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 1612 by gezelter, Fri Aug 12 19:59:56 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_.clear();
57 +    cellOffsets_.push_back( Vector3i(-1,-1,-1) );
58 +    cellOffsets_.push_back( Vector3i( 0,-1,-1) );
59 +    cellOffsets_.push_back( Vector3i( 1,-1,-1) );                          
60 +    cellOffsets_.push_back( Vector3i(-1, 0,-1) );
61 +    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
62 +    cellOffsets_.push_back( Vector3i( 1, 0,-1) );
63 +    cellOffsets_.push_back( Vector3i(-1, 1,-1) );
64 +    cellOffsets_.push_back( Vector3i( 0, 1,-1) );      
65 +    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
66 +    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
67 +    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
68 +    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
69 +    cellOffsets_.push_back( Vector3i(-1, 0, 0) );      
70 +    cellOffsets_.push_back( Vector3i( 0, 0, 0) );
71 +    cellOffsets_.push_back( Vector3i( 1, 0, 0) );
72 +    cellOffsets_.push_back( Vector3i(-1, 1, 0) );
73 +    cellOffsets_.push_back( Vector3i( 0, 1, 0) );
74 +    cellOffsets_.push_back( Vector3i( 1, 1, 0) );
75 +    cellOffsets_.push_back( Vector3i(-1,-1, 1) );
76 +    cellOffsets_.push_back( Vector3i( 0,-1, 1) );
77 +    cellOffsets_.push_back( Vector3i( 1,-1, 1) );
78 +    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
79 +    cellOffsets_.push_back( Vector3i( 0, 0, 1) );
80 +    cellOffsets_.push_back( Vector3i( 1, 0, 1) );
81 +    cellOffsets_.push_back( Vector3i(-1, 1, 1) );
82 +    cellOffsets_.push_back( Vector3i( 0, 1, 1) );
83 +    cellOffsets_.push_back( Vector3i( 1, 1, 1) );
84 + #endif    
85 +  }
86 +
87 +
88    /**
89     * distributeInitialData is essentially a copy of the older fortran
90     * SimulationSetup
91     */
54 –  
92    void ForceMatrixDecomposition::distributeInitialData() {
93      snap_ = sman_->getCurrentSnapshot();
94      storageLayout_ = sman_->getStorageLayout();
95      ff_ = info_->getForceField();
96      nLocal_ = snap_->getNumberOfAtoms();
97 <
97 >    
98      nGroups_ = info_->getNLocalCutoffGroups();
62 –    cerr << "in dId, nGroups = " << nGroups_ << "\n";
99      // gather the information for atomtype IDs (atids):
100 <    identsLocal = info_->getIdentArray();
100 >    idents = info_->getIdentArray();
101      AtomLocalToGlobal = info_->getGlobalAtomIndices();
102      cgLocalToGlobal = info_->getGlobalGroupIndices();
103      vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
104 +
105      massFactors = info_->getMassFactors();
69 –    PairList excludes = info_->getExcludedInteractions();
70 –    PairList oneTwo = info_->getOneTwoInteractions();
71 –    PairList oneThree = info_->getOneThreeInteractions();
72 –    PairList oneFour = info_->getOneFourInteractions();
106  
107 +    PairList* excludes = info_->getExcludedInteractions();
108 +    PairList* oneTwo = info_->getOneTwoInteractions();
109 +    PairList* oneThree = info_->getOneThreeInteractions();
110 +    PairList* oneFour = info_->getOneFourInteractions();
111 +
112   #ifdef IS_MPI
113  
114 <    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
115 <    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_);
114 >    MPI::Intracomm row = rowComm.getComm();
115 >    MPI::Intracomm col = colComm.getComm();
116  
117 <    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
118 <    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
119 <    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
120 <    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
121 <    AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
117 >    AtomPlanIntRow = new Plan<int>(row, nLocal_);
118 >    AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
119 >    AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
120 >    AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
121 >    AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
122  
123 <    cgCommIntRow = new Communicator<Row,int>(nGroups_);
124 <    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
125 <    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
126 <    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
123 >    AtomPlanIntColumn = new Plan<int>(col, nLocal_);
124 >    AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
125 >    AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
126 >    AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
127 >    AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
128  
129 <    nAtomsInRow_ = AtomCommIntRow->getSize();
130 <    nAtomsInCol_ = AtomCommIntColumn->getSize();
131 <    nGroupsInRow_ = cgCommIntRow->getSize();
132 <    nGroupsInCol_ = cgCommIntColumn->getSize();
129 >    cgPlanIntRow = new Plan<int>(row, nGroups_);
130 >    cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
131 >    cgPlanIntColumn = new Plan<int>(col, nGroups_);
132 >    cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
133  
134 +    nAtomsInRow_ = AtomPlanIntRow->getSize();
135 +    nAtomsInCol_ = AtomPlanIntColumn->getSize();
136 +    nGroupsInRow_ = cgPlanIntRow->getSize();
