[XviD-devel] SAD vs MSE for ME (warning! acronym overflow!)

Marc FD xvid-devel@xvid.org
Wed, 4 Sep 2002 17:15:16 +0200


----- Message d'origine -----
De : "Christoph Lampert" <chl@math.uni-bonn.de>
À : <xvid-devel@xvid.org>
Envoyé : mercredi 4 septembre 2002 16:53
Objet : Re: [XviD-devel] SAD vs MSE for ME (warning! acronym overflow!)


> On 4 Sep 2002, skal wrote:
>
> >
> > Hi everybody,
> >
> > just wanted to know: is there any reason to prefer
> > SAD as criterion for ME, instead of MSE, e.g.?
> > I'd tend to think SAD was faster to compute than
> > MSE with mults, but is there any other reasons?
>
> No, actually MSE/SSE would be the better choice (at least that's what
> theory says). MAD/SAD is only chosen because it's faster to compute and
> experiment (not mine) showed that it's not much of a difference.
>
> Christoph
>
>
> P.S. If we used SSE assembler for SSE calcalation, would this be in
> file sse_sse.s ? Or should we implement all, make a big file out of it
> and call that   mse_sse_mad_sad-mmx_xxm_sse_3dn_sse.s  ?

I've a sse_mmx.asm just here on my HDD (nasm file with) with a sse.h header
with XviD-like function pointers. I can post it when you want ;)