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 1589 by gezelter, Sun Jul 10 16:05:34 2011 UTC vs.
Revision 1771 by gezelter, Fri Jul 27 17:34:10 2012 UTC

# Line 36 | Line 36
36   * [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).            
37   * [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).          
38   * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
39 < * [4]  Vardeman & Gezelter, in progress (2009).                        
39 > * [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40 > * [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41   */
42   #include "parallel/ForceMatrixDecomposition.hpp"
43   #include "math/SquareMatrix3.hpp"
# Line 47 | Line 48 | namespace OpenMD {
48   using namespace std;
49   namespace OpenMD {
50  
51 +  ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52 +
53 +    // In a parallel computation, row and colum scans must visit all
54 +    // surrounding cells (not just the 14 upper triangular blocks that
55 +    // are used when the processor can see all pairs)
56 + #ifdef IS_MPI
57 +    cellOffsets_.clear();
58 +    cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59 +    cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60 +    cellOffsets_.push_back( Vector3i( 1,-1,-1) );                          
61 +    cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62 +    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63 +    cellOffsets_.push_back( Vector3i( 1, 0,-1) );
64 +    cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65 +    cellOffsets_.push_back( Vector3i( 0, 1,-1) );      
66 +    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67 +    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68 +    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69 +    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70 +    cellOffsets_.push_back( Vector3i(-1, 0, 0) );      
71 +    cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72 +    cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73 +    cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74 +    cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75 +    cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76 +    cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77 +    cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78 +    cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79 +    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80 +    cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81 +    cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82 +    cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83 +    cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84 +    cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85 + #endif    
86 +  }
87 +
88 +
89    /**
90     * distributeInitialData is essentially a copy of the older fortran
91     * SimulationSetup
92     */
54 –  
93    void ForceMatrixDecomposition::distributeInitialData() {
94      snap_ = sman_->getCurrentSnapshot();
95      storageLayout_ = sman_->getStorageLayout();
96      ff_ = info_->getForceField();
97      nLocal_ = snap_->getNumberOfAtoms();
98 <    
98 >  
99      nGroups_ = info_->getNLocalCutoffGroups();
100      // gather the information for atomtype IDs (atids):
101      idents = info_->getIdentArray();
# Line 71 | Line 109 | namespace OpenMD {
109      PairList* oneTwo = info_->getOneTwoInteractions();
110      PairList* oneThree = info_->getOneThreeInteractions();
111      PairList* oneFour = info_->getOneFourInteractions();
112 <
112 >    
113 >    if (needVelocities_)
114 >      snap_->cgData.setStorageLayout(DataStorage::dslPosition |
115 >                                     DataStorage::dslVelocity);
116 >    else
117 >      snap_->cgData.setStorageLayout(DataStorage::dslPosition);
118 >    
119   #ifdef IS_MPI
120  
121 <    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
122 <    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
79 <    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
80 <    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
81 <    AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
121 >    MPI::Intracomm row = rowComm.getComm();
122 >    MPI::Intracomm col = colComm.getComm();
123  
124 <    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
125 <    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
126 <    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
127 <    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
128 <    AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
124 >    AtomPlanIntRow = new Plan<int>(row, nLocal_);
125 >    AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
126 >    AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
127 >    AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
128 >    AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
129  
130 <    cgCommIntRow = new Communicator<Row,int>(nGroups_);
131 <    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
132 <    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
133 <    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
130 >    AtomPlanIntColumn = new Plan<int>(col, nLocal_);
131 >    AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
132 >    AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
133 >    AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
134 >    AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
135  
136 <    nAtomsInRow_ = AtomCommIntRow->getSize();
137 <    nAtomsInCol_ = AtomCommIntColumn->getSize();
138 <    nGroupsInRow_ = cgCommIntRow->getSize();
139 <    nGroupsInCol_ = cgCommIntColumn->getSize();
136 >    cgPlanIntRow = new Plan<int>(row, nGroups_);
137 >    cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
138 >    cgPlanIntColumn = new Plan<int>(col, nGroups_);
139 >    cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
140  
141 +    nAtomsInRow_ = AtomPlanIntRow->getSize();
142 +    nAtomsInCol_ = AtomPlanIntColumn->getSize();
143 +    nGroupsInRow_ = cgPlanIntRow->getSize();
144 +    nGroupsInCol_ = cgPlanIntColumn->getSize();
145 +
146      // Modify the data storage objects with the correct layouts and sizes:
147      atomRowData.resize(nAtomsInRow_);
148      atomRowData.setStorageLayout(storageLayout_);
# Line 104 | Line 151 | namespace OpenMD {
151      cgRowData.resize(nGroupsInRow_);
152      cgRowData.setStorageLayout(DataStorage::dslPosition);
153      cgColData.resize(nGroupsInCol_);
154 <    cgColData.setStorageLayout(DataStorage::dslPosition);
155 <        
154 >    if (needVelocities_)
155 >      // we only need column velocities if we need them.
