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 1582 by gezelter, Tue Jun 14 20:41:44 2011 UTC vs.
Revision 1591 by gezelter, Tue Jul 12 15:25:07 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 +    // allocate memory for the parallel objects
116 +    atypesRow.resize(nAtomsInRow_);
117 +    atypesCol.resize(nAtomsInCol_);
118 +
119 +    for (int i = 0; i < nAtomsInRow_; i++)
120 +      atypesRow[i] = ff_->getAtomType(identsRow[i]);
121 +    for (int i = 0; i < nAtomsInCol_; i++)
122 +      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
123 +
124 +    pot_row.resize(nAtomsInRow_);
125 +    pot_col.resize(nAtomsInCol_);
126 +
127 +    AtomRowToGlobal.resize(nAtomsInRow_);
128 +    AtomColToGlobal.resize(nAtomsInCol_);
129      AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
130      AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
131      
132 +    cgRowToGlobal.resize(nGroupsInRow_);
133 +    cgColToGlobal.resize(nGroupsInCol_);
134      cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
135      cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
136  
137 +    massFactorsRow.resize(nAtomsInRow_);
138 +    massFactorsCol.resize(nAtomsInCol_);
139      AtomCommRealRow->gather(massFactors, massFactorsRow);
140      AtomCommRealColumn->gather(massFactors, massFactorsCol);
141  
# Line 142 | Line 161 | namespace OpenMD {
161        }      
162      }
163  
164 <    skipsForAtom.clear();
165 <    skipsForAtom.resize(nAtomsInRow_);
164 >    excludesForAtom.clear();
165 >    excludesForAtom.resize(nAtomsInRow_);
166      toposForAtom.clear();
167      toposForAtom.resize(nAtomsInRow_);
168      topoDist.clear();
# Line 154 | Line 173 | namespace OpenMD {
173        for (int j = 0; j < nAtomsInCol_; j++) {
174          int jglob = AtomColToGlobal[j];
175  
176 <        if (excludes.hasPair(iglob, jglob))
177 <          skipsForAtom[i].push_back(j);      
176 >        if (excludes->hasPair(iglob, jglob))
177 >          excludesForAtom[i].push_back(j);      
178          
179 <        if (oneTwo.hasPair(iglob, jglob)) {
179 >        if (oneTwo->hasPair(iglob, jglob)) {
180            toposForAtom[i].push_back(j);
181            topoDist[i].push_back(1);
182          } else {
183 <          if (oneThree.hasPair(iglob, jglob)) {
183 >          if (oneThree->hasPair(iglob, jglob)) {
184              toposForAtom[i].push_back(j);
185              topoDist[i].push_back(2);
186            } else {
187 <            if (oneFour.hasPair(iglob, jglob)) {
187 >            if (oneFour->hasPair(iglob, jglob)) {
188                toposForAtom[i].push_back(j);
189                topoDist[i].push_back(3);
190              }
# Line 176 | Line 195 | namespace OpenMD {
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++) {
# Line 188 | Line 213 | namespace OpenMD {
213        }      
214      }
215  
216 <    skipsForAtom.clear();
217 <    skipsForAtom.resize(nLocal_);
216 >    excludesForAtom.clear();
217 >    excludesForAtom.resize(nLocal_);
218      toposForAtom.clear();
219      toposForAtom.resize(nLocal_);
220      topoDist.clear();
# Line 201 | Line 226 | namespace OpenMD {
226        for (int j = 0; j < nLocal_; j++) {
227          int jglob = AtomLocalToGlobal[j];
228  
229 <        if (excludes.hasPair(iglob, jglob))
230 <          skipsForAtom[i].push_back(j);              
229 >        if (excludes->hasPair(iglob, jglob))
230 >          excludesForAtom[i].push_back(j);              
231          
232 <        if (oneTwo.hasPair(iglob, jglob)) {
232 >        if (oneTwo->hasPair(iglob, jglob)) {
233            toposForAtom[i].push_back(j);
234            topoDist[i].push_back(1);
235          } else {
236 <          if (oneThree.hasPair(iglob, jglob)) {
236 >          if (oneThree->hasPair(iglob, jglob)) {
237              toposForAtom[i].push_back(j);
238              topoDist[i].push_back(2);
239            } else {
240 <            if (oneFour.hasPair(iglob, jglob)) {
240 >            if (oneFour->hasPair(iglob, jglob)) {
241                toposForAtom[i].push_back(j);
242                topoDist[i].push_back(3);
243              }
# Line 222 | Line 247 | namespace OpenMD {
247      }
248      
249      createGtypeCutoffMap();
250 +
251    }
252    
253    void ForceMatrixDecomposition::createGtypeCutoffMap() {
254 <
254 >    
255      RealType tol = 1e-6;
256      RealType rc;
257      int atid;
258      set<AtomType*> atypes = info_->getSimulatedAtomTypes();
