ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/OpenMD/trunk/src/parallel/ForceMatrixDecomposition.cpp
(Generate patch)

Comparing:
branches/development/src/parallel/ForceDecomposition.cpp (file contents), Revision 1538 by chuckv, Tue Jan 11 18:58:12 2011 UTC vs.
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1571 by gezelter, Fri May 27 16:45:44 2011 UTC

# Line 1 | Line 1
1 < /**
2 < * @file ForceDecomposition.cpp
3 < * @author Charles Vardeman <cvardema.at.nd.edu>
4 < * @date 08/18/2010
5 < * @time 11:56am
6 < * @version 1.0
1 > /*
2 > * Copyright (c) 2005 The University of Notre Dame. All Rights Reserved.
3   *
8 * @section LICENSE
9 * Copyright (c) 2010 The University of Notre Dame. All Rights Reserved.
10 *
4   * The University of Notre Dame grants you ("Licensee") a
5   * non-exclusive, royalty free, license to use, modify and
6   * redistribute this software in source and binary code form, provided
# Line 45 | Line 38
38   * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
39   * [4]  Vardeman & Gezelter, in progress (2009).                        
40   */
41 + #include "parallel/ForceMatrixDecomposition.hpp"
42 + #include "math/SquareMatrix3.hpp"
43 + #include "nonbonded/NonBondedInteraction.hpp"
44 + #include "brains/SnapshotManager.hpp"
45 + #include "brains/PairList.hpp"
46  
47 + using namespace std;
48 + namespace OpenMD {
49  
50 +  /**
51 +   * distributeInitialData is essentially a copy of the older fortran
52 +   * SimulationSetup
53 +   */
54 +  
55 +  void ForceMatrixDecomposition::distributeInitialData() {
56 +    snap_ = sman_->getCurrentSnapshot();
57 +    storageLayout_ = sman_->getStorageLayout();
58 +    ff_ = info_->getForceField();
59 +    nLocal_ = snap_->getNumberOfAtoms();
60 +    nGroups_ = snap_->getNumberOfCutoffGroups();
61  
62 < /*  -*- c++ -*-  */
63 < #include "config.h"
64 < #include <stdlib.h>
62 >    // gather the information for atomtype IDs (atids):
63 >    identsLocal = info_->getIdentArray();
64 >    AtomLocalToGlobal = info_->getGlobalAtomIndices();
65 >    cgLocalToGlobal = info_->getGlobalGroupIndices();
66 >    vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
67 >    vector<RealType> massFactorsLocal = info_->getMassFactors();
68 >    PairList excludes = info_->getExcludedInteractions();
69 >    PairList oneTwo = info_->getOneTwoInteractions();
70 >    PairList oneThree = info_->getOneThreeInteractions();
71 >    PairList oneFour = info_->getOneFourInteractions();
72 >    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
73 >
74   #ifdef IS_MPI
75 < #include <mpi.h>
76 < #endif
75 >
76 >    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
77 >    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
78 >    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
79 >    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
80  
81 < #include <iostream>
82 < #include <vector>
83 < #include <algorithm>
84 < #include <cmath>
62 < #include "parallel/ForceDecomposition.hpp"
81 >    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
82 >    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
83 >    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
84 >    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
85  
86 +    cgCommIntRow = new Communicator<Row,int>(nGroups_);
87 +    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
88 +    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
89 +    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
90  
91 < using namespace std;
92 < using namespace OpenMD;
91 >    nAtomsInRow_ = AtomCommIntRow->getSize();
92 >    nAtomsInCol_ = AtomCommIntColumn->getSize();
93 >    nGroupsInRow_ = cgCommIntRow->getSize();
94 >    nGroupsInCol_ = cgCommIntColumn->getSize();
95  
96 < //__static
97 < #ifdef IS_MPI
98 < static vector<MPI:Comm> communictors;
99 < #endif
96 >    // Modify the data storage objects with the correct layouts and sizes:
97 >    atomRowData.resize(nAtomsInRow_);
98 >    atomRowData.setStorageLayout(storageLayout_);
99 >    atomColData.resize(nAtomsInCol_);
100 >    atomColData.setStorageLayout(storageLayout_);
101 >    cgRowData.resize(nGroupsInRow_);
