On Mon, Jun 18, 2012 at 04:34:04PM +0200, Christophe Gisquet wrote:
> 2012/6/18 Kostya Shishkov <[email protected]>:
> >> How high are the weights typically? Same for the LUT values of idx2sym
> >> and sym2idx. num_syms seems to be set to 8 or 2, but there is probably
> >> something I missed in between.
> >
> > There's one model for all pixel values (i.e. 256 symbols).
> > Original decoder can have up to 1023 symbols but I have not found such large
> > model anywhere.
> >
> >> Basically, I am wondering if a smaller type could be used.
> >
> > It can, but I don't think it's worth the effort (even though in practice
> > probably every field can be made int8_t or uint8_t).
> 
> Is the -1 initialization of those LUTS really needed btw? If yes (I'm
> missing how, as negative values from the LUTs do not seem to be
> checked), then probably int16_t is needed.
>  
> How does the original decoder detect/what are the characteristics of a
> bitstream with such values? Some "need samples" warning may be added
> then.

weights[0] need it as a special guard value, others are just for internal
state checking in case something goes wrong.
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to