File: interface.hrst

package info (click to toggle)
gridtools 2.3.9-1
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid
  • size: 29,480 kB
  • sloc: cpp: 228,792; python: 17,561; javascript: 9,164; ansic: 4,101; sh: 850; makefile: 231; f90: 201
file content (247 lines) | stat: -rw-r--r-- 8,996 bytes parent folder | download | duplicates (3)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
Interfacing to other programming languages
============================================

|GT| provides an easy macro interface to generate bindings to C and Fortran. This library is available separately
at `GridTools/cpp_bindgen <https://github.com/GridTools/cpp_bindgen>`_.

To make the cpp_bindgen library available the recommended way is to use CMake's FetchContent as follows

.. code-block:: CMake

  include(FetchContent)
  FetchContent_Declare(
    cpp_bindgen
    GIT_REPOSITORY https://github.com/GridTools/cpp_bindgen.git
    GIT_TAG        master # consider replacing master by a tagged version
  )
  FetchContent_MakeAvailable(cpp_bindgen)

Suppose, the user wants to export the function ``add_impl``.

.. code-block:: gridtools

  int add_impl(int l, int r) {
      return l + r;
  }

The macros ``BINDGEN_EXPORT_*`` provide ways to generate bindings to functions. The different
flavours of this macro are explained below. The macro generates a wrapper around the function
``add_impl`` which is called ``add`` and registers the function to be exported.

.. code-block:: gridtools

  #include <cpp_bindgen/export.hpp>
  BINDGEN_EXPORT_BINDING_2(add, add_impl);

The user can generate a C header and a Fortran module that matches this header by
adding a call to ``bindgen_add_library`` in his CMake project:

.. code-block:: CMake

  bindgen_add_library(add_lib SOURCES add.cpp)

This will generate a library ``add_lib`` which contains the exported symbol ``add``,
and it will generate a target ``add_lib_declarations`` that generates the files
``add_lib.h`` and ``add_lib.f90`` containing the bindings that can be used from C
and Fortran.

The C header contains the exported function (boilerplate code removed):

.. code-block:: gridtools

  int add(int, int);

The generated Fortran module contains the corresponding declaration:

.. code-block:: fortran

  module add_lib
  implicit none
    interface
      integer(c_int) function add(arg0, arg1) bind(c)
        use iso_c_binding
        integer(c_int), value :: arg0
        integer(c_int), value :: arg1
      end function
    end interface
  end

------------------------------------------------
Exporting functions with no array-type arguments
------------------------------------------------

There exist various flavours of these macros. Functions which are non-templated or fully-specialized
can be exported with ``BINDGEN_EXPORT_BINDING``, for example:

.. code-block:: gridtools

  int add_impl(int, int) { return 0; }
  BINDGEN_EXPORT_BINDING_2(add, add_impl);

  template <typename T>
  T add_impl(T, T) { return {}; }
  BINDGEN_EXPORT_BINDING(2, add_impl<int, int>);

All functions exist in two flavours: Either you can pass the number of arguments as part of the name
of the macro (``BINDGEN_EXPORT_BINDING_2`` stands for two arguments), or you can pass it as a first argument
to the generic ``BINDGEN_EXPORT_BINDING``. The first flavours exist for up to 9 arguments.

Note that ``BINDGEN_EXPORT_BINDING_X`` requires a name and a function pointer as its arguments.
A lambda cannot be passed as function pointer; thus, the type of the arguments cannot be
deduced. In such cases, the functions can be exported with ``BINDGEN_EXPORT_BINDING_WITH_SIGNATURE_X``,
which additionally takes the function type as an argument:

.. code-block:: gridtools

  BINDGEN_EXPORT_BINDING_WITH_SIGNATURE_2(add, int(int, int), [](int l, int r) { return l + r; });

Templated functions can be exported for a given set of specializations using
``BINDGEN_EXPORT_GENERIC_BINDING_X``. In addition to the function name and the function pointer, it takes a list of
overloads for which the bindings are generated:

.. code-block:: gridtools

  BINDGEN_EXPORT_GENERIC_BINDING(2, add, add_impl, (int, int)(double, int));

In the generated Fortran module, generic bindings will produce an interface combining the different
overloads:

.. code-block:: fortran

  interface
    integer(c_int) function add_f0(arg0, arg1) bind(c)
      use iso_c_binding
      integer(c_int), value :: arg0
      integer(c_int), value :: arg1
    end function
    real(c_double) function add_f1(arg0, arg1) bind(c)
      use iso_c_binding
      real(c_double), value :: arg0
      integer(c_int), value :: arg1
    end function
  end interface
  interface add
    procedure add_f0, add_f1
  end interface

----------------
Complex types
----------------

Only a limited set of types can be passed from Fortran / C through the C bindings interface to C++,
namely integral and floating point types, booleans and pointers to those types.

Array references, |GT| storages, and any type that is `fortran_array_bindable`
appear as ``bindgen_fortran_array_descriptor`` in the C bindings. This structure allows
the user to describe the data that needs to be passed to C++.

It is possible to write bindings to functions that accept or return other types.
During the generation process, they are replaced with pointers to the type ``bindgen_handle``.

.. code-block:: gridtools

  std::vector<int> make_vector_impl() { return {}; }
  void use_vector_impl(std::vector<int>) {}
  BINDGEN_EXPORT_BINDING_0(make_vector, make_vector_impl);
  BINDGEN_EXPORT_BINDING_1(use_vector, use_vector_impl);

The code above will generate the following signatures in the C-header:

.. code-block:: C

  bindgen_handle* make_vector();
  void use_vector(bindgen_handle*);

The user needs to make sure that the types that stand behind ``bindgen_handle`` match, otherwise
an exception will be thrown.

--------------------------------------------------------
Exporting functions with array-type arguments to Fortran
--------------------------------------------------------

Special macros exist to export function that take array-like arguments to Fortran. While the normal
macros export such arguments as ``bindgen_fortran_array_descriptor``, the "wrapped" macros create
additional wrappers around the functions that fill the structures themselves.

.. code-block:: gridtools

  void dummy_impl(int (&a)[2][2]) {}
  BINDGEN_EXPORT_BINDING_WRAPPED_1(dummy, dummy_impl);

The function ``dummy_impl`` is taking a reference to an array. When exporting this function with
``BINDGEN_EXPORT_BINDING_WRAPPED_X``, an additional wrapper is generated in the Fortran bindings:

.. code-block:: fortran

  module add_lib
  implicit none
    interface
      subroutine dummy_impl(arg0) bind(c, name="dummy")
        use iso_c_binding
        use array_descriptor
        type(bindgen_fortran_array_descriptor) :: arg0
      end subroutine
    end interface
  contains
      subroutine dummy(arg0)
        use iso_c_binding
        use array_descriptor
        integer(c_int), dimension(:,:), target :: arg0
        type(bindgen_fortran_array_descriptor) :: descriptor0

        descriptor0%rank = 2
        descriptor0%type = 1
        descriptor0%dims = reshape(shape(arg0), &
          shape(descriptor0%dims), (/0/))
        descriptor0%data = c_loc(arg0(lbound(arg0, 1),lbound(arg0, 2)))

        call dummy_impl(descriptor0)
      end subroutine
  end

This allows to call the Fortran function ``dummy`` in a convenient way:

.. code-block:: fortran

  integer(c_int), dimension(:, :) :: some_array
  call dummy(some_array)

The bindings will take care that the rank matches and it will infer the size of the array automatically.

All additional macros behave as mentioned above, namely ``BINDGEN_EXPORT_BINDING_WITH_SIGNATURE_WRAPPED``,
and ``BINDGEN_EXPORT_BINDING_GENERIC_WRAPPED``.

Data types need to be `fortran_array_wrappable` in order to be compatible with these macros. Natively, only
C arrays and ``fortran_array_adapter`` are `fortran_array_wrappable`. The latter is an adapter between
Fortran arrays and |GT| storages, such that the user can pass a Fortran array to a C++ function,
which then can be transformed into a |GT| storage.

.. code-block:: gridtools

  #include <gridtools/storage/adapter/fortran_array_adapter.hpp>

  using storage_info_t = storage_traits<Backend>::storage_info_t<0, 3>;
  using data_store_t = storage_traits<Backend>::data_store_t<double, storage_info_t>;

  void modify_array_impl(fortran_array_adapter<data_store_t> inout) {
      data_store_t data_store{storage_info_t{10, 10, 10}};
      transform(data_store, inout);

      // use data_store

      transform(inout, data_store);
  }
  BINDGEN_EXPORT_BINDING_WRAPPED_1(modify_array, modify_array_impl)

-----------
CMake usage
-----------

A call to ``bindgen_add_library`` generates the libraries and the headers. By default,
the C header file and the Fortran file is written directly into the source tree. This choice was
taken to improve building in cross-build environments, because the process cannot rely
on generated binaries being executable on the host system. The output folder can be overwritten
by setting ``FORTRAN_OUTPUT_DIR`` and ``C_OUTPUT_DIR``.

By default, the name of the generated Fortran module is set to the name of the library. A different
name can be set with ``FORTRAN_MODULE_NAME``.