tree c4f25ecab0e048612bee75e5544f6f877d1d3198
parent 6f15d744ea3f34bd331f99f9a5cc6ba2ec73f7bd
author Jonathan Wright <jonathan.wright@arm.com> 1682528185 +0100
committer James Zern <jzern@google.com> 1683228025 +0000

Refactor Neon compound convolution functions 2/7

Refactor and tidy up the Armv8.0 MLA implementation of
av1_dist_wtd_convolve_x_neon:

1) Use a rounding-right-shift-and-accumulate instruction instead of
   separately adding both 'horiz_const' and 'round_offset' constants.
2) Return unsigned types from the convolution kernel inline functions
   - removing the need for a lot of bulky result vector type casting.

Change-Id: Ibb12f9cef1df2522b71add85bd4263c136a36178