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 1592 by gezelter, Tue Jul 12 20:33:14 2011 UTC vs.
Revision 1855 by gezelter, Tue Apr 2 18:31:51 2013 UTC

# Line 35 | Line 35
35   *                                                                      
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).                        
38 > * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008).          
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 189 | Line 244 | namespace OpenMD {
244                topoDist[i].push_back(3);
245              }
246            }
192 –        }
193 –      }      
194 –    }
195 –
196 – #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);
247          }
248        }      
249      }
250  
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 254 | Line 310 | namespace OpenMD {
310      
311      RealType tol = 1e-6;
312      largestRcut_ = 0.0;
257 –    RealType rc;
313      int atid;
314      set<AtomType*> atypes = info_->getSimulatedAtomTypes();
315      
# Line 339 | Line 394 | namespace OpenMD {
394        }
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;
# Line 364 | 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 408 | Line 463 | namespace OpenMD {
463      }
464    }
465  
411 –
466    groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
467      int i, j;  
468   #ifdef IS_MPI
# Line 422 | 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 432 | 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 448 | Line 504 | namespace OpenMD {
504           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
505  
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) {    
# Line 483 | 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 495 | 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 517 | 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    }
661    
# Line 557 | 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 +    // this isn't necessary if we don't have polarizable atoms, but
683 +    // we'll leave it here for now.
684 +    if (storageLayout_ & DataStorage::dslElectricField) {
685 +      
686 +      AtomPlanVectorRow->scatter(atomRowData.electricField,
687 +                                 snap_->atomData.electricField);
688 +      
689 +      int n = snap_->atomData.electricField.size();
690 +      vector<Vector3d> field_tmp(n, V3Zero);
691 +      AtomPlanVectorColumn->scatter(atomColData.electricField,
692 +                                    field_tmp);
693 +      for (int i = 0; i < n; i++)
694 +        snap_->atomData.electricField[i] += field_tmp[i];
695 +    }
696   #endif
697    }
698  
# Line 578 | Line 705 | namespace OpenMD {
705      storageLayout_ = sman_->getStorageLayout();
706   #ifdef IS_MPI
707      if (storageLayout_ & DataStorage::dslFunctional) {
708 <      AtomCommRealRow->gather(snap_->atomData.functional,
708 >      AtomPlanRealRow->gather(snap_->atomData.functional,
709                                atomRowData.functional);
710 <      AtomCommRealColumn->gather(snap_->atomData.functional,
710 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
711                                   atomColData.functional);
712      }
713      
714      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
715 <      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
715 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
716                                atomRowData.functionalDerivative);
717 <      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
717 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
718                                   atomColData.functionalDerivative);
719      }
720   #endif
# Line 601 | Line 728 | namespace OpenMD {
728      int n = snap_->atomData.force.size();
729      vector<Vector3d> frc_tmp(n, V3Zero);
730      
731 <    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
731 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
732      for (int i = 0; i < n; i++) {
733        snap_->atomData.force[i] += frc_tmp[i];
734        frc_tmp[i] = 0.0;
735      }
736      
737 <    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
738 <    for (int i = 0; i < n; i++)
737 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
738 >    for (int i = 0; i < n; i++) {
739        snap_->atomData.force[i] += frc_tmp[i];
740 +    }
741          
742      if (storageLayout_ & DataStorage::dslTorque) {
743  
744        int nt = snap_->atomData.torque.size();
745        vector<Vector3d> trq_tmp(nt, V3Zero);
746  
747 <      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
747 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
748        for (int i = 0; i < nt; i++) {
749          snap_->atomData.torque[i] += trq_tmp[i];
750          trq_tmp[i] = 0.0;
751        }
752        
753 <      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
753 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
754        for (int i = 0; i < nt; i++)
755          snap_->atomData.torque[i] += trq_tmp[i];
756      }
# Line 632 | Line 760 | namespace OpenMD {
760        int ns = snap_->atomData.skippedCharge.size();
