tree f6458ab62e23b7f3c8bbfe8d5a7cadd380d1c059
parent 63adafd2501bb0b7c10fed44c01ca4c767aec834
author David Turner <david.turner@argondesign.com> 1539267473 +0100
committer Debargha Mukherjee <debargha@google.com> 1540234029 +0000

SSE4.1 and AVX2 highbd_blend_a64_d16_mask

Add SSE4.1 and AVX2 SIMD optimised implementations of
highbd_blend_a64_d16_mask.

Unit tests speed-ups is approximately 2x to 2.2x for SSE4.1 and 2.8x to
3.6x for AVX2.

On 15 frames of crowd_run_360 at 10/12bd cpu=0 speed-up is about 3-4%.
At cpu=1 speed-up is about 3%.

I also corrected a comment at the top of blend_a64_mask.c which had not
been updated to reflect the rename of d32 blend functions to d16.

Change-Id: I789932a792c78e1bc64eabbf7143ecb0e67532da