tree 305c276b93a7c69bf2949c851c8eefe0557fe83c
parent 67142110b135d41f35516e1644159fb1f88b384c
author Yunqing Wang <yunqingwang@google.com> 1564763580 -0700
committer Yunqing Wang <yunqingwang@google.com> 1564788269 +0000

Reduce tx_size feature calculation

Rewrote the feature calculation for tx_size split prediction, and
reduced the complexity by eliminating 1 level mean/sse calculation.
get_blk_sse_sum_c can be written to SIMD to provide further speedup.
This CL didn't introduce bitstream change(borg test verified that).

Average speed 1 speedup over midres set is 0.2%(instruction count
result).

Change-Id: I188cd5dfe6d1d1de07ba60394499bc938c38e4b6