xserver ( git ) crashes with exa & compiz ( r200 )

Daniel Kasak dan at entropy.homelinux.org
Tue Nov 6 05:26:12 PST 2007


On Tue, 2007-11-06 at 13:48 +0100, Michel Dänzer wrote:
>
> On Tue, 2007-11-06 at 23:25 +1100, Daniel Kasak wrote:
> > On Tue, 2007-11-06 at 12:27 +0100, Michel Dänzer wrote:
> > 
> > > > If I enable EXA though, I get an xserver crash ( see attached log ) when
> > > > compiz starts.
> > > 
> > > Does the attached Mesa patch work?
> > > 
> > > The __glXDRIbindTexImage() fallback needs to be fixed to use GetImage
> > > instead of touching the pixmap data directly though.
> > 
> > Yes it does :)
> > 
> > In 24bit colour, compiz now works.
> 
> Great. How's performance?

Mostly very good, but interesting ...

It's certainly different - if I accidentally had exa enabled when I ran
compiz, it used to run *very* slowly. It's certainly not running very
slowly any more. But *some* things are. Eterm, for example, is glacial
at everything, from scrolling lots of text, to just displaying a simple
text-based animation ( ie alternating between the characters: \|/- to
producing a 'spinner' thing ... even this takes up 80% of CPU on an
Athlon 1800XP! ) So there are certainly some losers, but overall,
performance is much better. I've switched to using xterm and I'm now
recompiling mesa with your 2nd patch. The system seems more responsive
than it used to ( ie with XAA ). So I'd say it's an definite
improvement. Keep in mind I've only used it for a couple of minutes.

> > In 16bit colour, something *very* strange happens ( warning: large png
> > vs slow DSL ):
> > http://entropy.homelinux.org/Screenshot.png
> 
> Does the attached patch help for this?

It fixes the massive corruption, black & white, etc. Compiz starts in
16bit mode, but emerald doesn't. I tried running emerald manually, but
it doesn't give any errors - it just runs and doesn't do anything. Ah
well.

Thanks again for your help and for getting back to me so quickly. I'm
off to bed now, but I'll do some more testing tomorrow if I get time
after work.

Dan




More information about the xorg mailing list