Alex Deucher asked me to test a patch in a bugreport and I got a bit
carried away and did a lot of benchmarks. :-)
Now that I've made them, I thought someone else might find them interesting!

Data:
Bug: http://bugs.freedesktop.org/show_bug.cgi?id=26225
The test: http://alteredqualia.com/visualization/evolve/
Fix: http://cgit.freedesktop.org/xorg/dri...00849f27cc5d63
Microbenchmark: x11perf -getimage500
Card: Radeon 4770 pcie

The interesting numbers are for noaccel and lucid firefox 3.7a3,
specially if you compare them to karmic.

Regards
/Ernst

Code:
compiz off xserver 1.6.5 KARMIC with a ton of backports
                evolve
                3.5     3.6     3.7a3      gtkperf (Human) glxgears
2.6.34-rc1 (kms)
    -090929     2.7     3.7     3.9        14.8             400
    -100314     3.0     3.9     3.9         9.53           2630
    -100315    same    same    same         same           same
    noaccel    11.5    34.0    55.0         3.60            430
    
2.6.32+radeontest (kms)
    -090929     2.7     3.7     3.9        16.27            
    -100314     3.1     3.7     3.9        10.24           2030
    -100315    same    same    same        same            same



compiz off xserver 1.7.5.902 LUCID, newer gtk, ubuntu's -32 has drm -33
                evolve
                --      3.6    3.7a3   gtkperf (Clearlooks)glxgears
2.6.32-16.25 (kms)
    -6.12.191          10.0    26.0        4.94            2630
    -100314            10.0    26.0        4.77            2630
    -100315            12.0    48.0        4.8             2630

2.6.34-rc1 (kms)
    -100315            12.2    48.0        4.9             2630




100314 is 67e81c8f
100315 is 819b4015
090929 has drm 2.4.14 instead of 2.4.19...