102 >    cgRowData.setStorageLayout(DataStorage::dslPosition);
103 >    cgColData.resize(nGroupsInCol_);
104 >    cgColData.setStorageLayout(DataStorage::dslPosition);
105 >    
106 >    vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
107 >                                      vector<RealType> (nAtomsInRow_, 0.0));
108 >    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
109 >                                      vector<RealType> (nAtomsInCol_, 0.0));
110 >    
111 >    identsRow.reserve(nAtomsInRow_);
112 >    identsCol.reserve(nAtomsInCol_);
113 >    
114 >    AtomCommIntRow->gather(identsLocal, identsRow);
115 >    AtomCommIntColumn->gather(identsLocal, identsCol);
116 >    
117 >    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
118 >    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
119 >    
120 >    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
121 >    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
122  
123 < //____ MPITypeTraits
124 < template<typename T>
75 < struct MPITypeTraits;
123 >    AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
124 >    AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
125  
126 < #ifdef IS_MPI
127 < template<>
128 < struct MPITypeTraits<RealType> {
129 <  static const MPI::Datatype datatype;
130 < };
131 < const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL;
126 >    groupListRow_.clear();
127 >    groupListRow_.reserve(nGroupsInRow_);
128 >    for (int i = 0; i < nGroupsInRow_; i++) {
129 >      int gid = cgRowToGlobal[i];
130 >      for (int j = 0; j < nAtomsInRow_; j++) {
131 >        int aid = AtomRowToGlobal[j];
132 >        if (globalGroupMembership[aid] == gid)
133 >          groupListRow_[i].push_back(j);
134 >      }      
135 >    }
136  
137 < template<>
138 < struct MPITypeTraits<int> {
139 <  static const MPI::Datatype datatype;
140 < };
141 < const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT;
142 < #endif
137 >    groupListCol_.clear();
138 >    groupListCol_.reserve(nGroupsInCol_);
139 >    for (int i = 0; i < nGroupsInCol_; i++) {
140 >      int gid = cgColToGlobal[i];
141 >      for (int j = 0; j < nAtomsInCol_; j++) {
142 >        int aid = AtomColToGlobal[j];
143 >        if (globalGroupMembership[aid] == gid)
144 >          groupListCol_[i].push_back(j);
145 >      }      
146 >    }
147  
148 < /**
149 < * Constructor for ForceDecomposition Parallel Decomposition Method
150 < * Will try to construct a symmetric grid of processors. Ideally, the
151 < * number of processors will be a square ex: 4, 9, 16, 25.
152 < *
153 < */
148 >    skipsForRowAtom.clear();
149 >    skipsForRowAtom.reserve(nAtomsInRow_);
150 >    for (int i = 0; i < nAtomsInRow_; i++) {
151 >      int iglob = AtomRowToGlobal[i];
152 >      for (int j = 0; j < nAtomsInCol_; j++) {
153 >        int jglob = AtomColToGlobal[j];        
154 >        if (excludes.hasPair(iglob, jglob))
155 >          skipsForRowAtom[i].push_back(j);      
156 >      }      
157 >    }
158  
159 < ForceDecomposition::ForceDecomposition() {
159 >    toposForRowAtom.clear();
160 >    toposForRowAtom.reserve(nAtomsInRow_);
161 >    for (int i = 0; i < nAtomsInRow_; i++) {
162 >      int iglob = AtomRowToGlobal[i];
163 >      int nTopos = 0;
164 >      for (int j = 0; j < nAtomsInCol_; j++) {
165 >        int jglob = AtomColToGlobal[j];        
166 >        if (oneTwo.hasPair(iglob, jglob)) {
167 >          toposForRowAtom[i].push_back(j);
168 >          topoDistRow[i][nTopos] = 1;
169 >          nTopos++;
170 >        }
171 >        if (oneThree.hasPair(iglob, jglob)) {
172 >          toposForRowAtom[i].push_back(j);
173 >          topoDistRow[i][nTopos] = 2;
174 >          nTopos++;
175 >        }
176 >        if (oneFour.hasPair(iglob, jglob)) {
177 >          toposForRowAtom[i].push_back(j);
178 >          topoDistRow[i][nTopos] = 3;
179 >          nTopos++;
180 >        }
181 >      }      
182 >    }
183  
100 #ifdef IS_MPI
101  int nProcs = MPI::COMM_WORLD.Get_size();
102  int worldRank = MPI::COMM_WORLD.Get_rank();
184   #endif
185  
186 <  // First time through, construct column stride.
