On 09/03/26 22:48, Andi Kleen wrote:
External email: Use caution opening links or attachments
On Fri, Mar 06, 2026 at 01:23:27PM +0530, [email protected] wrote:
for component_in_prev in "$(components_in_prev)"; do \
perf_path=../prev-$$component_in_prev/$(PERF_DATA); \
- echo "Perf path:"; \
- echo $$perf_path; \
+ echo "Perf path:" $$perf_path "..." `[ -f $$perf_path ] && echo exists
|| echo does not exist`; \
if [ -f $$perf_path ]; then \
profile_name=cc1_$$component_in_prev.fda; \
- $(CREATE_GCOV) -binary ../stage1-gcc/cc1$(exeext) -gcov
$$profile_name -profile $$perf_path -gcov_version 3 || exit 1; \
+ if [ "x`perf buildid-list -i $$perf_path | grep '/cc1$$' && echo yes`" !=
"x" ]; then \
The open coded perfs should be also using AFDO_PROFILER
Or really just $(PERF)
Yeah, couldn't figure out a good way to have one consistent
perf invocation across the files I modified. Will try with $(PERF).
diff --git a/gcc/configure.ac b/gcc/configure.ac
index d12382d2fa3..04503800423 100644
--- a/gcc/configure.ac
+++ b/gcc/configure.ac
@@ -7894,6 +7894,21 @@ gif=`if test x$enable_x86_64_mfentry = xyes; then echo
1; else echo 0; fi`
AC_DEFINE_UNQUOTED(ENABLE_X86_64_MFENTRY, $gif,
[Define to enable -mfentry by default on x86-64.])
+AFDO_PROFILER=perf
+case "$cpu_type" in
+ aarch64)
+ AC_MSG_CHECKING(AArch64 build supports SPE profiling for
autoprofiledbootstrap)
+ aarch64_spe_support=no
+ if test -x /usr/bin/perf && test -n "$(/usr/bin/perf list | grep
arm_spe)"; then
+ AFDO_PROFILER=perf_spe
But I'm actually surprised about this, I thought the standard perf
supported spe. Are you sure most users have a perf_spe?
I am not entirely sure what you mean. perf_spe is an argument being passed
to create_gcov (--profiler=$(AFDO_PROFILER)). I guess the variable is
confusingly named, though I couldn't come up with anything better.
The rest looks good to me.
-Andi
--
Regards,
Dhruv
