Bug 10113

Summary: [modesetting] Broken OpenGL
Product: xorg Reporter: Frederic Riss <frederic.riss>
Component: Driver/intelAssignee: Eric Anholt <eric>
Status: RESOLVED FIXED QA Contact: Xorg Project Team <xorg-team>
Severity: normal    
Priority: medium CC: colin, tomi
Version: git   
Hardware: Other   
OS: All   
Whiteboard:
i915 platform: i915 features:
Bug Depends on:    
Bug Blocks: 10102    
Attachments:
Description Flags
xorg configuration file
none
Xorg.log in EFI mode (Jerky OpenGL)
none
Xorg.log file in BIOS mode (totally broken OpenGL)
none
intel-man.patch none

Description Frederic Riss 2007-02-27 10:47:22 UTC
With recent git changes, OpenGL seems broken. I'm using a full Xorg+mesa build from git, everything but xf86-video-intel from master branch.

I'm using this build on a Core Duo MacMini under Linux 2.6.20 (from Ubuntu Feisty) + git drm modules. I run this box in 2 modes: BIOS emulation and pure EFI mode. The behavior is indeed different between the 2 modes:
 - in EFI mode, the OpenGL runs, but it 'jerks' maybe once a second.
 - in BIOS mode, OpenGL is totally broken.

My OpenGL test application is Clutter (svn co http://svn.o-hand.com/repos/clutter/branches/clutter-0-2). For example the super-oh example ran smoothly with old driver versions and looks broken with current git (last tested commit 3e8e75e5d83a2fa7e9fc6e9a3fbb07dac548ea5a).

I'll attach the xorg.conf, an Xorg.log in EFI mode and an Xorg.log in BIOS mode. Just ask if you want something more.
Comment 1 Frederic Riss 2007-02-27 10:50:56 UTC
Created attachment 8890 [details]
xorg configuration file
Comment 2 Frederic Riss 2007-02-27 10:52:48 UTC
Created attachment 8891 [details]
Xorg.log in EFI mode (Jerky OpenGL)
Comment 3 Frederic Riss 2007-02-27 10:53:22 UTC
Created attachment 8892 [details]
Xorg.log file in BIOS mode (totally broken OpenGL)
Comment 4 Adam Williamson 2007-03-02 17:32:22 UTC
Possibly related: 3D performance with recent git snapshot (yesterday) is far worse than with a January snapshot for me (and for Colin Guthrie, see http://lists.freedesktop.org/archives/xorg/2007-March/022227.html). I can run GL stuff - glxgears, beryl - but it's very slow (glxgears 300fps, beryl pretty much unusable). With my January snapshot this was fine. I also went to git drm modules with no noticeable change.
Comment 5 Michel Dänzer 2007-03-05 00:55:14 UTC
First of all, does this still happen with the xf86-video-intel master branch after the modesetting merge?

Does anybody have an idea when the regression happened? git-bisect would be ideal, but even a date range might help. In the meantime, please attach the stderr output from running a GL application with LIBGL_DEBUG=verbose.
Comment 6 Michel Dänzer 2007-03-05 07:54:24 UTC
Actually, I'm seeing this as well now and have isolated it a little with git-bisect. xf86-video-intel commit a61a6b1db610a07060d0dcca54b66a4b2b3686cc is the last good one I could find for now, 5732dd7221b7bb24bf21e16066f014a0c7cb6b2a the first bad one. Everything in between crashes on server startup. My first guess would be it's some kind of regression of the video memory allocation rework.
Comment 7 Eric Anholt 2007-03-06 14:03:59 UTC
I fixed a performance regression just a moment ago, which may have caused your issue.  Please retest with the current code in master.
Comment 8 Todd Merrill 2007-03-06 21:01:35 UTC
The latest commits have significantly improved things on my i915, but performance is still lagging behind commit a61a6b1db610a07060d0dcca54b66a4b2b3686cc.
Comment 9 Michel Dänzer 2007-03-07 05:43:04 UTC
That seems to have been it, thanks!
Comment 10 Michel Dänzer 2007-03-07 05:55:40 UTC
(In reply to comment #8)
> The latest commits have significantly improved things on my i915, but
> performance is still lagging behind commit
> a61a6b1db610a07060d0dcca54b66a4b2b3686cc.

Is it using the same 3D driver as before? The current code will only use i915tex with Option "Legacy3D" "off".
Comment 11 Todd Merrill 2007-03-07 07:20:34 UTC
(In reply to comment #10)
> Is it using the same 3D driver as before? The current code will only use
> i915tex with Option "Legacy3D" "off".

Nope, it wasn't using the same 3D driver.  I didn't realize that this had changed.      The man page could use an update to reflect this.  Disabling Legacy3D has restored performance.
Comment 12 Todd Merrill 2007-03-07 07:39:00 UTC
Created attachment 9018 [details] [review]
intel-man.patch

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.