[clang-tools-extra] [compiler-rt] [clang] [PGO] Exposing PGO's Counter Reset and File Dumping APIs (PR #76471)
Qiongsi Wu via cfe-commits
cfe-commits at lists.llvm.org
Tue Jan 2 14:45:16 PST 2024
https://github.com/qiongsiwu updated https://github.com/llvm/llvm-project/pull/76471
>From 6c9381ec324595947237bd25642b03ab40b6a4df Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Wed, 27 Dec 2023 13:05:01 -0500
Subject: [PATCH 01/11] Initial commit
---
.../ExpandModularHeadersPPCallbacks.cpp | 2 +-
clang/include/clang/Frontend/Utils.h | 4 +-
clang/lib/Frontend/CompilerInstance.cpp | 2 +-
clang/lib/Frontend/InitPreprocessor.cpp | 12 ++--
compiler-rt/include/CMakeLists.txt | 1 +
.../include/profile/instr_prof_interface.h | 66 +++++++++++++++++++
compiler-rt/lib/profile/InstrProfiling.h | 32 +--------
7 files changed, 83 insertions(+), 36 deletions(-)
create mode 100644 compiler-rt/include/profile/instr_prof_interface.h
diff --git a/clang-tools-extra/clang-tidy/ExpandModularHeadersPPCallbacks.cpp b/clang-tools-extra/clang-tidy/ExpandModularHeadersPPCallbacks.cpp
index e414ac8c770508..5ecd4fb19131e4 100644
--- a/clang-tools-extra/clang-tidy/ExpandModularHeadersPPCallbacks.cpp
+++ b/clang-tools-extra/clang-tidy/ExpandModularHeadersPPCallbacks.cpp
@@ -100,7 +100,7 @@ ExpandModularHeadersPPCallbacks::ExpandModularHeadersPPCallbacks(
/*OwnsHeaderSearch=*/false);
PP->Initialize(Compiler.getTarget(), Compiler.getAuxTarget());
InitializePreprocessor(*PP, *PO, Compiler.getPCHContainerReader(),
- Compiler.getFrontendOpts());
+ Compiler.getFrontendOpts(), Compiler.getCodeGenOpts());
ApplyHeaderSearchOptions(*HeaderInfo, *HSO, LangOpts,
Compiler.getTarget().getTriple());
}
diff --git a/clang/include/clang/Frontend/Utils.h b/clang/include/clang/Frontend/Utils.h
index 143cf4359f00b5..604e42067a3f1e 100644
--- a/clang/include/clang/Frontend/Utils.h
+++ b/clang/include/clang/Frontend/Utils.h
@@ -43,12 +43,14 @@ class PCHContainerReader;
class Preprocessor;
class PreprocessorOptions;
class PreprocessorOutputOptions;
+class CodeGenOptions;
/// InitializePreprocessor - Initialize the preprocessor getting it and the
/// environment ready to process a single file.
void InitializePreprocessor(Preprocessor &PP, const PreprocessorOptions &PPOpts,
const PCHContainerReader &PCHContainerRdr,
- const FrontendOptions &FEOpts);
+ const FrontendOptions &FEOpts,
+ const CodeGenOptions &CodeGenOpts);
/// DoPrintPreprocessedInput - Implement -E mode.
void DoPrintPreprocessedInput(Preprocessor &PP, raw_ostream *OS,
diff --git a/clang/lib/Frontend/CompilerInstance.cpp b/clang/lib/Frontend/CompilerInstance.cpp
index 56bbef9697b650..ea44a26b6db7da 100644
--- a/clang/lib/Frontend/CompilerInstance.cpp
+++ b/clang/lib/Frontend/CompilerInstance.cpp
@@ -470,7 +470,7 @@ void CompilerInstance::createPreprocessor(TranslationUnitKind TUKind) {
// Predefine macros and configure the preprocessor.
InitializePreprocessor(*PP, PPOpts, getPCHContainerReader(),
- getFrontendOpts());
+ getFrontendOpts(), getCodeGenOpts());
// Initialize the header search object. In CUDA compilations, we use the aux
// triple (the host triple) to initialize our header search, since we need to
diff --git a/clang/lib/Frontend/InitPreprocessor.cpp b/clang/lib/Frontend/InitPreprocessor.cpp
index d83128adb511ef..009a67eea1eb52 100644
--- a/clang/lib/Frontend/InitPreprocessor.cpp
+++ b/clang/lib/Frontend/InitPreprocessor.cpp
@@ -1366,10 +1366,11 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
/// InitializePreprocessor - Initialize the preprocessor getting it and the
/// environment ready to process a single file.
-void clang::InitializePreprocessor(
- Preprocessor &PP, const PreprocessorOptions &InitOpts,
- const PCHContainerReader &PCHContainerRdr,
- const FrontendOptions &FEOpts) {
+void clang::InitializePreprocessor(Preprocessor &PP,
+ const PreprocessorOptions &InitOpts,
+ const PCHContainerReader &PCHContainerRdr,
+ const FrontendOptions &FEOpts,
+ const CodeGenOptions &CodeGenOpts) {
const LangOptions &LangOpts = PP.getLangOpts();
std::string PredefineBuffer;
PredefineBuffer.reserve(4080);
@@ -1416,6 +1417,9 @@ void clang::InitializePreprocessor(
InitializeStandardPredefinedMacros(PP.getTargetInfo(), PP.getLangOpts(),
FEOpts, Builder);
+ if (CodeGenOpts.hasProfileIRInstr())
+ Builder.defineMacro("__LLVM_INSTR_PROFILE_GENERATE");
+
// Add on the predefines from the driver. Wrap in a #line directive to report
// that they come from the command line.
