On 09/03/26 00:28, Andi Kleen wrote:
External email: Use caution opening links or attachments
On 2026-03-08 10:28, Jan Hubicka wrote:
From: Dhruv Chawla <[email protected]>
This is a partial fix for PR124075 which forces perf record to write
the
profile out to perf.data. This is required because I noticed on both
aarch64 and x86 that it was writing out the profile to stdout at
times,
which would cause profile information to be dropped. This did not fail
in the various create_fdas_for_* targets because they would only try
and
access the perf profiles if they existed at their paths.
Further work for this patch is to plumb the PERF_DATA make variable
into
the perf record invocation, but that is a stage 1 thing.
Autoprofilebootstrapped and regtested on aarch64-linux-gnu.
Signed-off-by: Dhruv Chawla <[email protected]>
gcc/ChangeLog:
PR gcov-profile/124075
* config/aarch64/gcc-auto-profile: Add "-o perf.data" to perf
record invocation.
* config/i386/gcc-auto-profile: Likewise.
I have similar change in my tree for some time, so i can do
profiledbootstraps without getting perf data accidentlly streamed into
redirected stdout/stderr. I do it in the Makefile.ac instead of
perf.data.
Again I would like Andy to have chance to comment, but the change looks
OK to me.
I don't think I've ever seen that issue. You get the perf.data binary
blob in the output? It shouldn't have anything to do with redirection or
not.
-o perf.data is default.
the only way I could think of this happening is when the passed command
has an -o /dev/stdout or -o /dev/stderr and maybe perf misparses it
somehow
(but normally it stops option parsing at the first command name), but
then
normally perf uses the last option as the winner, so you should still
see the same output.
If that's the case the correct fix would be to add a "--" before the
"$@" to
separate the perf options from others.
Hi Andi,
I had tried this initially as well, however there are additional arguments
passed to perf record that "--" breaks:
Makefile.in: AUTO_PROFILE = gcc-auto-profile --all -c 10000000
This then leads to perf record failing with "--all: command not found" or
something similar - I don't remember exactly.
If you want to change the output it would probably be better to pass in
a name with
the PID in it (perf.data.$$) from the caller, this would allow multiple
profiles
steps in the same directory eventually (but I think the profile process
only runs
once per directory atm)
Yeah, that makes sense. I do plan on integrating the PERF_DATA Makefile
variable as an argument to the script in the next stage 1, so this can be
done at that point.
-Andi
--
Regards,
Dhruv
