Cannot get correct FFT results

Hello,

I've been trying to batched FFT for a while and I can't get the correct result (but It works fine with skcuda). Could you tell me what is wrong with the way I am using pyculib. Or equivalently how to copy from pyculib to/from skcuda.

Here is a code example that does the same thing with numba, numpy and skcuda

```python
import numpy as np
from pyculib import fft
from scipy.io import loadmat
from scipy.fftpack import fft as cfft
import numba
from numba import cuda
import pycuda.autoinit
import pycuda.gpuarray as gpuarray
import numpy as np

import skcuda.fft as cu_fft

def test_skcuda():
print('Testing fft/ifft..')
x = INSERT_YOUR_OWN_SIGNAL

batch_size, N = x.shape

#numpy
xf = np.fft.fft(x, axis=1)
y = np.real(np.fft.ifft(xf))

#skcuda
x_gpu = gpuarray.to_gpu(x)
xf_gpu = gpuarray.empty((batch_size, N//2+1), np.complex64)
plan_forward = cu_fft.Plan(N, np.float32, np.complex64, batch_size)
cu_fft.fft(x_gpu, xf_gpu, plan_forward)

#numba
out = np.empty((batch_size, N//2 + 1), dtype=np.complex64)
gpu_temp = numba.cuda.to_device(out)
plan = fft.FFTPlan((N,), np.float32, np.complex64, batch_size)
fft.fft(x, gpu_temp)
result = gpu_temp.copy_to_host()

print(np.allclose(xf[:, :N//2+1], xf_gpu.get(), atol=1e-4)) #True
print(np.allclose(xf[:, :N//2+1], result, atol=1e-4)) #False

```

Run the supplied comparison with a concrete signal and inspect the pyculib.fft.FFTPlan call alongside test_skcuda(). Compare the NumPy, skcuda, and pyculib outputs and determine why the batched result differs. Done means the discrepancy is explained and the example produces the expected comparison.

Written by the indexing model from the issue text.