156 >      cgColData.setStorageLayout(DataStorage::dslPosition |
157 >                                 DataStorage::dslVelocity);
158 >    else    
159 >      cgColData.setStorageLayout(DataStorage::dslPosition);
160 >      
161      identsRow.resize(nAtomsInRow_);
162      identsCol.resize(nAtomsInCol_);
163      
164 <    AtomCommIntRow->gather(idents, identsRow);
165 <    AtomCommIntColumn->gather(idents, identsCol);
164 >    AtomPlanIntRow->gather(idents, identsRow);
165 >    AtomPlanIntColumn->gather(idents, identsCol);
166      
167      // allocate memory for the parallel objects
168 +    atypesRow.resize(nAtomsInRow_);
169 +    atypesCol.resize(nAtomsInCol_);
170 +
171 +    for (int i = 0; i < nAtomsInRow_; i++)
172 +      atypesRow[i] = ff_->getAtomType(identsRow[i]);
173 +    for (int i = 0; i < nAtomsInCol_; i++)
174 +      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
175 +
176 +    pot_row.resize(nAtomsInRow_);
177 +    pot_col.resize(nAtomsInCol_);
178 +
179 +    expot_row.resize(nAtomsInRow_);
180 +    expot_col.resize(nAtomsInCol_);
181 +
182      AtomRowToGlobal.resize(nAtomsInRow_);
183      AtomColToGlobal.resize(nAtomsInCol_);
184 +    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185 +    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186 +
187      cgRowToGlobal.resize(nGroupsInRow_);
188      cgColToGlobal.resize(nGroupsInCol_);
189 +    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190 +    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
191 +
192      massFactorsRow.resize(nAtomsInRow_);
193      massFactorsCol.resize(nAtomsInCol_);
194 <    pot_row.resize(nAtomsInRow_);
195 <    pot_col.resize(nAtomsInCol_);
194 >    AtomPlanRealRow->gather(massFactors, massFactorsRow);
195 >    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
196  
125 –    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
126 –    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
127 –    
128 –    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
129 –    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
130 –
131 –    AtomCommRealRow->gather(massFactors, massFactorsRow);
132 –    AtomCommRealColumn->gather(massFactors, massFactorsCol);
133 –
197      groupListRow_.clear();
198      groupListRow_.resize(nGroupsInRow_);
199      for (int i = 0; i < nGroupsInRow_; i++) {
# Line 185 | Line 248 | namespace OpenMD {
248        }      
249      }
250  
251 < #endif
189 <
190 <    groupList_.clear();
191 <    groupList_.resize(nGroups_);
192 <    for (int i = 0; i < nGroups_; i++) {
193 <      int gid = cgLocalToGlobal[i];
194 <      for (int j = 0; j < nLocal_; j++) {
195 <        int aid = AtomLocalToGlobal[j];
196 <        if (globalGroupMembership[aid] == gid) {
197 <          groupList_[i].push_back(j);
198 <        }
199 <      }      
200 <    }
201 <
251 > #else
252      excludesForAtom.clear();
253      excludesForAtom.resize(nLocal_);
254      toposForAtom.clear();
# Line 231 | Line 281 | namespace OpenMD {
281          }
282        }      
283      }
284 <    
284 > #endif
285 >
286 >    // allocate memory for the parallel objects
287 >    atypesLocal.resize(nLocal_);
288 >
289 >    for (int i = 0; i < nLocal_; i++)
290 >      atypesLocal[i] = ff_->getAtomType(idents[i]);
291 >
292 >    groupList_.clear();
293 >    groupList_.resize(nGroups_);
294 >    for (int i = 0; i < nGroups_; i++) {
295 >      int gid = cgLocalToGlobal[i];
296 >      for (int j = 0; j < nLocal_; j++) {
297 >        int aid = AtomLocalToGlobal[j];
298 >        if (globalGroupMembership[aid] == gid) {
299 >          groupList_[i].push_back(j);
300 >        }
301 >      }      
302 >    }
303 >
304 >
305      createGtypeCutoffMap();
306  
307    }
# Line 239 | Line 309 | namespace OpenMD {
309    void ForceMatrixDecomposition::createGtypeCutoffMap() {
310      
311      RealType tol = 1e-6;
312 <    RealType rc;
312 >    largestRcut_ = 0.0;
313      int atid;
314      set<AtomType*> atypes = info_->getSimulatedAtomTypes();
315 +    
316      map<int, RealType> atypeCutoff;
317        
318      for (set<AtomType*>::iterator at = atypes.begin();
# Line 249 | Line 320 | namespace OpenMD {
320        atid = (*at)->getIdent();
321        if (userChoseCutoff_)
322          atypeCutoff[atid] = userCutoff_;
323 <      else
323 >      else
324          atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
325      }
326 <
326 >    
327      vector<RealType> gTypeCutoffs;
328      // first we do a single loop over the cutoff groups to find the
329      // largest cutoff for any atypes present in this group.
