tree 69f84f3e1f27a681ca7109150606030f8fe01feb
parent f9fc16b1340a133f61d503f70bb3b660e8b91f81
author Jonathan Wright <jonathan.wright@arm.com> 1656545606 +0100
committer James Zern <jzern@google.com> 1662145170 +0000

Implement sad<w>xhx4d_neon functions using UDOT instruction

Add an alternative implementation of sad128xhx4d_neon,
sad64xhx4d_neon, sad32xhx4d_neon and sad16xhx4d_neon using ADB, UDOT
instead of ADB, UADALP. UDOT is higher throughput and lower latency
than UADALP on modern Arm CPUs and accumulating into 32-bit elements
removes any need for a reduction before the end of the loop.

The existing implementation is retained for CPUs that do not
implement the Armv8.4-A UDOT instruction, or those executing in
AArch32 mode. The availability of the UDOT instruction is indicated
by the feature macro __ARM_FEATURE_DOTPROD.

Change-Id: I7373251cc9071d82cae40a825acbcd64c767c9e7