[ANNOUNCE] xf86-video-ati-6.6.191

Xavier Bestel xavier.bestel at free.fr
Wed Apr 4 01:09:30 PDT 2007

On Wed, 2007-04-04 at 08:45 +0200, Michel Dänzer wrote:
> On Wed, 2007-04-04 at 06:56 +1000, Daniel Kasak wrote:
> > Xavier Bestel wrote:
> > 
> > >> Does this one works on r300 with EXA ? Esp. with 3D compositing WM ?
> > >>     
> > >
> > > Answering my own question here, for those interested: it's still
> > > extremely slow with EXA + beryl on r300, I had to revert to XAA for now
> > >   
> > As far as I know, there's no point in using EXA on r300, fullstop. The
> > latest xserver release candidates have some server-side EXA stuff that
> > make compositing usable ... almost ... but if you're using Beryl, then
> > you're getting than via another method, so I don't really see why you
> > want EXA anyway.
> You're right insofar as with the way GLX_EXT_texture_from_pixmap is
> currently implemented with AIGLX, 2D acceleration is getting in the way
> more than it helps. Even with XAA, Option "XaaNoOffscreenPixmaps" or the
> equivalent is generally needed for good performance, which basically
> disables 2D acceleration when a compositing manager runs as pretty much
> everything is a pixmap then. The same can be achieved with EXA with
> Option "FBTexPercent" "100", but there's little if any difference
> between them in that case, both will basically just pass through to the
> fb layer...
> However, this changes with something like
> http://people.freedesktop.org/~daenzer/aiglx-zero-copy-tfp/ , which is
> only possible with EXA.

Just out of curiosity: from looking at your code you're designing an API
to directly access a pixmap as a texture, which is then implemented for
r300 only. Does that mean that e.g. Intel chips or r200 chips don't need
that accelleration ? Because from the sound of it, people with these gfx
chips seem to report acceptable speed when running beryl.


More information about the xorg mailing list