Previously, the Accelerator Subsystem of Linux has been in a state of debate. Some kernel developers should develop new subsystems for this computing accelerator, while some developers believe that it should be a DRM subsystem or other subsystems. part.

And at this year’s LPC conference, upstream developers finally reached a consensus on how to deal with the accelerator subsystem: Given that there is a lot in common between various artificial intelligence accelerators and GPUs, this new “accel” core computes the accelerator subsystem. Infrastructure for the Direct Rendering Manager (DRM), but still brought into the Linux kernel as a separate subsystem.

Engineer Oded Gabbay from Intel/Habana Labs has been working on a patch for this new “accel” subsystem, with the fourth iteration released over the weekend. In the v4 announcement email, he confirmed that the v4 patch is the last version of the Linux Kernel Compute Accelerator Subsystem patch ready to be merged in the Linux 6.2 release.

According to foreign media Phoronix, when the 6.2 merge window opens, the Habana Labs AI accelerator driver will move from char/misc to the new acceleration area. In addition, there are many other candidate open source accelerator drivers under development, such as Intel Meteor Lake Multifunction Processing Unit, Toshiba DNN Visconti, NVIDIA NVDLA, Samsung Trinity NPU, Intel GNA, Qualcomm QAIC, etc.