tree a779f1f699a4dd9d4e8e08c4f1bb7c33a8c7d364
parent 78edddf20df2d90297551fe1d17db9757e437f31
author David Turner <david.turner@argondesign.com> 1538583052 +0100
committer Debargha Mukherjee <debargha@google.com> 1539294848 +0000

SSE4.1 and AVX2 highbd_pixel_proj_error

Add SSE4.1 and AVX2 implementations of the high bit-depth
pixel_proj_error function used in the self-guided filter encoder

SSE4.1 unit test speed-ups
r0=0 r1=0: 5.1x
r0=0 r1=1: 2.7x
r0=1 r1=0: 2.7x
r0=1 r1=1: 2.4x

AVX2 unit test speed-ups
r0=0 r1=0: 6.0x
r0=0 r1=1: 3.5x
r0=1 r1=0: 3.5x
r0=1 r1=1: 2.8x

Encoding 15 frames of ducks_take_off_1080p using AVX2:
cpu=5 bd=10 speed-up=3.6%
cpu=5 bd=12 speed-up=4.0%
cpu=3 bd=10 speed-up=2.4%
cpu=3 bd=12 speed-up=3.2%
cpu=2 bd=10 speed-up=1.0%
cpu=2 bd=12 speed-up=2.1%

Change-Id: I3aac63eeb8d1af33c1d6414dd910708d3dad1793