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/ForceMatrixDecomposition.cpp (file contents):
Revision 1581 by gezelter, Mon Jun 13 22:13:12 2011 UTC vs.
Revision 1593 by gezelter, Fri Jul 15 21:35:14 2011 UTC

# Line 47 | Line 47 | namespace OpenMD {
47   using namespace std;
48   namespace OpenMD {
49  
50 +  ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
51 +
52 +    // In a parallel computation, row and colum scans must visit all
53 +    // surrounding cells (not just the 14 upper triangular blocks that
54 +    // are used when the processor can see all pairs)
55 + #ifdef IS_MPI
56 +    cellOffsets_.push_back( Vector3i(-1, 0, 0) );
57 +    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
58 +    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
59 +    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
60 +    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
61 +    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
62 +    cellOffsets_.push_back( Vector3i(-1,-1,-1) );
63 +    cellOffsets_.push_back( Vector3i( 0,-1,-1) );
64 +    cellOffsets_.push_back( Vector3i( 1,-1,-1) );
65 +    cellOffsets_.push_back( Vector3i( 1, 0,-1) );
66 +    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67 +    cellOffsets_.push_back( Vector3i( 0, 1,-1) );
68 +    cellOffsets_.push_back( Vector3i(-1, 1,-1) );
69 + #endif    
70 +  }
71 +
72 +
73    /**
74     * distributeInitialData is essentially a copy of the older fortran
75     * SimulationSetup
76     */
54 –  
77    void ForceMatrixDecomposition::distributeInitialData() {
78      snap_ = sman_->getCurrentSnapshot();
79      storageLayout_ = sman_->getStorageLayout();
80      ff_ = info_->getForceField();
81      nLocal_ = snap_->getNumberOfAtoms();
82 <
82 >    
83      nGroups_ = info_->getNLocalCutoffGroups();
62 –    cerr << "in dId, nGroups = " << nGroups_ << "\n";
84      // gather the information for atomtype IDs (atids):
85 <    identsLocal = info_->getIdentArray();
85 >    idents = info_->getIdentArray();
86      AtomLocalToGlobal = info_->getGlobalAtomIndices();
87      cgLocalToGlobal = info_->getGlobalGroupIndices();
88      vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
89 +
90      massFactors = info_->getMassFactors();
69 –    PairList excludes = info_->getExcludedInteractions();
70 –    PairList oneTwo = info_->getOneTwoInteractions();
71 –    PairList oneThree = info_->getOneThreeInteractions();
72 –    PairList oneFour = info_->getOneFourInteractions();
91  
92 +    PairList* excludes = info_->getExcludedInteractions();
93 +    PairList* oneTwo = info_->getOneTwoInteractions();
94 +    PairList* oneThree = info_->getOneThreeInteractions();
95 +    PairList* oneFour = info_->getOneFourInteractions();
96 +
97   #ifdef IS_MPI
98  
99 <    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
100 <    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
78 <    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
79 <    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
80 <    AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
99 >    MPI::Intracomm row = rowComm.getComm();
100 >    MPI::Intracomm col = colComm.getComm();
101  
102 <    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
103 <    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
104 <    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
105 <    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
106 <    AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
102 >    AtomPlanIntRow = new Plan<int>(row, nLocal_);
103 >    AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
104 >    AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
105 >    AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
106 >    AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
107  
108 <    cgCommIntRow = new Communicator<Row,int>(nGroups_);
109 <    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
110 <    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
111 <    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
108 >    AtomPlanIntColumn = new Plan<int>(col, nLocal_);
109 >    AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
110 >    AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
111 >    AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
112 >    AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
113  
114 <    nAtomsInRow_ = AtomCommIntRow->getSize();
115 <    nAtomsInCol_ = AtomCommIntColumn->getSize();
116 <    nGroupsInRow_ = cgCommIntRow->getSize();
117 <    nGroupsInCol_ = cgCommIntColumn->getSize();
114 >    cgPlanIntRow = new Plan<int>(row, nGroups_);
115 >    cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
116 >    cgPlanIntColumn = new Plan<int>(col, nGroups_);
117 >    cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
118  
119 +    nAtomsInRow_ = AtomPlanIntRow->getSize();
120 +    nAtomsInCol_ = AtomPlanIntColumn->getSize();
121 +    nGroupsInRow_ = cgPlanIntRow->getSize();
122 +    nGroupsInCol_ = cgPlanIntColumn->getSize();
123 +
124      // Modify the data storage objects with the correct layouts and sizes:
125      atomRowData.resize(nAtomsInRow_);
126      atomRowData.setStorageLayout(storageLayout_);
# Line 108 | Line 134 | namespace OpenMD {
134      identsRow.resize(nAtomsInRow_);
135      identsCol.resize(nAtomsInCol_);
136      
137 <    AtomCommIntRow->gather(identsLocal, identsRow);
138 <    AtomCommIntColumn->gather(identsLocal, identsCol);
137 >    AtomPlanIntRow->gather(idents, identsRow);
138 >    AtomPlanIntColumn->gather(idents, identsCol);
139      
140 <    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
141 <    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
142 <    
117 <    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
118 <    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
140 >    // allocate memory for the parallel objects
141 >    atypesRow.resize(nAtomsInRow_);
142 >    atypesCol.resize(nAtomsInCol_);
143  
144 <    AtomCommRealRow->gather(massFactors, massFactorsRow);
145 <    AtomCommRealColumn->gather(massFactors, massFactorsCol);
144 >    for (int i = 0; i < nAtomsInRow_; i++)
145 >      atypesRow[i] = ff_->getAtomType(identsRow[i]);
146 >    for (int i = 0; i < nAtomsInCol_; i++)
