[Ffmpeg-devel] H.264 encoder

[Michael Niedermayer]

Hi

On Thu, Oct 05, 2006 at 01:39:16AM +0200, Panagiotis Issaris wrote:
[...]
> > > > [...]
> > > >
> > > >> +
> > > >> +// inblock is transposed, outblock isn't
> > > >> +void ff_h264_dct_c(DCTELEM inblock[4][4],DCTELEM outblock[4][4])
> > > >> +{
> > > >> +    DCTELEM pieces[4][4];
> > > >> +
> > > >> +    pieces[0][0] = inblock[0][0]+inblock[1][0]+inblock[2][0]+inblock[3][0];
> > > >> +    pieces[0][1] = inblock[0][1]+inblock[1][1]+inblock[2][1]+inblock[3][1];
> > > >> +    pieces[0][2] = inblock[0][2]+inblock[1][2]+inblock[2][2]+inblock[3][2];
> > > >> +    pieces[0][3] = inblock[0][3]+inblock[1][3]+inblock[2][3]+inblock[3][3];
> > > >> +
> > > >> +    pieces[1][0] = (inblock[0][0]<<1)+inblock[1][0]-inblock[2][0]-(inblock[3][0]<<1);
> > > >> +    pieces[1][1] = (inblock[0][1]<<1)+inblock[1][1]-inblock[2][1]-(inblock[3][1]<<1);
> > > >> +    pieces[1][2] = (inblock[0][2]<<1)+inblock[1][2]-inblock[2][2]-(inblock[3][2]<<1);
> > > >> +    pieces[1][3] = (inblock[0][3]<<1)+inblock[1][3]-inblock[2][3]-(inblock[3][3]<<1);
> > > >> +
> > > >> +    pieces[2][0] = inblock[0][0]-inblock[1][0]-inblock[2][0]+inblock[3][0];
> > > >> +    pieces[2][1] = inblock[0][1]-inblock[1][1]-inblock[2][1]+inblock[3][1];
> > > >> +    pieces[2][2] = inblock[0][2]-inblock[1][2]-inblock[2][2]+inblock[3][2];
> > > >> +    pieces[2][3] = inblock[0][3]-inblock[1][3]-inblock[2][3]+inblock[3][3];
> > > >> +
> > > >> +    pieces[3][0] = inblock[0][0]-(inblock[1][0]<<1)+(inblock[2][0]<<1)-inblock[3][0];
> > > >> +    pieces[3][1] = inblock[0][1]-(inblock[1][1]<<1)+(inblock[2][1]<<1)-inblock[3][1];
> > > >> +    pieces[3][2] = inblock[0][2]-(inblock[1][2]<<1)+(inblock[2][2]<<1)-inblock[3][2];
> > > >> +    pieces[3][3] = inblock[0][3]-(inblock[1][3]<<1)+(inblock[2][3]<<1)-inblock[3][3];
> > > >> +
> > > >> +    outblock[0][0] = pieces[0][0]+pieces[0][1]+pieces[0][2]+pieces[0][3];
> > > >> +    outblock[0][1] = pieces[1][0]+pieces[1][1]+pieces[1][2]+pieces[1][3];
> > > >> +    outblock[0][2] = pieces[2][0]+pieces[2][1]+pieces[2][2]+pieces[2][3];
> > > >> +    outblock[0][3] = pieces[3][0]+pieces[3][1]+pieces[3][2]+pieces[3][3];
> > > >> +
> > > >> +    outblock[1][0] = (pieces[0][0] << 1)+pieces[0][1]-pieces[0][2]-(pieces[0][3]<<1);
> > > >> +    outblock[1][1] = (pieces[1][0] << 1)+pieces[1][1]-pieces[1][2]-(pieces[1][3]<<1);
> > > >> +    outblock[1][2] = (pieces[2][0] << 1)+pieces[2][1]-pieces[2][2]-(pieces[2][3]<<1);
> > > >> +    outblock[1][3] = (pieces[3][0] << 1)+pieces[3][1]-pieces[3][2]-(pieces[3][3]<<1);
> > > >> +
> > > >> +    outblock[2][0] = pieces[0][0]-pieces[0][1]-pieces[0][2]+pieces[0][3];
> > > >> +    outblock[2][1] = pieces[1][0]-pieces[1][1]-pieces[1][2]+pieces[1][3];
> > > >> +    outblock[2][2] = pieces[2][0]-pieces[2][1]-pieces[2][2]+pieces[2][3];
> > > >> +    outblock[2][3] = pieces[3][0]-pieces[3][1]-pieces[3][2]+pieces[3][3];
> > > >> +
> > > >> +    outblock[3][0] = pieces[0][0]-(pieces[0][1]<<1)+(pieces[0][2]<<1)-pieces[0][3];
> > > >> +    outblock[3][1] = pieces[1][0]-(pieces[1][1]<<1)+(pieces[1][2]<<1)-pieces[1][3];
> > > >> +    outblock[3][2] = pieces[2][0]-(pieces[2][1]<<1)+(pieces[2][2]<<1)-pieces[2][3];
> > > >> +    outblock[3][3] = pieces[3][0]-(pieces[3][1]<<1)+(pieces[3][2]<<1)-pieces[3][3];
> > > >> +}
> > > >>
> > > >
> > > > h264.c h264_diff_dct_c() looks so much nicer, is this faster?
> The h264_diff_dct_c() function had been #if 0'ed out. Has it ever been
> tested and confirmed working correctly?
> 
> I'm asking because I'm getting image corruption with this function
> integrated, and knowing if this function is correct can help me find the
> problem.

hmm, if you see corruption then its likely not working correctly, so
remove it or merge it with your or whatever ...


> 
> > [...]
> > > +        if (topavail && w == 16 && h == 16 && srcleft->topblock != 0 && srcleft->topblock->available)
> > > +        {
> > > +            // Plane prediction
> > 
> > use h264.c:pred*x*_plane_c and the other ones from h264.c
> Fixed.

great :)

[...]
-- 
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