tree e52ed76cd18fda374ee7f9accfa196239d3bccf5
parent ff3923f48404c91ae3000cf66399f6a18f396ab0
author Jonathan Wright <jonathan.wright@arm.com> 1655248039 +0100
committer James Zern <jzern@google.com> 1662057839 +0000

Optimize implementation of aom_sum_squares_2d_i16_neon

Optimize the implementation of aom_sum_squares_2d_i16_neon by
refactoring the sum-of-squares helper functions. The main change is
to use vector accumulators that don't require some kind of reduction
on every loop iteration.

Also add new 32/64-bit agnostic reduction functions to sum_neon.h and
use them instead of #ifdef blocks in sum_squares_neon.c.

Change-Id: Icd7cab01a57621f567414b3637e5dea10a56e960