compiler to optimize some math from mul/div to shr/shl). Also add a cast to
uint32_t when calling decodeplane32(), this silences a compiler warning.
Lastly, in decodeplane8/32(), flatten a double-loop into a single-loop and
calculate the length once before entering the loop instead of during every
iteration (since it doesn't change).

Patch by Sebastian Vater <cdgs.basty googlemail com>.

Originally committed as revision 22974 to svn://svn.ffmpeg.org/ffmpeg/trunk