147 >      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
148  
149 +    pot_row.resize(nAtomsInRow_);
150 +    pot_col.resize(nAtomsInCol_);
151 +
152 +    AtomRowToGlobal.resize(nAtomsInRow_);
153 +    AtomColToGlobal.resize(nAtomsInCol_);
154 +    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
155 +    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
156 +
157 +    cerr << "Atoms in Local:\n";
158 +    for (int i = 0; i < AtomLocalToGlobal.size(); i++) {
159 +      cerr << "i =\t" << i << "\t localAt =\t" << AtomLocalToGlobal[i] << "\n";
160 +    }
161 +    cerr << "Atoms in Row:\n";
162 +    for (int i = 0; i < AtomRowToGlobal.size(); i++) {
163 +      cerr << "i =\t" << i << "\t rowAt =\t" << AtomRowToGlobal[i] << "\n";
164 +    }
165 +    cerr << "Atoms in Col:\n";
166 +    for (int i = 0; i < AtomColToGlobal.size(); i++) {
167 +      cerr << "i =\t" << i << "\t colAt =\t" << AtomColToGlobal[i] << "\n";
168 +    }
169 +
170 +    cgRowToGlobal.resize(nGroupsInRow_);
171 +    cgColToGlobal.resize(nGroupsInCol_);
172 +    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
173 +    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
174 +
175 +    cerr << "Gruops in Local:\n";
176 +    for (int i = 0; i < cgLocalToGlobal.size(); i++) {
177 +      cerr << "i =\t" << i << "\t localCG =\t" << cgLocalToGlobal[i] << "\n";
178 +    }
179 +    cerr << "Groups in Row:\n";
180 +    for (int i = 0; i < cgRowToGlobal.size(); i++) {
181 +      cerr << "i =\t" << i << "\t rowCG =\t" << cgRowToGlobal[i] << "\n";
182 +    }
183 +    cerr << "Groups in Col:\n";
184 +    for (int i = 0; i < cgColToGlobal.size(); i++) {
185 +      cerr << "i =\t" << i << "\t colCG =\t" << cgColToGlobal[i] << "\n";
186 +    }
187 +
188 +
189 +    massFactorsRow.resize(nAtomsInRow_);
190 +    massFactorsCol.resize(nAtomsInCol_);
191 +    AtomPlanRealRow->gather(massFactors, massFactorsRow);
192 +    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
193 +
194      groupListRow_.clear();
195      groupListRow_.resize(nGroupsInRow_);
196      for (int i = 0; i < nGroupsInRow_; i++) {
# Line 142 | Line 213 | namespace OpenMD {
213        }      
214      }
215  
216 <    skipsForAtom.clear();
217 <    skipsForAtom.resize(nAtomsInRow_);
216 >    excludesForAtom.clear();
217 >    excludesForAtom.resize(nAtomsInRow_);
218      toposForAtom.clear();
219      toposForAtom.resize(nAtomsInRow_);
220      topoDist.clear();
# Line 154 | Line 225 | namespace OpenMD {
225        for (int j = 0; j < nAtomsInCol_; j++) {
226          int jglob = AtomColToGlobal[j];
227  
228 <        if (excludes.hasPair(iglob, jglob))
229 <          skipsForAtom[i].push_back(j);      
228 >        if (excludes->hasPair(iglob, jglob))
229 >          excludesForAtom[i].push_back(j);      
230          
231 <        if (oneTwo.hasPair(iglob, jglob)) {
231 >        if (oneTwo->hasPair(iglob, jglob)) {
232            toposForAtom[i].push_back(j);
233            topoDist[i].push_back(1);
234          } else {
235 <          if (oneThree.hasPair(iglob, jglob)) {
235 >          if (oneThree->hasPair(iglob, jglob)) {
236              toposForAtom[i].push_back(j);
237              topoDist[i].push_back(2);
238            } else {
239 <            if (oneFour.hasPair(iglob, jglob)) {
239 >            if (oneFour->hasPair(iglob, jglob)) {
240                toposForAtom[i].push_back(j);
241                topoDist[i].push_back(3);
242              }
# Line 176 | Line 247 | namespace OpenMD {
247  
248   #endif
249  
250 +    // allocate memory for the parallel objects
251 +    atypesLocal.resize(nLocal_);
252 +
253 +    for (int i = 0; i < nLocal_; i++)
254 +      atypesLocal[i] = ff_->getAtomType(idents[i]);
255 +
256      groupList_.clear();
257      groupList_.resize(nGroups_);
258      for (int i = 0; i < nGroups_; i++) {
# Line 188 | Line 265 | namespace OpenMD {
265        }      
266      }
267  
268 <    skipsForAtom.clear();
269 <    skipsForAtom.resize(nLocal_);
268 >    excludesForAtom.clear();
269 >    excludesForAtom.resize(nLocal_);
270      toposForAtom.clear();
271      toposForAtom.resize(nLocal_);
272      topoDist.clear();
# Line 201 | Line 278 | namespace OpenMD {
278        for (int j = 0; j < nLocal_; j++) {
279          int jglob = AtomLocalToGlobal[j];
280  
281 <        if (excludes.hasPair(iglob, jglob))
282 <          skipsForAtom[i].push_back(j);              
281 >        if (excludes->hasPair(iglob, jglob))
282 >          excludesForAtom[i].push_back(j);              
283          
284 <        if (oneTwo.hasPair(iglob, jglob)) {
284 >        if (oneTwo->hasPair(iglob, jglob)) {
285            toposForAtom[i].push_back(j);
286            topoDist[i].push_back(1);
287          } else {
288 <          if (oneThree.hasPair(iglob, jglob)) {
288 >          if (oneThree->hasPair(iglob, jglob)) {
289              toposForAtom[i].push_back(j);
290              topoDist[i].push_back(2);
291            } else {
292 <            if (oneFour.hasPair(iglob, jglob)) {
292 >            if (oneFour->hasPair(iglob, jglob)) {
293                toposForAtom[i].push_back(j);
294                topoDist[i].push_back(3);
295              }
# Line 222 | Line 299 | namespace OpenMD {
299      }
300      
301      createGtypeCutoffMap();
302 +
303    }
304    
305    void ForceMatrixDecomposition::createGtypeCutoffMap() {
306 <
306 >    
307      RealType tol = 1e-6;
308 +    largestRcut_ = 0.0;
309      RealType rc;
310      int atid;
311      set<AtomType*> atypes = info_->getSimulatedAtomTypes();
312 <    vector<RealType> atypeCutoff;
313 <    atypeCutoff.resize( atypes.size() );
314 <
312 >    
313 >    map<int, RealType> atypeCutoff;
314 >      
315      for (set<AtomType*>::iterator at = atypes.begin();
316           at != atypes.end(); ++at){
238 –      rc = interactionMan_->getSuggestedCutoffRadius(*at);
317        atid = (*at)->getIdent();
318 <      atypeCutoff[atid] = rc;
318 >      if (userChoseCutoff_)
319 >        atypeCutoff[atid] = userCutoff_;
320 >      else
321 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
322      }
323 <
323 >    
324      vector<RealType> gTypeCutoffs;
244 –
325      // first we do a single loop over the cutoff groups to find the
326      // largest cutoff for any atypes present in this group.
