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

Comparing branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents):
Revision 1591 by gezelter, Tue Jul 12 15:25:07 2011 UTC vs.
Revision 1821 by gezelter, Mon Jan 7 20:05:43 2013 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_);
# Line 124 | Line 176 | namespace OpenMD {
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 <    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185 <    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186 <    
184 >    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185 >    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186 >
187      cgRowToGlobal.resize(nGroupsInRow_);
188      cgColToGlobal.resize(nGroupsInCol_);
189 <    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190 <    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
189 >    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190 >    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
191  
192      massFactorsRow.resize(nAtomsInRow_);
193      massFactorsCol.resize(nAtomsInCol_);
194 <    AtomCommRealRow->gather(massFactors, massFactorsRow);
195 <    AtomCommRealColumn->gather(massFactors, massFactorsCol);
194 >    AtomPlanRealRow->gather(massFactors, massFactorsRow);
195 >    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
196  
197      groupListRow_.clear();
198      groupListRow_.resize(nGroupsInRow_);
# Line 193 | Line 248 | namespace OpenMD {
248        }      
249      }
250  
251 < #endif
197 <
198 <    // allocate memory for the parallel objects
199 <    atypesLocal.resize(nLocal_);
200 <
201 <    for (int i = 0; i < nLocal_; i++)
202 <      atypesLocal[i] = ff_->getAtomType(idents[i]);
203 <
204 <    groupList_.clear();
205 <    groupList_.resize(nGroups_);
206 <    for (int i = 0; i < nGroups_; i++) {
207 <      int gid = cgLocalToGlobal[i];
208 <      for (int j = 0; j < nLocal_; j++) {
209 <        int aid = AtomLocalToGlobal[j];
210 <        if (globalGroupMembership[aid] == gid) {
211 <          groupList_[i].push_back(j);
212 <        }
213 <      }      
214 <    }
215 <
251 > #else
252      excludesForAtom.clear();
253      excludesForAtom.resize(nLocal_);
254      toposForAtom.clear();
# Line 245 | 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 253 | 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 263 | 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 326 | Line 383 | namespace OpenMD {
383      vector<RealType> groupCutoff(nGroups_, 0.0);
384      groupToGtype.resize(nGroups_);
385      for (int cg1 = 0; cg1 < nGroups_; cg1++) {
329 –
386        groupCutoff[cg1] = 0.0;
387        vector<int> atomList = getAtomsInGroupRow(cg1);
332 –
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];
339 –        }
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 365 | Line 419 | namespace OpenMD {
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 390 | Line 444 | namespace OpenMD {
444  
445          pair<int,int> key = make_pair(i,j);
446          gTypeCutoffMap[key].first = thisRcut;
393 –
447          if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
395 –
448          gTypeCutoffMap[key].second = thisRcut*thisRcut;
397 –        
449          gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
399 –
450          // sanity check
451          
452          if (userChoseCutoff_) {
# Line 413 | Line 463 | namespace OpenMD {
463      }
464    }
465  
416 –
466    groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
467      int i, j;  
468   #ifdef IS_MPI
# Line 427 | 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 437 | 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 455 | 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(),
517             0.0);
# Line 488 | Line 545 | namespace OpenMD {
545             atomColData.skippedCharge.end(), 0.0);
546      }
547  
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   #endif
563      // even in parallel, we need to zero out the local arrays:
564  
# Line 500 | Line 571 | namespace OpenMD {
571        fill(snap_->atomData.density.begin(),
572             snap_->atomData.density.end(), 0.0);
573      }
574 +
575      if (storageLayout_ & DataStorage::dslFunctional) {
576        fill(snap_->atomData.functional.begin(),
577             snap_->atomData.functional.end(), 0.0);
578      }
579 +
580      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
581        fill(snap_->atomData.functionalDerivative.begin(),
582             snap_->atomData.functionalDerivative.end(), 0.0);
583      }
584 +
585      if (storageLayout_ & DataStorage::dslSkippedCharge) {      
586        fill(snap_->atomData.skippedCharge.begin(),
587             snap_->atomData.skippedCharge.end(), 0.0);
588      }
589 <    
589 >
590 >    if (storageLayout_ & DataStorage::dslElectricField) {      
