Bug 82804 - [7/8/9 Regression] SPEC CPU2006 470.lbm ~5% performance deviation with r237185
Summary: [7/8/9 Regression] SPEC CPU2006 470.lbm ~5% performance deviation with r237185
Status: RESOLVED WONTFIX
Alias: None
Product: gcc
Classification: Unclassified
Component: target (show other bugs)
Version: 7.2.0
: P3 normal
Target Milestone: 7.4
Assignee: Not yet assigned to anyone
URL:
Keywords: missed-optimization
Depends on:
Blocks: spec 84613
  Show dependency tree / graph
 
Reported: 2017-11-02 08:28 UTC by Martin Liška
Modified: 2018-06-08 13:21 UTC (History)
2 users (show)

See Also:
Host:
Target: x86_64-*-*
Build:
Known to work:
Known to fail:
Last reconfirmed: 2017-11-02 00:00:00

Attachments
perf annotate (588.73 KB, application/x-bzip)
2017-11-02 08:37 UTC, Martin Liška 		Details
perf report (8.31 KB, text/plain)
2017-11-02 08:37 UTC, Martin Liška 		Details
perf diff (4.18 KB, text/plain)
2017-11-02 08:38 UTC, Martin Liška 		Details
perf annotate (637.04 KB, application/x-bzip)
2017-11-02 08:39 UTC, Martin Liška 		Details
perf report (8.20 KB, text/plain)
2017-11-02 08:39 UTC, Martin Liška 		Details
View All 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 Martin Liška 2017-11-02 08:28:03 UTC 
Starting with the revision we lost about ~5 on Zen wit -O2 -mtune=native.
Comment 1 Martin Liška 2017-11-02 08:37:32 UTC 
Created attachment 42520 [details]
perf annotate
Comment 2 Martin Liška 2017-11-02 08:37:58 UTC 
Created attachment 42521 [details]
perf report
Comment 3 Martin Liška 2017-11-02 08:38:20 UTC 
Created attachment 42522 [details]
perf diff
Comment 4 Martin Liška 2017-11-02 08:39:00 UTC 
Created attachment 42523 [details]
perf annotate
Comment 5 Martin Liška 2017-11-02 08:39:37 UTC 
Created attachment 42524 [details]
perf report
Comment 6 Martin Liška 2017-11-02 08:44:13 UTC 
Can't be reproduced on Haswell.
Comment 7 Richard Biener 2018-01-25 08:22:30 UTC 
GCC 7.3 is being released, adjusting target milestone.
Comment 8 Martin Jambor 2018-06-08 13:21:56 UTC 
This is a code layout issue with Zens (and a very quirky one, it is sometimes measurable and sometimes not, depending how the function happens to be placed on a particular system) and probably nothing that GCC can help with.