Skip to content

Commit d99baac

Browse files
committed
gh-xxxx: overhaul build rules for optimized binaries
This commit overhauls the make-based build system's rules for building optimized binaries. Along the way it fixes a myriad of bugs and shortcomings with the prior approach. The old way of producing optimized binaries had various limitations: * `make [all]` would do work when PGO was enabled because the phony `profile-opt` rule was non-empty. This prevented no-op PGO builds from working at all. This meant workflows like `make; make install` either incurred extra work or failed due to race conditions. * Same thing for BOLT, as its `bolt-opt` rule was also non-empty and always ran during `make [all]`. * BOLT could not be run multiple times without a full rebuild because `llvm-bolt` can't instrument binaries that have already received BOLT optimizations. * It was difficult to run BOLT on its own because of how various make targets and their dependencies were structured. * I found the old way that configure and make communicated the default targets to be confusing and hard to understand. There are essentially 2 major changes going on in this commit: 1. A rework of the high-level make targets for performing a build and how they are defined. 2. A rework of all the make logic related to profile-based optimization (read: PGO and BOLT). Build Target Rework ====================== Before, we essentially had `build_all`, `profile-opt`, `bolt-opt` and `build_wasm` as our 3 targets for performing a build. `all` would alias to one of these, as appropriate. And there was another definition for which _simple_ make target to evaluate for non-optimized builds. This was likely `build_all` or `all`. In the rework, we introduce 2 new high-level targets: * `build-plain` - Perform a build without optimizations. * `build-optimized` - Perform a build with optimizations. `build-plain` is aliased to `build_all` in all configurations except WASM, where it is `build_wasm`. `build-optimized` by default is aliased to a target that prints an error message when optimizations aren't enabled. If PGO or BOLT are enabled, it is aliased to their respective target. `build-optimized` is the logical successor to `profile-opt`. I felt it best to delete `profile-opt` completely, as the new `build-*` high-level targets feel more friendly to use. But if people lament its loss, we can add a `profile-opt: build-optimized` to achieve almost the same result. Profiled-Based Optimization Rework ================================== Most of the make logic related to profile-based optimization (read: PGO and BOLT) has been touched in this change. A major issue with the old way of doing things was we used phony, always-executed make rules. This is a bad practice in make because it undermines no-op builds. Another issue is that the separation between the rules and what order they ran in wasn't always clear. Both PGO and BOLT consist of the same 4 phase solution: instrument, run, analyze, and apply. However, these steps weren't clearly expressed in the make logic. This is especially true for BOLT, which only had 1 make rule. Another issue with BOLT is that it was really easy to get things into a bad state. e.g. if you applied BOLT to `pythonX.Y` you could not run BOLT again unless you rebuilt `pythonX.Y` from source. In the new world, we have separate `profile-<tool>-<stage>-stamp` rules defining the 4 distinct `instrument`, `run`, `analyze`, and `apply` stages for both PGO and BOLT. Each of these stages is tracked by a _stamp_ semaphore file so progress can be captured. This should all be pretty straightforward. There is some minimal complexity here to handle BOLT's optional dependency on PGO, as BOLT either depends on `build_all` or `profile-pgo-apply-stamp`. As part of the refactor to BOLT we also preserve the original input binary before BOLT is applied. This original file is restored if BOLT runs again. This greatly simplifies repeated BOLT invocations, as make doesn't perform needless work. However, this is all best effort, as it is possible for some make target evaluations to still get things in a bad state. Other Remarks ============= If this change perturbs any bugs, they are likely around cleaning behavior. The cleaning rules are a bit complicated and not clearly documented. And I'm unsure which targets CPython developers often iterate on. It is highly possible that state cleanup of PGO and/or BOLT files isn't as robust as it needs to be. I explicitly deleted some calls to PGO cleanup because those calls prevented no-op `make [all]` from working. It is certainly possible something somewhere (release automation?) relied on these files being deleted when they no longer are. We still have targets to purge profile files and it should be trivial to add these to appropriate make rules.
1 parent 90ec98a commit d99baac

File tree

4 files changed

+204
-95
lines changed

4 files changed

+204
-95
lines changed

.gitignore

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -122,8 +122,7 @@ Tools/unicode/data/
122122
# hendrikmuhs/ccache-action@v1
123123
/.ccache
124124
/platform
125-
/profile-clean-stamp
126-
/profile-run-stamp
125+
/profile-*-stamp
127126
/Python/deepfreeze/*.c
128127
/pybuilddir.txt
129128
/pyconfig.h

Makefile.pre.in

Lines changed: 127 additions & 51 deletions
Original file line numberDiff line numberDiff line change
@@ -601,12 +601,26 @@ LIBHACL_SHA2_HEADERS= \
601601
#########################################################################
602602
# Rules
603603