591 >      fill(snap_->atomData.electricField.begin(),
592 >           snap_->atomData.electricField.end(), V3Zero);
593 >    }
594    }
595  
596  
# Line 522 | Line 600 | namespace OpenMD {
600   #ifdef IS_MPI
601      
602      // gather up the atomic positions
603 <    AtomCommVectorRow->gather(snap_->atomData.position,
603 >    AtomPlanVectorRow->gather(snap_->atomData.position,
604                                atomRowData.position);
605 <    AtomCommVectorColumn->gather(snap_->atomData.position,
605 >    AtomPlanVectorColumn->gather(snap_->atomData.position,
606                                   atomColData.position);
607      
608      // gather up the cutoff group positions
609 <    cgCommVectorRow->gather(snap_->cgData.position,
609 >
610 >    cgPlanVectorRow->gather(snap_->cgData.position,
611                              cgRowData.position);
612 <    cgCommVectorColumn->gather(snap_->cgData.position,
612 >
613 >    cgPlanVectorColumn->gather(snap_->cgData.position,
614                                 cgColData.position);
615 +
616 +
617 +
618 +    if (needVelocities_) {
619 +      // gather up the atomic velocities
620 +      AtomPlanVectorColumn->gather(snap_->atomData.velocity,
621 +                                   atomColData.velocity);
622 +      
623 +      cgPlanVectorColumn->gather(snap_->cgData.velocity,
624 +                                 cgColData.velocity);
625 +    }
626 +
627      
628      // if needed, gather the atomic rotation matrices
629      if (storageLayout_ & DataStorage::dslAmat) {
630 <      AtomCommMatrixRow->gather(snap_->atomData.aMat,
630 >      AtomPlanMatrixRow->gather(snap_->atomData.aMat,
631                                  atomRowData.aMat);
632 <      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
632 >      AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
633                                     atomColData.aMat);
634      }
635 <    
636 <    // if needed, gather the atomic eletrostatic frames
637 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
638 <      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
639 <                                atomRowData.electroFrame);
640 <      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
641 <                                   atomColData.electroFrame);
635 >
636 >    // if needed, gather the atomic eletrostatic information
637 >    if (storageLayout_ & DataStorage::dslDipole) {
638 >      AtomPlanVectorRow->gather(snap_->atomData.dipole,
639 >                                atomRowData.dipole);
640 >      AtomPlanVectorColumn->gather(snap_->atomData.dipole,
641 >                                   atomColData.dipole);
642 >    }
643 >
644 >    if (storageLayout_ & DataStorage::dslQuadrupole) {
645 >      AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
646 >                                atomRowData.quadrupole);
647 >      AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
648 >                                   atomColData.quadrupole);
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    }
# Line 562 | 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,
690 +                                    field_tmp);
691 +      for (int i = 0; i < n; i++)
692 +        snap_->atomData.electricField[i] += field_tmp[i];
693 +    }
694   #endif
695    }
696  
# Line 583 | Line 703 | namespace OpenMD {
703      storageLayout_ = sman_->getStorageLayout();
704   #ifdef IS_MPI
705      if (storageLayout_ & DataStorage::dslFunctional) {
706 <      AtomCommRealRow->gather(snap_->atomData.functional,
706 >      AtomPlanRealRow->gather(snap_->atomData.functional,
707                                atomRowData.functional);
708 <      AtomCommRealColumn->gather(snap_->atomData.functional,
708 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
709                                   atomColData.functional);
710      }
711      
712      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
713 <      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
713 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
714                                atomRowData.functionalDerivative);
715 <      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
715 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
716                                   atomColData.functionalDerivative);
717      }
718   #endif
# Line 606 | Line 726 | namespace OpenMD {
726      int n = snap_->atomData.force.size();
727      vector<Vector3d> frc_tmp(n, V3Zero);
728      
729 <    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
729 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
730      for (int i = 0; i < n; i++) {
731        snap_->atomData.force[i] += frc_tmp[i];
732        frc_tmp[i] = 0.0;
733      }
734      
