lower complex number operations (view_as_real, view_as_complex, conj, abs)

[pitneitemeier]

🚀 Feature

Multiple operations on complex numbers default to cpu (visible by aten:: in metrics report) or just do not work.
Specifically I have found:
defaulting to cpu:

• torch.real(x), x.real
• torch.imag(x) x.imag
• torch.conj(x)

not working:

• torch.abs(x) when x.requires_grad==True
  prints: abs does not support automatic differentiation for outputs with complex dtype

There are probably many more that I have not found and i am happy to add to this list if other people finde more.

This is my First Issue so feel free to tell me if any Information is missing, or additional steps I can take.

[JackCaoG]

Lowering real and imag are pretty easy but lowering them as a view operator is a bit tircky. @bdhirsh Do you think it would make sense to wait for your functionalization change. Real and imag also changes dtype of the tensor (complexfloat -> f32 for example) which makes it a bit more tricky.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[jeffhataws]

Hi @JackCaoG do you know if this is already fixed?

[JackCaoG]

hmm I am not sure, if you use nightly do you see this issue?