[cairo] Re: [poppler] Re: poppler profiling on geode with cairo backend

Tomeu Vizoso tomeu at tomeuvizoso.net
Fri Jan 12 12:04:30 PST 2007


The pdf used to test is here:
http://tomeuvizoso.net/~tomeu/2005.02.02.rss.pdf

and the results of the profiling in the olpc are here:
http://tomeuvizoso.net/~tomeu/fbFetchTransformed.spf.gz

On Fri, 2007-01-12 at 18:54 +0100, Tomeu Vizoso wrote:
> It's much much faster now, thanks!
> 
> Trying a different pdf with big images showed another need for
> performance improvement.
> 
> After sysprofiling evince, pixman_compositeGeneral still accounted for a
> 22% of CPU time. Inside it, fbFetchTransformed consumed itself a 17.7%
> of total CPU time.
> 
> I can send the sysprof file and the pdf used to test it.
> 
> Thanks,
> 
> Tomeu
> 
> On Tue, 2007-01-09 at 15:48 -0800, Carl Worth wrote:
> > On Sat, 06 Jan 2007 13:56:09 +0100, Tomeu Vizoso wrote:
> > > So I guess that the patch for speeding composition using mmx is still
> > > not applied in the latest cairo.
> > >
> > > Can I do anything to speed up its inclusion?
> > 
> > Tomeu discovered that the unaligned-clip was scaling up by 10% with
> > every call to the drawing function (oops!). So now that that's fixed
> > it was easy to drop in Soeren's fix and get a 1.6x speedup for that
> > test case.
> > 
> > So try it out now and let us know if there is still any unexplained
> > slowness.
> > 
> > -Carl



More information about the cairo mailing list