735 <    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
736 <    for (int i = 0; i < n; i++)
735 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
736 >    for (int i = 0; i < n; i++) {
737        snap_->atomData.force[i] += frc_tmp[i];
738 +    }
739          
740      if (storageLayout_ & DataStorage::dslTorque) {
741  
742        int nt = snap_->atomData.torque.size();
743        vector<Vector3d> trq_tmp(nt, V3Zero);
744  
745 <      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
745 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
746        for (int i = 0; i < nt; i++) {
747          snap_->atomData.torque[i] += trq_tmp[i];
748          trq_tmp[i] = 0.0;
749        }
750        
751 <      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
751 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
752        for (int i = 0; i < nt; i++)
753          snap_->atomData.torque[i] += trq_tmp[i];
754      }
# Line 637 | Line 758 | namespace OpenMD {
758        int ns = snap_->atomData.skippedCharge.size();
759        vector<RealType> skch_tmp(ns, 0.0);
760  
761 <      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
761 >      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
762        for (int i = 0; i < ns; i++) {
763          snap_->atomData.skippedCharge[i] += skch_tmp[i];
764          skch_tmp[i] = 0.0;
765        }
766        
767 <      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
768 <      for (int i = 0; i < ns; i++)
767 >      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
768 >      for (int i = 0; i < ns; i++)
769          snap_->atomData.skippedCharge[i] += skch_tmp[i];
770 +            
771      }
772      
773 +    if (storageLayout_ & DataStorage::dslFlucQForce) {
774 +
775 +      int nq = snap_->atomData.flucQFrc.size();
776 +      vector<RealType> fqfrc_tmp(nq, 0.0);
777 +
778 +      AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
779 +      for (int i = 0; i < nq; i++) {
780 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
781 +        fqfrc_tmp[i] = 0.0;
782 +      }
783 +      
784 +      AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
785 +      for (int i = 0; i < nq; i++)
786 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
787 +            
788 +    }
789 +
790      nLocal_ = snap_->getNumberOfAtoms();
791  
792      vector<potVec> pot_temp(nLocal_,
793                              Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
794 +    vector<potVec> expot_temp(nLocal_,
795 +                              Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
796  
797      // scatter/gather pot_row into the members of my column
798            
799 <    AtomCommPotRow->scatter(pot_row, pot_temp);
799 >    AtomPlanPotRow->scatter(pot_row, pot_temp);
800 >    AtomPlanPotRow->scatter(expot_row, expot_temp);
801  
802 <    for (int ii = 0;  ii < pot_temp.size(); ii++ )
802 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
803        pairwisePot += pot_temp[ii];
804 <    
804 >
805 >    for (int ii = 0;  ii < expot_temp.size(); ii++ )
806 >      excludedPot += expot_temp[ii];
807 >        
808 >    if (storageLayout_ & DataStorage::dslParticlePot) {
809 >      // This is the pairwise contribution to the particle pot.  The
810 >      // embedding contribution is added in each of the low level
811 >      // non-bonded routines.  In single processor, this is done in
812 >      // unpackInteractionData, not in collectData.
813 >      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
814 >        for (int i = 0; i < nLocal_; i++) {
815 >          // factor of two is because the total potential terms are divided
816 >          // by 2 in parallel due to row/ column scatter      
817 >          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
818 >        }
819 >      }
820 >    }
821 >
822      fill(pot_temp.begin(), pot_temp.end(),
823           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
824 +    fill(expot_temp.begin(), expot_temp.end(),
825 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
826        
827 <    AtomCommPotColumn->scatter(pot_col, pot_temp);    
827 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
828 >    AtomPlanPotColumn->scatter(expot_col, expot_temp);    
829      
830      for (int ii = 0;  ii < pot_temp.size(); ii++ )
831        pairwisePot += pot_temp[ii];    
832 +
833 +    for (int ii = 0;  ii < expot_temp.size(); ii++ )
834 +      excludedPot += expot_temp[ii];    
835 +
836 +    if (storageLayout_ & DataStorage::dslParticlePot) {
837 +      // This is the pairwise contribution to the particle pot.  The
838 +      // embedding contribution is added in each of the low level
839 +      // non-bonded routines.  In single processor, this is done in
840 +      // unpackInteractionData, not in collectData.
