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

Comparing branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents):
Revision 1581 by gezelter, Mon Jun 13 22:13:12 2011 UTC vs.
Revision 1588 by gezelter, Sat Jul 9 15:05:59 2011 UTC

# Line 57 | Line 57 | namespace OpenMD {
57      storageLayout_ = sman_->getStorageLayout();
58      ff_ = info_->getForceField();
59      nLocal_ = snap_->getNumberOfAtoms();
60 <
60 >    
61      nGroups_ = info_->getNLocalCutoffGroups();
62 –    cerr << "in dId, nGroups = " << nGroups_ << "\n";
62      // gather the information for atomtype IDs (atids):
63 <    identsLocal = info_->getIdentArray();
63 >    idents = info_->getIdentArray();
64      AtomLocalToGlobal = info_->getGlobalAtomIndices();
65      cgLocalToGlobal = info_->getGlobalGroupIndices();
66      vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
67 +
68      massFactors = info_->getMassFactors();
69 –    PairList excludes = info_->getExcludedInteractions();
70 –    PairList oneTwo = info_->getOneTwoInteractions();
71 –    PairList oneThree = info_->getOneThreeInteractions();
72 –    PairList oneFour = info_->getOneFourInteractions();
69  
70 +    PairList* excludes = info_->getExcludedInteractions();
71 +    PairList* oneTwo = info_->getOneTwoInteractions();
72 +    PairList* oneThree = info_->getOneThreeInteractions();
73 +    PairList* oneFour = info_->getOneFourInteractions();
74 +
75   #ifdef IS_MPI
76  
77      AtomCommIntRow = new Communicator<Row,int>(nLocal_);
# Line 108 | Line 109 | namespace OpenMD {
109      identsRow.resize(nAtomsInRow_);
110      identsCol.resize(nAtomsInCol_);
111      
112 <    AtomCommIntRow->gather(identsLocal, identsRow);
113 <    AtomCommIntColumn->gather(identsLocal, identsCol);
112 >    AtomCommIntRow->gather(idents, identsRow);
113 >    AtomCommIntColumn->gather(idents, identsCol);
114      
115 +    vector<int>::iterator it;
116 +    for (it = AtomLocalToGlobal.begin(); it != AtomLocalToGlobal.end(); ++it) {
117 +      cerr << "my AtomLocalToGlobal = " << (*it) << "\n";
118 +    }
119      AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
120      AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
121      
# Line 142 | Line 147 | namespace OpenMD {
147        }      
148      }
149  
150 <    skipsForAtom.clear();
151 <    skipsForAtom.resize(nAtomsInRow_);
150 >    excludesForAtom.clear();
151 >    excludesForAtom.resize(nAtomsInRow_);
152      toposForAtom.clear();
153      toposForAtom.resize(nAtomsInRow_);
154      topoDist.clear();
# Line 154 | Line 159 | namespace OpenMD {
159        for (int j = 0; j < nAtomsInCol_; j++) {
160          int jglob = AtomColToGlobal[j];
161  
162 <        if (excludes.hasPair(iglob, jglob))
163 <          skipsForAtom[i].push_back(j);      
162 >        if (excludes->hasPair(iglob, jglob))
163 >          excludesForAtom[i].push_back(j);      
164          
165 <        if (oneTwo.hasPair(iglob, jglob)) {
165 >        if (oneTwo->hasPair(iglob, jglob)) {
166            toposForAtom[i].push_back(j);
167            topoDist[i].push_back(1);
168          } else {
169 <          if (oneThree.hasPair(iglob, jglob)) {
169 >          if (oneThree->hasPair(iglob, jglob)) {
170              toposForAtom[i].push_back(j);
171              topoDist[i].push_back(2);
172            } else {
173 <            if (oneFour.hasPair(iglob, jglob)) {
173 >            if (oneFour->hasPair(iglob, jglob)) {
174                toposForAtom[i].push_back(j);
175                topoDist[i].push_back(3);
176              }
# Line 188 | Line 193 | namespace OpenMD {
193        }      
194      }
195  
196 <    skipsForAtom.clear();
197 <    skipsForAtom.resize(nLocal_);
196 >    excludesForAtom.clear();
197 >    excludesForAtom.resize(nLocal_);
198      toposForAtom.clear();
199      toposForAtom.resize(nLocal_);
200      topoDist.clear();
# Line 201 | Line 206 | namespace OpenMD {
206        for (int j = 0; j < nLocal_; j++) {
207          int jglob = AtomLocalToGlobal[j];
208  
