| 1 |
// |
| 2 |
// SPDX-License-Identifier: BSD-3-Clause |
| 3 |
// Copyright Contributors to the OpenEXR Project. |
| 4 |
// |
| 5 |
|
| 6 |
// clang-format off |
| 7 |
|
| 8 |
#include <Python.h> |
| 9 |
#include <boost/python.hpp> |
| 10 |
#include <PyImath.h> |
| 11 |
#include <PyImathVec.h> |
| 12 |
#include <PyImathColor.h> |
| 13 |
#include <iostream> |
| 14 |
#include <boost/format.hpp> |
| 15 |
#define NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION |
| 16 |
#include <numpy/arrayobject.h> |
| 17 |
|
| 18 |
using namespace boost::python; |
| 19 |
using namespace PyImath; |
| 20 |
|
| 21 |
template <typename T> |
| 22 |
struct Holder |
| 23 |
{ |
| 24 |
Holder( T &a ) : m_val( a ) {} |
| 25 |
static void Cleanup (PyObject *capsule) |
| 26 |
{ |
| 27 |
Holder* h = static_cast<Holder*> (PyCapsule_GetPointer (capsule, NULL)); |
| 28 |
delete h; |
| 29 |
} |
| 30 |
T m_val; |
| 31 |
}; |
| 32 |
|
| 33 |
template <typename T> |
| 34 |
static void |
| 35 |
setBaseObject (PyObject* nparr, T& arr) |
| 36 |
{ |
| 37 |
using holder = Holder<T>; |
| 38 |
|
| 39 |
holder* ph = new holder (arr); |
| 40 |
PyObject* capsule = PyCapsule_New (ph, NULL, holder::Cleanup); |
| 41 |
PyArray_SetBaseObject ((PyArrayObject*) nparr, capsule); |
| 42 |
} |
| 43 |
|
| 44 |
template <typename T> struct NumpyTypeFromType { BOOST_STATIC_CONSTANT(int, typeEnum=NPY_NOTYPE); }; |
| 45 |
template <> struct NumpyTypeFromType<signed char> { BOOST_STATIC_CONSTANT(int, typeEnum=NPY_INT8); }; |
| 46 |
template <> struct NumpyTypeFromType<unsigned char> { BOOST_STATIC_CONSTANT(int, typeEnum=NPY_UINT8); }; |
| 47 |
template <> struct NumpyTypeFromType<short> { BOOST_STATIC_CONSTANT(int, typeEnum=NPY_INT16); }; |
| 48 |
template <> struct NumpyTypeFromType<unsigned short> { BOOST_STATIC_CONSTANT(int, typeEnum=NPY_UINT16); }; |
| 49 |
template <> struct NumpyTypeFromType<int> { BOOST_STATIC_CONSTANT(int, typeEnum=NPY_INT32); }; |
| 50 |
template <> struct NumpyTypeFromType<unsigned int> { BOOST_STATIC_CONSTANT(int, typeEnum=NPY_UINT32); }; |
| 51 |
template <> struct NumpyTypeFromType<float> { BOOST_STATIC_CONSTANT(int, typeEnum=NPY_FLOAT); }; |
| 52 |
template <> struct NumpyTypeFromType<double> { BOOST_STATIC_CONSTANT(int, typeEnum=NPY_DOUBLE); }; |
| 53 |
|
| 54 |
template <typename T> struct BaseTypeFrom2DArray { typedef void type; }; |
| 55 |
template <> struct BaseTypeFrom2DArray<IntArray2D> { typedef int type; }; |
| 56 |
template <> struct BaseTypeFrom2DArray<FloatArray2D> { typedef float type; }; |
| 57 |
template <> struct BaseTypeFrom2DArray<DoubleArray2D> { typedef double type; }; |
| 58 |
template <> struct BaseTypeFrom2DArray<Color4cArray> { typedef IMATH_NAMESPACE::Color4c type; }; |
| 59 |
template <> struct BaseTypeFrom2DArray<Color4fArray> { typedef IMATH_NAMESPACE::Color4f type; }; |
| 60 |
|
| 61 |
template <typename T> |
| 62 |
object |
| 63 |
arrayToNumpy_scalar(T &sa) |
| 64 |
{ |
| 65 |
typedef typename T::BaseType PodType; |
| 66 |
|
| 67 |
if (sa.stride() != 1) |
| 68 |
throw std::logic_error("Unable to make numpy wrapping of strided arrays"); |
| 69 |
|
| 70 |
int type = NumpyTypeFromType<PodType>::typeEnum; |
| 71 |
npy_intp dims = sa.len(); |
| 72 |
PodType *data = &sa[0]; |
| 73 |
PyObject *a = PyArray_SimpleNewFromData(1, &dims, type, data); |
