Bug 101198 - Major performance regression for AMD card
Summary: Major performance regression for AMD card
Status: RESOLVED FIXED
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/Gallium/radeonsi (show other bugs)
Version: git
Hardware: x86-64 (AMD64) All
: medium major
Assignee: Default DRI bug account
QA Contact: Default DRI bug account
URL:
Whiteboard:
Keywords: bisected, regression
Depends on:
Blocks:
 
Reported: 2017-05-26 11:55 UTC by Fabian Maurer
Modified: 2017-05-26 15:02 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Attachments

Description Fabian Maurer 2017-05-26 11:55:17 UTC
With the latest git mesa, there are major performance drops. First noticed the issue with the 32bit steam game "Move or die", but it seems to affect the whole system. It basically makes the game unplayable.

Bisecting pointed me to the following commit:

commit 524d4fff9e0e58e812e71bc5ba2c92ffa73bee16
Author: Leo Liu <leo.liu@amd.com>
Date:   Wed May 3 14:06:35 2017 -0400

    amd/common: add vcn dec ip info query
    
    Signed-off-by: Leo Liu <leo.liu@amd.com>
    Reviewed-by: Christian König <christian.koenig@amd.com>

I tried reverting it, but it doesn't apply cleanly.


Running Arch Linux 64
Linux 4.11.2

System the issue was tested on:
- Arch Linux 64bit
- Linux 4.11.2, AMDGPU driver
- Radeon R9 285
Comment 1 Fabian Maurer 2017-05-26 12:09:42 UTC
Just noticed, this seems to be reverted already during the time I need to setup and bisect mesa.
Comment 2 Dieter Nützel 2017-05-26 12:10:14 UTC
Hi Fabian,

David reverted it and Leo Liu sent an update just in time:
Please try with this one:
https://lists.freedesktop.org/archives/mesa-dev/2017-May/156930.html

Dicussion was here:
https://lists.freedesktop.org/archives/mesa-dev/2017-May/156907.html
https://lists.freedesktop.org/archives/mesa-dev/2017-May/156931.html

Cheers,
Dieter
Comment 3 Kai 2017-05-26 12:12:56 UTC
(In reply to Fabian Maurer from comment #0)
> With the latest git mesa, [...]

Can't be the latest, since the commit you've identified is already reverted: <https://cgit.freedesktop.org/mesa/mesa/commit/?id=e1409f730251bf9d7e70046f58392de6e4e98815>. I find it helpful giving exact commit IDs of the builds I'm using (or the version number, whatever is more accurate). Otherwise bug reports don't tend to age well. ;-)

In any case, if you're bisection is correct, you should check a build of current HEAD (e1409f730251bf9d7e70046f58392de6e4e98815) and close this bug if appropriate. Otherwise a new biscetion would be required.
Comment 4 Dieter Nützel 2017-05-26 12:17:18 UTC
(In reply to Dieter Nützel from comment #2)
> Hi Fabian,
> 
> David reverted it and Leo Liu sent an update just in time:
> Please try with this one:
> https://lists.freedesktop.org/archives/mesa-dev/2017-May/156930.html
> 
> Dicussion was here:
> https://lists.freedesktop.org/archives/mesa-dev/2017-May/156907.html
> https://lists.freedesktop.org/archives/mesa-dev/2017-May/156931.html
> 
> Cheers,
> Dieter

s/Discussion...

Bonus:
When you've tested it fine, you can send four (first) 'Tested-by:' ;-)
Comment 5 Fabian Maurer 2017-05-26 12:32:19 UTC
Thanks, just tested, it works fine after the commit was reverted. Marking this as resolved.
Comment 6 leoxsliu 2017-05-26 15:02:54 UTC
I can reproduce the problem with kernel that has no vcn patch, and the fix sent to mesa-dev list was tested and have issue fixed.

Thanks,
Leo


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.