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

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1601 by gezelter, Thu Aug 4 20:04:35 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1879 by gezelter, Sun Jun 16 15:15:42 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 53 | Line 54 | namespace OpenMD {
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_.push_back( Vector3i(-1, 0, 0) );
57 <    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
58 <    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
59 <    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
60 <    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
61 <    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
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) );
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) );
66 –    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67 –    cellOffsets_.push_back( Vector3i( 0, 1,-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  
# Line 79 | Line 95 | namespace OpenMD {
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 93 | 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      MPI::Intracomm row = rowComm.getComm();
# Line 129 | 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      
# Line 149 | 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      AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
# Line 218 | Line 248 | namespace OpenMD {
248        }      
249      }
250  
251 < #endif
222 <
223 <    // allocate memory for the parallel objects
224 <    atypesLocal.resize(nLocal_);
225 <
226 <    for (int i = 0; i < nLocal_; i++)
227 <      atypesLocal[i] = ff_->getAtomType(idents[i]);
228 <
229 <    groupList_.clear();
230 <    groupList_.resize(nGroups_);
231 <    for (int i = 0; i < nGroups_; i++) {
232 <      int gid = cgLocalToGlobal[i];
233 <      for (int j = 0; j < nLocal_; j++) {
234 <        int aid = AtomLocalToGlobal[j];
235 <        if (globalGroupMembership[aid] == gid) {
236 <          groupList_[i].push_back(j);
237 <        }
238 <      }      
239 <    }
240 <
251 > #else
252      excludesForAtom.clear();
253      excludesForAtom.resize(nLocal_);
254      toposForAtom.clear();
# Line 270 | 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 279 | Line 310 | namespace OpenMD {
310      
311      RealType tol = 1e-6;
312      largestRcut_ = 0.0;
282 –    RealType rc;
313      int atid;
314      set<AtomType*> atypes = info_->getSimulatedAtomTypes();
315      
# Line 364 | 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 389 | 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 433 | Line 463 | namespace OpenMD {
463      }
464    }
465  
436 –
466    groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
467      int i, j;  
468   #ifdef IS_MPI
# Line 447 | 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 457 | 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 475 | 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 508 | 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 520 | 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 555 | Line 613 | namespace OpenMD {
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) {
# Line 563 | Line 632 | namespace OpenMD {
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 <      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
639 <                                atomRowData.electroFrame);
640 <      AtomPlanMatrixColumn->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 594 | Line 678 | namespace OpenMD {
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 668 | Line 767 | namespace OpenMD {
767        }
768        
769        AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
770 <      for (int i = 0; i < ns; i++)
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      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      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;
# Line 699 | Line 893 | namespace OpenMD {
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 739 | Line 975 | namespace OpenMD {
975      d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
976   #endif
977      
978 <    snap_->wrapVector(d);
978 >    if (usePeriodicBoundaryConditions_) {
979 >      snap_->wrapVector(d);
980 >    }
981      return d;    
982    }
983  
984 +  Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
985 + #ifdef IS_MPI
986 +    return cgColData.velocity[cg2];
987 + #else
988 +    return snap_->cgData.velocity[cg2];
989 + #endif
990 +  }
991  
992 +  Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
993 + #ifdef IS_MPI
994 +    return atomColData.velocity[atom2];
995 + #else
996 +    return snap_->atomData.velocity[atom2];
997 + #endif
998 +  }
999 +
1000 +
1001    Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
1002  
1003      Vector3d d;
# Line 753 | Line 1007 | namespace OpenMD {
1007   #else
1008      d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
1009   #endif
1010 <
1011 <    snap_->wrapVector(d);
1010 >    if (usePeriodicBoundaryConditions_) {
1011 >      snap_->wrapVector(d);
1012 >    }
1013      return d;    
1014    }
1015    
# Line 766 | Line 1021 | namespace OpenMD {
1021   #else
1022      d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1023   #endif
1024 <    
1025 <    snap_->wrapVector(d);
1024 >    if (usePeriodicBoundaryConditions_) {
1025 >      snap_->wrapVector(d);
1026 >    }
1027      return d;    
1028    }
1029  
# Line 796 | Line 1052 | namespace OpenMD {
1052   #else
1053      d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1054   #endif
1055 <
1056 <    snap_->wrapVector(d);
1055 >    if (usePeriodicBoundaryConditions_) {
1056 >      snap_->wrapVector(d);
1057 >    }
1058      return d;    
1059    }
1060  
# Line 809 | Line 1066 | namespace OpenMD {
1066     * We need to exclude some overcounted interactions that result from
1067     * the parallel decomposition.
