https://github.com/tstellar updated
https://github.com/llvm/llvm-project/pull/119896
>From dd1f6807904691586c715d447fff54915a46c751 Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Tue, 3 Dec 2024 22:32:46 +0000
Subject: [PATCH 01/10] [clang][cmake] Apply bolt optimizations as part of the
clang target
This change removes the need to call the clang-bolt target in order
to apply bolt optimizations to clang. Now running `ninja clang` will
build a clang with bolt optimizations, and `ninja check-clang` and
`ninja install-clang` will test and install bolt optimized clang too.
The clang-bolt target has been kept for compatibilty reasons, but it
is now just an alias to the clang target.
Also, this new design for applying the bolt optimizations to
clang will be easier to generalize and use to optimize other
binaries/libraries in the project.
---
clang/CMakeLists.txt | 52 ------------------
clang/tools/driver/CMakeLists.txt | 52 ++++++++++++++++++
clang/utils/perf-training/CMakeLists.txt | 26 +--------
clang/utils/perf-training/perf-helper.py | 69 ++++++++++++++++++++++++
4 files changed, 122 insertions(+), 77 deletions(-)
diff --git a/clang/CMakeLists.txt b/clang/CMakeLists.txt
index 27e8095534a65c..b7741ce49b46ca 100644
--- a/clang/CMakeLists.txt
+++ b/clang/CMakeLists.txt
@@ -873,58 +873,6 @@ if (CLANG_ENABLE_BOOTSTRAP)
endforeach()
endif()
-set(CLANG_BOLT OFF CACHE STRING "Apply BOLT optimization to Clang. \
- May be specified as Instrument or Perf or LBR to use a particular profiling \
- mechanism.")
-string(TOUPPER "${CLANG_BOLT}" CLANG_BOLT)
-
-if (CLANG_BOLT AND NOT LLVM_BUILD_INSTRUMENTED)
- set(CLANG_PATH ${LLVM_RUNTIME_OUTPUT_INTDIR}/clang)
- set(CLANG_INSTRUMENTED
${LLVM_RUNTIME_OUTPUT_INTDIR}/${CLANG_BOLT_INSTRUMENTED})
- set(BOLT_FDATA ${CMAKE_CURRENT_BINARY_DIR}/utils/perf-training/prof.fdata)
-
- # Pass extra flag in no-LBR mode
- if (CLANG_BOLT STREQUAL "PERF")
- set(BOLT_NO_LBR "-nl")
- endif()
-
- if (CLANG_BOLT STREQUAL "INSTRUMENT")
- # Instrument clang with BOLT
- add_custom_target(clang-instrumented
- DEPENDS ${CLANG_INSTRUMENTED}
- )
- add_custom_command(OUTPUT ${CLANG_INSTRUMENTED}
- DEPENDS clang llvm-bolt
- COMMAND llvm-bolt ${CLANG_PATH} -o ${CLANG_INSTRUMENTED}
- -instrument --instrumentation-file-append-pid
- --instrumentation-file=${BOLT_FDATA}
- COMMENT "Instrumenting clang binary with BOLT"
- USES_TERMINAL
- VERBATIM
- )
- add_custom_target(clang-bolt-training-deps DEPENDS clang-instrumented)
- else() # perf or LBR
- add_custom_target(clang-bolt-training-deps DEPENDS clang)
- endif()
-
- # Optimize original (pre-bolt) Clang using the collected profile
- add_custom_target(clang-bolt
- DEPENDS clang-bolt-profile
- COMMAND ${CMAKE_COMMAND} -E rename $<TARGET_FILE:clang>
${CLANG_PATH}-prebolt
- COMMAND ${CMAKE_COMMAND} -E create_symlink ${CLANG_PATH}-prebolt
${CLANG_PATH}++-prebolt
- COMMAND llvm-bolt ${CLANG_PATH}-prebolt
- -o $<TARGET_FILE:clang>
- -data ${BOLT_FDATA}
- -reorder-blocks=ext-tsp -reorder-functions=cdsort -split-functions
- -split-all-cold -split-eh -dyno-stats -use-gnu-stack
- -update-debug-sections
- ${BOLT_NO_LBR}
- COMMENT "Optimizing Clang with BOLT"
- USES_TERMINAL
- VERBATIM
- )
-endif()
-
if (LLVM_ADD_NATIVE_VISUALIZERS_TO_SOLUTION)
add_subdirectory(utils/ClangVisualizers)
endif()
diff --git a/clang/tools/driver/CMakeLists.txt
b/clang/tools/driver/CMakeLists.txt
index a4debc2dd2e895..ad336fcc45b600 100644
--- a/clang/tools/driver/CMakeLists.txt
+++ b/clang/tools/driver/CMakeLists.txt
@@ -23,6 +23,18 @@ if(CLANG_PLUGIN_SUPPORT)
set(support_plugins SUPPORT_PLUGINS)
endif()
+set(CLANG_BOLT OFF CACHE STRING "Apply BOLT optimization to Clang. \
+ May be specified as Instrument or Perf or LBR to use a particular profiling \
+ mechanism.")