# Line 312 | Line 383 | namespace OpenMD {
383      vector<RealType> groupCutoff(nGroups_, 0.0);
384      groupToGtype.resize(nGroups_);
385      for (int cg1 = 0; cg1 < nGroups_; cg1++) {
315 –
386        groupCutoff[cg1] = 0.0;
387        vector<int> atomList = getAtomsInGroupRow(cg1);
318 –
388        for (vector<int>::iterator ia = atomList.begin();
389             ia != atomList.end(); ++ia) {            
390          int atom1 = (*ia);
391          atid = idents[atom1];
392 <        if (atypeCutoff[atid] > groupCutoff[cg1]) {
392 >        if (atypeCutoff[atid] > groupCutoff[cg1])
393            groupCutoff[cg1] = atypeCutoff[atid];
325 –        }
394        }
395 <
395 >      
396        bool gTypeFound = false;
397 <      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
397 >      for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
398          if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
399            groupToGtype[cg1] = gt;
400            gTypeFound = true;
401          }
402        }
403 <      if (!gTypeFound) {
403 >      if (!gTypeFound) {      
404          gTypeCutoffs.push_back( groupCutoff[cg1] );
405          groupToGtype[cg1] = gTypeCutoffs.size() - 1;
406        }      
# Line 341 | Line 409 | namespace OpenMD {
409  
410      // Now we find the maximum group cutoff value present in the simulation
411  
412 <    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
412 >    RealType groupMax = *max_element(gTypeCutoffs.begin(),
413 >                                     gTypeCutoffs.end());
414  
415   #ifdef IS_MPI
416 <    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
416 >    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
417 >                              MPI::MAX);
418   #endif
419      
420      RealType tradRcut = groupMax;
421  
422 <    for (int i = 0; i < gTypeCutoffs.size();  i++) {
423 <      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
422 >    for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) {
423 >      for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) {      
424          RealType thisRcut;
425          switch(cutoffPolicy_) {
426          case TRADITIONAL:
# Line 374 | Line 444 | namespace OpenMD {
444  
445          pair<int,int> key = make_pair(i,j);
446          gTypeCutoffMap[key].first = thisRcut;
377 –
447          if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
379 –
448          gTypeCutoffMap[key].second = thisRcut*thisRcut;
381 –        
449          gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
383 –
450          // sanity check
451          
452          if (userChoseCutoff_) {
# Line 397 | Line 463 | namespace OpenMD {
463      }
464    }
465  
400 –
466    groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
467      int i, j;  
468   #ifdef IS_MPI
# Line 411 | Line 476 | namespace OpenMD {
476    }
477  
478    int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
479 <    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
479 >    for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
480        if (toposForAtom[atom1][j] == atom2)
481          return topoDist[atom1][j];
482      }
# Line 421 | Line 486 | namespace OpenMD {
486    void ForceMatrixDecomposition::zeroWorkArrays() {
487      pairwisePot = 0.0;
488      embeddingPot = 0.0;
489 +    excludedPot = 0.0;
490 +    excludedSelfPot = 0.0;
491  
492   #ifdef IS_MPI
493      if (storageLayout_ & DataStorage::dslForce) {
# Line 439 | Line 506 | namespace OpenMD {
506      fill(pot_col.begin(), pot_col.end(),
507           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
508  
509 +    fill(expot_row.begin(), expot_row.end(),
510 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
511 +
512 +    fill(expot_col.begin(), expot_col.end(),
513 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
514 +
515      if (storageLayout_ & DataStorage::dslParticlePot) {    
516 <      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
517 <      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
516 >      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
517 >           0.0);
518 >      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
519 >           0.0);
520      }
521  
522      if (storageLayout_ & DataStorage::dslDensity) {      
# Line 450 | Line 525 | namespace OpenMD {
525      }
526  
527      if (storageLayout_ & DataStorage::dslFunctional) {  
528 <      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
529 <      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
528 >      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
529 >           0.0);
530 >      fill(atomColData.functional.begin(), atomColData.functional.end(),
531 >           0.0);
532      }
533  
534      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
# Line 468 | Line 545 | namespace OpenMD {
545             atomColData.skippedCharge.end(), 0.0);
546      }
547  
548 < #else
549 <    
548 >    if (storageLayout_ & DataStorage::dslFlucQForce) {      
549 >      fill(atomRowData.flucQFrc.begin(),
550 >           atomRowData.flucQFrc.end(), 0.0);
551 >      fill(atomColData.flucQFrc.begin(),
552 >           atomColData.flucQFrc.end(), 0.0);
553 >    }
554 >
555 >    if (storageLayout_ & DataStorage::dslElectricField) {    
556 >      fill(atomRowData.electricField.begin(),
557 >           atomRowData.electricField.end(), V3Zero);
558 >      fill(atomColData.electricField.begin(),
559 >           atomColData.electricField.end(), V3Zero);
560 >    }
561 >
562 >    if (storageLayout_ & DataStorage::dslFlucQForce) {    
563 >      fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(),
564 >           0.0);
565 >      fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.end(),
566 >           0.0);
567 >    }
568 >
