"Clément Bœsch" <ubitux at gmail.com> wrote: >Current dithering only use the first 4w instead of the whole 8 random >values. I think you should benchmark it, I suspect leaving the comparison in the middle instead of having the C version handle more pixels might be faster.