259 <    vector<RealType> atypeCutoff;
260 <    atypeCutoff.resize( atypes.size() );
235 <
259 >    map<int, RealType> atypeCutoff;
260 >      
261      for (set<AtomType*>::iterator at = atypes.begin();
262           at != atypes.end(); ++at){
238 –      rc = interactionMan_->getSuggestedCutoffRadius(*at);
263        atid = (*at)->getIdent();
264 <      atypeCutoff[atid] = rc;
264 >      if (userChoseCutoff_)
265 >        atypeCutoff[atid] = userCutoff_;
266 >      else
267 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
268      }
269  
270      vector<RealType> gTypeCutoffs;
244 –
271      // first we do a single loop over the cutoff groups to find the
272      // largest cutoff for any atypes present in this group.
273   #ifdef IS_MPI
# Line 299 | Line 325 | namespace OpenMD {
325  
326      vector<RealType> groupCutoff(nGroups_, 0.0);
327      groupToGtype.resize(nGroups_);
302 –
303 –    cerr << "nGroups = " << nGroups_ << "\n";
328      for (int cg1 = 0; cg1 < nGroups_; cg1++) {
329  
330        groupCutoff[cg1] = 0.0;
# Line 309 | Line 333 | namespace OpenMD {
333        for (vector<int>::iterator ia = atomList.begin();
334             ia != atomList.end(); ++ia) {            
335          int atom1 = (*ia);
336 <        atid = identsLocal[atom1];
336 >        atid = idents[atom1];
337          if (atypeCutoff[atid] > groupCutoff[cg1]) {
338            groupCutoff[cg1] = atypeCutoff[atid];
339          }
# Line 329 | Line 353 | namespace OpenMD {
353      }
354   #endif
355  
332 –    cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
356      // Now we find the maximum group cutoff value present in the simulation
357  
358 <    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
358 >    RealType groupMax = *max_element(gTypeCutoffs.begin(),
359 >                                     gTypeCutoffs.end());
360  
361   #ifdef IS_MPI
362 <    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
362 >    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
363 >                              MPI::MAX);
364   #endif
365      
366      RealType tradRcut = groupMax;
# Line 378 | Line 403 | namespace OpenMD {
403            if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
404              sprintf(painCave.errMsg,
405                      "ForceMatrixDecomposition::createGtypeCutoffMap "
406 <                    "user-specified rCut does not match computed group Cutoff\n");
406 >                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
407              painCave.severity = OPENMD_ERROR;
408              painCave.isFatal = 1;
409              simError();            
# Line 410 | Line 435 | namespace OpenMD {
435    }
436  
437    void ForceMatrixDecomposition::zeroWorkArrays() {
438 +    pairwisePot = 0.0;
439 +    embeddingPot = 0.0;
440  
414 –    for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
415 –      longRangePot_[j] = 0.0;
416 –    }
417 –
441   #ifdef IS_MPI
442      if (storageLayout_ & DataStorage::dslForce) {
443        fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
# Line 430 | Line 453 | namespace OpenMD {
453           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
454  
455      fill(pot_col.begin(), pot_col.end(),
456 <         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
434 <    
435 <    pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
456 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
457  
458      if (storageLayout_ & DataStorage::dslParticlePot) {    
459 <      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
460 <      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
459 >      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
460 >           0.0);
461 >      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
462 >           0.0);
463      }
464  
465      if (storageLayout_ & DataStorage::dslDensity) {      
# Line 445 | Line 468 | namespace OpenMD {
468      }
469  
470      if (storageLayout_ & DataStorage::dslFunctional) {  
471 <      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
472 <      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
471 >      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
472 >           0.0);
473 >      fill(atomColData.functional.begin(), atomColData.functional.end(),
474 >           0.0);
475      }
476  
477      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
# Line 456 | Line 481 | namespace OpenMD {
481             atomColData.functionalDerivative.end(), 0.0);