1068     */
1069 <  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1069 >  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1070      int unique_id_1, unique_id_2;
1071 <    
1071 >        
1072   #ifdef IS_MPI
1073      // in MPI, we have to look up the unique IDs for each atom
1074      unique_id_1 = AtomRowToGlobal[atom1];
1075      unique_id_2 = AtomColToGlobal[atom2];
1076 +    // group1 = cgRowToGlobal[cg1];
1077 +    // group2 = cgColToGlobal[cg2];
1078 + #else
1079 +    unique_id_1 = AtomLocalToGlobal[atom1];
1080 +    unique_id_2 = AtomLocalToGlobal[atom2];
1081 +    int group1 = cgLocalToGlobal[cg1];
1082 +    int group2 = cgLocalToGlobal[cg2];
1083 + #endif  
1084  
820 –    // this situation should only arise in MPI simulations
1085      if (unique_id_1 == unique_id_2) return true;
1086 <    
1086 >
1087 > #ifdef IS_MPI
1088      // this prevents us from doing the pair on multiple processors
1089      if (unique_id_1 < unique_id_2) {
1090        if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1091      } else {
1092 <      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1092 >      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1093      }
1094 + #endif    
1095 +
1096 + #ifndef IS_MPI
1097 +    if (group1 == group2) {
1098 +      if (unique_id_1 < unique_id_2) return true;
1099 +    }
1100   #endif
1101 +    
1102      return false;
1103    }
1104  
# Line 840 | Line 1112 | namespace OpenMD {
1112     * field) must still be handled for these pairs.
1113     */
1114    bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1115 <    int unique_id_2;
1116 < #ifdef IS_MPI
1117 <    // in MPI, we have to look up the unique IDs for the row atom.
846 <    unique_id_2 = AtomColToGlobal[atom2];
847 < #else
848 <    // in the normal loop, the atom numbers are unique
849 <    unique_id_2 = atom2;
850 < #endif
1115 >
1116 >    // excludesForAtom was constructed to use row/column indices in the MPI
1117 >    // version, and to use local IDs in the non-MPI version:
1118      
1119      for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1120           i != excludesForAtom[atom1].end(); ++i) {
1121 <      if ( (*i) == unique_id_2 ) return true;
1121 >      if ( (*i) == atom2 ) return true;
1122      }
1123  
1124      return false;
# Line 890 | Line 1157 | namespace OpenMD {
1157        idat.A2 = &(atomColData.aMat[atom2]);
1158      }
1159      
893 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
894 –      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
895 –      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
896 –    }
897 –
1160      if (storageLayout_ & DataStorage::dslTorque) {
1161        idat.t1 = &(atomRowData.torque[atom1]);
1162        idat.t2 = &(atomColData.torque[atom2]);
1163      }
1164  
1165 +    if (storageLayout_ & DataStorage::dslDipole) {
1166 +      idat.dipole1 = &(atomRowData.dipole[atom1]);
1167 +      idat.dipole2 = &(atomColData.dipole[atom2]);
1168 +    }
1169 +
1170 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1171 +      idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1172 +      idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1173 +    }
1174 +
1175      if (storageLayout_ & DataStorage::dslDensity) {
1176        idat.rho1 = &(atomRowData.density[atom1]);
1177        idat.rho2 = &(atomColData.density[atom2]);
# Line 925 | Line 1197 | namespace OpenMD {
1197        idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1198      }
1199  
1200 < #else
1200 >    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1201 >      idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1202 >      idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1203 >    }
1204  
1205 + #else
1206 +    
1207      idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
931 –    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
932 –    //                         ff_->getAtomType(idents[atom2]) );
1208  
1209      if (storageLayout_ & DataStorage::dslAmat) {
1210        idat.A1 = &(snap_->atomData.aMat[atom1]);
1211        idat.A2 = &(snap_->atomData.aMat[atom2]);
1212      }
1213  
939 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
940 –      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
