[FFmpeg-devel] [PATCH] VP8: idct_mb optimizations

Måns Rullgård mans
Tue Feb 8 23:01:29 CET 2011

"Ronald S. Bultje" <rsbultje at gmail.com> writes:

> Hi,
> On Sun, Feb 6, 2011 at 12:23 AM, Jason Garrett-Glaser <jason at x264.com> wrote:
>> Currently uses AV_RL32 instead of AV_RL32A, as the latter doesn't exist yet.
>> ---
>> ?libavcodec/vp8.c | ? 30 ++++++++++++------------------
>> ?1 files changed, 12 insertions(+), 18 deletions(-)
> Looks interesting to me. How many cycles does this save? Maybe someone
> on BE can confirm that this doesn't make it slower (unlikely, but
> would be nice to confirm)?

The only BE architecture likely to matter is PPC, and it has
little-endian load instructions used in those macros.

M?ns Rullg?rd
mans at mansr.com

More information about the ffmpeg-devel mailing list