841 +      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
842 +        for (int i = 0; i < nLocal_; i++) {
843 +          // factor of two is because the total potential terms are divided
844 +          // by 2 in parallel due to row/ column scatter      
845 +          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
846 +        }
847 +      }
848 +    }
849 +    
850 +    if (storageLayout_ & DataStorage::dslParticlePot) {
851 +      int npp = snap_->atomData.particlePot.size();
852 +      vector<RealType> ppot_temp(npp, 0.0);
853 +
854 +      // This is the direct or embedding contribution to the particle
855 +      // pot.
856 +      
857 +      AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
858 +      for (int i = 0; i < npp; i++) {
859 +        snap_->atomData.particlePot[i] += ppot_temp[i];
860 +      }
861 +
862 +      fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
863 +      
864 +      AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
865 +      for (int i = 0; i < npp; i++) {
866 +        snap_->atomData.particlePot[i] += ppot_temp[i];
867 +      }
868 +    }
869 +
870 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
871 +      RealType ploc1 = pairwisePot[ii];
872 +      RealType ploc2 = 0.0;
873 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
874 +      pairwisePot[ii] = ploc2;
875 +    }
876 +
877 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
878 +      RealType ploc1 = excludedPot[ii];
879 +      RealType ploc2 = 0.0;
880 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
881 +      excludedPot[ii] = ploc2;
882 +    }
883 +
884 +    // Here be dragons.
885 +    MPI::Intracomm col = colComm.getComm();
886 +
887 +    col.Allreduce(MPI::IN_PLACE,
888 +                  &snap_->frameData.conductiveHeatFlux[0], 3,
889 +                  MPI::REALTYPE, MPI::SUM);
890 +
891 +
892   #endif
893  
894    }
895  
896 +  /**
897 +   * Collects information obtained during the post-pair (and embedding
898 +   * functional) loops onto local data structures.
899 +   */
900 +  void ForceMatrixDecomposition::collectSelfData() {
901 +    snap_ = sman_->getCurrentSnapshot();
902 +    storageLayout_ = sman_->getStorageLayout();
903 +
904 + #ifdef IS_MPI
905 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
906 +      RealType ploc1 = embeddingPot[ii];
907 +      RealType ploc2 = 0.0;
908 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
909 +      embeddingPot[ii] = ploc2;
910 +    }    
911 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
912 +      RealType ploc1 = excludedSelfPot[ii];
913 +      RealType ploc2 = 0.0;
914 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
915 +      excludedSelfPot[ii] = ploc2;
916 +    }    
917 + #endif
918 +    
919 +  }
920 +
921 +
922 +
923    int ForceMatrixDecomposition::getNAtomsInRow() {  
924   #ifdef IS_MPI
925      return nAtomsInRow_;
# Line 711 | Line 960 | namespace OpenMD {
960      return d;    
961    }
962  
963 +  Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
964 + #ifdef IS_MPI
965 +    return cgColData.velocity[cg2];
966 + #else
967 +    return snap_->cgData.velocity[cg2];
968 + #endif
969 +  }
970  
971 +  Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
972 + #ifdef IS_MPI
973 +    return atomColData.velocity[atom2];
974 + #else
975 +    return snap_->atomData.velocity[atom2];
976 + #endif
977 +  }
978 +
979 +
980    Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
981  
982      Vector3d d;
# Line 777 | Line 1042 | namespace OpenMD {
1042     * We need to exclude some overcounted interactions that result from
1043     * the parallel decomposition.
1044     */
1045 <  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1045 >  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1046      int unique_id_1, unique_id_2;
1047 <
1047 >        
1048   #ifdef IS_MPI
1049      // in MPI, we have to look up the unique IDs for each atom
1050      unique_id_1 = AtomRowToGlobal[atom1];
1051      unique_id_2 = AtomColToGlobal[atom2];
1052 +    // group1 = cgRowToGlobal[cg1];
1053 +    // group2 = cgColToGlobal[cg2];
1054 + #else
1055 +    unique_id_1 = AtomLocalToGlobal[atom1];
1056 +    unique_id_2 = AtomLocalToGlobal[atom2];
1057 +    int group1 = cgLocalToGlobal[cg1];
1058 +    int group2 = cgLocalToGlobal[cg2];
1059 + #endif  
1060  
788 –    // this situation should only arise in MPI simulations
1061      if (unique_id_1 == unique_id_2) return true;
1062 <    
1062 >
1063 > #ifdef IS_MPI
1064      // this prevents us from doing the pair on multiple processors
1065      if (unique_id_1 < unique_id_2) {
1066        if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1067      } else {
1068 <      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1068 >      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1069      }
1070 + #endif    
1071 +
1072 + #ifndef IS_MPI
1073 +    if (group1 == group2) {
1074 +      if (unique_id_1 < unique_id_2) return true;
1075 +    }
1076   #endif
1077 +    
1078      return false;
1079    }
1080  
# Line 808 | Line 1088 | namespace OpenMD {
1088     * field) must still be handled for these pairs.
