Skip to content

Blend AVX2

Henrik Gramner requested to merge gramner/dav1d:blend_avx2 into master

Benchmarks results on Skylake-X, 64-bit Linux:

blend_w2_ms0_8bpc_c: 324.9
blend_w2_ms0_8bpc_avx2: 51.6
blend_w2_ms1_8bpc_c: 351.3
blend_w2_ms1_8bpc_avx2: 67.1
blend_w2_ms2_8bpc_c: 324.3
blend_w2_ms2_8bpc_avx2: 55.4
blend_w4_ms0_8bpc_c: 644.5
blend_w4_ms0_8bpc_avx2: 65.7
blend_w4_ms1_8bpc_c: 620.2
blend_w4_ms1_8bpc_avx2: 92.9
blend_w4_ms4_8bpc_c: 570.4
blend_w4_ms4_8bpc_avx2: 75.9
blend_w8_ms0_8bpc_c: 894.6
blend_w8_ms0_8bpc_avx2: 70.5
blend_w8_ms1_8bpc_c: 953.9
blend_w8_ms1_8bpc_avx2: 92.5
blend_w8_ms8_8bpc_c: 893.1
blend_w8_ms8_8bpc_avx2: 96.6
blend_w16_ms0_8bpc_c: 1558.9
blend_w16_ms0_8bpc_avx2: 92.4
blend_w16_ms1_8bpc_c: 1648.3
blend_w16_ms1_8bpc_avx2: 101.8
blend_w16_ms16_8bpc_c: 1558.3
blend_w16_ms16_8bpc_avx2: 106.5
blend_w32_ms0_8bpc_c: 3088.4
blend_w32_ms0_8bpc_avx2: 122.8
blend_w32_ms1_8bpc_c: 3131.0
blend_w32_ms1_8bpc_avx2: 161.5
blend_w32_ms32_8bpc_c: 3143.7
blend_w32_ms32_8bpc_avx2: 188.4
blend_w64_ms0_8bpc_c: 2052.4
blend_w64_ms0_8bpc_avx2: 93.5
blend_w64_ms1_8bpc_c: 1911.1
blend_w64_ms1_8bpc_avx2: 95.6
blend_w64_ms64_8bpc_c: 2052.3
blend_w64_ms64_8bpc_avx2: 117.7
blend_w128_ms0_8bpc_c: 4899.0
blend_w128_ms0_8bpc_avx2: 216.5
blend_w128_ms1_8bpc_c: 4452.8
blend_w128_ms1_8bpc_avx2: 194.4
blend_w128_ms128_8bpc_c: 4901.7
blend_w128_ms128_8bpc_avx2: 262.8

Benchmarks results on Coffee Lake-R, 64-bit Windows:

blend_w2_ms0_8bpc_c: 274.2
blend_w2_ms0_8bpc_avx2: 48.5
blend_w2_ms1_8bpc_c: 289.1
blend_w2_ms1_8bpc_avx2: 64.1
blend_w2_ms2_8bpc_c: 271.5
blend_w2_ms2_8bpc_avx2: 54.6
blend_w4_ms0_8bpc_c: 625.2
blend_w4_ms0_8bpc_avx2: 61.7
blend_w4_ms1_8bpc_c: 633.2
blend_w4_ms1_8bpc_avx2: 89.5
blend_w4_ms4_8bpc_c: 624.6
blend_w4_ms4_8bpc_avx2: 74.2
blend_w8_ms0_8bpc_c: 991.6
blend_w8_ms0_8bpc_avx2: 65.6
blend_w8_ms1_8bpc_c: 999.7
blend_w8_ms1_8bpc_avx2: 85.9
blend_w8_ms8_8bpc_c: 984.5
blend_w8_ms8_8bpc_avx2: 85.5
blend_w16_ms0_8bpc_c: 1502.9
blend_w16_ms0_8bpc_avx2: 67.9
blend_w16_ms1_8bpc_c: 1717.7
blend_w16_ms1_8bpc_avx2: 85.5
blend_w16_ms16_8bpc_c: 1505.2
blend_w16_ms16_8bpc_avx2: 84.1
blend_w32_ms0_8bpc_c: 2984.0
blend_w32_ms0_8bpc_avx2: 102.3
blend_w32_ms1_8bpc_c: 3060.5
blend_w32_ms1_8bpc_avx2: 148.0
blend_w32_ms32_8bpc_c: 2983.0
blend_w32_ms32_8bpc_avx2: 153.4
blend_w64_ms0_8bpc_c: 2043.7
blend_w64_ms0_8bpc_avx2: 74.5
blend_w64_ms1_8bpc_c: 2002.8
blend_w64_ms1_8bpc_avx2: 86.0
blend_w64_ms64_8bpc_c: 2042.1
blend_w64_ms64_8bpc_avx2: 100.1
blend_w128_ms0_8bpc_c: 4914.5
blend_w128_ms0_8bpc_avx2: 168.1
blend_w128_ms1_8bpc_c: 4675.4
blend_w128_ms1_8bpc_avx2: 177.7
blend_w128_ms128_8bpc_c: 4877.8
blend_w128_ms128_8bpc_avx2: 232.7
Edited by Henrik Gramner

Merge request reports