1 // Copyright (C) 2007-2015 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
19 // Author : Anthony Geay (CEA/DEN)
21 #include "OverlapDEC.hxx"
22 #include "CommInterface.hxx"
23 #include "ParaFIELD.hxx"
24 #include "MPIProcessorGroup.hxx"
25 #include "OverlapElementLocator.hxx"
26 #include "OverlapInterpolationMatrix.hxx"
31 \anchor OverlapDEC-det
34 \section OverlapDEC-over Overview
36 The \c OverlapDEC enables the \ref InterpKerRemapGlobal "conservative remapping" of fields between
37 two parallel codes. This remapping is based on the computation of intersection volumes on
38 a \b single \b processor \b group. On this processor group are defined two field-templates called A
39 and B. The computation is possible for 3D meshes, 2D meshes, 3D-surface meshes, 1D meshes and
40 2D-curve meshes. Dimensions must be similar for the distribution templates A and B.
42 The main difference with \ref InterpKernelDEC-det "InterpKernelDEC" is that this
43 \ref para-dec "DEC" works with a *single* processor group, in which processors will share the work.
44 Consequently each processor manages two \ref MEDCouplingFieldTemplatesPage "field templates" (A and B)
45 called source and target.
46 Furthermore all processors in the processor group cooperate in the global interpolation matrix
47 computation. In this respect \c InterpKernelDEC is a specialization of \c OverlapDEC.
49 \section ParaMEDMEMOverlapDECAlgorithmDescription Algorithm description
51 Let's consider the following use case that is ran in ParaMEDMEMTest_OverlapDEC.cxx to describes
52 the different steps of the computation. The processor group contains 3 processors.
53 \anchor ParaMEDMEMOverlapDECImgTest1
54 \image html OverlapDEC1.png "Example split of the source and target mesh among the 3 procs"
56 \subsection ParaMEDMEMOverlapDECAlgoStep1 Step 1 : Bounding box exchange and global interaction between procs computation.
58 In order to reduce as much as possible the amount of communications between distant processors,
59 every processor computes a bounding box for A and B. Then a AllToAll communication is performed
61 every processor can compute the \b global interactions between processor.
62 This computation leads every processor to compute the same global TODO list expressed as a list
63 of pair. A pair ( x, y ) means that proc \b x fieldtemplate A can interact with fieltemplate B of
64 proc \b y because the two bounding boxes interact.
65 In the \ref ParaMEDMEMOverlapDECImgTest1 "example above" this computation leads to the following
66 a \b global TODO list :
68 \b (0,0),(0,1),(1,0),(1,2),(2,0),(2,1),(2,2)
70 Here the pair (0,2) does not appear because the bounding box of fieldtemplateA of proc#2 does
71 not intersect that of fieldtemplate B on proc#0.
73 Stage performed by ParaMEDMEM::OverlapElementLocator::computeBoundingBoxes.
75 \subsection ParaMEDMEMOverlapDECAlgoStep2 Step 2 : Computation of local TODO list
77 Starting from the global interaction previously computed in \ref ParaMEDMEMOverlapDECAlgoStep1
78 "Step 1", each proc computes the TODO list per proc.
79 The following rules is chosen : a pair (x,y) can be treated by either proc \#x or proc \#y,
80 in order to reduce the amount of data transfert among
81 processors. The algorithm chosen for load balancing is the following : Each processor has
82 an empty \b local TODO list at the beginning. Then for each pair (k,m) in
83 \b global TODO list, if proc\#k has less temporary local list than proc\#m pair, (k,m) is added
84 to temparary local TODO list of proc\#k.
85 If proc\#m has less temporary local TODO list than proc\#k pair, (k,m) is added to temporary
86 local TODO list of proc\#m.
87 If proc\#k and proc\#m have the same amount of temporary local TODO list pair, (k,m) is added to
88 temporary local TODO list of proc\#k.
90 In the \ref ParaMEDMEMOverlapDECImgTest1 "example above" this computation leads to the following
94 - proc\#1 : (0,1),(1,0)
95 - proc\#2 : (1,2),(2,0),(2,1),(2,2)
97 The algorithm described here is not perfect for this use case, we hope to enhance it soon.
99 At this stage each proc knows precisely its \b local TODO list (with regard to interpolation).
100 The \b local TODO list of other procs than local
101 is kept for future computations.
103 \subsection ParaMEDMEMOverlapDECAlgoStep3 Step 3 : Matrix echange between procs
105 Knowing the \b local TODO list, the aim now is to exchange field-templates between procs.
106 Each proc computes knowing TODO list per
107 proc computed in \ref ParaMEDMEMOverlapDECAlgoStep2 "Step 2" the exchange TODO list :
109 In the \ref ParaMEDMEMOverlapDECImgTest1 "example above" the exchange TODO list gives the
112 Sending TODO list per proc :
114 - proc \#0 : Send fieldtemplate A to Proc\#1, Send fieldtemplate B to Proc\#1, Send fieldtemplate
116 - Proc \#1 : Send fieldtemplate A to Proc\#2, Send fieldtemplate B to Proc\#2
117 - Proc \#2 : No send.