941 –      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
942 –    }
943 –
1214      if (storageLayout_ & DataStorage::dslTorque) {
1215        idat.t1 = &(snap_->atomData.torque[atom1]);
1216        idat.t2 = &(snap_->atomData.torque[atom2]);
1217      }
1218  
1219 +    if (storageLayout_ & DataStorage::dslDipole) {
1220 +      idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1221 +      idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1222 +    }
1223 +
1224 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1225 +      idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1226 +      idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1227 +    }
1228 +
1229      if (storageLayout_ & DataStorage::dslDensity) {    
1230        idat.rho1 = &(snap_->atomData.density[atom1]);
1231        idat.rho2 = &(snap_->atomData.density[atom2]);
# Line 970 | Line 1250 | namespace OpenMD {
1250        idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1251        idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1252      }
1253 +
1254 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1255 +      idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1256 +      idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1257 +    }
1258 +
1259   #endif
1260    }
1261  
1262    
1263    void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1264   #ifdef IS_MPI
1265 <    pot_row[atom1] += 0.5 *  *(idat.pot);
1266 <    pot_col[atom2] += 0.5 *  *(idat.pot);
1265 >    pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1266 >    pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1267 >    expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1268 >    expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1269  
1270      atomRowData.force[atom1] += *(idat.f1);
1271      atomColData.force[atom2] -= *(idat.f1);
1272 +
1273 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1274 +      atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1275 +      atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1276 +    }
1277 +
1278 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1279 +      atomRowData.electricField[atom1] += *(idat.eField1);
1280 +      atomColData.electricField[atom2] += *(idat.eField2);
1281 +    }
1282 +
1283   #else
1284      pairwisePot += *(idat.pot);
1285 +    excludedPot += *(idat.excludedPot);
1286  
1287      snap_->atomData.force[atom1] += *(idat.f1);
1288      snap_->atomData.force[atom2] -= *(idat.f1);
1289 +
1290 +    if (idat.doParticlePot) {
1291 +      // This is the pairwise contribution to the particle pot.  The
1292 +      // embedding contribution is added in each of the low level
1293 +      // non-bonded routines.  In parallel, this calculation is done
1294 +      // in collectData, not in unpackInteractionData.
1295 +      snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1296 +      snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1297 +    }
1298 +    
1299 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1300 +      snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1301 +      snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1302 +    }
1303 +
1304 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1305 +      snap_->atomData.electricField[atom1] += *(idat.eField1);
1306 +      snap_->atomData.electricField[atom2] += *(idat.eField2);
1307 +    }
1308 +
1309   #endif
1310      
1311    }
# Line 1002 | Line 1322 | namespace OpenMD {
1322      groupCutoffs cuts;
1323      bool doAllPairs = false;
1324  
1325 +    RealType rList_ = (largestRcut_ + skinThickness_);
1326 +    Snapshot* snap_ = sman_->getCurrentSnapshot();
1327 +    Mat3x3d box;
1328 +    Mat3x3d invBox;
1329 +
1330 +    Vector3d rs, scaled, dr;
1331 +    Vector3i whichCell;
1332 +    int cellIndex;
1333 +
1334   #ifdef IS_MPI
1335      cellListRow_.clear();
1336      cellListCol_.clear();
1337   #else
1338      cellList_.clear();
1339   #endif
1340 <
1341 <    RealType rList_ = (largestRcut_ + skinThickness_);
1342 <    RealType rl2 = rList_ * rList_;
1343 <    Snapshot* snap_ = sman_->getCurrentSnapshot();
1344 <    Mat3x3d Hmat = snap_->getHmat();
1345 <    Vector3d Hx = Hmat.getColumn(0);
1346 <    Vector3d Hy = Hmat.getColumn(1);
