[FFmpeg-devel] [RFC] Pixel format descriptors and hw-accelerated pixel formats

[Michael Niedermayer]

On Sun, Oct 25, 2009 at 06:41:42PM +0100, Stefano Sabatini wrote:
> On date Sunday 2009-05-24 02:56:05 +0200, Michael Niedermayer encoded:
> > On Sun, May 24, 2009 at 01:23:48AM +0200, Stefano Sabatini wrote:
> > > On date Saturday 2009-04-25 20:32:22 +0200, Stefano Sabatini encoded:
> > > > Hi,
> > > > 
> > > > now that we have all the pixdescs for the non HW-accelerated pixdescs
> > > > I think it's time to think about the HW-accelerated ones, then we can
> > > > get rid of the PixFmtInfo stuff.
> > > > 
> > > > I have no idea what's the better way to deal with them, even if having
> > > > a pixdesc for them is a good idea at all, so suggestions are welcome.
> > > 
> > > Ping.
> > 
> > Neither do i have an idea.
> > 
> > Maybe if you describe a problem one could think of a solution the other
> > way around is hard.
> > Or if there is no problem maybe it does not need a solution and we could
> > spend our time on actual problems?
> 
> I'm trying to understand what the HW-stuff is about.
> 
> Now I see for example I can select in a filterchain:
> -vfilters "format=xvmcmc"
> 
> For example I got this:
> ./ffplay in.avi -loglevel debug -vfilters "format=xvmcmc"
> [...]
> [format @ 0x8da9610]auto-inserting filter 'scale'q=    0B    
> [ffplay_output @ 0x8d9b2a0]auto-inserting filter 'scale'
> swScaler: xvmc_mpeg2_mc is not supported as output pixel format
> 
> This doesn't seem to make much sense, as well as it doesn't look to
> make much sense to have in an hwaccel all the informations about plane,
> step, offset, shift.
> 
> So my question is:
> 1) does it makes sense to fill the pixdesc plane, step, chroma_w/h,
> and components of an AVPixFmtDescriptor of an HW accelerated pixfmt?

i guess it doesnt hurt to fill in what is semantically meaningfull.


> 
> I see in imgconvert the x/y chroma are used at least for VAAPI
> formats.
> 

> So my idea is to add another flag and use that in the definition of
> the HW pixfmts descriptions, see the attached patch with only the XVMC
> definitions filled.

no objection to this



> 
> Apart from this, what is still missing pixdesc to be enabled in SVN?

as internal API, i guess nothing
as public API id like to go over the API once again and it should be asked
if anyone has any suggestions for improvments first


> 
> BTW I also wonder if it would make more sense into moving that to
> libavutil, 

possibly yes

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

In a rich man's house there is no place to spit but his face.
-- Diogenes of Sinope
-------------- 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/ffmpeg-devel/attachments/20091025/1f14a73f/attachment.pgp>