gh-101282: rework the BOLT build process

(This change is a quick and dirty way to merge some of the build system
improvements I'm proposing in gh-101093 before the 3.12 feature freeze.
I wanted to scope bloat myself to fix some longstanding deficiencies in
the build system around profile-guided builds. But I'm getting soft
resistance to the reviews so close to the freeze deadline and it is
obvious that we need a simpler solution to hit the 3.12 deadline. While
this change is quick and dirty, it attempts to not make things worse.)

Before this change, we only applied bolt to the main python binary.
After this change, we apply bolt to libpython if it is configured. In
shared library builds, most of the C code is in libpython so it is
critical to apply bolt to libpython to realize bolt benefits.

This change also reworks how bolt instrumentation is applied. It
effectively removes the readelf based logic added in gh-101525 and
replaces it with a mechanism that saves a copy of the pre-bolt binary
and restores that copy when necessary. This allows us to perform
bolt optimizations without having to manually delete the output binary
to force a new bolt run.

We also add a new make target for purging bolt files and hook it up
to `clean` so bolt state is purged when appropriate.

`.gitignore` rules have been added to ignore files related to bolt.

Before and after this refactor, `make` will no-op after a previous run.
Both versions should also share common make DAG deficiencies where
targets fail to trigger as often as they need to or can trigger
prematurely in certain scenarios. e.g. after this change you may need
to `rm profile-bolt-stamp` to force a bolt run because there aren't
appropriate non-phony targets for bolt's make target to depend on.
Fixing this is a non-trivial amount of work that will likely have to
wait until the 3.13 window.

To make it easier to iterate on custom BOLT settings, the flags to
pass to instrumentation and application are now defined in configure
and can be overridden by passing `BOLT_INSTRUMENT_FLAGS` and
`BOLT_APPLY_FLAGS`.

Author: indygreg

.gitignore

@@ -23,6 +23,10 @@
 *.gc??
 *.profclang?
 *.profraw
+# Copies of binaries before BOLT optimizations.
+*.prebolt
+# BOLT profile data.
+*.fdata
 *.dyn
 .gdb_history
 .purify
@@ -124,6 +128,7 @@ Tools/unicode/data/
 /platform
 /profile-clean-stamp
 /profile-run-stamp
+/profile-bolt-stamp
 /Python/deepfreeze/*.c
 /pybuilddir.txt
 /pyconfig.h

Doc/using/configure.rst

@@ -313,7 +313,13 @@ also be used to improve performance.
    experimental for now. Because this tool operates on machine code its success
    is dependent on a combination of the build environment + the other
    optimization configure args + the CPU architecture, and not all combinations
-   are supported.
+   are supported. BOLT versions before LLVM 16 are known to crash BOLT under
+   some scenarios. Use of LLVM 16 or newer for BOLT optimization is stronger
+   encouraged.
+
+   The ``BOLT_INSTRUMENT_FLAGS`` and ``BOLT_APPLY_FLAGS`` configure variables
+   can be defined to override the default set of arguments for ``llvm-bolt``
+   to instrument and apply BOLT data to binaries, respectively.
 
    .. versionadded:: 3.12
 

Makefile.pre.in

@@ -672,21 +672,55 @@ profile-opt: profile-run-stamp
 	-rm -f profile-clean-stamp
 	$(MAKE) @DEF_MAKE_RULE@ CFLAGS_NODIST="$(CFLAGS_NODIST) $(PGO_PROF_USE_FLAG)" LDFLAGS_NODIST="$(LDFLAGS_NODIST)"
 