209 <        if (excludes.hasPair(iglob, jglob))
210 <          skipsForAtom[i].push_back(j);              
209 >        if (excludes->hasPair(iglob, jglob))
210 >          excludesForAtom[i].push_back(j);              
211          
212 <        if (oneTwo.hasPair(iglob, jglob)) {
212 >        if (oneTwo->hasPair(iglob, jglob)) {
213            toposForAtom[i].push_back(j);
214            topoDist[i].push_back(1);
215          } else {
216 <          if (oneThree.hasPair(iglob, jglob)) {
216 >          if (oneThree->hasPair(iglob, jglob)) {
217              toposForAtom[i].push_back(j);
218              topoDist[i].push_back(2);
219            } else {
220 <            if (oneFour.hasPair(iglob, jglob)) {
220 >            if (oneFour->hasPair(iglob, jglob)) {
221                toposForAtom[i].push_back(j);
222                topoDist[i].push_back(3);
223              }
# Line 222 | Line 227 | namespace OpenMD {
227      }
228      
229      createGtypeCutoffMap();
230 +
231    }
232    
233    void ForceMatrixDecomposition::createGtypeCutoffMap() {
234 <
234 >    
235      RealType tol = 1e-6;
236      RealType rc;
237      int atid;
238      set<AtomType*> atypes = info_->getSimulatedAtomTypes();
239 <    vector<RealType> atypeCutoff;
240 <    atypeCutoff.resize( atypes.size() );
235 <
239 >    map<int, RealType> atypeCutoff;
240 >      
241      for (set<AtomType*>::iterator at = atypes.begin();
242           at != atypes.end(); ++at){
238 –      rc = interactionMan_->getSuggestedCutoffRadius(*at);
243        atid = (*at)->getIdent();
244 <      atypeCutoff[atid] = rc;
244 >      if (userChoseCutoff_)
245 >        atypeCutoff[atid] = userCutoff_;
246 >      else
247 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
248      }
249  
250      vector<RealType> gTypeCutoffs;
244 –
251      // first we do a single loop over the cutoff groups to find the
252      // largest cutoff for any atypes present in this group.
253   #ifdef IS_MPI
# Line 299 | Line 305 | namespace OpenMD {
305  
306      vector<RealType> groupCutoff(nGroups_, 0.0);
307      groupToGtype.resize(nGroups_);
302 –
303 –    cerr << "nGroups = " << nGroups_ << "\n";
308      for (int cg1 = 0; cg1 < nGroups_; cg1++) {
309  
310        groupCutoff[cg1] = 0.0;
# Line 309 | Line 313 | namespace OpenMD {
313        for (vector<int>::iterator ia = atomList.begin();
314             ia != atomList.end(); ++ia) {            
315          int atom1 = (*ia);
316 <        atid = identsLocal[atom1];
316 >        atid = idents[atom1];
317          if (atypeCutoff[atid] > groupCutoff[cg1]) {
318            groupCutoff[cg1] = atypeCutoff[atid];
319          }
# Line 329 | Line 333 | namespace OpenMD {
333      }
334   #endif
335  
332 –    cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
336      // Now we find the maximum group cutoff value present in the simulation
337  
338      RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
# Line 378 | Line 381 | namespace OpenMD {
381            if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
382              sprintf(painCave.errMsg,
383                      "ForceMatrixDecomposition::createGtypeCutoffMap "
384 <                    "user-specified rCut does not match computed group Cutoff\n");
384 >                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
385              painCave.severity = OPENMD_ERROR;
386              painCave.isFatal = 1;
387              simError();            
# Line 410 | Line 413 | namespace OpenMD {
413    }
414  
415    void ForceMatrixDecomposition::zeroWorkArrays() {
416 +    pairwisePot = 0.0;
417 +    embeddingPot = 0.0;
418  
414 –    for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
415 –      longRangePot_[j] = 0.0;
416 –    }
417 –
419   #ifdef IS_MPI
420      if (storageLayout_ & DataStorage::dslForce) {
421        fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
# Line 430 | Line 431 | namespace OpenMD {
431           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
432  
433      fill(pot_col.begin(), pot_col.end(),
