System information

1.15dev

What is the problem that this feature solves?

PyTorch accepts a dynamic list (https://github.com/pytorch/pytorch/blob/e773f28ee307e2a246a4b765f3a51117661b45ba/aten/src/ATen/native/native_functions.yaml#L2914) of axes as input to its _fft_c2c operator. Currently DFT-17 requires axis to be static (https://onnx.ai/onnx/operators/onnx__DFT.html). To fully support _fft_c2c we need axis to be dynamic in DFT.

With this we are also able to express a n-d DFT with functions (with a loop inside).

Alternatives considered

No response

Describe the feature

No response

Will this influence the current api (Y/N)?

N

Feature Area

operators

Are you willing to contribute it (Y/N)

Yes

Notes

No response