How to use CUDA const array in Python

howto_cuda_const_array_python.py

import random

from numba import cuda
import numpy as np
import cupy as cp

H_DATA = np.random.uniform(0, 40, 128).astype(np.float32)


@cuda.jit
def kernel(input_data, output_data):
  thread_id = cuda.grid(1)
  if thread_id >= output_data.shape[0]:
    return

  data = cuda.const.array_like(H_DATA)  # const data inside the device. copied from the host during compilation

  input_value = input_data[thread_id]
  data_value = data[thread_id % data.shape[0]]

  output_data[thread_id] = input_value * data_value


def main():
  data_size: int = 1_000_000

  d_input_data = cp.array(np.random.randint(0, 40, size=data_size, dtype=np.uint8))
  d_output_data = cp.array(np.zeros(shape=data_size, dtype=np.float32))

  block_size: int = 256
  grid_size: int = (data_size + (block_size - 1)) // block_size

  kernel[grid_size, block_size](d_input_data, d_output_data)
  cuda.synchronize()

  h_output_data = d_output_data.get()

  print(h_output_data)


if __name__ == '__main__':
  main()