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 1713 by gezelter, Sat May 19 14:21:02 2012 UTC

# Line 36 | Line 36
36   * [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).            
37   * [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).          
38   * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
39 < * [4]  Vardeman & Gezelter, in progress (2009).                        
39 > * [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40 > * [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41   */
42   #include "parallel/ForceMatrixDecomposition.hpp"
43   #include "math/SquareMatrix3.hpp"
# Line 47 | Line 48 | namespace OpenMD {
48   using namespace std;
49   namespace OpenMD {
50  
51 +  ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52 +
53 +    // In a parallel computation, row and colum scans must visit all
54 +    // surrounding cells (not just the 14 upper triangular blocks that
55 +    // are used when the processor can see all pairs)
56 + #ifdef IS_MPI
57 +    cellOffsets_.clear();
58 +    cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59 +    cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60 +    cellOffsets_.push_back( Vector3i( 1,-1,-1) );                          
61 +    cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62 +    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63 +    cellOffsets_.push_back( Vector3i( 1, 0,-1) );
64 +    cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65 +    cellOffsets_.push_back( Vector3i( 0, 1,-1) );      
66 +    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67 +    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68 +    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69 +    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70 +    cellOffsets_.push_back( Vector3i(-1, 0, 0) );      
71 +    cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72 +    cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73 +    cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74 +    cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75 +    cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76 +    cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77 +    cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78 +    cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79 +    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80 +    cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81 +    cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82 +    cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83 +    cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84 +    cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85 + #endif    
86 +  }
87 +
88 +
89    /**
90     * distributeInitialData is essentially a copy of the older fortran
91     * SimulationSetup
92     */
54 –  
93    void ForceMatrixDecomposition::distributeInitialData() {
94      snap_ = sman_->getCurrentSnapshot();
95      storageLayout_ = sman_->getStorageLayout();
96      ff_ = info_->getForceField();
97      nLocal_ = snap_->getNumberOfAtoms();
98 <
98 >    
99      nGroups_ = info_->getNLocalCutoffGroups();
62 –    cerr << "in dId, nGroups = " << nGroups_ << "\n";
100      // gather the information for atomtype IDs (atids):
101 <    identsLocal = info_->getIdentArray();
101 >    idents = info_->getIdentArray();
102      AtomLocalToGlobal = info_->getGlobalAtomIndices();
103      cgLocalToGlobal = info_->getGlobalGroupIndices();
104      vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
105 +
106      massFactors = info_->getMassFactors();
69 –    PairList excludes = info_->getExcludedInteractions();
70 –    PairList oneTwo = info_->getOneTwoInteractions();
71 –    PairList oneThree = info_->getOneThreeInteractions();
72 –    PairList oneFour = info_->getOneFourInteractions();
107  
108 +    PairList* excludes = info_->getExcludedInteractions();
109 +    PairList* oneTwo = info_->getOneTwoInteractions();
110 +    PairList* oneThree = info_->getOneThreeInteractions();
111 +    PairList* oneFour = info_->getOneFourInteractions();
112 +
113   #ifdef IS_MPI
114  
115 <    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
116 <    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
78 <    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
79 <    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
80 <    AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
115 >    MPI::Intracomm row = rowComm.getComm();
