[MPlayer-dev-eng] [PATCH] Support for XvMC VLD on S3 Unichrome

Michael Niedermayer michaelni at gmx.at
Tue Feb 19 04:31:48 CET 2008


On Sat, Feb 16, 2008 at 08:33:33PM +0200, Ivan Kalvachev wrote:
> On Feb 16, 2008 3:43 AM, Timothy Lee <timothy.lee at siriushk.com> wrote:
> > Compn wrote:
> > > On Sat, 5 Jan 2008 10:21:33 +0000 (UTC), Carl Eugen Hoyos wrote:
> > >
> > >> Timothy Lee <timothy.lee <at> siriushk.com> writes:
> > >>
> > >>> It would be great to get XvMC VLD support on the trunk.   Looking
> > >>> forward to your comments!
> > >>>
> > >> Ivan, did you see this mail? Could you comment?
> > >>
> > >> Carl Eugen
> > >>
> > > iive said he is writing a big review for it, please wait
> > >
> > > -compn
> > >
> > How's the review coming along?
> 
> Sorry for the delay I wanted to check few things, but time never was enough.
> 
> Ok, patch is rejected. Same reason as the original one.
> FFmpeg should not call XvMC functions directly.
> 
> The biggest issue seems to be qmatrix and binfo (XvMCQMatrix &
> XvMCMpegControl).  I think that these belong to the xvmc_render
> structure (per surface).  They should be set by the codec and the vo
> would call the XvMC functions they are destined to, on its first run
> of draw_slice() .
> I'm not sure if xvmc_render should include them fully, pointer to
> shared or pointer to malloc-ed, it's up to you to deside:)
> 
> Now about the changes and improvements that could be done:
> (file by file, it may be a little chaotic, I didn't want to post the
> full patch).
> 
> * xvmcvideo.c:
> - I cannot understand what findPastSurface() and findFutureSurface()
> do. They just seem to dublicate what ffmpeg12 already does. The frames
> and surfaces types are well know by the codec. Also they do some
> checks that are kind of strange . E.g. they check if the prediction
> frame is B-picture. The only way this could happen is in interlace
> video when the second decoded field is B, but in this case the other
> field is still valid source for prediction (afair).

I dont think b is ever valid for prediction.


[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

I do not agree with what you have to say, but I'll defend to the death your
right to say it. -- Voltaire
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/attachments/20080219/ef74e595/attachment.pgp>


More information about the MPlayer-dev-eng mailing list