Make calculation of mask_edge free of branches, faster of course but probably

little effect overall as this is not that often executed.

Originally committed as revision 21366 to svn://svn.ffmpeg.org/ffmpeg/trunk