EXA performance ( XCOMPOSITE and XRENDER ) performance regression from xorg 7.0 to 7.1 ( r200 )

Michel Dänzer michel at tungstengraphics.com
Sun Jul 2 05:58:39 PDT 2006


On Sat, 2006-07-01 at 21:50 -0700, James Blanford wrote:
> 
> I'm running a Radeon 8500, also an r200, known to be dog slow with XAA,
> but seems to be fast with EXA.  It seemed to speed up if anything when I
> switched to cvs (git?) xserver and drivers.  

Yes, we've been making big improvements since 7.1. I'm using the
following setup:

      * xserver git with exa-damagetrack branch merged to my-master
        branch (based on master) and Option "MigrationHeuristic"
        "always".
      * xf86-video-ati git with Option "AccelDFS" (enabled by default
        with PCI(e), but not AGP) and Option "FBTexPercent" "0".

This is significantly faster than 7.0 or 7.1 for me with xfwm4's
built-in compositor, YMMV.

If you don't feel like playing with this many experimental bits, maybe
Option "MigrationHeuristic" "greedy" makes a difference for your
workload.


> I tried 15 windows on one desktop and 10 in another and didn't notice any 
> slowdown or high cpu usage.  I didn't use transset, though. I used fluxbox's 
> built-in transparency control.

That probably uses the same mechanism though.


> I've noticed huge differences in performance among the various
> compositing and window managers.  Window moves are jerky and peg my cpu
> with metacity and libcm.  

With AIGLX, that's expected due to the still inefficient implementations
of GLX_EXT_texture_from_pixmap and glTexSubImage. Again, xserver git may
be somewhat better though.

> Compiz is almost unbelievably fast and responsive.  Unfortunately, I get 
> no window decorations.

Is one of its window decorators running?


-- 
Earthling Michel Dänzer           |          http://tungstengraphics.com
Libre software enthusiast         |          Debian, X and DRI developer





More information about the xorg mailing list