1 // Copyright (C) 2007-2016 CEA/DEN, EDF R&D
3 // This library is free software; you can redistribute it and/or
4 // modify it under the terms of the GNU Lesser General Public
5 // License as published by the Free Software Foundation; either
6 // version 2.1 of the License, or (at your option) any later version.
8 // This library is distributed in the hope that it will be useful,
9 // but WITHOUT ANY WARRANTY; without even the implied warranty of
10 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 // Lesser General Public License for more details.
13 // You should have received a copy of the GNU Lesser General Public
14 // License along with this library; if not, write to the Free Software
15 // Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 // See http://www.salome-platform.org/ or email : webmaster.salome@opencascade.com
21 #include "CommInterface.hxx"
22 #include "Topology.hxx"
23 #include "BlockTopology.hxx"
24 #include "ComponentTopology.hxx"
25 #include "ParaFIELD.hxx"
26 #include "MPIProcessorGroup.hxx"
27 #include "ParaMESH.hxx"
29 #include "InterpolationMatrix.hxx"
30 #include "InterpKernelDEC.hxx"
31 #include "ElementLocator.hxx"
37 \anchor InterpKernelDEC-det
38 \class InterpKernelDEC
40 \section InterpKernelDEC-over Overview
42 The InterpKernelDEC enables the \ref InterpKerRemapGlobal "remapping" (or interpolation) of fields between
46 methodology is based on the algorithms of %INTERP_KERNEL, that is to say, they work in a similar fashion than
47 what the \ref remapper "sequential remapper" does. The following \ref discretization "projection methods"
48 are supported: P0->P0 (the most common case), P1->P0, P0->P1.
50 The computation is possible for 3D meshes, 2D meshes, and 3D-surface
51 meshes. Dimensions must be identical for code A and code B (for instance, though it could be
52 desirable, it is not yet possible to couple 3D surfaces with 2D surfaces).
54 The name "InterpKernelDEC" comes from the fact that this class uses exactly the same algorithms
55 as the sequential remapper. Both this class and the sequential
56 \ref MEDCoupling::MEDCouplingRemapper "MEDCouplingRemapper" are built on top of the %INTERP_KERNEL
57 algorithms (notably the computation of the intersection volumes).
59 Among the important properties inherited from the parent abstract class \ref DisjointDEC-det "DisjointDEC",
60 the two \ref MPIProcessorGroup-det "processor groups" (source and target) must have a void intersection.
62 \image html NonCoincident_small.png "Transfer of a field supported by a quadrangular mesh to a triangular mesh".
64 \image latex NonCoincident_small.eps "Transfer of a field supported by a quadrangular mesh to a triangular mesh"
66 In the figure above we see the transfer of a field based on a quadrangular mesh to a new field supported by
67 a triangular mesh. In a P0-P0 interpolation, to obtain the value on a triangle, the values on the
68 quadrangles are weighted by their intersection area and summed.
70 A typical use of InterpKernelDEC encompasses two distinct phases :
71 - A setup phase during which the intersection volumes are computed and the communication structures are
72 setup. This corresponds to calling the InterpKernelDEC::synchronize() method.
73 - A running phase during which the projections are actually performed. This corresponds to the calls to
74 sendData() and recvData() which actually trigger the data exchange. The data exchange are synchronous
75 in the current version of the library so that recvData() and sendData() calls must be synchronized
76 on code A and code B processor groups.
78 The following code excerpt illustrates a typical use of the InterpKernelDEC class.
82 InterpKernelDEC dec(groupA, groupB);
83 dec.attachLocalField(field);
85 if (groupA.containsMyRank())
87 else if (groupB.containsMyRank())
91 A \ref InterpKerRemapGlobal "remapping" of the field from the source mesh to the target mesh is performed by
92 the function synchronise(), which computes the interpolation matrix.
