[PATCH] Ensure blitter quiescience before reading pixels from the framebuffer

Bernardo Innocenti bernie at codewiz.org
Fri Aug 3 16:05:31 PDT 2007


Dan Amelang wrote:

> On 8/3/07, Bernardo Innocenti <bernie at codewiz.org> wrote:
>> Indeed, we've got very counter-intuitive results with oprofile.
>> Over 30% of the time is being spent in fbFetch_r5g6b5(), which is
>> funny given that the screen is 24bpp!
> 
> Hmm...I know certain models of the OLPC (at least the model I have)
> use 16bpp. I see fbFetch_r5g6b5 at the top on mine also, for the
> obvious reason that there are few "fast paths" in pixman for r5g6b5
> (so you hit the slow, general code). Are you positive that your model
> isn't using 16bpp?

The default is 16bpp, but I changed that and I double checked that my
xorg.conf really says Depth 24.

We're considering switching to 24bit (32bpp) because format conversions
are expensive and Cairo works mostly with a8r8b8g8.

The downside is that we waste double as much video memory and simple
operations such as fills and copies instantly become 2x slower.

-- 
   // Bernardo Innocenti - Develer S.r.l., R&D dept.
 \X/  http://www.develer.com/




More information about the xorg mailing list