| 74 |
|
| 75 |
if (!a) |
| 76 |
throw_error_already_set(); |
| 77 |
|
| 78 |
setBaseObject (a, sa); |
| 79 |
|
| 80 |
object retval = object(handle<>(a)); |
| 81 |
return retval; |
| 82 |
} |
| 83 |
|
| 84 |
template<typename T> |
| 85 |
object |
| 86 |
arrayToNumpy_vector(T &va) |
| 87 |
{ |
| 88 |
typedef typename T::BaseType BaseType; |
| 89 |
typedef typename BaseType::BaseType PodType; |
| 90 |
|
| 91 |
if (va.stride() != 1) |
| 92 |
throw std::logic_error("Unable to make numpy wrapping of strided arrays"); |
| 93 |
|
| 94 |
int type = NumpyTypeFromType<PodType>::typeEnum; |
| 95 |
npy_intp dims[2]{ va.len(), BaseType::dimensions()}; |
| 96 |
PodType *data = &va[0][0]; |
| 97 |
PyObject *a = PyArray_SimpleNewFromData(2, dims, type, data); |
| 98 |
|
| 99 |
if (!a) |
| 100 |
throw_error_already_set(); |
| 101 |
|
| 102 |
setBaseObject (a, va); |
| 103 |
|
| 104 |
object retval = object (handle<> (a)); |
| 105 |
return retval; |
| 106 |
} |
| 107 |
|
| 108 |
template<typename T> |
| 109 |
object |
| 110 |
arrayToNumpy_scalar2D(T &ca) |
| 111 |
{ |
| 112 |
typedef typename BaseTypeFrom2DArray<T>::type PodType; |
| 113 |
|
| 114 |
// Numpy is matrix indexed based - the first index indicates the row and |
| 115 |
// the second index indicates the column. |
| 116 |
// Zeno data is image indexed based - the first index indicates the column |
| 117 |
// and the second indicates the row. |
| 118 |
// Swap the dimensions so Numpy handles data natively with matrix based |
| 119 |
// indexing. |
| 120 |
IMATH_NAMESPACE::Vec2<size_t> len = ca.len(); |
| 121 |
int type = NumpyTypeFromType<PodType>::typeEnum; |
| 122 |
npy_intp dims[2]{ static_cast<npy_intp>(len.y), |
| 123 |
static_cast<npy_intp>(len.x) }; |
| 124 |
PodType *data = &ca(0, 0); |
| 125 |
PyObject *a = PyArray_SimpleNewFromData(2, dims, type, data); |
| 126 |
|
| 127 |
if (!a) |
| 128 |
throw_error_already_set(); |
| 129 |
|
| 130 |
setBaseObject (a, ca); |
| 131 |
|
| 132 |
object retval = object(handle<>(a)); |
| 133 |
return retval; |
| 134 |
} |
| 135 |
|
| 136 |
template<typename T> |
| 137 |
object |
| 138 |
arrayToNumpy_vector2D(T &ca) |
| 139 |
{ |
| 140 |
typedef typename BaseTypeFrom2DArray<T>::type BaseType; |
| 141 |
typedef typename BaseType::BaseType PodType; |
| 142 |
|
| 143 |
// Numpy is matrix indexed based - the first index indicates the row and |
| 144 |
// the second index indicates the column. |
| 145 |
// Zeno data is image indexed based - the first index indicates the column |
| 146 |
// and the second indicates the row. |
| 147 |
// Swap the dimensions so Numpy handles data natively with matrix based |
| 148 |
// indexing. |
| 149 |
IMATH_NAMESPACE::Vec2<size_t> len = ca.len(); |
| 150 |
int type = NumpyTypeFromType<PodType>::typeEnum; |
| 151 |
npy_intp dims[3]{ static_cast<npy_intp>(len.y), |
| 152 |
static_cast<npy_intp>(len.x), |
| 153 |
BaseType::dimensions() }; |
| 154 |
PodType *data = &ca(0, 0)[0]; |
| 155 |
PyObject *a = PyArray_SimpleNewFromData(3, dims, type, data); |
| 156 |
|
| 157 |
if (!a) |
| 158 |
throw_error_already_set(); |
| 159 |
|
| 160 |
setBaseObject (a, ca); |
| 161 |
|
| 162 |
object retval = object(handle<>(a)); |
| 163 |
return retval; |
| 164 |
} |
| 165 |
|
| 166 |
#if PY_MAJOR_VERSION > 2 |
| 167 |
static void *apply_import() |
| 168 |
{ |