+string(TOUPPER "${CLANG_BOLT}" CLANG_BOLT)
+
+if (CLANG_BOLT AND NOT LLVM_BUILD_INSTRUMENTED)
+ set(CLANG_BOLT_DEPS clear-bolt-fdata llvm-bolt llvm-readobj)
+ if (NOT CLANG_BOLT STREQUAL "INSTRUMENT")
+ list(APPEND CLANG_BOLT_DEPS clear-perf-data)
+ endif()
+endif()
+
add_clang_tool(clang
driver.cpp
cc1_main.cpp
@@ -35,6 +47,7 @@ add_clang_tool(clang
ARMTargetParserTableGen
AArch64TargetParserTableGen
${support_plugins}
+ ${CLANG_BOLT_DEPS}
GENERATE_DRIVER
)
@@ -134,3 +147,42 @@ if(CLANG_ORDER_FILE AND
set_target_properties(clang PROPERTIES LINK_DEPENDS ${CLANG_ORDER_FILE})
endif()
endif()
+
+if (CLANG_BOLT AND NOT LLVM_BUILD_INSTRUMENTED)
+ # Add a clang-bolt target for backwards compatibility.
+ add_custom_target(clang-bolt DEPENDS clang)
+
+ set(CLANG_BOLT_INSTRUMENTED "clang-bolt.inst" CACHE STRING
+ "Name of BOLT-instrumented Clang binary")
+ set(CLANG_INSTRUMENTED
${LLVM_RUNTIME_OUTPUT_INTDIR}/${CLANG_BOLT_INSTRUMENTED})
+ set(PERF_TRAINING_BINARY_DIR
${CMAKE_CURRENT_BINARY_DIR}/../../utils/perf-training)
+ set(BOLT_FDATA ${PERF_TRAINING_BINARY_DIR}/prof.fdata)
+ get_llvm_lit_path(
+ lit_base_dir
+ lit_file_name
+ ALLOW_EXTERNAL
+ )
+ set(LIT_COMMAND "${lit_base_dir}/${lit_file_name}")
+
+ # This POST_BUILD command is executed unconditionally even if the clang
target
+ # is already built. We need to wrap the whole bolt optimization process in
+ # a single python wrapper, so that we can first check if the binary has
+ # already been optimized and then exit early with a 0 status if it has.
