Add aten::polar and its variants

[yucai-intel]

No description provided.

[fengyuan14]

Looks we have better dtype coverage than expected?

[yucai-intel]

The data types are consistent with those supported by CUDA.

[fengyuan14]

The data types are consistent with those supported by CUDA.

The error log is saying unexpected data type is supported in Polar backward. Do yo know which operator is invoked in backward? Can you have an investigation, if the backward op supports bfloat16, does not aligned with CUDA, you can skip the case and add error log and list the backward operator as comments.

[yucai-intel]

Polar's backward is calculated using complex(), which does not support bfloat16, so the test of this datatype is skipped.

[fengyuan14]

o

If we align with CUDA, CUDA should fail as well?

[yucai-intel]

Cuda fails with same error:
RuntimeError: Expected both inputs to be Half, Float or Double tensors but got BFloat16 and BFloat16.