137 +    nGroupsInCol_ = cgPlanIntColumn->getSize();
138 +
139      // Modify the data storage objects with the correct layouts and sizes:
140      atomRowData.resize(nAtomsInRow_);
141      atomRowData.setStorageLayout(storageLayout_);
# Line 108 | Line 149 | namespace OpenMD {
149      identsRow.resize(nAtomsInRow_);
150      identsCol.resize(nAtomsInCol_);
151      
152 <    AtomCommIntRow->gather(identsLocal, identsRow);
153 <    AtomCommIntColumn->gather(identsLocal, identsCol);
152 >    AtomPlanIntRow->gather(idents, identsRow);
153 >    AtomPlanIntColumn->gather(idents, identsCol);
154      
155 <    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
156 <    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
157 <    
117 <    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
118 <    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
155 >    // allocate memory for the parallel objects
156 >    atypesRow.resize(nAtomsInRow_);
157 >    atypesCol.resize(nAtomsInCol_);
158  
159 <    AtomCommRealRow->gather(massFactors, massFactorsRow);
160 <    AtomCommRealColumn->gather(massFactors, massFactorsCol);
159 >    for (int i = 0; i < nAtomsInRow_; i++)
160 >      atypesRow[i] = ff_->getAtomType(identsRow[i]);
161 >    for (int i = 0; i < nAtomsInCol_; i++)
162 >      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
163  
164 +    pot_row.resize(nAtomsInRow_);
165 +    pot_col.resize(nAtomsInCol_);
166 +
167 +    AtomRowToGlobal.resize(nAtomsInRow_);
168 +    AtomColToGlobal.resize(nAtomsInCol_);
169 +    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
170 +    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
171 +
172 +    cgRowToGlobal.resize(nGroupsInRow_);
173 +    cgColToGlobal.resize(nGroupsInCol_);
174 +    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
175 +    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
176 +
177 +    massFactorsRow.resize(nAtomsInRow_);
178 +    massFactorsCol.resize(nAtomsInCol_);
179 +    AtomPlanRealRow->gather(massFactors, massFactorsRow);
180 +    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
181 +
182      groupListRow_.clear();
183      groupListRow_.resize(nGroupsInRow_);
184      for (int i = 0; i < nGroupsInRow_; i++) {
# Line 142 | Line 201 | namespace OpenMD {
201        }      
202      }
203  
204 <    skipsForAtom.clear();
205 <    skipsForAtom.resize(nAtomsInRow_);
204 >    excludesForAtom.clear();
205 >    excludesForAtom.resize(nAtomsInRow_);
206      toposForAtom.clear();
207      toposForAtom.resize(nAtomsInRow_);
208      topoDist.clear();
# Line 154 | Line 213 | namespace OpenMD {
213        for (int j = 0; j < nAtomsInCol_; j++) {
214          int jglob = AtomColToGlobal[j];
215  
216 <        if (excludes.hasPair(iglob, jglob))
217 <          skipsForAtom[i].push_back(j);      
216 >        if (excludes->hasPair(iglob, jglob))
217 >          excludesForAtom[i].push_back(j);      
218          
219 <        if (oneTwo.hasPair(iglob, jglob)) {
219 >        if (oneTwo->hasPair(iglob, jglob)) {
220            toposForAtom[i].push_back(j);
221            topoDist[i].push_back(1);
222          } else {
223 <          if (oneThree.hasPair(iglob, jglob)) {
223 >          if (oneThree->hasPair(iglob, jglob)) {
224              toposForAtom[i].push_back(j);
225              topoDist[i].push_back(2);
226            } else {
227 <            if (oneFour.hasPair(iglob, jglob)) {
227 >            if (oneFour->hasPair(iglob, jglob)) {
228                toposForAtom[i].push_back(j);
229                topoDist[i].push_back(3);
230              }
# Line 176 | Line 235 | namespace OpenMD {
235  
236   #endif
237  
238 +    // allocate memory for the parallel objects
239 +    atypesLocal.resize(nLocal_);
240 +
241 +    for (int i = 0; i < nLocal_; i++)
242 +      atypesLocal[i] = ff_->getAtomType(idents[i]);
243 +
244      groupList_.clear();
245      groupList_.resize(nGroups_);
246      for (int i = 0; i < nGroups_; i++) {
# Line 188 | Line 253 | namespace OpenMD {
253        }      
254      }
255  
256 <    skipsForAtom.clear();
257 <    skipsForAtom.resize(nLocal_);
256 >    excludesForAtom.clear();
257 >    excludesForAtom.resize(nLocal_);
258      toposForAtom.clear();
259      toposForAtom.resize(nLocal_);
260      topoDist.clear();
# Line 201 | Line 266 | namespace OpenMD {
266        for (int j = 0; j < nLocal_; j++) {
267          int jglob = AtomLocalToGlobal[j];
268  
269 <        if (excludes.hasPair(iglob, jglob))
270 <          skipsForAtom[i].push_back(j);              
269 >        if (excludes->hasPair(iglob, jglob))
270 >          excludesForAtom[i].push_back(j);              
271          
272 <        if (oneTwo.hasPair(iglob, jglob)) {
272 >        if (oneTwo->hasPair(iglob, jglob)) {
273            toposForAtom[i].push_back(j);
274            topoDist[i].push_back(1);
275          } else {
276 <          if (oneThree.hasPair(iglob, jglob)) {
276 >          if (oneThree->hasPair(iglob, jglob)) {
277              toposForAtom[i].push_back(j);
278              topoDist[i].push_back(2);
279            } else {
280 <            if (oneFour.hasPair(iglob, jglob)) {