116 >    MPI::Intracomm col = colComm.getComm();
117  
118 <    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
119 <    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
120 <    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
121 <    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
122 <    AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
118 >    AtomPlanIntRow = new Plan<int>(row, nLocal_);
119 >    AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
120 >    AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
121 >    AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
122 >    AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
123  
124 <    cgCommIntRow = new Communicator<Row,int>(nGroups_);
125 <    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
126 <    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
127 <    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
124 >    AtomPlanIntColumn = new Plan<int>(col, nLocal_);
125 >    AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
126 >    AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
127 >    AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
128 >    AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
129  
130 <    nAtomsInRow_ = AtomCommIntRow->getSize();
131 <    nAtomsInCol_ = AtomCommIntColumn->getSize();
132 <    nGroupsInRow_ = cgCommIntRow->getSize();
133 <    nGroupsInCol_ = cgCommIntColumn->getSize();
130 >    cgPlanIntRow = new Plan<int>(row, nGroups_);
131 >    cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
132 >    cgPlanIntColumn = new Plan<int>(col, nGroups_);
133 >    cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
134  
135 +    nAtomsInRow_ = AtomPlanIntRow->getSize();
136 +    nAtomsInCol_ = AtomPlanIntColumn->getSize();
137 +    nGroupsInRow_ = cgPlanIntRow->getSize();
138 +    nGroupsInCol_ = cgPlanIntColumn->getSize();
139 +
140      // Modify the data storage objects with the correct layouts and sizes:
141      atomRowData.resize(nAtomsInRow_);
142      atomRowData.setStorageLayout(storageLayout_);
# Line 108 | Line 150 | namespace OpenMD {
150      identsRow.resize(nAtomsInRow_);
151      identsCol.resize(nAtomsInCol_);
152      
153 <    AtomCommIntRow->gather(identsLocal, identsRow);
154 <    AtomCommIntColumn->gather(identsLocal, identsCol);
153 >    AtomPlanIntRow->gather(idents, identsRow);
154 >    AtomPlanIntColumn->gather(idents, identsCol);
155      
156 <    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
157 <    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
158 <    
117 <    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
118 <    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
156 >    // allocate memory for the parallel objects
157 >    atypesRow.resize(nAtomsInRow_);
158 >    atypesCol.resize(nAtomsInCol_);
159  
160 <    AtomCommRealRow->gather(massFactors, massFactorsRow);
161 <    AtomCommRealColumn->gather(massFactors, massFactorsCol);
160 >    for (int i = 0; i < nAtomsInRow_; i++)
161 >      atypesRow[i] = ff_->getAtomType(identsRow[i]);
162 >    for (int i = 0; i < nAtomsInCol_; i++)
163 >      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
164 >
165 >    pot_row.resize(nAtomsInRow_);
166 >    pot_col.resize(nAtomsInCol_);
167 >
168 >    AtomRowToGlobal.resize(nAtomsInRow_);
169 >    AtomColToGlobal.resize(nAtomsInCol_);
170 >    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
171 >    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
172 >
173 >    cgRowToGlobal.resize(nGroupsInRow_);
174 >    cgColToGlobal.resize(nGroupsInCol_);
175 >    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
176 >    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
177  
178 +    massFactorsRow.resize(nAtomsInRow_);
179 +    massFactorsCol.resize(nAtomsInCol_);
180 +    AtomPlanRealRow->gather(massFactors, massFactorsRow);
181 +    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
182 +
183      groupListRow_.clear();
184      groupListRow_.resize(nGroupsInRow_);
185      for (int i = 0; i < nGroupsInRow_; i++) {
# Line 142 | Line 202 | namespace OpenMD {
202        }      
203      }
204  
205 <    skipsForAtom.clear();
206 <    skipsForAtom.resize(nAtomsInRow_);
205 >    excludesForAtom.clear();
206 >    excludesForAtom.resize(nAtomsInRow_);
207      toposForAtom.clear();
208      toposForAtom.resize(nAtomsInRow_);
209      topoDist.clear();
# Line 154 | Line 214 | namespace OpenMD {
214        for (int j = 0; j < nAtomsInCol_; j++) {
