Compute Accelerator Subsystem Hopes To Be Ready For Linux 6.2
At this year's LPC event there was finally a consensus reached among upstream developers how to handle an accelerator subsystem moving forward following the long debate of a new subsystem vs. DRM subsystem vs. the existing char/misc status quo. This new subsystem will leveraging the Direct Rendering Manager (DRM) infrastructure given there being much in common between various AI accelerators and GPUs.
Intel-owned Habana Labs has been on the leading edge of open-source AI accelerator driver enablement.
Oded Gabbay of Intel / Habana Labs has been working on the patches for this new "accel" subsystem and this weekend marked the fourth iteration of them. In the v4 announcement he confirmed that it's hopefully the last version of the patches and that it is believed to be ready for merging.
Oded also confirmed his Linux 6.2 kernel hopes recently on Twitter:
Thanks for the coverage. Indeed, I want to get this merged asap as there are a couple of drivers trying to get upstreamed and are waiting for the new accelerator subsystem. I believe v4 will be a good candidate for merging to drm-next and hopefully for kernel 6.2 merge window https://t.co/XPKATbI2Am— Oded Gabbay (@OdedGabbay) November 8, 2022
So with some luck, the new accelerator subsystem is now ready to be introduced with Linux 6.2. From there the Habana Labs AI accelerator driver will work to move from char/misc to the new accel area. There are also a number of other candidate open-source accelerator drivers in the works like for the Intel Meteor Lake Versatile Processing Unit, Toshiba DNN Visconti, NVIDIA NVDLA, Samsung Trinity NPU, Intel GNA, Qualcomm QAIC, and others.