Intel-numpy ffts are leaking (as hell)

[MGRAV]

Hi,

I'm not an expert in Python.

When intel-numpy ffts I experiment with (fft2 and fftn), the memory increases until to be full.

I try a small code that generates the same issue:

import numpy
import math
import gc


import numpy.fft as fft

while True :
	test=numpy.fft.fft2(numpy.ones((200,200)));
	del test
	gc.collect()

I am doing something wrong, or it's a bug in the package ?

I run the experimentations on a Ubuntu en a CentOS with Python 3.6

Thanks, in advance for the highlight,

Mathieu

PS. mkl_fft do the same

[Oleksandr_P_Intel]

Hi Mathieu, 

what version of `mkl_fft` and Intel(R) MKL are you using? Did you install it through conda, pip, or built it yourself? 

Thank you,
Oleksandr

[MGRAV]

Hi,

I get the version number with mkl_fft.__version__

On Ubuntu I installed with pip, and I get the version: '1.0.6'

On CentOS with pip, then I remove it and install with yum , and I get the same version  '1.0.6', with the same problem

Mathieu

[Oleksandr_P_Intel]

Mathieu, I have file the issue on your behalf on GitHub, https://github.com/IntelPython/mkl_fft/issues/39

I have identified the leak, and will commit the fix shortly.

[MGRAV]

Thanks a lot,

Mathieu