1089     */
1090    bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1091 <    int unique_id_2;
1092 <    
1093 < #ifdef IS_MPI
814 <    // in MPI, we have to look up the unique IDs for the row atom.
815 <    unique_id_2 = AtomColToGlobal[atom2];
816 < #else
817 <    // in the normal loop, the atom numbers are unique
818 <    unique_id_2 = atom2;
819 < #endif
1091 >
1092 >    // excludesForAtom was constructed to use row/column indices in the MPI
1093 >    // version, and to use local IDs in the non-MPI version:
1094      
1095      for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1096           i != excludesForAtom[atom1].end(); ++i) {
1097 <      if ( (*i) == unique_id_2 ) return true;
1097 >      if ( (*i) == atom2 ) return true;
1098      }
1099  
1100      return false;
# Line 859 | Line 1133 | namespace OpenMD {
1133        idat.A2 = &(atomColData.aMat[atom2]);
1134      }
1135      
862 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
863 –      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
864 –      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
865 –    }
866 –
1136      if (storageLayout_ & DataStorage::dslTorque) {
1137        idat.t1 = &(atomRowData.torque[atom1]);
1138        idat.t2 = &(atomColData.torque[atom2]);
1139      }
1140  
1141 +    if (storageLayout_ & DataStorage::dslDipole) {
1142 +      idat.dipole1 = &(atomRowData.dipole[atom1]);
1143 +      idat.dipole2 = &(atomColData.dipole[atom2]);
1144 +    }
1145 +
1146 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1147 +      idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1148 +      idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1149 +    }
1150 +
1151      if (storageLayout_ & DataStorage::dslDensity) {
1152        idat.rho1 = &(atomRowData.density[atom1]);
1153        idat.rho2 = &(atomColData.density[atom2]);
# Line 894 | Line 1173 | namespace OpenMD {
1173        idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1174      }
1175  
1176 < #else
1176 >    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1177 >      idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1178 >      idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1179 >    }
1180  
1181 + #else
1182 +    
1183      idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
900 –    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
901 –    //                         ff_->getAtomType(idents[atom2]) );
1184  
1185      if (storageLayout_ & DataStorage::dslAmat) {
1186        idat.A1 = &(snap_->atomData.aMat[atom1]);
1187        idat.A2 = &(snap_->atomData.aMat[atom2]);
1188      }
1189  
1190 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
909 <      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
910 <      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
911 <    }
1190 >    RealType ct = dot(idat.A1->getColumn(2), idat.A2->getColumn(2));
1191  
1192      if (storageLayout_ & DataStorage::dslTorque) {
1193        idat.t1 = &(snap_->atomData.torque[atom1]);
1194        idat.t2 = &(snap_->atomData.torque[atom2]);
1195      }
1196  
1197 +    if (storageLayout_ & DataStorage::dslDipole) {
1198 +      idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1199 +      idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1200 +    }
1201 +
1202 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1203 +      idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1204 +      idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1205 +    }
1206 +
1207      if (storageLayout_ & DataStorage::dslDensity) {    
1208        idat.rho1 = &(snap_->atomData.density[atom1]);
1209        idat.rho2 = &(snap_->atomData.density[atom2]);
# Line 939 | Line 1228 | namespace OpenMD {
1228        idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1229        idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1230      }
1231 +
1232 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1233 +      idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1234 +      idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1235 +    }
1236 +
1237   #endif
1238    }
1239  
1240    
1241    void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1242   #ifdef IS_MPI
1243 <    pot_row[atom1] += 0.5 *  *(idat.pot);
1244 <    pot_col[atom2] += 0.5 *  *(idat.pot);
1243 >    pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1244 >    pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1245 >    expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1246 >    expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1247  
1248      atomRowData.force[atom1] += *(idat.f1);
1249      atomColData.force[atom2] -= *(idat.f1);
1250 +
1251 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1252 +      atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1253 +      atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1254 +    }
1255 +
1256 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1257 +      atomRowData.electricField[atom1] += *(idat.eField1);
1258 +      atomColData.electricField[atom2] += *(idat.eField2);
1259 +    }
1260 +
1261   #else
1262      pairwisePot += *(idat.pot);
1263 +    excludedPot += *(idat.excludedPot);
1264  
1265      snap_->atomData.force[atom1] += *(idat.f1);
1266      snap_->atomData.force[atom2] -= *(idat.f1);
1267 +
1268 +    if (idat.doParticlePot) {
1269 +      // This is the pairwise contribution to the particle pot.  The
1270 +      // embedding contribution is added in each of the low level
1271 +      // non-bonded routines.  In parallel, this calculation is done
1272 +      // in collectData, not in unpackInteractionData.