Builder.append("# 1 \"<command line>\" 1");
diff --git a/compiler-rt/include/CMakeLists.txt b/compiler-rt/include/CMakeLists.txt
index 78427beedb3cc4..7a100c66bbcfda 100644
--- a/compiler-rt/include/CMakeLists.txt
+++ b/compiler-rt/include/CMakeLists.txt
@@ -44,6 +44,7 @@ endif(COMPILER_RT_BUILD_ORC)
if (COMPILER_RT_BUILD_PROFILE)
set(PROFILE_HEADERS
profile/InstrProfData.inc
+ profile/instr_prof_interface.h
)
endif(COMPILER_RT_BUILD_PROFILE)
diff --git a/compiler-rt/include/profile/instr_prof_interface.h b/compiler-rt/include/profile/instr_prof_interface.h
new file mode 100644
index 00000000000000..6cbf6b414f3af1
--- /dev/null
+++ b/compiler-rt/include/profile/instr_prof_interface.h
@@ -0,0 +1,66 @@
+/*===---- instr_profiling.h - Instrumentation PGO User Program API ----------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ *
+ * This header provides a public interface for user programs to provide
+ * fine-grained control of profile dumping.
+ *
+\*===---------------------------------------------------------------------===*/
+
+#ifndef COMPILER_RT_INSTR_PROFILING
+#define COMPILER_RT_INSTR_PROFILING
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef __LLVM_INSTR_PROFILE_GENERATE
+// Profile file reset and dump interfaces.
+// Only defined when `-fprofile-generate` is in effect.
+
+/*!
+ * \brief Interface to set all PGO counters to zero for the current process.
+ *
+ */
+void __llvm_profile_reset_counters(void);
+
+/*!
+ * \brief this is a wrapper interface to \c __llvm_profile_write_file.
+ * After this interface is invoked, an already dumped flag will be set
+ * so that profile won't be dumped again during program exit.
+ * Invocation of interface __llvm_profile_reset_counters will clear
+ * the flag. This interface is designed to be used to collect profile
+ * data from user selected hot regions. The use model is
+ * __llvm_profile_reset_counters();
+ * ... hot region 1
+ * __llvm_profile_dump();
+ * .. some other code
+ * __llvm_profile_reset_counters();
+ * ... hot region 2
+ * __llvm_profile_dump();
+ *
+ * It is expected that on-line profile merging is on with \c %m specifier
+ * used in profile filename . If merging is not turned on, user is expected
+ * to invoke __llvm_profile_set_filename to specify different profile names
+ * for different regions before dumping to avoid profile write clobbering.
+ */
+int __llvm_profile_dump(void);
+
+// Interface to dump the current process' order file to disk.
+int __llvm_orderfile_dump(void);
+
+#else
+#define __llvm_profile_reset_counters()
+#define __llvm_profile_dump()
+#define __llvm_orderfile_dump()
+#endif
+
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
+#endif
diff --git a/compiler-rt/lib/profile/InstrProfiling.h b/compiler-rt/lib/profile/InstrProfiling.h
index 137115996748ce..3f2c75b5df9064 100644
--- a/compiler-rt/lib/profile/InstrProfiling.h
+++ b/compiler-rt/lib/profile/InstrProfiling.h
@@ -12,6 +12,9 @@
#include "InstrProfilingPort.h"
#include <stdio.h>
+#define __LLVM_INSTR_PROFILE_GENERATE
+#include "profile/instr_prof_interface.h"
+
#define INSTR_PROF_VISIBILITY COMPILER_RT_VISIBILITY
#include "profile/InstrProfData.inc"
@@ -100,12 +103,6 @@ ValueProfNode *__llvm_profile_begin_vnodes();
ValueProfNode *__llvm_profile_end_vnodes();
uint32_t *__llvm_profile_begin_orderfile();
-/*!
- * \brief Clear profile counters to zero.
- *
- */
-void __llvm_profile_reset_counters(void);
-
/*!
* \brief Merge profile data from buffer.
*
@@ -156,29 +153,6 @@ void __llvm_profile_instrument_target_value(uint64_t TargetValue, void *Data,
int __llvm_profile_write_file(void);
int __llvm_orderfile_write_file(void);
-/*!
- * \brief this is a wrapper interface to \c __llvm_profile_write_file.
- * After this interface is invoked, an already dumped flag will be set
- * so that profile won't be dumped again during program exit.
- * Invocation of interface __llvm_profile_reset_counters will clear
- * the flag. This interface is designed to be used to collect profile
- * data from user selected hot regions. The use model is
- * __llvm_profile_reset_counters();
- * ... hot region 1
- * __llvm_profile_dump();
- * .. some other code
- * __llvm_profile_reset_counters();
- * ... hot region 2
- * __llvm_profile_dump();
- *
- * It is expected that on-line profile merging is on with \c %m specifier
- * used in profile filename . If merging is not turned on, user is expected
- * to invoke __llvm_profile_set_filename to specify different profile names
- * for different regions before dumping to avoid profile write clobbering.
- */
-int __llvm_profile_dump(void);
-
-int __llvm_orderfile_dump(void);
/*!
* \brief Set the filename for writing instrumentation data.
>From a6c3b4619ea93cec2be962c60181985a379f8553 Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Wed, 27 Dec 2023 14:02:16 -0500
Subject: [PATCH 02/11] Frontend minor cleanup and adding a test
---
clang/include/clang/Basic/CodeGenOptions.h | 6 ++++++
clang/lib/Frontend/InitPreprocessor.cpp | 12 ++++++++++--
clang/test/Preprocessor/pgo-init.c | 5 +++++
3 files changed, 21 insertions(+), 2 deletions(-)
create mode 100644 clang/test/Preprocessor/pgo-init.c
diff --git a/clang/include/clang/Basic/CodeGenOptions.h b/clang/include/clang/Basic/CodeGenOptions.h
index 6952b48e898a81..e06f1094784caf 100644
--- a/clang/include/clang/Basic/CodeGenOptions.h
+++ b/clang/include/clang/Basic/CodeGenOptions.h
@@ -494,6 +494,12 @@ class CodeGenOptions : public CodeGenOptionsBase {
return getProfileInstr() == ProfileCSIRInstr;
}
+ /// Check if any form of instrumentation is on.