1347 <    Vector3d Hz = Hmat.getColumn(2);
1348 <
1349 <    nCells_.x() = (int) ( Hx.length() )/ rList_;
1350 <    nCells_.y() = (int) ( Hy.length() )/ rList_;
1351 <    nCells_.z() = (int) ( Hz.length() )/ rList_;
1352 <
1340 >    
1341 >    if (!usePeriodicBoundaryConditions_) {
1342 >      box = snap_->getBoundingBox();
1343 >      invBox = snap_->getInvBoundingBox();
1344 >    } else {
1345 >      box = snap_->getHmat();
1346 >      invBox = snap_->getInvHmat();
1347 >    }
1348 >    
1349 >    Vector3d boxX = box.getColumn(0);
1350 >    Vector3d boxY = box.getColumn(1);
1351 >    Vector3d boxZ = box.getColumn(2);
1352 >    
1353 >    nCells_.x() = (int) ( boxX.length() )/ rList_;
1354 >    nCells_.y() = (int) ( boxY.length() )/ rList_;
1355 >    nCells_.z() = (int) ( boxZ.length() )/ rList_;
1356 >    
1357      // handle small boxes where the cell offsets can end up repeating cells
1358      
1359      if (nCells_.x() < 3) doAllPairs = true;
1360      if (nCells_.y() < 3) doAllPairs = true;
1361      if (nCells_.z() < 3) doAllPairs = true;
1362 <
1030 <    Mat3x3d invHmat = snap_->getInvHmat();
1031 <    Vector3d rs, scaled, dr;
1032 <    Vector3i whichCell;
1033 <    int cellIndex;
1362 >    
1363      int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1364 <
1364 >    
1365   #ifdef IS_MPI
1366      cellListRow_.resize(nCtot);
1367      cellListCol_.resize(nCtot);
1368   #else
1369      cellList_.resize(nCtot);
1370   #endif
1371 <
1371 >    
1372      if (!doAllPairs) {
1373   #ifdef IS_MPI
1374 <
1374 >      
1375        for (int i = 0; i < nGroupsInRow_; i++) {
1376          rs = cgRowData.position[i];
1377          
1378          // scaled positions relative to the box vectors
1379 <        scaled = invHmat * rs;
1379 >        scaled = invBox * rs;
1380          
1381          // wrap the vector back into the unit box by subtracting integer box
1382          // numbers
1383          for (int j = 0; j < 3; j++) {
1384            scaled[j] -= roundMe(scaled[j]);
1385            scaled[j] += 0.5;
1386 +          // Handle the special case when an object is exactly on the
1387 +          // boundary (a scaled coordinate of 1.0 is the same as
1388 +          // scaled coordinate of 0.0)
1389 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1390          }
1391          
1392          // find xyz-indices of cell that cutoffGroup is in.
# Line 1071 | Line 1404 | namespace OpenMD {
1404          rs = cgColData.position[i];
1405          
1406          // scaled positions relative to the box vectors
1407 <        scaled = invHmat * rs;
1407 >        scaled = invBox * rs;
1408          
1409          // wrap the vector back into the unit box by subtracting integer box
1410          // numbers
1411          for (int j = 0; j < 3; j++) {
1412            scaled[j] -= roundMe(scaled[j]);
1413            scaled[j] += 0.5;
1414 +          // Handle the special case when an object is exactly on the
1415 +          // boundary (a scaled coordinate of 1.0 is the same as
1416 +          // scaled coordinate of 0.0)
1417 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1418          }
1419          
1420          // find xyz-indices of cell that cutoffGroup is in.
# Line 1091 | Line 1428 | namespace OpenMD {
1428          // add this cutoff group to the list of groups in this cell;
1429          cellListCol_[cellIndex].push_back(i);
1430        }
1431 +      
1432   #else
1433        for (int i = 0; i < nGroups_; i++) {
1434          rs = snap_->cgData.position[i];
1435          
1436          // scaled positions relative to the box vectors
1437 <        scaled = invHmat * rs;
1437 >        scaled = invBox * rs;
1438          
1439          // wrap the vector back into the unit box by subtracting integer box
1440          // numbers
1441          for (int j = 0; j < 3; j++) {
1442            scaled[j] -= roundMe(scaled[j]);
1443            scaled[j] += 0.5;
1444 +          // Handle the special case when an object is exactly on the
1445 +          // boundary (a scaled coordinate of 1.0 is the same as
1446 +          // scaled coordinate of 0.0)
1447 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1448          }
1449          
1450          // find xyz-indices of cell that cutoffGroup is in.