280 >            if (oneFour->hasPair(iglob, jglob)) {
281                toposForAtom[i].push_back(j);
282                topoDist[i].push_back(3);
283              }
# Line 222 | Line 287 | namespace OpenMD {
287      }
288      
289      createGtypeCutoffMap();
290 +
291    }
292    
293    void ForceMatrixDecomposition::createGtypeCutoffMap() {
294 <
294 >    
295      RealType tol = 1e-6;
296 +    largestRcut_ = 0.0;
297      RealType rc;
298      int atid;
299      set<AtomType*> atypes = info_->getSimulatedAtomTypes();
300 <    vector<RealType> atypeCutoff;
301 <    atypeCutoff.resize( atypes.size() );
302 <
300 >    
301 >    map<int, RealType> atypeCutoff;
302 >      
303      for (set<AtomType*>::iterator at = atypes.begin();
304           at != atypes.end(); ++at){
238 –      rc = interactionMan_->getSuggestedCutoffRadius(*at);
305        atid = (*at)->getIdent();
306 <      atypeCutoff[atid] = rc;
306 >      if (userChoseCutoff_)
307 >        atypeCutoff[atid] = userCutoff_;
308 >      else
309 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
310      }
311 <
311 >    
312      vector<RealType> gTypeCutoffs;
244 –
313      // first we do a single loop over the cutoff groups to find the
314      // largest cutoff for any atypes present in this group.
315   #ifdef IS_MPI
# Line 299 | Line 367 | namespace OpenMD {
367  
368      vector<RealType> groupCutoff(nGroups_, 0.0);
369      groupToGtype.resize(nGroups_);
302 –
303 –    cerr << "nGroups = " << nGroups_ << "\n";
370      for (int cg1 = 0; cg1 < nGroups_; cg1++) {
305 –
371        groupCutoff[cg1] = 0.0;
372        vector<int> atomList = getAtomsInGroupRow(cg1);
308 –
373        for (vector<int>::iterator ia = atomList.begin();
374             ia != atomList.end(); ++ia) {            
375          int atom1 = (*ia);
376 <        atid = identsLocal[atom1];
377 <        if (atypeCutoff[atid] > groupCutoff[cg1]) {
376 >        atid = idents[atom1];
377 >        if (atypeCutoff[atid] > groupCutoff[cg1])
378            groupCutoff[cg1] = atypeCutoff[atid];
315 –        }
379        }
380 <
380 >      
381        bool gTypeFound = false;
382        for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
383          if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
# Line 322 | Line 385 | namespace OpenMD {
385            gTypeFound = true;
386          }
387        }
388 <      if (!gTypeFound) {
388 >      if (!gTypeFound) {      
389          gTypeCutoffs.push_back( groupCutoff[cg1] );
390          groupToGtype[cg1] = gTypeCutoffs.size() - 1;
391        }      
392      }
393   #endif
394  
332 –    cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
395      // Now we find the maximum group cutoff value present in the simulation
396  
397 <    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
397 >    RealType groupMax = *max_element(gTypeCutoffs.begin(),
398 >                                     gTypeCutoffs.end());
399  
400   #ifdef IS_MPI
401 <    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
401 >    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
402 >                              MPI::MAX);
403   #endif
404      
405      RealType tradRcut = groupMax;
# Line 365 | Line 429 | namespace OpenMD {
429  
430          pair<int,int> key = make_pair(i,j);
431          gTypeCutoffMap[key].first = thisRcut;
368 –
432          if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
370 –
433          gTypeCutoffMap[key].second = thisRcut*thisRcut;
372 –        
434          gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
374 –
435          // sanity check
436          
437          if (userChoseCutoff_) {
438            if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
439              sprintf(painCave.errMsg,
440                      "ForceMatrixDecomposition::createGtypeCutoffMap "
441 <                    "user-specified rCut does not match computed group Cutoff\n");
441 >                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
442              painCave.severity = OPENMD_ERROR;
443              painCave.isFatal = 1;
444              simError();            
# Line 410 | Line 470 | namespace OpenMD {
470    }
471  
472    void ForceMatrixDecomposition::zeroWorkArrays() {
473 +    pairwisePot = 0.0;
474 +    embeddingPot = 0.0;
475  
414 –    for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
415 –      longRangePot_[j] = 0.0;
416 –    }
417 –
476   #ifdef IS_MPI
477      if (storageLayout_ & DataStorage::dslForce) {
478        fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
# Line 430 | Line 488 | namespace OpenMD {
488           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
489  
490      fill(pot_col.begin(), pot_col.end(),
491 <         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
434 <    
435 <    pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
491 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
492  
493      if (storageLayout_ & DataStorage::dslParticlePot) {    
494 <      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
495 <      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
494 >      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
