ati driver falling back to software

[Michel Dänzer]

On Wed, 2008-06-18 at 19:30 -0700, Ray Van Dolson wrote:
> > > The xfwm4 compositor is *disabled*.  I have just tried it with only one
> > > screen attached and the virtual screen size set to 1600x1200.  I no
> > > longer see the R300CheckComposite errors in my Xorg logfile and XFCE
> > > Terminal resize is *definitely* better.  xchat resize seems about the
> > > same though (aka pretty bad), even though I don't see any errors at all
> > > in the Xorg logfile.
> > 
> > It's probably falling back somewhere else then. You can try rebuilding
> > xserver with DEBUG_TRACE_FALL defined to 1 in exa/exa_priv.h, though
> > that may generate a lot of debugging output.
> > 
> > BTW, Option "AccelDFS" may alleviate the migration overhead for
> > fallbacks somewhat.
> > 
> > > An oprofile run resizing xchat for about 20 seconds still shows a lot
> > > of calls to memcpy (being called by exaCopyDirty):
> > 
> > Thanks for the data, but I have to admit I can't make sense of oprofile
> > callgraphs. Can you provide data from sysprof or at least opreport
> > -gdf / op2calltree?
> > 
> 
> Alright, I've modified my Xorg config file to use AccelDFS 

I guess that didn't make a difference though? The profile doesn't seem
to show any significant fallbacks.

> and rebuilt Xorg with DEBUG_TRACE_FALL set.  In addition I also re-ran a profile
> using sysprof.  Output of everything can be found here:
> 
>   http://rayvd.fedorapeople.org/xorg/20080618.1/

That's a much more useful profile, thanks.

It looks like most cycles are spent for text rendering. Have you tried
or could you try the xserver Git master branch to see if the text
rendering and other optimizations there help?


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