+ bool hasProfileInstr() const {
+ return hasProfileClangInstr() || hasProfileIRInstr() ||
+ hasProfileCSIRInstr();
+ }
+
/// Check if Clang profile use is on.
bool hasProfileClangUse() const {
return getProfileUse() == ProfileClangInstr;
diff --git a/clang/lib/Frontend/InitPreprocessor.cpp b/clang/lib/Frontend/InitPreprocessor.cpp
index 009a67eea1eb52..0386a75ac429fa 100644
--- a/clang/lib/Frontend/InitPreprocessor.cpp
+++ b/clang/lib/Frontend/InitPreprocessor.cpp
@@ -1364,6 +1364,12 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
TI.getTargetDefines(LangOpts, Builder);
}
+static void InitializePGOProfileMacros(const CodeGenOptions &CodeGenOpts,
+ MacroBuilder &Builder) {
+ if (CodeGenOpts.hasProfileInstr())
+ Builder.defineMacro("__LLVM_INSTR_PROFILE_GENERATE");
+}
+
/// InitializePreprocessor - Initialize the preprocessor getting it and the
/// environment ready to process a single file.
void clang::InitializePreprocessor(Preprocessor &PP,
@@ -1417,8 +1423,10 @@ void clang::InitializePreprocessor(Preprocessor &PP,
InitializeStandardPredefinedMacros(PP.getTargetInfo(), PP.getLangOpts(),
FEOpts, Builder);
- if (CodeGenOpts.hasProfileIRInstr())
- Builder.defineMacro("__LLVM_INSTR_PROFILE_GENERATE");
+ // The PGO instrumentation profile macros are driven by options
+ // -fprofile[-instr]-generate/-fcs-profile-generate/-fprofile[-instr]-use,
+ // hence they are not guarded by InitOpts.UsePredefines.
+ InitializePGOProfileMacros(CodeGenOpts, Builder);
// Add on the predefines from the driver. Wrap in a #line directive to report
// that they come from the command line.
diff --git a/clang/test/Preprocessor/pgo-init.c b/clang/test/Preprocessor/pgo-init.c
new file mode 100644
index 00000000000000..6ac28708fddf71
--- /dev/null
+++ b/clang/test/Preprocessor/pgo-init.c
@@ -0,0 +1,5 @@
+// RUN: %clang -fprofile-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGEN %s
+// RUN: %clang -fprofile-instr-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGEN %s
+// RUN: %clang -fcs-profile-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGEN %s
+//
+// PROFGEN:#define __LLVM_INSTR_PROFILE_GENERATE 1
>From 18ec4ac08a08473f146e842dbeefbe7171ce3eec Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Wed, 27 Dec 2023 16:01:22 -0500
Subject: [PATCH 03/11] Refactoring tests, adding profile-use macro
---
clang/lib/Frontend/InitPreprocessor.cpp | 3 +++
clang/test/Preprocessor/pgo-init.c | 5 -----
clang/test/Profile/c-general.c | 9 ++++++++
compiler-rt/test/profile/instrprof-api.c | 26 ++++++++++++++++++++++++
4 files changed, 38 insertions(+), 5 deletions(-)
delete mode 100644 clang/test/Preprocessor/pgo-init.c
create mode 100644 compiler-rt/test/profile/instrprof-api.c
diff --git a/clang/lib/Frontend/InitPreprocessor.cpp b/clang/lib/Frontend/InitPreprocessor.cpp
index 0386a75ac429fa..fe0fd3614113c4 100644
--- a/clang/lib/Frontend/InitPreprocessor.cpp
+++ b/clang/lib/Frontend/InitPreprocessor.cpp
@@ -1368,6 +1368,9 @@ static void InitializePGOProfileMacros(const CodeGenOptions &CodeGenOpts,
MacroBuilder &Builder) {
if (CodeGenOpts.hasProfileInstr())
Builder.defineMacro("__LLVM_INSTR_PROFILE_GENERATE");
+
+ if (CodeGenOpts.hasProfileIRUse() || CodeGenOpts.hasProfileClangUse())
+ Builder.defineMacro("__LLVM_INSTR_PROFILE_USE");
}
/// InitializePreprocessor - Initialize the preprocessor getting it and the
diff --git a/clang/test/Preprocessor/pgo-init.c b/clang/test/Preprocessor/pgo-init.c
deleted file mode 100644
index 6ac28708fddf71..00000000000000
--- a/clang/test/Preprocessor/pgo-init.c
+++ /dev/null
@@ -1,5 +0,0 @@
-// RUN: %clang -fprofile-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGEN %s
-// RUN: %clang -fprofile-instr-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGEN %s
-// RUN: %clang -fcs-profile-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGEN %s
-//
-// PROFGEN:#define __LLVM_INSTR_PROFILE_GENERATE 1
diff --git a/clang/test/Profile/c-general.c b/clang/test/Profile/c-general.c
index b841f9c3d2a1d1..fa9eb7ebc78274 100644
--- a/clang/test/Profile/c-general.c
+++ b/clang/test/Profile/c-general.c
@@ -9,6 +9,15 @@
// Also check compatibility with older profiles.