495 >           0.0);
496 >      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
497 >           0.0);
498      }
499  
500      if (storageLayout_ & DataStorage::dslDensity) {      
# Line 445 | Line 503 | namespace OpenMD {
503      }
504  
505      if (storageLayout_ & DataStorage::dslFunctional) {  
506 <      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
507 <      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
506 >      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
507 >           0.0);
508 >      fill(atomColData.functional.begin(), atomColData.functional.end(),
509 >           0.0);
510      }
511  
512      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
# Line 456 | Line 516 | namespace OpenMD {
516             atomColData.functionalDerivative.end(), 0.0);
517      }
518  
519 < #else
520 <    
519 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
520 >      fill(atomRowData.skippedCharge.begin(),
521 >           atomRowData.skippedCharge.end(), 0.0);
522 >      fill(atomColData.skippedCharge.begin(),
523 >           atomColData.skippedCharge.end(), 0.0);
524 >    }
525 >
526 > #endif
527 >    // even in parallel, we need to zero out the local arrays:
528 >
529      if (storageLayout_ & DataStorage::dslParticlePot) {      
530        fill(snap_->atomData.particlePot.begin(),
531             snap_->atomData.particlePot.end(), 0.0);
# Line 475 | Line 543 | namespace OpenMD {
543        fill(snap_->atomData.functionalDerivative.begin(),
544             snap_->atomData.functionalDerivative.end(), 0.0);
545      }
546 < #endif
546 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
547 >      fill(snap_->atomData.skippedCharge.begin(),
548 >           snap_->atomData.skippedCharge.end(), 0.0);
549 >    }
550      
551    }
552  
# Line 486 | Line 557 | namespace OpenMD {
557   #ifdef IS_MPI
558      
559      // gather up the atomic positions
560 <    AtomCommVectorRow->gather(snap_->atomData.position,
560 >    AtomPlanVectorRow->gather(snap_->atomData.position,
561                                atomRowData.position);
562 <    AtomCommVectorColumn->gather(snap_->atomData.position,
562 >    AtomPlanVectorColumn->gather(snap_->atomData.position,
563                                   atomColData.position);
564      
565      // gather up the cutoff group positions
566 <    cgCommVectorRow->gather(snap_->cgData.position,
566 >
567 >    cgPlanVectorRow->gather(snap_->cgData.position,
568                              cgRowData.position);
569 <    cgCommVectorColumn->gather(snap_->cgData.position,
569 >
570 >    cgPlanVectorColumn->gather(snap_->cgData.position,
571                                 cgColData.position);
572 +
573      
574      // if needed, gather the atomic rotation matrices
575      if (storageLayout_ & DataStorage::dslAmat) {
576 <      AtomCommMatrixRow->gather(snap_->atomData.aMat,
576 >      AtomPlanMatrixRow->gather(snap_->atomData.aMat,
577                                  atomRowData.aMat);
578 <      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
578 >      AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
579                                     atomColData.aMat);
580      }
581      
582      // if needed, gather the atomic eletrostatic frames
583      if (storageLayout_ & DataStorage::dslElectroFrame) {
584 <      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
584 >      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
585                                  atomRowData.electroFrame);
586 <      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
586 >      AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
587                                     atomColData.electroFrame);
588      }
589 +
590   #endif      
591    }
592    
# Line 525 | Line 600 | namespace OpenMD {
600      
601      if (storageLayout_ & DataStorage::dslDensity) {
602        
603 <      AtomCommRealRow->scatter(atomRowData.density,
603 >      AtomPlanRealRow->scatter(atomRowData.density,
604                                 snap_->atomData.density);
605        
606        int n = snap_->atomData.density.size();
607        vector<RealType> rho_tmp(n, 0.0);
608 <      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
608 >      AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
609        for (int i = 0; i < n; i++)
610          snap_->atomData.density[i] += rho_tmp[i];
611      }
# Line 546 | Line 621 | namespace OpenMD {
621      storageLayout_ = sman_->getStorageLayout();
622   #ifdef IS_MPI
623      if (storageLayout_ & DataStorage::dslFunctional) {
624 <      AtomCommRealRow->gather(snap_->atomData.functional,
624 >      AtomPlanRealRow->gather(snap_->atomData.functional,
625                                atomRowData.functional);
626 <      AtomCommRealColumn->gather(snap_->atomData.functional,
626 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
627                                   atomColData.functional);
628      }
629      
630      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
631 <      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
631 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