94 Computing the field on the receiving side can be expressed in terms of a matrix-vector product :
95 \f$ \phi_t=W.\phi_s\f$, with \f$ \phi_t \f$ the field on the target side and \f$ \phi_s \f$ the field
97 When remapping a 3D surface to another 3D surface, a projection phase is necessary to match elements
98 from both sides. Care must be taken when defining this projection to obtain a
99 \ref InterpKerRemapGlobal "conservative remapping".
101 In the P0-P0 case, this matrix is a plain rectangular matrix with coefficients equal to the
102 intersection areas between triangle and quadrangles. For instance, in the above figure, the matrix
106 \begin{tabular}{|cccc|}
107 0.72 & 0 & 0.2 & 0 \\
108 0.46 & 0 & 0.51 & 0.03\\
109 0.42 & 0.53 & 0 & 0.05\\
110 0 & 0 & 0.92 & 0.05 \\
114 \section InterpKernelDEC-options Options
115 On top of the usual \ref MEDCoupling::DECOptions "DEC options", the options supported by %InterpKernelDEC objects are
116 related to the underlying \ref InterpKerIntersectors "intersector class".
117 All the options available in the intersector objects are
118 available for the %InterpKernelDEC object. The various options available for intersectors can
119 be reviewed in \ref InterpKerIntersectors.
123 InterpKernelDEC dec(source_group, target_group);
124 dec.attachLocalField(field);
125 dec.setDoRotate(false);
126 dec.setPrecision(1e-12);
130 \warning{ Options must be set before calling the synchronize method. }
133 InterpKernelDEC::InterpKernelDEC():
135 _nb_distant_points(0), _distant_coords(0),
136 _distant_locations(0), _interpolation_matrix(0)
141 This constructor creates an InterpKernelDEC which has \a source_group as a working side
142 and \a target_group as an idle side. All the processors will actually participate, but intersection computations will be performed on the working side during the \a synchronize() phase.
143 The constructor must be called synchronously on all processors of both processor groups.
144 The source group and target group MUST form a partition of all the procs within the communicator passed as 'world_comm'
145 when building the group.
147 \param source_group working side ProcessorGroup
148 \param target_group lazy side ProcessorGroup
151 InterpKernelDEC::InterpKernelDEC(ProcessorGroup& source_group, ProcessorGroup& target_group):
152 DisjointDEC(source_group, target_group),
153 _nb_distant_points(0), _distant_coords(0),
154 _distant_locations(0), _interpolation_matrix(0)
161 * Creates an InterpKernelDEC from a set of source procs IDs and target group IDs.
162 * The difference with the ctor using groups is that the set of procs might not cover entirely MPI_COMM_WORLD
163 * (a sub-communicator holding the union of source and target procs is recreated internally).
165 InterpKernelDEC::InterpKernelDEC(const std::set<int>& src_ids, const std::set<int>& trg_ids,
166 const MPI_Comm& world_comm):
167 DisjointDEC(src_ids,trg_ids,world_comm),
168 _nb_distant_points(0), _distant_coords(0),
169 _distant_locations(0), _interpolation_matrix(0)
173 InterpKernelDEC::~InterpKernelDEC()
175 if (_interpolation_matrix !=0)
176 delete _interpolation_matrix;
180 \brief Synchronization process for exchanging topologies.
182 This method prepares all the structures necessary for sending data from a processor group to the other. It uses the mesh
183 underlying the fields that have been set with attachLocalField method.
184 It works in four steps :
185 -# Bounding boxes are computed for each sub-domain,
186 -# The lazy side mesh parts that are likely to intersect the working side local processor are sent to the working side,
187 -# The working side calls the interpolation kernel to compute the intersection between local and imported mesh.
188 -# The lazy side is updated so that it knows the structure of the data that will be sent by
189 the working side during a \a sendData() call.