761        vector<RealType> skch_tmp(ns, 0.0);
762  
763 <      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
763 >      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
764        for (int i = 0; i < ns; i++) {
765          snap_->atomData.skippedCharge[i] += skch_tmp[i];
766          skch_tmp[i] = 0.0;
767        }
768        
769 <      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
770 <      for (int i = 0; i < ns; i++)
769 >      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
770 >      for (int i = 0; i < ns; i++)
771          snap_->atomData.skippedCharge[i] += skch_tmp[i];
772 +            
773      }
774      
775 +    if (storageLayout_ & DataStorage::dslFlucQForce) {
776 +
777 +      int nq = snap_->atomData.flucQFrc.size();
778 +      vector<RealType> fqfrc_tmp(nq, 0.0);
779 +
780 +      AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
781 +      for (int i = 0; i < nq; i++) {
782 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
783 +        fqfrc_tmp[i] = 0.0;
784 +      }
785 +      
786 +      AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
787 +      for (int i = 0; i < nq; i++)
788 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
789 +            
790 +    }
791 +
792 +    if (storageLayout_ & DataStorage::dslElectricField) {
793 +
794 +      int nef = snap_->atomData.electricField.size();
795 +      vector<Vector3d> efield_tmp(nef, V3Zero);
796 +
797 +      AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
798 +      for (int i = 0; i < nef; i++) {
799 +        snap_->atomData.electricField[i] += efield_tmp[i];
800 +        efield_tmp[i] = 0.0;
801 +      }
802 +      
803 +      AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
804 +      for (int i = 0; i < nef; i++)
805 +        snap_->atomData.electricField[i] += efield_tmp[i];
806 +    }
807 +
808 +
809      nLocal_ = snap_->getNumberOfAtoms();
810  
811      vector<potVec> pot_temp(nLocal_,
812                              Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
813 +    vector<potVec> expot_temp(nLocal_,
814 +                              Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
815  
816      // scatter/gather pot_row into the members of my column
817            
818 <    AtomCommPotRow->scatter(pot_row, pot_temp);
818 >    AtomPlanPotRow->scatter(pot_row, pot_temp);
819 >    AtomPlanPotRow->scatter(expot_row, expot_temp);
820  
821 <    for (int ii = 0;  ii < pot_temp.size(); ii++ )
821 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
822        pairwisePot += pot_temp[ii];
823 <    
823 >
824 >    for (int ii = 0;  ii < expot_temp.size(); ii++ )
825 >      excludedPot += expot_temp[ii];
826 >        
827 >    if (storageLayout_ & DataStorage::dslParticlePot) {
828 >      // This is the pairwise contribution to the particle pot.  The
829 >      // embedding contribution is added in each of the low level
830 >      // non-bonded routines.  In single processor, this is done in
831 >      // unpackInteractionData, not in collectData.
832 >      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
833 >        for (int i = 0; i < nLocal_; i++) {
834 >          // factor of two is because the total potential terms are divided
835 >          // by 2 in parallel due to row/ column scatter      
836 >          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
837 >        }
838 >      }
839 >    }
840 >
841      fill(pot_temp.begin(), pot_temp.end(),
842           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
843 +    fill(expot_temp.begin(), expot_temp.end(),
844 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
845        
846 <    AtomCommPotColumn->scatter(pot_col, pot_temp);    
846 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
847 >    AtomPlanPotColumn->scatter(expot_col, expot_temp);    
848      
849      for (int ii = 0;  ii < pot_temp.size(); ii++ )
850        pairwisePot += pot_temp[ii];    
851 +
852 +    for (int ii = 0;  ii < expot_temp.size(); ii++ )
853 +      excludedPot += expot_temp[ii];    
854 +
855 +    if (storageLayout_ & DataStorage::dslParticlePot) {
856 +      // This is the pairwise contribution to the particle pot.  The
857 +      // embedding contribution is added in each of the low level
858 +      // non-bonded routines.  In single processor, this is done in
859 +      // unpackInteractionData, not in collectData.
860 +      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
861 +        for (int i = 0; i < nLocal_; i++) {
862 +          // factor of two is because the total potential terms are divided
863 +          // by 2 in parallel due to row/ column scatter      
864 +          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
865 +        }
866 +      }
867 +    }
868 +    
869 +    if (storageLayout_ & DataStorage::dslParticlePot) {
870 +      int npp = snap_->atomData.particlePot.size();
871 +      vector<RealType> ppot_temp(npp, 0.0);
872 +
873 +      // This is the direct or embedding contribution to the particle
874 +      // pot.
