[Ffmpeg-devel] [RFC] VC-1 I-frames decoder

[Kostya]

On Sat, Jun 17, 2006 at 12:52:34PM +0200, Michael Niedermayer wrote:
> Hi
> 
> On Sat, Jun 17, 2006 at 07:43:18AM +0300, Kostya wrote:
> > Here is my VC-1 decoder which decodes I-frames only. And not all of them are decoded properly too.
> > Nevertheless I want to get opinions on this implementation before moving to P-frames support.
> > P.S. If somebody will find and fix bugs in AC prediction it would be incredibly great.
> 
> hmm, are ac predicted coeffs dequantized in non-coded blocks?

If AC prediction is present then some AC coeffs are copied from previous block and dequantized
no matter was block coded or not.
> 
> 
> the VC9->VC1 cosmetic should be a seperate patch/change
> 
> 
> > +/** @name MV types for B frames */
> > +//@{
> > +#define BMV_TYPE_BACKWARD          0
> > +#define BMV_TYPE_BACKWARD          0
> > +#define BMV_TYPE_FORWARD           1
> > +#define BMV_TYPE_INTERPOLATED      3
> > +//@}
> 
> BMV_TYPE_BACKWARD occurs twice and this and all the other tables could maybe
> be changed to enums, IMHO that would be cleaner, as you could even have
> functions which take enum bmv_type as parameter ...
> 
Copied from old vc9, will correct.
> 
> > +    src0 = block[0]; src1 = block[1];
> > +    for(i = 0; i < 8; i++) {
> > +        for(j = 0; j < 8; j++) {
> > +            Y[j]     = clip_uint8(src0[j]);
> > +            Y[j + 8] = clip_uint8(src1[j]);
> > +        }
> > +        src0 += 8;
> > +        src1 += 8;
> > +        Y += ys;
> > +    }
> 
> DSPContext.put_pixels_clamped()
> 
> 
> vc1_inv_trans() should at some point be moved to DSPContext

That's a quick hack to test. I just want to keep everything in one place for now.
> 
> [...]
> 
> -- 
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
> 
> In the past you could go to a library and read, borrow or copy any book
> Today you'd get arrested for mere telling someone where the library is
>