I am trying to speed up my Numpy code and decided that I wanted to implement one particular function where my code spent most of the time in C.
I'm actually a rookie in C, but I managed to write the function which normalizes every row in a matrix to sum to 1. I can compile it and I tested it with some data (in C) and it does what I want. At that point I was very proud of myself.
Now I'm trying to call my glorious function from Python where it should accept a 2d-Numpy array.
The various things I've tried are
SWIG
SWIG + numpy.i
ctypes
My function has the prototype
void normalize_logspace_matrix(size_t nrow, size_t ncol, double mat[nrow][ncol]);
So it takes a pointer to a variable-length array and modifies it in place.
I tried the following pure SWIG interface file:
%module c_utils
%{
extern void normalize_logspace_matrix(size_t, size_t, double mat[*][*]);
%}
extern void normalize_logspace_matrix(size_t, size_t, double** mat);
Then I would do (on Mac OS X 64bit):
> swig -python c-utils.i
> gcc -fPIC c-utils_wrap.c -o c-utils_wrap.o \
-I/Library/Frameworks/Python.framework/Versions/6.2/include/python2.6/ \
-L/Library/Frameworks/Python.framework/Versions/6.2/lib/python2.6/ -c
c-utils_wrap.c: In function ‘_wrap_normalize_logspace_matrix’:
c-utils_wrap.c:2867: warning: passing argument 3 of ‘normalize_logspace_matrix’ from incompatible pointer type
> g++ -dynamiclib c-utils.o -o _c_utils.so
In Python I then get the following error on importing my module:
>>> import c_utils
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ImportError: dynamic module does not define init function (initc_utils)
Next I tried this approach using SWIG + numpy.i:
%module c_utils
%{
#define SWIG_FILE_WITH_INIT
#include "c-utils.h"
%}
%include "numpy.i"
%init %{
import_array();
%}
%apply ( int DIM1, int DIM2, DATA_TYPE* INPLACE_ARRAY2 )
{(size_t nrow, size_t ncol, double* mat)};
%include "c-utils.h"
However, I don't get any further than this:
> swig -python c-utils.i
c-utils.i:13: Warning 453: Can't apply (int DIM1,int DIM2,DATA_TYPE *INPLACE_ARRAY2). No typemaps are defined.
SWIG doesn't seem to find the typemaps defined in numpy.i, but I don't understand why, because numpy.i is in the same directory and SWIG doesn't complain that it can't find it.
With ctypes I didn't get very far, but got lost in the docs pretty quickly since I couldn't figure out how to pass it a 2d-array and then get the result back.
So could somebody show me the magic trick how to make my function available in Python/Numpy?
Unless you have a really good reason not to, you should use cython to interface C and python. (We are starting to use cython instead of raw C inside numpy/scipy themselves).
You can see a simple example in my scikits talkbox (since cython has improved quite a bit since then, I think you could write it better today).
def cslfilter(c_np.ndarray b, c_np.ndarray a, c_np.ndarray x):
"""Fast version of slfilter for a set of frames and filter coefficients.
More precisely, given rank 2 arrays for coefficients and input, this
computes:
for i in range(x.shape[0]):
y[i] = lfilter(b[i], a[i], x[i])
This is mostly useful for processing on a set of windows with variable
filters, e.g. to compute LPC residual from a signal chopped into a set of
windows.
Parameters
----------
b: array
recursive coefficients
a: array
non-recursive coefficients
x: array
signal to filter
Note
----
This is a specialized function, and does not handle other types than
double, nor initial conditions."""