482      }
483  
484 < #else
485 <    
484 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
485 >      fill(atomRowData.skippedCharge.begin(),
486 >           atomRowData.skippedCharge.end(), 0.0);
487 >      fill(atomColData.skippedCharge.begin(),
488 >           atomColData.skippedCharge.end(), 0.0);
489 >    }
490 >
491 > #endif
492 >    // even in parallel, we need to zero out the local arrays:
493 >
494      if (storageLayout_ & DataStorage::dslParticlePot) {      
495        fill(snap_->atomData.particlePot.begin(),
496             snap_->atomData.particlePot.end(), 0.0);
# Line 475 | Line 508 | namespace OpenMD {
508        fill(snap_->atomData.functionalDerivative.begin(),
509             snap_->atomData.functionalDerivative.end(), 0.0);
510      }
511 < #endif
511 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
512 >      fill(snap_->atomData.skippedCharge.begin(),
513 >           snap_->atomData.skippedCharge.end(), 0.0);
514 >    }
515      
516    }
517  
# Line 512 | Line 548 | namespace OpenMD {
548        AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
549                                     atomColData.electroFrame);
550      }
551 +
552   #endif      
553    }
554    
# Line 578 | Line 615 | namespace OpenMD {
615      AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
616      for (int i = 0; i < n; i++)
617        snap_->atomData.force[i] += frc_tmp[i];
618 <    
582 <    
618 >        
619      if (storageLayout_ & DataStorage::dslTorque) {
620  
621 <      int nt = snap_->atomData.force.size();
621 >      int nt = snap_->atomData.torque.size();
622        vector<Vector3d> trq_tmp(nt, V3Zero);
623  
624        AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
625 <      for (int i = 0; i < n; i++) {
625 >      for (int i = 0; i < nt; i++) {
626          snap_->atomData.torque[i] += trq_tmp[i];
627          trq_tmp[i] = 0.0;
628        }
629        
630        AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
631 <      for (int i = 0; i < n; i++)
631 >      for (int i = 0; i < nt; i++)
632          snap_->atomData.torque[i] += trq_tmp[i];
633 +    }
634 +
635 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
636 +
637 +      int ns = snap_->atomData.skippedCharge.size();
638 +      vector<RealType> skch_tmp(ns, 0.0);
639 +
640 +      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
641 +      for (int i = 0; i < ns; i++) {
642 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
643 +        skch_tmp[i] = 0.0;
644 +      }
645 +      
646 +      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
647 +      for (int i = 0; i < ns; i++)
648 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
649      }
650      
651      nLocal_ = snap_->getNumberOfAtoms();
# Line 606 | Line 658 | namespace OpenMD {
658      AtomCommPotRow->scatter(pot_row, pot_temp);
659  
660      for (int ii = 0;  ii < pot_temp.size(); ii++ )
661 <      pot_local += pot_temp[ii];
661 >      pairwisePot += pot_temp[ii];
662      
663      fill(pot_temp.begin(), pot_temp.end(),
664           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
# Line 614 | Line 666 | namespace OpenMD {
666      AtomCommPotColumn->scatter(pot_col, pot_temp);    
667      
668      for (int ii = 0;  ii < pot_temp.size(); ii++ )
669 <      pot_local += pot_temp[ii];
618 <    
669 >      pairwisePot += pot_temp[ii];    
670   #endif
671 +
672    }
673  
674    int ForceMatrixDecomposition::getNAtomsInRow() {  
# Line 717 | Line 769 | namespace OpenMD {
769      return d;    
770    }
771  
772 <  vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
773 <    return skipsForAtom[atom1];
772 >  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
773 >    return excludesForAtom[atom1];
774    }
775  
776    /**
777 <   * 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
777 >   * We need to exclude some overcounted interactions that result from
778     * the parallel decomposition.
779     */
780    bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
# Line 745 | Line 794 | namespace OpenMD {
794      } else {
795        if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
796      }
797 + #endif
798 +    return false;
799 +  }
800 +
801 +  /**
802 +   * We need to handle the interactions for atoms who are involved in
803 +   * the same rigid body as well as some short range interactions
804 +   * (bonds, bends, torsions) differently from other interactions.
