Those are highly incomplete, especially as they aren't unrolled.
But it already decreases libmpeg2 CPU usage by 20%.