Speed of `cupyx.scipy.ndimage.affine_transform` is not consistent

[Soan Thị Minh Dương]

Hi,

I try to warp an image by using `cupyx.scipy.ndimage.affine_transform`. And I realize the the speed of this function is not consistent (the running times are significant different after running the same code). Here is the code that I use to warp the image and measure running time.

```

# Cupy warping
t_start = time.time()
W_cupy = cp.asnumpy(ndimage_cpx.affine_transform(cp.array(M_vol), cp.array(T), output_shape=F_shape, output=float, order=1, cval=np.min(M_vol)))
t_end = time.time()
print('Cupy warping:', t_end - t_start)

```

More information:

- I use cupy 11.4.0

- Fixed/moving vol. shape: (512, 512, 586) /(512, 512, 281)

- Running times after 3 consecutive runs:

Cupy warping: 0.3929164409637451         

        Cupy warping: 21.550039052963257         

        Cupy warping: 0.33153533935546875

Do you know how to get the running time consistency after every runs?

Thank you very much!

Best,

Soan