- It fixes a small bug where a byte value is divided by 255.0 to convert to a float within [0.0, 1.0] and later multiplied by 256.0 to convert back. This makes the luminance lookup table more correct, although the visual difference is relatively small. - speedup of inner loop, using dst[i] instead of *dst++ based on patch by Linards Ticmanis <ticmanis@coli.uni-sb.de>