434 <         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
434 <    
435 <    pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
434 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
435  
436      if (storageLayout_ & DataStorage::dslParticlePot) {    
437        fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
# Line 456 | Line 455 | namespace OpenMD {
455             atomColData.functionalDerivative.end(), 0.0);
456      }
457  
458 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
459 +      fill(atomRowData.skippedCharge.begin(),
460 +           atomRowData.skippedCharge.end(), 0.0);
461 +      fill(atomColData.skippedCharge.begin(),
462 +           atomColData.skippedCharge.end(), 0.0);
463 +    }
464 +
465   #else
466      
467      if (storageLayout_ & DataStorage::dslParticlePot) {      
# Line 475 | Line 481 | namespace OpenMD {
481        fill(snap_->atomData.functionalDerivative.begin(),
482             snap_->atomData.functionalDerivative.end(), 0.0);
483      }
484 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
485 +      fill(snap_->atomData.skippedCharge.begin(),
486 +           snap_->atomData.skippedCharge.end(), 0.0);
487 +    }
488   #endif
489      
490    }
# Line 582 | Line 592 | namespace OpenMD {
592      
593      if (storageLayout_ & DataStorage::dslTorque) {
594  
595 <      int nt = snap_->atomData.force.size();
595 >      int nt = snap_->atomData.torque.size();
596        vector<Vector3d> trq_tmp(nt, V3Zero);
597  
598        AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
599 <      for (int i = 0; i < n; i++) {
599 >      for (int i = 0; i < nt; i++) {
600          snap_->atomData.torque[i] += trq_tmp[i];
601          trq_tmp[i] = 0.0;
602        }
603        
604        AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
605 <      for (int i = 0; i < n; i++)
605 >      for (int i = 0; i < nt; i++)
606          snap_->atomData.torque[i] += trq_tmp[i];
607 +    }
608 +
609 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
610 +
611 +      int ns = snap_->atomData.skippedCharge.size();
612 +      vector<RealType> skch_tmp(ns, 0.0);
613 +
614 +      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
615 +      for (int i = 0; i < ns; i++) {
616 +        snap_->atomData.skippedCharge[i] = skch_tmp[i];
617 +        skch_tmp[i] = 0.0;
618 +      }
619 +      
620 +      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
621 +      for (int i = 0; i < ns; i++)
622 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
623      }
624      
625      nLocal_ = snap_->getNumberOfAtoms();
# Line 606 | Line 632 | namespace OpenMD {
632      AtomCommPotRow->scatter(pot_row, pot_temp);
633  
634      for (int ii = 0;  ii < pot_temp.size(); ii++ )
635 <      pot_local += pot_temp[ii];
635 >      pairwisePot += pot_temp[ii];
636      
637      fill(pot_temp.begin(), pot_temp.end(),
638           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
# Line 614 | Line 640 | namespace OpenMD {
640      AtomCommPotColumn->scatter(pot_col, pot_temp);    
641      
642      for (int ii = 0;  ii < pot_temp.size(); ii++ )
643 <      pot_local += pot_temp[ii];
618 <    
643 >      pairwisePot += pot_temp[ii];    
644   #endif
645 +
646    }
647  
648    int ForceMatrixDecomposition::getNAtomsInRow() {  
# Line 717 | Line 743 | namespace OpenMD {
743      return d;    
744    }
745  
746 <  vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
747 <    return skipsForAtom[atom1];
746 >  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
747 >    return excludesForAtom[atom1];
748    }
749  
750    /**
751 <   * There are a number of reasons to skip a pair or a
726 <   * particle. Mostly we do this to exclude atoms who are involved in
727 <   * short range interactions (bonds, bends, torsions), but we also
728 <   * need to exclude some overcounted interactions that result from
751 >   * We need to exclude some overcounted interactions that result from
752     * the parallel decomposition.