604-
# Default target
605-
all: @DEF_MAKE_ALL_RULE@
604+
# Default target.
605+
# Likely either `build-plain` or `build-optimized`.
606+
all: @MAKE_TARGET_ALL@
606607
# First target in Makefile is implicit default. So .PHONY needs to come after
607608
# all.
608609
.PHONY: all
609610

611+
# Build without any optimizations or instrumented binaries.
612+
.PHONY: build-plain
613+
build-plain: @MAKE_TARGET_BUILD_PLAIN@
614+
615+
# Build with optimizations (PGO, BOLT, etc).
616+
.PHONY: build-optimized
617+
build-optimized: @MAKE_TARGET_BUILD_OPTIMIZED@
618+
619+
.PHONY: build-optimized-not-enabled
620+
build-optimized-not-enabled:
621+
@echo "build-optimized requires --enable-optimizations in configure; aborting"
622+
@exit 1
623+
610624
.PHONY: build_all
611625
build_all: check-clean-src $(BUILDPYTHON) platform sharedmods \
612626
gdbhooks Programs/_testembed scripts checksharedmods rundsymutil
@@ -627,69 +641,145 @@ check-clean-src:
627641
exit 1; \
628642
fi
629643

630-
# Profile generation build must start from a clean tree.
644+
# Profile-based optimization.
645+
#
646+
# PGO and BOLT profile-based optimization is supported. For each optimization,
647+
# roughly the following steps are done:
648+
#
649+
# 1. "Instrument" binaries with run-time data collection (e.g. build or modify
650+
# a variant of the binary.)
651+
# 2. "Run" instrumented binaries (via subset of test suite) to collect data.
652+
# 3. "Analyze" / collect / merge data files from previous step.
653+
# 4. "Apply" collected data from above. (e.g. rebuild or modify a binary).
654+
#
655+
# 0, 1, or multiple profile based optimizations can be enabled.
656+
#
657+
# We track the progress of profile-based optimization using various "stamp"
658+
# files. An empty stamp file tracks the stage of optimization we're in.
659+
# Each *-stamp rule that follows is defined in execution / dependency order.
660+
661+
# Remove files produced by or used for tracking profile-guided optimization.
662+
.PHONY: profile-remove
663+
profile-remove: clean-bolt
664+
find . -name '*.gc??' -exec rm -f {} ';'
665+
find . -name '*.profclang?' -exec rm -f {} ';'
666+
find . -name '*.dyn' -exec rm -f {} ';'
667+
rm -f $(COVERAGE_INFO)
668+
rm -rf $(COVERAGE_REPORT)
669+
# Remove all progress tracking stamps to ensure a clean slate.
670+
rm -f profile-*-stamp
671+
672+
# Profile-based optimization requires a fresh build environment.
631673
profile-clean-stamp:
632-
$(MAKE) clean
674+
$(MAKE) clean profile-remove
633675
touch $@
634676

635-
# Compile with profile generation enabled.
636-
profile-gen-stamp: profile-clean-stamp
677+
# Build with PGO instrumentation enabled.
678+
profile-pgo-instrument-stamp: profile-clean-stamp
637679
@if [ $(LLVM_PROF_ERR) = yes ]; then \
638680
echo "Error: Cannot perform PGO build because llvm-profdata was not found in PATH" ;\
639681
echo "Please add it to PATH and run ./configure again" ;\
640682
exit 1;\
641683
fi
642684
@echo "Building with support for profile generation:"
643-
$(MAKE) @DEF_MAKE_RULE@ CFLAGS_NODIST="$(CFLAGS_NODIST) $(PGO_PROF_GEN_FLAG)" LDFLAGS_NODIST="$(LDFLAGS_NODIST) $(PGO_PROF_GEN_FLAG)" LIBS="$(LIBS)"
685+
$(MAKE) @MAKE_TARGET_BUILD_PLAIN@ CFLAGS_NODIST="$(CFLAGS_NODIST) $(PGO_PROF_GEN_FLAG)" LDFLAGS_NODIST="$(LDFLAGS_NODIST) $(PGO_PROF_GEN_FLAG)" LIBS="$(LIBS)"
644686
touch $@
645687

