A=cuda.to_device(a)
fgriddim, blockdim
b=A.copy_to_host()

cuda{
blockDim:[x],
blockIdx:[x],

threadDim:[x],
threadIdx:[x],

gridDim:[x],

}

from numba import cuda
import numpy as np

@cuda.jit
def addGPU(A,B,C):
    i=cuda.blockIdx.x
    j=cuda.threadIdx.x
    C[i][j]=A[i][j]+B[i][j]
@cuda.jit
def mulGPU(A,B,C,k):
    i=cuda.blockIdx.x
    j=cuda.threadIdx.x
    C[i][j]=0
    for k0 in range(k):
        C[i][j]=C[i][j]+A[i][k0]*B[k0][j]

if __name__ == '__main__':
    
    print("start")
    N=3
    A=np.array([
        [1,0,0],
        [0,2,0],
        [0,0,1]
    ])
    B=np.array([
        [1,0,0],
        [0,1,1],
        [0,1,1]
    ])
    d_A = cuda.to_device(A)
    d_B = cuda.to_device(B)
    d_C = cuda.to_device(np.zeros((N,N)))
    print("A:\n",A,"\nB:\n",B)
    addGPU[N, N](d_A,d_B,d_C)
    C = d_C.copy_to_host()
    print("add:\n",C)
    mulGPU[N, N](d_A,d_B,d_C,3)
    C = d_C.copy_to_host()
    print("mul:\n",C)

参考：https://blog.csdn.net/sazass/article/details/108649728