Numba python3 get error [GPU ufunc requires array arguments to have the exact types.]

Question

I'm trying to use numba to do np.diff on my GPU.
Here is the script I use ;

import numpy as np
import numba

@numba.vectorize(["float32(float32, float32)"], target='cuda')
def vector_diff_axis0(a, b):
    return a + b

def my_diff(A, axis=0):
    if (axis == 0):
        return vector_diff_axis0(A[1:], A[:-1])
    if (axis == 1):
        return vector_diff_axis0(A[:,1:], A[:,:-1])

A = np.matrix([
    [0, 1, 2, 3, 4],
    [5, 6, 7, 8, 9],
    [9, 8, 7, 6, 5],
    [4, 3, 2, 1, 0],
    [0, 2, 4, 6, 8]
    ], dtype='float32')

C = my_diff(A, axis=1)
print (str(C))

And here is the error I get ;

TypeError: No matching version.  GPU ufunc requires array arguments  
to have the exact types.  This behaves like regular ufunc with casting='no'.

Does anybody knows the reason for this?

PS : I used this video to do my script ; https://youtu.be/jKV1m8APttU?t=388

EDIT : Thanks for the fast answers!

I added the dtype='float32' in np.matrix but now I have this error ; Known signatures: * (float32, float32) -> float32 File "", line 5 [1] During: resolving callee type: Function( signature=(float32, float32) -> float32>) [2] During: typing of call at (5)

I also tried to use float32[:] into the signature but it doesn't work and in the video I followed they don't do that

Answer 1

The dtype of your matrix will be int32, which does not match the signature of vector_diff_axis0 since it requires float32. You need to make the matrix float32, i.e. pass the argument dtype='float32' when you call np.matrix.