Just released Mesa 7.10.1
Official Ubuntu 10.10 Xorg
In some areas of Lightsmark 2008 benchmark fps is not as good as older more stable Mesa versions such as Mesa 7.7.1, 7.8.3-rc1, 7.9-series.
some like half fps.
What's your hardware? resolution?
Official Ubuntu Xorg
xf86-video-intel 2.13.0 release
Intel Dual Core 2.20 GHz
RAM 3 GB
Intel GMA 4500M 256 MB
Three major bugs for Mesa 7.10.1, kicking me back to stay with Mesa 7.8.3-rc1
Bug 32949 - cannot play Windows games with WINE with GLSL enabled
This bug itself - which I guess has performance impact on other games
Bug 33926 - I do not know other Linux native games are affected by this bug
And please include the actual measurements.
Author: Kenneth Graunke <firstname.lastname@example.org>
Date: Tue Mar 1 11:15:34 2011 -0800
intel: Support glCopyTexImage() from ARGB8888 to XRGB8888.
Nexuiz was hitting a software fallback.
also fixes a fallback that Lightsmark was hitting. And Lightsmark is also sensitive to libgomp.so, so make sure you are using the same version of the compiler whilst comparing.
still low at beginning and other areas as well, like before with patched Mesa 7.10.1.
but I do not know about libgomp.so, which version I have to use and how to compare. I did not compile libgomp.so, it comes with Ubuntu 10.10.
Closing the bug, since the bug is not verifiable.
After playing with Mesa for 3 hours, I found that
Mesa 7.10.1 and Mesa 7.11-devel (mesa git) has this bug:
what: LightsMark: http://dee.cz/lightsmark/
benchmark has half fps in some areas (like 5 fps but before that was 11 fps).
This bug was not present in Mesa 7.9 series but 7.10.
This bug was fixed by that commit 1991d92207cf629ba4ceead4bfc3f768d7b9e402
but reintroduced by commit 0f7325b89038937bd428f7c89ed9859189a0ab0b
If I revert this commit in Mesa 7.10 and Mesa 7.11-devel then everything becomes normal again like before with Mesa 7.9 (11 fps).
changing importance to high so that this issue does not effect other programs
Author: Eric Anholt <email@example.com>
Date: Mon May 23 09:12:07 2011 -0700
i965/fs: Track fixed GRF regs separate from allocated GRF file in scheduling.
There's an assumption here that fixed GRFs will never intersect with
the allocated GRFs. That's true today, though it might change some
day if we decide to register-allocate the regs containing push
constants once they're dead.
This fixes a regression in 0f7325b89038937bd428f7c89ed9859189a0ab0b in
Lightsmark from the texture instructions now containing g0 references
instead of having that be implied. Performance is improved 15.2% +/-