875 +      
876 +      AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
877 +      for (int i = 0; i < npp; i++) {
878 +        snap_->atomData.particlePot[i] += ppot_temp[i];
879 +      }
880 +
881 +      fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
882 +      
883 +      AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
884 +      for (int i = 0; i < npp; i++) {
885 +        snap_->atomData.particlePot[i] += ppot_temp[i];
886 +      }
887 +    }
888 +
889 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
890 +      RealType ploc1 = pairwisePot[ii];
891 +      RealType ploc2 = 0.0;
892 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
893 +      pairwisePot[ii] = ploc2;
894 +    }
895 +
896 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
897 +      RealType ploc1 = excludedPot[ii];
898 +      RealType ploc2 = 0.0;
899 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
900 +      excludedPot[ii] = ploc2;
901 +    }
902 +
903 +    // Here be dragons.
904 +    MPI::Intracomm col = colComm.getComm();
905 +
906 +    col.Allreduce(MPI::IN_PLACE,
907 +                  &snap_->frameData.conductiveHeatFlux[0], 3,
908 +                  MPI::REALTYPE, MPI::SUM);
909 +
910 +
911   #endif
912  
913    }
914  
915 +  /**
916 +   * Collects information obtained during the post-pair (and embedding
917 +   * functional) loops onto local data structures.
918 +   */
919 +  void ForceMatrixDecomposition::collectSelfData() {
920 +    snap_ = sman_->getCurrentSnapshot();
921 +    storageLayout_ = sman_->getStorageLayout();
922 +
923 + #ifdef IS_MPI
924 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
925 +      RealType ploc1 = embeddingPot[ii];
926 +      RealType ploc2 = 0.0;
927 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
928 +      embeddingPot[ii] = ploc2;
929 +    }    
930 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
931 +      RealType ploc1 = excludedSelfPot[ii];
932 +      RealType ploc2 = 0.0;
933 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
934 +      excludedSelfPot[ii] = ploc2;
935 +    }    
936 + #endif
937 +    
938 +  }
939 +
940 +
941 +
942    int ForceMatrixDecomposition::getNAtomsInRow() {  
943   #ifdef IS_MPI
944      return nAtomsInRow_;
# Line 706 | Line 979 | namespace OpenMD {
979      return d;    
980    }
981  
982 +  Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
983 + #ifdef IS_MPI
984 +    return cgColData.velocity[cg2];
985 + #else
986 +    return snap_->cgData.velocity[cg2];
987 + #endif
988 +  }
989  
990 +  Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
991 + #ifdef IS_MPI
992 +    return atomColData.velocity[atom2];
993 + #else
994 +    return snap_->atomData.velocity[atom2];
995 + #endif
996 +  }
997 +
998 +
999    Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
1000  
1001      Vector3d d;
# Line 772 | Line 1061 | namespace OpenMD {
1061     * We need to exclude some overcounted interactions that result from
1062     * the parallel decomposition.
1063     */
1064 <  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1064 >  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1065      int unique_id_1, unique_id_2;
1066 <
1066 >        
1067   #ifdef IS_MPI
1068      // in MPI, we have to look up the unique IDs for each atom
1069      unique_id_1 = AtomRowToGlobal[atom1];
1070      unique_id_2 = AtomColToGlobal[atom2];
1071 +    // group1 = cgRowToGlobal[cg1];
1072 +    // group2 = cgColToGlobal[cg2];
1073 + #else
1074 +    unique_id_1 = AtomLocalToGlobal[atom1];
1075 +    unique_id_2 = AtomLocalToGlobal[atom2];
1076 +    int group1 = cgLocalToGlobal[cg1];
1077 +    int group2 = cgLocalToGlobal[cg2];
1078 + #endif  
1079  
783 –    // this situation should only arise in MPI simulations
1080      if (unique_id_1 == unique_id_2) return true;
1081 <    
1081 >
1082 > #ifdef IS_MPI
1083      // this prevents us from doing the pair on multiple processors
1084      if (unique_id_1 < unique_id_2) {
1085        if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1086      } else {
1087 <      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1087 >      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1088      }
1089 + #endif    
1090 +
1091 + #ifndef IS_MPI
1092 +    if (group1 == group2) {
1093 +      if (unique_id_1 < unique_id_2) return true;
1094 +    }
1095   #endif
1096 +    
1097      return false;
1098    }
1099  
# Line 803 | Line 1107 | namespace OpenMD {
1107     * field) must still be handled for these pairs.
