[x264-devel] Re: x264: svn commit r388 (pengvado)

[Loren Merritt]

On Tue, 20 Dec 2005, Tuukka Toivonen wrote:

> On Tue, 20 Dec 2005, Subversion daemon wrote:
>
>> Successive elimination motion search: same as exhaustive search, but 2-3x faster.
>
> Wow. Something I have been researching once.
>
> Any coding results? I suspect it's not so much better than UMH
> to be really worth it.
>
> Considered MSEA (multilevel SEA [successive elimination algorithm])? In
> theory its much better but I'm not sure how well it translates to MMX etc.

Adding the 2nd level of MSEA does reduce the number of SADs performed. 
But it does not improve overall speed, because 4 SEA checks aren't 
enough faster than a 16x16 MMX SAD.

Changing the order of evaluation from raster to spiral makes negligible 
difference. Probably because we already use lots of mv predictors.

Partial distortion elimination helps, I'll commit that once it's cleaned up.

--Loren Merritt

-- 
This is the x264-devel mailing-list
To unsubscribe, go to: http://developers.videolan.org/lists.html