| Summary: | [Bisected SNB] 30% of 3D performance regression on glbenchmark/Egypt and lightsmark | ||
|---|---|---|---|
| Product: | Mesa | Reporter: | libo <bo.c.li> |
| Component: | Drivers/DRI/i965 | Assignee: | Eric Anholt <eric> |
| Status: | VERIFIED FIXED | QA Contact: | |
| Severity: | major | ||
| Priority: | high | CC: | chadversary |
| Version: | git | ||
| Hardware: | All | ||
| OS: | Linux (All) | ||
| Whiteboard: | |||
| i915 platform: | i915 features: | ||
| Bug Depends on: | |||
| Bug Blocks: | 42993 | ||
|
Description
libo
2011-11-29 02:04:52 UTC
update:
For glbenchmark Egypt, the frames value changed from 11384.0 to 8128.0,
about 28% regression.
For lightsmark, the fps value changed from 28.69 to 24.15. about 14% regression.
I have put some patches and bisected again,get the result as follow:
The first bad commit could be any of:
1 d7b33309fe160212f2eb73f471f3aedcb5d0b5c1
2 da2816a45e6e3a33246a341fee72e6f893f315d9
3 005149d5860ad55c5e58e2de8a138e3a763f2036
4 c2ddde70daece24d1eb7185946032158993a9e4d
5 278e77a1192d5251c5b70a555e676f72b446e8e1
6 c3c7cbd15418293208034e8970d626b5998abd4b
These commit couldn't be build successfully.
for bisecting, the one before these commits is bad commit :
commit: 24da7335b22432ef4c2d57cab86e4b8fbe8733d5
Author: Chad Versace <chad.versace@linux.intel.com>
Date: Tue Nov 15 10:05:21 2011 -0800
and the next one of these commit is good commit:
commit: 19faa12bb751eabc1590d042adbd345b89542de0
Author: Chad Versace <chad.versace@linux.intel.com>
Date: Sun Nov 13 23:05:47 2011 -0800
By the way, I find 3DMMES/uimark has about 43.67 % regression , changed from 44.9563 to 25.3248 . I'm surprised to discover that this merge caused a performance regression. I'm not denying that it did, just stating my surprise. I tested Egypt's performance on my HiZ branch *before* the merge, and there existed a performance *improvement*. Also, Phoronix performed several benchmarks after the merge, and discovered no performance degradations, only improvements. I will try look into this on Monday. What git url should I use to clone 3dmmes? you can download from: http://tinderbox.sh.intel.com/games/glbenchmark.tar.gz after extract it ,you can run it as follow steps: 1. cd /glbenchmark/projects/Linux/ 2. rm -f GLBenchTestInput.txt 3. cp Egypt.txt GLBenchTestInput.txt 4. start X and gnome-session 5. ./glbenchmark_nogui (In reply to comment #3) > I'm surprised to discover that this merge caused a performance regression. I'm > not denying that it did, just stating my surprise. I tested Egypt's performance > on my HiZ branch *before* the merge, and there existed a performance > *improvement*. Also, Phoronix performed several benchmarks after the merge, and > discovered no performance degradations, only improvements. > > I will try look into this on Monday. What git url should I use to clone 3dmmes? (In reply to comment #3) You can also download 3DMMES from: http://tinderbox.sh.intel.com/games/3DMMES.tar.gz > I'm surprised to discover that this merge caused a performance regression. I'm > not denying that it did, just stating my surprise. I tested Egypt's performance > on my HiZ branch *before* the merge, and there existed a performance > *improvement*. Also, Phoronix performed several benchmarks after the merge, and > discovered no performance degradations, only improvements. > > I will try look into this on Monday. What git url should I use to clone 3dmmes? commit 7f854a5028ef00a0e041e78a770456204dc58eef Author: Eric Anholt <eric@anholt.net> Date: Fri Dec 16 13:18:55 2011 -0800 intel: Fix performance regression in Lightsmark since HiZ changes. Since the refactor in d7b33309fe160212f2eb73f471f3aedcb5d0b5c1, depth in the miptree changed from 1 to 6, so we always decided it didn't match, and we would relayout to something that would still not "match". Improves performance 23.8% (+/- 1.1%, n=4) Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=43329 |
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.