[MPlayer-users] Re: [Unichrome-users] Mplayer XvMC failure with Unichrome

[Thomas Hellström]

Hi.

> Thomas Hellstrom writes:
>  > The unichrome XvMC implementation does not support the standard
> "MOCOMP"
>  > and "IDCT" levels of acceleration but only a non-standard "VLD" higher
>  > level of acceleration.
>
> By the way, is this because the CLE266 does not have APIs for
> MC and IDCT, or because there is no motivation to support it ?
>
> In the former case, it might be possible to implement an
> adaptation layer which receives IDCT coefficients and motion
> vectors from the application, encodes them into a slice
> bitstream, and feeds that to the chip.
>

It is because of the former case. The hardware does not support it, And
you are quite right that an option would be to reconstruct a slice
bitstream and  implement at least the IDCT level of XvMC. I don't
currently have the mpeg knowledge to do that, however. Also the decoding
process will not be as efficient. If somebody would like to take a look at
it, however, It would mean automatic but slightly suboptimal CLE266
support in a number of players. Pascal, what do you think? Maybe one could
steal some code from mpeg2 encoders?

It seems that for the CN400 VIA claim mpeg4 _acceleration_ vs mpeg2
_decoding_, so if and when we can extend unichrome XvMC to mpeg4 we might
be looking at the other acceleration levels.


> Of course this is only useful if there are other codecs than
> MPEG1/2 which can take advantage of MC and IDCT acceleration.
>
> -- Pascal
>


/Thomas