569 > #endif
570 >    // even in parallel, we need to zero out the local arrays:
571 >
572      if (storageLayout_ & DataStorage::dslParticlePot) {      
573        fill(snap_->atomData.particlePot.begin(),
574             snap_->atomData.particlePot.end(), 0.0);
# Line 479 | Line 578 | namespace OpenMD {
578        fill(snap_->atomData.density.begin(),
579             snap_->atomData.density.end(), 0.0);
580      }
581 +
582      if (storageLayout_ & DataStorage::dslFunctional) {
583        fill(snap_->atomData.functional.begin(),
584             snap_->atomData.functional.end(), 0.0);
585      }
586 +
587      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
588        fill(snap_->atomData.functionalDerivative.begin(),
589             snap_->atomData.functionalDerivative.end(), 0.0);
590      }
591 +
592      if (storageLayout_ & DataStorage::dslSkippedCharge) {      
593        fill(snap_->atomData.skippedCharge.begin(),
594             snap_->atomData.skippedCharge.end(), 0.0);
595      }
596 < #endif
597 <    
596 >
597 >    if (storageLayout_ & DataStorage::dslElectricField) {      
598 >      fill(snap_->atomData.electricField.begin(),
599 >           snap_->atomData.electricField.end(), V3Zero);
600 >    }
601    }
602  
603  
# Line 502 | Line 607 | namespace OpenMD {
607   #ifdef IS_MPI
608      
609      // gather up the atomic positions
610 <    AtomCommVectorRow->gather(snap_->atomData.position,
610 >    AtomPlanVectorRow->gather(snap_->atomData.position,
611                                atomRowData.position);
612 <    AtomCommVectorColumn->gather(snap_->atomData.position,
612 >    AtomPlanVectorColumn->gather(snap_->atomData.position,
613                                   atomColData.position);
614      
615      // gather up the cutoff group positions
616 <    cgCommVectorRow->gather(snap_->cgData.position,
617 <                            cgRowData.position);
618 <    cgCommVectorColumn->gather(snap_->cgData.position,
616 >
617 >    cgPlanVectorRow->gather(snap_->cgData.position,
618 >                            cgRowData.position);
619 >
620 >    cgPlanVectorColumn->gather(snap_->cgData.position,
621                                 cgColData.position);
622 +
623 +
624 +
625 +    if (needVelocities_) {
626 +      // gather up the atomic velocities
627 +      AtomPlanVectorColumn->gather(snap_->atomData.velocity,
628 +                                   atomColData.velocity);
629 +      
630 +      cgPlanVectorColumn->gather(snap_->cgData.velocity,
631 +                                 cgColData.velocity);
632 +    }
633 +
634      
635      // if needed, gather the atomic rotation matrices
636      if (storageLayout_ & DataStorage::dslAmat) {
637 <      AtomCommMatrixRow->gather(snap_->atomData.aMat,
637 >      AtomPlanMatrixRow->gather(snap_->atomData.aMat,
638                                  atomRowData.aMat);
639 <      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
639 >      AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
640                                     atomColData.aMat);
641      }
642      
643      // if needed, gather the atomic eletrostatic frames
644      if (storageLayout_ & DataStorage::dslElectroFrame) {
645 <      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
645 >      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
646                                  atomRowData.electroFrame);
647 <      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
647 >      AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
648                                     atomColData.electroFrame);
649      }
650 +
651 +    // if needed, gather the atomic fluctuating charge values
652 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {
653 +      AtomPlanRealRow->gather(snap_->atomData.flucQPos,
654 +                              atomRowData.flucQPos);
655 +      AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
656 +                                 atomColData.flucQPos);
657 +    }
658 +
659   #endif      
660    }
661    
# Line 541 | Line 669 | namespace OpenMD {
669      
670      if (storageLayout_ & DataStorage::dslDensity) {
671        
672 <      AtomCommRealRow->scatter(atomRowData.density,
672 >      AtomPlanRealRow->scatter(atomRowData.density,
673                                 snap_->atomData.density);
674        
675        int n = snap_->atomData.density.size();
676        vector<RealType> rho_tmp(n, 0.0);
677 <      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
677 >      AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
678        for (int i = 0; i < n; i++)
679          snap_->atomData.density[i] += rho_tmp[i];
680      }
681 +
682 +    if (storageLayout_ & DataStorage::dslElectricField) {
683 +      
684 +      AtomPlanVectorRow->scatter(atomRowData.electricField,
685 +                                 snap_->atomData.electricField);
686 +      
687 +      int n = snap_->atomData.electricField.size();
688 +      vector<Vector3d> field_tmp(n, V3Zero);
689 +      AtomPlanVectorColumn->scatter(atomColData.electricField, field_tmp);
690 +      for (int i = 0; i < n; i++)
691 +        snap_->atomData.electricField[i] += field_tmp[i];
692 +    }
693   #endif
694    }
695  
# Line 562 | Line 702 | namespace OpenMD {
702      storageLayout_ = sman_->getStorageLayout();
703   #ifdef IS_MPI
704      if (storageLayout_ & DataStorage::dslFunctional) {
705 <      AtomCommRealRow->gather(snap_->atomData.functional,
705 >      AtomPlanRealRow->gather(snap_->atomData.functional,
706                                atomRowData.functional);
707 <      AtomCommRealColumn->gather(snap_->atomData.functional,
707 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
708                                   atomColData.functional);
709      }