| 169 |
import_array(); |
| 170 |
return 0; |
| 171 |
} |
| 172 |
#endif |
| 173 |
|
| 174 |
// Convenience macro for wrapping scalar PyImath array types. |
| 175 |
#define WRAP_SCALAR_ARRAY(ARRAY_TYPE) \ |
| 176 |
def("arrayToNumpy", &arrayToNumpy_scalar<ARRAY_TYPE>, \ |
| 177 |
"arrayToNumpy(array) - wrap the given " #ARRAY_TYPE " as a numpy array", \ |
| 178 |
(arg("array"))); |
| 179 |
#define WRAP_SCALAR_ARRAY_2D(ARRAY_TYPE) \ |
| 180 |
def("arrayToNumpy", &arrayToNumpy_scalar2D<ARRAY_TYPE>, \ |
| 181 |
"arrayToNumpy(array) - wrap the given " #ARRAY_TYPE " as a numpy array", \ |
| 182 |
(arg("array"))); |
| 183 |
|
| 184 |
// Convenience macro for wrapping vector PyImath array types. |
| 185 |
#define WRAP_VECTOR_ARRAY(ARRAY_TYPE) \ |
| 186 |
def("arrayToNumpy", &arrayToNumpy_vector<ARRAY_TYPE>, \ |
| 187 |
"arrayToNumpy(array) - wrap the given " #ARRAY_TYPE " as a numpy array", \ |
| 188 |
(arg("array"))); |
| 189 |
#define WRAP_VECTOR_ARRAY_2D(ARRAY_TYPE) \ |
| 190 |
def("arrayToNumpy", &arrayToNumpy_vector2D<ARRAY_TYPE>, \ |
| 191 |
"arrayToNumpy(array) - wrap the given " #ARRAY_TYPE " as a numpy array", \ |
| 192 |
(arg("array"))); |
| 193 |
|
| 194 |
BOOST_PYTHON_MODULE(imathnumpy) |
| 195 |
{ |
| 196 |
scope().attr("__doc__") = "Imathnumpy module"; |
| 197 |
scope().attr("__version__") = IMATH_VERSION_STRING; |
| 198 |
|
| 199 |
handle<> imath(PyImport_ImportModule("imath")); |
| 200 |
if (PyErr_Occurred()) throw_error_already_set(); |
| 201 |
scope().attr("imath") = imath; |
| 202 |
|
| 203 |
handle<> numpy(PyImport_ImportModule("numpy")); |
| 204 |
if (PyErr_Occurred()) throw_error_already_set(); |
| 205 |
scope().attr("numpy") = numpy; |
| 206 |
|
| 207 |
#if PY_MAJOR_VERSION > 2 |
| 208 |
// seems like numpy expects this to be used in a scenario |
| 209 |
// where there is a return value in python3... |
| 210 |
(void)apply_import(); |
| 211 |
#else |
| 212 |
import_array(); |
| 213 |
#endif |
| 214 |
|
| 215 |
scope().attr("__doc__") = "Array wrapping module to overlay imath array data with numpy arrays"; |
| 216 |
|
| 217 |
WRAP_SCALAR_ARRAY(SignedCharArray) |
| 218 |
WRAP_SCALAR_ARRAY(UnsignedCharArray) |
| 219 |
WRAP_SCALAR_ARRAY(ShortArray) |
| 220 |
WRAP_SCALAR_ARRAY(UnsignedShortArray) |
| 221 |
WRAP_SCALAR_ARRAY(IntArray) |
| 222 |
WRAP_SCALAR_ARRAY(UnsignedIntArray) |
| 223 |
WRAP_SCALAR_ARRAY(FloatArray) |
| 224 |
WRAP_SCALAR_ARRAY(DoubleArray) |
| 225 |
|
| 226 |
WRAP_VECTOR_ARRAY(V2sArray) |
| 227 |
WRAP_VECTOR_ARRAY(V2iArray) |
| 228 |
WRAP_VECTOR_ARRAY(V2fArray) |
| 229 |
WRAP_VECTOR_ARRAY(V2dArray) |
| 230 |
|
| 231 |
WRAP_VECTOR_ARRAY(V3sArray) |
| 232 |
WRAP_VECTOR_ARRAY(V3iArray) |
| 233 |
WRAP_VECTOR_ARRAY(V3fArray) |
| 234 |
WRAP_VECTOR_ARRAY(V3dArray) |
| 235 |
|
| 236 |
WRAP_VECTOR_ARRAY(V4sArray) |
| 237 |
WRAP_VECTOR_ARRAY(V4iArray) |
| 238 |
WRAP_VECTOR_ARRAY(V4fArray) |
| 239 |
WRAP_VECTOR_ARRAY(V4dArray) |
| 240 |
|
| 241 |
WRAP_VECTOR_ARRAY(C3cArray) |
| 242 |
WRAP_VECTOR_ARRAY(C3fArray) |
| 243 |
WRAP_VECTOR_ARRAY(C4cArray) |
| 244 |
WRAP_VECTOR_ARRAY(C4fArray) |
| 245 |
|
| 246 |
WRAP_SCALAR_ARRAY_2D(IntArray2D) |
| 247 |
WRAP_SCALAR_ARRAY_2D(FloatArray2D) |
| 248 |
WRAP_SCALAR_ARRAY_2D(DoubleArray2D) |
| 249 |
|
| 250 |
WRAP_VECTOR_ARRAY_2D(Color4cArray) |
| 251 |
WRAP_VECTOR_ARRAY_2D(Color4fArray) |
| 252 |
} |