[FFmpeg-devel] [PATCH] VP8 MMX idct_add

[Ronald S. Bultje]

Hi,

as per $subj. Performance goes from approx. 850-900 cycles C to
195-200 cycles with MMX.

Note how I'm pulling in x86util.asm for a 4x4 transpose and sumsub
macro, these are pretty simple but I don't want to duplicate them.
Loren, please consider LGPL'ing these (trivial) parts of x86util.asm
(into a separate file with LGPL-only code), otherwise I'll have to
duplicate them somewhere else in libavcodec/x86/ and screw up
x86util.asm to include this file and de-duplicate it accordingly.

Ronald
-------------- next part --------------
A non-text attachment was scrubbed...
Name: vp8_more_mmx.patch
Type: application/octet-stream
Size: 4439 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20100628/47c52b82/attachment.obj>