215          int jglob = AtomColToGlobal[j];
216  
217 <        if (excludes.hasPair(iglob, jglob))
218 <          skipsForAtom[i].push_back(j);      
217 >        if (excludes->hasPair(iglob, jglob))
218 >          excludesForAtom[i].push_back(j);      
219          
220 <        if (oneTwo.hasPair(iglob, jglob)) {
220 >        if (oneTwo->hasPair(iglob, jglob)) {
221            toposForAtom[i].push_back(j);
222            topoDist[i].push_back(1);
223          } else {
224 <          if (oneThree.hasPair(iglob, jglob)) {
224 >          if (oneThree->hasPair(iglob, jglob)) {
225              toposForAtom[i].push_back(j);
226              topoDist[i].push_back(2);
227            } else {
228 <            if (oneFour.hasPair(iglob, jglob)) {
228 >            if (oneFour->hasPair(iglob, jglob)) {
229                toposForAtom[i].push_back(j);
230                topoDist[i].push_back(3);
231              }
# Line 174 | Line 234 | namespace OpenMD {
234        }      
235      }
236  
237 < #endif
238 <
239 <    groupList_.clear();
180 <    groupList_.resize(nGroups_);
181 <    for (int i = 0; i < nGroups_; i++) {
182 <      int gid = cgLocalToGlobal[i];
183 <      for (int j = 0; j < nLocal_; j++) {
184 <        int aid = AtomLocalToGlobal[j];
185 <        if (globalGroupMembership[aid] == gid) {
186 <          groupList_[i].push_back(j);
187 <        }
188 <      }      
189 <    }
190 <
191 <    skipsForAtom.clear();
192 <    skipsForAtom.resize(nLocal_);
237 > #else
238 >    excludesForAtom.clear();
239 >    excludesForAtom.resize(nLocal_);
240      toposForAtom.clear();
241      toposForAtom.resize(nLocal_);
242      topoDist.clear();
# Line 201 | Line 248 | namespace OpenMD {
248        for (int j = 0; j < nLocal_; j++) {
249          int jglob = AtomLocalToGlobal[j];
250  
251 <        if (excludes.hasPair(iglob, jglob))
252 <          skipsForAtom[i].push_back(j);              
251 >        if (excludes->hasPair(iglob, jglob))
252 >          excludesForAtom[i].push_back(j);              
253          
254 <        if (oneTwo.hasPair(iglob, jglob)) {
254 >        if (oneTwo->hasPair(iglob, jglob)) {
255            toposForAtom[i].push_back(j);
256            topoDist[i].push_back(1);
257          } else {
258 <          if (oneThree.hasPair(iglob, jglob)) {
258 >          if (oneThree->hasPair(iglob, jglob)) {
259              toposForAtom[i].push_back(j);
260              topoDist[i].push_back(2);
261            } else {
262 <            if (oneFour.hasPair(iglob, jglob)) {
262 >            if (oneFour->hasPair(iglob, jglob)) {
263                toposForAtom[i].push_back(j);
264                topoDist[i].push_back(3);
265              }
# Line 220 | Line 267 | namespace OpenMD {
267          }
268        }      
269      }
270 <    
270 > #endif
271 >
272 >    // allocate memory for the parallel objects
273 >    atypesLocal.resize(nLocal_);
274 >
275 >    for (int i = 0; i < nLocal_; i++)
276 >      atypesLocal[i] = ff_->getAtomType(idents[i]);
277 >
278 >    groupList_.clear();
279 >    groupList_.resize(nGroups_);
280 >    for (int i = 0; i < nGroups_; i++) {
281 >      int gid = cgLocalToGlobal[i];
282 >      for (int j = 0; j < nLocal_; j++) {
283 >        int aid = AtomLocalToGlobal[j];
284 >        if (globalGroupMembership[aid] == gid) {
285 >          groupList_[i].push_back(j);
286 >        }
287 >      }      
288 >    }
289 >
290 >
291      createGtypeCutoffMap();
292 +
293    }
294    
295    void ForceMatrixDecomposition::createGtypeCutoffMap() {
296 <
296 >    
297      RealType tol = 1e-6;
298 +    largestRcut_ = 0.0;
299      RealType rc;
300      int atid;
301      set<AtomType*> atypes = info_->getSimulatedAtomTypes();
302 <    vector<RealType> atypeCutoff;
303 <    atypeCutoff.resize( atypes.size() );
304 <
302 >    
303 >    map<int, RealType> atypeCutoff;
304 >      
305      for (set<AtomType*>::iterator at = atypes.begin();
306           at != atypes.end(); ++at){
238 –      rc = interactionMan_->getSuggestedCutoffRadius(*at);
307        atid = (*at)->getIdent();
308 <      atypeCutoff[atid] = rc;
308 >      if (userChoseCutoff_)
309 >        atypeCutoff[atid] = userCutoff_;
310 >      else
311 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
312      }
313 <
313 >    
314      vector<RealType> gTypeCutoffs;
244 –
315      // first we do a single loop over the cutoff groups to find the
316      // largest cutoff for any atypes present in this group.