1108     */
1109    bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1110 <    int unique_id_2;
1110 >
1111 >    // excludesForAtom was constructed to use row/column indices in the MPI
1112 >    // version, and to use local IDs in the non-MPI version:
1113      
808 – #ifdef IS_MPI
809 –    // in MPI, we have to look up the unique IDs for the row atom.
810 –    unique_id_2 = AtomColToGlobal[atom2];
811 – #else
812 –    // in the normal loop, the atom numbers are unique
813 –    unique_id_2 = atom2;
814 – #endif
815 –    
1114      for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1115           i != excludesForAtom[atom1].end(); ++i) {
1116 <      if ( (*i) == unique_id_2 ) return true;
1116 >      if ( (*i) == atom2 ) return true;
1117      }
1118  
1119      return false;
# Line 854 | Line 1152 | namespace OpenMD {
1152        idat.A2 = &(atomColData.aMat[atom2]);
1153      }
1154      
857 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
858 –      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
859 –      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
860 –    }
861 –
1155      if (storageLayout_ & DataStorage::dslTorque) {
1156        idat.t1 = &(atomRowData.torque[atom1]);
1157        idat.t2 = &(atomColData.torque[atom2]);
1158      }
1159  
1160 +    if (storageLayout_ & DataStorage::dslDipole) {
1161 +      idat.dipole1 = &(atomRowData.dipole[atom1]);
1162 +      idat.dipole2 = &(atomColData.dipole[atom2]);
1163 +    }
1164 +
1165 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1166 +      idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1167 +      idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1168 +    }
1169 +
1170      if (storageLayout_ & DataStorage::dslDensity) {
1171        idat.rho1 = &(atomRowData.density[atom1]);
1172        idat.rho2 = &(atomColData.density[atom2]);
# Line 889 | Line 1192 | namespace OpenMD {
1192        idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1193      }
1194  
1195 < #else
1195 >    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1196 >      idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1197 >      idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1198 >    }
1199  
1200 + #else
1201 +    
1202      idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
895 –    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
896 –    //                         ff_->getAtomType(idents[atom2]) );
1203  
1204      if (storageLayout_ & DataStorage::dslAmat) {
1205        idat.A1 = &(snap_->atomData.aMat[atom1]);
1206        idat.A2 = &(snap_->atomData.aMat[atom2]);
1207      }
1208  
903 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
904 –      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
905 –      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
906 –    }
907 –
1209      if (storageLayout_ & DataStorage::dslTorque) {
1210        idat.t1 = &(snap_->atomData.torque[atom1]);
1211        idat.t2 = &(snap_->atomData.torque[atom2]);
1212      }
1213  
1214 +    if (storageLayout_ & DataStorage::dslDipole) {
1215 +      idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1216 +      idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1217 +    }
1218 +
1219 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1220 +      idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1221 +      idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1222 +    }
1223 +
1224      if (storageLayout_ & DataStorage::dslDensity) {    
1225        idat.rho1 = &(snap_->atomData.density[atom1]);
1226        idat.rho2 = &(snap_->atomData.density[atom2]);
# Line 934 | Line 1245 | namespace OpenMD {
1245        idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1246        idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1247      }
1248 +
1249 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1250 +      idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1251 +      idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1252 +    }
1253 +
1254   #endif
1255    }
1256  
1257    
1258    void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1259   #ifdef IS_MPI
1260 <    pot_row[atom1] += 0.5 *  *(idat.pot);
1261 <    pot_col[atom2] += 0.5 *  *(idat.pot);
1260 >    pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1261 >    pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1262 >    expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1263 >    expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1264  
1265      atomRowData.force[atom1] += *(idat.f1);
1266      atomColData.force[atom2] -= *(idat.f1);
1267 +
1268 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1269 +      atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1270 +      atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1271 +    }
1272 +
1273 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1274 +      atomRowData.electricField[atom1] += *(idat.eField1);
1275 +      atomColData.electricField[atom2] += *(idat.eField2);
1276 +    }
1277 +
1278   #else
1279      pairwisePot += *(idat.pot);
1280 +    excludedPot += *(idat.excludedPot);
1281  
1282      snap_->atomData.force[atom1] += *(idat.f1);
1283      snap_->atomData.force[atom2] -= *(idat.f1);
1284 +
1285 +    if (idat.doParticlePot) {
1286 +      // This is the pairwise contribution to the particle pot.  The
1287 +      // embedding contribution is added in each of the low level
1288 +      // non-bonded routines.  In parallel, this calculation is done
1289 +      // in collectData, not in unpackInteractionData.
