Open
Description
Describe the issue:
The following example for calculating numpy.fft.fftn fails.
N-D FFTs are calculated in a loop over axes, and when s is passed the shape of the intermediate result in each step will vary depending on the value of s. In this scenario, out which has the shape of final result cannot be used in intermediate steps.
It is expected that ifftn, rfftn to to fail in a similar scenario but irfftn will works fin since out is not used (because of its dtype) in intermediate steps.
Reproduce the code example:
import numpy
a = numpy.random.random((3,4,5)) + 1j * numpy.random.random((3,4,5))
res_without_out = numpy.fft.fftn(a, axes=(0, 1, 2), s=(7, 8, 9))
out = numpy.empty(res_without_out.shape, dtype=numpy.complex128)
res_with_out = numpy.fft.fftn(a, axes=(0, 1, 2), s=(7, 8, 9), out=out)Error message:
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
Cell In[15], line 1
----> 1 res_with_out = numpy.fft.fftn(a, axes=(0, 1, 2), s=(7, 8, 9), out=out)
File /miniforge/envs/fft/lib/python3.10/site-packages/numpy/fft/_pocketfft.py:878, in fftn(a, s, axes, norm, out)
749 @array_function_dispatch(_fftn_dispatcher)
750 def fftn(a, s=None, axes=None, norm=None, out=None):
751 """
752 Compute the N-dimensional discrete Fourier Transform.
753
(...)
876
877 """
--> 878 return _raw_fftnd(a, s, axes, fft, norm, out=out)
File /miniforge/envs/fft/lib/python3.10/site-packages/numpy/fft/_pocketfft.py:741, in _raw_fftnd(a, s, axes, function, norm, out)
739 itl.reverse()
740 for ii in itl:
--> 741 a = function(a, n=s[ii], axis=axes[ii], norm=norm, out=out)
742 return a
File miniforge/envs/fft/lib/python3.10/site-packages/numpy/fft/_pocketfft.py:209, in fft(a, n, axis, norm, out)
207 if n is None:
208 n = a.shape[axis]
--> 209 output = _raw_fft(a, n, axis, False, True, norm, out)
210 return output
File miniforge/envs/fft/lib/python3.10/site-packages/numpy/fft/_pocketfft.py:94, in _raw_fft(a, n, axis, is_real, is_forward, norm, out)
90 elif ((shape := getattr(out, "shape", None)) is not None
91 and (len(shape) != a.ndim or shape[axis] != n_out)):
92 raise ValueError("output array has wrong shape.")
---> 94 return ufunc(a, fct, axes=[(axis,), (), (axis,)], out=out)
ValueError: operands could not be broadcast together with remapped shapes [original->remapped]: (3,4,5)->(3,4,newaxis) ()->() (7,8,9)->(7,8,9) and requested shape (9)Python and NumPy Versions:
2.2.3
3.10.16 | packaged by conda-forge | (main, Dec 5 2024, 14:16:10) [GCC 13.3.0]
Runtime Environment:
No response
Context for the issue:
No response