632                                atomRowData.functionalDerivative);
633 <      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
633 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
634                                   atomColData.functionalDerivative);
635      }
636   #endif
# Line 569 | Line 644 | namespace OpenMD {
644      int n = snap_->atomData.force.size();
645      vector<Vector3d> frc_tmp(n, V3Zero);
646      
647 <    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
647 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
648      for (int i = 0; i < n; i++) {
649        snap_->atomData.force[i] += frc_tmp[i];
650        frc_tmp[i] = 0.0;
651      }
652      
653 <    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
654 <    for (int i = 0; i < n; i++)
653 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
654 >    for (int i = 0; i < n; i++) {
655        snap_->atomData.force[i] += frc_tmp[i];
656 <    
657 <    
656 >    }
657 >        
658      if (storageLayout_ & DataStorage::dslTorque) {
659  
660 <      int nt = snap_->atomData.force.size();
660 >      int nt = snap_->atomData.torque.size();
661        vector<Vector3d> trq_tmp(nt, V3Zero);
662  
663 <      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
664 <      for (int i = 0; i < n; i++) {
663 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
664 >      for (int i = 0; i < nt; i++) {
665          snap_->atomData.torque[i] += trq_tmp[i];
666          trq_tmp[i] = 0.0;
667        }
668        
669 <      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
670 <      for (int i = 0; i < n; i++)
669 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
670 >      for (int i = 0; i < nt; i++)
671          snap_->atomData.torque[i] += trq_tmp[i];
672      }
673 +
674 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
675 +
676 +      int ns = snap_->atomData.skippedCharge.size();
677 +      vector<RealType> skch_tmp(ns, 0.0);
678 +
679 +      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
680 +      for (int i = 0; i < ns; i++) {
681 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
682 +        skch_tmp[i] = 0.0;
683 +      }
684 +      
685 +      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
686 +      for (int i = 0; i < ns; i++)
687 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
688 +    }
689      
690      nLocal_ = snap_->getNumberOfAtoms();
691  
# Line 603 | Line 694 | namespace OpenMD {
694  
695      // scatter/gather pot_row into the members of my column
696            
697 <    AtomCommPotRow->scatter(pot_row, pot_temp);
697 >    AtomPlanPotRow->scatter(pot_row, pot_temp);
698  
699      for (int ii = 0;  ii < pot_temp.size(); ii++ )
700 <      pot_local += pot_temp[ii];
700 >      pairwisePot += pot_temp[ii];
701      
702      fill(pot_temp.begin(), pot_temp.end(),
703           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
704        
705 <    AtomCommPotColumn->scatter(pot_col, pot_temp);    
705 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
706      
707      for (int ii = 0;  ii < pot_temp.size(); ii++ )
708 <      pot_local += pot_temp[ii];
708 >      pairwisePot += pot_temp[ii];    
709      
710 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
711 +      RealType ploc1 = pairwisePot[ii];
712 +      RealType ploc2 = 0.0;
713 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
714 +      pairwisePot[ii] = ploc2;
715 +    }
716 +
717   #endif
718 +
719    }
720  
721    int ForceMatrixDecomposition::getNAtomsInRow() {  
# Line 717 | Line 816 | namespace OpenMD {
816      return d;    
817    }
818  
819 <  vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
820 <    return skipsForAtom[atom1];
819 >  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
820 >    return excludesForAtom[atom1];
821    }
822  
823    /**
824 <   * 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
824 >   * We need to exclude some overcounted interactions that result from
825     * the parallel decomposition.
826     */
827    bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
828      int unique_id_1, unique_id_2;
829 <
829 >    
830   #ifdef IS_MPI
831      // in MPI, we have to look up the unique IDs for each atom
832      unique_id_1 = AtomRowToGlobal[atom1];
# Line 745 | Line 841 | namespace OpenMD {
841      } else {
842        if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
843      }
844 + #endif
845 +    return false;
846 +  }
847 +
848 +  /**
849 +   * We need to handle the interactions for atoms who are involved in
850 +   * the same rigid body as well as some short range interactions
851 +   * (bonds, bends, torsions) differently from other interactions.
852 +   * We'll still visit the pairwise routines, but with a flag that
853 +   * tells those routines to exclude the pair from direct long range
854 +   * interactions.  Some indirect interactions (notably reaction
855 +   * field) must still be handled for these pairs.
856 +   */
857 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
858 +    int unique_id_2;
859 + #ifdef IS_MPI
860 +    // in MPI, we have to look up the unique IDs for the row atom.
