[Help-gsl] Fourier Transform with different results

[Goo Creations]

Hi all

I'm trying to re-implement numpy's np.fft.fft2 in C++ by using GSL.
Currently I'm using gsl_fft_complex_radix2_forward for first computing a
row-based and then a column-based transform.

When using a small dataset (eg 4x4) the results from numpy and GSL are
exactly the same. However, the bigger the matrix gets, the greater the
difference in results.
With a 256x256 matrix the results of numpy are totally different (sometimes
even more than twice as big).

Does anyone have an explanation for this? I first thought that Python's and
C++ decimal rounding are different,, but would this have such a huge
different on the result?

Thanks for any help!
Chris