When global-motion is enabled, a considerable amount
of encoder time is spent in the functions in corner_match.c.
This patch optimizes those functions to be 3.5-4x as fast,
leading to an end-to-end encoder speed improvement
(on 20 frames of tempete_cif.y4m) of:

 200kbps: ~26% faster
 800kbps: ~19% faster
2800kbps: ~12% faster

Change-Id: I04d3f87484c36c41eb5a1e86e814f2accbe86297