[Unichrome-devel] Re: [mythtv] [PATCH] xvmcvld picture quality

[Ivor Hewitt]

Terry Barnaby wrote:

> Some points from what I can see:
> 1. The Default QMatrix is setup using:
>     s->intra_matrix[s->dsp.idct_permutation[i]] = default_matrix[i];
> 2. The QMatrix is setup from MPEG using:
>     s->intra_matrix[s->dsp.idct_permutation[ff_zigzag_direct[i]]] = 
> default_matrix[i];
>     I assume the MPEG data streams QMATRIX is sent in a zigzag pattern ?
>
> It appears that MythTv sets separate scan tables (inter_scantable ?) 
> for alternate
> scan's for the various IDCT algorithems and that intra_matrix is 
> unchanged.
>
> So I would have assumed we can set the viaXvMC using:
>     qmatrix.intra_quantiser_matrix[i] = 
> s->intra_matrix[s->dsp.idct_permutation[i]];
> This seems to match Ivors last patch.
> Does this make sense ?

I think it still needs to choose between scan tables based on whether 
it's alternate or zig-zag. I've worked up a new patch but I can't try it 
out till tonight.

> Also I note that MythTv can and does set separate chroma qmatrixes if 
> present
> int the MPEG stream. If they are not present it appears to use
> the standard qmatrix's in the MPEG stream for them.
> I also note that the Via hardware decoder, from the datasheet has the 
> ability
> to set these Matrixes. Should we be doing this ??

Absolutely. Although there was only ever code setting the first matrix 
in the ddmpeg code. Wonder if the chipset actually supports it? :-)