327   #ifdef IS_MPI
# Line 299 | Line 379 | namespace OpenMD {
379  
380      vector<RealType> groupCutoff(nGroups_, 0.0);
381      groupToGtype.resize(nGroups_);
302 –
303 –    cerr << "nGroups = " << nGroups_ << "\n";
382      for (int cg1 = 0; cg1 < nGroups_; cg1++) {
305 –
383        groupCutoff[cg1] = 0.0;
384        vector<int> atomList = getAtomsInGroupRow(cg1);
308 –
385        for (vector<int>::iterator ia = atomList.begin();
386             ia != atomList.end(); ++ia) {            
387          int atom1 = (*ia);
388 <        atid = identsLocal[atom1];
389 <        if (atypeCutoff[atid] > groupCutoff[cg1]) {
388 >        atid = idents[atom1];
389 >        if (atypeCutoff[atid] > groupCutoff[cg1])
390            groupCutoff[cg1] = atypeCutoff[atid];
315 –        }
391        }
392 <
392 >      
393        bool gTypeFound = false;
394        for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
395          if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
# Line 322 | Line 397 | namespace OpenMD {
397            gTypeFound = true;
398          }
399        }
400 <      if (!gTypeFound) {
400 >      if (!gTypeFound) {      
401          gTypeCutoffs.push_back( groupCutoff[cg1] );
402          groupToGtype[cg1] = gTypeCutoffs.size() - 1;
403        }      
404      }
405   #endif
406  
332 –    cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
407      // Now we find the maximum group cutoff value present in the simulation
408  
409 <    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
409 >    RealType groupMax = *max_element(gTypeCutoffs.begin(),
410 >                                     gTypeCutoffs.end());
411  
412   #ifdef IS_MPI
413 <    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
413 >    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
414 >                              MPI::MAX);
415   #endif
416      
417      RealType tradRcut = groupMax;
# Line 365 | Line 441 | namespace OpenMD {
441  
442          pair<int,int> key = make_pair(i,j);
443          gTypeCutoffMap[key].first = thisRcut;
368 –
444          if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
370 –
445          gTypeCutoffMap[key].second = thisRcut*thisRcut;
372 –        
446          gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
374 –
447          // sanity check
448          
449          if (userChoseCutoff_) {
450            if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
451              sprintf(painCave.errMsg,
452                      "ForceMatrixDecomposition::createGtypeCutoffMap "
453 <                    "user-specified rCut does not match computed group Cutoff\n");
453 >                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
454              painCave.severity = OPENMD_ERROR;
455              painCave.isFatal = 1;
456              simError();            
# Line 410 | Line 482 | namespace OpenMD {
482    }
483  
484    void ForceMatrixDecomposition::zeroWorkArrays() {
485 +    pairwisePot = 0.0;
486 +    embeddingPot = 0.0;
487  
414 –    for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
415 –      longRangePot_[j] = 0.0;
416 –    }
417 –
488   #ifdef IS_MPI
489      if (storageLayout_ & DataStorage::dslForce) {
490        fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
# Line 430 | Line 500 | namespace OpenMD {
500           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
501  
502      fill(pot_col.begin(), pot_col.end(),
503 <         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
434 <    
435 <    pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
503 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
504  
505      if (storageLayout_ & DataStorage::dslParticlePot) {    
506 <      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
507 <      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
506 >      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
507 >           0.0);
508 >      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
509 >           0.0);
510      }
511  
512      if (storageLayout_ & DataStorage::dslDensity) {      
# Line 445 | Line 515 | namespace OpenMD {
515      }
516  
517      if (storageLayout_ & DataStorage::dslFunctional) {  
518 <      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
519 <      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
518 >      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
519 >           0.0);
520 >      fill(atomColData.functional.begin(), atomColData.functional.end(),
521 >           0.0);
522      }
523  
524      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
# Line 456 | Line 528 | namespace OpenMD {
528             atomColData.functionalDerivative.end(), 0.0);
529      }
530  
531 < #else
532 <    
531 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
532 >      fill(atomRowData.skippedCharge.begin(),
533 >           atomRowData.skippedCharge.end(), 0.0);
534 >      fill(atomColData.skippedCharge.begin(),
535 >           atomColData.skippedCharge.end(), 0.0);
536 >    }
537 >
538 > #endif
539 >    // even in parallel, we need to zero out the local arrays:
540 >
541      if (storageLayout_ & DataStorage::dslParticlePot) {      
542        fill(snap_->atomData.particlePot.begin(),
