[Ffmpeg-cvslog] r5427 - in trunk: libavcodec/dsputil.c libavcodec/snow.c libavcodec/snow.h tests/ffmpeg.regression.ref tests/rotozoom.regression.ref

[Loren Merritt]

On Tue, 30 May 2006, Oded Shimon wrote:

> On Tue, May 30, 2006 at 07:44:23AM +0200, lorenm wrote:
>> Author: lorenm
>> Date: Tue May 30 07:44:22 2006
>> New Revision: 5427
>>
>> Modified:
>>    trunk/libavcodec/dsputil.c
>>    trunk/libavcodec/snow.c
>>    trunk/libavcodec/snow.h
>>    trunk/tests/ffmpeg.regression.ref
>>    trunk/tests/rotozoom.regression.ref
>>
>> Log:
>> Snow: fix subband weighting in wavelet cmp functions. use 32x32 cmp in iterative motion estimation.
>
> cmp=11 and cmp=12 will actually give good results now?...
> Is this also an encoder speed optimization?

Yes, cmp=w53/w97 are now better than sad/ssd/satd.
It's no faster than before. Still needs a simd dwt.

--Loren Merritt