187 <  if (communicators.size() == 0)
188 <  {
189 <    int nColumnsMax = (int) round(sqrt((float) nProcs));
190 <    for (int i = 0; i < nProcs; ++i)
191 <    {
192 <      if (nProcs%i==0) nColumns=i;
186 >    groupList_.clear();
187 >    groupList_.reserve(nGroups_);
188 >    for (int i = 0; i < nGroups_; i++) {
189 >      int gid = cgLocalToGlobal[i];
190 >      for (int j = 0; j < nLocal_; j++) {
191 >        int aid = AtomLocalToGlobal[j];
192 >        if (globalGroupMembership[aid] == gid)
193 >          groupList_[i].push_back(j);
194 >      }      
195      }
196  
197 <    int nRows = nProcs/nColumns;    
198 <    myRank_ = (int) worldRank%nColumns;
116 <  }
117 <  else
118 <  {
119 <    myRank_ = myRank/nColumns;
120 <  }
121 <  MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0);
122 <  
123 <  isColumn_ = false;
124 <  
125 < }
126 <
127 < ForceDecomposition::gather(sendbuf, receivebuf){
128 <  communicators(myIndex_).Allgatherv();
129 < }
197 >    skipsForLocalAtom.clear();
198 >    skipsForLocalAtom.reserve(nLocal_);
199  
200 +    for (int i = 0; i < nLocal_; i++) {
201 +      int iglob = AtomLocalToGlobal[i];
202 +      for (int j = 0; j < nLocal_; j++) {
203 +        int jglob = AtomLocalToGlobal[j];        
204 +        if (excludes.hasPair(iglob, jglob))
205 +          skipsForLocalAtom[i].push_back(j);      
206 +      }      
207 +    }
208  
209 +    toposForLocalAtom.clear();
210 +    toposForLocalAtom.reserve(nLocal_);
211 +    for (int i = 0; i < nLocal_; i++) {
212 +      int iglob = AtomLocalToGlobal[i];
213 +      int nTopos = 0;
214 +      for (int j = 0; j < nLocal_; j++) {
215 +        int jglob = AtomLocalToGlobal[j];        
216 +        if (oneTwo.hasPair(iglob, jglob)) {
217 +          toposForLocalAtom[i].push_back(j);
218 +          topoDistLocal[i][nTopos] = 1;
219 +          nTopos++;
220 +        }
221 +        if (oneThree.hasPair(iglob, jglob)) {
222 +          toposForLocalAtom[i].push_back(j);
223 +          topoDistLocal[i][nTopos] = 2;
224 +          nTopos++;
225 +        }
226 +        if (oneFour.hasPair(iglob, jglob)) {
227 +          toposForLocalAtom[i].push_back(j);
228 +          topoDistLocal[i][nTopos] = 3;
229 +          nTopos++;
230 +        }
231 +      }      
232 +    }
233 +  }
234 +  
235 +  void ForceMatrixDecomposition::distributeData()  {
236 +    snap_ = sman_->getCurrentSnapshot();
237 +    storageLayout_ = sman_->getStorageLayout();
238 + #ifdef IS_MPI
239 +    
240 +    // gather up the atomic positions
241 +    AtomCommVectorRow->gather(snap_->atomData.position,
242 +                              atomRowData.position);
243 +    AtomCommVectorColumn->gather(snap_->atomData.position,
244 +                                 atomColData.position);
245 +    
246 +    // gather up the cutoff group positions
247 +    cgCommVectorRow->gather(snap_->cgData.position,
248 +                            cgRowData.position);
249 +    cgCommVectorColumn->gather(snap_->cgData.position,
250 +                               cgColData.position);
251 +    
252 +    // if needed, gather the atomic rotation matrices
253 +    if (storageLayout_ & DataStorage::dslAmat) {
254 +      AtomCommMatrixRow->gather(snap_->atomData.aMat,
255 +                                atomRowData.aMat);
256 +      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
257 +                                   atomColData.aMat);
258 +    }
259 +    
260 +    // if needed, gather the atomic eletrostatic frames
261 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
262 +      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
263 +                                atomRowData.electroFrame);
264 +      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
