Highlights
https://www.youtube.com/watch?v=IUo0UwZOaRw
- FFmpeg and VLC achieve 62x speedup over C through handwritten assembly that directly leverages SIMD instructions without compiler abstractions, with FFmpeg containing 100k lines of assembly across codecs and David (AV1 decoder) using 240k lines of assembly (79.9%) versus only 30k lines of C (90.6%) for 720p decoding on 1-2 cores.