cdef int na, nb, nfr, i, nx
cdef double *raw_x, *raw_a, *raw_b, *raw_y
cdef c_np.ndarray[double, ndim=2] tb
cdef c_np.ndarray[double, ndim=2] ta
cdef c_np.ndarray[double, ndim=2] tx
cdef c_np.ndarray[double, ndim=2] ty
dt = np.common_type(a, b, x)
if not dt == np.float64:
raise ValueError("Only float64 supported for now")
if not x.ndim == 2:
raise ValueError("Only input of rank 2 support")
if not b.ndim == 2:
raise ValueError("Only b of rank 2 support")
if not a.ndim == 2:
raise ValueError("Only a of rank 2 support")
nfr = a.shape[0]
if not nfr == b.shape[0]:
raise ValueError("Number of filters should be the same")
if not nfr == x.shape[0]:
raise ValueError, \
"Number of filters and number of frames should be the same"
tx = np.ascontiguousarray(x, dtype=dt)
ty = np.ones((x.shape[0], x.shape[1]), dt)
na = a.shape[1]
nb = b.shape[1]
nx = x.shape[1]
ta = np.ascontiguousarray(np.copy(a), dtype=dt)
tb = np.ascontiguousarray(np.copy(b), dtype=dt)
raw_x = <double*>tx.data
raw_b = <double*>tb.data
raw_a = <double*>ta.data
raw_y = <double*>ty.data
for i in range(nfr):
filter_double(raw_b, nb, raw_a, na, raw_x, nx, raw_y)
raw_b += nb
raw_a += na
raw_x += nx
raw_y += nx
return ty
As you can see, besides the usual argument checking you would do in python, it is almost the same thing (filter_double is a function which can be written in pure C in a separate library if you want to). Of course, since it is compiled code, failing to check your argument will crash your interpreter instead of raising exception (there are several levels of safety vs speed tradeoffs available with recent cython, though).
To answer the real question: SWIG doesn't tell you it can't find any typemaps. It tells you it can't apply the typemap (int DIM1,int DIM2,DATA_TYPE *INPLACE_ARRAY2), which is because there is no typemap defined for DATA_TYPE *. You need to tell it you want to apply it to a double*:
%apply ( int DIM1, int DIM2, double* INPLACE_ARRAY2 )
{(size_t nrow, size_t ncol, double* mat)};
First, are you sure that you were writing the fastest possible numpy code? If by normalise you mean divide the whole row by its sum, then you can write fast vectorised code which looks something like this:
matrix /= matrix.sum(axis=0)
If this is not what you had in mind and you are still sure that you need a fast C extension, I would strongly recommend you write it in cython instead of C. This will save you all the overhead and difficulties in wrapping code, and allow you to write something which looks like python code but which can be made to run as fast as C in most circumstances.
I agree with others that a little Cython is well worth learning.
But if you must write C or C++, use a 1d array which overlays the 2d, like this:
// sum1rows.cpp: 2d A as 1d A1
// Unfortunately
// void f( int m, int n, double a[m][n] ) { ... }
// is valid c but not c++ .
// See also
// http://stackoverflow.com/questions/3959457/high-performance-c-multi-dimensional-arrays
// http://stackoverflow.com/questions/tagged/multidimensional-array c++
#include <stdio.h>
void sum1( int n, double x[] ) // x /= sum(x)
{
float sum = 0;
for( int j = 0; j < n; j ++ )
sum += x[j];
for( int j = 0; j < n; j ++ )
x[j] /= sum;
}
void sum1rows( int nrow, int ncol, double A1[] ) // 1d A1 == 2d A[nrow][ncol]
{
for( int j = 0; j < nrow*ncol; j += ncol )
sum1( ncol, &A1[j] );
}
int main( int argc, char** argv )
{
int nrow = 100, ncol = 10;
double A[nrow][ncol];
for( int j = 0; j < nrow; j ++ )
for( int k = 0; k < ncol; k ++ )
A[j][k] = (j+1) * k;
double* A1 = &A[0][0]; // A as 1d array -- bad practice
sum1rows( nrow, ncol, A1 );
for( int j = 0; j < 2; j ++ ){
for( int k = 0; k < ncol; k ++ ){
printf( "%.2g ", A[j][k] );
}
printf( "\n" );
}
}
Added 8 Nov: as you probably know, numpy.reshape can overlay a numpy 2d array with a 1d view to pass to sum1rows, like this:
import numpy as np
A = np.arange(10).reshape((2,5))
A1 = A.reshape(A.size) # a 1d view of A, not a copy
# sum1rows( 2, 5, A1 )
A[1,1] += 10
print "A:", A
print "A1:", A1
SciPy has an extension tutorial with example code for arrays.
http://docs.scipy.org/doc/numpy/user/c-info.how-to-extend.html
Related
Currently I'm learning about C types. My goal is to generate an numpy
array A in python from 0 to 4*pi in 500 steps. That array is passed to
C code which calculates the tangent of those values. The C code also
passes those values back to an numpy array B in python.
Yesterday I tried simply to convert one value from python to C and
(after some help) succeeded. Today I try to pass a whole array, not a
value.
