[MPlayer-users] VDPAU errors, flickering black video display on aspect change in MPEG2 TS

[Reimar Döffinger]

On Sun, Dec 08, 2013 at 04:30:16PM +0100, wm4 wrote:
> On Sun, 8 Dec 2013 16:27:50 +0100
> Reimar Döffinger <Reimar.Doeffinger at gmx.de> wrote:
> 
> > On Sun, Dec 08, 2013 at 04:22:48PM +0100, wm4 wrote:
> > > On Sun, 8 Dec 2013 16:01:17 +0100
> > > Reimar Döffinger <Reimar.Doeffinger at gmx.de> wrote:
> > > 
> > > > On Sun, Dec 08, 2013 at 03:49:12PM +0100, wm4 wrote:
> > > > > On Sun, 8 Dec 2013 14:52:43 +0100
> > > > > Reimar Döffinger <Reimar.Doeffinger at gmx.de> wrote:
> > > > > 
> > > > > > My best explanation is that this is a NVidia driver bug, there is no
> > > > > > reason the decoder should care about aspect changes.
> > > > > 
> > > > > Of what nature would this bug be? I can see no problems on mpeg2 vdpau
> > > > > decoding with nvidia drivers on non-mplayer. Is it triggered by
> > > > > libavcodec threading? (I think I heard something about threading.)
> > > > 
> > > > The driver claims that there is a size mismatch even though no size
> > > > anywhere changed, only aspect.
> > > > More specifically, reinitializing the decoder _with exactly identical
> > > > settings_ fixes a VDP_STATUS_INVALID_SIZE error, which doesn't make any
> > > > sense.
> > > > You probably see no issues anywhere because everyone else uses the same
> > > > overkill approach of doing a complete reinitialization for no good
> > > > reason on an aspect change.
> > > 
> > > Well, personally I reinitialize the decoder _only_ if the actual size
> > > changes. So I see no bug here.
> > 
> > Well, I would need a sample for this specific issue to make sure,
> > but how do you explain this bug then?
> > In particular, the fix of just creating a decoder with _exactly the same
> > settings_ fixes a VDP_STATUS_INVALID_SIZE, can you come up with any
> > explanation for that, that is not a driver bug?
> 
> Sounds like it could be caused by conflating get_buffer and VO reinit
> (which are really completely different things). Though you said the
> decoder should be properly flushed.

_If_ there had been a size change, the first patch would not have made
any difference at all since we'd never have gone into the "aspect change
only" case. But that is the reason why I said I'd need to
look at a sample, to make sure there really wasn't any size change,
since things are indeed quite non-obvious there.