tree 5cbe63aec9615478abb066d8974d6d4448c5c61b
parent f38717653c29dc5648a1dd7ab8f8639c0cbdcb2e
author Jonathan Wright <jonathan.wright@arm.com> 1681394520 +0100
committer James Zern <jzern@google.com> 1681492911 +0000

Optimize Armv8.0 Neon SAD4D 16xh, 32xh, 64xh and 128xh functions

Add a widening 4D reduction function operating on uint16x8_t vectors
and use it to optimize the final reduction in Armv8.0 Neon standard
bitdepth 32xh SAD4D computations.

In the 16xh, 64xh and 128xh SAD4D functions we use a combination of
widening pair-wise addition instructions before a final 4D reduction
operating on uint32x4_t vectors.

Change-Id: Ide120ecc71f706eef2a603958927edbcef929d4b