265 +                                   atomColData.electroFrame);
266 +    }
267 + #endif      
268 +  }
269 +  
270 +  void ForceMatrixDecomposition::collectIntermediateData() {
271 +    snap_ = sman_->getCurrentSnapshot();
272 +    storageLayout_ = sman_->getStorageLayout();
273 + #ifdef IS_MPI
274 +    
275 +    if (storageLayout_ & DataStorage::dslDensity) {
276 +      
277 +      AtomCommRealRow->scatter(atomRowData.density,
278 +                               snap_->atomData.density);
279 +      
280 +      int n = snap_->atomData.density.size();
281 +      std::vector<RealType> rho_tmp(n, 0.0);
282 +      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
283 +      for (int i = 0; i < n; i++)
284 +        snap_->atomData.density[i] += rho_tmp[i];
285 +    }
286 + #endif
287 +  }
288 +  
289 +  void ForceMatrixDecomposition::distributeIntermediateData() {
290 +    snap_ = sman_->getCurrentSnapshot();
291 +    storageLayout_ = sman_->getStorageLayout();
292 + #ifdef IS_MPI
293 +    if (storageLayout_ & DataStorage::dslFunctional) {
294 +      AtomCommRealRow->gather(snap_->atomData.functional,
295 +                              atomRowData.functional);
296 +      AtomCommRealColumn->gather(snap_->atomData.functional,
297 +                                 atomColData.functional);
298 +    }
299 +    
300 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
301 +      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
302 +                              atomRowData.functionalDerivative);
303 +      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
304 +                                 atomColData.functionalDerivative);
305 +    }
306 + #endif
307 +  }
308 +  
309 +  
310 +  void ForceMatrixDecomposition::collectData() {
311 +    snap_ = sman_->getCurrentSnapshot();
312 +    storageLayout_ = sman_->getStorageLayout();
313 + #ifdef IS_MPI    
314 +    int n = snap_->atomData.force.size();
315 +    vector<Vector3d> frc_tmp(n, V3Zero);
316 +    
317 +    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
318 +    for (int i = 0; i < n; i++) {
319 +      snap_->atomData.force[i] += frc_tmp[i];
320 +      frc_tmp[i] = 0.0;
321 +    }
322 +    
323 +    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
324 +    for (int i = 0; i < n; i++)
325 +      snap_->atomData.force[i] += frc_tmp[i];
326 +    
327 +    
328 +    if (storageLayout_ & DataStorage::dslTorque) {
329  
330 < ForceDecomposition::scatter(sbuffer, rbuffer){
331 <  communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM);
332 < }
330 >      int nt = snap_->atomData.force.size();
331 >      vector<Vector3d> trq_tmp(nt, V3Zero);
332 >
333 >      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
334 >      for (int i = 0; i < n; i++) {
335 >        snap_->atomData.torque[i] += trq_tmp[i];
336 >        trq_tmp[i] = 0.0;
337 >      }
338 >      
339 >      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
340 >      for (int i = 0; i < n; i++)
341 >        snap_->atomData.torque[i] += trq_tmp[i];
342 >    }
343 >    
344 >    nLocal_ = snap_->getNumberOfAtoms();
345 >
346 >    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
347 >                                       vector<RealType> (nLocal_, 0.0));
348 >    
349 >    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
350 >      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
351 >      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
352 >        pot_local[i] += pot_temp[i][ii];
353 >      }
354 >    }
355 > #endif
356 >  }
357 >
358 >  int ForceMatrixDecomposition::getNAtomsInRow() {  
359 > #ifdef IS_MPI
360 >    return nAtomsInRow_;
361 > #else
362 >    return nLocal_;
363 > #endif
364 >  }
365 >
366 >  /**
367 >   * returns the list of atoms belonging to this group.  
