[Bug middle-end/110973] New: 9% namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[Bug middle-end/110973] New: 9% namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110973

            Bug ID: 110973
           Summary: 9% namd regression between g:c2a447d840476dbd
                    (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09
                    20:17)
           Product: gcc
           Version: 13.1.0
            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: ---

All is -Ofast -march=native
zen2
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=300.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=301.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=301.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=288.120.0
Icelake
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=789.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=790.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=785.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=791.120.0

[Bug middle-end/110973] 9% 444.namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110973

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization,
                   |                            |needs-bisection
             Target|                            |x86_64-*-*
                 CC|                            |rguenth at gcc dot gnu.org
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=110972

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Probably the same cause as PR110972, but I can only confirm a 2.5% regression
on Zen2 due to r14-3078-gd9f3ea61fe36e2.  Looks like the bigger regression was
for generic?  I can confirm a 5.5% regression for -Ofast generic on Zen2 due to
r14-3078-gd9f3ea61fe36e2.

[Bug middle-end/110973] 9% 444.namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110973

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2023-08-11
     Ever confirmed|0                           |1

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
For generic the biggest difference is in calc_pair_energy_fullelect:

Samples: 1M of event 'cycles', Event count (approx.): 2222678752091             
Overhead       Samples  Command          Shared Object             Symbol       
   6.68%        132170  namd_peak.amd64  namd_peak.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_energy_fullelect                         #
   6.16%        121796  namd_base.amd64  namd_base.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_fullelect                                #
   6.16%        121875  namd_peak.amd64  namd_peak.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_fullelect                                #
   5.98%        118457  namd_base.amd64  namd_base.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_energy_fullelect                         #
   5.33%        105302  namd_peak.amd64  namd_peak.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair                                          #
   5.19%        102880  namd_peak.amd64  namd_peak.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_energy                                   #
   5.15%        101892  namd_base.amd64  namd_base.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_energy_merge_fullelect                   #
   5.15%        102119  namd_peak.amd64  namd_peak.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_energy_merge_fullelect                   #
   5.02%         99415  namd_base.amd64  namd_base.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_energy                                   #
   5.00%         98964  namd_peak.amd64  namd_peak.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_merge_fullelect                          #
   4.98%         98257  namd_base.amd64  namd_base.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair                                          #
   4.90%         96943  namd_base.amd64  namd_base.amd64-m64-mine  [.]
ComputeNonbondedUtil::calc_pair_merge_fullelect

it's again very difficult to pin-point an offender :/

[Bug middle-end/110973] 9% 444.namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110973

Martin Jambor <jamborm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |fkastl at suse dot cz,
                   |                            |jamborm at gcc dot gnu.org

--- Comment #3 from Martin Jambor <jamborm at gcc dot gnu.org> ---
There was also a 7.7% regression on zen3 with generic march (these measurements
are without LTO):

https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=466.120.0

[Bug middle-end/110973] 9% 444.namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[fkastl at suse dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110973

--- Comment #4 from Filip Kastl <fkastl at suse dot cz> ---
(In reply to Martin Jambor from comment #3)
> There was also a 7.7% regression on zen3 with generic march (these
> measurements are without LTO):
> 
> https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=466.120.0

I think this slowdown is caused by the r14-3078-gd9f3ea61fe36e2 commit that
Richard mentioned. I measured it on another zen3 machine and got ~7% slowdown.

[Bug middle-end/110973] 9% 444.namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110973

--- Comment #5 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Note that some (not all?) namd scores seems to be back to pre-regression
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=798.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=791.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=299.120.0
between 2a0b19f52596d75b (2023-08-07 00:16) and b0894a12e9e04dea (2023-08-10
13:29)

[Bug middle-end/110973] 9% 444.namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110973

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

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

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Likely r14-3440-ge80f7c13f64e10 fixed it.

[Bug middle-end/110973] 9% 444.namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[fkastl at suse dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110973

--- Comment #7 from Filip Kastl <fkastl at suse dot cz> ---
Not all measurements are back to pre-regression. The Ofast zen3 generic score
that Martin mentioned
(https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=466.120.0) is still
higher than before.

[Bug middle-end/110973] 9% 444.namd regression between g:c2a447d840476dbd (2023-08-03 18:47) and g:73da34a538ddc2ad (2023-08-09 20:17)

[fkastl at suse dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110973

--- Comment #8 from Filip Kastl <fkastl at suse dot cz> ---
(In reply to Filip Kastl from comment #7)
> Not all measurements are back to pre-regression. The Ofast zen3 generic
> score that Martin mentioned
> (https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=466.120.0) is
> still higher than before.

There's also these two:
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=789.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=300.120.0

And these benchmarks haven't run yet since the commit that Richard thinks fixes
the regression:
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=785.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=790.120.0
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=288.120.0