Re[2]: [XviD-devel] I haven't been here for 4 days and what I see? :)

[Radek 'sysKin' Czyz]

> sounds exceptionally good. commit.

Done :D

I also found a correct way for inter/inter4v decision - I simply had
to 'take d_mv_bits seriously' even if 8x8 match was perfect. A simple
bias fixed the problem, inter4v is back to it's normal performance.
And it's commited, too.

I also put a gradiend-descent GME procedure, as a second step to GME.
I oryginally planned it for more then one warping point, but it should
also benefit if there is only one (it does benefit now, that's for
sure).

I was planning the following way for GME to work:
- first, we have to do a simple ME to have the vectors. We do it now if
max_bframes >= 0, not if it's -1. We can also use first-pass MV data
for this. Anyway, I don't think we can have a decent GMC in one-pass
Pframes only encoding... fix me if I'm wrong.
- then, we should find GMC using the vectors. It's done now, I think.
- and then, we should refine GMC vector using real pictures and real
SADs. This is the slow part.
- Finally, we can make normal ME. We should have gmced-SAD already,
ready for SKIP-decisions and mcsel comparsion.

What do you think?
GMC is going to be my next playground :)

Radek