tree a9c75104e51b28937c5b179a7b82a47f242d7b7c
parent b2eb1abe6e1dda0516d7d5d5992d11689deab88d
author Satish Kumar Suman <satish.suman@ittiam.com> 1539778909 +0530
committer Yunqing Wang <yunqingwang@google.com> 1539878732 +0000

Optimize highbd fwd_txfm modules

Enabled sse4_1 optimizations for tx_sizes 16x64,
64x16,32x64 and 64x32

Module level gains:
Tx_size    Gain w.r.t. C
16x64        4.4x
64x16        4.0x
32x64        3.3x
64x32        3.5x

When tested for 10 frames of crowd_run_1080p_10 at 6 mbps
for speed=1 preset, observed ~1.00% reduction in encoder time.

Change-Id: I5a1a504a9915bf6f81f3df95c8bdd21689636e38