This allows simd optimized routines to work in steps of 8 pixels
without going over the linesize. (this matters for yuv->rgb24 for example)
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>