I think it's an good idea to add another function to the C library to
process the array. The new function should in a loop pass each value
of A to the function tan1() and store that value in array B.
I have two issues:
writing the function that processes the numpy array A
Passing the numpy array between python and C code.
I read the following info:
https://nenadmarkus.com/p/numpy-to-native/
How to use NumPy array with ctypes?
Helpful, but I still don't know how to solve my problem.
C code (Only the piece that seems relevant):
double tan1(f) double f;
{
return sin1(f)/cos1(f);
}
void loop(double A, int n);
{
double *B;
B = (double*) malloc(n * sizeof(double));
for(i=0; i<= n, i++)
{
B[i] = tan1(A[i])
}
}
Python code:
import numpy as np
import ctypes
A = np.array(np.linspace(0,4*np.pi,500), dtype=np.float64)
testlib = ctypes.CDLL('./testlib.so')
testlib.loop.argtypes = ctypes.c_double,
testlib.loop.restype = ctypes.c_double
#print(testlib.tan1(3))
I'm aware that ctypes.c_double is wrong in this context, but that is what I had in the 1 value version and don't know yet for what to substitute.
Could I please get some feedback on how to achieve this goal?
You need to return the dynamically allocated memory, e.g. change your C code to something like:
#include <math.h>
#include <stdlib.h>
#include <stdio.h>
double tan1(double f) {
return sin(f)/cos(f);
}
double *loop(double *arr, int n) {
double *b = malloc(n * sizeof(double));
for(int i = 0; i < n; i++) {
b[i] = tan(arr[i]);
}
return b;
}
void freeArray(double *b) {
free(b);
}
On the Python side you have to declare parameter and return types. As mentioned by others in comments, you should also free dynamically allocated memory. Note that on the C side, arrays always decay into pointers. Therefore, you need an additional parameter which tells you the number of elements in the array.
Also if you return a pointer to double to the Python page, you must specify the size of the array. With np.frombuffer you can work with the data without making a copy of it.
import numpy as np
from ctypes import *
testlib = ctypes.CDLL('./testlib.so')
n = 500
dtype = np.float64
input_array = np.array(np.linspace(0, 4 * np.pi, n), dtype=dtype)
input_ptr = input_array.ctypes.data_as(POINTER(c_double))
testlib.loop.argtypes = (POINTER(c_double), c_int)
testlib.loop.restype = POINTER(c_double * n)
testlib.freeArray.argtypes = POINTER(c_double * n),
result_ptr = testlib.loop(input_ptr, n)
result_array = np.frombuffer(result_ptr.contents)
# ...do some processing
for value in result_array:
print(value)
# free buffer
testlib.freeArray(result_ptr)
I am working on a project involving object detection through deep learning, with the underlying detection code written in C. Due to the requirements of the project, this code has a Python wrapper around it, which interfaces with the required C functions through ctypes. Images are read from Python, and then transferred into C to be processed as a batch.
In its current state, the code is very unoptimized: the images (640x360x3 each) are read using cv2.imread then stacked into a numpy array. For example, for a batch size of 16, the dimensions of this array are (16,360,640,3). Once this is done, a pointer to this array is passed through ctypes into C where the array is parsed, pixel values are normalized and rearranged into a 2D array. The dimensions of the 2D array are 16x691200 (16x(640*360*3)), arranged as follows.
row [0]: Image 0: (B)r0(B)r1(B)r2.... (G)r0(G)r1(G)r2.... (R)r0(R)r1(R)r2....
row [1]: Image 1: (B)r0(B)r1(B)r2.... (G)r0(G)r1(G)r2.... (R)r0(R)r1(R)r2....
.
.
row [15]: Image 15: (B)r0(B)r1(B)r2.... (G)r0(G)r1(G)r2.... (R)r0(R)r1(R)r2....
