Bug 12908 - Bad bumblebee/primus performance probably due to mesa regression
Summary: Bad bumblebee/primus performance probably due to mesa regression
Status: RESOLVED WORKSFORME
Alias: None
Product: Mageia
Classification: Unclassified
Component: RPM Packages (show other bugs)
Version: 4
Hardware: x86_64 Linux
Priority: Normal normal
Target Milestone: ---
Assignee: Rémi Verschelde
QA Contact:
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2014-03-01 05:41 CET by Pavel Kreuzt
Modified: 2015-01-10 15:26 CET (History)
1 user (show)

See Also:
Source RPM: bumblebee
CVE:
Status comment:


Attachments

Description Pavel Kreuzt 2014-03-01 05:41:11 CET
Description of problem: ( From this thread: https://bugs.mageia.org/show_bug.cgi?id=10218 ) Bumblebee with virtualgl bridge and proxy method, or primus bridge, shows very bad performance, half than with Intel integrated graphics card alone, and never higher than 25fps. Since same version of bumblebee/primus worked quite well on Mga3 and mesa 9, I believe it's some problem with mesa 10. General issue looks similar to this https://bugzilla.novell.com/show_bug.cgi?id=828341 but that proposed workaround doesn't help.


Version-Release number of selected component (if applicable): mesa-10.0.3-1.mga4.tainted


How reproducible: Always when using bumblebee/primus


Steps to Reproduce:
1. Test glxspheres or glxgears with integrated graphics
2. Test glxspheres or glxgears with primusrun/optirun
3. See the output and different framerates


Reproducible: 

Steps to Reproduce:
Comment 1 Thomas Backlund 2014-03-01 12:55:08 CET
It also may be the pile of hacks that bumblebee is that is not properly coded/tested for mesa 10 as mesa 10 without bubmlebee works good here.

so assigning to bumblebee until someone can prove mesa regression _without_ bumblebee

not to mention we dont support bumblebee in any stable release so far, and to be honest I hope we never will

CC: (none) => tmb
Assignee: bugsquad => remi
Source RPM: mesa-10.0.3-1.mga4.tainted.src.rpm => bumblebee

Comment 2 Pavel Kreuzt 2014-03-01 21:48:14 CET
Can you tell me which tests I can use to have a look to mesa 10 behaviour? Something in mesa-demos? I looked for subtexrate test mentioned in novell's bugzilla but didn't find it in Mga4.
Comment 3 Pavel Kreuzt 2014-03-14 14:45:13 CET
Posted an issue to primus' github https://github.com/amonakov/primus/issues/135
Comment 4 Pavel Kreuzt 2014-04-06 04:17:02 CEST
Any news on this? According to primus' author it's an upload-to-mesa issue (see https://github.com/amonakov/primus/issues/135 ) So it seems clear mesa is responsible. Some other issues on primus' github say the regression pressent in mesa 9.2 related to texture upload (that affects primus too) was only *partially* solved in mesa 10.0, and in mesa 10.1 there is a new method that doesn't need to use textures at all (although it has it's own problems). Even so, mesa 10.0 teoretically has a performance similar to 9.1, so we've got a problem here. Amonakov remarks this kind of issues affect mostly rpm-based distros, so maybe a compile option of our mesa is interfering?
Comment 5 Rémi Verschelde 2014-12-28 15:55:42 CET
Do you still experience the same issue with mesa-10.0.5-1.1.mga4.tainted from the Updates repo? I was not able to reproduce such troubles with bumblebee or Mageia 4, but I haven't used it long since I moved to Cauldron months ago.

But most of the reports that I got about the bumblebee package in backports_testing were quite good, so I guess there's something specific to your hardware.
Comment 6 Rémi Verschelde 2015-01-10 15:26:41 CET
Closing as WORKSFORME for now, I'm not sure this is a bug for which I could really provide support. If upstream comes up with a patch, I'll gladly include it of course.

Status: NEW => RESOLVED
Resolution: (none) => WORKSFORME


Note You need to log in before you can comment on or make changes to this bug.