805 +   * We'll still visit the pairwise routines, but with a flag that
806 +   * tells those routines to exclude the pair from direct long range
807 +   * interactions.  Some indirect interactions (notably reaction
808 +   * field) must still be handled for these pairs.
809 +   */
810 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
811 +    int unique_id_2;
812 +    
813 + #ifdef IS_MPI
814 +    // in MPI, we have to look up the unique IDs for the row atom.
815 +    unique_id_2 = AtomColToGlobal[atom2];
816   #else
817      // in the normal loop, the atom numbers are unique
750 –    unique_id_1 = atom1;
818      unique_id_2 = atom2;
819   #endif
820      
821 <    for (vector<int>::iterator i = skipsForAtom[atom1].begin();
822 <         i != skipsForAtom[atom1].end(); ++i) {
821 >    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
822 >         i != excludesForAtom[atom1].end(); ++i) {
823        if ( (*i) == unique_id_2 ) return true;
824 <    }    
824 >    }
825  
826 +    return false;
827    }
828  
829  
# Line 777 | Line 845 | namespace OpenMD {
845  
846      // filling interaction blocks with pointers
847    void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
848 <                                                     int atom1, int atom2) {    
848 >                                                     int atom1, int atom2) {
849 >
850 >    idat.excluded = excludeAtomPair(atom1, atom2);
851 >  
852   #ifdef IS_MPI
853 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
854 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
855 +    //                         ff_->getAtomType(identsCol[atom2]) );
856      
783 –    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
784 –                             ff_->getAtomType(identsCol[atom2]) );
785 –    
857      if (storageLayout_ & DataStorage::dslAmat) {
858        idat.A1 = &(atomRowData.aMat[atom1]);
859        idat.A2 = &(atomColData.aMat[atom2]);
# Line 818 | Line 889 | namespace OpenMD {
889        idat.particlePot2 = &(atomColData.particlePot[atom2]);
890      }
891  
892 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
893 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
894 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
895 +    }
896 +
897   #else
898  
899 <    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
900 <                             ff_->getAtomType(identsLocal[atom2]) );
899 >    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
900 >    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
901 >    //                         ff_->getAtomType(idents[atom2]) );
902  
903      if (storageLayout_ & DataStorage::dslAmat) {
904        idat.A1 = &(snap_->atomData.aMat[atom1]);
# Line 838 | Line 915 | namespace OpenMD {
915        idat.t2 = &(snap_->atomData.torque[atom2]);
916      }
917  
918 <    if (storageLayout_ & DataStorage::dslDensity) {
918 >    if (storageLayout_ & DataStorage::dslDensity) {    
919        idat.rho1 = &(snap_->atomData.density[atom1]);
920        idat.rho2 = &(snap_->atomData.density[atom2]);
921      }
# Line 858 | Line 935 | namespace OpenMD {
935        idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
936      }
937  
938 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
939 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
940 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
941 +    }
942   #endif
943    }
944  
# Line 870 | Line 951 | namespace OpenMD {
951      atomRowData.force[atom1] += *(idat.f1);
952      atomColData.force[atom2] -= *(idat.f1);
953   #else
954 <    longRangePot_ += *(idat.pot);
955 <    
954 >    pairwisePot += *(idat.pot);
955 >
956      snap_->atomData.force[atom1] += *(idat.f1);
957      snap_->atomData.force[atom2] -= *(idat.f1);
958   #endif
959 <
959 >    
960    }
961  
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 –
962    /*
963     * buildNeighborList
964     *
# Line 918 | Line 969 | namespace OpenMD {
969        
970      vector<pair<int, int> > neighborList;
971      groupCutoffs cuts;
972 +    bool doAllPairs = false;
973 +
974   #ifdef IS_MPI