543             snap_->atomData.particlePot.end(), 0.0);
# Line 475 | Line 555 | namespace OpenMD {
555        fill(snap_->atomData.functionalDerivative.begin(),
556             snap_->atomData.functionalDerivative.end(), 0.0);
557      }
558 < #endif
558 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
559 >      fill(snap_->atomData.skippedCharge.begin(),
560 >           snap_->atomData.skippedCharge.end(), 0.0);
561 >    }
562      
563    }
564  
# Line 486 | Line 569 | namespace OpenMD {
569   #ifdef IS_MPI
570      
571      // gather up the atomic positions
572 <    AtomCommVectorRow->gather(snap_->atomData.position,
572 >    AtomPlanVectorRow->gather(snap_->atomData.position,
573                                atomRowData.position);
574 <    AtomCommVectorColumn->gather(snap_->atomData.position,
574 >    AtomPlanVectorColumn->gather(snap_->atomData.position,
575                                   atomColData.position);
576      
577      // gather up the cutoff group positions
578 <    cgCommVectorRow->gather(snap_->cgData.position,
578 >
579 >    cerr  << "before gather\n";
580 >    for (int i = 0; i < snap_->cgData.position.size(); i++) {
581 >      cerr << "cgpos = " << snap_->cgData.position[i] << "\n";
582 >    }
583 >
584 >    cgPlanVectorRow->gather(snap_->cgData.position,
585                              cgRowData.position);
586 <    cgCommVectorColumn->gather(snap_->cgData.position,
586 >
587 >    cerr  << "after gather\n";
588 >    for (int i = 0; i < cgRowData.position.size(); i++) {
589 >      cerr << "cgRpos = " << cgRowData.position[i] << "\n";
590 >    }
591 >
592 >    cgPlanVectorColumn->gather(snap_->cgData.position,
593                                 cgColData.position);
594 +    for (int i = 0; i < cgColData.position.size(); i++) {
595 +      cerr << "cgCpos = " << cgColData.position[i] << "\n";
596 +    }
597 +
598      
599      // if needed, gather the atomic rotation matrices
600      if (storageLayout_ & DataStorage::dslAmat) {
601 <      AtomCommMatrixRow->gather(snap_->atomData.aMat,
601 >      AtomPlanMatrixRow->gather(snap_->atomData.aMat,
602                                  atomRowData.aMat);
603 <      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
603 >      AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
604                                     atomColData.aMat);
605      }
606      
607      // if needed, gather the atomic eletrostatic frames
608      if (storageLayout_ & DataStorage::dslElectroFrame) {
609 <      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
609 >      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
610                                  atomRowData.electroFrame);
611 <      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
611 >      AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
612                                     atomColData.electroFrame);
613      }
614 +
615   #endif      
616    }
617    
# Line 525 | Line 625 | namespace OpenMD {
625      
626      if (storageLayout_ & DataStorage::dslDensity) {
627        
628 <      AtomCommRealRow->scatter(atomRowData.density,
628 >      AtomPlanRealRow->scatter(atomRowData.density,
629                                 snap_->atomData.density);
630        
631        int n = snap_->atomData.density.size();
632        vector<RealType> rho_tmp(n, 0.0);
633 <      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
633 >      AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
634        for (int i = 0; i < n; i++)
635          snap_->atomData.density[i] += rho_tmp[i];
636      }
# Line 546 | Line 646 | namespace OpenMD {
646      storageLayout_ = sman_->getStorageLayout();
647   #ifdef IS_MPI
648      if (storageLayout_ & DataStorage::dslFunctional) {
649 <      AtomCommRealRow->gather(snap_->atomData.functional,
649 >      AtomPlanRealRow->gather(snap_->atomData.functional,
650                                atomRowData.functional);
651 <      AtomCommRealColumn->gather(snap_->atomData.functional,
651 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
652                                   atomColData.functional);
653      }
654      
655      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
656 <      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
656 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
657                                atomRowData.functionalDerivative);
658 <      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
658 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
659                                   atomColData.functionalDerivative);
660      }
661   #endif
# Line 569 | Line 669 | namespace OpenMD {
669      int n = snap_->atomData.force.size();
670      vector<Vector3d> frc_tmp(n, V3Zero);
671      
672 <    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
672 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
673      for (int i = 0; i < n; i++) {
674        snap_->atomData.force[i] += frc_tmp[i];
675        frc_tmp[i] = 0.0;
676      }
677      
678 <    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
679 <    for (int i = 0; i < n; i++)
678 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
679 >    for (int i = 0; i < n; i++) {