1273 +      snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1274 +      snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1275 +    }
1276 +    
1277 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1278 +      snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1279 +      snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1280 +    }
1281 +
1282 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1283 +      snap_->atomData.electricField[atom1] += *(idat.eField1);
1284 +      snap_->atomData.electricField[atom2] += *(idat.eField2);
1285 +    }
1286 +
1287   #endif
1288      
1289    }
# Line 1023 | Line 1352 | namespace OpenMD {
1352          for (int j = 0; j < 3; j++) {
1353            scaled[j] -= roundMe(scaled[j]);
1354            scaled[j] += 0.5;
1355 +          // Handle the special case when an object is exactly on the
1356 +          // boundary (a scaled coordinate of 1.0 is the same as
1357 +          // scaled coordinate of 0.0)
1358 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1359          }
1360          
1361          // find xyz-indices of cell that cutoffGroup is in.
# Line 1036 | Line 1369 | namespace OpenMD {
1369          // add this cutoff group to the list of groups in this cell;
1370          cellListRow_[cellIndex].push_back(i);
1371        }
1039 –      
1372        for (int i = 0; i < nGroupsInCol_; i++) {
1373          rs = cgColData.position[i];
1374          
# Line 1048 | Line 1380 | namespace OpenMD {
1380          for (int j = 0; j < 3; j++) {
1381            scaled[j] -= roundMe(scaled[j]);
1382            scaled[j] += 0.5;
1383 +          // Handle the special case when an object is exactly on the
1384 +          // boundary (a scaled coordinate of 1.0 is the same as
1385 +          // scaled coordinate of 0.0)
1386 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1387          }
1388          
1389          // find xyz-indices of cell that cutoffGroup is in.
# Line 1061 | Line 1397 | namespace OpenMD {
1397          // add this cutoff group to the list of groups in this cell;
1398          cellListCol_[cellIndex].push_back(i);
1399        }
1400 +    
1401   #else
1402        for (int i = 0; i < nGroups_; i++) {
1403          rs = snap_->cgData.position[i];
# Line 1073 | Line 1410 | namespace OpenMD {
1410          for (int j = 0; j < 3; j++) {
1411            scaled[j] -= roundMe(scaled[j]);
1412            scaled[j] += 0.5;
1413 +          // Handle the special case when an object is exactly on the
1414 +          // boundary (a scaled coordinate of 1.0 is the same as
1415 +          // scaled coordinate of 0.0)
1416 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1417          }
1418          
1419          // find xyz-indices of cell that cutoffGroup is in.
