Merged today to Mesa 26.1-devel is unifying of the AMD video decode implementation between the RadeonSI Gallium3D and RADV Vulkan drivers.
GPU-accelerated video decoding on the AMD open-source driver stack has traditionally been done with the RadeonSI Gallium3D driver using the video acceleration “VA” state tracker for VA-API support. The RADV Vulkan driver has been supporting Vulkan Video and that cross-platform video encode/decode API is beginning to see more adoption by different applications as well as maturing of the implementations by Vulkan drivers. Now thankfully the AMD video decode implementation is being unified between RadeonSI and RADV for video acceleration.
David Rosca of AMD who has been leading many of the open-source Radeon Linux video acceleration improvements devised a shared video decode implementation. This common video decode interface for AMD graphics hardware is used now by both RadeonSI and RADV while covering the Video Core Next (VCN), VCN JPEG, and Unified Video Decode (UVD) engines.
As an immediate initial benefit of this shared implementation is RADV Vulkan Video now being supported on aging Hawaii GPUs and older.
More details on this new AMD shared video decode implementation via this merge now part of Mesa 26.1 for its stable debut in Q2. This unified implementation shifts around more than six thousand lines of code but ends up reducing the line count by around 1.4k lines in the end.
