Since Raven/Picasso APUs and Navi GPUs there is Video Core Next (VCN) as the modern unified video encode/decode block for Radeon graphics. But for those with older Radeon GPUs where there are the Unified Video Decode (UVD) and Video Coding Engine (VCE) blocks, a set of Mesa patches is looking to enhance the video acceleration support on Linux systems.

Open-source developer David Rosca who has been working on many improvements to the AMD Radeon video acceleration support for their Linux drivers opened a Mesa merge request to rework the VCE/UVD functionality.

This newly opened merge request adds a number of features for older AMD GPUs with VCE and UVD to better match functionality that until now was only supported with the newer VCN hardware. The features being enabled for these older AMD video coding IP blocks include:

VCE and UVD:

- App DPB management (long term references, P hierarchy, reference invalidation, ...)

- Slice encoding (128 maximum slices)

- VBAQ

- Quality presets (Speed, Balanced, Quality)

- Min/Max QP

- Max frame size

- Intra refresh

- Raw packed headers

- Encode latency

UVD only:

- Pre-Encode

- Temporal layer rate control

[1]This merge request made up of 32 patches adjusting more than two thousand lines of code provides the new Radeon VCE/UVD functionality for older generations of GPUs. Hopefully these UVD/VCE improvements will be merged in time for the Mesa 25.0 release later this quarter.



[1] https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32829