710      
711      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
712 <      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
712 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
713                                atomRowData.functionalDerivative);
714 <      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
714 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
715                                   atomColData.functionalDerivative);
716      }
717   #endif
# Line 585 | Line 725 | namespace OpenMD {
725      int n = snap_->atomData.force.size();
726      vector<Vector3d> frc_tmp(n, V3Zero);
727      
728 <    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
728 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
729      for (int i = 0; i < n; i++) {
730        snap_->atomData.force[i] += frc_tmp[i];
731        frc_tmp[i] = 0.0;
732      }
733      
734 <    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
735 <    for (int i = 0; i < n; i++)
734 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
735 >    for (int i = 0; i < n; i++) {
736        snap_->atomData.force[i] += frc_tmp[i];
737 <    
738 <    
737 >    }
738 >        
739      if (storageLayout_ & DataStorage::dslTorque) {
740  
741        int nt = snap_->atomData.torque.size();
742        vector<Vector3d> trq_tmp(nt, V3Zero);
743  
744 <      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
744 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
745        for (int i = 0; i < nt; i++) {
746          snap_->atomData.torque[i] += trq_tmp[i];
747          trq_tmp[i] = 0.0;
748        }
749        
750 <      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
750 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
751        for (int i = 0; i < nt; i++)
752          snap_->atomData.torque[i] += trq_tmp[i];
753      }
# Line 617 | Line 757 | namespace OpenMD {
757        int ns = snap_->atomData.skippedCharge.size();
758        vector<RealType> skch_tmp(ns, 0.0);
759  
760 <      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
760 >      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
761        for (int i = 0; i < ns; i++) {
762 <        snap_->atomData.skippedCharge[i] = skch_tmp[i];
762 >        snap_->atomData.skippedCharge[i] += skch_tmp[i];
763          skch_tmp[i] = 0.0;
764        }
765        
766 <      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
767 <      for (int i = 0; i < ns; i++)
766 >      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
767 >      for (int i = 0; i < ns; i++)
768          snap_->atomData.skippedCharge[i] += skch_tmp[i];
769 +            
770      }
771      
772 +    if (storageLayout_ & DataStorage::dslFlucQForce) {
773 +
774 +      int nq = snap_->atomData.flucQFrc.size();
775 +      vector<RealType> fqfrc_tmp(nq, 0.0);
776 +
777 +      AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
778 +      for (int i = 0; i < nq; i++) {
779 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
780 +        fqfrc_tmp[i] = 0.0;
781 +      }
782 +      
783 +      AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
784 +      for (int i = 0; i < nq; i++)
785 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
786 +            
787 +    }
788 +
789      nLocal_ = snap_->getNumberOfAtoms();
790  
791      vector<potVec> pot_temp(nLocal_,
792                              Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
793 +    vector<potVec> expot_temp(nLocal_,
794 +                              Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
795  
796      // scatter/gather pot_row into the members of my column
797            
798 <    AtomCommPotRow->scatter(pot_row, pot_temp);
798 >    AtomPlanPotRow->scatter(pot_row, pot_temp);
799 >    AtomPlanPotRow->scatter(expot_row, expot_temp);
800  
801 <    for (int ii = 0;  ii < pot_temp.size(); ii++ )
801 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
802        pairwisePot += pot_temp[ii];
803 <    
803 >
804 >    for (int ii = 0;  ii < expot_temp.size(); ii++ )
805 >      excludedPot += expot_temp[ii];
806 >        
807 >    if (storageLayout_ & DataStorage::dslParticlePot) {
808 >      // This is the pairwise contribution to the particle pot.  The
809 >      // embedding contribution is added in each of the low level
810 >      // non-bonded routines.  In single processor, this is done in
811 >      // unpackInteractionData, not in collectData.
812 >      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
813 >        for (int i = 0; i < nLocal_; i++) {
814 >          // factor of two is because the total potential terms are divided
815 >          // by 2 in parallel due to row/ column scatter      
816 >          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
817 >        }
818 >      }
819 >    }
820 >
821      fill(pot_temp.begin(), pot_temp.end(),
822           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
823 +    fill(expot_temp.begin(), expot_temp.end(),
824 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
825        
826 <    AtomCommPotColumn->scatter(pot_col, pot_temp);    
826 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
827 >    AtomPlanPotColumn->scatter(expot_col, expot_temp);    
828      
829      for (int ii = 0;  ii < pot_temp.size(); ii++ )
830        pairwisePot += pot_temp[ii];    
831 +
832 +    for (int ii = 0;  ii < expot_temp.size(); ii++ )
833 +      excludedPot += expot_temp[ii];    
834 +
835 +    if (storageLayout_ & DataStorage::dslParticlePot) {
836 +      // This is the pairwise contribution to the particle pot.  The
837 +      // embedding contribution is added in each of the low level
838 +      // non-bonded routines.  In single processor, this is done in
839 +      // unpackInteractionData, not in collectData.