680        snap_->atomData.force[i] += frc_tmp[i];
681 <    
682 <    
681 >    }
682 >        
683      if (storageLayout_ & DataStorage::dslTorque) {
684  
685 <      int nt = snap_->atomData.force.size();
685 >      int nt = snap_->atomData.torque.size();
686        vector<Vector3d> trq_tmp(nt, V3Zero);
687  
688 <      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
689 <      for (int i = 0; i < n; i++) {
688 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
689 >      for (int i = 0; i < nt; i++) {
690          snap_->atomData.torque[i] += trq_tmp[i];
691          trq_tmp[i] = 0.0;
692        }
693        
694 <      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
695 <      for (int i = 0; i < n; i++)
694 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
695 >      for (int i = 0; i < nt; i++)
696          snap_->atomData.torque[i] += trq_tmp[i];
697      }
698 +
699 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
700 +
701 +      int ns = snap_->atomData.skippedCharge.size();
702 +      vector<RealType> skch_tmp(ns, 0.0);
703 +
704 +      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
705 +      for (int i = 0; i < ns; i++) {
706 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
707 +        skch_tmp[i] = 0.0;
708 +      }
709 +      
710 +      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
711 +      for (int i = 0; i < ns; i++)
712 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
713 +    }
714      
715      nLocal_ = snap_->getNumberOfAtoms();
716  
# Line 603 | Line 719 | namespace OpenMD {
719  
720      // scatter/gather pot_row into the members of my column
721            
722 <    AtomCommPotRow->scatter(pot_row, pot_temp);
722 >    AtomPlanPotRow->scatter(pot_row, pot_temp);
723  
724      for (int ii = 0;  ii < pot_temp.size(); ii++ )
725 <      pot_local += pot_temp[ii];
725 >      pairwisePot += pot_temp[ii];
726      
727      fill(pot_temp.begin(), pot_temp.end(),
728           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
729        
730 <    AtomCommPotColumn->scatter(pot_col, pot_temp);    
730 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
731      
732      for (int ii = 0;  ii < pot_temp.size(); ii++ )
733 <      pot_local += pot_temp[ii];
618 <    
733 >      pairwisePot += pot_temp[ii];    
734   #endif
735 +
736 +    cerr << "pairwisePot = " <<  pairwisePot << "\n";
737    }
738  
739    int ForceMatrixDecomposition::getNAtomsInRow() {  
# Line 651 | Line 768 | namespace OpenMD {
768      
769   #ifdef IS_MPI
770      d = cgColData.position[cg2] - cgRowData.position[cg1];
771 +    cerr << "cg1 = " << cg1 << "\tcg1p = " << cgRowData.position[cg1] << "\n";
772 +    cerr << "cg2 = " << cg2 << "\tcg2p = " << cgColData.position[cg2] << "\n";
773   #else
774      d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
775 +    cerr << "cg1 = " << cg1 << "\tcg1p = " << snap_->cgData.position[cg1] << "\n";
776 +    cerr << "cg2 = " << cg2 << "\tcg2p = " << snap_->cgData.position[cg2] << "\n";
777   #endif
778      
779      snap_->wrapVector(d);
# Line 717 | Line 838 | namespace OpenMD {
838      return d;    
839    }
840  
841 <  vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
842 <    return skipsForAtom[atom1];
841 >  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
842 >    return excludesForAtom[atom1];
843    }
844  
845    /**
846 <   * There are a number of reasons to skip a pair or a
726 <   * particle. Mostly we do this to exclude atoms who are involved in
727 <   * short range interactions (bonds, bends, torsions), but we also
728 <   * need to exclude some overcounted interactions that result from
846 >   * We need to exclude some overcounted interactions that result from
847     * the parallel decomposition.
848     */
849    bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
850      int unique_id_1, unique_id_2;
851 +    
852  
853 +    cerr << "sap with atom1, atom2 =\t" << atom1 << "\t" << atom2 << "\n";
854   #ifdef IS_MPI
855      // in MPI, we have to look up the unique IDs for each atom
856      unique_id_1 = AtomRowToGlobal[atom1];
857      unique_id_2 = AtomColToGlobal[atom2];
858  
859 +    cerr << "sap with uid1, uid2 =\t" << unique_id_1 << "\t" << unique_id_2 << "\n";
860      // this situation should only arise in MPI simulations
861      if (unique_id_1 == unique_id_2) return true;
862      
# Line 745 | Line 866 | namespace OpenMD {
866      } else {
867        if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
868      }
869 + #endif
870 +    return false;
871 +  }
872 +
873 +  /**
874 +   * We need to handle the interactions for atoms who are involved in
875 +   * the same rigid body as well as some short range interactions
876 +   * (bonds, bends, torsions) differently from other interactions.
877 +   * We'll still visit the pairwise routines, but with a flag that
878 +   * tells those routines to exclude the pair from direct long range
879 +   * interactions.  Some indirect interactions (notably reaction
880 +   * field) must still be handled for these pairs.
881 +   */
882 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
883 +    int unique_id_2;
884 + #ifdef IS_MPI
885 +    // in MPI, we have to look up the unique IDs for the row atom.