975      cellListRow_.clear();
976      cellListCol_.clear();
# Line 937 | Line 990 | namespace OpenMD {
990      nCells_.y() = (int) ( Hy.length() )/ rList_;
991      nCells_.z() = (int) ( Hz.length() )/ rList_;
992  
993 +    // handle small boxes where the cell offsets can end up repeating cells
994 +    
995 +    if (nCells_.x() < 3) doAllPairs = true;
996 +    if (nCells_.y() < 3) doAllPairs = true;
997 +    if (nCells_.z() < 3) doAllPairs = true;
998 +
999      Mat3x3d invHmat = snap_->getInvHmat();
1000      Vector3d rs, scaled, dr;
1001      Vector3i whichCell;
# Line 950 | Line 1009 | namespace OpenMD {
1009      cellList_.resize(nCtot);
1010   #endif
1011  
1012 +    if (!doAllPairs) {
1013   #ifdef IS_MPI
954 –    for (int i = 0; i < nGroupsInRow_; i++) {
955 –      rs = cgRowData.position[i];
1014  
1015 <      // scaled positions relative to the box vectors
1016 <      scaled = invHmat * rs;
1017 <
1018 <      // wrap the vector back into the unit box by subtracting integer box
1019 <      // numbers
1020 <      for (int j = 0; j < 3; j++) {
1021 <        scaled[j] -= roundMe(scaled[j]);
1022 <        scaled[j] += 0.5;
1015 >      for (int i = 0; i < nGroupsInRow_; i++) {
1016 >        rs = cgRowData.position[i];
1017 >        
1018 >        // scaled positions relative to the box vectors
1019 >        scaled = invHmat * rs;
1020 >        
1021 >        // wrap the vector back into the unit box by subtracting integer box
1022 >        // numbers
1023 >        for (int j = 0; j < 3; j++) {
1024 >          scaled[j] -= roundMe(scaled[j]);
1025 >          scaled[j] += 0.5;
1026 >        }
1027 >        
1028 >        // find xyz-indices of cell that cutoffGroup is in.
1029 >        whichCell.x() = nCells_.x() * scaled.x();
1030 >        whichCell.y() = nCells_.y() * scaled.y();
1031 >        whichCell.z() = nCells_.z() * scaled.z();
1032 >        
1033 >        // find single index of this cell:
1034 >        cellIndex = Vlinear(whichCell, nCells_);
1035 >        
1036 >        // add this cutoff group to the list of groups in this cell;
1037 >        cellListRow_[cellIndex].push_back(i);
1038        }
1039 <    
1040 <      // find xyz-indices of cell that cutoffGroup is in.
1041 <      whichCell.x() = nCells_.x() * scaled.x();
1042 <      whichCell.y() = nCells_.y() * scaled.y();
1043 <      whichCell.z() = nCells_.z() * scaled.z();
1044 <
1045 <      // find single index of this cell:
1046 <      cellIndex = Vlinear(whichCell, nCells_);
1047 <
1048 <      // add this cutoff group to the list of groups in this cell;
1049 <      cellListRow_[cellIndex].push_back(i);
1050 <    }
1051 <
1052 <    for (int i = 0; i < nGroupsInCol_; i++) {
1053 <      rs = cgColData.position[i];
1054 <
1055 <      // scaled positions relative to the box vectors
1056 <      scaled = invHmat * rs;
1057 <
1058 <      // wrap the vector back into the unit box by subtracting integer box
1059 <      // numbers
1060 <      for (int j = 0; j < 3; j++) {
1061 <        scaled[j] -= roundMe(scaled[j]);
1062 <        scaled[j] += 0.5;
1039 >      
1040 >      for (int i = 0; i < nGroupsInCol_; i++) {
1041 >        rs = cgColData.position[i];
1042 >        
1043 >        // scaled positions relative to the box vectors
1044 >        scaled = invHmat * rs;
1045 >        
1046 >        // wrap the vector back into the unit box by subtracting integer box
1047 >        // numbers
1048 >        for (int j = 0; j < 3; j++) {
1049 >          scaled[j] -= roundMe(scaled[j]);
1050 >          scaled[j] += 0.5;
1051 >        }
1052 >        
1053 >        // find xyz-indices of cell that cutoffGroup is in.
1054 >        whichCell.x() = nCells_.x() * scaled.x();
1055 >        whichCell.y() = nCells_.y() * scaled.y();
1056 >        whichCell.z() = nCells_.z() * scaled.z();
1057 >        
1058 >        // find single index of this cell:
1059 >        cellIndex = Vlinear(whichCell, nCells_);
1060 >        
1061 >        // add this cutoff group to the list of groups in this cell;
1062 >        cellListCol_[cellIndex].push_back(i);
1063        }
991 –
992 –      // find xyz-indices of cell that cutoffGroup is in.
