Skip to content

x86: Add cdef_filter_4x4 AVX-512 (Ice Lake) asm

Henrik Gramner requested to merge gramner/dav1d:cdef_filter_4x4_avx512icl into master 
cdef_filter_4x4_8bpc_c: 228.6
cdef_filter_4x4_8bpc_sse2: 51.5
cdef_filter_4x4_8bpc_ssse3: 39.2
cdef_filter_4x4_8bpc_sse4: 35.6
cdef_filter_4x4_8bpc_avx2: 27.8
cdef_filter_4x4_8bpc_avx512icl: 14.5

Does everything, including edge handling, in 8-bit.

Utilizes most of the AVX-512 subsets added in Ice Lake: VNNI, (AI! neural networks! deep learning!), VBMI, VBMI2, BITALG, GFNI (cryptography!)

Merge request reports

VideoLAN code repository instance