646-
# Run task with profile generation build to create profile information.
647-
profile-run-stamp:
688+
# Run PGO instrumented binaries and collect profile data.
689+
profile-pgo-run-stamp: profile-pgo-instrument-stamp
648690
@echo "Running code to generate profile data (this can take a while):"
649-
# First, we need to create a clean build with profile generation
650-
# enabled.
651-
$(MAKE) profile-gen-stamp
652-
# Next, run the profile task to generate the profile information.
653691
@ # FIXME: can't run for a cross build
654692
$(LLVM_PROF_FILE) $(RUNSHARED) ./$(BUILDPYTHON) $(PROFILE_TASK) || true
693+
touch $@
694+
695+
# Collect data files produced by running PGO instrumented binaries.
696+
profile-pgo-analyze-stamp: profile-pgo-run-stamp
655697
$(LLVM_PROF_MERGER)
656698
# Remove profile generation binary since we are done with it.
657699
$(MAKE) clean-retain-profile
658-
# This is an expensive target to build and it does not have proper
659-
# makefile dependency information. So, we create a "stamp" file
660-
# to record its completion and avoid re-running it.
661700
touch $@
662701

663-
# Compile Python binary with profile guided optimization.
664-
# To force re-running of the profile task, remove the profile-run-stamp file.
665-
.PHONY: profile-opt
666-
profile-opt: profile-run-stamp
702+
# Use collected PGO data to influence rebuild of binaries.
703+
profile-pgo-apply-stamp: profile-pgo-analyze-stamp
667704
@echo "Rebuilding with profile guided optimizations:"
668-
-rm -f profile-clean-stamp
669-
$(MAKE) @DEF_MAKE_RULE@ CFLAGS_NODIST="$(CFLAGS_NODIST) $(PGO_PROF_USE_FLAG)" LDFLAGS_NODIST="$(LDFLAGS_NODIST)"
670-
671-
.PHONY: bolt-opt
672-
bolt-opt: @PREBOLT_RULE@
673-
rm -f *.fdata
674-
@if $(READELF) -p .note.bolt_info $(BUILDPYTHON) | grep BOLT > /dev/null; then\
675-
echo "skip: $(BUILDPYTHON) is already BOLTed."; \
676-
else \
677-
@LLVM_BOLT@ ./$(BUILDPYTHON) -instrument -instrumentation-file-append-pid -instrumentation-file=$(abspath $(BUILDPYTHON).bolt) -o $(BUILDPYTHON).bolt_inst; \
678-
./$(BUILDPYTHON).bolt_inst $(PROFILE_TASK) || true; \
679-
@MERGE_FDATA@ $(BUILDPYTHON).*.fdata > $(BUILDPYTHON).fdata; \
680-
@LLVM_BOLT@ ./$(BUILDPYTHON) -o $(BUILDPYTHON).bolt -data=$(BUILDPYTHON).fdata -update-debug-sections -reorder-blocks=ext-tsp -reorder-functions=hfsort+ -split-functions -icf=1 -inline-all -split-eh -reorder-functions-use-hot-size -peepholes=none -jump-tables=aggressive -inline-ap -indirect-call-promotion=all -dyno-stats -use-gnu-stack -frame-opt=hot; \
681-
rm -f *.fdata; \
682-
rm -f $(BUILDPYTHON).bolt_inst; \
683-
mv $(BUILDPYTHON).bolt $(BUILDPYTHON); \
705+
# Need to purge PGO instrumented build to force a rebuild.
706+
$(MAKE) clean-retain-profile
707+
$(MAKE) @MAKE_TARGET_BUILD_PLAIN@ CFLAGS_NODIST="$(CFLAGS_NODIST) $(PGO_PROF_USE_FLAG)" LDFLAGS_NODIST="$(LDFLAGS_NODIST)"
708+
touch $@
709+
710+
# BOLT supports instrumenting and applying changes to standalone binaries
711+
# without having to recompile.
712+
#
713+
# BOLT can run independently or in addition to PGO. If running with PGO,
714+
# it always runs after PGO. Care needs to be taken to preserve PGO state
715+
# when running BOLT so make doesn't re-apply PGO.
716+
#
717+
# BOLT also can't instrument binaries that have already had BOLT applied
718+
# to them. So we make an attempt to preserve and re-use the pristine
719+
# pre-BOLT binaries so developers can iterate on just BOLT optimization
720+
# passes.
721+
722+
# List of binaries that BOLT runs on.
723+
BOLT_BINARIES = $(BUILDPYTHON)
724+
725+
# Remove traces of bolt.
726+
.PHONY: clean-bolt
727+
clean-bolt:
728+
# Instrumented binaries.
729+
find . -name '*.bolt_inst' -exec rm -f {} ';'
730+
# The data files they produce.
731+
find . -name '*.fdata' -exec rm -f {} ';'
732+
# Copied of binaries before BOLT application.
733+
find . -name '*.prebolt' -exec rm -f {} ';'
734+
735+
# BOLTs dependencies are a bit wonky.
736+
#
737+
# If PGO is enabled, we can take a native rule dependency on a stamp file.
738+
# If PGO isn't enabled, we don't have a stamp to key off of and the phony
739+
# target (e.g. build_all) will always force rebuilds. So we call out to
740+
# make externally to sidestep the dependency.
741+
#
742+
# We can simplify this hack if we ever get stamp files for plain builds.
743+
profile-bolt-prebuild-stamp: @MAKE_BOLT_NATIVE_DEPENDENCY@
744+
if [ -n "@MAKE_BOLT_MAKE_DEPENDENCY@" ]; then \
745+
$(MAKE) @MAKE_BOLT_MAKE_DEPENDENCY@; \
684746
fi
747+
touch $@
685748