993 –      whichCell.x() = nCells_.x() * scaled.x();
994 –      whichCell.y() = nCells_.y() * scaled.y();
995 –      whichCell.z() = nCells_.z() * scaled.z();
996 –
997 –      // find single index of this cell:
998 –      cellIndex = Vlinear(whichCell, nCells_);
999 –
1000 –      // add this cutoff group to the list of groups in this cell;
1001 –      cellListCol_[cellIndex].push_back(i);
1002 –    }
1064   #else
1065 <    for (int i = 0; i < nGroups_; i++) {
1066 <      rs = snap_->cgData.position[i];
1067 <
1068 <      // scaled positions relative to the box vectors
1069 <      scaled = invHmat * rs;
1070 <
1071 <      // wrap the vector back into the unit box by subtracting integer box
1072 <      // numbers
1073 <      for (int j = 0; j < 3; j++) {
1074 <        scaled[j] -= roundMe(scaled[j]);
1075 <        scaled[j] += 0.5;
1065 >      for (int i = 0; i < nGroups_; i++) {
1066 >        rs = snap_->cgData.position[i];
1067 >        
1068 >        // scaled positions relative to the box vectors
1069 >        scaled = invHmat * rs;
1070 >        
1071 >        // wrap the vector back into the unit box by subtracting integer box
1072 >        // numbers
1073 >        for (int j = 0; j < 3; j++) {
1074 >          scaled[j] -= roundMe(scaled[j]);
1075 >          scaled[j] += 0.5;
1076 >        }
1077 >        
1078 >        // find xyz-indices of cell that cutoffGroup is in.
1079 >        whichCell.x() = nCells_.x() * scaled.x();
1080 >        whichCell.y() = nCells_.y() * scaled.y();
1081 >        whichCell.z() = nCells_.z() * scaled.z();
1082 >        
1083 >        // find single index of this cell:
1084 >        cellIndex = Vlinear(whichCell, nCells_);      
1085 >        
1086 >        // add this cutoff group to the list of groups in this cell;
1087 >        cellList_[cellIndex].push_back(i);
1088        }
1016 –
1017 –      // find xyz-indices of cell that cutoffGroup is in.
1018 –      whichCell.x() = nCells_.x() * scaled.x();
1019 –      whichCell.y() = nCells_.y() * scaled.y();
1020 –      whichCell.z() = nCells_.z() * scaled.z();
1021 –
1022 –      // find single index of this cell:
1023 –      cellIndex = Vlinear(whichCell, nCells_);      
1024 –
1025 –      // add this cutoff group to the list of groups in this cell;
1026 –      cellList_[cellIndex].push_back(i);
1027 –    }
1089   #endif
1090  
1091 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1092 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1093 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1094 <          Vector3i m1v(m1x, m1y, m1z);
1095 <          int m1 = Vlinear(m1v, nCells_);
1035 <
1036 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1037 <               os != cellOffsets_.end(); ++os) {
1091 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1092 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1093 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1094 >            Vector3i m1v(m1x, m1y, m1z);
1095 >            int m1 = Vlinear(m1v, nCells_);
1096              
1097 <            Vector3i m2v = m1v + (*os);
1098 <            
1099 <            if (m2v.x() >= nCells_.x()) {
1100 <              m2v.x() = 0;          
1101 <            } else if (m2v.x() < 0) {
1102 <              m2v.x() = nCells_.x() - 1;
1103 <            }
1104 <            
1105 <            if (m2v.y() >= nCells_.y()) {
1106 <              m2v.y() = 0;          
1107 <            } else if (m2v.y() < 0) {
1108 <              m2v.y() = nCells_.y() - 1;
1109 <            }
1110 <            
1111 <            if (m2v.z() >= nCells_.z()) {
1112 <              m2v.z() = 0;          
1113 <            } else if (m2v.z() < 0) {
1114 <              m2v.z() = nCells_.z() - 1;
1115 <            }
1116 <            
1117 <            int m2 = Vlinear (m2v, nCells_);
1118 <
1097 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1098 >                 os != cellOffsets_.end(); ++os) {
1099 >              
1100 >              Vector3i m2v = m1v + (*os);
1101 >              
1102 >              if (m2v.x() >= nCells_.x()) {
1103 >                m2v.x() = 0;          
1104 >              } else if (m2v.x() < 0) {
1105 >                m2v.x() = nCells_.x() - 1;
1106 >              }
1107 >              
1108 >              if (m2v.y() >= nCells_.y()) {
1109 >                m2v.y() = 0;          
1110 >              } else if (m2v.y() < 0) {
1111 >                m2v.y() = nCells_.y() - 1;
1112 >              }
1113 >              
1114 >              if (m2v.z() >= nCells_.z()) {
1115 >                m2v.z() = 0;          
1116 >              } else if (m2v.z() < 0) {
1117 >                m2v.z() = nCells_.z() - 1;
1118 >              }
1119 >              
1120 >              int m2 = Vlinear (m2v, nCells_);
1121 >              
1122   #ifdef IS_MPI
1123 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1124 <                 j1 != cellListRow_[m1].end(); ++j1) {
