tree c4ad39e09175b84f25efbed46b77997023efabf8
parent d192cdfc229d3d4edf6a0acd2e5b71fb4880d28e
author George Steed <george.steed@arm.com> 1681902686 +0100
committer James Zern <jzern@google.com> 1682624126 +0000

Add Neon implementations for remaining v predictors

We already have Neon implementations of the v predictors for all square
block sizes except 64x64, so add the remaining cases and update
tests/speed to match.

On Neoverse V1, these new implementations are worth about a 10%
improvement for the largest and smallest cases on both Clang 15 and GCC
12, and around parity for the remainder with both compilers.

Change-Id: I945f550608e75e978a48b253c2d81e04955c0be5