[XviD-devel] discussion continue...

[Marc FD]

----- Message d'origine -----
De : "Marco Al" <m.f.al@student.utwente.nl>
À : <xvid-devel@xvid.org>
Envoyé : mardi 10 septembre 2002 20:58
Objet : Re: [XviD-devel] discussion continue...


> Marco Al wrote:
>
> >> i used full butterflyed vertical (2x32) and horizontal(8x8).
> >> i thinked it was slower because i'm not using
> >> only vertical + transpose (faster?)
> >
> > I assume he is suggesting that using butterflies for the columns
> > using vector processing, and full matrix multiply (which can be
> > fully parallized with MMX without transposing) for the rows.
>
> Sorry for my poor reading comprehension and quoting, let me
> rephrase that ... he is suggesting you do exactly what you are
> doing :)
>
> Marco

thx for the link. i start to understand.
but i don't see how "doing exactly what i'm doing" can help me ;)

i think i should take a deep look in is asm code.
i don't understand how it really works (too much macros ;)

MarcFD  marc.fd@libertysurf.fr