`
The C code for doing this currently looks like this, where the pixel values are accessed through strides and arranged sequentially per image. nb is the total number of images in the batch (usually 16); h, w, c are 360,640 and 3 respectively.
matrix ndarray_to_matrix(unsigned char* src, long* shape, long* strides)
{
int nb = shape[0];
int h = shape[1];
int w = shape[2];
int c = shape[3];
matrix X = make_matrix(nb, h*w*c);
int step_b = strides[0];
int step_h = strides[1];
int step_w = strides[2];
int step_c = strides[3];
int b, i, j, k;
int index1, index2 = 0;
for(b = 0; b < nb ; ++b) {
for(i = 0; i < h; ++i) {
for(k= 0; k < c; ++k) {
for(j = 0; j < w; ++j) {
index1 = k*w*h + i*w + j;
index2 = step_b*b + step_h*i + step_w*j + step_c*k;
X.vals[b][index1] = src[index2]/255.;
}
}
}
}
return X;
}
And the corresponding Python code that calls this function: (array is the original numpy array)
for i in range(start, end):
imgName = imgDir + '/' + allImageName[i]
img = cv2.imread(imgName, 1)
batchImageData[i-start,:,:] = img[:,:]
data = batchImageData.ctypes.data_as(POINTER(c_ubyte))
resmatrix = self.ndarray_to_matrix(data, batchImageData.ctypes.shape, batchImageData.ctypes.strides)
As of now, this ctypes implementation takes about 35 ms for a batch of 16 images. I'm working on a very FPS critical image processing pipeline, so is there a more efficient way of doing these operations? Specifically:
Can I read the image directly as a 'strided' one dimensional array in Python from disk, thus avoiding the iterative access and copying?
I have looked into numpy operations such as:
np.ascontiguousarray(img.transpose(2,0,1).flat, dtype=float)/255. which should achieve something similar, but this is actually taking more time possibly because of it being called in Python.
Would Cython help anywhere during the read operation?
Regarding the ascontiguousarray method, I'm assuming that it's pretty slow as python has to do some memory works to return a C-like contiguous array.
EDIT 1:
I saw this answer, apparently openCV's imread function should already return a contiguous array.
I am not very familiar with ctypes, but happen to use the PyBind library and can only recommend using it. It implements Python's buffer protocol hence allowing you to interact with python data with almost no overhead.
I've answered a question explaining how to pass a numpy array from Python to C/C++, do something dummy to it in C++ and return a dynamically created array back to Python.
EDIT 2 : I've added a simple example that receives a Numpy array, send it to C and prints it from C. You can find it here. Hope it helps!
EDIT 3 :
To answer your last comment, yes you can definitely do that.
You could modify your code to (1) instantiate a 2D numpy array in C++, (2) pass its reference to the data to your C function that will modify it instead of declaring a Matrix and (3) return that instance to Python by reference.
Your function would become:
void ndarray_to_matrix(unsigned char* src, double * x, long* shape, long* strides)
{
int nb = shape[0];
int h = shape[1];
int w = shape[2];
int c = shape[3];
int step_b = strides[0];
int step_h = strides[1];
int step_w = strides[2];
int step_c = strides[3];
int b, i, j, k;
int index1, index2 = 0;
for(b = 0; b < nb ; ++b) {
for(i = 0; i < h; ++i) {
for(k= 0; k < c; ++k) {
for(j = 0; j < w; ++j) {
index1 = k*w*h + i*w + j;
index2 = step_b*b + step_h*i + step_w*j + step_c*k;
X.vals[b][index1] = src[index2]/255.;
}
}
}
}
}
And you'd add, in your C++ wrapper code
// Instantiate the output array, assuming we know b, h, c,w
py::array_t<double> x = py::array_t<double>(b*h*c*w);
py::buffer_info bufx = x.request();
double*ptrx = (double *) bufx.ptr;
// Call to your C function with ptrx as input
ndarray_to_matrix(src, ptrx, shape, strides);
// now reshape x
x.reshape({b, h*c*w});
Do not forget to modify the prototype of the C++ wrapper function to return a numpy array like:
py::array_t<double> read_matrix(...){}...
This should work, I didn't test it though :)
I have some code writen in Python for which the output is a numpy array, and now I want to send that output to C++ code, where the heavy part of the calculations will be performed.
I have tried using cython's public cdef, but I am running on some issues. I would appreciate your help! Here goes my code:
pymodule.pyx:
from pythonmodule import result # result is my numpy array
import numpy as np
cimport numpy as np
cimport cython
#cython.boundscheck(False)
#cython.wraparound(False)
cdef public void cfunc():
print 'I am in here!!!'
cdef np.ndarray[np.float64_t, ndim=2, mode='c'] res = result
print res
Once this is cythonized, I call:
pymain.c:
#include <Python.h>
#include <numpy/arrayobject.h>
#include "pymodule.h"
int main() {
Py_Initialize();
initpymodule();
test(2);
Py_Finalize();
}
int test(int a)
{
Py_Initialize();
initpymodule();
cfunc();
return 0;
}
I am getting a NameError for the result variable at C++. I have tried defining it with pointers and calling it indirectly from other functions, but the array remains invisible. I am pretty sure the answer is quite simple, but I just do not get it. Thanks for your help!
