在Cython中使用2D NumPy数组可以避免使用Python吗？

Question

I'm trying to minimize Python dependency in a while loop for large data series (>200MM values).

Comparing or setting a 1D NumPy array in Cython can be done completely with C, but doing the same with a 2D NumPy array devolves into expensive Python.

Does anyone know how to avoid this?

Minimal code sample:

#!python
#cython: boundscheck=False
#cython: wraparound=False
#cython: cdivision=True

import numpy as np
cimport numpy as np

def sometest():
    cdef np.ndarray[np.float64_t, ndim=1] arr1d = np.zeros((10))
    cdef np.ndarray[np.float64_t, ndim=2] arr2d = np.zeros((10))

    if arr1d[0] > 1:
        arr1d[0] = 1

    if arr2d[0][0] > 1:
        arr2d[0][0] = 1

Cython annotation:

Expanded Cython annotation:

Thank you in advance for any ideas.

Answer 1

Thanks @hpaulj.

Accessing with arr2d[0,0] instead of arr2d[0][0] indeed does not degrade to Python.

Updated code:

#!python
#cython: boundscheck=False
#cython: wraparound=False
#cython: cdivision=True

import numpy as np
cimport numpy as np

def sometest():
    cdef np.ndarray[np.float64_t, ndim=1] arr1d = np.zeros((10))
    cdef np.ndarray[np.float64_t, ndim=2] arr2d = np.zeros((10, 10))

    if arr1d[0] > 1:
        arr1d[0] = 1

    if arr2d[0][0] > 1:
        arr2d[0][0] = 1

    if arr2d[0,0] > 1:
        arr2d[0,0] = 1

Cython annotation: