[MPlayer-dev-eng] XviD

[Michael Niedermayer]

Hi

On Thursday 07 March 2002 14:23, Arpi wrote:
> Hi,
>
> > On Thursday 07 March 2002 11:19, Christoph Lampert wrote:
> > [...]
> >
> > > 3) We had a version including quartel-pel ME, but it was slow and did
> > > not give much improvement. Maybe it was buggy, anyhow it did not make
> > > it into CVS. When B-frames are in, we'll come back to that. Any other
> > > features new of DivX5 are unknown to me at the moment, if you know
> > > exactly what elements of MPEG4 they are using, please inform everyone.
> >
> > ill post a summary of my findings to the xvid forum, does anyone want me
> > to post the stuff here too?
>
> yes.
ok, so heres the list (ill post it to the xvid forum in a minute)
mostly according to the mpeg4 header ... (only new features)

1. quarter sample mc
2. lots of s(gmc)-vop instead of p-vop
but thes seem to only support a tiny subset of gmc ... nah i didnt expect 
them to do per pixel mc anyway ;)
all s-vops i saw did had 2 sprite-warping points but the second one was 
allways (0,0) so the gmc does only shift the image (if i understand the mpeg4 
equations correctly) but according to the warping accuracy they do need 1/16 
pel mc (iam not sure if this is supposed to be bilinearly interpolated or 
lowpass filtered like 1/4 pel, but imho its not lowpassed)
the bitstream part is allready decoded in ffmpeg cvs ...
2b. a marker bit is missing after the x vector for the sprite trajectory
2c. the mpeg4 std has a typo in the definition of W' and H' it says they are 
of the form 2*alpha and 2*beta but imho they should be 2^alpha & 2^beta
3. b-vop
they seem to have a variable number of b-vops between s-vops (iirc i saw a 
much as 7 b-vops between s-vops

>
> > > 4) XviD is FAST for encoding, much faster than DivX4 and still faster
> > > than DivX5. It's somewhat slower for decoding than ffodivx, true.
> > > But that's no reason to not make it DivX5 compatible, is it?
> >
> > well, i guess both ffmpeg & xvid will be divx5 compatible soon ... well i
> > do hope at least ;)
>
> IMHO xvid should concentrate to encoding, and libavcodec for decoding.
>
> about the nasm stuff... i'm against nasm, as it has many portability
> problems in unix world (name mangling) and sometimes has worse performance
> as the c compiler cannot optimize in/out registers of asm stuff...
> and users don't like to install nasm, or they have nasm compiled with gcc
> 2.96 which is broken (at least SDL compiled with nasm compiled with 2.96
> causes strange sig11 in yv12->rgb converter - old bug see archives).
ok, so no nasm for ffmpeg, i didnt really expect that this would be 
accepted/possible/good idea anyway, just wanted to ask ...

[...]

Michael

PS: i only have 1 divx5 avi at the moment, does anyone know a url with a few 
more samples (i dont use windoz anymore and divx5 encoding doesnt work in 
linux yet afaik)