317   #ifdef IS_MPI
# Line 299 | Line 369 | namespace OpenMD {
369  
370      vector<RealType> groupCutoff(nGroups_, 0.0);
371      groupToGtype.resize(nGroups_);
302 –
303 –    cerr << "nGroups = " << nGroups_ << "\n";
372      for (int cg1 = 0; cg1 < nGroups_; cg1++) {
305 –
373        groupCutoff[cg1] = 0.0;
374        vector<int> atomList = getAtomsInGroupRow(cg1);
308 –
375        for (vector<int>::iterator ia = atomList.begin();
376             ia != atomList.end(); ++ia) {            
377          int atom1 = (*ia);
378 <        atid = identsLocal[atom1];
379 <        if (atypeCutoff[atid] > groupCutoff[cg1]) {
378 >        atid = idents[atom1];
379 >        if (atypeCutoff[atid] > groupCutoff[cg1])
380            groupCutoff[cg1] = atypeCutoff[atid];
315 –        }
381        }
382 <
382 >      
383        bool gTypeFound = false;
384        for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
385          if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
# Line 322 | Line 387 | namespace OpenMD {
387            gTypeFound = true;
388          }
389        }
390 <      if (!gTypeFound) {
390 >      if (!gTypeFound) {      
391          gTypeCutoffs.push_back( groupCutoff[cg1] );
392          groupToGtype[cg1] = gTypeCutoffs.size() - 1;
393        }      
394      }
395   #endif
396  
332 –    cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
397      // Now we find the maximum group cutoff value present in the simulation
398  
399 <    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
399 >    RealType groupMax = *max_element(gTypeCutoffs.begin(),
400 >                                     gTypeCutoffs.end());
401  
402   #ifdef IS_MPI
403 <    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
403 >    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
404 >                              MPI::MAX);
405   #endif
406      
407      RealType tradRcut = groupMax;
# Line 365 | Line 431 | namespace OpenMD {
431  
432          pair<int,int> key = make_pair(i,j);
433          gTypeCutoffMap[key].first = thisRcut;
368 –
434          if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
370 –
435          gTypeCutoffMap[key].second = thisRcut*thisRcut;
372 –        
436          gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
374 –
437          // sanity check
438          
439          if (userChoseCutoff_) {
440            if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
441              sprintf(painCave.errMsg,
442                      "ForceMatrixDecomposition::createGtypeCutoffMap "
443 <                    "user-specified rCut does not match computed group Cutoff\n");
443 >                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
444              painCave.severity = OPENMD_ERROR;
445              painCave.isFatal = 1;
446              simError();            
# Line 410 | Line 472 | namespace OpenMD {
472    }
473  
474    void ForceMatrixDecomposition::zeroWorkArrays() {
475 +    pairwisePot = 0.0;
476 +    embeddingPot = 0.0;
477  
414 –    for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
415 –      longRangePot_[j] = 0.0;
416 –    }
417 –
478   #ifdef IS_MPI
479      if (storageLayout_ & DataStorage::dslForce) {
480        fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
# Line 430 | Line 490 | namespace OpenMD {
490           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
491  
492      fill(pot_col.begin(), pot_col.end(),
493 <         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
434 <    
435 <    pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
493 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
494  
495      if (storageLayout_ & DataStorage::dslParticlePot) {    
496 <      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
497 <      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
496 >      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
497 >           0.0);
498 >      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
499 >           0.0);
500      }
501  
502      if (storageLayout_ & DataStorage::dslDensity) {      
# Line 445 | Line 505 | namespace OpenMD {
505      }
506  
507      if (storageLayout_ & DataStorage::dslFunctional) {  
508 <      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
509 <      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
508 >      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
509 >           0.0);
510 >      fill(atomColData.functional.begin(), atomColData.functional.end(),
511 >           0.0);
512      }
513  
514      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
# Line 456 | Line 518 | namespace OpenMD {
518             atomColData.functionalDerivative.end(), 0.0);
519      }
520  
521 < #else
522 <    
