numba gpu:如何计算两个阵列的最大相对误差？

import numpy as np
from numba import cuda

@cuda.jit
def max_relative_error(arr1, arr2, result):
    # 计算每个元素的相对误差并存储在result数组中
    i = cuda.grid(1)
    if i < arr1.size:
        result[i] = abs(arr1[i] - arr2[i]) / abs(arr1[i])

arr1_gpu = cuda.to_device(arr1)
arr2_gpu = cuda.to_device(arr2)

result_gpu = cuda.device_array_like(arr1)

threads_per_block = 128
blocks_per_grid = (arr1.size + (threads_per_block - 1)) // threads_per_block
max_relative_error[blocks_per_grid, threads_per_block](arr1_gpu, arr2_gpu, result_gpu)

result = result_gpu.copy_to_host()

max_error = np.max(result)