Intel Talks Up Their oneVPL Acceleration Within FFmpeg

Last month Intel began landing oneVPL support within FFmpeg as their video processing and acceleration library that is part of their oneAPI toolkit. The oneVPL Video Processing Library supports CPU-based execution as well as native Intel GPU acceleration for their latest Gen12/Xe hardware with a focus on Arc Graphics / DG2 hardware, targeting the Intel Media SDK for their older GPUs, and can be adapted for other possible back-ends.

This Intel oneVPL support in FFmpeg complements the Video Acceleration API (VA-API) support that has long been within FFmpeg and is also supported by Intel GPUs.

Intel oneVPL with FFmpeg.

For those curious about this FFmpeg oneVPL integration, Intel has now published a new Intel.com developer article outlining the integration for those interested.

While this oneVPL integration is within upstream FFmpeg, one thing I have learned from my Arc Graphics A380 Linux testing is Intel's FFmpeg "Cartwheel" repository. Intel's FFmpeg Cartwheel repo is where they have been queuing up their patches prior to upstreaming. The cartwheel-ffmpeg GitHub repository is Intel's staging area for patches they are working on upstreaming to FFmpeg. There are dozens of patches affecting VA-API, QSV, and more. Dozens of Intel patches still working to be upstreamed. Particularly for DG2/Alchemist, I am told this is where to get the best support until all this work has made it into FFmpeg itself.

I'm still working on some Arc Graphics A380 AV1 encode/decode benchmarks and hope to have those wrapped up in the next week or two.