368 >   */
369 >  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
370 > #ifdef IS_MPI
371 >    return groupListRow_[cg1];
372 > #else
373 >    return groupList_[cg1];
374 > #endif
375 >  }
376 >
377 >  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
378 > #ifdef IS_MPI
379 >    return groupListCol_[cg2];
380 > #else
381 >    return groupList_[cg2];
382 > #endif
383 >  }
384 >  
385 >  Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
386 >    Vector3d d;
387 >    
388 > #ifdef IS_MPI
389 >    d = cgColData.position[cg2] - cgRowData.position[cg1];
390 > #else
391 >    d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
392 > #endif
393 >    
394 >    snap_->wrapVector(d);
395 >    return d;    
396 >  }
397 >
398 >
399 >  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
400 >
401 >    Vector3d d;
402 >    
403 > #ifdef IS_MPI
404 >    d = cgRowData.position[cg1] - atomRowData.position[atom1];
405 > #else
406 >    d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
407 > #endif
408 >
409 >    snap_->wrapVector(d);
410 >    return d;    
411 >  }
412 >  
413 >  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
414 >    Vector3d d;
415 >    
416 > #ifdef IS_MPI
417 >    d = cgColData.position[cg2] - atomColData.position[atom2];
418 > #else
419 >    d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
420 > #endif
421 >    
422 >    snap_->wrapVector(d);
423 >    return d;    
424 >  }
425  
426 +  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
427 + #ifdef IS_MPI
428 +    return massFactorsRow[atom1];
429 + #else
430 +    return massFactorsLocal[atom1];
431 + #endif
432 +  }
433  
434 +  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
435 + #ifdef IS_MPI
436 +    return massFactorsCol[atom2];
437 + #else
438 +    return massFactorsLocal[atom2];
439 + #endif
440 +
441 +  }
442 +    
443 +  Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
444 +    Vector3d d;
445 +    
446 + #ifdef IS_MPI
447 +    d = atomColData.position[atom2] - atomRowData.position[atom1];
448 + #else
449 +    d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
450 + #endif
451 +
452 +    snap_->wrapVector(d);
453 +    return d;    
454 +  }
455 +
456 +  vector<int> ForceMatrixDecomposition::getSkipsForRowAtom(int atom1) {
457 + #ifdef IS_MPI
458 +    return skipsForRowAtom[atom1];
459 + #else
460 +    return skipsForLocalAtom[atom1];
461 + #endif
462 +  }
463 +
464 +  /**
465 +   * there are a number of reasons to skip a pair or a particle mostly
466 +   * we do this to exclude atoms who are involved in short range
467 +   * interactions (bonds, bends, torsions), but we also need to
468 +   * exclude some overcounted interactions that result from the
469 +   * parallel decomposition.
