September 2025 (FFmpeg/FFmpeg) focused on performance optimization for 8-bit VP9 decoding on AVX2-capable CPUs, plus targeted maintenance to simplify the codebase. Delivered two AVX2-based optimizations for 8-bit VP9 intra prediction and inverse transforms and removed an obsolete 8-bit AVX2 VP9 inverse transform implementation to reduce code size and compilation time. These changes improve decoding throughput on supported hardware, lower maintenance burden, and illustrate robust low-level optimization and refactoring capabilities.

August 2025: FFmpeg/FFmpeg delivered a high-impact performance optimization for VP9 sub-pixel motion compensation using AVX-512ICL. The change introduces AVX-512ICL assembly optimizations for 8-bit-per-pixel sub-pixel interpolation, along with new helper functions/macros and updates to initialization routines to cover multiple sub-pixel scenarios. Expected throughput improvements on AVX-512-capable CPUs for VP9 workloads; commit referenced below. This work strengthens encoding/decoding efficiency and contributes to better streaming performance on modern hardware.

May 2025 — ossrs/ffmpeg-webrtc: Delivered VP9 AVX-512ICL optimization, targeting 16x16 and 32x32 blocks for 8-bit and 10-bit decoding. The change includes new decoding paths and inverse transforms, enabling faster VP9 decode on AVX-512ICL-capable hardware and improving real-time WebRTC throughput.