[DOC] Questions on Integrating a New CPU Operator into TorchAO？

[Zijie-Tian]

I'm working on integrating a CPU operator into TorchAO and have a few questions regarding the process:

How can I add a New CPU Operator in 'torchao/csrc':

• What is the recommended approach for adding a new CPU operator in the 'csrc' directory?

• Are there any specific guidelines or templates I should follow to ensure compatibility with the existing codebase?

How can I Remove or Disable current CUDA Operators:

• How can I remove or disable all existing CUDA operators in the codebase?

• Are there any configuration flags or build options that can be used to exclude CUDA-related code during compilation?

How can I Move Experimental MPS and CPU Code to TorchAO:

• I noticed that there is experimental code for MPS and CPU in the repository (torchao/experimental/kernels'). What is the process for moving this code into the main TorchAO module?

• Are there any specific considerations or steps I should follow to ensure a smooth transition?

Thank you for your help!

[supriyar]

@metascroy can you help answer these since you have the most context?