861 +    unique_id_2 = AtomColToGlobal[atom2];
862   #else
863      // in the normal loop, the atom numbers are unique
750 –    unique_id_1 = atom1;
864      unique_id_2 = atom2;
865   #endif
866      
867 <    for (vector<int>::iterator i = skipsForAtom[atom1].begin();
868 <         i != skipsForAtom[atom1].end(); ++i) {
867 >    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
868 >         i != excludesForAtom[atom1].end(); ++i) {
869        if ( (*i) == unique_id_2 ) return true;
870 <    }    
870 >    }
871  
872 +    return false;
873    }
874  
875  
# Line 776 | Line 890 | namespace OpenMD {
890    }
891  
892      // filling interaction blocks with pointers
893 <  void ForceMatrixDecomposition::fillInteractionData(InteractionData idat,
894 <                                                     int atom1, int atom2) {    
893 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
894 >                                                     int atom1, int atom2) {
895 >
896 >    idat.excluded = excludeAtomPair(atom1, atom2);
897 >  
898   #ifdef IS_MPI
899 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
900 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
901 +    //                         ff_->getAtomType(identsCol[atom2]) );
902      
783 –    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
784 –                             ff_->getAtomType(identsCol[atom2]) );
785 –    
903      if (storageLayout_ & DataStorage::dslAmat) {
904        idat.A1 = &(atomRowData.aMat[atom1]);
905        idat.A2 = &(atomColData.aMat[atom2]);
# Line 818 | Line 935 | namespace OpenMD {
935        idat.particlePot2 = &(atomColData.particlePot[atom2]);
936      }
937  
938 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
939 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
940 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
941 +    }
942 +
943   #else
944  
945 <    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
946 <                             ff_->getAtomType(identsLocal[atom2]) );
945 >    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
946 >    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
947 >    //                         ff_->getAtomType(idents[atom2]) );
948  
949      if (storageLayout_ & DataStorage::dslAmat) {
950        idat.A1 = &(snap_->atomData.aMat[atom1]);
# Line 838 | Line 961 | namespace OpenMD {
961        idat.t2 = &(snap_->atomData.torque[atom2]);
962      }
963  
964 <    if (storageLayout_ & DataStorage::dslDensity) {
964 >    if (storageLayout_ & DataStorage::dslDensity) {    
965        idat.rho1 = &(snap_->atomData.density[atom1]);
966        idat.rho2 = &(snap_->atomData.density[atom2]);
967      }
# Line 858 | Line 981 | namespace OpenMD {
981        idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
982      }
983  
984 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
985 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
986 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
987 +    }
988   #endif
989    }
990  
991    
992 <  void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {    
992 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
993   #ifdef IS_MPI
994      pot_row[atom1] += 0.5 *  *(idat.pot);
995      pot_col[atom2] += 0.5 *  *(idat.pot);
# Line 870 | Line 997 | namespace OpenMD {
997      atomRowData.force[atom1] += *(idat.f1);
998      atomColData.force[atom2] -= *(idat.f1);
999   #else
1000 <    longRangePot_ += *(idat.pot);
1001 <    
1000 >    pairwisePot += *(idat.pot);
1001 >
1002      snap_->atomData.force[atom1] += *(idat.f1);
1003      snap_->atomData.force[atom2] -= *(idat.f1);
1004   #endif
1005 <
1005 >    
1006    }
1007  
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 –
1008    /*
1009     * buildNeighborList
1010     *
# Line 918 | Line 1015 | namespace OpenMD {
1015        
1016      vector<pair<int, int> > neighborList;
1017      groupCutoffs cuts;
1018 +    bool doAllPairs = false;
1019 +
1020   #ifdef IS_MPI
1021      cellListRow_.clear();
1022      cellListCol_.clear();
# Line 937 | Line 1036 | namespace OpenMD {
1036      nCells_.y() = (int) ( Hy.length() )/ rList_;
1037      nCells_.z() = (int) ( Hz.length() )/ rList_;
1038  
1039 +    // handle small boxes where the cell offsets can end up repeating cells
1040 +    
1041 +    if (nCells_.x() < 3) doAllPairs = true;
1042 +    if (nCells_.y() < 3) doAllPairs = true;
1043 +    if (nCells_.z() < 3) doAllPairs = true;
1044 +
1045      Mat3x3d invHmat = snap_->getInvHmat();
1046      Vector3d rs, scaled, dr;
1047      Vector3i whichCell;
1048      int cellIndex;
1049      int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1050  
946 –    cerr << "flag1\n";
1051   #ifdef IS_MPI
1052      cellListRow_.resize(nCtot);
1053      cellListCol_.resize(nCtot);
1054   #else
1055      cellList_.resize(nCtot);
1056   #endif
1057 <    cerr << "flag2\n";
1057 >
1058 >    if (!doAllPairs) {
1059   #ifdef IS_MPI
955 –    for (int i = 0; i < nGroupsInRow_; i++) {
956 –      rs = cgRowData.position[i];
1060  
1061 <      // scaled positions relative to the box vectors
1062 <      scaled = invHmat * rs;
1063 <
1064 <      // wrap the vector back into the unit box by subtracting integer box
1065 <      // numbers
1066 <      for (int j = 0; j < 3; j++) {
1067 <        scaled[j] -= roundMe(scaled[j]);
1068 <        scaled[j] += 0.5;
1061 >      for (int i = 0; i < nGroupsInRow_; i++) {
1062 >        rs = cgRowData.position[i];
1063 >        
1064 >        // scaled positions relative to the box vectors
1065 >        scaled = invHmat * rs;
1066 >        
1067 >        // wrap the vector back into the unit box by subtracting integer box
1068 >        // numbers
1069 >        for (int j = 0; j < 3; j++) {
1070 >          scaled[j] -= roundMe(scaled[j]);
1071 >          scaled[j] += 0.5;
1072 >        }
1073 >        
1074 >        // find xyz-indices of cell that cutoffGroup is in.