Short Answer
The NameError was cause by the fact that Python couldn't find the module, the working directory isn't automatically added to your PYTHONPATH. Using setenv with setenv("PYTHONPATH", ".", 1); in your C/C++ code fixes this.
Longer Answer
There's an easy way to do this, apparently. With a python module pythonmodule.py containing an already created array:
import numpy as np
result = np.arange(20, dtype=np.float).reshape((2, 10))
You can structure your pymodule.pyx to export that array by using the public keyword. By adding some auxiliary functions, you'll generally won't need to touch neither the Python, nor the Numpy C-API:
from pythonmodule import result
from libc.stdlib cimport malloc
import numpy as np
cimport numpy as np
cdef public np.ndarray getNPArray():
""" Return array from pythonmodule. """
return <np.ndarray>result
cdef public int getShape(np.ndarray arr, int shape):
""" Return Shape of the Array based on shape par value. """
return <int>arr.shape[1] if shape else <int>arr.shape[0]
cdef public void copyData(float *** dst, np.ndarray src):
""" Copy data from src numpy array to dst. """
cdef float **tmp
cdef int i, j, m = src.shape[0], n=src.shape[1];
# Allocate initial pointer
tmp = <float **>malloc(m * sizeof(float *))
if not tmp:
raise MemoryError()
# Allocate rows
for j in range(m):
tmp[j] = <float *>malloc(n * sizeof(float))
if not tmp[j]:
raise MemoryError()
# Copy numpy Array
for i in range(m):
for j in range(n):
tmp[i][j] = src[i, j]
# Assign pointer to dst
dst[0] = tmp
Function getNPArray and getShape return the array and its shape, respectively. copyData was added in order to just extract the ndarray.data and copy it so you can then finalize Python and work without having the interpreter initialized.
A sample program (in C, C++ should look identical) would look like this:
#include <Python.h>
#include "numpy/arrayobject.h"
#include "pyxmod.h"
#include <stdio.h>
void printArray(float **arr, int m, int n);
void getArray(float ***arr, int * m, int * n);
int main(int argc, char **argv){
// Holds data and shapes.
float **data = NULL;
int m, n;
// Gets array and then prints it.
getArray(&data, &m, &n);
printArray(data, m, n);
return 0;
}
void getArray(float ***data, int * m, int * n){
// setenv is important, makes python find
// modules in working directory
setenv("PYTHONPATH", ".", 1);
// Initialize interpreter and module
Py_Initialize();
initpyxmod();
// Use Cython functions.
PyArrayObject *arr = getNPArray();
*m = getShape(arr, 0);
*n = getShape(arr, 1);
copyData(data, arr);
if (data == NULL){ //really redundant.
fprintf(stderr, "Data is NULL\n");
return ;
}
Py_DECREF(arr);
Py_Finalize();
}
void printArray(float **arr, int m, int n){
int i, j;
for(i=0; i < m; i++){
for(j=0; j < n; j++)
printf("%f ", arr[i][j]);
printf("\n");
}
}
Always remember to set:
setenv("PYTHONPATH", ".", 1);
before you call Py_Initialize so Python can find modules in the working directory.
The rest is pretty straight-forward. It might need some additional error-checking and definitely needs a function to free the allocated memmory.
Alternate Way w/o Cython:
Doing it the way you are attempting is way hassle than it's worth, you would probably be better off using numpy.save to save your array in a npy binary file and then use some C++ library that reads that file for you.
I have written a good bit of code in python and it works great. But now I'm scaling up the size of the problems that I'm analyzing and python is dreadfully slow. The slow part of the python code is
for i in range(0,H,1):
x1 = i - length
x2 = i + length
for j in range(0,W,1):
#print i, ',', j # check the limits
y1 = j - length
y2 = j + length
IntRed[i,j] = np.mean(RawRed[x1:x2,y1:y2])
With H and W equal to 1024 the function takes around 5 minutes to excute. I've written a simple c++ program/function that performs the same computation and it excutes in less than a second with the same data size.
double summ = 0;
double total_num = 0;
double tmp_num = 0 ;
int avesize = 2;
for( i = 0+avesize; i <X-avesize ;i++)
for(j = 0+avesize;j<Y-avesize;j++)
{
// loop through sub region of the matrix
// if the value is not zero add it to the sum
// and increment the counter.
for( int ii = -2; ii < 2; ii ++)
{
int iii = i + ii;
for( int jj = -2; jj < 2 ; jj ++ )
{
int jjj = j + jj;
tmp_num = gsl_matrix_get(m,iii,jjj);
if(tmp_num != 0 )
{
summ = summ + tmp_num;
total_num++;
}
}
}
gsl_matrix_set(Matrix_mean,i,j,summ/total_num);
summ = 0;
total_num = 0;
}
I have some other methods to perform on the 2D array. The one listed is a simple examples.
What I want to do is pass a python 2D array to my c++ function and return a 2D array back to python.
I've read a bit about swig, and have sereached pervious questions, and it seems like it's a possible solution. But I can't seem to figure out what I actually need to do.
Can I get any help? Thanks
You can use arrays as it is described here: Doc - 5.4.5 Arrays, the carray.i or std_vector.i from the SWIG library.
I find it easier to work with std::vector from the SWIG library std_vector.i to send a python list to a C++ SWIG extension. Though in your case where optimization matters, it may not be the optimal.
In your case you can define:
test.i
%module test
%{
#include "test.h"
%}
%include "std_vector.i"
namespace std {
%template(Line) vector < int >;
%template(Array) vector < vector < int> >;
}
void print_array(std::vector< std::vector < int > > myarray);
test.h
#ifndef TEST_H__
#define TEST_H__
#include <stdio.h>
#include <vector>
void print_array(std::vector< std::vector < int > > myarray);
#endif /* TEST_H__ */
test.cpp
#include "test.h"
void print_array(std::vector< std::vector < int > > myarray)
{
for (int i=0; i<2; i++)
for (int j=0; j<2; j++)
printf("[%d][%d] = [%d]\n", i, j, myarray[i][j]);
}
If you run the following python code (I used python 2.6.5), you can see that the C++ function can access the python list:
>>> import test
>>> a = test.Array()
>>> a = [[0, 1], [2, 3]]
>>> test.print_array(a)
[0][0] = [0]
[0][1] = [1]
[1][0] = [2]
[1][1] = [3]
How does one properly initialize and return a Cython array? For instance:
cdef public double* cyTest(double[] input):
cdef double output[3]
for i in xrange(3):
output[i] = input[i]**2
print 'loop: ' + str(output[i])
return output
cdef double* test = [1,2,3]
cdef double* results = cyTest(test)
for i in xrange(3):
print 'return: ' + str(results[i])
This returns:
loop: 1.0->1.0
loop: 2.0->4.0
loop: 3.0->9.0
return: 1.88706086937e-299
return: 9.7051011575e+236
return: 1.88706086795e-299
So obviously, results still points only to garbage instead of the values it should point to. Admittedly, I am slightly confused with mixing the pointer and array syntax and which one is preferable/possible in a Cython context.
In the end, I want to call cyTest from a pure C++ function:
#include <iostream>
#include <Python.h>
#include "cyTest.h"
void main() {
Py_Initialize();
initcyTest();
double input[3] = {1,2,3};
double* output = cyTest(input);
for(int i = 0; i < 3; i++)
std::cout << "cout: " << output[i] << std::endl;
Py_Finalize();
}
This returns similar results:
loop: 1.0->1.0
loop: 2.0->4.0
loop: 3.0->9.0
cout: 1
cout: 6.30058e+077
cout: 6.39301e-308
Anyone care to explain what error I'm making? I'd like to keep it as simple as possible. It's just returning an array from Cython to C++ after all. I'll deal with dynamic memory allocation later, if not necessary.
You are returning reference to local array ( output ), which will not work.
Try changing your script to:
from cpython.mem cimport PyMem_Malloc
cdef public double * cyTest(double[] input):
cdef double * output = < double * >PyMem_Malloc( sizeof(double) * 3 )
for i in xrange(3):
output[i] = input[i]**2
print 'loop: ' + str(output[i])
return output
And in your c++ code,
after you done using double* output issue free( output );
If you want to use cdef double* results = cyTest(test) in your pyx script then don't forget to use PyMem_Free(results)