753     */
754    bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
# Line 745 | Line 768 | namespace OpenMD {
768      } else {
769        if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
770      }
771 + #endif
772 +    return false;
773 +  }
774 +
775 +  /**
776 +   * We need to handle the interactions for atoms who are involved in
777 +   * the same rigid body as well as some short range interactions
778 +   * (bonds, bends, torsions) differently from other interactions.
779 +   * We'll still visit the pairwise routines, but with a flag that
780 +   * tells those routines to exclude the pair from direct long range
781 +   * interactions.  Some indirect interactions (notably reaction
782 +   * field) must still be handled for these pairs.
783 +   */
784 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
785 +    int unique_id_2;
786 +    
787 + #ifdef IS_MPI
788 +    // in MPI, we have to look up the unique IDs for the row atom.
789 +    unique_id_2 = AtomColToGlobal[atom2];
790   #else
791      // in the normal loop, the atom numbers are unique
750 –    unique_id_1 = atom1;
792      unique_id_2 = atom2;
793   #endif
794      
795 <    for (vector<int>::iterator i = skipsForAtom[atom1].begin();
796 <         i != skipsForAtom[atom1].end(); ++i) {
795 >    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
796 >         i != excludesForAtom[atom1].end(); ++i) {
797        if ( (*i) == unique_id_2 ) return true;
798 <    }    
798 >    }
799  
800 +    return false;
801    }
802  
803  
# Line 776 | Line 818 | namespace OpenMD {
818    }
819  
820      // filling interaction blocks with pointers
821 <  void ForceMatrixDecomposition::fillInteractionData(InteractionData idat,
822 <                                                     int atom1, int atom2) {    
821 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
822 >                                                     int atom1, int atom2) {
823 >
824 >    idat.excluded = excludeAtomPair(atom1, atom2);
825 >  
826   #ifdef IS_MPI
827      
828      idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
# Line 818 | Line 863 | namespace OpenMD {
863        idat.particlePot2 = &(atomColData.particlePot[atom2]);
864      }
865  
866 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
867 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
868 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
869 +    }
870 +
871   #else
872  
873 <    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
874 <                             ff_->getAtomType(identsLocal[atom2]) );
873 >    idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
874 >                             ff_->getAtomType(idents[atom2]) );
875  
876      if (storageLayout_ & DataStorage::dslAmat) {
877        idat.A1 = &(snap_->atomData.aMat[atom1]);
# Line 838 | Line 888 | namespace OpenMD {
888        idat.t2 = &(snap_->atomData.torque[atom2]);
889      }
890  
891 <    if (storageLayout_ & DataStorage::dslDensity) {
891 >    if (storageLayout_ & DataStorage::dslDensity) {    
892        idat.rho1 = &(snap_->atomData.density[atom1]);
893        idat.rho2 = &(snap_->atomData.density[atom2]);
894      }
# Line 858 | Line 908 | namespace OpenMD {
908        idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
909      }
910  
911 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
912 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
913 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
914 +    }
915   #endif
916    }
917  
918    