+ add_custom_command(
+ TARGET clang POST_BUILD
+ COMMAND "${Python3_EXECUTABLE}"
${CMAKE_CURRENT_SOURCE_DIR}/../../utils/perf-training/perf-helper.py
+ bolt-optimize
+ --method ${CLANG_BOLT}
+ --input $<TARGET_FILE:clang>
+ --instrumented-output ${CLANG_INSTRUMENTED}
+ --fdata ${BOLT_FDATA}
+ --perf-training-binary-dir ${PERF_TRAINING_BINARY_DIR}
+ --readelf $<TARGET_FILE:llvm-readobj>
+ --bolt $<TARGET_FILE:llvm-bolt>
+ --lit "${LIT_COMMAND}"
+ --merge-fdata $<TARGET_FILE:merge-fdata>
+ COMMENT "Optimizing Clang with BOLT"
+ USES_TERMINAL
+ VERBATIM
+ )
+endif()
diff --git a/clang/utils/perf-training/CMakeLists.txt
b/clang/utils/perf-training/CMakeLists.txt
index 49673790ff6e84..4aed086563ee92 100644
--- a/clang/utils/perf-training/CMakeLists.txt
+++ b/clang/utils/perf-training/CMakeLists.txt
@@ -83,8 +83,6 @@ if(APPLE AND DTRACE AND NOT LLVM_TOOL_LLVM_DRIVER_BUILD)
endif()
if(CLANG_BOLT AND NOT LLVM_BUILD_INSTRUMENTED)
- set(CLANG_BOLT_INSTRUMENTED "clang-bolt.inst" CACHE STRING
- "Name of BOLT-instrumented Clang binary")
configure_lit_site_cfg(
${CMAKE_CURRENT_SOURCE_DIR}/bolt.lit.site.cfg.in
${CMAKE_CURRENT_BINARY_DIR}/bolt-fdata/lit.site.cfg
@@ -93,7 +91,7 @@ if(CLANG_BOLT AND NOT LLVM_BUILD_INSTRUMENTED)
add_lit_testsuite(generate-bolt-fdata "Generating BOLT profile for Clang"
${CMAKE_CURRENT_BINARY_DIR}/bolt-fdata/
EXCLUDE_FROM_CHECK_ALL
- DEPENDS clang-bolt-training-deps clear-bolt-fdata clear-perf-data
+ DEPENDS clear-bolt-fdata clear-perf-data
)
add_custom_target(clear-bolt-fdata
@@ -104,26 +102,4 @@ if(CLANG_BOLT AND NOT LLVM_BUILD_INSTRUMENTED)
COMMAND "${Python3_EXECUTABLE}" ${CMAKE_CURRENT_SOURCE_DIR}/perf-helper.py
clean ${CMAKE_CURRENT_BINARY_DIR} perf.data
COMMENT "Clearing old perf data")
- string(TOUPPER "${CLANG_BOLT}" CLANG_BOLT)
- if (CLANG_BOLT STREQUAL "LBR")
- set(BOLT_LBR "--lbr")
- endif()
-
- add_custom_target(merge-fdata-deps)
- if (CLANG_BOLT STREQUAL "INSTRUMENT")
- add_dependencies(merge-fdata-deps generate-bolt-fdata)
- else()
- # Convert perf profiles into fdata
- add_custom_target(convert-perf-fdata
- COMMAND "${Python3_EXECUTABLE}"
${CMAKE_CURRENT_SOURCE_DIR}/perf-helper.py perf2bolt $<TARGET_FILE:llvm-bolt>
${CMAKE_CURRENT_BINARY_DIR} $<TARGET_FILE:clang> ${BOLT_LBR}
- COMMENT "Converting perf files to BOLT fdata"
- DEPENDS llvm-bolt generate-bolt-fdata)
- add_dependencies(merge-fdata-deps convert-perf-fdata)
- endif()
-
- # Merge profiles into one using merge-fdata
- add_custom_target(clang-bolt-profile
- COMMAND "${Python3_EXECUTABLE}" ${CMAKE_CURRENT_SOURCE_DIR}/perf-helper.py
merge-fdata $<TARGET_FILE:merge-fdata> ${CMAKE_CURRENT_BINARY_DIR}/prof.fdata
${CMAKE_CURRENT_BINARY_DIR}
- COMMENT "Merging BOLT fdata"
- DEPENDS merge-fdata merge-fdata-deps)
endif()
diff --git a/clang/utils/perf-training/perf-helper.py
b/clang/utils/perf-training/perf-helper.py
index d76c6ede3fe5a2..098c5973096122 100644
--- a/clang/utils/perf-training/perf-helper.py
+++ b/clang/utils/perf-training/perf-helper.py
@@ -16,6 +16,8 @@
import bisect
import shlex
import tempfile
+import re
+import shutil
test_env = {"PATH": os.environ["PATH"]}
@@ -558,7 +560,74 @@ def genOrderFile(args):
return 0
+def bolt_optimize(args):
+ parser = argparse.ArgumentParser("%prog [options] ")
+ parser.add_argument('--method')
+ parser.add_argument('--input')
+ parser.add_argument('--instrumented-output')
+ parser.add_argument('--fdata')
+ parser.add_argument('--perf-training-binary-dir')
+ parser.add_argument('--readelf')
+ parser.add_argument('--bolt')
+ parser.add_argument('--lit')
+ parser.add_argument('--merge-fdata')
+
+ opts = parser.parse_args(args)
+
+ readelf = opts.readelf
+ bolt = opts.bolt
+ lit = opts.lit
+
+ output = subprocess.check_output(
+ [readelf, "-WS", opts.input], universal_newlines=True
+ )
+
+ # This binary has already been bolt-optimized, so skip further processing.
