Skip to content

AVX2 MSAC adapt16 SIMD

Lynne requested to merge lynne/dav1d:avx2_msac into master

With #define BENCH_RUNS (1 << 22):

msac_decode_symbol_adapt16_c: 55.7
msac_decode_symbol_adapt16_sse2: 30.1
msac_decode_symbol_adapt16_avx2: 28.0

Overall speed increase varies by frame size but for small 360p videos its around 1%.

Edited by Lynne

Merge request reports

VideoLAN code repository instance