919 <  void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {    
919 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
920   #ifdef IS_MPI
921      pot_row[atom1] += 0.5 *  *(idat.pot);
922      pot_col[atom2] += 0.5 *  *(idat.pot);
# Line 870 | Line 924 | namespace OpenMD {
924      atomRowData.force[atom1] += *(idat.f1);
925      atomColData.force[atom2] -= *(idat.f1);
926   #else
927 <    longRangePot_ += *(idat.pot);
928 <    
927 >    pairwisePot += *(idat.pot);
928 >
929      snap_->atomData.force[atom1] += *(idat.f1);
930      snap_->atomData.force[atom2] -= *(idat.f1);
931   #endif
932 <
932 >    
933    }
934  
881 –
882 –  void ForceMatrixDecomposition::fillSkipData(InteractionData idat,
883 –                                              int atom1, int atom2) {
884 – #ifdef IS_MPI
885 –    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
886 –                             ff_->getAtomType(identsCol[atom2]) );
887 –
888 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
889 –      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
890 –      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
891 –    }
892 –    if (storageLayout_ & DataStorage::dslTorque) {
893 –      idat.t1 = &(atomRowData.torque[atom1]);
894 –      idat.t2 = &(atomColData.torque[atom2]);
895 –    }
896 – #else
897 –    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
898 –                             ff_->getAtomType(identsLocal[atom2]) );
899 –
900 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
901 –      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
902 –      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
903 –    }
904 –    if (storageLayout_ & DataStorage::dslTorque) {
905 –      idat.t1 = &(snap_->atomData.torque[atom1]);
906 –      idat.t2 = &(snap_->atomData.torque[atom2]);
907 –    }
908 – #endif    
909 –  }
910 –
935    /*
936     * buildNeighborList
937     *
# Line 918 | Line 942 | namespace OpenMD {
942        
943      vector<pair<int, int> > neighborList;
944      groupCutoffs cuts;
945 +    bool doAllPairs = false;
946 +
947   #ifdef IS_MPI
948      cellListRow_.clear();
949      cellListCol_.clear();
# Line 937 | Line 963 | namespace OpenMD {
963      nCells_.y() = (int) ( Hy.length() )/ rList_;
964      nCells_.z() = (int) ( Hz.length() )/ rList_;
965  
966 +    // handle small boxes where the cell offsets can end up repeating cells
967 +    
968 +    if (nCells_.x() < 3) doAllPairs = true;
969 +    if (nCells_.y() < 3) doAllPairs = true;
970 +    if (nCells_.z() < 3) doAllPairs = true;
971 +
972      Mat3x3d invHmat = snap_->getInvHmat();
973      Vector3d rs, scaled, dr;
974      Vector3i whichCell;
975      int cellIndex;
976      int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
977  
946 –    cerr << "flag1\n";
978   #ifdef IS_MPI
979      cellListRow_.resize(nCtot);
980      cellListCol_.resize(nCtot);
981   #else
982      cellList_.resize(nCtot);
983   #endif
984 <    cerr << "flag2\n";
984 >
985 >    if (!doAllPairs) {
986   #ifdef IS_MPI
955 –    for (int i = 0; i < nGroupsInRow_; i++) {
956 –      rs = cgRowData.position[i];
987  
988 <      // scaled positions relative to the box vectors
989 <      scaled = invHmat * rs;
990 <
991 <      // wrap the vector back into the unit box by subtracting integer box
992 <      // numbers
993 <      for (int j = 0; j < 3; j++) {
994 <        scaled[j] -= roundMe(scaled[j]);
995 <        scaled[j] += 0.5;
996 <      }
997 <    
998 <      // find xyz-indices of cell that cutoffGroup is in.