+ if re.search('\\.bolt\\.org\\.text', output, re.MULTILINE):
+ return 0
+
+ if opts.method == 'INSTRUMENT':
+ process = subprocess.run(
+ [bolt, opts.input, '-o', opts.instrumented_output, '-instrument',
'--instrumentation-file-append-pid',
+ f'--instrumentation-file={opts.fdata}'],
+ stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text = True)
+
+ print(process.args)
+ for line in process.stdout:
+ sys.stdout.write(line)
+ process.check_returncode()
+
+ process = subprocess.run(
+ [sys.executable, lit, os.path.join(opts.perf_training_binary_dir,
'bolt-fdata')],
+ stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text = True)
+
+ print(process.args)
+ for line in process.stdout:
+ sys.stdout.write(line)
+ process.check_returncode()
+
+ if opts.method == 'PERF':
+ perf2bolt([bolt, opts.perf_training_binary_dir, opts.input])
+
+ merge_fdata([opts.merge_fdata, opts.fdata, opts.perf_training_binary_dir])
+
+ shutil.copy(opts.input, f'{opts.input}-prebolt')
+
+ process = subprocess.run(
+ [bolt, f'{opts.input}-prebolt', '-o', opts.input, '-data', opts.fdata,
+ '-reorder-blocks=ext-tsp', '-reorder-functions=cdsort',
'-split-functions',
+ '-split-all-cold', '-split-eh', '-dyno-stats', '-use-gnu-stack',
'-update-debug-sections',
+ '-nl' if opts.method == 'PERF' else ''],
+ stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text = True)
+
+ print(process.args)
+ for line in process.stdout:
+ sys.stdout.write(line)
+ process.check_returncode()
+
+
commands = {
+ "bolt-optimize" : bolt_optimize,
"clean": clean,
"merge": merge,
"dtrace": dtrace,
>From dab4b8ddb49fcdc7c91467fc256516e1d216659f Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Tue, 17 Dec 2024 16:30:16 -0800
Subject: [PATCH 02/10] Fix formatting
---
clang/utils/perf-training/perf-helper.py | 88 ++++++++++++++++--------
1 file changed, 60 insertions(+), 28 deletions(-)
diff --git a/clang/utils/perf-training/perf-helper.py
b/clang/utils/perf-training/perf-helper.py
index 098c5973096122..e628907fdc85d6 100644
--- a/clang/utils/perf-training/perf-helper.py
+++ b/clang/utils/perf-training/perf-helper.py
@@ -562,16 +562,16 @@ def genOrderFile(args):
def bolt_optimize(args):
parser = argparse.ArgumentParser("%prog [options] ")
- parser.add_argument('--method')
- parser.add_argument('--input')
- parser.add_argument('--instrumented-output')
- parser.add_argument('--fdata')
- parser.add_argument('--perf-training-binary-dir')
- parser.add_argument('--readelf')
- parser.add_argument('--bolt')
- parser.add_argument('--lit')
- parser.add_argument('--merge-fdata')
-
+ parser.add_argument("--method")
+ parser.add_argument("--input")
+ parser.add_argument("--instrumented-output")
+ parser.add_argument("--fdata")
+ parser.add_argument("--perf-training-binary-dir")
+ parser.add_argument("--readelf")
+ parser.add_argument("--bolt")
+ parser.add_argument("--lit")
+ parser.add_argument("--merge-fdata")
+
opts = parser.parse_args(args)
readelf = opts.readelf
@@ -581,16 +581,25 @@ def bolt_optimize(args):
output = subprocess.check_output(
[readelf, "-WS", opts.input], universal_newlines=True
)
