[Pixman] [PATCH 0/8] Performance statistics analyzer

Taekyun Kim podain77 at gmail.com
Tue Sep 27 20:24:21 PDT 2011


thanks for reviewing these patches!
I wrote some comments below.

On 09/27/2011 11:01 PM, Søren Sandmann wrote:
> The one highlevel comment I have is about fast_composite_solid_fill(),
> where you change from calling pixman_fill() to directly calling the new
> pixman_fill_fast_path(). This is problematic because pixman_fill() could
> call the SSE2 (or NEON) version of fill, whereas pixman_fill_fast_path()
> is always C only.
I also worried about this. The reason why I modified this to simply call 
specific fast path rather than pixman_fill() is that there seem to be no 
where SSE2 or NEON fill function is called through C fast paths. I 
thought that
fast_composite_solid_fill() would be invoked only when it is delegated from
upper level of implementation, that is something like SSE2 or NEON do not
have solid fill fast path.

But if we use more complicated fast path mechanism in the future, I also
think that this modification can cause problems. So your below suggestion
seem to be more proper way of grabbing perfstat data. That is also good for
simple repeat which is basically C fast path but invokes other fast 
paths. So
I will modify this based on your nested begin/end suggestion.

Taekyun Kim
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/pixman/attachments/20110928/5b08c9ec/attachment.html>

More information about the Pixman mailing list