886 +    unique_id_2 = AtomColToGlobal[atom2];
887   #else
888      // in the normal loop, the atom numbers are unique
750 –    unique_id_1 = atom1;
889      unique_id_2 = atom2;
890   #endif
891      
892 <    for (vector<int>::iterator i = skipsForAtom[atom1].begin();
893 <         i != skipsForAtom[atom1].end(); ++i) {
892 >    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
893 >         i != excludesForAtom[atom1].end(); ++i) {
894        if ( (*i) == unique_id_2 ) return true;
895 <    }    
895 >    }
896  
897 +    return false;
898    }
899  
900  
# Line 776 | Line 915 | namespace OpenMD {
915    }
916  
917      // filling interaction blocks with pointers
918 <  void ForceMatrixDecomposition::fillInteractionData(InteractionData idat,
919 <                                                     int atom1, int atom2) {    
918 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
919 >                                                     int atom1, int atom2) {
920 >
921 >    idat.excluded = excludeAtomPair(atom1, atom2);
922 >  
923   #ifdef IS_MPI
924 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
925 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
926 +    //                         ff_->getAtomType(identsCol[atom2]) );
927      
783 –    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
784 –                             ff_->getAtomType(identsCol[atom2]) );
785 –    
928      if (storageLayout_ & DataStorage::dslAmat) {
929        idat.A1 = &(atomRowData.aMat[atom1]);
930        idat.A2 = &(atomColData.aMat[atom2]);
# Line 818 | Line 960 | namespace OpenMD {
960        idat.particlePot2 = &(atomColData.particlePot[atom2]);
961      }
962  
963 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
964 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
965 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
966 +    }
967 +
968   #else
969  
970 <    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
971 <                             ff_->getAtomType(identsLocal[atom2]) );
970 >    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
971 >    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
972 >    //                         ff_->getAtomType(idents[atom2]) );
973  
974      if (storageLayout_ & DataStorage::dslAmat) {
975        idat.A1 = &(snap_->atomData.aMat[atom1]);
# Line 838 | Line 986 | namespace OpenMD {
986        idat.t2 = &(snap_->atomData.torque[atom2]);
987      }
988  
989 <    if (storageLayout_ & DataStorage::dslDensity) {
989 >    if (storageLayout_ & DataStorage::dslDensity) {    
990        idat.rho1 = &(snap_->atomData.density[atom1]);
991        idat.rho2 = &(snap_->atomData.density[atom2]);
992      }
# Line 858 | Line 1006 | namespace OpenMD {
1006        idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1007      }
1008  
1009 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
1010 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1011 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1012 +    }
1013   #endif
1014    }
1015  
1016    
1017 <  void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {    
1017 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1018   #ifdef IS_MPI
1019      pot_row[atom1] += 0.5 *  *(idat.pot);
1020      pot_col[atom2] += 0.5 *  *(idat.pot);
# Line 870 | Line 1022 | namespace OpenMD {
1022      atomRowData.force[atom1] += *(idat.f1);
1023      atomColData.force[atom2] -= *(idat.f1);
1024   #else
1025 <    longRangePot_ += *(idat.pot);
1026 <    
1025 >    pairwisePot += *(idat.pot);
1026 >
1027      snap_->atomData.force[atom1] += *(idat.f1);
1028      snap_->atomData.force[atom2] -= *(idat.f1);
1029   #endif
1030 <
1030 >    
1031    }
1032  
881 –
882 –  void ForceMatrixDecomposition::fillSkipData(InteractionData idat,
883 –                                              int atom1, int atom2) {
884 – #ifdef IS_MPI
885 –    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
886 –                             ff_->getAtomType(identsCol[atom2]) );
887 –
888 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
889 –      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
890 –      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
891 –    }
892 –    if (storageLayout_ & DataStorage::dslTorque) {
893 –      idat.t1 = &(atomRowData.torque[atom1]);
894 –      idat.t2 = &(atomColData.torque[atom2]);
895 –    }
896 – #else
897 –    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
898 –                             ff_->getAtomType(identsLocal[atom2]) );
899 –
900 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
901 –      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
902 –      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
903 –    }
904 –    if (storageLayout_ & DataStorage::dslTorque) {
905 –      idat.t1 = &(snap_->atomData.torque[atom1]);
906 –      idat.t2 = &(snap_->atomData.torque[atom2]);
907 –    }
908 – #endif    
909 –  }
910 –
1033    /*
1034     * buildNeighborList
1035     *
# Line 918 | Line 1040 | namespace OpenMD {
1040        
1041      vector<pair<int, int> > neighborList;
1042      groupCutoffs cuts;
1043 +    bool doAllPairs = false;
1044 +
1045   #ifdef IS_MPI
1046      cellListRow_.clear();
1047      cellListCol_.clear();
# Line 937 | Line 1061 | namespace OpenMD {
1061      nCells_.y() = (int) ( Hy.length() )/ rList_;
1062      nCells_.z() = (int) ( Hz.length() )/ rList_;
1063  
1064 +    // handle small boxes where the cell offsets can end up repeating cells
1065 +    
1066 +    if (nCells_.x() < 3) doAllPairs = true;
1067 +    if (nCells_.y() < 3) doAllPairs = true;
1068 +    if (nCells_.z() < 3) doAllPairs = true;
1069 +
1070      Mat3x3d invHmat = snap_->getInvHmat();
1071      Vector3d rs, scaled, dr;
1072      Vector3i whichCell;
1073      int cellIndex;
1074      int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1075  
946 –    cerr << "flag1\n";
1076   #ifdef IS_MPI
1077      cellListRow_.resize(nCtot);
1078      cellListCol_.resize(nCtot);
1079   #else
1080      cellList_.resize(nCtot);
1081   #endif
1082 <    cerr << "flag2\n";
1082 >
1083 >    if (!doAllPairs) {
1084   #ifdef IS_MPI
955 –    for (int i = 0; i < nGroupsInRow_; i++) {
956 –      rs = cgRowData.position[i];
1085  
1086 <      // scaled positions relative to the box vectors
1087 <      scaled = invHmat * rs;
1088 <
1089 <      // wrap the vector back into the unit box by subtracting integer box
1090 <      // numbers
1091 <      for (int j = 0; j < 3; j++) {
1092 <        scaled[j] -= roundMe(scaled[j]);
1093 <        scaled[j] += 0.5;
1086 >      for (int i = 0; i < nGroupsInRow_; i++) {
1087 >        rs = cgRowData.position[i];
1088 >        
1089 >        // scaled positions relative to the box vectors
1090 >        scaled = invHmat * rs;
1091 >        
1092 >        // wrap the vector back into the unit box by subtracting integer box
1093 >        // numbers
1094 >        for (int j = 0; j < 3; j++) {
1095 >          scaled[j] -= roundMe(scaled[j]);
1096 >          scaled[j] += 0.5;
1097 >        }
1098 >        
1099 >        // find xyz-indices of cell that cutoffGroup is in.
