Bug 96825 - [11/12 Regression] Commit r11-2645 degrades CPU2017 548.exchange2_r by 35%
Summary: [11/12 Regression] Commit r11-2645 degrades CPU2017 548.exchange2_r by 35%
Status: RESOLVED DUPLICATE of bug 98782
Alias: None
Product: gcc
Classification: Unclassified
Component: ipa (show other bugs)
Version: 11.0
: P3 normal
Target Milestone: 11.3
Assignee: Not yet assigned to anyone
URL:
Keywords: missed-optimization
Depends on:
Blocks: spec
  Show dependency tree / graph
 
Reported: 2020-08-27 20:12 UTC by pthaugen
Modified: 2021-12-03 11:44 UTC (History)
15 users (show)

See Also:
Host: powerpc64-unknown-linux-gnu
Target: powerpc64-unknown-linux-gnu
Build: powerpc64-unknown-linux-gnu
Known to work:
Known to fail:
Last reconfirmed: 2020-08-28 00:00:00

Attachments
Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description pthaugen 2020-08-27 20:12:48 UTC 
The given commit (1118a3ff9d3ad6a64bba25dc01e7703325e23d92) causes a 35% degradation for exchange2_r on Power9 built with the options "-O2 -mcpu=power9". Switching to -O3 results in a 44% degradation. The degradation occurs in __brute_force_MOD_digits_2().
Comment 1 Richard Biener 2020-08-28 06:18:16 UTC 
Yep, known.  Still bugs need to be fixed.
Comment 2 ktkachov 2021-03-17 14:51:50 UTC 
See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98782 for more analysis
Comment 3 Bill Schmidt 2021-03-17 15:06:20 UTC 
Is this going to be addressed in GCC 11?  Should this be only a P3?
Comment 4 Richard Biener 2021-04-09 07:56:14 UTC 
I believe there have been improvements recently - can you re-assess the magnitude of the problem?  The corresponding ARM PR got re-targeted to GCC 12 (for a RA fix), I think Martin has improved the IPA CP parts, maybe not fully though.

Martin?
Comment 5 Martin Jambor 2021-04-09 09:25:37 UTC 
I have not benchmark results from Power, but the reported regression has been fixed/mitigated on Zens, see:

https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=275.407.0&plot.1=397.407.0&plot.2=294.407.0&

or

https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=260.407.0&plot.1=361.407.0&plot.2=33.407.0&

We can still do better even at -Ofast and have an -O2 regression with that benchmark, I hope that both are covered by PR 98782 (which is IMHO quite generic, not ARM specific).

So I think this is fixed and we should deal with the existing problems in the other bug (but it would be nice if someone confirmed that Power also no longer regresses this bad).
Comment 6 pthaugen 2021-04-09 15:28:05 UTC 
(In reply to Richard Biener from comment #4)
> I believe there have been improvements recently - can you re-assess the
> magnitude of the problem?  The corresponding ARM PR got re-targeted to GCC
> 12 (for a RA fix), I think Martin has improved the IPA CP parts, maybe not
> fully though.

There has been no improvement seen on Power since the degradation appeared.
Comment 7 Jakub Jelinek 2021-04-27 11:39:18 UTC 
GCC 11.1 has been released, retargeting bugs to GCC 11.2.
Comment 8 Richard Biener 2021-07-28 07:05:02 UTC 
GCC 11.2 is being released, retargeting bugs to GCC 11.3
Comment 9 Jan Hubicka 2021-12-03 11:44:36 UTC 
We have two PRs about the same revision causing regression on exchange2.  It is understood now that it is more of an IRA issue (which does not like more realistic profile sadly)

*** This bug has been marked as a duplicate of bug 98782 ***