470 +   */
471 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
472 +    int unique_id_1, unique_id_2;
473 +
474 + #ifdef IS_MPI
475 +    // in MPI, we have to look up the unique IDs for each atom
476 +    unique_id_1 = AtomRowToGlobal[atom1];
477 +    unique_id_2 = AtomColToGlobal[atom2];
478 +
479 +    // this situation should only arise in MPI simulations
480 +    if (unique_id_1 == unique_id_2) return true;
481 +    
482 +    // this prevents us from doing the pair on multiple processors
483 +    if (unique_id_1 < unique_id_2) {
484 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
485 +    } else {
486 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
487 +    }
488 + #else
489 +    // in the normal loop, the atom numbers are unique
490 +    unique_id_1 = atom1;
491 +    unique_id_2 = atom2;
492 + #endif
493 +    
494 + #ifdef IS_MPI
495 +    for (vector<int>::iterator i = skipsForRowAtom[atom1].begin();
496 +         i != skipsForRowAtom[atom1].end(); ++i) {
497 +      if ( (*i) == unique_id_2 ) return true;
498 +    }    
499 + #else
500 +    for (vector<int>::iterator i = skipsForLocalAtom[atom1].begin();
501 +         i != skipsForLocalAtom[atom1].end(); ++i) {
502 +      if ( (*i) == unique_id_2 ) return true;
503 +    }    
504 + #endif
505 +  }
506 +
507 +  int ForceMatrixDecomposition::getTopoDistance(int atom1, int atom2) {
508 +    
509 + #ifdef IS_MPI
510 +    for (int i = 0; i < toposForRowAtom[atom1].size(); i++) {
511 +      if ( toposForRowAtom[atom1][i] == atom2 ) return topoDistRow[atom1][i];
512 +    }
513 + #else
514 +    for (int i = 0; i < toposForLocalAtom[atom1].size(); i++) {
515 +      if ( toposForLocalAtom[atom1][i] == atom2 ) return topoDistLocal[atom1][i];
516 +    }
517 + #endif
518 +
519 +    // zero is default for unconnected (i.e. normal) pair interactions
520 +    return 0;
521 +  }
522 +
523 +  void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
524 + #ifdef IS_MPI
525 +    atomRowData.force[atom1] += fg;
526 + #else
527 +    snap_->atomData.force[atom1] += fg;
528 + #endif
529 +  }
530 +
531 +  void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
532 + #ifdef IS_MPI
533 +    atomColData.force[atom2] += fg;
534 + #else
535 +    snap_->atomData.force[atom2] += fg;
536 + #endif
537 +  }
538 +
539 +    // filling interaction blocks with pointers
540 +  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
541 +    InteractionData idat;
542 +
543 + #ifdef IS_MPI
544 +    
545 +    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
546 +                             ff_->getAtomType(identsCol[atom2]) );
547 +
548 +    if (storageLayout_ & DataStorage::dslAmat) {
549 +      idat.A1 = &(atomRowData.aMat[atom1]);
550 +      idat.A2 = &(atomColData.aMat[atom2]);
551 +    }
552 +    
553 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
554 +      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
555 +      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
556 +    }
557 +
558 +    if (storageLayout_ & DataStorage::dslTorque) {
559 +      idat.t1 = &(atomRowData.torque[atom1]);
560 +      idat.t2 = &(atomColData.torque[atom2]);
561 +    }
562 +
563 +    if (storageLayout_ & DataStorage::dslDensity) {
564 +      idat.rho1 = &(atomRowData.density[atom1]);
565 +      idat.rho2 = &(atomColData.density[atom2]);
566 +    }
567 +
568 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
569 +      idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
570 +      idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
571 +    }
572 +
573 + #else
574 +
575 +    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
576 +                             ff_->getAtomType(identsLocal[atom2]) );
577 +
578 +    if (storageLayout_ & DataStorage::dslAmat) {
579 +      idat.A1 = &(snap_->atomData.aMat[atom1]);
580 +      idat.A2 = &(snap_->atomData.aMat[atom2]);
581 +    }
582 +
583 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
584 +      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
585 +      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
586 +    }
587 +
588 +    if (storageLayout_ & DataStorage::dslTorque) {
589 +      idat.t1 = &(snap_->atomData.torque[atom1]);
590 +      idat.t2 = &(snap_->atomData.torque[atom2]);
591 +    }
592 +
593 +    if (storageLayout_ & DataStorage::dslDensity) {