// RUN: %clang_cc1 -triple x86_64-apple-macosx10.9 -main-file-name c-general.c %s -o - -emit-llvm -fprofile-instrument-use-path=%S/Inputs/c-general.profdata.v1 | FileCheck -allow-deprecated-dag-overlap -check-prefix=PGOUSE %s
+// RUN: %clang -fprofile-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGENMACRO %s
+// RUN: %clang -fprofile-instr-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGENMACRO %s
+// RUN: %clang -fcs-profile-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGENMACRO %s
+//
+// RUN: %clang -fprofile-use=%t.profdata -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFUSEMACRO %s
+
+// PROFGENMACRO:#define __LLVM_INSTR_PROFILE_GENERATE 1
+// PROFUSEMACRO:#define __LLVM_INSTR_PROFILE_USE 1
+
// PGOGEN: @[[SLC:__profc_simple_loops]] = private global [4 x i64] zeroinitializer
// PGOGEN: @[[IFC:__profc_conditionals]] = private global [13 x i64] zeroinitializer
// PGOGEN: @[[EEC:__profc_early_exits]] = private global [9 x i64] zeroinitializer
diff --git a/compiler-rt/test/profile/instrprof-api.c b/compiler-rt/test/profile/instrprof-api.c
new file mode 100644
index 00000000000000..600fbb575feea3
--- /dev/null
+++ b/compiler-rt/test/profile/instrprof-api.c
@@ -0,0 +1,26 @@
+// RUN: %clang_profgen %s -S -emit-llvm -o - | FileCheck %s --check-prefix=PROFGEN
+// RUN: %clang_profgen -o %t %s
+// RUN: env LLVM_PROFILE_FILE=%t.profraw %run %t
+// RUN: llvm-profdata merge -o %t.profdata %t.profraw
+// RUN: %clang_profuse=%t.profdata %s -S -emit-llvm -o - | FileCheck %s --check-prefix=PROFUSE
+#include "profile/instr_profiling.h"
+
+__attribute__((noinline)) int bar() { return 4; }
+
+int foo() {
+ __llvm_profile_reset_counters();
+ // PROFGEN: call void @__llvm_profile_reset_counters()
+ // PROFUSE-NOT: call void @__llvm_profile_reset_counters()
+ return bar();
+}
+
+int main() {
+ int z = foo() + 3;
+ __llvm_profile_dump();
+ // PROFGEN: %call1 = call signext i32 @__llvm_profile_dump()
+ // PROFUSE-NOT: %call1 = call signext i32 @__llvm_profile_dump()
+ __llvm_orderfile_dump();
+ // PROFGEN: %call2 = call signext i32 @__llvm_orderfile_dump()
+ // PROFUSE-NOT: %call2 = call signext i32 @__llvm_orderfile_dump()
+ return z + bar() - 11;
+}
>From b741a557715eb17b69fa5b4013f2efeadcb80fda Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Wed, 27 Dec 2023 16:08:25 -0500
Subject: [PATCH 04/11] Fix header comment
---
compiler-rt/include/profile/instr_prof_interface.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/compiler-rt/include/profile/instr_prof_interface.h b/compiler-rt/include/profile/instr_prof_interface.h
index 6cbf6b414f3af1..7d0516e5f6eaec 100644
--- a/compiler-rt/include/profile/instr_prof_interface.h
+++ b/compiler-rt/include/profile/instr_prof_interface.h
@@ -1,4 +1,4 @@
-/*===---- instr_profiling.h - Instrumentation PGO User Program API ----------===
+/*===---- instr_prof_interface.h - Instrumentation PGO User Program API ----===
*
* Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
* See https://llvm.org/LICENSE.txt for license information.
@@ -7,7 +7,7 @@
*===-----------------------------------------------------------------------===
*
* This header provides a public interface for user programs to provide
- * fine-grained control of profile dumping.
+ * fine-grained control of counter reset and profile dumping.
*
\*===---------------------------------------------------------------------===*/
>From 13a4bb39165df84668b6958cecf6aa46a1235731 Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Wed, 27 Dec 2023 16:42:04 -0500
Subject: [PATCH 05/11] Fix test case after header renaming.
---
compiler-rt/test/profile/instrprof-api.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/compiler-rt/test/profile/instrprof-api.c b/compiler-rt/test/profile/instrprof-api.c
index 600fbb575feea3..8d21dca69a5d60 100644
--- a/compiler-rt/test/profile/instrprof-api.c
+++ b/compiler-rt/test/profile/instrprof-api.c
@@ -3,7 +3,7 @@
// RUN: env LLVM_PROFILE_FILE=%t.profraw %run %t
// RUN: llvm-profdata merge -o %t.profdata %t.profraw
// RUN: %clang_profuse=%t.profdata %s -S -emit-llvm -o - | FileCheck %s --check-prefix=PROFUSE
-#include "profile/instr_profiling.h"
+#include "profile/instr_prof_interface.h"
__attribute__((noinline)) int bar() { return 4; }
>From cce25f062c00424c153254fc159309808441414c Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Wed, 27 Dec 2023 18:25:14 -0500
Subject: [PATCH 06/11] Address review comments
---
compiler-rt/include/profile/instr_prof_interface.h | 4 ++--
compiler-rt/lib/profile/InstrProfiling.h | 10 ++++++++++
compiler-rt/test/profile/instrprof-api.c | 6 ++++--
3 files changed, 16 insertions(+), 4 deletions(-)
diff --git a/compiler-rt/include/profile/instr_prof_interface.h b/compiler-rt/include/profile/instr_prof_interface.h
index 7d0516e5f6eaec..4905b913503dff 100644
--- a/compiler-rt/include/profile/instr_prof_interface.h
+++ b/compiler-rt/include/profile/instr_prof_interface.h
@@ -55,8 +55,8 @@ int __llvm_orderfile_dump(void);
#else
#define __llvm_profile_reset_counters()
-#define __llvm_profile_dump()
-#define __llvm_orderfile_dump()
+#define __llvm_profile_dump() (0)
+#define __llvm_orderfile_dump() (0)
#endif
#ifdef __cplusplus
diff --git a/compiler-rt/lib/profile/InstrProfiling.h b/compiler-rt/lib/profile/InstrProfiling.h
index 3f2c75b5df9064..f921c361c17e2b 100644
--- a/compiler-rt/lib/profile/InstrProfiling.h
+++ b/compiler-rt/lib/profile/InstrProfiling.h
@@ -12,8 +12,18 @@
#include "InstrProfilingPort.h"
#include <stdio.h>
+// Make sure __LLVM_INSTR_PROFILE_GENERATE is always defined before
+// including instr_prof_interface.h so the interface functions are
+// declared correctly for the runtime. Additionally, make sure
+// that __LLVM_INSTR_PROFILE_GENERATE is undefined only when it is
+// not explicitly defined somewhere else.
