# Threads or processors per block
threads_per_block = 512
# Blocks per grid (or blocks per GPU)
blocks_per_grid = 36
# Send coord1 to GPU
coord1_gpu = cuda.to_device(coord1)
# Send coord2 to GPU
coord2_gpu = cuda.to_device(coord2)
# Create an output on GPU
out_gpu = cuda.device_array(shape=(n,), dtype=np.int32)
# Oof, okay, function[block_def,thread_def](*args)
get_nearby_kernel[blocks_per_grid, threads_per_block](coord1_gpu, coord2_gpu, 1.0, out_gpu)
# Get the result!
out_gpu.copy_to_host()