[Bug middle-end/110004] New: large increase in profile mismatches on tramp3d

[Bug middle-end/110004] New: large increase in profile mismatches on tramp3d

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110004

            Bug ID: 110004
           Summary: large increase in profile mismatches on tramp3d
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hubicka at gcc dot gnu.org
  Target Milestone: ---

https://lnt.opensuse.org/db_default/v4/CPP/latest_runs_report

it is between g:c5300bf3110b44e2742b36f49c2a380abd08d9c5 and
g:d6b756447cd58bcca20e6892790582308b869817

[Bug middle-end/110004] large increase in profile mismatches on tramp3d

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110004

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-*-*
            Version|unknown                     |14.0
           Keywords|                            |needs-bisection

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
isn't that just from 1 mismatch to 2 for some RTL passes?!

[Bug middle-end/110004] large increase in profile mismatches on tramp3d

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110004

--- Comment #2 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Linkes the statistics for outedges (that probabilities sums to 100).
Here are statistic for inedges.

We track mismatches after alignments RTL pass since it is last one, so it
cummulates all the mismatches through the pipeline. Martin was objecting agains
tracing all the passes independently.

https://lnt.opensuse.org/db_default/v4/CPP/graph?highlight_run=42409&plot.0=353.964.7

[Bug middle-end/110004] large increase in profile mismatches on tramp3d

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110004

--- Comment #3 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
It seems that fixing (part of) the jump threading profile mismatches has
improved stats a lot. Still there are some more visible regressions that can be
understood.  I looked into range for first and did not see obvious candidate.

2022-05-04 g:d7de15fd93fc654e to 2022-05-06 g:37f57a3f4e48fe1a
2022-07-02 g:ed974488991256c5 to 2022-07-06 g:d1f3a3fffeb40e64
2022-09-28 g:5d7be27bf7ef63c7 to 2022-09-30 g:43faf3e5445b5717