+#ifndef __LLVM_INSTR_PROFILE_GENERATE
#define __LLVM_INSTR_PROFILE_GENERATE
#include "profile/instr_prof_interface.h"
+#undef __LLVM_INSTR_PROFILE_GENERATE
+#else
+#include "profile/instr_prof_interface.h"
+#endif
#define INSTR_PROF_VISIBILITY COMPILER_RT_VISIBILITY
#include "profile/InstrProfData.inc"
diff --git a/compiler-rt/test/profile/instrprof-api.c b/compiler-rt/test/profile/instrprof-api.c
index 8d21dca69a5d60..894e4b92e8348d 100644
--- a/compiler-rt/test/profile/instrprof-api.c
+++ b/compiler-rt/test/profile/instrprof-api.c
@@ -1,8 +1,10 @@
-// RUN: %clang_profgen %s -S -emit-llvm -o - | FileCheck %s --check-prefix=PROFGEN
+// RUN: %clang_profgen %s --target=ppc64le-unknown-linux-gnu -S \
+// RUN: -emit-llvm -o - | FileCheck %s --check-prefix=PROFGEN
// RUN: %clang_profgen -o %t %s
// RUN: env LLVM_PROFILE_FILE=%t.profraw %run %t
// RUN: llvm-profdata merge -o %t.profdata %t.profraw
-// RUN: %clang_profuse=%t.profdata %s -S -emit-llvm -o - | FileCheck %s --check-prefix=PROFUSE
+// RUN: %clang_profuse=%t.profdata %s --target=ppc64le-unknown-linux-gnu -S \
+// RUN: -emit-llvm -o - | FileCheck %s --check-prefix=PROFUSE
#include "profile/instr_prof_interface.h"
__attribute__((noinline)) int bar() { return 4; }
>From ec4786808381b1a85e051d034408c8009290d8b4 Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Thu, 28 Dec 2023 09:40:08 -0500
Subject: [PATCH 07/11] Adding a new test to cover weak symbols. Adding a new
interface function to support setting profile file name.
---
.../include/profile/instr_prof_interface.h | 34 ++++++++++++++++---
compiler-rt/lib/profile/InstrProfiling.h | 21 ------------
.../profile/Linux/instrprof-weak-symbol.c | 17 ++++++++++
compiler-rt/test/profile/instrprof-api.c | 12 ++++++-
4 files changed, 58 insertions(+), 26 deletions(-)
create mode 100644 compiler-rt/test/profile/Linux/instrprof-weak-symbol.c
diff --git a/compiler-rt/include/profile/instr_prof_interface.h b/compiler-rt/include/profile/instr_prof_interface.h
index 4905b913503dff..06f31210d73bea 100644
--- a/compiler-rt/include/profile/instr_prof_interface.h
+++ b/compiler-rt/include/profile/instr_prof_interface.h
@@ -6,8 +6,9 @@
*
*===-----------------------------------------------------------------------===
*
- * This header provides a public interface for user programs to provide
- * fine-grained control of counter reset and profile dumping.
+ * This header provides a public interface for fine-grained control of counter
+ * reset and profile dumping. These interface functions can be directly called
+ * in user programs.
*
\*===---------------------------------------------------------------------===*/
@@ -20,7 +21,29 @@ extern "C" {
#ifdef __LLVM_INSTR_PROFILE_GENERATE
// Profile file reset and dump interfaces.
-// Only defined when `-fprofile-generate` is in effect.
+// When `-fprofile[-instr]-generate`/`-fcs-profile-generate` is in effect,
+// clang defines __LLVM_INSTR_PROFILE_GENERATE to pick up the API calls.
+
+/*!
+ * \brief Set the filename for writing instrumentation data.
+ *
+ * Sets the filename to be used for subsequent calls to
+ * \a __llvm_profile_write_file().
+ *
+ * \c Name is not copied, so it must remain valid. Passing NULL resets the
+ * filename logic to the default behaviour.
+ *
+ * Note: There may be multiple copies of the profile runtime (one for each
+ * instrumented image/DSO). This API only modifies the filename within the
+ * copy of the runtime available to the calling image.
+ *
+ * Warning: This is a no-op if continuous mode (\ref
+ * __llvm_profile_is_continuous_mode_enabled) is on. The reason for this is
+ * that in continuous mode, profile counters are mmap()'d to the profile at
+ * program initialization time. Support for transferring the mmap'd profile
+ * counts to a new file has not been implemented.
+ */
+void __llvm_profile_set_filename(const char *Name);
/*!
* \brief Interface to set all PGO counters to zero for the current process.
@@ -40,7 +63,7 @@ void __llvm_profile_reset_counters(void);
* __llvm_profile_dump();
* .. some other code
* __llvm_profile_reset_counters();
- * ... hot region 2
+ * ... hot region 2
* __llvm_profile_dump();
*
* It is expected that on-line profile merging is on with \c %m specifier
@@ -54,9 +77,12 @@ int __llvm_profile_dump(void);
int __llvm_orderfile_dump(void);
#else
+
+#define __llvm_profile_set_filename(Name)
#define __llvm_profile_reset_counters()
#define __llvm_profile_dump() (0)
#define __llvm_orderfile_dump() (0)
+
#endif
#ifdef __cplusplus
diff --git a/compiler-rt/lib/profile/InstrProfiling.h b/compiler-rt/lib/profile/InstrProfiling.h
index f921c361c17e2b..2210e048210637 100644
--- a/compiler-rt/lib/profile/InstrProfiling.h
+++ b/compiler-rt/lib/profile/InstrProfiling.h
@@ -164,27 +164,6 @@ int __llvm_profile_write_file(void);
int __llvm_orderfile_write_file(void);
-/*!
