python - numba.typeinfer.TypingError : Untyped global name 'child_launch' when using CUDA Dynamic Parallelism in Python ( Anaconda ) on NVIDIA GPU

Question

我的代码在这里:

import numpy as np
from numbapro import cuda

@cuda.autojit
def child_launch(data):
    data[cuda.threadIdx.x] = data[cuda.threadIdx.x] + 100

@cuda.autojit
def parent_launch(data):
    data[cuda.threadIdx.x] = cuda.threadIdx.x
    cuda.syncthreads()

    if cuda.threadIdx.x == 0:
        child_launch[1, 256](data)
        cuda.synchronize()

    cuda.syncthreads()

data = np.zeros(256)
print data
parent_launch[ 1, 256 ](data)
print data

运行结果如下:

Traceback (most recent call last):
......
numba.typeinfer.TypingError: Untyped global name 'child_launch'
File "PythonOpenMPTestMain.py", line 14

我只是不知道为什么我无法启动子内核。请帮我。非常感谢。

Answer 1

I just don't know why I can't launch the child kernel.

因为它不受支持(即使在 2018 年中期)。

引用自 Numba 0.39 documentation

Numba does not implement all features of CUDA, yet. Some missing features are listed below:

• dynamic parallelism
• texture memory