4

my problem "should" be simple but I am still not able to solve it.

I am currently working on a project that requires some heavy computations (done in C++) and some post-simulations data analysis (done in Python).

However, now I am changing the main algorithm and I will need to "cycle" some computations back and forth from C++ and Python. That is, I will need to move back and forth from C++ and Python a matrix of doubles.

In C++ the matrix of data is a "gsl_matrix" object while in python the same matrix is implemented as a "numpy array".

At the moment, I am running my C++ code, saving the matrix to file, reading it from python, writing it back to file and then opening it back again in C++ for further computations.

Since this is VERY inefficient, I would like to ask if somebody can give me an example on how to do it in a "clean" way.

I have been reading (and trying for 10 days) SWIG, Cython, Boost.Python and Boost.Numpy but I'm still not able to crack it.

Does anyone have a worked example to share?

Thanks!

Rene

4

1 回答 1

1

我认为您不需要自己实现包装器，因为您可以使用pygsl。如果您真的想实现自己的版本，这里是 pygsl 的例程，可能对您有价值

%{
#include <gsl/gsl_matrix_double.h>
#include <gsl/gsl_matrix_complex_double.h>
%}

%include typemaps.i

// gsl_matrix typemaps
%typemap(in) gsl_matrix* %{
  PyArrayObject *_PyMatrix$argnum;
  gsl_matrix_view matrix$argnum;
  {
    _PyMatrix$argnum = (PyArrayObject*)
      PyArray_ContiguousFromObject($input, PyArray_DOUBLE, 2, 2);
    if (_PyMatrix$argnum == NULL)
      return NULL;
    matrix$argnum
      = gsl_matrix_view_array((double*)_PyMatrix$argnum->data,
                  _PyMatrix$argnum->dimensions[0],
                  _PyMatrix$argnum->dimensions[1]);    
    $1 = &matrix$argnum.matrix;
  }
%}
于 2013-09-11T22:53:58.180 回答