为什么 dask 在 CUDA 函数上返回 none？

Question

我试图在我的 cuda 函数之上添加 dask，但是当 dask returns 我得到一个 NoneType 对象。

from numba import cuda
import numpy as np
from dask.distributed import Client, LocalCluster


@cuda.jit()
def addingNumbersCUDA (big_array, big_array2, save_array):
    i = cuda.grid(1)
    if i < big_array.shape[0]:
        for j in range (big_array.shape[1]):
            save_array[i][j] = big_array[i][j] * big_array2[i][j]


if __name__ == "__main__":
    cluster = LocalCluster()
    client = Client(cluster)


    big_array = np.random.random_sample((100, 3000))
    big_array2  = np.random.random_sample((100, 3000))
    save_array = np.zeros(shape=(100, 3000))


    arraysize = 100
    threadsperblock = 64
    blockspergrid = (arraysize + (threadsperblock - 1))

    x = client.submit(addingNumbersCUDA[blockspergrid, threadsperblock], big_array, big_array2, save_array)


    y = client.gather(x)
    print(y)

我知道你实际上并没有在 cuda 函数中 return 并且结果被推回你调用的数组。这就是我得到 noneType 的原因，还是因为我对 cuda 使用 dask 错误？

Answer 1

正如 Matthew Rocklin 在这个问题中指出的：，dask 无法处理 in-place 操作。为了解决这个问题，最好添加一个处理 gpu 代码的附加函数。

from numba import cuda
import numpy as np
from dask.distributed import Client, LocalCluster


@cuda.jit()
def addingNumbersCUDA (big_array, big_array2, save_array):
    i = cuda.grid(1)
    if i < big_array.shape[0]:
        for j in range (big_array.shape[1]):
            save_array[i][j] = big_array[i][j] * big_array2[i][j]

def toCUDA (big_array, big_array2, save_array):
    arraysize = 100
    threadsperblock = 64
    blockspergrid = (arraysize + (threadsperblock - 1))

    d_big_array = cuda.to_device(big_array)
    d_big_array2 = cuda.to_device(big_array2)
    d_save_array = cuda.to_device(save_array)

    addingNumbersCUDA[blockspergrid, threadsperblock](d_big_array, d_big_array2, d_save_array)

    save_array = d_save_array.copy_to_host()
    return save_array

if __name__ == "__main__":
    cluster = LocalCluster()
    client = Client(cluster)

    big_array = np.random.random_sample((100, 3000))
    big_array2  = np.random.random_sample((100, 3000))
    save_array = np.zeros(shape=(100, 3000))

    x = client.submit(toCUDA, big_array, big_array2, save_array)


    y = client.gather(x)
    print(y)

为什么 dask 在 CUDA 函数上返回 none？

Why is dask returning none on a CUDA function?

python

dask

dask-distributed