Unexpected result of 1D FFT on Fortran arrays of array rank > 2

[oleksandr-pavlyk]

FFT transformation on different transpositions of the same array should give the same result, within a reasonbale threshold.

This is currently (v1.0.2) not the case as verified by:

import numpy as np
import mkl_fft

d_ccont = np.random.randn(2, 3, 2)
assert d_ccont.flags['C'] and not d_ccont.flags['F']
d_fcont = np.asfortranarray(d_ccont)
assert d_fcont.flags['F'] and not d_fcont.flags['C']

assert np.allclose(d_ccont, d_fcont) # data as the same

f1 = mkl_fft.fft(d_ccont)
f2 = mkl_fft.fft(d_fcont)

assert np.allclose(f1, f2) # this checks fails to pass

[oleksandr-pavlyk]

As another manifestation of this same bug:

import numpy as np
from mkl_fft._numpy_fft import (rtfft, irfft)

d = np.asfortranarray(np.random.randn(12,16,14))
f = rfft(d)
i = irfft(f)

assert np.allclose(d, i, rtol = 1e-3)

[oleksandr-pavlyk]

Fixed with pull request #12.