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

[Bernardo Innocenti]

Carl Worth wrote:

>> 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 :-)
> 
> CPUs are pretty impressive that way. Single-stepping through the code
> is definitely not the way to find bottlenecks. It all seems horribly
> slow when you single step through piles of code, (but that doesn't
> necessarily mean that those piles are the things that are causing
> slowdowns).

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!

We're probably doing something silly, but it's hard to tell because
we can't get useful call graphs out of oprofile.  Tomorrow we'll give
it another shot, maybe with sysprof as recommended by Michel.

I'm attaching the testcase in case someone is interested.

-- 
   // Bernardo Innocenti
 \X/  http://www.codewiz.org/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bench.py
Type: text/x-python
Size: 1171 bytes
Desc: not available
URL: <http://lists.x.org/archives/xorg/attachments/20070803/6310bbc5/attachment.py>