- * \brief Set the filename for writing instrumentation data.
- *
- * Sets the filename to be used for subsequent calls to
- * \a __llvm_profile_write_file().
- *
- * \c Name is not copied, so it must remain valid. Passing NULL resets the
- * filename logic to the default behaviour.
- *
- * Note: There may be multiple copies of the profile runtime (one for each
- * instrumented image/DSO). This API only modifies the filename within the
- * copy of the runtime available to the calling image.
- *
- * Warning: This is a no-op if continuous mode (\ref
- * __llvm_profile_is_continuous_mode_enabled) is on. The reason for this is
- * that in continuous mode, profile counters are mmap()'d to the profile at
- * program initialization time. Support for transferring the mmap'd profile
- * counts to a new file has not been implemented.
- */
-void __llvm_profile_set_filename(const char *Name);
-
/*!
* \brief Set the FILE object for writing instrumentation data. Return 0 if set
* successfully or return 1 if failed.
diff --git a/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c b/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c
new file mode 100644
index 00000000000000..88292974bdada3
--- /dev/null
+++ b/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c
@@ -0,0 +1,17 @@
+// RUN: %clang_pgogen -o %t %s
+// RUN: not %t
+// RUN: %clang -o %t %s
+// RUN: %t
+
+__attribute__((weak)) void __llvm_profile_reset_counters(void);
+
+__attribute__((noinline)) int bar() { return 4; }
+int foo() {
+ if (__llvm_profile_reset_counters) {
+ __llvm_profile_reset_counters();
+ return 0;
+ }
+ return bar();
+}
+
+int main() { return foo() - 4; }
diff --git a/compiler-rt/test/profile/instrprof-api.c b/compiler-rt/test/profile/instrprof-api.c
index 894e4b92e8348d..8b235f7b5e674c 100644
--- a/compiler-rt/test/profile/instrprof-api.c
+++ b/compiler-rt/test/profile/instrprof-api.c
@@ -16,9 +16,15 @@ int foo() {
return bar();
}
+// PROFUSE-NOT: declare void @__llvm_profile_reset_counters()
+
int main() {
int z = foo() + 3;
- __llvm_profile_dump();
+ __llvm_profile_set_filename("rawprof.profraw");
+ // PROFGEN: call void @__llvm_profile_set_filename(ptr noundef @.str)
+ // PROFUSE-NOT: call void @__llvm_profile_set_filename(ptr noundef @.str)
+ if (__llvm_profile_dump())
+ return 2;
// PROFGEN: %call1 = call signext i32 @__llvm_profile_dump()
// PROFUSE-NOT: %call1 = call signext i32 @__llvm_profile_dump()
__llvm_orderfile_dump();
@@ -26,3 +32,7 @@ int main() {
// PROFUSE-NOT: %call2 = call signext i32 @__llvm_orderfile_dump()
return z + bar() - 11;
}
+
+// PROFUSE-NOT: declare void @__llvm_profile_set_filename(ptr noundef)
+// PROFUSE-NOT: declare signext i32 @__llvm_profile_dump()
+// PROFUSE-NOT: declare signext i32 @__llvm_orderfile_dump()
>From 29086fc6a28be54999afab4787eeaf9334e40b5c Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Thu, 28 Dec 2023 20:02:00 -0500
Subject: [PATCH 08/11] Address code review
---
clang/docs/UsersManual.rst | 91 +++++++++++++++++++
clang/test/Profile/c-general.c | 1 +
compiler-rt/lib/profile/InstrProfiling.h | 10 +-
.../profile/Linux/instrprof-weak-symbol.c | 9 +-
4 files changed, 99 insertions(+), 12 deletions(-)
diff --git a/clang/docs/UsersManual.rst b/clang/docs/UsersManual.rst
index 7c30570437e8b0..82add00423a026 100644
--- a/clang/docs/UsersManual.rst
+++ b/clang/docs/UsersManual.rst
@@ -2809,6 +2809,97 @@ indexed format, regardeless whether it is produced by frontend or the IR pass.
overhead. ``prefer-atomic`` will be transformed to ``atomic`` when supported
by the target, or ``single`` otherwise.
+Fine Tuning Profile Collection
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+The PGO infrastructure provides user program knobs to fine tune profile
+collection. Specifically, the PGO runtime provides the following functions
+that can be used to control the regions in the program where profiles should
+be collected.
+
+ * ``void __llvm_profile_set_filename(const char *Name)``: changes the name of
+ the profile file to ``Name``.
+ * ``void __llvm_profile_reset_counters(void)``: resets all counters to zero.
+ * ``int __llvm_profile_dump(void)``: write the profile data to disk.
+ * ``int __llvm_orderfile_dump(void)``: write the order file to disk.
+
+For example, the following pattern can be used to skip profiling program
+initialization, profile two specific hot regions, and skip profiling program
+cleanup:
+
+.. code-block:: c
+ int main() {
+ initialize();
+
+ // Reset all profile counters to 0 to omit profile collected during
+ // initialize()'s execution.
+ __llvm_profile_reset_counters();
+ ... hot region 1
+ // Dump the profile for hot region 1.
+ __llvm_profile_set_filename("region1.profraw");
+ __llvm_profile_dump();
+
+ // Reset counters before proceeding to hot region 2.
+ __llvm_profile_reset_counters();
+ ... hot region 2
+ // Dump the profile for hot region 2.
+ __llvm_profile_set_filename("region2.profraw");
+ __llvm_profile_dump();
+
+ // Since the profile has been dumped, no further profile data
+ // will be collected beyond the above __llvm_profile_dump().
+ cleanup();
+ return 0;
+ }
+
+These APIs' names can be introduced to user programs in two ways.