594 +      idat.rho1 = &(snap_->atomData.density[atom1]);
595 +      idat.rho2 = &(snap_->atomData.density[atom2]);
596 +    }
597 +
598 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
599 +      idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
600 +      idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
601 +    }
602 + #endif
603 +    return idat;
604 +  }
605 +
606 +  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
607 +
608 +    InteractionData idat;
609 + #ifdef IS_MPI
610 +    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
611 +                             ff_->getAtomType(identsCol[atom2]) );
612 +
613 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
614 +      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
615 +      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
616 +    }
617 +    if (storageLayout_ & DataStorage::dslTorque) {
618 +      idat.t1 = &(atomRowData.torque[atom1]);
619 +      idat.t2 = &(atomColData.torque[atom2]);
620 +    }
621 +    if (storageLayout_ & DataStorage::dslForce) {
622 +      idat.t1 = &(atomRowData.force[atom1]);
623 +      idat.t2 = &(atomColData.force[atom2]);
624 +    }
625 + #else
626 +    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
627 +                             ff_->getAtomType(identsLocal[atom2]) );
628 +
629 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
630 +      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
631 +      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
632 +    }
633 +    if (storageLayout_ & DataStorage::dslTorque) {
634 +      idat.t1 = &(snap_->atomData.torque[atom1]);
635 +      idat.t2 = &(snap_->atomData.torque[atom2]);
636 +    }
637 +    if (storageLayout_ & DataStorage::dslForce) {
638 +      idat.t1 = &(snap_->atomData.force[atom1]);
639 +      idat.t2 = &(snap_->atomData.force[atom2]);
640 +    }
641 + #endif    
642 +  }
643 +
644 +  /*
645 +   * buildNeighborList
646 +   *
647 +   * first element of pair is row-indexed CutoffGroup
648 +   * second element of pair is column-indexed CutoffGroup
649 +   */
650 +  vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
651 +      
652 +    vector<pair<int, int> > neighborList;
653 + #ifdef IS_MPI
654 +    cellListRow_.clear();
655 +    cellListCol_.clear();
656 + #else
657 +    cellList_.clear();
658 + #endif
659 +
660 +    // dangerous to not do error checking.
661 +    RealType rCut_;
662 +
663 +    RealType rList_ = (rCut_ + skinThickness_);
664 +    RealType rl2 = rList_ * rList_;
665 +    Snapshot* snap_ = sman_->getCurrentSnapshot();
666 +    Mat3x3d Hmat = snap_->getHmat();
667 +    Vector3d Hx = Hmat.getColumn(0);
668 +    Vector3d Hy = Hmat.getColumn(1);
669 +    Vector3d Hz = Hmat.getColumn(2);
670 +
671 +    nCells_.x() = (int) ( Hx.length() )/ rList_;
672 +    nCells_.y() = (int) ( Hy.length() )/ rList_;
673 +    nCells_.z() = (int) ( Hz.length() )/ rList_;
674 +
675 +    Mat3x3d invHmat = snap_->getInvHmat();
676 +    Vector3d rs, scaled, dr;
677 +    Vector3i whichCell;
678 +    int cellIndex;
679 +
680 + #ifdef IS_MPI
681 +    for (int i = 0; i < nGroupsInRow_; i++) {
682 +      rs = cgRowData.position[i];
683 +      // scaled positions relative to the box vectors
684 +      scaled = invHmat * rs;
685 +      // wrap the vector back into the unit box by subtracting integer box
686 +      // numbers
687 +      for (int j = 0; j < 3; j++)
688 +        scaled[j] -= roundMe(scaled[j]);
689 +    
690 +      // find xyz-indices of cell that cutoffGroup is in.
691 +      whichCell.x() = nCells_.x() * scaled.x();
692 +      whichCell.y() = nCells_.y() * scaled.y();
693 +      whichCell.z() = nCells_.z() * scaled.z();
694 +
695 +      // find single index of this cell:
696 +      cellIndex = Vlinear(whichCell, nCells_);
697 +      // add this cutoff group to the list of groups in this cell;
698 +      cellListRow_[cellIndex].push_back(i);
699 +    }
700 +
701 +    for (int i = 0; i < nGroupsInCol_; i++) {
702 +      rs = cgColData.position[i];
703 +      // scaled positions relative to the box vectors
704 +      scaled = invHmat * rs;
705 +      // wrap the vector back into the unit box by subtracting integer box
706 +      // numbers
707 +      for (int j = 0; j < 3; j++)
708 +        scaled[j] -= roundMe(scaled[j]);
709 +
710 +      // find xyz-indices of cell that cutoffGroup is in.