# Line 1116 | Line 1458 | namespace OpenMD {
1458          // add this cutoff group to the list of groups in this cell;
1459          cellList_[cellIndex].push_back(i);
1460        }
1461 +
1462   #endif
1463  
1464        for (int m1z = 0; m1z < nCells_.z(); m1z++) {
# Line 1128 | Line 1471 | namespace OpenMD {
1471                   os != cellOffsets_.end(); ++os) {
1472                
1473                Vector3i m2v = m1v + (*os);
1474 <              
1474 >            
1475 >
1476                if (m2v.x() >= nCells_.x()) {
1477                  m2v.x() = 0;          
1478                } else if (m2v.x() < 0) {
# Line 1146 | Line 1490 | namespace OpenMD {
1490                } else if (m2v.z() < 0) {
1491                  m2v.z() = nCells_.z() - 1;
1492                }
1493 <              
1493 >
1494                int m2 = Vlinear (m2v, nCells_);
1495                
1496   #ifdef IS_MPI
# Line 1155 | Line 1499 | namespace OpenMD {
1499                  for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1500                       j2 != cellListCol_[m2].end(); ++j2) {
1501                    
1502 <                  // In parallel, we need to visit *all* pairs of row &
1503 <                  // column indicies and will truncate later on.
1502 >                  // In parallel, we need to visit *all* pairs of row
1503 >                  // & column indicies and will divide labor in the
1504 >                  // force evaluation later.
1505                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1506 <                  snap_->wrapVector(dr);
1506 >                  if (usePeriodicBoundaryConditions_) {
1507 >                    snap_->wrapVector(dr);
1508 >                  }
1509                    cuts = getGroupCutoffs( (*j1), (*j2) );
1510                    if (dr.lengthSquare() < cuts.third) {
1511                      neighborList.push_back(make_pair((*j1), (*j2)));
# Line 1166 | Line 1513 | namespace OpenMD {
1513                  }
1514                }
1515   #else
1169 –              
1516                for (vector<int>::iterator j1 = cellList_[m1].begin();
1517                     j1 != cellList_[m1].end(); ++j1) {
1518                  for (vector<int>::iterator j2 = cellList_[m2].begin();
1519                       j2 != cellList_[m2].end(); ++j2) {
1520 <                  
1520 >    
1521                    // Always do this if we're in different cells or if
1522 <                  // we're in the same cell and the global index of the
1523 <                  // j2 cutoff group is less than the j1 cutoff group
1524 <                  
1525 <                  if (m2 != m1 || (*j2) < (*j1)) {
1522 >                  // we're in the same cell and the global index of
1523 >                  // the j2 cutoff group is greater than or equal to
1524 >                  // the j1 cutoff group.  Note that Rappaport's code
1525 >                  // has a "less than" conditional here, but that
1526 >                  // deals with atom-by-atom computation.  OpenMD
1527 >                  // allows atoms within a single cutoff group to
1528 >                  // interact with each other.
1529 >
1530 >                  if (m2 != m1 || (*j2) >= (*j1) ) {
1531 >
1532                      dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1533 <                    snap_->wrapVector(dr);
1533 >                    if (usePeriodicBoundaryConditions_) {
1534 >                      snap_->wrapVector(dr);
1535 >                    }
1536                      cuts = getGroupCutoffs( (*j1), (*j2) );
1537                      if (dr.lengthSquare() < cuts.third) {
1538                        neighborList.push_back(make_pair((*j1), (*j2)));
# Line 1195 | Line 1549 | namespace OpenMD {
1549        // branch to do all cutoff group pairs
1550   #ifdef IS_MPI
1551        for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1552 <        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1552 >        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
1553            dr = cgColData.position[j2] - cgRowData.position[j1];
1554 <          snap_->wrapVector(dr);
1554 >          if (usePeriodicBoundaryConditions_) {
1555 >            snap_->wrapVector(dr);
1556 >          }
1557            cuts = getGroupCutoffs( j1, j2 );
1558            if (dr.lengthSquare() < cuts.third) {
1559              neighborList.push_back(make_pair(j1, j2));
1560            }
1561          }
1562 <      }
1562 >      }      
1563   #else
1564 <      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1565 <        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1564 >      // include all groups here.
1565 >      for (int j1 = 0; j1 < nGroups_; j1++) {
1566 >        // include self group interactions j2 == j1
1567 >        for (int j2 = j1; j2 < nGroups_; j2++) {
1568            dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1569 <          snap_->wrapVector(dr);
1569 >          if (usePeriodicBoundaryConditions_) {
1570 >            snap_->wrapVector(dr);
1571 >          }
1572            cuts = getGroupCutoffs( j1, j2 );
1573            if (dr.lengthSquare() < cuts.third) {
1574              neighborList.push_back(make_pair(j1, j2));
1575            }
1576 <        }
1577 <      }        
1576 >        }    
1577 >      }
1578   #endif
1579      }
1580        

Diff Legend

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