[MPlayer-users] VIA CLE266 MPEG2 decoding

[Ivan Kalvachev]

Dermot McGahon said:
> On Sat, 19 Jun 2004 22:06:22 +0100, Ian Molton <spyro at f2s.com> wrote:
>
>> Hi.
>>
>> Is this possible on a 2.6 kernel? if so is there a howto? I can only
>> find 2.4 stuff.
>
> Good question. Are you referring to this:
>
> http://www.ivor.it/cle266/
>
> Maybe Ivor Hewitt would be the person to ask (I've cc'd him).
>
> With respect to mplayer, my understanding is that the devs were
> approached by VIA over a year ago with an offer of specs, but in
> the end nothing happened (NDA problems?).

We even didn't reach the NDA discusion.
VIA promised to supply an board to the interested developers.
I was interested. VIA said they have to delay it a little because they
are making major rewrite. I waited. When they had done the rewrite they
very kindly refused to supply an board. Saying that they will work with
developers that have already bought it. So far I know at least 2 persons
that were interested in such development. But nobody had send patches
to MPlayer.
After VIA refuse, I bought cheap NVidia card and implemented XvMC.

Now About Unichrome Drivers.
I am not going to support them. Not in their current form.
The problem is that they are not using XvMC, instead they
incorporate an VIA accelearation library with new functions
that start with XvMC string.
The drivers in the current form requare an compleatly different calling
scheme and I am not going to extend the current XvMC code in MPlayer to
support it. The current XvMC is already big hack in both FFmpeg and MPlayer.

Well I have few ideas how to make unichrome an more XvMC like,
but I still haven't put them together and I haven't talked with the
driver developer.
(e.g. use XvMCSlice instead of XvMCMacroBlocks arrays, and use the good old
XvMCRenderSurface, and it may not even need memcpy. One of the good sides
of XvMC is that it doesn't requare to have explict Begin/End Frame handling.)

Anyway, without hardware to test I cannot do any sugnificant development.

Wish You Best
   Ivan Kalvachev
  iive


>
> The xine devs did work with VIA on this and the result was
> vexp (VIA enabled xine player). I'm not sure how well maintained
> vexp is. Reports seem to indicate that it is buggy.
>
> As the webpage above suggest, four parts are needed for this to
> be supported:
>
> (1) V4L support - I need to check if this has been committed to
>      mainline or whether patches are uptodate (or can easily be
>      made uptodate for 2.6.
>
> (2) XFree support - again, has this been submitted to mainline;
>      and before the XOrg fork? The comments suggest that with some
>      changes to V4L that this patch would become unnecessary. Who
>      could make those changes?
>
> (3) libddmpeg.so - should be just a matter of compiling and
>      installing the reverse engineered version (really excellent
>      work Ivor).
>
> (4) MPlayer support. What would need to be done here? A new
>      video codec, decoding mpeg-2 frames using the libddmpeg
>      API? Surely relatively easy to implement compared to a
>      full s/w codec.
>
> A little more info:
>
> http://www.epiawiki.org/wiki/tiki-index.php?page=EpiaMpeg
>
>
> I am also interested in seeing this work and willing to
> contribute code to make that happen. Anybody else interested
> in seeing this and willing to collaborate?
>
>
> Dermot.
> --