1290 +      snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1291 +      snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1292 +    }
1293 +    
1294 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1295 +      snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1296 +      snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1297 +    }
1298 +
1299 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1300 +      snap_->atomData.electricField[atom1] += *(idat.eField1);
1301 +      snap_->atomData.electricField[atom2] += *(idat.eField2);
1302 +    }
1303 +
1304   #endif
1305      
1306    }
# Line 974 | Line 1325 | namespace OpenMD {
1325   #endif
1326  
1327      RealType rList_ = (largestRcut_ + skinThickness_);
977 –    RealType rl2 = rList_ * rList_;
1328      Snapshot* snap_ = sman_->getCurrentSnapshot();
1329      Mat3x3d Hmat = snap_->getHmat();
1330      Vector3d Hx = Hmat.getColumn(0);
# Line 1018 | Line 1368 | namespace OpenMD {
1368          for (int j = 0; j < 3; j++) {
1369            scaled[j] -= roundMe(scaled[j]);
1370            scaled[j] += 0.5;
1371 +          // Handle the special case when an object is exactly on the
1372 +          // boundary (a scaled coordinate of 1.0 is the same as
1373 +          // scaled coordinate of 0.0)
1374 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1375          }
1376          
1377          // find xyz-indices of cell that cutoffGroup is in.
# Line 1031 | Line 1385 | namespace OpenMD {
1385          // add this cutoff group to the list of groups in this cell;
1386          cellListRow_[cellIndex].push_back(i);
1387        }
1034 –      
1388        for (int i = 0; i < nGroupsInCol_; i++) {
1389          rs = cgColData.position[i];
1390          
# Line 1043 | Line 1396 | namespace OpenMD {
1396          for (int j = 0; j < 3; j++) {
1397            scaled[j] -= roundMe(scaled[j]);
1398            scaled[j] += 0.5;
1399 +          // Handle the special case when an object is exactly on the
1400 +          // boundary (a scaled coordinate of 1.0 is the same as
1401 +          // scaled coordinate of 0.0)
1402 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1403          }
1404          
1405          // find xyz-indices of cell that cutoffGroup is in.
# Line 1056 | Line 1413 | namespace OpenMD {
1413          // add this cutoff group to the list of groups in this cell;
1414          cellListCol_[cellIndex].push_back(i);
1415        }
1416 +    
1417   #else
1418        for (int i = 0; i < nGroups_; i++) {
1419          rs = snap_->cgData.position[i];
# Line 1068 | Line 1426 | namespace OpenMD {
1426          for (int j = 0; j < 3; j++) {
1427            scaled[j] -= roundMe(scaled[j]);
1428            scaled[j] += 0.5;
1429 +          // Handle the special case when an object is exactly on the
1430 +          // boundary (a scaled coordinate of 1.0 is the same as
1431 +          // scaled coordinate of 0.0)
1432 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1433          }
1434          
1435          // find xyz-indices of cell that cutoffGroup is in.
