On 12/1/19 11:45 AM, Jan Hubicka wrote:
Hi.
I'm sending v3 of the patch where I changed:
- function.cold sections are properly put into .text.unlikely and
not into a .text.sorted.XYZ section
I've just finished measurements and I still have the original speed up
for tramp3d:
Total runs: 10, before: 13.92, after: 13.82, cmp: 99.219%
Hi,
I have updated binutils to current head on the Firefox testing patch and
run FDO build with tp first run ordering and call chain clustering.
https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=1313e6a4d74ebff702afa7594684beb83c01d95f&newProject=try&newRevision=1c2d53b10b042aaaac15edbe7bd26e2740641840&framework=1
Hello.
Thank you for the testing.
It seems there are no differences in performance. The two binaries can
be downloaded at
w/o patch:
https://firefox-ci-tc.services.mozilla.com/api/queue/v1/task/MK-7DC3FQcevZC_Nvlnq8Q/runs/0/artifacts/public/build/target.tar.bz2
with call chain clustering.
https://firefox-ci-tc.services.mozilla.com/api/queue/v1/task/UVh6iNILT-qb8sYM5vxVCQ/runs/0/artifacts/public/build/target.tar.bz2
I would ideally need output of -fdump-ipa-reorder which prints sorted symbols
and
so that I can compare it with resulting assembly.
Since Firefox is quite sensitive to code size I would expect to be able
to measure some benefits here. Any idea what may have go wrong?
That's a pity.
I checked that the binaries seems generally sane - out of 58MB text
segment there is 34MB cold section. It is possible that system ld is
used instead of provided one, but that would be weird. I will try to
find way to double-check that updating binutils really updated them for
GCC.
I can double check once having the dump file.
Martin
Honza