711 +      whichCell.x() = nCells_.x() * scaled.x();
712 +      whichCell.y() = nCells_.y() * scaled.y();
713 +      whichCell.z() = nCells_.z() * scaled.z();
714 +
715 +      // find single index of this cell:
716 +      cellIndex = Vlinear(whichCell, nCells_);
717 +      // add this cutoff group to the list of groups in this cell;
718 +      cellListCol_[cellIndex].push_back(i);
719 +    }
720 + #else
721 +    for (int i = 0; i < nGroups_; i++) {
722 +      rs = snap_->cgData.position[i];
723 +      // scaled positions relative to the box vectors
724 +      scaled = invHmat * rs;
725 +      // wrap the vector back into the unit box by subtracting integer box
726 +      // numbers
727 +      for (int j = 0; j < 3; j++)
728 +        scaled[j] -= roundMe(scaled[j]);
729 +
730 +      // find xyz-indices of cell that cutoffGroup is in.
731 +      whichCell.x() = nCells_.x() * scaled.x();
732 +      whichCell.y() = nCells_.y() * scaled.y();
733 +      whichCell.z() = nCells_.z() * scaled.z();
734 +
735 +      // find single index of this cell:
736 +      cellIndex = Vlinear(whichCell, nCells_);
737 +      // add this cutoff group to the list of groups in this cell;
738 +      cellList_[cellIndex].push_back(i);
739 +    }
740 + #endif
741 +
742 +
743 +
744 +    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
745 +      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
746 +        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
747 +          Vector3i m1v(m1x, m1y, m1z);
748 +          int m1 = Vlinear(m1v, nCells_);
749 +
750 +          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
751 +               os != cellOffsets_.end(); ++os) {
752 +            
753 +            Vector3i m2v = m1v + (*os);
754 +            
755 +            if (m2v.x() >= nCells_.x()) {
756 +              m2v.x() = 0;          
757 +            } else if (m2v.x() < 0) {
758 +              m2v.x() = nCells_.x() - 1;
759 +            }
760 +            
761 +            if (m2v.y() >= nCells_.y()) {
762 +              m2v.y() = 0;          
763 +            } else if (m2v.y() < 0) {
764 +              m2v.y() = nCells_.y() - 1;
765 +            }
766 +            
767 +            if (m2v.z() >= nCells_.z()) {
768 +              m2v.z() = 0;          
769 +            } else if (m2v.z() < 0) {
770 +              m2v.z() = nCells_.z() - 1;
771 +            }
772 +            
773 +            int m2 = Vlinear (m2v, nCells_);
774 +
775 + #ifdef IS_MPI
776 +            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
777 +                 j1 != cellListRow_[m1].end(); ++j1) {
778 +              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
779 +                   j2 != cellListCol_[m2].end(); ++j2) {
780 +                              
781 +                // Always do this if we're in different cells or if
782 +                // we're in the same cell and the global index of the
783 +                // j2 cutoff group is less than the j1 cutoff group
784 +
785 +                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
786 +                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
787 +                  snap_->wrapVector(dr);
788 +                  if (dr.lengthSquare() < rl2) {
789 +                    neighborList.push_back(make_pair((*j1), (*j2)));
790 +                  }
791 +                }
792 +              }
793 +            }
794 + #else
795 +            for (vector<int>::iterator j1 = cellList_[m1].begin();
796 +                 j1 != cellList_[m1].end(); ++j1) {
797 +              for (vector<int>::iterator j2 = cellList_[m2].begin();
798 +                   j2 != cellList_[m2].end(); ++j2) {
799 +                              
800 +                // Always do this if we're in different cells or if
801 +                // we're in the same cell and the global index of the
802 +                // j2 cutoff group is less than the j1 cutoff group
803 +
804 +                if (m2 != m1 || (*j2) < (*j1)) {
805 +                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
806 +                  snap_->wrapVector(dr);
807 +                  if (dr.lengthSquare() < rl2) {
808 +                    neighborList.push_back(make_pair((*j1), (*j2)));
809 +                  }
810 +                }
811 +              }
812 +            }
813 + #endif
814 +          }
815 +        }
816 +      }
817 +    }
818 +
819 +    // save the local cutoff group positions for the check that is
820 +    // done on each loop:
821 +    saved_CG_positions_.clear();
822 +    for (int i = 0; i < nGroups_; i++)
823 +      saved_CG_positions_.push_back(snap_->cgData.position[i]);
824 +
825 +    return neighborList;
826 +  }
827 + } //end namespace OpenMD

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines