https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116743
--- Comment #22 from Eugene Rozenfeld ---
Rama, thank you for confirming this fixes your perf regression. I'll be on
vacation for the rest of December but will get this completed and ported after
New Year.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116743
--- Comment #19 from Eugene Rozenfeld ---
I investigated what happens in the compiler.
In afdo_annotate_cfg we have these lines:
cgraph_node::get (current_function_decl)->count
= profile_count::from_gcov_type (s->head_count ()).afdo ();
ENT
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116743
--- Comment #18 from Eugene Rozenfeld ---
Hi Rama.
I started looking at this. Below is what the info looks like for
ha_innobase::build_template in the gcov file you shared. If you look near the
end, you can see counts for calls to build_templat
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116743
--- Comment #17 from Eugene Rozenfeld ---
Rama, just a note that I have this on my list and will look into this as soon
as I can.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116743
--- Comment #16 from Eugene Rozenfeld ---
Sorry for the delay, Rama. I should have time to look into this by the end of
the week.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116743
--- Comment #13 from Eugene Rozenfeld ---
Hi Rama. Sorry for the delay. I hope I'll have time to review your repro in the
next couple of weeks.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116743
--- Comment #4 from Eugene Rozenfeld ---
AutoFDO does work. I made a number of fixes and improvements over the last
several years, both in GCC (including fixing autoprofiledbootstrap) and in
google/autofdo that's used to build create_gcov. The c
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113765
--- Comment #7 from Eugene Rozenfeld ---
I'll prepare a patch this week.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113765
--- Comment #4 from Eugene Rozenfeld ---
auto-profile currently doesn't guarantee that it will set probabilities on all
edges because of zero basic block counts. Normally those edges just have
probabilities set by the preceding profile_estimate
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108000
Eugene Rozenfeld changed:
What|Removed |Added
Status|ASSIGNED|RESOLVED
Resolution|---
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108000
--- Comment #3 from Eugene Rozenfeld ---
Repro steps:
gcc -O2 -g -o ./pr108000 ./pr108000.c
perf record -e cpu/event=0xC4,umask=0x20/puk -b -m8 -o perf.data ./pr108000
create_gcov --binary ./pr108000 --profile=./perf.data -gcov_version=2
--gc
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108000
--- Comment #2 from Eugene Rozenfeld ---
I actually have a fix for this, just wanted to file a simple test case. I
initially ran into this failure while reviving autoprofiledbootstrap build. The
failure happens while using the profile and the bu
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108000
Bug ID: 108000
Summary: Assert during ipa-cp with AutoFDO
Product: gcc
Version: 13.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: ipa
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107197
--- Comment #9 from Eugene Rozenfeld ---
107193 was fixed by
https://gcc.gnu.org/g:80f414e6d73f9f1683f93d83ce63a6a482e54bee
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107193
--- Comment #8 from Eugene Rozenfeld ---
Created attachment 53690
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53690&action=edit
Proposed patch
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107193
--- Comment #7 from Eugene Rozenfeld ---
No, locus won't be changed by the loop. But the purpose of the loop is to
change statement locations (by adding discriminators) in this line:
gimple_set_location (stmt, dloc);
I think the code would m
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107193
--- Comment #4 from Eugene Rozenfeld ---
Yes, that's the problem. Sorry about that, will send a patch with the fix
shortly.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107169
--- Comment #2 from Eugene Rozenfeld ---
When -gstatement-frontiers is on, the IR coming from the front end may be
different with and without debug information turned on. That may cause e.g.,
different discriminator values and -fcompare-debug fa
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96674
--- Comment #9 from Eugene Rozenfeld ---
I used XXX_MIN for consistency with comments on other patterns. If TYPE_MIN is
preferable, the change should be made in all of those comments as well.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97223
--- Comment #4 from Eugene Rozenfeld ---
The commit that fixed this:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=32ee472864ada44ef05b2a3b087b8ce413bee282
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96674
--- Comment #7 from Eugene Rozenfeld ---
Thank you for the feedback, Gabriel and Jakub. I re-worked the patch based on
your suggestions. I attached the new patch and also sent it to gcc-patches.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96674
Eugene Rozenfeld changed:
What|Removed |Added
Attachment #49940|0 |1
is obsolete|
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96674
--- Comment #3 from Eugene Rozenfeld ---
Both are used but it looks like __attribute__((noinline)) is used more
frequently.
Under gcc/testsuite there are 1537 instances of __attribute__((noipa)) and 3794
instances of __attribute__((noinline)).
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96674
Eugene Rozenfeld changed:
What|Removed |Added
CC||erozen at microsoft dot com
--- Comme
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86698
Eugene Rozenfeld changed:
What|Removed |Added
CC||erozen at microsoft dot com
--- Comme
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97223
--- Comment #3 from Eugene Rozenfeld ---
The proposed patch was committed to the trunk by Jeff Law.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96701
Eugene Rozenfeld changed:
What|Removed |Added
CC||erozen at microsoft dot com
--- Comme
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97223
Eugene Rozenfeld changed:
What|Removed |Added
CC||erozen at microsoft dot com
--- Comme
28 matches
Mail list logo
