[Bug target/110758] [14 Regression] 8% hmmer regression on zen1/3 with -Ofast -march=native -flto between g:8377cf1bf41a0a9d (2023-07-05 01:46) and g:3a61ca1b9256535e (2023-07-06 16:56); g:d76d19c9bc5

[hubicka at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110758

Jan Hubicka  changed:

   What|Removed |Added

 Resolution|--- |FIXED
 Status|NEW |RESOLVED

--- Comment #4 from Jan Hubicka  ---
Fixing profile update in loop distribution fixed the regression and we get
better result than before.  The first tester still shows regression compared to
July last year, but we have PR106293 to track this.

[Bug target/110758] [14 Regression] 8% hmmer regression on zen1/3 with -Ofast -march=native -flto between g:8377cf1bf41a0a9d (2023-07-05 01:46) and g:3a61ca1b9256535e (2023-07-06 16:56); g:d76d19c9bc5

[hubicka at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110758

Jan Hubicka  changed:

   What|Removed |Added

   Last reconfirmed||2023-07-26
 Status|UNCONFIRMED |NEW
 Ever confirmed|0   |1

--- Comment #3 from Jan Hubicka  ---
Regression between  
  g:d76d19c9bc5ef113 (2023-07-16 00:16) and 
  g:a5088dc3f5ef73c8 (2023-07-17 03:24)  
seems to be gone. 


The range is:
  1c6231c05bdccab3 (2023-07-21 03:06) and
  f33fdf9e7c038639 (2023-07-23 00:17)

Looking at the patches in range, it is likely the flat profile detection in 
vectorizer:

Author: Jan Hubicka 
Date:   Fri Jul 21 19:38:26 2023 +0200

Avoid scaling flat loop profiles of vectorized loops

[Bug target/110758] [14 Regression] 8% hmmer regression on zen1/3 with -Ofast -march=native -flto between g:8377cf1bf41a0a9d (2023-07-05 01:46) and g:3a61ca1b9256535e (2023-07-06 16:56); g:d76d19c9bc5

[hubicka at ucw dot cz via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110758

--- Comment #2 from Jan Hubicka  ---
> I suspect this is most likely the profile updates changes ...
Quite possibly. The goal of this excercise is to figure out if there are
some bugs in profile estimate or whether passes somehow preffer broken
profile or if it is just back luck.

Looking at sphinx and fatigue it seems that LRA really may preffer
increased profile counts in peeled vectorized loop since it does not
understand the fact that putting spill on critical path through the
dependnecy graph of the code is not good for out of order execution.

[Bug target/110758] [14 Regression] 8% hmmer regression on zen1/3 with -Ofast -march=native -flto between g:8377cf1bf41a0a9d (2023-07-05 01:46) and g:3a61ca1b9256535e (2023-07-06 16:56); g:d76d19c9bc5

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110758

--- Comment #1 from Andrew Pinski  ---
I suspect this is most likely the profile updates changes ...

[Bug target/110758] [14 Regression] 8% hmmer regression on zen1/3 with -Ofast -march=native -flto between g:8377cf1bf41a0a9d (2023-07-05 01:46) and g:3a61ca1b9256535e (2023-07-06 16:56); g:d76d19c9bc5

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110758

Andrew Pinski  changed:

   What|Removed |Added

 Blocks||26163
Version|13.1.0  |14.0
   Keywords||missed-optimization
   Target Milestone|--- |14.0
Summary|8% hmmer regression on zen1 |[14 Regression] 8% hmmer
   |and zen3 with -Ofast|regression on zen1/3 with
   |-march=native -flto between |-Ofast -march=native -flto
   |g:8377cf1bf41a0a9d  |between g:8377cf1bf41a0a9d
   |(2023-07-05 01:46) and  |(2023-07-05 01:46) and
   |g:3a61ca1b9256535e  |g:3a61ca1b9256535e
   |(2023-07-06 16:56) and  |(2023-07-06 16:56);
   |g:d76d19c9bc5ef113  |g:d76d19c9bc5ef113
   |(2023-07-16 00:16) and  |(2023-07-16 00:16) and
   |g:a5088dc3f5ef73c8  |g:a5088dc3f5ef73c8
   |(2023-07-17 03:24)  |(2023-07-17 03:24)


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=26163
[Bug 26163] [meta-bug] missed optimization in SPEC (2k17, 2k and 2k6 and 95)