tree 4717f66a6107f7496a6f3497d1e66d9c7e03078b
parent e802b78d5adaf1d4686623f5b20624d291a1c96a
author Jonathan Wright <jonathan.wright@arm.com> 1684877539 +0100
committer James Zern <jzern@google.com> 1685485209 +0000

Optimize Neon implementation of aom_int_pro_col

Unroll to operate on 4 rows per iteration. This unrolling also allows
us to use a 4D reduction that uses pairwise-add instructions, rather
than (slower) full-vector reduction instructions.

Change-Id: Ibd31c8c62a5b47c0fbe9d0ec1c241e236846967f