[cairo] [PATCH v2] Use the new pixman_glyph_cache_t API that will be in pixman 0.28.0

[Chris Wilson]

On Sun, 03 Jun 2012 00:44:22 +0200, sandmann at cs.au.dk (=?utf-8?Q?S=C3=B8ren?= Sandmann) wrote:
> Chris Wilson <chris at chris-wilson.co.uk> writes:
> 
> > On Sat,  2 Jun 2012 09:43:51 -0400, Søren Sandmann <sandmann at cs.au.dk> wrote:
> >> From: Søren Sandmann Pedersen <ssp at redhat.com>
> >> - The slowdown in poppler-reseau is real and stable across runs. I'm
> >>   not too concerned about it because this benchmark is only one run
> >>   and so it is dominated by glyph cache setup costs and FreeType
> >>   rasterizing.
> >
> > Can you please include a couple of runs with
> > cairo-traces/full/firefox-36*? Those traces are taken whilst browsing
> > the international news section of the BBC webseite and though they
> > deliberately thrash the glyph cache should be reasonably interesting.
> 
> Sure,
> 
> [ # ]  backend                         test   min(s) median(s)  stddev. count
> 
> Before:
> [  0]    image          firefox-36-20090609   18.979   19.000   0.09%   4/5
> [  1]    image          firefox-36-20090611   10.165   10.208   0.33%   5/5
> 
> Ater:
> [  0]    image          firefox-36-20090609   19.017   19.028   0.06%   3/5
> [  1]    image          firefox-36-20090611   10.203   10.207   0.21%   5/5
> 
> It's difficult to see any meaningful difference.

I think it helps confirm your suggestion that poppler-reseau is a
pathological case. I'm a bit surprised that it suffers though,
poppler-reseau is more or less a street map with an index -- should be
enough repetitive text there to benefit from a cache, but poppler does
have the habit of only sending single characters at a time.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre