Bug 93394 - [BDW, BSW] GLbenchmark_V3.0_gl_driver/gl_driver_offscreen/Synmark2_v6_Batch6/Batch7 show performance dropped by 5%-12%
Summary: [BDW, BSW] GLbenchmark_V3.0_gl_driver/gl_driver_offscreen/Synmark2_v6_Batch6/...
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/DRI/i965 (show other bugs)
Version: unspecified
Hardware: Other All
: medium normal
Assignee: Kenneth Graunke
QA Contact: Intel 3D Bugs Mailing List
Keywords: bisected
Depends on:
Reported: 2015-12-16 08:21 UTC by wendy.wang
Modified: 2016-11-02 07:04 UTC (History)
1 user (show)

See Also:
i915 platform:
i915 features:

xorg.log (18.36 KB, text/plain)
2015-12-16 08:21 UTC, wendy.wang

Description wendy.wang 2015-12-16 08:21:05 UTC
Created attachment 120536 [details]

Test environment:
Failed platforms: BDW/BSW

Mesa Regression: Yes, bisect is under progress

Bad gfx sw stack info:
Mesa: git-7752bbc
Kernel: 4.4-rc4-drm-intel-nightly_ad271e_20151213+
xf83: 2.99.917-517-g6c4ffba

GLbenchmark_V3.0_gl_driver/gl_driver_offscreen/Synmark2_v6_Batch6/Batch7 show performance dropped by 5%-12%

Reproduce steps:
1. cd /gfxbench3_desktop/build/linux/gfxbench_Release/mainapp
2. export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/games/OpenGL/gfxbench3_desktop/install/linux/lib
3. ./mainapp vblank_mode=0 -t gl_alu -fullscreen 1 -w 1920 -h 1080

xorg file has been attached.
Comment 1 ye.tian 2015-12-18 06:59:39 UTC
It's mesa regression.By bisect, shows the first bad mesa commit is c59d1b1.

commit c59d1b1fd1ac5dca3b769f42fe0e42a11c10d4b8
Author:     Kenneth Graunke <kenneth@whitecape.org>
AuthorDate: Fri Oct 9 17:07:23 2015 -0700
Commit:     Kenneth Graunke <kenneth@whitecape.org>
CommitDate: Fri Dec 11 13:11:15 2015 -0800

    i965: Add tessellation shader sampler support.

    Based on code by Chris Forbes and Fabian Bieler.

    Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
    Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Comment 2 Matt Turner 2016-11-02 05:11:35 UTC
I suspect the commit just caused the benchmark to do more work. Ken, is there anything to do here?
Comment 3 Kenneth Graunke 2016-11-02 07:04:25 UTC
The benchmark should do the same thing, but adding additional tracked state atoms increased the draw call overhead.  And these are CPU overhead benchmarks.

I don't think it's worth tracking as a regression.  We can't exactly retract core features.  We just need to optimize away draw call CPU overhead.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.