119 Receiving TODO list per proc :
121 - proc \#0 : No receiving
122 - proc \#1 : receiving fieldtemplate A from Proc\#0, receiving fieldtemplate B from Proc\#0
123 - proc \#2 : receiving fieldtemplate B from Proc\#0, receiving fieldtemplate A from Proc\#1,
124 receiving fieldtemplate B from Proc\#1
126 To avoid as much as possible large volumes of transfers between procs, only relevant parts of
127 meshes are sent. In order for proc\#k to send fieldtemplate A to fieldtemplate B
128 of proc \#m., proc\#k computes the part of mesh A contained in the boundingbox B of proc\#m. It
129 implies that the corresponding cellIds or nodeIds of the
130 corresponding part are sent to proc \#m too.
132 Let's consider the couple (k,m) in the TODO list. This couple is treated by either k or m as
133 seen in \ref ParaMEDMEMOverlapDECAlgoStep2 "here in Step2".
135 As will be dealt in Step 6, for final matrix-vector computations, the resulting matrix of the
136 couple (k,m) whereever it is computed (proc \#k or proc \#m)
137 will be stored in \b proc\#m.
139 - If proc \#k is in charge (performs the matrix computation) for this couple (k,m), target ids
140 (cells or nodes) of the mesh in proc \#m are renumbered, because proc \#m has seelected a sub mesh
141 of the target mesh to avoid large amounts of data to transfer. In this case as proc \#m is ultimately
142 in charge of the matrix, proc \#k must keep preciously the
143 source ids needed to be sent to proc\#m. No problem will appear for matrix assembling in proc m
144 for source ids because no restriction was done.
145 Concerning source ids to be sent for the matrix-vector computation, proc k will know precisely
146 which source ids field values to send to proc \#m.
147 This is embodied by OverlapMapping::keepTracksOfTargetIds in proc m.
149 - If proc \#m is in charge (performs matrix computation) for this couple (k,m), source ids (cells
150 or nodes) of the mesh in proc \#k are renumbered, because proc \#k has selected a sub mesh of the
151 source mesh to avoid large amounts of data to transfer. In this case as proc \#k is ultimately
152 in charge of the matrix, proc \#m receives the source ids
153 from remote proc \#k, and thus the matrix is directly correct, no need for renumbering as
154 in \ref ParaMEDMEMOverlapDECAlgoStep5 "Step 5". However proc \#k must
155 keep track of the ids sent to proc \#m for te matrix-vector computation.
156 This is incarnated by OverlapMapping::keepTracksOfSourceIds in proc k.
158 This step is performed in ParaMEDMEM::OverlapElementLocator::exchangeMeshes method.
160 \subsection ParaMEDMEMOverlapDECAlgoStep4 Step 4 : Computation of the interpolation matrix
162 After mesh exchange in \ref ParaMEDMEMOverlapDECAlgoStep3 "Step3" each processor has all the
163 required information to treat its \b local TODO list computed in
164 \ref ParaMEDMEMOverlapDECAlgoStep2 "Step2". This step is potentially CPU costly, which is why
165 the \b local TODO list per proc is expected to
166 be as well balanced as possible.
168 The interpolation is performed as the \ref ParaMEDMEM::MEDCouplingRemapper "remapper" does.
170 This operation is performed by OverlapInterpolationMatrix::addContribution method.
172 \subsection ParaMEDMEMOverlapDECAlgoStep5 Step 5 : Global matrix construction.
174 After having performed the TODO list at the end of \ref ParaMEDMEMOverlapDECAlgoStep4 "Step4"
175 we need to assemble the final matrix.
177 The final aim is to have a distributed matrix \f$ M_k \f$ on each proc\#k. In order to reduce
178 data exchange during the matrix product process,
179 \f$ M_k \f$ is built using sizeof(Proc group) \c std::vector< \c std::map<int,double> \c >.
181 For a proc\#k, it is necessary to fetch info of all matrices built in
182 \ref ParaMEDMEMOverlapDECAlgoStep4 "Step4" where the first element in pair (i,j)
185 After this step, the matrix repartition is the following after a call to
186 ParaMEDMEM::OverlapMapping::prepare :
188 - proc\#0 : (0,0),(1,0),(2,0)
189 - proc\#1 : (0,1),(2,1)
190 - proc\#2 : (1,2),(2,2)
192 Tuple (2,1) computed on proc 2 is stored in proc 1 after execution of the function
193 "prepare". This is an example of item 0 in \ref ParaMEDMEMOverlapDECAlgoStep2 "Step2".
