Supporting new devices, where to start

Hi,

where to start if I want to add support for a new parallel computing device to pytorch?

Found it here Extending dispatcher for a new backend in C++ — PyTorch Tutorials 1.10.1+cu102 documentation