peterbell10

`overwrite_x` I believe comes from the original `f2py` wrapper code, or at least is consistent with that style. It was kept mainly for backwards-compatibility, I doubt it belongs in the...

@pytorchbot merge -l

@pytorchbot merge

@pytorchbot merge

I don't think any of the matmul family supports multiple dtypes, so I'm not sure this change makes sense on its own.

This is what happens when you try to do a mixed dtype matmul currently: ``` In [1]: import torch ...: a = torch.rand((2, 4), dtype=torch.bfloat16) ...: b = torch.rand((4, 2),...

Yes, by "mixed dtype transforms" I mean multiple dtypes used in the same function call e.g. `sbgemm` does `addmm(float32, bfloat16, bfloat16)`.

This is already implemented in `scipy.fft` and dates back to `scipy.fftpack`. `numpy.fft` does support having `None` values inside the tuple as a placeholder, but I don't think this is documented...

My wording was a bit too strong, it's possible but would require either some significant work to implement `List[Optional[T]]` in PyTorch's C++ code-generation, or require extra python wrapper code that...

@malfet ping