How to declare 2D c-arrays dynamically in Cython

You just need to stop doing bounds checking:

with cython.boundscheck(False):    thesum += x_view[i,j]

that brings the speed basically up to par.

If you really want a C array from it, try:

import numpy as numpyfrom numpy import int32from numpy cimport int32_tnumpy_array = numpy.array([[]], dtype=int32)cdef:    int32_t[:, :] cython_view = numpy_array    int32_t *c_integers_array = &cython_view[0, 0]    int32_t[4] *c_2d_array = <int32_t[4] *>c_integers_array

First you get a Numpy array. You use that to get a memory view. Then you get a pointer to its data, which you cast to pointers of the desired stride.

So after invaluable help from @Veedrac (Many thanks!) I finally came up with a script that demonstrates the use of both, memory views and c-arrays to speed up calculations in Cython. They both go down to similar speeds and so I personally think using memory-views is MUCH easier.

Here is an example cython script that "accepts" a numpy array and converts it to memory view or c-array and then performs simple array summation via c-level functions:

# cython: boundscheck=Falsecimport cythonimport numpy as npcimport numpy as npfrom numpy import int32from numpy cimport int32_t#Generate numpy array:narr = np.array([[1,2,3,4],[5,6,7,8],[9,10,11,12],[13,14,15,16]], dtype=np.dtype("i"))cdef int a = np.shape(narr)[0]cdef int b = np.shape(narr)[1]cdef int i, jtestsum = np.sum(narr)print "Test summation: np.sum(narr) =", testsum#Generate the memory view:cdef int [:,:] x_view = narr#Generate the 2D c-array and its pointer:cdef:    int32_t[:, :] cython_view = narr    int32_t *c_integers_array = &cython_view[0, 0]    int32_t[4] *c_arr = <int32_t[4] *>c_integers_arraydef test1():    speed_test_mview(x_view)  def test2():    speed_test_carray(&c_arr[0][0], a, b)cdef int speed_test_mview(int[:,:] x_view):    cdef int n, i, j, thesum    # Define the view:    for n in range(10000):        thesum = 0        for i in range(a):            for j in range(b):                thesum += x_view[i, j]        cdef int speed_test_carray(int32_t *c_Arr, int a, int b):    cdef int n, i, j, thesum    for n in range(10000):        thesum = 0        for i in range(a):            for j in range(b):                thesum += c_Arr[(i*b)+j]

Then using ipython shell timing tests reveal similar speeds:

import testlib as tTest summation: np.sum(narr) = 136%timeit t.test1()10000000 loops, best of 3: 46.3 ns per loop%timeit t.test2()10000000 loops, best of 3: 46 ns per loop

Oh and for comparison - using numpy arrays in this example took 125 ms (not shown).