192 void InterpKernelDEC::synchronize()
196 delete _interpolation_matrix;
197 _interpolation_matrix = new InterpolationMatrix (_local_field, *_source_group,*_target_group,*this,*this);
199 //setting up the communication DEC on both sides
200 if (_source_group->containsMyRank())
202 //locate the distant meshes
203 ElementLocator locator(*_local_field, *_target_group, *_source_group);
204 //transferring option from InterpKernelDEC to ElementLocator
205 locator.copyOptions(*this);
206 MEDCouplingPointSet* distant_mesh=0;
208 std::string distantMeth;
209 for (int i=0; i<_target_group->size(); i++)
211 // int idistant_proc = (i+_source_group->myRank())%_target_group->size();
214 //gathers pieces of the target meshes that can intersect the local mesh
215 locator.exchangeMesh(idistant_proc,distant_mesh,distant_ids);
216 if (distant_mesh !=0)
218 locator.exchangeMethod(_method,idistant_proc,distantMeth);
219 //adds the contribution of the distant mesh on the local one
220 int idistant_proc_in_union=_union_group->translateRank(_target_group,idistant_proc);
221 //std::cout <<"add contribution from proc "<<idistant_proc_in_union<<" to proc "<<_union_group->myRank()<<std::endl;
222 _interpolation_matrix->addContribution(*distant_mesh,idistant_proc_in_union,distant_ids,_method,distantMeth);
223 distant_mesh->decrRef();
224 delete [] distant_ids;
229 _interpolation_matrix->finishContributionW(locator);
232 if (_target_group->containsMyRank())
234 ElementLocator locator(*_local_field, *_source_group, *_target_group);
235 //transferring option from InterpKernelDEC to ElementLocator
236 locator.copyOptions(*this);
237 MEDCouplingPointSet* distant_mesh=0;
239 for (int i=0; i<_source_group->size(); i++)
241 // int idistant_proc = (i+_target_group->myRank())%_source_group->size();
243 //gathers pieces of the target meshes that can intersect the local mesh
244 locator.exchangeMesh(idistant_proc,distant_mesh,distant_ids);
245 //std::cout << " Data sent from "<<_union_group->myRank()<<" to source proc "<< idistant_proc<<std::endl;
248 std::string distantMeth;
249 locator.exchangeMethod(_method,idistant_proc,distantMeth);
250 distant_mesh->decrRef();
251 delete [] distant_ids;
256 _interpolation_matrix->finishContributionL(locator);
258 _interpolation_matrix->prepare();
263 Receives the data whether the processor is on the working side or on the lazy side. It must match a \a sendData() call on the other side.
265 void InterpKernelDEC::recvData()
267 if (_source_group->containsMyRank())
268 _interpolation_matrix->transposeMultiply(*_local_field->getField());
269 else if (_target_group->containsMyRank())
271 _interpolation_matrix->multiply(*_local_field->getField());
272 if (getForcedRenormalization())
273 renormalizeTargetField(getMeasureAbsStatus());
279 Receives the data at time \a time in asynchronous mode. The value of the field
280 will be time-interpolated from the field values received.
281 \param time time at which the value is desired
283 void InterpKernelDEC::recvData( double time )
285 _interpolation_matrix->getAccessDEC()->setTime(time);
290 Sends the data whether the processor is on the working side or on the lazy side.
291 It must match a recvData() call on the other side.
293 void InterpKernelDEC::sendData()
295 if (_source_group->containsMyRank())
298 _interpolation_matrix->multiply(*_local_field->getField());
299 if (getForcedRenormalization())
300 renormalizeTargetField(getMeasureAbsStatus());
303 else if (_target_group->containsMyRank())
304 _interpolation_matrix->transposeMultiply(*_local_field->getField());
308 Sends the data available at time \a time in asynchronous mode.
309 \param time time at which the value is available
310 \param deltatime time interval between the value presently sent and the next one.
312 void InterpKernelDEC::sendData( double time , double deltatime )
314 _interpolation_matrix->getAccessDEC()->setTime(time,deltatime);