1100 >        whichCell.x() = nCells_.x() * scaled.x();
1101 >        whichCell.y() = nCells_.y() * scaled.y();
1102 >        whichCell.z() = nCells_.z() * scaled.z();
1103 >        
1104 >        // find single index of this cell:
1105 >        cellIndex = Vlinear(whichCell, nCells_);
1106 >        
1107 >        // add this cutoff group to the list of groups in this cell;
1108 >        cellListRow_[cellIndex].push_back(i);
1109        }
1110 <    
1111 <      // find xyz-indices of cell that cutoffGroup is in.
1112 <      whichCell.x() = nCells_.x() * scaled.x();
1113 <      whichCell.y() = nCells_.y() * scaled.y();
1114 <      whichCell.z() = nCells_.z() * scaled.z();
1115 <
1116 <      // find single index of this cell:
1117 <      cellIndex = Vlinear(whichCell, nCells_);
1118 <
1119 <      // add this cutoff group to the list of groups in this cell;
1120 <      cellListRow_[cellIndex].push_back(i);
1121 <    }
1122 <
1123 <    for (int i = 0; i < nGroupsInCol_; i++) {
1124 <      rs = cgColData.position[i];
1125 <
1126 <      // scaled positions relative to the box vectors
1127 <      scaled = invHmat * rs;
1128 <
1129 <      // wrap the vector back into the unit box by subtracting integer box
1130 <      // numbers
1131 <      for (int j = 0; j < 3; j++) {
1132 <        scaled[j] -= roundMe(scaled[j]);
990 <        scaled[j] += 0.5;
1110 >      for (int i = 0; i < nGroupsInCol_; i++) {
1111 >        rs = cgColData.position[i];
1112 >        
1113 >        // scaled positions relative to the box vectors
1114 >        scaled = invHmat * rs;
1115 >        
1116 >        // wrap the vector back into the unit box by subtracting integer box
1117 >        // numbers
1118 >        for (int j = 0; j < 3; j++) {
1119 >          scaled[j] -= roundMe(scaled[j]);
1120 >          scaled[j] += 0.5;
1121 >        }
1122 >        
1123 >        // find xyz-indices of cell that cutoffGroup is in.
1124 >        whichCell.x() = nCells_.x() * scaled.x();
1125 >        whichCell.y() = nCells_.y() * scaled.y();
1126 >        whichCell.z() = nCells_.z() * scaled.z();
1127 >        
1128 >        // find single index of this cell:
1129 >        cellIndex = Vlinear(whichCell, nCells_);
1130 >        
1131 >        // add this cutoff group to the list of groups in this cell;
1132 >        cellListCol_[cellIndex].push_back(i);
1133        }
992 –
993 –      // find xyz-indices of cell that cutoffGroup is in.
994 –      whichCell.x() = nCells_.x() * scaled.x();
995 –      whichCell.y() = nCells_.y() * scaled.y();
996 –      whichCell.z() = nCells_.z() * scaled.z();
997 –
998 –      // find single index of this cell:
999 –      cellIndex = Vlinear(whichCell, nCells_);
1000 –
1001 –      // add this cutoff group to the list of groups in this cell;
1002 –      cellListCol_[cellIndex].push_back(i);
1003 –    }
1134   #else
1135 <    for (int i = 0; i < nGroups_; i++) {
1136 <      rs = snap_->cgData.position[i];
1137 <
1138 <      // scaled positions relative to the box vectors
1139 <      scaled = invHmat * rs;
1140 <
1141 <      // wrap the vector back into the unit box by subtracting integer box
1142 <      // numbers
1143 <      for (int j = 0; j < 3; j++) {
1144 <        scaled[j] -= roundMe(scaled[j]);
1145 <        scaled[j] += 0.5;
1135 >      for (int i = 0; i < nGroups_; i++) {
1136 >        rs = snap_->cgData.position[i];
1137 >        
1138 >        // scaled positions relative to the box vectors
1139 >        scaled = invHmat * rs;
1140 >        
1141 >        // wrap the vector back into the unit box by subtracting integer box
1142 >        // numbers
1143 >        for (int j = 0; j < 3; j++) {
1144 >          scaled[j] -= roundMe(scaled[j]);
1145 >          scaled[j] += 0.5;
1146 >        }
1147 >        
1148 >        // find xyz-indices of cell that cutoffGroup is in.
1149 >        whichCell.x() = nCells_.x() * scaled.x();
1150 >        whichCell.y() = nCells_.y() * scaled.y();
1151 >        whichCell.z() = nCells_.z() * scaled.z();
1152 >        
1153 >        // find single index of this cell:
1154 >        cellIndex = Vlinear(whichCell, nCells_);
1155 >        
1156 >        // add this cutoff group to the list of groups in this cell;
1157 >        cellList_[cellIndex].push_back(i);
1158        }
1017 –
1018 –      // find xyz-indices of cell that cutoffGroup is in.