+They can be declared as weak symbols on platforms which support
+treating weak symbols as ``null`` during linking. For example, the user can
+have
+
+.. code-block:: c
+ __attribute__((weak)) int __llvm_profile_dump(void);
+
+ // Then later in the same source file
+ if (__llvm_profile_dump)
+ if (__llvm_profile_dump() != 0) { ... }
+ // The first if condition tests if the symbol is actually defined.
+ // Profile dumping only happens if the symbol is defined. Hence,
+ // the user program works correctly during normal (not profile-generate)
+ // executions.
+
+Alternatively, the user program can include the header
+``profile/instr_prof_interface.h``, which contains the API names. For example,
+
+.. code-block:: c
+ #include "profile/instr_prof_interface.h"
+
+ // Then later in the same source file
+ if (__llvm_profile_dump() != 0) { ... }
+
+The user code does not need to check if the API names are defined, because
+these names are automatically replaced by ``(0)`` or the equivalence of noop
+if the ``clang`` is not compiling for profile generation.
+
+Such replacement can happen because ``clang`` adds one of two macros depending
+on the ``-fprofile-generate`` and the ``-fprofile-use`` flags.
+
+ * ``__LLVM_INSTR_PROFILE_GENERATE``: defined when one of
+ ``-fprofile[-instr]-generate``/``-fcs-profile-generate`` is in effect.
+ * ``__LLVM_INSTR_PROFILE_USE``: defined when one of
+ ``-fprofile-use``/``-fprofile-instr-use`` is in effect.
+
+The two macros can be used to provide more flexibiilty so a user program
+can execute code specifically intended for profile generate or profile use.
+For example, a user program can have special logging during profile generate:
+
+.. code-block:: c
+ #if __LLVM_INSTR_PROFILE_GENERATE
+ expensive_logging_of_full_program_state();
+ #endif
+
+The logging is automatically excluded during a normal build of the program,
+hence it does not impact performance during a normal execution.
+
Disabling Instrumentation
^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/clang/test/Profile/c-general.c b/clang/test/Profile/c-general.c
index fa9eb7ebc78274..2f621ec9b0bf9d 100644
--- a/clang/test/Profile/c-general.c
+++ b/clang/test/Profile/c-general.c
@@ -14,6 +14,7 @@
// RUN: %clang -fcs-profile-generate -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFGENMACRO %s
//
// RUN: %clang -fprofile-use=%t.profdata -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFUSEMACRO %s
+// RUN: %clang -fprofile-instr-use=%t.profdata -E -dM %s | FileCheck -match-full-lines -check-prefix=PROFUSEMACRO %s
// PROFGENMACRO:#define __LLVM_INSTR_PROFILE_GENERATE 1
// PROFUSEMACRO:#define __LLVM_INSTR_PROFILE_USE 1
diff --git a/compiler-rt/lib/profile/InstrProfiling.h b/compiler-rt/lib/profile/InstrProfiling.h
index 2210e048210637..01239083369187 100644
--- a/compiler-rt/lib/profile/InstrProfiling.h
+++ b/compiler-rt/lib/profile/InstrProfiling.h
@@ -14,16 +14,14 @@
// Make sure __LLVM_INSTR_PROFILE_GENERATE is always defined before
// including instr_prof_interface.h so the interface functions are
-// declared correctly for the runtime. Additionally, make sure
-// that __LLVM_INSTR_PROFILE_GENERATE is undefined only when it is
-// not explicitly defined somewhere else.
+// declared correctly for the runtime.
+// __LLVM_INSTR_PROFILE_GENERATE is always `#undef`ed after the header,
+// because compiler-rt does not support profiling the profiling runtime itself.
#ifndef __LLVM_INSTR_PROFILE_GENERATE
#define __LLVM_INSTR_PROFILE_GENERATE
+#endif
#include "profile/instr_prof_interface.h"
#undef __LLVM_INSTR_PROFILE_GENERATE
-#else
-#include "profile/instr_prof_interface.h"
-#endif
#define INSTR_PROF_VISIBILITY COMPILER_RT_VISIBILITY
#include "profile/InstrProfData.inc"
diff --git a/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c b/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c
index 88292974bdada3..4af69fecb34f63 100644
--- a/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c
+++ b/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c
@@ -5,13 +5,10 @@
__attribute__((weak)) void __llvm_profile_reset_counters(void);
-__attribute__((noinline)) int bar() { return 4; }
-int foo() {
+int main() {
if (__llvm_profile_reset_counters) {
__llvm_profile_reset_counters();
- return 0;
+ return 1;
}
- return bar();
+ return 0;
}
-
-int main() { return foo() - 4; }
>From 8e2bf6952c651e5f3094ea6ba0057bcf3e0b0e52 Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Thu, 28 Dec 2023 22:25:15 -0500
Subject: [PATCH 09/11] Fixing documentation build error.
---
clang/docs/UsersManual.rst | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/clang/docs/UsersManual.rst b/clang/docs/UsersManual.rst
index 82add00423a026..2612a029165389 100644
--- a/clang/docs/UsersManual.rst
+++ b/clang/docs/UsersManual.rst
@@ -2827,6 +2827,7 @@ initialization, profile two specific hot regions, and skip profiling program
cleanup:
.. code-block:: c
+
int main() {
initialize();
@@ -2857,6 +2858,7 @@ treating weak symbols as ``null`` during linking. For example, the user can
have
.. code-block:: c
+
__attribute__((weak)) int __llvm_profile_dump(void);
// Then later in the same source file
@@ -2871,6 +2873,7 @@ Alternatively, the user program can include the header
``profile/instr_prof_interface.h``, which contains the API names. For example,
.. code-block:: c
+
#include "profile/instr_prof_interface.h"
// Then later in the same source file
@@ -2893,6 +2896,7 @@ can execute code specifically intended for profile generate or profile use.