521 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
522 >      fill(atomRowData.skippedCharge.begin(),
523 >           atomRowData.skippedCharge.end(), 0.0);
524 >      fill(atomColData.skippedCharge.begin(),
525 >           atomColData.skippedCharge.end(), 0.0);
526 >    }
527 >
528 >    if (storageLayout_ & DataStorage::dslElectricField) {    
529 >      fill(atomRowData.electricField.begin(),
530 >           atomRowData.electricField.end(), V3Zero);
531 >      fill(atomColData.electricField.begin(),
532 >           atomColData.electricField.end(), V3Zero);
533 >    }
534 >    if (storageLayout_ & DataStorage::dslFlucQForce) {    
535 >      fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(),
536 >           0.0);
537 >      fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.end(),
538 >           0.0);
539 >    }
540 >
541 > #endif
542 >    // even in parallel, we need to zero out the local arrays:
543 >
544      if (storageLayout_ & DataStorage::dslParticlePot) {      
545        fill(snap_->atomData.particlePot.begin(),
546             snap_->atomData.particlePot.end(), 0.0);
# Line 467 | Line 550 | namespace OpenMD {
550        fill(snap_->atomData.density.begin(),
551             snap_->atomData.density.end(), 0.0);
552      }
553 +
554      if (storageLayout_ & DataStorage::dslFunctional) {
555        fill(snap_->atomData.functional.begin(),
556             snap_->atomData.functional.end(), 0.0);
557      }
558 +
559      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
560        fill(snap_->atomData.functionalDerivative.begin(),
561             snap_->atomData.functionalDerivative.end(), 0.0);
562      }
563 < #endif
564 <    
563 >
564 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
565 >      fill(snap_->atomData.skippedCharge.begin(),
566 >           snap_->atomData.skippedCharge.end(), 0.0);
567 >    }
568 >
569 >    if (storageLayout_ & DataStorage::dslElectricField) {      
570 >      fill(snap_->atomData.electricField.begin(),
571 >           snap_->atomData.electricField.end(), V3Zero);
572 >    }
573    }
574  
575  
# Line 486 | Line 579 | namespace OpenMD {
579   #ifdef IS_MPI
580      
581      // gather up the atomic positions
582 <    AtomCommVectorRow->gather(snap_->atomData.position,
582 >    AtomPlanVectorRow->gather(snap_->atomData.position,
583                                atomRowData.position);
584 <    AtomCommVectorColumn->gather(snap_->atomData.position,
584 >    AtomPlanVectorColumn->gather(snap_->atomData.position,
585                                   atomColData.position);
586      
587      // gather up the cutoff group positions
588 <    cgCommVectorRow->gather(snap_->cgData.position,
588 >
589 >    cgPlanVectorRow->gather(snap_->cgData.position,
590                              cgRowData.position);
591 <    cgCommVectorColumn->gather(snap_->cgData.position,
591 >
592 >    cgPlanVectorColumn->gather(snap_->cgData.position,
593                                 cgColData.position);
594 +
595      
596      // if needed, gather the atomic rotation matrices
597      if (storageLayout_ & DataStorage::dslAmat) {
598 <      AtomCommMatrixRow->gather(snap_->atomData.aMat,
598 >      AtomPlanMatrixRow->gather(snap_->atomData.aMat,
599                                  atomRowData.aMat);
600 <      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
600 >      AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
601                                     atomColData.aMat);
602      }
603      
604      // if needed, gather the atomic eletrostatic frames
605      if (storageLayout_ & DataStorage::dslElectroFrame) {
606 <      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
606 >      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
607                                  atomRowData.electroFrame);
608 <      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
608 >      AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
609                                     atomColData.electroFrame);
610      }
611 +
612 +    // if needed, gather the atomic fluctuating charge values
613 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {
614 +      AtomPlanRealRow->gather(snap_->atomData.flucQPos,
615 +                              atomRowData.flucQPos);
616 +      AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
617 +                                 atomColData.flucQPos);
618 +    }
619 +
620   #endif      
621    }
622    
# Line 525 | Line 630 | namespace OpenMD {
630      
631      if (storageLayout_ & DataStorage::dslDensity) {
632        
633 <      AtomCommRealRow->scatter(atomRowData.density,
633 >      AtomPlanRealRow->scatter(atomRowData.density,
634                                 snap_->atomData.density);
635        
636        int n = snap_->atomData.density.size();
