replace MULL with asm too, no significnat speedup but its probably better to not take any chances, some versions of gcc will almost certainly mess it up too if they can

Originally committed as revision 6049 to svn://svn.ffmpeg.org/ffmpeg/trunk