1075 >        whichCell.x() = nCells_.x() * scaled.x();
1076 >        whichCell.y() = nCells_.y() * scaled.y();
1077 >        whichCell.z() = nCells_.z() * scaled.z();
1078 >        
1079 >        // find single index of this cell:
1080 >        cellIndex = Vlinear(whichCell, nCells_);
1081 >        
1082 >        // add this cutoff group to the list of groups in this cell;
1083 >        cellListRow_[cellIndex].push_back(i);
1084        }
1085 <    
1086 <      // find xyz-indices of cell that cutoffGroup is in.
1087 <      whichCell.x() = nCells_.x() * scaled.x();
1088 <      whichCell.y() = nCells_.y() * scaled.y();
1089 <      whichCell.z() = nCells_.z() * scaled.z();
1090 <
1091 <      // find single index of this cell:
1092 <      cellIndex = Vlinear(whichCell, nCells_);
1093 <
1094 <      // add this cutoff group to the list of groups in this cell;
1095 <      cellListRow_[cellIndex].push_back(i);
1096 <    }
1097 <
1098 <    for (int i = 0; i < nGroupsInCol_; i++) {
1099 <      rs = cgColData.position[i];
1100 <
1101 <      // scaled positions relative to the box vectors
1102 <      scaled = invHmat * rs;
1103 <
1104 <      // wrap the vector back into the unit box by subtracting integer box
1105 <      // numbers
1106 <      for (int j = 0; j < 3; j++) {
1107 <        scaled[j] -= roundMe(scaled[j]);
990 <        scaled[j] += 0.5;
1085 >      for (int i = 0; i < nGroupsInCol_; i++) {
1086 >        rs = cgColData.position[i];
1087 >        
1088 >        // scaled positions relative to the box vectors
1089 >        scaled = invHmat * rs;
1090 >        
1091 >        // wrap the vector back into the unit box by subtracting integer box
1092 >        // numbers
1093 >        for (int j = 0; j < 3; j++) {
1094 >          scaled[j] -= roundMe(scaled[j]);
1095 >          scaled[j] += 0.5;
1096 >        }
1097 >        
1098 >        // find xyz-indices of cell that cutoffGroup is in.
1099 >        whichCell.x() = nCells_.x() * scaled.x();
1100 >        whichCell.y() = nCells_.y() * scaled.y();
1101 >        whichCell.z() = nCells_.z() * scaled.z();
1102 >        
1103 >        // find single index of this cell:
1104 >        cellIndex = Vlinear(whichCell, nCells_);
1105 >        
1106 >        // add this cutoff group to the list of groups in this cell;
1107 >        cellListCol_[cellIndex].push_back(i);
1108        }
1109 <
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 <    }
1109 >    
1110   #else
1111 <    for (int i = 0; i < nGroups_; i++) {
1112 <      rs = snap_->cgData.position[i];
1113 <
1114 <      // scaled positions relative to the box vectors
1115 <      scaled = invHmat * rs;
1116 <
1117 <      // wrap the vector back into the unit box by subtracting integer box
1118 <      // numbers
1119 <      for (int j = 0; j < 3; j++) {
1120 <        scaled[j] -= roundMe(scaled[j]);
1121 <        scaled[j] += 0.5;
1111 >      for (int i = 0; i < nGroups_; i++) {
1112 >        rs = snap_->cgData.position[i];
1113 >        
1114 >        // scaled positions relative to the box vectors
1115 >        scaled = invHmat * rs;
1116 >        
1117 >        // wrap the vector back into the unit box by subtracting integer box
1118 >        // numbers
1119 >        for (int j = 0; j < 3; j++) {
1120 >          scaled[j] -= roundMe(scaled[j]);
1121 >          scaled[j] += 0.5;
1122 >        }
1123 >        
1124 >        // find xyz-indices of cell that cutoffGroup is in.
