prefix=PROF
7+
// RUN: llvm-cov show %t -instr-profile %t.profdata | FileCheck %s -check-prefix=COV
8+
// RUN: llvm-nm %t | FileCheck %s -check-prefix=NM
9+
// RUN: llvm-readelf -x __llvm_prf_names %t | FileCheck %s -check-prefix=PRF_NAMES
10+
// RUN: llvm-readelf -x __llvm_prf_cnts %t | FileCheck %s -check-prefix=PRF_CNTS
11+
12+
// RUN: %clang_lto_profgen=%t.lto.profraw -fuse-ld=lld -fcoverage-mapping -mllvm -enable-name-compression=false -DCODE=1 -ffunction-sections -fdata-sections -Wl,--gc-sections -flto -o %t.lto %s
13+
// RUN: %run %t.lto
14+
// RUN: llvm-profdata merge -o %t.lto.profdata %t.lto.profraw
15+
// RUN: llvm-profdata show --all-functions %t.lto.profdata | FileCheck %s -check-prefix=PROF
16+
// RUN: llvm-cov show %t.lto -instr-profile %t.lto.profdata | FileCheck %s -check-prefix=COV
17+
// RUN: llvm-nm %t.lto | FileCheck %s -check-prefix=NM
18+
// RUN: llvm-readelf -x __llvm_prf_names %t.lto | FileCheck %s -check-prefix=PRF_NAMES
19+
// RUN: llvm-readelf -x __llvm_prf_cnts %t.lto | FileCheck %s -check-prefix=PRF_CNTS
20+
21+
// Note: We expect foo() and some of the profiling data associated with it to
22+
// be garbage collected.
23+
24+
// Note: When there is no code in a program, we expect to see the exact same
25+
// set of external functions provided by the profile runtime.
26+
27+
// RUN: %clang_profgen -fcoverage-mapping -ffunction-sections -fdata-sections -Wl,--gc-sections -shared -o %t.nocode.so %s
28+
// RUN: llvm-nm -jgU %t.nocode.so | grep -vE "__start_.*|__stop_.*" > %t.nocode.syms
29+
// RUN: llvm-nm -jgU %t | grep -vE "main|foo|_start|__libc_.*" > %t.code.syms
30+
// RUN: diff %t.nocode.syms %t.code.syms
31+
32+
// Note: We also check the IR instrumentation and expect foo() to be garbage
33+
// collected as well.
34+
35+
// RUN: %clang_pgogen=%t.pgo.profraw -fuse-ld=lld -DCODE=1 -ffunction-sections -fdata-sections -Wl,--gc-sections -o %t.pgo %s
36+
// RUN: %run %t.pgo
37+
// RUN: llvm-profdata merge -o %t.pgo.profdata %t.pgo.profraw
38+
// RUN: llvm-profdata show --all-functions %t.pgo.profdata | FileCheck %s -check-prefix=PGO
39+
// RUN: llvm-nm %t.pgo | FileCheck %s -check-prefix=NM
40+
41+
#ifdef CODE
42+
43+
// COV: [[@LINE+1]]{{ *}}|{{ *}}0|void foo()
44+
void foo() {}
45+
46+
// COV: [[@LINE+1]]{{ *}}|{{ *}}1|int main
47+
int main() { return 0; }
48+
49+
#endif // CODE
50+
51+
// NM-NOT: foo
52+
53+
// PROF: Counters:
54+
// PROF-NEXT: main:
55+
// PROF-NEXT: Hash:
56+
// PROF-NEXT: Counters: 1
57+
// PROF-NEXT: Function count: 1
58+
// PROF-NEXT: Instrumentation level: Front-end
59+
// PROF-NEXT: Functions shown: 1
60+
// PROF-NEXT: Total functions: 1
61+
// PROF-NEXT: Maximum function count:
62+
// PROF-NEXT: Maximum internal block count:
63+
64+
// Note: We don't expect the names of garbage collected functions to disappear
65+
// from __llvm_prf_names, because collectPGOFuncNameStrings() glues the names
66+
// together.
67+
68+
// PRF_NAMES: Hex dump of section '__llvm_prf_names':
69+
// PRF_NAMES-NEXT: {{.*}} 0800666f 6f016d61 696e {{.*$}}
70+
// | | f o o # m a i n
71+
// | |___________|
72+
// | |
73+
// UncompressedLen = 8 |
74+
// |
75+
// CompressedLen = 0
76+
77+
// Note: We expect the profile counters for garbage collected functions to also
78+
// be garbage collected.
79+
80+
// PRF_CNTS: Hex dump of section '__llvm_prf_cnts':
81+
// PRF_CNTS-NEXT: {{.*}} 00000000 00000000 {{.*$}}
82+
83+
// PGO: Counters:
84+
// PGO-NEXT: main:
85+
// PGO-NEXT: Hash:
86+
// PGO-NEXT: Counters: 1
87+
// PGO-NEXT: Instrumentation level: IR
88+
// PGO-NEXT: Functions shown: 1
89+
// PGO-NEXT: Total functions: 1
90+
// PGO-NEXT: Maximum function count:
91+
// PGO-NEXT: Maximum internal block count:

‎llvm/lib/Transforms/Instrumentation/InstrProfiling.cpp

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -844,6 +844,8 @@ InstrProfiling::getOrCreateRegionCounters(InstrProfIncrementInst *Inc) {
844844
CounterPtr->setAlignment(Align(8));
845845
MaybeSetComdat(CounterPtr);
846846
CounterPtr->setLinkage(Linkage);
847+
CounterPtr->setMetadata(LLVMContext::MD_associated,
848+
MDNode::get(Ctx, ValueAsMetadata::get(CounterPtr)));
847849

848850
auto *Int8PtrTy = Type::getInt8PtrTy(Ctx);
849851
// Allocate statically the array of pointers to value profile nodes for
@@ -865,6 +867,8 @@ InstrProfiling::getOrCreateRegionCounters(InstrProfIncrementInst *Inc) {
865867
getInstrProfSectionName(IPSK_vals, TT.getObjectFormat()));
866868
ValuesVar->setAlignment(Align(8));
867869
MaybeSetComdat(ValuesVar);
870+
ValuesVar->setMetadata(LLVMContext::MD_associated,
871+
MDNode::get(Ctx, ValueAsMetadata::get(CounterPtr)));
868872
ValuesPtrExpr =
869873
ConstantExpr::getBitCast(ValuesVar, Type::getInt8PtrTy(Ctx));
870874
}
@@ -899,6 +903,8 @@ InstrProfiling::getOrCreateRegionCounters(InstrProfIncrementInst *Inc) {
899903
Data->setAlignment(Align(INSTR_PROF_DATA_ALIGNMENT));
900904
MaybeSetComdat(Data);
901905
Data->setLinkage(Linkage);
906+
Data->setMetadata(LLVMContext::MD_associated,
907+
MDNode::get(Ctx, ValueAsMetadata::get(CounterPtr)));
902908

903909
PD.RegionCounters = CounterPtr;
904910
PD.DataVar = Data;

‎llvm/test/Instrumentation/InstrProfiling/associated.ll

Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,16 @@
1+
; RUN: opt < %s -instrprof -S | FileCheck %s
2+
; RUN: opt < %s -passes=instrprof -S | FileCheck %s
3+
4+
@__profn_foo = hidden constant [3 x i8] c"foo"
5+
6+
; CHECK: @__profc_foo = hidden global [1 x i64] zeroinitializer, section "__llvm_prf_cnts", align 8, !associated !0
7+
; CHECK: @__profd_foo = hidden global {{.*}}, section "__llvm_prf_data", align 8, !associated !0
8+
9+
define void @foo() {
10+
call void @llvm.instrprof.increment(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @__profn_foo, i32 0, i32 0), i64 0, i32 1, i32 0)
11+
ret void
12+
}
13+
14+
declare void @llvm.instrprof.increment(i8*, i64, i32, i32)
15+
16+
; CHECK: !0 = !{[1 x i64]* @__profc_foo}

‎llvm/test/Transforms/PGOProfile/associated.ll

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,11 @@
1+
; RUN: opt < %s -pgo-instr-gen -instrprof -S | FileCheck %s
2+
; RUN: opt < %s -passes=pgo-instr-gen,instrprof -S | FileCheck %s
3+
4+
; CHECK: @__profc_foo = private global [1 x i64] zeroinitializer, section "__llvm_prf_cnts", align 8, !associated !0
5+
; CHECK: @__profd_foo = private global {{.*}}, section "__llvm_prf_data", align 8, !associated !0
6+
7+
define void @foo() {
8+
ret void
9+
}
10+
11+
; CHECK: !0 = !{[1 x i64]* @__profc_foo}

‎llvm/test/Transforms/PGOProfile/counter_promo.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -60,7 +60,7 @@ bb12: ; preds = %bb9
6060
; ATOMIC_PROMO: atomicrmw add {{.*}} @__profc_foo{{.*}}0), i64 %[[LIVEOUT1]] seq_cst
6161
; ATOMIC_PROMO-NEXT: atomicrmw add {{.*}} @__profc_foo{{.*}}1), i64 %[[LIVEOUT2]] seq_cst
6262
; ATOMIC_PROMO-NEXT: atomicrmw add {{.*}} @__profc_foo{{.*}}2), i64 %[[LIVEOUT3]] seq_cst
63-
; PROMO-NOT: @__profc_foo
63+
; PROMO-NOT: @__profc_foo{{.*}})
6464

6565

6666
}

‎llvm/test/Transforms/PGOProfile/counter_promo_mexits.ll

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -69,7 +69,7 @@ bb15_0: ; preds = %bb11
6969
; PROMO-NEXT: %pgocount{{.*}} = load {{.*}} @__profc_foo{{.*}} 4)
7070
; PROMO-NEXT: add
7171
; PROMO-NEXT: store {{.*}}@__profc_foo{{.*}}4)
72-
; PROMO-NOT: @__profc_foo
72+
; PROMO-NOT: @__profc_foo{{.*}})
7373

7474

7575
bb15: ; preds = %bb14, %bb4

0 commit comments

Comments
 (0)