840 +      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
841 +        for (int i = 0; i < nLocal_; i++) {
842 +          // factor of two is because the total potential terms are divided
843 +          // by 2 in parallel due to row/ column scatter      
844 +          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
845 +        }
846 +      }
847 +    }
848 +    
849 +    if (storageLayout_ & DataStorage::dslParticlePot) {
850 +      int npp = snap_->atomData.particlePot.size();
851 +      vector<RealType> ppot_temp(npp, 0.0);
852 +
853 +      // This is the direct or embedding contribution to the particle
854 +      // pot.
855 +      
856 +      AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
857 +      for (int i = 0; i < npp; i++) {
858 +        snap_->atomData.particlePot[i] += ppot_temp[i];
859 +      }
860 +
861 +      fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
862 +      
863 +      AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
864 +      for (int i = 0; i < npp; i++) {
865 +        snap_->atomData.particlePot[i] += ppot_temp[i];
866 +      }
867 +    }
868 +
869 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
870 +      RealType ploc1 = pairwisePot[ii];
871 +      RealType ploc2 = 0.0;
872 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
873 +      pairwisePot[ii] = ploc2;
874 +    }
875 +
876 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
877 +      RealType ploc1 = excludedPot[ii];
878 +      RealType ploc2 = 0.0;
879 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
880 +      excludedPot[ii] = ploc2;
881 +    }
882 +
883 +    // Here be dragons.
884 +    MPI::Intracomm col = colComm.getComm();
885 +
886 +    col.Allreduce(MPI::IN_PLACE,
887 +                  &snap_->frameData.conductiveHeatFlux[0], 3,
888 +                  MPI::REALTYPE, MPI::SUM);
889 +
890 +
891   #endif
892  
893    }
894  
895 +  /**
896 +   * Collects information obtained during the post-pair (and embedding
897 +   * functional) loops onto local data structures.
898 +   */
899 +  void ForceMatrixDecomposition::collectSelfData() {
900 +    snap_ = sman_->getCurrentSnapshot();
901 +    storageLayout_ = sman_->getStorageLayout();
902 +
903 + #ifdef IS_MPI
904 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
905 +      RealType ploc1 = embeddingPot[ii];
906 +      RealType ploc2 = 0.0;
907 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
908 +      embeddingPot[ii] = ploc2;
909 +    }    
910 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
911 +      RealType ploc1 = excludedSelfPot[ii];
912 +      RealType ploc2 = 0.0;
913 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
914 +      excludedSelfPot[ii] = ploc2;
915 +    }    
916 + #endif
917 +    
918 +  }
919 +
920 +
921 +
922    int ForceMatrixDecomposition::getNAtomsInRow() {  
923   #ifdef IS_MPI
924      return nAtomsInRow_;
# Line 691 | Line 959 | namespace OpenMD {
959      return d;    
960    }
961  
962 +  Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
963 + #ifdef IS_MPI
964 +    return cgColData.velocity[cg2];
965 + #else
966 +    return snap_->cgData.velocity[cg2];
967 + #endif
968 +  }
969  
970 +  Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
971 + #ifdef IS_MPI
972 +    return atomColData.velocity[atom2];
973 + #else
974 +    return snap_->atomData.velocity[atom2];
975 + #endif
976 +  }
977 +
978 +
979    Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
980  
981      Vector3d d;
# Line 757 | Line 1041 | namespace OpenMD {
1041     * We need to exclude some overcounted interactions that result from
1042     * the parallel decomposition.
1043     */
1044 <  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1045 <    int unique_id_1, unique_id_2;
1046 <
1044 >  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1045 >    int unique_id_1, unique_id_2, group1, group2;
1046 >        
1047   #ifdef IS_MPI
1048      // in MPI, we have to look up the unique IDs for each atom
1049      unique_id_1 = AtomRowToGlobal[atom1];
1050      unique_id_2 = AtomColToGlobal[atom2];
1051 <
1052 <    // this situation should only arise in MPI simulations
1051 >    group1 = cgRowToGlobal[cg1];
1052 >    group2 = cgColToGlobal[cg2];
1053 > #else
1054 >    unique_id_1 = AtomLocalToGlobal[atom1];
1055 >    unique_id_2 = AtomLocalToGlobal[atom2];
1056 >    group1 = cgLocalToGlobal[cg1];
1057 >    group2 = cgLocalToGlobal[cg2];
1058 > #endif  
1059 >
1060      if (unique_id_1 == unique_id_2) return true;
1061 <    
1061 >
1062 > #ifdef IS_MPI
1063      // this prevents us from doing the pair on multiple processors
1064      if (unique_id_1 < unique_id_2) {
1065        if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1066      } else {
1067 <      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1067 >      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1068      }
1069 + #endif    
1070 +
1071 + #ifndef IS_MPI
1072 +    if (group1 == group2) {
1073 +      if (unique_id_1 < unique_id_2) return true;
1074 +    }
1075   #endif
1076 +    
1077      return false;
1078    }
1079  
# Line 788 | Line 1087 | namespace OpenMD {
1087     * field) must still be handled for these pairs.
