cdef: Improve cdef_filter_block_4x4_{8,16}.
commitf1ba814554837bcff4b9d24d43b409208fde93cf
authorScott LaVarnway <slavarnway@google.com>
Thu, 23 Dec 2021 17:37:01 +0000 (23 12:37 -0500)
committerScott LaVarnway <slavarnway@google.com>
Tue, 11 Jan 2022 13:41:03 +0000 (11 13:41 +0000)
treec94367626ab2eaf69a9cdba58b33f66002b5c72f
parent6195b8d0d753f0b29da1c5a03e64831356404782
cdef: Improve cdef_filter_block_4x4_{8,16}.

Move shared code into a common function. Use a flag to choose bitdepth
specific code.

For cdef_filter_block_4x4_8(), use v256 until storing the final output. Eliminates
instructions during constrain.

For cdef_filter_block_4x4_16(), reduce the number of instructions used to find max.

Change-Id: I1d37931b27897a9969eb0e6b2f25b0ec9e7ee26e
av1/common/cdef_block_simd.h