Using the current infrastructure (kernel 2.6.29-rc3, xorg-server 22.214.171.1242, xf86-video-intel 2.6.1, mesa head from a few days ago), 3D acceleration doesn't work at all on an i945GM - e.g. starting tuxracer draws just its light blue background screen.
An identical installation works perfectly with a Q35.
Problem still present in 2.6.29-rc4 with latest patches from drm-intel-next applied.
This is on an Acer Aspire One (8086:27ae - 1025:015b); there is no indication of anything going wrong in dmesg or in the output of the application even with LIBGL_DEBUG=verbose
I'm starting to suspect the DRI driver is drawing on the VGA port or something else that happens to be not in use.
All the usual information is required (dmesg, Xorg.0.log).
When did it start happening? Can you bisect?
Created attachment 28407 [details]
Xorg log file
Xorg log file
Created attachment 28408 [details]
dmesg log file
On eeebox b202: atom n270 + 945GME,
using ubuntu 9.04 with kernel 126.96.36.199
using lastest intel drivers (2009 Q2)
2D looks good performance
3D is catastrophic!
128 frames/secondes in glxgears against 284 before
unable to use real 3D application, like Google Earth
What is the problem?
(In reply to comment #5)
> On eeebox b202: atom n270 + 945GME,
> using ubuntu 9.04 with kernel 188.8.131.52
> using lastest intel drivers (2009 Q2)
> 2D looks good performance
> 3D is catastrophic!
> 128 frames/secondes in glxgears against 284 before
> unable to use real 3D application, like Google Earth
> What is the problem?
I was wrong: 2D is worse too....slower
Are you using Mesa 7.5?
(In reply to comment #7)
> Are you using Mesa 7.5?
Yes 7.5, compiled with make argument "linux-dri-x86", something like that.
By the way, all CGLAGS and CXXFLAGS are ignored, since the makefile oblige us to make a "make realclean", deleting user config options created by configure script.
Is there a way to compile with user CFLAGS option? Like "-march=..."?
FYI, i use the 1900*1080 mode, and xvmc option in xorg.conf is enabled.
Mass version move, cvs -> git
given that this isn't the behavior we see on 945GM systems, I'll leave this as NEEDINFO until we can get a bisect or other clue as to what makes your system special.
If the CPU is busy when you have slow rendering, then sysprof would be a tool to use.
Feedback timeout from original submitter.