I followed this tutorial using this test as an example to integrate my custom backend via PrivateUse1. Everything is working as expected in my PyTorch implementation.
However, I wanted to implement a similar dev environment in C++ in the hopes that I’d be able to set breakpoints and step into my custom kernel implementations for debugging. To my understanding in PyTorch, my custom backend is registered when I call torch.utils.cpp_extension.load method where I pass in my registration and kernel implementation code. What is the process that needs to be followed to register my custom backend to libtorch?