1125 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1126 <                   j2 != cellListCol_[m2].end(); ++j2) {
1127 <                              
1128 <                // Always do this if we're in different cells or if
1129 <                // we're in the same cell and the global index of the
1130 <                // j2 cutoff group is less than the j1 cutoff group
1131 <
1132 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1133 <                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1134 <                  snap_->wrapVector(dr);
1135 <                  cuts = getGroupCutoffs( (*j1), (*j2) );
1136 <                  if (dr.lengthSquare() < cuts.third) {
1137 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1123 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1124 >                   j1 != cellListRow_[m1].end(); ++j1) {
1125 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1126 >                     j2 != cellListCol_[m2].end(); ++j2) {
1127 >                  
1128 >                  // Always do this if we're in different cells or if
1129 >                  // we're in the same cell and the global index of the
1130 >                  // j2 cutoff group is less than the j1 cutoff group
1131 >                  
1132 >                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1133 >                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1134 >                    snap_->wrapVector(dr);
1135 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1136 >                    if (dr.lengthSquare() < cuts.third) {
1137 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1138 >                    }
1139                    }
1140                  }
1141                }
1080 –            }
1142   #else
1143 <
1144 <            for (vector<int>::iterator j1 = cellList_[m1].begin();
1145 <                 j1 != cellList_[m1].end(); ++j1) {
1146 <              for (vector<int>::iterator j2 = cellList_[m2].begin();
1147 <                   j2 != cellList_[m2].end(); ++j2) {
1148 <
1149 <                // Always do this if we're in different cells or if
1150 <                // we're in the same cell and the global index of the
1151 <                // j2 cutoff group is less than the j1 cutoff group
1152 <
1153 <                if (m2 != m1 || (*j2) < (*j1)) {
1154 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1155 <                  snap_->wrapVector(dr);
1156 <                  cuts = getGroupCutoffs( (*j1), (*j2) );
1157 <                  if (dr.lengthSquare() < cuts.third) {
1158 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1143 >              
1144 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1145 >                   j1 != cellList_[m1].end(); ++j1) {
1146 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1147 >                     j2 != cellList_[m2].end(); ++j2) {
1148 >                  
1149 >                  // Always do this if we're in different cells or if
1150 >                  // we're in the same cell and the global index of the
1151 >                  // j2 cutoff group is less than the j1 cutoff group
1152 >                  
1153 >                  if (m2 != m1 || (*j2) < (*j1)) {
1154 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1155 >                    snap_->wrapVector(dr);
1156 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1157 >                    if (dr.lengthSquare() < cuts.third) {
1158 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1159 >                    }
1160                    }
1161                  }
1162                }
1101 –            }
1163   #endif
1164 +            }
1165            }
1166          }
1167        }
1168 +    } else {
1169 +      // branch to do all cutoff group pairs
1170 + #ifdef IS_MPI
1171 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1172 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1173 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1174 +          snap_->wrapVector(dr);
1175 +          cuts = getGroupCutoffs( j1, j2 );
1176 +          if (dr.lengthSquare() < cuts.third) {
1177 +            neighborList.push_back(make_pair(j1, j2));
1178 +          }
1179 +        }
1180 +      }
1181 + #else
1182 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1183 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1184 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1185 +          snap_->wrapVector(dr);
1186 +          cuts = getGroupCutoffs( j1, j2 );
1187 +          if (dr.lengthSquare() < cuts.third) {
1188 +            neighborList.push_back(make_pair(j1, j2));
1189 +          }
1190 +        }
1191 +      }        
1192 + #endif
1193      }
1194 <    
1194 >      
1195      // save the local cutoff group positions for the check that is
1196      // done on each loop:
1197      saved_CG_positions_.clear();

Diff Legend

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