[cairo] [pixman] C optimization for x888->0565
vladimir at pobox.com
Wed Sep 10 12:06:27 PDT 2008
Maybe, but I don't think it's critical -- on most processors where MMX/
SSE2 are available, chances are we'll be running with a 24bpp
display. Though I guess that might not be true given Intel's Atom...
Or, I guess, xscale with WiMMX. I mean, it can't hurt, but I wouldn't
make it a priority :)
On Sep 10, 2008, at 11:37 AM, André Tupinambá wrote:
> Hi Vlad,
> Makes sense to create the MMX and SSE2 versions of this fast paths?
> André Tupinambá
> On Sat, Sep 6, 2008 at 9:57 PM, Vladimir Vukicevic
> <vladimir at pobox.com> wrote:
>> The x8r8g8b8 -> r5g6b5 with operator source case was going through
>> composite_general, complete with fbFetch_x8r8g8b8's and
>> Youch. This case happens pretty commonly when an app is using a 24-
>> surface as a backbuffer when it does the final blit to a 16bpp X
>> server as
>> part of expose handling.
>> This can maybe be done faster on arm with assembly (where it'll get
>> the most
>> use), but Jeff can do that better than me if it's necessary --
>> looking at
>> gcc's code, it doesn't look awful.
>> - Vlad
>> cairo mailing list
>> cairo at cairographics.org
More information about the cairo