FFmpeg Is Ripe For More AVX-512 Optimizations
There is already some AVX-512 optimizations for software contained within the FFmpeg multimedia library, but there is room for greater AVX-512 usage especially now with AMD supporting this Advanced Vector Extensions expansion and the latest Intel Xeon CPUs having little associated cost now with AVX-512 use.
FFmpeg developer Kieran Kunhya presented at FOSDEM last weekend in Brussels around the AVX-512 use within FFmpeg. AVX-512 instructions can be very valuable for bettering the performance especially among multimedia encode/transcode software. Kieran Kunhya looked at existing use within FFmpeg and the dav1d AV1 decoder and went over some of the basics for AVX-512 use.
The hope though is to ultimately see more AVX-512 use within FFmpeg. It's unfortunate that Intel dropped AVX-512 with Alder Lake from their Core processors while at least with AMD Zen 4 there is now AVX-512 support across the desktop/laptop and server product stack -- Zen 4's AVX-512 implementation has proven very efficient both for Ryzen and EPYC use. Intel Sapphire Rapids AVX-512 is also in great shape.
Those wanting to learn more about AVX-512 possibilities for FFmpeg can see Kunhya's FOSDEM 2023 presentation embedded below along with the slide deck.