-
+
# This binary has already been bolt-optimized, so skip further processing.
- if re.search('\\.bolt\\.org\\.text', output, re.MULTILINE):
+ if re.search("\\.bolt\\.org\\.text", output, re.MULTILINE):
return 0
- if opts.method == 'INSTRUMENT':
+ if opts.method == "INSTRUMENT":
process = subprocess.run(
- [bolt, opts.input, '-o', opts.instrumented_output, '-instrument',
'--instrumentation-file-append-pid',
- f'--instrumentation-file={opts.fdata}'],
- stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text = True)
+ [
+ bolt,
+ opts.input, "-o",
+ opts.instrumented_output,
+ "-instrument",
+ "--instrumentation-file-append-pid",
+ f'--instrumentation-file={opts.fdata}'
+ ],
+ stdout=subprocess.PIPE,
+ stderr=subprocess.STDOUT,
+ text=True
+ )
print(process.args)
for line in process.stdout:
@@ -598,36 +607,59 @@ def bolt_optimize(args):
process.check_returncode()
process = subprocess.run(
- [sys.executable, lit, os.path.join(opts.perf_training_binary_dir,
'bolt-fdata')],
- stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text = True)
+ [
+ sys.executable,
+ lit,
+ os.path.join(opts.perf_training_binary_dir, "bolt-fdata")
+ ],
+ stdout=subprocess.PIPE,
+ stderr=subprocess.STDOUT,
+ text=True
+ )
print(process.args)
for line in process.stdout:
sys.stdout.write(line)
process.check_returncode()
- if opts.method == 'PERF':
+ if opts.method == "PERF":
perf2bolt([bolt, opts.perf_training_binary_dir, opts.input])
merge_fdata([opts.merge_fdata, opts.fdata, opts.perf_training_binary_dir])
-
- shutil.copy(opts.input, f'{opts.input}-prebolt')
+
+ shutil.copy(opts.input, f"{opts.input}-prebolt")
process = subprocess.run(
- [bolt, f'{opts.input}-prebolt', '-o', opts.input, '-data', opts.fdata,
- '-reorder-blocks=ext-tsp', '-reorder-functions=cdsort',
'-split-functions',
- '-split-all-cold', '-split-eh', '-dyno-stats', '-use-gnu-stack',
'-update-debug-sections',
- '-nl' if opts.method == 'PERF' else ''],
- stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text = True)
+ [
+ bolt,
+ f"{opts.input}-prebolt",
+ "-o",
+ opts.input,
+ "-data",
+ opts.fdata,
+ "-reorder-blocks=ext-tsp",
+ "-reorder-functions=cdsort",
+ "-split-functions",
+ "-split-all-cold",
+ "-split-eh",
+ "-dyno-stats",
+ "-use-gnu-stack",
+ "-update-debug-sections",
+ "-nl" if opts.method == "PERF" else ""
+ ],
+ stdout=subprocess.PIPE,
+ stderr=subprocess.STDOUT,
+ text=True
+ )
print(process.args)
for line in process.stdout:
sys.stdout.write(line)
process.check_returncode()
-
+
commands = {
- "bolt-optimize" : bolt_optimize,
+ "bolt-optimize": bolt_optimize,
"clean": clean,
"merge": merge,
"dtrace": dtrace,
>From f2bfbdee31953fa2f50568f87c19555e2feddab9 Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Tue, 17 Dec 2024 16:40:10 -0800
Subject: [PATCH 03/10] Fix formatting
---
clang/utils/perf-training/perf-helper.py | 15 ++++++++-------
1 file changed, 8 insertions(+), 7 deletions(-)
diff --git a/clang/utils/perf-training/perf-helper.py
b/clang/utils/perf-training/perf-helper.py
index e628907fdc85d6..1949ab75a21825 100644
--- a/clang/utils/perf-training/perf-helper.py
+++ b/clang/utils/perf-training/perf-helper.py
@@ -590,15 +590,16 @@ def bolt_optimize(args):
process = subprocess.run(
[
bolt,
- opts.input, "-o",
+ opts.input,
+ "-o",
opts.instrumented_output,
"-instrument",
"--instrumentation-file-append-pid",
- f'--instrumentation-file={opts.fdata}'
+ f'--instrumentation-file={opts.fdata}',
],
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
- text=True
+ text=True,
)
print(process.args)
@@ -610,7 +611,7 @@ def bolt_optimize(args):
[
sys.executable,
lit,
- os.path.join(opts.perf_training_binary_dir, "bolt-fdata")
+ os.path.join(opts.perf_training_binary_dir, "bolt-fdata"),
],
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
@@ -631,7 +632,7 @@ def bolt_optimize(args):
process = subprocess.run(
[
- bolt,
+ bolt,
f"{opts.input}-prebolt",
"-o",
opts.input,
@@ -645,11 +646,11 @@ def bolt_optimize(args):
"-dyno-stats",
"-use-gnu-stack",
"-update-debug-sections",
- "-nl" if opts.method == "PERF" else ""
+ "-nl" if opts.method == "PERF" else "",
],
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
- text=True
+ text=True,
)
print(process.args)
>From 7a421453ed7b62a806b1a5aa0845f9001e584c23 Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Tue, 17 Dec 2024 16:45:29 -0800
