[Bug c/114468] New: unixbench compilation performance results vary with different gcc compilers

[Bug c/114468] New: unixbench compilation performance results vary with different gcc compilers

[jiangchuanganghw at outlook dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114468

            Bug ID: 114468
           Summary: unixbench compilation performance results vary with
                    different gcc compilers
           Product: gcc
           Version: 10.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jiangchuanganghw at outlook dot com
  Target Milestone: ---

We used Unixbench to compare gcc 10.3.0 with gcc8.5 on an intel Sapphire Rapids
CPU and found that the performance of the dhry2reg test project compiled with
gcc 10.3.0 deteriorated by about 10%. We found that this commit affected
performance, and after rolling back the changes in that patch, performance was
restored.
https://github.com/gcc-mirror/gcc/commit/3133bed5d0327e8a9cd0a601b7ecdb9de4fc825d

We expect to find the cause of this result.Looking forward to your help.

[Bug target/114468] unixbench compilation performance results vary with different gcc compilers

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114468

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Gcc 10 is no longer supported. Can you test a newer gcc? Register allocation is
always improving between versions so this might be already fixed.

[Bug target/114468] unixbench compilation performance results vary with different gcc compilers

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114468

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-03-25

[Bug target/114468] unixbench compilation performance results vary with different gcc compilers (r10-6804-g3133bed5d0327e)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114468

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=26944

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Also one should note that register allocation is NP complete problem. So
obvious tuning heurstics for one case will most likely cause issues for others.

Also dhrystone benchmark is very much not a decent benchmark for real code.