[Ffmpeg-devel] ADI Blackfin port and optimization

[Michael Benjamin]

On 8/23/06, Michael Niedermayer <michaelni at gmx.at> wrote:
>
> > +    fdctCoef[0] = 0x5a82;  //  C4
> > +    fdctCoef[1] = 0x5a82;  //  C4
> > +    fdctCoef[2] = 0x30Fb;  //cos(3pi/8)  C6
> > +    fdctCoef[3] = 0x7641;  //cos(pi/8)   C2
> > +    fdctCoef[4] = 0x18F8;  //cos(7pi/16)
> > +    fdctCoef[5] = 0x7D8A;  //cos(pi/16)
> > +    fdctCoef[6] = 0x471c;  //cos(5pi/16)
> > +    fdctCoef[7] = 0x6A6d;  //cos(3pi/16)
> > +
> > +    r8x8dct_bfin( block, fdctCoef, computeTemp );
>
> what is computeTemp good for ?


ideally, computeTemp would be in L1 SRAM (in core clock domain) on the
Blackfin (along with DCT/IDCT) for fast execution. right now whole code/data
segments can be put into L1 SRAM, putting specific functions/data with gcc
attributes is not yet implemented.

[...]
> > +    if(idct_algo==FF_IDCT_AUTO||idct_algo==FF_IDCT_BFIN){
> > +        c->idct = simple_idct_bfin;
> > +    }
>
> IDCTs which are selected automatically by default (especially for
> encoding)
> must
> 1. provide a binary identical portable implementation in C


true for DCT too?