Subject: [PATCH 04/10] Fix formatting
---
clang/utils/perf-training/perf-helper.py | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/clang/utils/perf-training/perf-helper.py
b/clang/utils/perf-training/perf-helper.py
index 1949ab75a21825..1e099fcea90b41 100644
--- a/clang/utils/perf-training/perf-helper.py
+++ b/clang/utils/perf-training/perf-helper.py
@@ -595,7 +595,7 @@ def bolt_optimize(args):
opts.instrumented_output,
"-instrument",
"--instrumentation-file-append-pid",
- f'--instrumentation-file={opts.fdata}',
+ f"--instrumentation-file={opts.fdata}",
],
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
@@ -615,7 +615,7 @@ def bolt_optimize(args):
],
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
- text=True
+ text=True,
)
print(process.args)
>From f0bb473f753a2776380bd2f38a137901ed189069 Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Wed, 18 Dec 2024 17:43:23 -0800
Subject: [PATCH 05/10] Update clang/utils/perf-training/perf-helper.py
Co-authored-by: Amir Ayupov <fad...@gmail.com>
---
clang/utils/perf-training/perf-helper.py | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/utils/perf-training/perf-helper.py
b/clang/utils/perf-training/perf-helper.py
index 1e099fcea90b41..42c78f91642685 100644
--- a/clang/utils/perf-training/perf-helper.py
+++ b/clang/utils/perf-training/perf-helper.py
@@ -623,7 +623,7 @@ def bolt_optimize(args):
sys.stdout.write(line)
process.check_returncode()
- if opts.method == "PERF":
+ if opts.method in ["PERF", "LBR"]:
perf2bolt([bolt, opts.perf_training_binary_dir, opts.input])
merge_fdata([opts.merge_fdata, opts.fdata, opts.perf_training_binary_dir])
>From 543dffe5e19657808138b245fa76205228331bca Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Thu, 19 Dec 2024 06:34:10 -0800
Subject: [PATCH 06/10] Update clang/tools/driver/CMakeLists.txt
Co-authored-by: Petr Hosek <pho...@google.com>
---
clang/tools/driver/CMakeLists.txt | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/tools/driver/CMakeLists.txt
b/clang/tools/driver/CMakeLists.txt
index ad336fcc45b600..e3e6c1bbfad1ee 100644
--- a/clang/tools/driver/CMakeLists.txt
+++ b/clang/tools/driver/CMakeLists.txt
@@ -177,7 +177,7 @@ if (CLANG_BOLT AND NOT LLVM_BUILD_INSTRUMENTED)
--instrumented-output ${CLANG_INSTRUMENTED}
--fdata ${BOLT_FDATA}
--perf-training-binary-dir ${PERF_TRAINING_BINARY_DIR}
- --readelf $<TARGET_FILE:llvm-readobj>
+ --readelf $<TARGET_FILE:llvm-readelf>
--bolt $<TARGET_FILE:llvm-bolt>
--lit "${LIT_COMMAND}"
--merge-fdata $<TARGET_FILE:merge-fdata>
>From b85039dec9b532abe59a466eb345786e045e25e5 Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Thu, 19 Dec 2024 06:34:25 -0800
Subject: [PATCH 07/10] Update clang/utils/perf-training/perf-helper.py
Co-authored-by: Petr Hosek <pho...@google.com>
---
clang/utils/perf-training/perf-helper.py | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/utils/perf-training/perf-helper.py
b/clang/utils/perf-training/perf-helper.py
index 42c78f91642685..e16475f2317823 100644
--- a/clang/utils/perf-training/perf-helper.py
+++ b/clang/utils/perf-training/perf-helper.py
@@ -562,7 +562,7 @@ def genOrderFile(args):
def bolt_optimize(args):
parser = argparse.ArgumentParser("%prog [options] ")
- parser.add_argument("--method")
+ parser.add_argument("--method", choices=["INSTRUMENT", "PERF", "LBR"])
parser.add_argument("--input")
parser.add_argument("--instrumented-output")
parser.add_argument("--fdata")
>From 7a8c6914fa82a92b5c4e44140b35f42539e8a586 Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Thu, 19 Dec 2024 14:37:09 +0000
Subject: [PATCH 08/10] Inline some variables
---
clang/utils/perf-training/perf-helper.py | 12 ++++--------
1 file changed, 4 insertions(+), 8 deletions(-)
diff --git a/clang/utils/perf-training/perf-helper.py
b/clang/utils/perf-training/perf-helper.py
index e16475f2317823..6c795bdfc53c02 100644
--- a/clang/utils/perf-training/perf-helper.py
+++ b/clang/utils/perf-training/perf-helper.py
@@ -574,12 +574,8 @@ def bolt_optimize(args):
opts = parser.parse_args(args)
- readelf = opts.readelf
- bolt = opts.bolt
- lit = opts.lit
-
output = subprocess.check_output(
- [readelf, "-WS", opts.input], universal_newlines=True
+ [opts.readelf, "-WS", opts.input], universal_newlines=True
)