749+
profile-bolt-instrument-stamp: profile-bolt-prebuild-stamp
750+
for bin in $(BOLT_BINARIES); do \
751+
if [ -e "$${bin}.prebolt" ]; then \
752+
echo "Restoring pre-BOLT binary $${bin}.prebolt"; \
753+
mv "$${bin}.prebolt" "$${bin}"; \
754+
fi \
755+
done
756+
# Ensure prior BOLT state is purged.
757+
$(MAKE) clean-bolt
758+
@LLVM_BOLT@ ./$(BUILDPYTHON) -instrument -instrumentation-file-append-pid -instrumentation-file=$(abspath $(BUILDPYTHON).bolt) -o $(BUILDPYTHON).bolt_inst
759+
touch $@
760+
761+
profile-bolt-run-stamp: profile-bolt-instrument-stamp
762+
./$(BUILDPYTHON).bolt_inst $(PROFILE_TASK) || true
763+
touch $@
764+
765+
profile-bolt-analyze-stamp: profile-bolt-run-stamp
766+
@MERGE_FDATA@ $(BUILDPYTHON).*.fdata > $(BUILDPYTHON).fdata
767+
touch $@
768+
769+
profile-bolt-apply-stamp: profile-bolt-analyze-stamp
770+
@LLVM_BOLT@ ./$(BUILDPYTHON) -o $(BUILDPYTHON).bolt -data=$(BUILDPYTHON).fdata -update-debug-sections -reorder-blocks=ext-tsp -reorder-functions=hfsort+ -split-functions=3 -icf=1 -inline-all -split-eh -reorder-functions-use-hot-size -peepholes=all -jump-tables=aggressive -inline-ap -indirect-call-promotion=all -dyno-stats -use-gnu-stack -frame-opt=hot
771+
mv $(BUILDPYTHON) $(BUILDPYTHON).prebolt
772+
mv $(BUILDPYTHON).bolt $(BUILDPYTHON)
773+
touch $@
774+
775+
# End of profile-based optimization rules.
686776

687777
# Compile and run with gcov
688778
.PHONY: coverage
689779
coverage:
690780
@echo "Building with support for coverage checking:"
691781
$(MAKE) clean
692-
$(MAKE) @DEF_MAKE_RULE@ CFLAGS="$(CFLAGS) -O0 -pg --coverage" LDFLAGS="$(LDFLAGS) --coverage"
782+
$(MAKE) @MAKE_SIMPLE_BUILD_TARGET@ CFLAGS="$(CFLAGS) -O0 -pg --coverage" LDFLAGS="$(LDFLAGS) --coverage"
693783

694784
.PHONY: coverage-lcov
695785
coverage-lcov:
@@ -2608,23 +2698,9 @@ clean-retain-profile: pycremoval
26082698
-rm -f Python/frozen_modules/MANIFEST
26092699
-find build -type f -a ! -name '*.gc??' -exec rm -f {} ';'
26102700
-rm -f Include/pydtrace_probes.h
2611-
-rm -f profile-gen-stamp
2612-
2613-
.PHONY: profile-removal
2614-
profile-removal:
2615-
find . -name '*.gc??' -exec rm -f {} ';'
2616-
find . -name '*.profclang?' -exec rm -f {} ';'
2617-
find . -name '*.dyn' -exec rm -f {} ';'
2618-
rm -f $(COVERAGE_INFO)
2619-
rm -rf $(COVERAGE_REPORT)
2620-
rm -f profile-run-stamp
26212701

26222702
.PHONY: clean
26232703
clean: clean-retain-profile
2624-
@if test @DEF_MAKE_ALL_RULE@ = profile-opt; then \
2625-
rm -f profile-gen-stamp profile-clean-stamp; \
2626-
$(MAKE) profile-removal; \
2627-
fi
26282704

26292705
.PHONY: clobber
26302706
clobber: clean

configure

Lines changed: 39 additions & 21 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

0 commit comments

Comments
 (0)