Ati radeon performances

[Michel Dänzer]

On Mon, 2008-04-14 at 22:21 +0200, Matthieu Baechler wrote:
> Le lundi 14 avril 2008 à 13:57 -0400, Alex Deucher a écrit :
> > On Mon, Apr 14, 2008 at 1:49 PM, Matthieu Baechler <m.baechler at free.fr> wrote:
> > > Le lundi 14 avril 2008 à 16:31 +0200, Michel Dänzer a écrit :
> > >
> > >  > > Is there something we can do to avoid these migration ?
> > >  >
> > >  > A good start is usually finding out why it's falling back to software.
> > >  > You can change RADEON_TRACE_FALL to 1 in
> > >  > xf86-video-ati/src/radeon_exa.c, rebuild the driver, re-run the test and
> > >  > put up the log file (compressed if it's very big) somewhere for us to
> > >  > look at.
> > >
> > >  I put my Xorg.log there : http://pastebin.com/m2717acea
> > 
> > see bugs:
> > http://bugs.freedesktop.org/show_bug.cgi?id=15371
> > http://bugs.freedesktop.org/show_bug.cgi?id=15334
> 
> I patched my radeon driver with patches from #15371, and it reduces
> migration a lot.
> The same gnome-system-monitor test case dropped from 20% to 5% of CPU.
> I didn't noticed any regression. Is there something preventing from
> pushing it to master ?

I don't think so, somebody just needs to get around to it.

> While I am at this, the next CPU killer seems to be this one (browsing
> the internet triggers this) :
> 
> R200CheckCompositeTexture: Unsupported picture format 0x1011000
> 
> This time I checked bugzilla, I found nothing about it.

That's the a1 (single bit alpha) format, which EXA doesn't accelerate
yet. This could be related to non-antialiased fonts.

BTW, as you're using current xserver code, Option "EXAOptimizeMigration"
might help, but read its description in the exa manpage first.


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