1019 –      whichCell.x() = nCells_.x() * scaled.x();
1020 –      whichCell.y() = nCells_.y() * scaled.y();
1021 –      whichCell.z() = nCells_.z() * scaled.z();
1022 –
1023 –      // find single index of this cell:
1024 –      cellIndex = Vlinear(whichCell, nCells_);      
1025 –
1026 –      // add this cutoff group to the list of groups in this cell;
1027 –      cellList_[cellIndex].push_back(i);
1028 –    }
1159   #endif
1160  
1161 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1162 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1163 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1164 <          Vector3i m1v(m1x, m1y, m1z);
1165 <          int m1 = Vlinear(m1v, nCells_);
1036 <
1037 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1038 <               os != cellOffsets_.end(); ++os) {
1161 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1162 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1163 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1164 >            Vector3i m1v(m1x, m1y, m1z);
1165 >            int m1 = Vlinear(m1v, nCells_);
1166              
1167 <            Vector3i m2v = m1v + (*os);
1168 <            
1169 <            if (m2v.x() >= nCells_.x()) {
1170 <              m2v.x() = 0;          
1171 <            } else if (m2v.x() < 0) {
1172 <              m2v.x() = nCells_.x() - 1;
1173 <            }
1174 <            
1175 <            if (m2v.y() >= nCells_.y()) {
1176 <              m2v.y() = 0;          
1177 <            } else if (m2v.y() < 0) {
1178 <              m2v.y() = nCells_.y() - 1;
1179 <            }
1180 <            
1181 <            if (m2v.z() >= nCells_.z()) {
1182 <              m2v.z() = 0;          
1183 <            } else if (m2v.z() < 0) {
1184 <              m2v.z() = nCells_.z() - 1;
1185 <            }
1186 <            
1187 <            int m2 = Vlinear (m2v, nCells_);
1188 <
1167 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1168 >                 os != cellOffsets_.end(); ++os) {
1169 >              
1170 >              Vector3i m2v = m1v + (*os);
1171 >              
1172 >              if (m2v.x() >= nCells_.x()) {
1173 >                m2v.x() = 0;          
1174 >              } else if (m2v.x() < 0) {
1175 >                m2v.x() = nCells_.x() - 1;
1176 >              }
1177 >              
1178 >              if (m2v.y() >= nCells_.y()) {
1179 >                m2v.y() = 0;          
1180 >              } else if (m2v.y() < 0) {
1181 >                m2v.y() = nCells_.y() - 1;
1182 >              }
1183 >              
1184 >              if (m2v.z() >= nCells_.z()) {
1185 >                m2v.z() = 0;          
1186 >              } else if (m2v.z() < 0) {
1187 >                m2v.z() = nCells_.z() - 1;
1188 >              }
1189 >              
1190 >              int m2 = Vlinear (m2v, nCells_);
1191 >              
1192   #ifdef IS_MPI
1193 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1194 <                 j1 != cellListRow_[m1].end(); ++j1) {
1195 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1196 <                   j2 != cellListCol_[m2].end(); ++j2) {
1197 <                              
1198 <                // Always do this if we're in different cells or if
1199 <                // we're in the same cell and the global index of the
1070 <                // j2 cutoff group is less than the j1 cutoff group
1071 <
1072 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1193 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1194 >                   j1 != cellListRow_[m1].end(); ++j1) {
1195 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1196 >                     j2 != cellListCol_[m2].end(); ++j2) {
1197 >                  
1198 >                  // In parallel, we need to visit *all* pairs of row &
1199 >                  // column indicies and will truncate later on.
1200                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1201                    snap_->wrapVector(dr);
1202                    cuts = getGroupCutoffs( (*j1), (*j2) );
1203                    if (dr.lengthSquare() < cuts.third) {
1204                      neighborList.push_back(make_pair((*j1), (*j2)));
1205 <                  }
1205 >                  }                  
1206                  }
1207                }
1081 –            }
1208   #else
1209 <
1210 <            for (vector<int>::iterator j1 = cellList_[m1].begin();
1211 <                 j1 != cellList_[m1].end(); ++j1) {
1212 <              for (vector<int>::iterator j2 = cellList_[m2].begin();
1213 <                   j2 != cellList_[m2].end(); ++j2) {
1214 <
1215 <                // Always do this if we're in different cells or if
1216 <                // we're in the same cell and the global index of the
1217 <                // j2 cutoff group is less than the j1 cutoff group
1218 <
1219 <                if (m2 != m1 || (*j2) < (*j1)) {
1220 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1221 <                  snap_->wrapVector(dr);
1222 <                  cuts = getGroupCutoffs( (*j1), (*j2) );
1223 <                  if (dr.lengthSquare() < cuts.third) {
1224 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1209 >              
1210 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1211 >                   j1 != cellList_[m1].end(); ++j1) {
1212 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1213 >                     j2 != cellList_[m2].end(); ++j2) {
1214 >                  
1215 >                  // Always do this if we're in different cells or if
1216 >                  // we're in the same cell and the global index of the
1217 >                  // j2 cutoff group is less than the j1 cutoff group
1218 >                  
1219 >                  if (m2 != m1 || (*j2) < (*j1)) {
1220 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1221 >                    snap_->wrapVector(dr);
1222 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1223 >                    if (dr.lengthSquare() < cuts.third) {
1224 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1225 >                    }
1226                    }
1227                  }
1228                }
1102 –            }
1229   #endif
1230 +            }
1231            }
1232          }
1233        }
1234 +    } else {
1235 +      // branch to do all cutoff group pairs
1236 + #ifdef IS_MPI
1237 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1238 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1239 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1240 +          snap_->wrapVector(dr);
1241 +          cuts = getGroupCutoffs( j1, j2 );
1242 +          if (dr.lengthSquare() < cuts.third) {
1243 +            neighborList.push_back(make_pair(j1, j2));
1244 +          }
1245 +        }
1246 +      }
1247 + #else
1248 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1249 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1250 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1251 +          snap_->wrapVector(dr);
1252 +          cuts = getGroupCutoffs( j1, j2 );
1253 +          if (dr.lengthSquare() < cuts.third) {
1254 +            neighborList.push_back(make_pair(j1, j2));
1255 +          }
1256 +        }
1257 +      }        
1258 + #endif
1259      }
1260 <    
1260 >      
1261      // save the local cutoff group positions for the check that is
1262      // done on each loop:
1263      saved_CG_positions_.clear();

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines