• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright Jim Bosch 2011-2012.
2 // Distributed under the Boost Software License, Version 1.0.
3 //    (See accompanying file LICENSE_1_0.txt or copy at
4 //          http://www.boost.org/LICENSE_1_0.txt)
5 
6 /**
7  *  A simple example showing how to wrap a couple of C++ functions that
8  *  operate on 2-d arrays into Python functions that take NumPy arrays
9  *  as arguments.
10  *
11  *  If you find have a lot of such functions to wrap, you may want to
12  *  create a C++ array type (or use one of the many existing C++ array
13  *  libraries) that maps well to NumPy arrays and create Boost.Python
14  *  converters.  There's more work up front than the approach here,
15  *  but much less boilerplate per function.  See the "Gaussian" example
16  *  included with Boost.NumPy for an example of custom converters, or
17  *  take a look at the "ndarray" project on GitHub for a more complete,
18  *  high-level solution.
19  *
20  *  Note that we're using embedded Python here only to make a convenient
21  *  self-contained example; you could just as easily put the wrappers
22  *  in a regular C++-compiled module and imported them in regular
23  *  Python.  Again, see the Gaussian demo for an example.
24  */
25 
26 #include <boost/python/numpy.hpp>
27 #include <boost/scoped_array.hpp>
28 #include <iostream>
29 
30 namespace p = boost::python;
31 namespace np = boost::python::numpy;
32 
33 // This is roughly the most efficient way to write a C/C++ function that operates
34 // on a 2-d NumPy array - operate directly on the array by incrementing a pointer
35 // with the strides.
fill1(double * array,int rows,int cols,int row_stride,int col_stride)36 void fill1(double * array, int rows, int cols, int row_stride, int col_stride) {
37     double * row_iter = array;
38     double n = 0.0; // just a counter we'll fill the array with.
39     for (int i = 0; i < rows; ++i, row_iter += row_stride) {
40         double * col_iter = row_iter;
41         for (int j = 0; j < cols; ++j, col_iter += col_stride) {
42             *col_iter = ++n;
43         }
44     }
45 }
46 
47 // Here's a simple wrapper function for fill1.  It requires that the passed
48 // NumPy array be exactly what we're looking for - no conversion from nested
49 // sequences or arrays with other data types, because we want to modify it
50 // in-place.
wrap_fill1(np::ndarray const & array)51 void wrap_fill1(np::ndarray const & array) {
52     if (array.get_dtype() != np::dtype::get_builtin<double>()) {
53         PyErr_SetString(PyExc_TypeError, "Incorrect array data type");
54         p::throw_error_already_set();
55     }
56     if (array.get_nd() != 2) {
57         PyErr_SetString(PyExc_TypeError, "Incorrect number of dimensions");
58         p::throw_error_already_set();
59     }
60     fill1(reinterpret_cast<double*>(array.get_data()),
61           array.shape(0), array.shape(1),
62           array.strides(0) / sizeof(double), array.strides(1) / sizeof(double));
63 }
64 
65 // Another fill function that takes a double**.  This is less efficient, because
66 // it's not the native NumPy data layout, but it's common enough in C/C++ that
67 // it's worth its own example.  This time we don't pass the strides, and instead
68 // in wrap_fill2 we'll require the C_CONTIGUOUS bitflag, which guarantees that
69 // the column stride is 1 and the row stride is the number of columns.  That
70 // restricts the arrays that can be passed to fill2 (it won't work on most
71 // subarray views or transposes, for instance).
fill2(double ** array,int rows,int cols)72 void fill2(double ** array, int rows, int cols) {
73     double n = 0.0; // just a counter we'll fill the array with.
74     for (int i = 0; i < rows; ++i) {
75         for (int j = 0; j < cols; ++j) {
76             array[i][j] = ++n;
77         }
78     }
79 }
80 // Here's the wrapper for fill2; it's a little more complicated because we need
81 // to check the flags and create the array of pointers.
wrap_fill2(np::ndarray const & array)82 void wrap_fill2(np::ndarray const & array) {
83     if (array.get_dtype() != np::dtype::get_builtin<double>()) {
84         PyErr_SetString(PyExc_TypeError, "Incorrect array data type");
85         p::throw_error_already_set();
86     }
87     if (array.get_nd() != 2) {
88         PyErr_SetString(PyExc_TypeError, "Incorrect number of dimensions");
89         p::throw_error_already_set();
90     }
91     if (!(array.get_flags() & np::ndarray::C_CONTIGUOUS)) {
92         PyErr_SetString(PyExc_TypeError, "Array must be row-major contiguous");
93         p::throw_error_already_set();
94     }
95     double * iter = reinterpret_cast<double*>(array.get_data());
96     int rows = array.shape(0);
97     int cols = array.shape(1);
98     boost::scoped_array<double*> ptrs(new double*[rows]);
99     for (int i = 0; i < rows; ++i, iter += cols) {
100         ptrs[i] = iter;
101     }
102     fill2(ptrs.get(), array.shape(0), array.shape(1));
103 }
104 
BOOST_PYTHON_MODULE(example)105 BOOST_PYTHON_MODULE(example) {
106     np::initialize();  // have to put this in any module that uses Boost.NumPy
107     p::def("fill1", wrap_fill1);
108     p::def("fill2", wrap_fill2);
109 }
110 
main(int argc,char ** argv)111 int main(int argc, char **argv)
112 {
113     // This line makes our module available to the embedded Python intepreter.
114 # if PY_VERSION_HEX >= 0x03000000
115     PyImport_AppendInittab("example", &PyInit_example);
116 # else
117     PyImport_AppendInittab("example", &initexample);
118 # endif
119     // Initialize the Python runtime.
120     Py_Initialize();
121 
122     PyRun_SimpleString(
123         "import example\n"
124         "import numpy\n"
125         "z1 = numpy.zeros((5,6), dtype=float)\n"
126         "z2 = numpy.zeros((4,3), dtype=float)\n"
127         "example.fill1(z1)\n"
128         "example.fill2(z2)\n"
129         "print z1\n"
130         "print z2\n"
131     );
132     Py_Finalize();
133 }
134 
135 
136