[Bug middle-end/98173] New: -fno-tree-pta improves tfft2 benchmark by 50% on zen and -march=natie.

[Bug middle-end/98173] New: -fno-tree-pta improves tfft2 benchmark by 50% on zen and -march=natie.

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98173

            Bug ID: 98173
           Summary: -fno-tree-pta improves tfft2 benchmark by 50% on zen
                    and -march=natie.
           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: ---

I got curious about overall effect of PTA.
This compares -fno-tree-pta to -ftree-pta.  
There is significant regression on tfft2 (and some with tramp3d) with
-march=native on zen.

https://lnt.opensuse.org/db_default/v4/CPP/latest_runs_report?younger_in_days=14&older_in_days=0&min_percentage_change=0.02&revisions=03f48fb0d37c769661d30aa0f08d91bb7174cd98%2C68ee6d12fe9882223f47f81f93616577ab4e36da&include_user_branches=on

https://lnt.opensuse.org/db_default/v4/SPEC/latest_runs_report?younger_in_days=14&older_in_days=0&min_percentage_change=0.02&revisions=03f48fb0d37c769661d30aa0f08d91bb7174cd98%2C68ee6d12fe9882223f47f81f93616577ab4e36da&include_user_branches=on

[Bug middle-end/98173] -fno-tree-pta improves tfft2 benchmark by 50% on zen and -march=natie.

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98173

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Version|unknown                     |11.0
           Keywords|                            |missed-optimization

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
PTA certainly can increase register pressure by means of more disambiguations
and thus more store-motion or PRE.  But you hardly can blame PTA for that ;)

[Bug middle-end/98173] -fno-tree-pta improves tfft2 benchmark by 50% on zen and -march=natie.

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98173

--- Comment #2 from Jan Hubicka <hubicka at ucw dot cz> ---
> PTA certainly can increase register pressure by means of more disambiguations
> and thus more store-motion or PRE.  But you hardly can blame PTA for that ;)
It seems microarch specific, so it also can be just enabled
vectorization...

[Bug middle-end/98173] -fno-tree-pta improves tfft2 benchmark by 50% on zen and -march=natie.

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98173

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|missed-optimization         |
            Version|11.0                        |unknown

--- Comment #3 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Note that also nbench/FP EMULATION regresses on both zens but not on cabylake.