Skip to content

x86: Add high bit-depth cdef_filter AVX-512 (Ice Lake) asm

Henrik Gramner requested to merge gramner/dav1d:cdef16_avx512icl into master

Checkasm numbers with !1422 (merged) applied:

cdef_filter_4x4_01_16bpc_avx2:       61.3
cdef_filter_4x4_01_16bpc_avx512icl:  32.4
cdef_filter_4x4_10_16bpc_avx2:       55.7
cdef_filter_4x4_10_16bpc_avx512icl:  26.8
cdef_filter_4x4_11_16bpc_avx2:       80.5
cdef_filter_4x4_11_16bpc_avx512icl:  47.5

cdef_filter_4x8_01_16bpc_avx2:       81.6
cdef_filter_4x8_01_16bpc_avx512icl:  49.9
cdef_filter_4x8_10_16bpc_avx2:       69.3
cdef_filter_4x8_10_16bpc_avx512icl:  35.9
cdef_filter_4x8_11_16bpc_avx2:      116.9
cdef_filter_4x8_11_16bpc_avx512icl:  80.4

cdef_filter_8x8_01_16bpc_avx2:      113.0
cdef_filter_8x8_01_16bpc_avx512icl:  95.3
cdef_filter_8x8_10_16bpc_avx2:       81.0
cdef_filter_8x8_10_16bpc_avx512icl:  76.3
cdef_filter_8x8_11_16bpc_avx2:      184.0
cdef_filter_8x8_11_16bpc_avx512icl: 148.6
Edited by Henrik Gramner

Merge request reports