-.PHONY: bolt-opt
-bolt-opt: @PREBOLT_RULE@
+# List of binaries that BOLT runs on.
+BOLT_BINARIES := @BOLT_BINARIES@
+
+BOLT_INSTRUMENT_FLAGS := @BOLT_INSTRUMENT_FLAGS@
+BOLT_APPLY_FLAGS := @BOLT_APPLY_FLAGS@
+
+.PHONY: clean-bolt
+clean-bolt:
+	# Profile data.
 	rm -f *.fdata
-	@if $(READELF) -p .note.bolt_info $(BUILDPYTHON) | grep BOLT > /dev/null; then\
-		echo "skip: $(BUILDPYTHON) is already BOLTed."; \
-	else \
-		@LLVM_BOLT@ ./$(BUILDPYTHON) -instrument -instrumentation-file-append-pid -instrumentation-file=$(abspath $(BUILDPYTHON).bolt) -o $(BUILDPYTHON).bolt_inst; \
-		./$(BUILDPYTHON).bolt_inst $(PROFILE_TASK) || true; \
-		@MERGE_FDATA@ $(BUILDPYTHON).*.fdata > $(BUILDPYTHON).fdata; \
-		@LLVM_BOLT@ ./$(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; \
-		rm -f *.fdata; \
-		rm -f $(BUILDPYTHON).bolt_inst; \
-		mv $(BUILDPYTHON).bolt $(BUILDPYTHON); \
-	fi
+	# Pristine binaries before BOLT optimization.
+	rm -f *.prebolt
+	# BOLT instrumented binaries.
+	rm -f *.bolt_inst
+
+profile-bolt-stamp: $(BUILDPYTHON)
+	# Ensure a pristine, pre-BOLT copy of the binary and no profile data from last run.
+	for bin in $(BOLT_BINARIES); do \
+	  prebolt="$${bin}.prebolt"; \
+	  if [ -e "$${prebolt}" ]; then \
+	    echo "Restoring pre-BOLT binary $${prebolt}"; \
+	    mv "$${bin}.prebolt" "$${bin}"; \
+	  fi; \
+	  cp "$${bin}" "$${prebolt}"; \
+	  rm -f $${bin}.bolt.*.fdata $${bin}.fdata; \
+	done
+	# Instrument each binary.
+	for bin in $(BOLT_BINARIES); do \
+	  @LLVM_BOLT@ "$${bin}" -instrument -instrumentation-file-append-pid -instrumentation-file=$(abspath $${bin}.bolt) -o $${bin}.bolt_inst $(BOLT_INSTRUMENT_FLAGS); \
+	  mv "$${bin}.bolt_inst" "$${bin}"; \
+	done
+	# Run instrumented binaries to collect data.
+	$(RUNSHARED) ./$(BUILDPYTHON) $(PROFILE_TASK) || true
+	# Merge all the data files together.
+	for bin in $(BOLT_BINARIES); do \
+	  @MERGE_FDATA@ $${bin}.*.fdata > "$${bin}.fdata"; \
+	  rm -f $${bin}.*.fdata; \
+	done
+	# Run bolt against the merged data to produce an optimized binary.
+	for bin in $(BOLT_BINARIES); do \
+	  @LLVM_BOLT@ "$${bin}.prebolt" -o "$${bin}.bolt" -data="$${bin}.fdata" $(BOLT_APPLY_FLAGS); \
+	  mv "$${bin}.bolt" "$${bin}"; \
+	done
+	touch $@
 
+.PHONY: bolt-opt
+bolt-opt:
+	$(MAKE) @PREBOLT_RULE@
+	$(MAKE) profile-bolt-stamp
 
 # Compile and run with gcov
 .PHONY: coverage
@@ -2623,10 +2657,11 @@ profile-removal:
 	rm -f $(COVERAGE_INFO)
 	rm -rf $(COVERAGE_REPORT)
 	rm -f profile-run-stamp
+	rm -f profile-bolt-stamp
 
 .PHONY: clean
-clean: clean-retain-profile
-	@if test @DEF_MAKE_ALL_RULE@ = profile-opt; then \
+clean: clean-retain-profile clean-bolt
+	@if test @DEF_MAKE_ALL_RULE@ = profile-opt -o @DEF_MAKE_ALL_RULE@ = bolt-opt; then \
 		rm -f profile-gen-stamp profile-clean-stamp; \
 		$(MAKE) profile-removal; \
 	fi

Misc/NEWS.d/next/Build/2023-05-20-16-09-59.gh-issue-101282.FvRARb.rst

@@ -0,0 +1,4 @@
+BOLT optimization is now applied to the libpython shared library if building
+a shared library. BOLT instrumentation and application settings can now be
+influenced via the ``BOLT_INSTRUMENT_FLAGS`` and ``BOLT_APPLY_FLAGS``
+configure variables.

configure

@@ -2028,13 +2028,6 @@ if test "$Py_BOLT" = 'true' ; then
   DEF_MAKE_ALL_RULE="bolt-opt"
   DEF_MAKE_RULE="build_all"
 
-  AC_SUBST(READELF)
-  AC_CHECK_TOOLS(READELF, [readelf], "notfound")
-  if test "$READELF" == "notfound"
-  then
-    AC_MSG_ERROR([readelf is required for a --enable-bolt build but could not be found.])
-  fi
-
   # -fno-reorder-blocks-and-partition is required for bolt to work.
   # Possibly GCC only.
   AX_CHECK_COMPILE_FLAG([-fno-reorder-blocks-and-partition],[
@@ -2067,6 +2060,30 @@ if test "$Py_BOLT" = 'true' ; then
   fi
 fi
 
+# Enable BOLT of libpython if built.
+AC_SUBST(BOLT_BINARIES)
+BOLT_BINARIES='$(BUILDPYTHON)'
+if test "${enable_shared}" = "yes"
+then
+  BOLT_BINARIES="${BOLT_BINARIES} \$(INSTSONAME)"
+fi
+
+AC_ARG_VAR(BOLT_INSTRUMENT_FLAGS, Arguments to llvm-bolt when instrumenting binaries)
+AC_MSG_CHECKING(BOLT_INSTRUMENT_FLAGS)
+if test -z "${BOLT_INSTRUMENT_FLAGS}"
+then
+  BOLT_INSTRUMENT_FLAGS=
+fi
+AC_MSG_RESULT($BOLT_INSTRUMENT_FLAGS)
+
+AC_ARG_VAR(BOLT_APPLY_FLAGS, Arguments to llvm-bolt when creating a BOLT optimized binary)
+AC_MSG_CHECKING(BOLT_APPLY_FLAGS)
+if test -z "${BOLT_APPLY_FLAGS}"
+then
+  BOLT_APPLY_FLAGS="-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"
+fi
+AC_MSG_RESULT($BOLT_APPLY_FLAGS)
+
 # XXX Shouldn't the code above that fiddles with BASECFLAGS and OPT be
 # merged with this chunk of code?