# This binary has already been bolt-optimized, so skip further processing.
@@ -589,7 +585,7 @@ def bolt_optimize(args):
if opts.method == "INSTRUMENT":
process = subprocess.run(
[
- bolt,
+ opts.bolt,
opts.input,
"-o",
opts.instrumented_output,
@@ -610,7 +606,7 @@ def bolt_optimize(args):
process = subprocess.run(
[
sys.executable,
- lit,
+ opts.lit,
os.path.join(opts.perf_training_binary_dir, "bolt-fdata"),
],
stdout=subprocess.PIPE,
@@ -624,7 +620,7 @@ def bolt_optimize(args):
process.check_returncode()
if opts.method in ["PERF", "LBR"]:
- perf2bolt([bolt, opts.perf_training_binary_dir, opts.input])
+ perf2bolt([opts.bolt, opts.perf_training_binary_dir, opts.input])
merge_fdata([opts.merge_fdata, opts.fdata, opts.perf_training_binary_dir])
>From 9abb676f748766eb58a81b6740ad4ee3b0e2fea3 Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Thu, 19 Dec 2024 14:39:21 +0000
Subject: [PATCH 09/10] Revert "Update clang/tools/driver/CMakeLists.txt"
This reverts commit 543dffe5e19657808138b245fa76205228331bca.
---
clang/tools/driver/CMakeLists.txt | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/tools/driver/CMakeLists.txt
b/clang/tools/driver/CMakeLists.txt
index e3e6c1bbfad1ee..ad336fcc45b600 100644
--- a/clang/tools/driver/CMakeLists.txt
+++ b/clang/tools/driver/CMakeLists.txt
@@ -177,7 +177,7 @@ if (CLANG_BOLT AND NOT LLVM_BUILD_INSTRUMENTED)
--instrumented-output ${CLANG_INSTRUMENTED}
--fdata ${BOLT_FDATA}
--perf-training-binary-dir ${PERF_TRAINING_BINARY_DIR}
- --readelf $<TARGET_FILE:llvm-readelf>
+ --readelf $<TARGET_FILE:llvm-readobj>
--bolt $<TARGET_FILE:llvm-bolt>
--lit "${LIT_COMMAND}"
--merge-fdata $<TARGET_FILE:merge-fdata>
>From bd670071651c04d4db98b994e57758191f3e484f Mon Sep 17 00:00:00 2001
From: Tom Stellard <tstel...@redhat.com>
Date: Thu, 19 Dec 2024 19:10:23 +0000
Subject: [PATCH 10/10] Inline another variable
---
clang/utils/perf-training/perf-helper.py | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/utils/perf-training/perf-helper.py
b/clang/utils/perf-training/perf-helper.py
index 6c795bdfc53c02..55c5160a71c4ff 100644
--- a/clang/utils/perf-training/perf-helper.py
+++ b/clang/utils/perf-training/perf-helper.py
@@ -628,7 +628,7 @@ def bolt_optimize(args):
process = subprocess.run(
[
- bolt,
+ opts.bolt,
f"{opts.input}-prebolt",
"-o",
opts.input,
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