# Line 1076 | Line 1438 | namespace OpenMD {
1438          whichCell.z() = nCells_.z() * scaled.z();
1439          
1440          // find single index of this cell:
1441 <        cellIndex = Vlinear(whichCell, nCells_);      
1441 >        cellIndex = Vlinear(whichCell, nCells_);
1442          
1443          // add this cutoff group to the list of groups in this cell;
1444          cellList_[cellIndex].push_back(i);
1445        }
1446 +
1447   #endif
1448  
1449        for (int m1z = 0; m1z < nCells_.z(); m1z++) {
# Line 1093 | Line 1456 | namespace OpenMD {
1456                   os != cellOffsets_.end(); ++os) {
1457                
1458                Vector3i m2v = m1v + (*os);
1459 <              
1459 >            
1460 >
1461                if (m2v.x() >= nCells_.x()) {
1462                  m2v.x() = 0;          
1463                } else if (m2v.x() < 0) {
# Line 1111 | Line 1475 | namespace OpenMD {
1475                } else if (m2v.z() < 0) {
1476                  m2v.z() = nCells_.z() - 1;
1477                }
1478 <              
1478 >
1479                int m2 = Vlinear (m2v, nCells_);
1480                
1481   #ifdef IS_MPI
# Line 1120 | Line 1484 | namespace OpenMD {
1484                  for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1485                       j2 != cellListCol_[m2].end(); ++j2) {
1486                    
1487 <                  // Always do this if we're in different cells or if
1488 <                  // we're in the same cell and the global index of the
1489 <                  // j2 cutoff group is less than the j1 cutoff group
1490 <                  
1491 <                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1492 <                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1493 <                    snap_->wrapVector(dr);
1494 <                    cuts = getGroupCutoffs( (*j1), (*j2) );
1495 <                    if (dr.lengthSquare() < cuts.third) {
1132 <                      neighborList.push_back(make_pair((*j1), (*j2)));
1133 <                    }
1134 <                  }
1487 >                  // In parallel, we need to visit *all* pairs of row
1488 >                  // & column indicies and will divide labor in the
1489 >                  // force evaluation later.
1490 >                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1491 >                  snap_->wrapVector(dr);
1492 >                  cuts = getGroupCutoffs( (*j1), (*j2) );
1493 >                  if (dr.lengthSquare() < cuts.third) {
1494 >                    neighborList.push_back(make_pair((*j1), (*j2)));
1495 >                  }                  
1496                  }
1497                }
1498   #else
1138 –              
1499                for (vector<int>::iterator j1 = cellList_[m1].begin();
1500                     j1 != cellList_[m1].end(); ++j1) {
1501                  for (vector<int>::iterator j2 = cellList_[m2].begin();
1502                       j2 != cellList_[m2].end(); ++j2) {
1503 <                  
1503 >    
1504                    // Always do this if we're in different cells or if
1505 <                  // we're in the same cell and the global index of the
1506 <                  // j2 cutoff group is less than the j1 cutoff group
1507 <                  
1508 <                  if (m2 != m1 || (*j2) < (*j1)) {
1505 >                  // we're in the same cell and the global index of
1506 >                  // the j2 cutoff group is greater than or equal to
1507 >                  // the j1 cutoff group.  Note that Rappaport's code
1508 >                  // has a "less than" conditional here, but that
1509 >                  // deals with atom-by-atom computation.  OpenMD
1510 >                  // allows atoms within a single cutoff group to
1511 >                  // interact with each other.
1512 >
1513 >
1514 >
1515 >                  if (m2 != m1 || (*j2) >= (*j1) ) {
1516 >
1517                      dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1518                      snap_->wrapVector(dr);
1519                      cuts = getGroupCutoffs( (*j1), (*j2) );
# Line 1164 | Line 1532 | namespace OpenMD {
1532        // branch to do all cutoff group pairs
1533   #ifdef IS_MPI
1534        for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1535 <        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1535 >        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
1536            dr = cgColData.position[j2] - cgRowData.position[j1];
1537            snap_->wrapVector(dr);
1538            cuts = getGroupCutoffs( j1, j2 );
# Line 1172 | Line 1540 | namespace OpenMD {
1540              neighborList.push_back(make_pair(j1, j2));
1541            }
1542          }
1543 <      }
1543 >      }      
1544   #else
1545 <      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1546 <        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1545 >      // include all groups here.
1546 >      for (int j1 = 0; j1 < nGroups_; j1++) {
1547 >        // include self group interactions j2 == j1
1548 >        for (int j2 = j1; j2 < nGroups_; j2++) {
1549            dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1550            snap_->wrapVector(dr);
1551            cuts = getGroupCutoffs( j1, j2 );
1552            if (dr.lengthSquare() < cuts.third) {
1553              neighborList.push_back(make_pair(j1, j2));
1554            }
1555 <        }
1556 <      }        
1555 >        }    
1556 >      }
1557   #endif
1558      }
1559        

Diff Legend

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