[cairo] Tessellator performance patch

Baz brian.ewins at gmail.com
Sun Dec 10 17:50:58 PST 2006


On 12/10/06, Baz <brian.ewins at gmail.com> wrote:
> I got to wondering about why this is placed before the call to
> _slope_compare() as well - to avoid the 64bit arithmetic? Couldn't the
> same lower precision check benefit slope_compare too? ie:

So I tried this (after cleaning up my busted code), it does make a big
difference to some tests (not tried perfdiff yet)

../HEAD.perf:[222]    image-rgba             tessellate-256-100
1252663   67.925   70.072  1.65%   6
../HEAD.perf:[222]    image-rgb              tessellate-256-100
1233750   66.900   69.056  2.44%   6
../TEST.perf:[222]    image-rgba             tessellate-256-100
244636   13.268   17.472 14.14%  95
../TEST.perf:[222]    image-rgb              tessellate-256-100
261724   14.194   18.439 14.77%  94

The changes for the other tests I checked aren't so impressive, under
10% change on the 'zrusin' tests and just noise on the world map. BTW
this is with just the changes to _slope_compare, without adding the
_may_intersect code.

-Baz


More information about the cairo mailing list