[libdv-dev] IDCT-248 mmx

[Stefan L. <lu...@sn...>]

Hi,

is there someone working on this ?
I just started with an mmx version of quant_248_inverse.


For encoding:
there is an error in weight_248:
  for (z=0;z<8;z++)
    for (x=0;x<8;x++) {
should be
  for (z=0;z<4;z++)
    for (x=0;x<8;x++) {
to avoid out of bounds addressing of:
      block[z*8+x] *= W[x] * W[2*z] / 2;
      block[(z+4)*8+x] *= W[x] * W[2*z] / 2;

-- 

mfg

  Stefan Lucke (lu...@sn...)