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

[Bernardo Innocenti]

Michel Dänzer wrote:

>> afaik, as of today XAA is still the default on all mayor Linux
>> distros.  Simple bug fixes are probably still wanted.
> 
> So are you volunteering to maintain XAA? :)

:-)

No, but I will contribute bugfixes like this one for problems that directly
affect me.  I depend on XAA because EXA is still unusably slow on all hardware
I ever tried it with.  I wonder if there's someone with a different experience.


>> I'm unsure how we could eliminate those 1x1 pixmaps used for solid fills,
>> but they're certainly a big performance hit.  In some cases, we even upload
>> them to the framebuffer by *dma*, then read the pixel with CPU :-)
> 
> Adam Jackson fixed that in GIT.

This patch?

 http://gitweb.freedesktop.org/?p=xorg/xserver.git;a=commitdiff;h=486fd4145aed93093d1f1655de40c0a8582bb8b1

That's not really a fix, rather a workaround: we still upload the pixmap to the
framebuffer, and we still allocate and initialize it in memory, which is also
unfortunate.

Moreover, even when we hit the fast path, exaGetPixmapFirstPixel() is a rather
slow implementation for a simple "read pixel" operation: it calls into 3 other
non inlineable functions, one of which rather complex.


> If there's still a measurable penalty in some cases with the above change,
> this is probably the way to go.

I always wanted to run oprofile on one of Cairo's benchmark to see how the
overhead is distributed.  But Carl Worth already provided plenty of proof.

The overhead is quite visible also with the naked eye: after walking through the
server-side code for drawing one trapezoid, I'm actually surprised it still runs
so fast :-)

-- 
   // Bernardo Innocenti
 \X/  http://www.codewiz.org/