BUG: FFT operation returns C_CONTIGUOUS even if input is F_CONTUGIOUS

[vahidmech]

Describe the issue:

In Numpy_2, FFT operation returns C_CONTIGUOUS even if input is F_CONTIGUOUS while In NumPy 1.26 output was F_CONTIGUOUS for F_CONTIGUOUS input.

Reproduce the code example:

>>> import numpy as np
>>> a = np.ones((10,10), dtype=numpy.complex64, order="F")
>>> numpy.fft.fft(a).flags
  C_CONTIGUOUS : True
  F_CONTIGUOUS : False
  OWNDATA : True
  WRITEABLE : True
  ALIGNED : True
  WRITEBACKIFCOPY : False

Error message:

No response

Python and NumPy Versions:

2.0.0
3.12.4 | packaged by conda-forge | (main, Jun 17 2024, 10:13:44) [Clang 16.0.6 ]

Runtime Environment:

No response

Context for the issue:

No response

[ngoldbaum]

Ping @mhvk this smells related to the pocketfft changes.

[mhvk]

Hmm, I don't think that is a conscious choice, but rather something the gufunc machinery does automatically. I'm not sure how easy it is to set that - @seberg?

[mhvk]

p.s. Am on holidays so will not be able to react very quickly.

[vahidmech]

Hmm, I don't think that is a conscious choice, but rather something the gufunc machinery does automatically. I'm not sure how easy it is to set that - @seberg?

Isn't it because here out array is defined with default order="C"?
https://github.com/numpy/numpy/blob/main/numpy/fft/_pocketfft.py#L88
out = empty(a.shape[:axis] + (n_out,) + a.shape[axis+1:], dtype=out_dtype)

If an appropriate order is passed, it might work fine.

[mhvk]

@vahidmech - indeed, that seems plausible! Maybe that should just become an empty_like, so that only shape and dtype are overridden? Do you want to try and make a quick PR if it works? (I'm on holidays myself so it will be a bit before I get around to it.)