For example, a user program can have special logging during profile generate:
.. code-block:: c
+
#if __LLVM_INSTR_PROFILE_GENERATE
expensive_logging_of_full_program_state();
#endif
>From 6f9a443065df12193947d9b39014996bd3e76be7 Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Thu, 28 Dec 2023 22:28:17 -0500
Subject: [PATCH 10/11] Adding missing line break in documentation.
---
clang/docs/UsersManual.rst | 1 +
1 file changed, 1 insertion(+)
diff --git a/clang/docs/UsersManual.rst b/clang/docs/UsersManual.rst
index 2612a029165389..615883840d0685 100644
--- a/clang/docs/UsersManual.rst
+++ b/clang/docs/UsersManual.rst
@@ -2811,6 +2811,7 @@ indexed format, regardeless whether it is produced by frontend or the IR pass.
Fine Tuning Profile Collection
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
The PGO infrastructure provides user program knobs to fine tune profile
collection. Specifically, the PGO runtime provides the following functions
that can be used to control the regions in the program where profiles should
>From ae9d57de9504731f8522901fd8888a06def75716 Mon Sep 17 00:00:00 2001
From: Qiongsi Wu <qwu at ibm.com>
Date: Tue, 2 Jan 2024 17:44:59 -0500
Subject: [PATCH 11/11] Address code review comments.
---
.../include/profile/instr_prof_interface.h | 4 +--
.../profile/Linux/instrprof-weak-symbol.c | 2 ++
compiler-rt/test/profile/instrprof-api.c | 26 ++++++++++++-------
3 files changed, 21 insertions(+), 11 deletions(-)
diff --git a/compiler-rt/include/profile/instr_prof_interface.h b/compiler-rt/include/profile/instr_prof_interface.h
index 06f31210d73bea..be40f2685934be 100644
--- a/compiler-rt/include/profile/instr_prof_interface.h
+++ b/compiler-rt/include/profile/instr_prof_interface.h
@@ -67,8 +67,8 @@ void __llvm_profile_reset_counters(void);
* __llvm_profile_dump();
*
* It is expected that on-line profile merging is on with \c %m specifier
- * used in profile filename . If merging is not turned on, user is expected
- * to invoke __llvm_profile_set_filename to specify different profile names
+ * used in profile filename . If merging is not turned on, user is expected
+ * to invoke __llvm_profile_set_filename to specify different profile names
* for different regions before dumping to avoid profile write clobbering.
*/
int __llvm_profile_dump(void);
diff --git a/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c b/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c
index 4af69fecb34f63..eda299cb6610e4 100644
--- a/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c
+++ b/compiler-rt/test/profile/Linux/instrprof-weak-symbol.c
@@ -1,3 +1,5 @@
+// Test the linker feature that treats undefined weak symbols as null values.
+
// RUN: %clang_pgogen -o %t %s
// RUN: not %t
// RUN: %clang -o %t %s
diff --git a/compiler-rt/test/profile/instrprof-api.c b/compiler-rt/test/profile/instrprof-api.c
index 8b235f7b5e674c..680c3f8577c7d4 100644
--- a/compiler-rt/test/profile/instrprof-api.c
+++ b/compiler-rt/test/profile/instrprof-api.c
@@ -1,10 +1,18 @@
-// RUN: %clang_profgen %s --target=ppc64le-unknown-linux-gnu -S \
-// RUN: -emit-llvm -o - | FileCheck %s --check-prefix=PROFGEN
+// Testing profile generate.
+// RUN: %clang_profgen %s -S -emit-llvm -o - | FileCheck %s --check-prefix=PROFGEN
+// RUN: %clang_pgogen %s -S -emit-llvm -o - | FileCheck %s --check-prefix=PROFGEN
+
+// Testing profile use. Generate some profile file first.
+// RUN: rm -rf rawprof.profraw
// RUN: %clang_profgen -o %t %s
+// RUN: %run %t
+// RUN: llvm-profdata merge -o %t1.profdata rawprof.profraw
+// RUN: %clang_profuse=%t.profdata %s -S -emit-llvm -o - | FileCheck %s --check-prefix=PROFUSE
+// RUN: rm -rf rawprof.profraw
+// RUN: %clang_pgogen -o %t %s
// RUN: env LLVM_PROFILE_FILE=%t.profraw %run %t
-// RUN: llvm-profdata merge -o %t.profdata %t.profraw
-// RUN: %clang_profuse=%t.profdata %s --target=ppc64le-unknown-linux-gnu -S \
-// RUN: -emit-llvm -o - | FileCheck %s --check-prefix=PROFUSE
+// RUN: llvm-profdata merge -o %t2.profdata rawprof.profraw
+// RUN: %clang_pgouse=%t.profdata %s -S -emit-llvm -o - | FileCheck %s --check-prefix=PROFUSE
#include "profile/instr_prof_interface.h"
__attribute__((noinline)) int bar() { return 4; }
@@ -25,11 +33,11 @@ int main() {
// PROFUSE-NOT: call void @__llvm_profile_set_filename(ptr noundef @.str)
if (__llvm_profile_dump())
return 2;
- // PROFGEN: %call1 = call signext i32 @__llvm_profile_dump()
- // PROFUSE-NOT: %call1 = call signext i32 @__llvm_profile_dump()
+ // PROFGEN: %call1 = call {{(signext )*}}i32 @__llvm_profile_dump()
+ // PROFUSE-NOT: %call1 = call {{(signext )*}}i32 @__llvm_profile_dump()
__llvm_orderfile_dump();
- // PROFGEN: %call2 = call signext i32 @__llvm_orderfile_dump()
- // PROFUSE-NOT: %call2 = call signext i32 @__llvm_orderfile_dump()
+ // PROFGEN: %call2 = call {{(signext )*}}i32 @__llvm_orderfile_dump()
+ // PROFUSE-NOT: %call2 = call {{(signext )*}}i32 @__llvm_orderfile_dump()
return z + bar() - 11;
}
More information about the cfe-commits
mailing list