https://gcc.gnu.org/bugzilla/show_bug.cgi?id=72517

            Bug ID: 72517
           Summary: 436.cactusADM: More than 40% regression in O3 and
                    Ofast on AMD bdver4 m/c.
           Product: gcc
           Version: 7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: Amit.Pawar at amd dot com
  Target Milestone: ---

Starting from git commit id 5695a690b29a98205a8ae29a6072fd2e42cf1c84 cactusADM
showing >40% regression with generic and > 8% with native on AMD BDVER4
machine. Following table describes the flags used.

Git bisecting shows 

5695a690b29a98205a8ae29a6072fd2e42cf1c84 is the first bad commit
commit 5695a690b29a98205a8ae29a6072fd2e42cf1c84
Author: rguenth <rguenth@138bc75d-0d04-0410-961f-82ee72b054a4>
Date:   Wed Jun 15 09:53:07 2016 +0000

    2016-06-15  Richard Biener  <rguent...@suse.de>

        * tree-vect-data-refs.c (vect_analyze_data_ref_dependence): Do
        not consider dependences between accesses that belong to the
        same group.
        (vect_analyze_data_ref_dependences): Do not analyze read-read
        or self-dependences.

        * gcc.dg/vect/bb-slp-pattern-2.c: Disable loop vectorization.


    git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@237473
138bc75d-0d04-0410-961f-82ee72b054a4


Git id "61958e92f84b20ec65157388ac221d0f23291b03" (good trunk) produces good
score on bdver4. Tested on bdver2,bdver3 and bdver4 machines and shows only on
bdver4.

Flags           Good Trunk      Trunk        Difference
O3              37.69           20.2            -46%
O3 bdver4       36.55           32.2            -12%
Ofast           40.36           24              -41%
Ofast bdver4    38.31           35.4            -8%

Reply via email to