194 Tuple (0,1) computed on proc 1 is stored in proc 1 too. This is an example of item 1 in \ref ParaMEDMEMOverlapDECAlgoStep2 "Step2".
196 In the end ParaMEDMEM::OverlapMapping::_proc_ids_to_send_vector_st will contain :
202 In the end ParaMEDMEM::OverlapMapping::_proc_ids_to_recv_vector_st will contain :
208 The method in charge to perform this is : ParaMEDMEM::OverlapMapping::prepare.
210 OverlapDEC::OverlapDEC(const std::set<int>& procIds, const MPI_Comm& world_comm):
211 _own_group(true),_interpolation_matrix(0), _locator(0),
212 _source_field(0),_own_source_field(false),
213 _target_field(0),_own_target_field(false),
214 _default_field_value(0.0),
217 ParaMEDMEM::CommInterface comm;
218 int *ranks_world=new int[procIds.size()]; // ranks of sources and targets in world_comm
219 std::copy(procIds.begin(),procIds.end(),ranks_world);
220 MPI_Group group,world_group;
221 comm.commGroup(world_comm,&world_group);
222 comm.groupIncl(world_group,procIds.size(),ranks_world,&group);
223 delete [] ranks_world;
224 comm.commCreate(world_comm,group,&_comm);
225 comm.groupFree(&group);
226 comm.groupFree(&world_group);
227 if(_comm==MPI_COMM_NULL)
232 std::set<int> idsUnion;
233 for(std::size_t i=0;i<procIds.size();i++)
235 _group=new MPIProcessorGroup(comm,idsUnion,_comm);
238 OverlapDEC::~OverlapDEC()
242 if(_own_source_field)
243 delete _source_field;
244 if(_own_target_field)
245 delete _target_field;
246 delete _interpolation_matrix;
248 if (_comm != MPI_COMM_NULL)
250 ParaMEDMEM::CommInterface comm;
251 comm.commFree(&_comm);
255 void OverlapDEC::sendRecvData(bool way)
263 void OverlapDEC::sendData()
265 _interpolation_matrix->multiply(_default_field_value);
268 void OverlapDEC::recvData()
270 throw INTERP_KERNEL::Exception("Not implemented yet !!!!");
271 //_interpolation_matrix->transposeMultiply();
274 void OverlapDEC::synchronize()
278 // Check number of components of field on both side (for now allowing void field/mesh on one proc is not allowed)
279 if (!_source_field || !_source_field->getField())
280 throw INTERP_KERNEL::Exception("OverlapDEC::synchronize(): currently, having a void source field on a proc is not allowed!");
281 if (!_target_field || !_target_field->getField())
282 throw INTERP_KERNEL::Exception("OverlapDEC::synchronize(): currently, having a void target field on a proc is not allowed!");
283 if (_target_field->getField()->getNumberOfComponents() != _source_field->getField()->getNumberOfComponents())
284 throw INTERP_KERNEL::Exception("OverlapDEC::synchronize(): source and target field have different number of components!");
285 delete _interpolation_matrix;
286 _locator = new OverlapElementLocator(_source_field,_target_field,*_group, getBoundingBoxAdjustmentAbs());
287 _interpolation_matrix=new OverlapInterpolationMatrix(_source_field,_target_field,*_group,*this,*this, *_locator);
288 _locator->copyOptions(*this);
289 _locator->exchangeMeshes(*_interpolation_matrix);
290 std::vector< std::pair<int,int> > jobs=_locator->getToDoList();
291 std::string srcMeth=_locator->getSourceMethod();
292 std::string trgMeth=_locator->getTargetMethod();
293 for(std::vector< std::pair<int,int> >::const_iterator it=jobs.begin();it!=jobs.end();it++)
295 const MEDCouplingPointSet *src=_locator->getSourceMesh((*it).first);
296 const DataArrayInt *srcIds=_locator->getSourceIds((*it).first);
297 const MEDCouplingPointSet *trg=_locator->getTargetMesh((*it).second);
298 const DataArrayInt *trgIds=_locator->getTargetIds((*it).second);
299 _interpolation_matrix->addContribution(src,srcIds,srcMeth,(*it).first,trg,trgIds,trgMeth,(*it).second);
301 _interpolation_matrix->prepare(_locator->getProcsToSendFieldData());
302 _interpolation_matrix->computeDeno();
305 void OverlapDEC::attachSourceLocalField(ParaFIELD *field, bool ownPt)
309 if(_own_source_field)
310 delete _source_field;
312 _own_source_field=ownPt;
315 void OverlapDEC::attachTargetLocalField(ParaFIELD *field, bool ownPt)
319 if(_own_target_field)
320 delete _target_field;
322 _own_target_field=ownPt;
325 bool OverlapDEC::isInGroup() const
329 return _group->containsMyRank();