999 <      whichCell.x() = nCells_.x() * scaled.x();
1000 <      whichCell.y() = nCells_.y() * scaled.y();
1001 <      whichCell.z() = nCells_.z() * scaled.z();
1002 <
1003 <      // find single index of this cell:
1004 <      cellIndex = Vlinear(whichCell, nCells_);
1005 <
1006 <      // add this cutoff group to the list of groups in this cell;
1007 <      cellListRow_[cellIndex].push_back(i);
1008 <    }
1009 <
1010 <    for (int i = 0; i < nGroupsInCol_; i++) {
981 <      rs = cgColData.position[i];
982 <
983 <      // scaled positions relative to the box vectors
984 <      scaled = invHmat * rs;
985 <
986 <      // wrap the vector back into the unit box by subtracting integer box
987 <      // numbers
988 <      for (int j = 0; j < 3; j++) {
989 <        scaled[j] -= roundMe(scaled[j]);
990 <        scaled[j] += 0.5;
988 >      for (int i = 0; i < nGroupsInRow_; i++) {
989 >        rs = cgRowData.position[i];
990 >        
991 >        // scaled positions relative to the box vectors
992 >        scaled = invHmat * rs;
993 >        
994 >        // wrap the vector back into the unit box by subtracting integer box
995 >        // numbers
996 >        for (int j = 0; j < 3; j++) {
997 >          scaled[j] -= roundMe(scaled[j]);
998 >          scaled[j] += 0.5;
999 >        }
1000 >        
1001 >        // find xyz-indices of cell that cutoffGroup is in.
1002 >        whichCell.x() = nCells_.x() * scaled.x();
1003 >        whichCell.y() = nCells_.y() * scaled.y();
1004 >        whichCell.z() = nCells_.z() * scaled.z();
1005 >        
1006 >        // find single index of this cell:
1007 >        cellIndex = Vlinear(whichCell, nCells_);
1008 >        
1009 >        // add this cutoff group to the list of groups in this cell;
1010 >        cellListRow_[cellIndex].push_back(i);
1011        }
1012 <
1013 <      // find xyz-indices of cell that cutoffGroup is in.
1014 <      whichCell.x() = nCells_.x() * scaled.x();
1015 <      whichCell.y() = nCells_.y() * scaled.y();
1016 <      whichCell.z() = nCells_.z() * scaled.z();
1017 <
1018 <      // find single index of this cell:
1019 <      cellIndex = Vlinear(whichCell, nCells_);
1020 <
1021 <      // add this cutoff group to the list of groups in this cell;
1022 <      cellListCol_[cellIndex].push_back(i);
1023 <    }
1012 >      
1013 >      for (int i = 0; i < nGroupsInCol_; i++) {
1014 >        rs = cgColData.position[i];
1015 >        
1016 >        // scaled positions relative to the box vectors
1017 >        scaled = invHmat * rs;
1018 >        
1019 >        // wrap the vector back into the unit box by subtracting integer box
1020 >        // numbers
1021 >        for (int j = 0; j < 3; j++) {
1022 >          scaled[j] -= roundMe(scaled[j]);
1023 >          scaled[j] += 0.5;
1024 >        }
1025 >        
1026 >        // find xyz-indices of cell that cutoffGroup is in.
1027 >        whichCell.x() = nCells_.x() * scaled.x();
1028 >        whichCell.y() = nCells_.y() * scaled.y();
1029 >        whichCell.z() = nCells_.z() * scaled.z();
1030 >        
1031 >        // find single index of this cell:
1032 >        cellIndex = Vlinear(whichCell, nCells_);
1033 >        
1034 >        // add this cutoff group to the list of groups in this cell;
1035 >        cellListCol_[cellIndex].push_back(i);
1036 >      }
1037   #else
1038 <    for (int i = 0; i < nGroups_; i++) {
1039 <      rs = snap_->cgData.position[i];
1040 <
1041 <      // scaled positions relative to the box vectors
1042 <      scaled = invHmat * rs;
1043 <
1044 <      // wrap the vector back into the unit box by subtracting integer box
1045 <      // numbers
1046 <      for (int j = 0; j < 3; j++) {
1047 <        scaled[j] -= roundMe(scaled[j]);
1048 <        scaled[j] += 0.5;
1038 >      for (int i = 0; i < nGroups_; i++) {
1039 >        rs = snap_->cgData.position[i];
1040 >        
1041 >        // scaled positions relative to the box vectors
1042 >        scaled = invHmat * rs;
1043 >        
1044 >        // wrap the vector back into the unit box by subtracting integer box
1045 >        // numbers
1046 >        for (int j = 0; j < 3; j++) {
1047 >          scaled[j] -= roundMe(scaled[j]);
1048 >          scaled[j] += 0.5;
1049 >        }
1050 >        
1051 >        // find xyz-indices of cell that cutoffGroup is in.