1125 >        whichCell.x() = nCells_.x() * scaled.x();
1126 >        whichCell.y() = nCells_.y() * scaled.y();
1127 >        whichCell.z() = nCells_.z() * scaled.z();
1128 >        
1129 >        // find single index of this cell:
1130 >        cellIndex = Vlinear(whichCell, nCells_);
1131 >        
1132 >        // add this cutoff group to the list of groups in this cell;
1133 >        cellList_[cellIndex].push_back(i);
1134        }
1135  
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 –    }
1136   #endif
1137  
1138 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1139 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1140 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1141 <          Vector3i m1v(m1x, m1y, m1z);
1142 <          int m1 = Vlinear(m1v, nCells_);
1036 <
1037 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1038 <               os != cellOffsets_.end(); ++os) {
1138 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1139 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1140 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1141 >            Vector3i m1v(m1x, m1y, m1z);
1142 >            int m1 = Vlinear(m1v, nCells_);
1143              
1144 <            Vector3i m2v = m1v + (*os);
1145 <            
1146 <            if (m2v.x() >= nCells_.x()) {
1147 <              m2v.x() = 0;          
1148 <            } else if (m2v.x() < 0) {
1045 <              m2v.x() = nCells_.x() - 1;
1046 <            }
1047 <            
1048 <            if (m2v.y() >= nCells_.y()) {
1049 <              m2v.y() = 0;          
1050 <            } else if (m2v.y() < 0) {
1051 <              m2v.y() = nCells_.y() - 1;
1052 <            }
1053 <            
1054 <            if (m2v.z() >= nCells_.z()) {
1055 <              m2v.z() = 0;          
1056 <            } else if (m2v.z() < 0) {
1057 <              m2v.z() = nCells_.z() - 1;
1058 <            }
1059 <            
1060 <            int m2 = Vlinear (m2v, nCells_);
1144 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1145 >                 os != cellOffsets_.end(); ++os) {
1146 >              
1147 >              Vector3i m2v = m1v + (*os);
1148 >            
1149  
1150 < #ifdef IS_MPI
1151 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1152 <                 j1 != cellListRow_[m1].end(); ++j1) {
1153 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1154 <                   j2 != cellListCol_[m2].end(); ++j2) {
1155 <                              
1156 <                // Always do this if we're in different cells or if
1157 <                // we're in the same cell and the global index of the
1158 <                // j2 cutoff group is less than the j1 cutoff group
1150 >              if (m2v.x() >= nCells_.x()) {
1151 >                m2v.x() = 0;          
1152 >              } else if (m2v.x() < 0) {
1153 >                m2v.x() = nCells_.x() - 1;
1154 >              }
1155 >              
1156 >              if (m2v.y() >= nCells_.y()) {
1157 >                m2v.y() = 0;          
1158 >              } else if (m2v.y() < 0) {
1159 >                m2v.y() = nCells_.y() - 1;
1160 >              }
1161 >              
1162 >              if (m2v.z() >= nCells_.z()) {
1163 >                m2v.z() = 0;          
1164 >              } else if (m2v.z() < 0) {
1165 >                m2v.z() = nCells_.z() - 1;
1166 >              }
1167  
1168 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1168 >              int m2 = Vlinear (m2v, nCells_);
1169 >              
1170 > #ifdef IS_MPI
1171 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1172 >                   j1 != cellListRow_[m1].end(); ++j1) {
1173 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1174 >                     j2 != cellListCol_[m2].end(); ++j2) {
1175 >                  
1176 >                  // In parallel, we need to visit *all* pairs of row
1177 >                  // & column indicies and will divide labor in the
1178 >                  // force evaluation later.
1179                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1180                    snap_->wrapVector(dr);
1181                    cuts = getGroupCutoffs( (*j1), (*j2) );
1182                    if (dr.lengthSquare() < cuts.third) {
1183                      neighborList.push_back(make_pair((*j1), (*j2)));
1184 <                  }
1184 >                  }                  
1185                  }
1186                }
1081 –            }
1187   #else
1188 <
1189 <            for (vector<int>::iterator j1 = cellList_[m1].begin();
1190 <                 j1 != cellList_[m1].end(); ++j1) {
1191 <              for (vector<int>::iterator j2 = cellList_[m2].begin();
1192 <                   j2 != cellList_[m2].end(); ++j2) {
1193 <
1194 <                // Always do this if we're in different cells or if
1195 <                // we're in the same cell and the global index of the
1196 <                // j2 cutoff group is less than the j1 cutoff group
1197 <
1198 <                if (m2 != m1 || (*j2) < (*j1)) {
1199 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1200 <                  snap_->wrapVector(dr);
1201 <                  cuts = getGroupCutoffs( (*j1), (*j2) );
1202 <                  if (dr.lengthSquare() < cuts.third) {
1203 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1188 >              
1189 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1190 >                   j1 != cellList_[m1].end(); ++j1) {
1191 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1192 >                     j2 != cellList_[m2].end(); ++j2) {
1193 >                  
1194 >                  // Always do this if we're in different cells or if
1195 >                  // we're in the same cell and the global index of the
1196 >                  // j2 cutoff group is less than the j1 cutoff group
1197 >                  
1198 >                  if (m2 != m1 || (*j2) < (*j1)) {
1199 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1200 >                    snap_->wrapVector(dr);
1201 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1202 >                    if (dr.lengthSquare() < cuts.third) {
1203 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1204 >                    }
1205                    }
1206                  }
1207                }
1102 –            }
1208   #endif
1209 +            }
1210            }
1211          }
1212        }
1213 +    } else {
1214 +      // branch to do all cutoff group pairs
1215 + #ifdef IS_MPI
1216 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1217 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1218 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1219 +          snap_->wrapVector(dr);
1220 +          cuts = getGroupCutoffs( j1, j2 );
1221 +          if (dr.lengthSquare() < cuts.third) {
1222 +            neighborList.push_back(make_pair(j1, j2));
1223 +          }
1224 +        }
1225 +      }
1226 + #else
1227 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1228 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1229 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1230 +          snap_->wrapVector(dr);
1231 +          cuts = getGroupCutoffs( j1, j2 );
1232 +          if (dr.lengthSquare() < cuts.third) {
1233 +            neighborList.push_back(make_pair(j1, j2));
1234 +          }
1235 +        }
1236 +      }        
1237 + #endif
1238      }
1239 <    
1239 >      
1240      // save the local cutoff group positions for the check that is
1241      // done on each loop:
1242      saved_CG_positions_.clear();

Diff Legend

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