1088     */
1089    bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1090 <    int unique_id_2;
1090 >
1091 >    // excludesForAtom was constructed to use row/column indices in the MPI
1092 >    // version, and to use local IDs in the non-MPI version:
1093      
793 – #ifdef IS_MPI
794 –    // in MPI, we have to look up the unique IDs for the row atom.
795 –    unique_id_2 = AtomColToGlobal[atom2];
796 – #else
797 –    // in the normal loop, the atom numbers are unique
798 –    unique_id_2 = atom2;
799 – #endif
800 –    
1094      for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1095           i != excludesForAtom[atom1].end(); ++i) {
1096 <      if ( (*i) == unique_id_2 ) return true;
1096 >      if ( (*i) == atom2 ) return true;
1097      }
1098  
1099      return false;
# Line 830 | Line 1123 | namespace OpenMD {
1123      idat.excluded = excludeAtomPair(atom1, atom2);
1124    
1125   #ifdef IS_MPI
1126 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1127 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1128 +    //                         ff_->getAtomType(identsCol[atom2]) );
1129      
834 –    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
835 –                             ff_->getAtomType(identsCol[atom2]) );
836 –    
1130      if (storageLayout_ & DataStorage::dslAmat) {
1131        idat.A1 = &(atomRowData.aMat[atom1]);
1132        idat.A2 = &(atomColData.aMat[atom2]);
# Line 874 | Line 1167 | namespace OpenMD {
1167        idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1168      }
1169  
1170 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1171 +      idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1172 +      idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1173 +    }
1174 +
1175   #else
1176 +    
1177 +    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1178  
879 –    idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
880 –                             ff_->getAtomType(idents[atom2]) );
881 –
1179      if (storageLayout_ & DataStorage::dslAmat) {
1180        idat.A1 = &(snap_->atomData.aMat[atom1]);
1181        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 918 | Line 1215 | namespace OpenMD {
1215        idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1216        idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1217      }
1218 +
1219 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1220 +      idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1221 +      idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1222 +    }
1223 +
1224   #endif
1225    }
1226  
1227    
1228    void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1229   #ifdef IS_MPI
1230 <    pot_row[atom1] += 0.5 *  *(idat.pot);
1231 <    pot_col[atom2] += 0.5 *  *(idat.pot);
1230 >    pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1231 >    pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1232 >    expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1233 >    expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1234  
1235      atomRowData.force[atom1] += *(idat.f1);
1236      atomColData.force[atom2] -= *(idat.f1);
1237 +
1238 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1239 +      atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1240 +      atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1241 +    }
1242 +
1243 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1244 +      atomRowData.electricField[atom1] += *(idat.eField1);
1245 +      atomColData.electricField[atom2] += *(idat.eField2);
1246 +    }
1247 +
1248   #else
1249      pairwisePot += *(idat.pot);
1250 +    excludedPot += *(idat.excludedPot);
1251  
1252      snap_->atomData.force[atom1] += *(idat.f1);
1253      snap_->atomData.force[atom2] -= *(idat.f1);
1254 +
1255 +    if (idat.doParticlePot) {
1256 +      // This is the pairwise contribution to the particle pot.  The
1257 +      // embedding contribution is added in each of the low level
1258 +      // non-bonded routines.  In parallel, this calculation is done
1259 +      // in collectData, not in unpackInteractionData.
1260 +      snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1261 +      snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1262 +    }
1263 +    
1264 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1265 +      snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1266 +      snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1267 +    }
1268 +
1269 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1270 +      snap_->atomData.electricField[atom1] += *(idat.eField1);
1271 +      snap_->atomData.electricField[atom2] += *(idat.eField2);
1272 +    }
1273 +
1274   #endif
1275      
1276    }
# Line 1015 | Line 1352 | namespace OpenMD {
1352          // add this cutoff group to the list of groups in this cell;
1353          cellListRow_[cellIndex].push_back(i);
1354        }
1018 –      
1355        for (int i = 0; i < nGroupsInCol_; i++) {
1356          rs = cgColData.position[i];
1357          
# Line 1040 | Line 1376 | namespace OpenMD {
1376          // add this cutoff group to the list of groups in this cell;
1377          cellListCol_[cellIndex].push_back(i);
1378        }
1379 +    
1380   #else
1381        for (int i = 0; i < nGroups_; i++) {
1382          rs = snap_->cgData.position[i];
# Line 1052 | Line 1389 | namespace OpenMD {
1389          for (int j = 0; j < 3; j++) {
1390            scaled[j] -= roundMe(scaled[j]);
1391            scaled[j] += 0.5;
1392 +          // Handle the special case when an object is exactly on the
1393 +          // boundary (a scaled coordinate of 1.0 is the same as
1394 +          // scaled coordinate of 0.0)
1395 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1396          }
1397          
1398          // find xyz-indices of cell that cutoffGroup is in.