1052 >        whichCell.x() = nCells_.x() * scaled.x();
1053 >        whichCell.y() = nCells_.y() * scaled.y();
1054 >        whichCell.z() = nCells_.z() * scaled.z();
1055 >        
1056 >        // find single index of this cell:
1057 >        cellIndex = Vlinear(whichCell, nCells_);      
1058 >        
1059 >        // add this cutoff group to the list of groups in this cell;
1060 >        cellList_[cellIndex].push_back(i);
1061        }
1017 –
1018 –      // find xyz-indices of cell that cutoffGroup is in.
1019 –      whichCell.x() = nCells_.x() * scaled.x();
1020 –      whichCell.y() = nCells_.y() * scaled.y();
1021 –      whichCell.z() = nCells_.z() * scaled.z();
1022 –
1023 –      // find single index of this cell:
1024 –      cellIndex = Vlinear(whichCell, nCells_);      
1025 –
1026 –      // add this cutoff group to the list of groups in this cell;
1027 –      cellList_[cellIndex].push_back(i);
1028 –    }
1062   #endif
1063  
1064 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1065 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1066 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1067 <          Vector3i m1v(m1x, m1y, m1z);
1068 <          int m1 = Vlinear(m1v, nCells_);
1036 <
1037 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1038 <               os != cellOffsets_.end(); ++os) {
1064 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1065 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1066 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1067 >            Vector3i m1v(m1x, m1y, m1z);
1068 >            int m1 = Vlinear(m1v, nCells_);
1069              
1070 <            Vector3i m2v = m1v + (*os);
1071 <            
1072 <            if (m2v.x() >= nCells_.x()) {
1073 <              m2v.x() = 0;          
1074 <            } else if (m2v.x() < 0) {
1075 <              m2v.x() = nCells_.x() - 1;
1076 <            }
1077 <            
1078 <            if (m2v.y() >= nCells_.y()) {
1079 <              m2v.y() = 0;          
1080 <            } else if (m2v.y() < 0) {
1081 <              m2v.y() = nCells_.y() - 1;
1082 <            }
1083 <            
1084 <            if (m2v.z() >= nCells_.z()) {
1085 <              m2v.z() = 0;          
1086 <            } else if (m2v.z() < 0) {
1087 <              m2v.z() = nCells_.z() - 1;
1088 <            }
1089 <            
1090 <            int m2 = Vlinear (m2v, nCells_);
1091 <
1070 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1071 >                 os != cellOffsets_.end(); ++os) {
1072 >              
1073 >              Vector3i m2v = m1v + (*os);
1074 >              
1075 >              if (m2v.x() >= nCells_.x()) {
1076 >                m2v.x() = 0;          
1077 >              } else if (m2v.x() < 0) {
1078 >                m2v.x() = nCells_.x() - 1;
1079 >              }
1080 >              
1081 >              if (m2v.y() >= nCells_.y()) {
1082 >                m2v.y() = 0;          
1083 >              } else if (m2v.y() < 0) {
1084 >                m2v.y() = nCells_.y() - 1;
1085 >              }
1086 >              
1087 >              if (m2v.z() >= nCells_.z()) {
1088 >                m2v.z() = 0;          
1089 >              } else if (m2v.z() < 0) {
1090 >                m2v.z() = nCells_.z() - 1;
1091 >              }
1092 >              
1093 >              int m2 = Vlinear (m2v, nCells_);
1094 >              
1095   #ifdef IS_MPI
1096 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1097 <                 j1 != cellListRow_[m1].end(); ++j1) {
1098 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1099 <                   j2 != cellListCol_[m2].end(); ++j2) {
1100 <                              
1101 <                // Always do this if we're in different cells or if
1102 <                // we're in the same cell and the global index of the
1103 <                // j2 cutoff group is less than the j1 cutoff group
1104 <
1105 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1106 <                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1107 <                  snap_->wrapVector(dr);
1108 <                  cuts = getGroupCutoffs( (*j1), (*j2) );
1109 <                  if (dr.lengthSquare() < cuts.third) {
