Attempting numpy conversion when not needed in cupy

Question

Using python3.8 with cupy-cuda111

creating a cupy array and trying to cast to cp.int_ causes cupy assume an implicit conversion to numpy array

import cupy as cp

def test_function(x):
    y = cp.int_(x * 10)
    return(y)

x = cp.array([0.1, 0.2, 0.3, 0.4, 0.5])
y = test_function(x)

print(y)

This I assumed would multiply the scalar and return ([1, 2, 3, 4, 5], dtype=cp.int_) but instead gives:

TypeError: Implicit conversion to a NumPy array is not allowed. Please use `.get()` to construct a NumPy array explicitly.

Why does this error get generated?

How can I multiply and cast to int_ the cp.array?

Answer 1

You can do

x = x * 10
y = x.astype(cp.int_)
return y

The problem is cp.int_ is np.int_, so calling that would invoke a host function to operate on a GPU array, which is not allowed.

Answer 2

Now, data is operated in GPU, and you can check data in the GPU or CPU by device function.

Error show that you can convert it to CPU by the get() function, and then, check data's type by the type() function, will find it's numpy.ndarray, but not cupy.core.core.ndarray

Finally:

def test_function(x):
    y = (x * 10).astype(cp.int_)
    return y