# Line 1060 | Line 1401 | namespace OpenMD {
1401          whichCell.z() = nCells_.z() * scaled.z();
1402          
1403          // find single index of this cell:
1404 <        cellIndex = Vlinear(whichCell, nCells_);      
1404 >        cellIndex = Vlinear(whichCell, nCells_);
1405          
1406          // add this cutoff group to the list of groups in this cell;
1407          cellList_[cellIndex].push_back(i);
1408        }
1409 +
1410   #endif
1411  
1412        for (int m1z = 0; m1z < nCells_.z(); m1z++) {
# Line 1077 | Line 1419 | namespace OpenMD {
1419                   os != cellOffsets_.end(); ++os) {
1420                
1421                Vector3i m2v = m1v + (*os);
1422 <              
1422 >            
1423 >
1424                if (m2v.x() >= nCells_.x()) {
1425                  m2v.x() = 0;          
1426                } else if (m2v.x() < 0) {
# Line 1095 | Line 1438 | namespace OpenMD {
1438                } else if (m2v.z() < 0) {
1439                  m2v.z() = nCells_.z() - 1;
1440                }
1441 <              
1441 >
1442                int m2 = Vlinear (m2v, nCells_);
1443                
1444   #ifdef IS_MPI
# Line 1104 | Line 1447 | namespace OpenMD {
1447                  for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1448                       j2 != cellListCol_[m2].end(); ++j2) {
1449                    
1450 <                  // Always do this if we're in different cells or if
1451 <                  // we're in the same cell and the global index of the
1452 <                  // j2 cutoff group is less than the j1 cutoff group
1453 <                  
1454 <                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1455 <                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1456 <                    snap_->wrapVector(dr);
1457 <                    cuts = getGroupCutoffs( (*j1), (*j2) );
1458 <                    if (dr.lengthSquare() < cuts.third) {
1116 <                      neighborList.push_back(make_pair((*j1), (*j2)));
1117 <                    }
1118 <                  }
1450 >                  // In parallel, we need to visit *all* pairs of row
1451 >                  // & column indicies and will divide labor in the
1452 >                  // force evaluation later.
1453 >                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1454 >                  snap_->wrapVector(dr);
1455 >                  cuts = getGroupCutoffs( (*j1), (*j2) );
1456 >                  if (dr.lengthSquare() < cuts.third) {
1457 >                    neighborList.push_back(make_pair((*j1), (*j2)));
1458 >                  }                  
1459                  }
1460                }
1461   #else
1122 –              
1462                for (vector<int>::iterator j1 = cellList_[m1].begin();
1463                     j1 != cellList_[m1].end(); ++j1) {
1464                  for (vector<int>::iterator j2 = cellList_[m2].begin();
1465                       j2 != cellList_[m2].end(); ++j2) {
1466 <                  
1466 >    
1467                    // Always do this if we're in different cells or if
1468 <                  // we're in the same cell and the global index of the
1469 <                  // j2 cutoff group is less than the j1 cutoff group
1470 <                  
1471 <                  if (m2 != m1 || (*j2) < (*j1)) {
1468 >                  // we're in the same cell and the global index of
1469 >                  // the j2 cutoff group is greater than or equal to
1470 >                  // the j1 cutoff group.  Note that Rappaport's code
1471 >                  // has a "less than" conditional here, but that
1472 >                  // deals with atom-by-atom computation.  OpenMD
1473 >                  // allows atoms within a single cutoff group to
1474 >                  // interact with each other.
1475 >
1476 >
1477 >
1478 >                  if (m2 != m1 || (*j2) >= (*j1) ) {
1479 >
1480                      dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1481                      snap_->wrapVector(dr);
1482                      cuts = getGroupCutoffs( (*j1), (*j2) );
# Line 1148 | Line 1495 | namespace OpenMD {
1495        // branch to do all cutoff group pairs
1496   #ifdef IS_MPI
1497        for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1498 <        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1498 >        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
1499            dr = cgColData.position[j2] - cgRowData.position[j1];
1500            snap_->wrapVector(dr);
1501            cuts = getGroupCutoffs( j1, j2 );
# Line 1156 | Line 1503 | namespace OpenMD {
1503              neighborList.push_back(make_pair(j1, j2));
1504            }
1505          }
1506 <      }
1506 >      }      
1507   #else
1508 <      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1509 <        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1508 >      // include all groups here.
1509 >      for (int j1 = 0; j1 < nGroups_; j1++) {
1510 >        // include self group interactions j2 == j1
1511 >        for (int j2 = j1; j2 < nGroups_; j2++) {
1512            dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1513            snap_->wrapVector(dr);
1514            cuts = getGroupCutoffs( j1, j2 );
1515            if (dr.lengthSquare() < cuts.third) {
1516              neighborList.push_back(make_pair(j1, j2));
1517            }
1518 <        }
1519 <      }        
1518 >        }    
1519 >      }
1520   #endif
1521      }
1522        

Diff Legend

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