On 1/9/2023 5:23 AM, Jacek Lawrynowicz wrote:
Hi, This patchset contains a new Linux* Kernel Driver for Intel® VPUs. VPU stands for Versatile Processing Unit and it is an AI inference accelerator integrated with Intel non-server CPUs starting from 14th generation. VPU enables efficient execution of Deep Learning applications like object detection, classification etc. The whole driver is licensed under GPL-2.0-only except for two headers imported from the firmware that are MIT licensed. User mode driver was open sourced in December 2022 and it is available at: https://github.com/intel/linux-vpu-driver
I feel like I forgot to mention this earlier because I can't find a reference to it in my mails. I would like to see something in Documentation for this driver/device. Would be nice to get an overview of how it works (system block diagram?), how it is intended to be used, etc. Include relevant references. This would be a great place to document the UMD and the compiler (I am positive you have mentioned the compiler before, but I am currently failing to find a reference to it).
I feel that traditional DRM gets away from not needing this since their stuff is pretty well established. Everyone uses Mesa/igt and so how things are structured/used is fairly well implied. Accel is brand new and doesn't have that yet so I suspect we'll be well situated if we take the extra effort to spell out these things which might be just assumed elsewhere. Hopefully, over time, such documentation helps in identifying useful areas to build up the common code of the subsystem.
I can't justify holding up this series for this though. Please put it on some todo list :)
-Jeff
