[XviD-devel] pMB->pmvs []

[Christoph Lampert]

On Tue, 5 Nov 2002, peter ross wrote:

> >From: Christoph Lampert <chl@math.uni-bonn.de>
> >after fiddling with ME another time, I decided I'd like to get rid
> >of the pMB->pmvs[] field in macroblock structure. Instead, the relative
> >MV (pMB->mvs[]-predictor) could be calculated just before MBCoding,
> >when it's neede for the first (and only) time.
> >
> >This means we might have to call get_pmv() another time, but it would save
> >us setting the stupid pmvs[0-3] fields every time we change motion vectors
> >or encoding type.
> >
> >When I finally implement a faster get_pmv is might even save us a tiny
> >little bit of time.
> >
> >What do you think?
> >
> 
> sure; what ever makes things simplers.
> how about calculating the predictor once (for each mb) within motest,
> and then then subtract it from the actual mvs from during mbcoding.

Doesn't help, because predictor may change when neighbours' MV change.

gruel