[XviD-devel] [In progress] devapi4 -- mpeg matrices

[Edouard Gomez]

skal (skal at planet-d.net) wrote:
> 	oops... sorry for not answering, i'm somehow busy lately...

Not a problem.
 
> > sklmp4, do you mind  if i steal^Wuse it in XviD ?  Of course it would be
> > GPL, so you have to grant us this license change.
> 
> 	Sure, no pb. But beware:
> 
> 	a) i'm not sure it's bug-free (especially the h263 (de)quant)

Unit tests will show if it's  bug free, anyway, i was not thinking about
changing h263 family  of functions. These ones work  well. I just wanted
to use the mpeg4 ones.

> 	b) It might take more time to adapt it to xvid than simply
> 	re-using the idea: if you can't pass 4 pointers, just pass 1
> 	pointers, with all the four matrices packed together
> 	sequentially. (That is: pass "Q[4][64]" instead of "Q1[64], 
> 	Q2[64], Q3[64], Q4[64]"). Another idea: precompute the matrices
> 	for the 31 possible values of Q (even lazily). Memory cost
> 	is not humongous, and it simplifies inner loops...

The problem  isn't really code adaption, the  current MMX implementation
is broken (see  the 2dB impact) we need correct ones.  If yours prove to
be right, then the work is worth it. No matter how long it could take.

> > just a  bit complicated sometimes because
> > of  the C++ readability
> 
> 	my fault: i was too lazy systematically prefixing
> 	the internal attributes with a '_'.
> 

Cette modestie :-)

-- 
Edouard Gomez