1110 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1096 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1097 >                   j1 != cellListRow_[m1].end(); ++j1) {
1098 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1099 >                     j2 != cellListCol_[m2].end(); ++j2) {
1100 >                  
1101 >                  // Always do this if we're in different cells or if
1102 >                  // we're in the same cell and the global index of the
1103 >                  // j2 cutoff group is less than the j1 cutoff group
1104 >                  
1105 >                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1106 >                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1107 >                    snap_->wrapVector(dr);
1108 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1109 >                    if (dr.lengthSquare() < cuts.third) {
1110 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1111 >                    }
1112                    }
1113                  }
1114                }
1081 –            }
1115   #else
1116 <
1117 <            for (vector<int>::iterator j1 = cellList_[m1].begin();
1118 <                 j1 != cellList_[m1].end(); ++j1) {
1119 <              for (vector<int>::iterator j2 = cellList_[m2].begin();
1120 <                   j2 != cellList_[m2].end(); ++j2) {
1121 <
1122 <                // Always do this if we're in different cells or if
1123 <                // we're in the same cell and the global index of the
1124 <                // j2 cutoff group is less than the j1 cutoff group
1125 <
1126 <                if (m2 != m1 || (*j2) < (*j1)) {
1127 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1128 <                  snap_->wrapVector(dr);
1129 <                  cuts = getGroupCutoffs( (*j1), (*j2) );
1130 <                  if (dr.lengthSquare() < cuts.third) {
1131 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1116 >              
1117 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1118 >                   j1 != cellList_[m1].end(); ++j1) {
1119 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1120 >                     j2 != cellList_[m2].end(); ++j2) {
1121 >                  
1122 >                  // Always do this if we're in different cells or if
1123 >                  // we're in the same cell and the global index of the
1124 >                  // j2 cutoff group is less than the j1 cutoff group
1125 >                  
1126 >                  if (m2 != m1 || (*j2) < (*j1)) {
1127 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1128 >                    snap_->wrapVector(dr);
1129 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1130 >                    if (dr.lengthSquare() < cuts.third) {
1131 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1132 >                    }
1133                    }
1134                  }
1135                }
1102 –            }
1136   #endif
1137 +            }
1138            }
1139          }
1140        }
1141 +    } else {
1142 +      // branch to do all cutoff group pairs
1143 + #ifdef IS_MPI
1144 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1145 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1146 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1147 +          snap_->wrapVector(dr);
1148 +          cuts = getGroupCutoffs( j1, j2 );
1149 +          if (dr.lengthSquare() < cuts.third) {
1150 +            neighborList.push_back(make_pair(j1, j2));
1151 +          }
1152 +        }
1153 +      }
1154 + #else
1155 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1156 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1157 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1158 +          snap_->wrapVector(dr);
1159 +          cuts = getGroupCutoffs( j1, j2 );
1160 +          if (dr.lengthSquare() < cuts.third) {
1161 +            neighborList.push_back(make_pair(j1, j2));
1162 +          }
1163 +        }
1164 +      }        
1165 + #endif
1166      }
1167 <    
1167 >      
1168      // save the local cutoff group positions for the check that is
1169      // done on each loop:
1170      saved_CG_positions_.clear();

Diff Legend

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