Hello, I am in the process of adding a new hardware backend for MetaSchedule to enable tuning on a hardware target I am also working on. I’m struggling a bit with understanding what I need to provide for this, so I hope someone can help me out. I want to use microTVM so I am using Relay → Meta-Schedle → TIR for the compilation flow and I have a function to do the scheduling. What else do I need to provide here? It seems like I need a builder, and a runner, but I am not sure how to build these components, if there is some pre-existing work in this direction I would be happy for a pointer.
Finally, I also need a way to measure the performance. Does this strictly require an RPC? Because for my use-case that would be difficult to achieve. Ideally, I can just use the performance counters my hardware provides in a simpler manner.
Thanks in advance!