[Bug tree-optimization/102809] New: Some TSVC benchmarks are slower after r12-4195-gec0124e0acb556cd

[Bug tree-optimization/102809] New: Some TSVC benchmarks are slower after r12-4195-gec0124e0acb556cd

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102809

            Bug ID: 102809
           Summary: Some TSVC benchmarks are slower after
                    r12-4195-gec0124e0acb556cd
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: marxin at gcc dot gnu.org
                CC: aldyh at gcc dot gnu.org
  Target Milestone: ---

Tested with https://github.com/UoB-HPC/TSVC_2/pull/2 where one can run a single
benchmark.

Happens with -Ofast -march=native on znver1 machine:

gcc-bisect.py 'gcc *.c -lm -Ofast -march=native -fuse-ld=bfd && timeout 100.7
./a.out va' -l -e r12-4194-g1f51e9af7b615838 -s r12-4195-gec0124e0acb556cd

Bisecting latest revisions
  ec0124e0acb556cd(05 Oct 2021 18:24)(aldyh@redhat.com): [took: 7.050s] result:
OK
Loop    Time(sec)       Checksum
   va        3.719      1.644725
  1f51e9af7b615838(05 Oct 2021 16:47)(jwakely@redhat.com): [took: 5.433s]
result: OK
Loop    Time(sec)       Checksum
   va        2.006      1.644725

and can be seen here:
https://lnt.opensuse.org/db_default/v4/CPP/graph?plot.0=14.920.0
https://lnt.opensuse.org/db_default/v4/CPP/graph?plot.0=14.887.0
https://lnt.opensuse.org/db_default/v4/CPP/graph?plot.0=14.943.0

[Bug tree-optimization/102809] Some TSVC benchmarks are slower after r12-4195-gec0124e0acb556cd

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102809

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-10-18
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

[Bug tree-optimization/102809] Some TSVC benchmarks are slower after r12-4195-gec0124e0acb556cd

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102809

--- Comment #1 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
There's a pending patch in this area that restricts loop threading:

https://gcc.gnu.org/pipermail/gcc-patches/2021-October/581894.html

Does this fix the problem?

[Bug tree-optimization/102809] Some TSVC benchmarks are slower after r12-4195-gec0124e0acb556cd

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102809

--- Comment #2 from Martin Liška <marxin at gcc dot gnu.org> ---
> Does this fix the problem?

Yes, it helps! Thank you for the patch.

[Bug tree-optimization/102809] Some TSVC benchmarks are slower after r12-4195-gec0124e0acb556cd

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102809

--- Comment #3 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
(In reply to Martin Liška from comment #2)
> > Does this fix the problem?
> 
> Yes, it helps! Thank you for the patch.

Thanks for all your help here, and sorry for all the noise.  Getting the jump
threading restrictions in the presence of loops has been chore.

It seems we have finally agreed on what should be allowed, and Jeff is about to
approve the last bits.

Unfortunately, the next step is playing with the pipeline to see if we can
remove some threading passes.  This will probably unsettle things, but
hopefully not as much.

Again, thanks.

[Bug tree-optimization/102809] [12 Regression] Some TSVC benchmarks are slower after r12-4195-gec0124e0acb556cd

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102809

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
   Target Milestone|---                         |12.0
            Summary|Some TSVC benchmarks are    |[12 Regression] Some TSVC
                   |slower after                |benchmarks are slower after
                   |r12-4195-gec0124e0acb556cd  |r12-4195-gec0124e0acb556cd

[Bug tree-optimization/102809] [12 Regression] Some TSVC benchmarks are slower after r12-4195-gec0124e0acb556cd

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102809

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
This should be fixed, at least the change is in.