Skip to content

prep bilin avx512

---------------------------------------
mct_bilinear_w4_0_8bpc_avx2:      3.9
mct_bilinear_w4_0_8bpc_avx512icl: 3.6
------------------
mct_bilinear_w8_0_8bpc_avx2:      5.1
mct_bilinear_w8_0_8bpc_avx512icl: 4.9
---------------------
mct_bilinear_w16_0_8bpc_avx2:      9.7
mct_bilinear_w16_0_8bpc_avx512icl: 8.3
---------------------
mct_bilinear_w32_0_8bpc_avx2:      29.3
mct_bilinear_w32_0_8bpc_avx512icl: 16.7
---------------------
mct_bilinear_w64_0_8bpc_avx2:      68.2
mct_bilinear_w64_0_8bpc_avx512icl: 37.7
---------------------
mct_bilinear_w128_0_8bpc_avx2:      173.6
mct_bilinear_w128_0_8bpc_avx512icl: 134.6
------------------------------------------
mct_bilinear_w4_h_8bpc_avx2:      4.0
mct_bilinear_w4_h_8bpc_avx512icl: 3.9
---------------------
mct_bilinear_w8_h_8bpc_avx2:      5.4
mct_bilinear_w8_h_8bpc_avx512icl: 5.1
---------------------
mct_bilinear_w16_h_8bpc_avx2:      12.6
mct_bilinear_w16_h_8bpc_avx512icl:  8.8
---------------------
mct_bilinear_w32_h_8bpc_avx2:      42.1
mct_bilinear_w32_h_8bpc_avx512icl: 20.0
---------------------
mct_bilinear_w64_h_8bpc_avx2:      94.9
mct_bilinear_w64_h_8bpc_avx512icl: 35.9
---------------------
mct_bilinear_w128_h_8bpc_avx2:      238.8
mct_bilinear_w128_h_8bpc_avx512icl: 149.6
------------------------------------------
mct_bilinear_w4_v_8bpc_avx2:      4.1
mct_bilinear_w4_v_8bpc_avx512icl: 3.7
---------------------
mct_bilinear_w8_v_8bpc_avx2:      6.2
mct_bilinear_w8_v_8bpc_avx512icl: 5.4
---------------------
mct_bilinear_w16_v_8bpc_avx2:      11.4
mct_bilinear_w16_v_8bpc_avx512icl: 10.1
---------------------
mct_bilinear_w32_v_8bpc_avx2:      29.3
mct_bilinear_w32_v_8bpc_avx512icl: 25.4
---------------------
mct_bilinear_w64_v_8bpc_avx2:      64.0
mct_bilinear_w64_v_8bpc_avx512icl: 41.6
---------------------
mct_bilinear_w128_v_8bpc_avx2:      197.6
mct_bilinear_w128_v_8bpc_avx512icl: 135.0
------------------------------------------
mct_bilinear_w4_hv_8bpc_avx2:      5.6
mct_bilinear_w4_hv_8bpc_avx512icl: 5.2
---------------------
mct_bilinear_w8_hv_8bpc_avx2:      8.0
mct_bilinear_w8_hv_8bpc_avx512icl: 6.9
---------------------
mct_bilinear_w16_hv_8bpc_avx2:      19.2
mct_bilinear_w16_hv_8bpc_avx512icl: 12.0
---------------------
mct_bilinear_w32_hv_8bpc_avx2:      69.2
mct_bilinear_w32_hv_8bpc_avx512icl: 31.7
---------------------
mct_bilinear_w64_hv_8bpc_avx2:      164.5
mct_bilinear_w64_hv_8bpc_avx512icl:  71.1
---------------------
mct_bilinear_w128_hv_8bpc_avx2:      408.4
mct_bilinear_w128_hv_8bpc_avx512icl: 187.9
------------------------------------------
Edited by Victorien Le Couviour--Tuffet

Merge request reports