Using Ctypes with double pointers

There is very little on the web describing how to pass data to a C function that has a double pointer as an input variable. The following example passes a 2D Numpy matrix to a C library using ctypes.

Below is a function that has a double pointer as an input variable.

// my_lib.c

#include <stdio.h>

void cFunc (int **x, int xLim, int yLim)
{
    int nx, ny;
    for (nx = 0; nx < xLim; nx++)
        for (ny = 0; ny < yLim; ny++)
            x[nx][ny] = nx+ny*100;
}

To compile with Linux/GCC, run the below commands.

gcc -c -fPIC -O3 my_lib.c
gcc -shared -Wl,-soname,my_lib.so -o my_lib.so my_lib.o

This is a Python function that passes a matrix to the C function.

# dbl_ptr_example.py

import numpy as np
import ctypes as ct

lib = ct.CDLL('./my_lib.so')

toDblPtr = lambda x: \
x.ctypes.data + x.strides[0]*np.arange(x.shape[0], dtype=np.uintp)

def callCLib(matrix):
    [xSz, ySz] = matrix.shape
    lib.cFunc.restype = None
    lib.cFunc.argtypes = \
    [np.ctypeslib.ndpointer(dtype=np.uintp, ndim=1)] +  [ct.c_uint]*2
    lib.cFunc(toDblPtr(matrix), xSz, ySz)
    return matrix

At the Python terminal, run the below commands.

import numpy as np
import dbl_ptr_example as dp_ex

x = np.zeros([5, 7], np.int32, order='C')
y = dp_ex.callCLib(x)

print(y)

The resulting output should look like this:

[[  0 100 200 300 400 500 600]
 [  1 101 201 301 401 501 601]
 [  2 102 202 302 402 502 602]
 [  3 103 203 303 403 503 603]
 [  4 104 204 304 404 504 604]]