# Line 1081 | Line 1422 | namespace OpenMD {
1422          whichCell.z() = nCells_.z() * scaled.z();
1423          
1424          // find single index of this cell:
1425 <        cellIndex = Vlinear(whichCell, nCells_);      
1425 >        cellIndex = Vlinear(whichCell, nCells_);
1426          
1427          // add this cutoff group to the list of groups in this cell;
1428          cellList_[cellIndex].push_back(i);
1429        }
1430 +
1431   #endif
1432  
1433        for (int m1z = 0; m1z < nCells_.z(); m1z++) {
# Line 1098 | Line 1440 | namespace OpenMD {
1440                   os != cellOffsets_.end(); ++os) {
1441                
1442                Vector3i m2v = m1v + (*os);
1443 <              
1443 >            
1444 >
1445                if (m2v.x() >= nCells_.x()) {
1446                  m2v.x() = 0;          
1447                } else if (m2v.x() < 0) {
# Line 1116 | Line 1459 | namespace OpenMD {
1459                } else if (m2v.z() < 0) {
1460                  m2v.z() = nCells_.z() - 1;
1461                }
1462 <              
1462 >
1463                int m2 = Vlinear (m2v, nCells_);
1464                
1465   #ifdef IS_MPI
# Line 1125 | Line 1468 | namespace OpenMD {
1468                  for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1469                       j2 != cellListCol_[m2].end(); ++j2) {
1470                    
1471 <                  // Always do this if we're in different cells or if
1472 <                  // we're in the same cell and the global index of the
1473 <                  // j2 cutoff group is less than the j1 cutoff group
1474 <                  
1475 <                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1476 <                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1477 <                    snap_->wrapVector(dr);
1478 <                    cuts = getGroupCutoffs( (*j1), (*j2) );
1479 <                    if (dr.lengthSquare() < cuts.third) {
1137 <                      neighborList.push_back(make_pair((*j1), (*j2)));
1138 <                    }
1139 <                  }
1471 >                  // In parallel, we need to visit *all* pairs of row
1472 >                  // & column indicies and will divide labor in the
1473 >                  // force evaluation later.
1474 >                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1475 >                  snap_->wrapVector(dr);
1476 >                  cuts = getGroupCutoffs( (*j1), (*j2) );
1477 >                  if (dr.lengthSquare() < cuts.third) {
1478 >                    neighborList.push_back(make_pair((*j1), (*j2)));
1479 >                  }                  
1480                  }
1481                }
1482   #else
1143 –              
1483                for (vector<int>::iterator j1 = cellList_[m1].begin();
1484                     j1 != cellList_[m1].end(); ++j1) {
1485                  for (vector<int>::iterator j2 = cellList_[m2].begin();
1486                       j2 != cellList_[m2].end(); ++j2) {
1487 <                  
1487 >    
1488                    // Always do this if we're in different cells or if
1489 <                  // we're in the same cell and the global index of the
1490 <                  // j2 cutoff group is less than the j1 cutoff group
1491 <                  
1492 <                  if (m2 != m1 || (*j2) < (*j1)) {
1489 >                  // we're in the same cell and the global index of
1490 >                  // the j2 cutoff group is greater than or equal to
1491 >                  // the j1 cutoff group.  Note that Rappaport's code
1492 >                  // has a "less than" conditional here, but that
1493 >                  // deals with atom-by-atom computation.  OpenMD
1494 >                  // allows atoms within a single cutoff group to
1495 >                  // interact with each other.
1496 >
1497 >
1498 >
1499 >                  if (m2 != m1 || (*j2) >= (*j1) ) {
1500 >
1501                      dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1502                      snap_->wrapVector(dr);
1503                      cuts = getGroupCutoffs( (*j1), (*j2) );
# Line 1169 | Line 1516 | namespace OpenMD {
1516        // branch to do all cutoff group pairs
1517   #ifdef IS_MPI
1518        for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1519 <        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1519 >        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
1520            dr = cgColData.position[j2] - cgRowData.position[j1];
1521            snap_->wrapVector(dr);
1522            cuts = getGroupCutoffs( j1, j2 );
# Line 1177 | Line 1524 | namespace OpenMD {
1524              neighborList.push_back(make_pair(j1, j2));
1525            }
1526          }
1527 <      }
1527 >      }      
1528   #else
1529 <      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1530 <        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1529 >      // include all groups here.
1530 >      for (int j1 = 0; j1 < nGroups_; j1++) {
1531 >        // include self group interactions j2 == j1
1532 >        for (int j2 = j1; j2 < nGroups_; j2++) {
1533            dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1534            snap_->wrapVector(dr);
1535            cuts = getGroupCutoffs( j1, j2 );
1536            if (dr.lengthSquare() < cuts.third) {
1537              neighborList.push_back(make_pair(j1, j2));
1538            }
1539 <        }
1540 <      }        
1539 >        }    
1540 >      }
1541   #endif
1542      }
1543        

Diff Legend

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