How to find out which operators are supported by HIP

hardware-backends AMD ROCm
1

torch version: 1.10
Operating System: Linux

When I use an AMD graphics card, I found torch.frexp is not available. However, according to the PyTorch documentation, HIP is meant to reuse the CUDA interface, which seems to imply that the operators available on CUDA should also be available on HIP (rocm). Now I have three questions:

  1. Is it the design goal of the torch HIP backend to support all the operators supported by CUDA?
  2. How can I find out which operators are supported by HIP?
  3. If torch.frexp is available on CUDA but not on HIP, is it a bug?

What I have done on find operators that suppored by HIP:

  1. Search docs on pytorch.org
  2. Search in aten/src/ATen/native/native_functions.yaml