637        vector<RealType> rho_tmp(n, 0.0);
638 <      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
638 >      AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
639        for (int i = 0; i < n; i++)
640          snap_->atomData.density[i] += rho_tmp[i];
641 +    }
642 +
643 +    if (storageLayout_ & DataStorage::dslElectricField) {
644 +      
645 +      AtomPlanVectorRow->scatter(atomRowData.electricField,
646 +                                 snap_->atomData.electricField);
647 +      
648 +      int n = snap_->atomData.electricField.size();
649 +      vector<Vector3d> field_tmp(n, V3Zero);
650 +      AtomPlanVectorColumn->scatter(atomColData.electricField, field_tmp);
651 +      for (int i = 0; i < n; i++)
652 +        snap_->atomData.electricField[i] += field_tmp[i];
653      }
654   #endif
655    }
# Line 546 | Line 663 | namespace OpenMD {
663      storageLayout_ = sman_->getStorageLayout();
664   #ifdef IS_MPI
665      if (storageLayout_ & DataStorage::dslFunctional) {
666 <      AtomCommRealRow->gather(snap_->atomData.functional,
666 >      AtomPlanRealRow->gather(snap_->atomData.functional,
667                                atomRowData.functional);
668 <      AtomCommRealColumn->gather(snap_->atomData.functional,
668 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
669                                   atomColData.functional);
670      }
671      
672      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
673 <      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
673 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
674                                atomRowData.functionalDerivative);
675 <      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
675 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
676                                   atomColData.functionalDerivative);
677      }
678   #endif
# Line 569 | Line 686 | namespace OpenMD {
686      int n = snap_->atomData.force.size();
687      vector<Vector3d> frc_tmp(n, V3Zero);
688      
689 <    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
689 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
690      for (int i = 0; i < n; i++) {
691        snap_->atomData.force[i] += frc_tmp[i];
692        frc_tmp[i] = 0.0;
693      }
694      
695 <    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
696 <    for (int i = 0; i < n; i++)
695 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
696 >    for (int i = 0; i < n; i++) {
697        snap_->atomData.force[i] += frc_tmp[i];
698 <    
699 <    
698 >    }
699 >        
700      if (storageLayout_ & DataStorage::dslTorque) {
701  
702 <      int nt = snap_->atomData.force.size();
702 >      int nt = snap_->atomData.torque.size();
703        vector<Vector3d> trq_tmp(nt, V3Zero);
704  
705 <      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
706 <      for (int i = 0; i < n; i++) {
705 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
706 >      for (int i = 0; i < nt; i++) {
707          snap_->atomData.torque[i] += trq_tmp[i];
708          trq_tmp[i] = 0.0;
709        }
710        
711 <      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
712 <      for (int i = 0; i < n; i++)
711 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
712 >      for (int i = 0; i < nt; i++)
713          snap_->atomData.torque[i] += trq_tmp[i];
714      }
715 +
716 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
717 +
718 +      int ns = snap_->atomData.skippedCharge.size();
719 +      vector<RealType> skch_tmp(ns, 0.0);
720 +
721 +      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
722 +      for (int i = 0; i < ns; i++) {
723 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
724 +        skch_tmp[i] = 0.0;
725 +      }
726 +      
727 +      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
728 +      for (int i = 0; i < ns; i++)
729 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
730 +            
731 +    }
732      
733 +    if (storageLayout_ & DataStorage::dslFlucQForce) {
734 +
735 +      int nq = snap_->atomData.flucQFrc.size();
736 +      vector<RealType> fqfrc_tmp(nq, 0.0);
737 +
738 +      AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
739 +      for (int i = 0; i < nq; i++) {
740 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
741 +        fqfrc_tmp[i] = 0.0;
742 +      }
743 +      
744 +      AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
745 +      for (int i = 0; i < nq; i++)
746 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
747 +            
748 +    }
749 +
750      nLocal_ = snap_->getNumberOfAtoms();
751  
752      vector<potVec> pot_temp(nLocal_,
# Line 603 | Line 754 | namespace OpenMD {
754  
755      // scatter/gather pot_row into the members of my column
756            
757 <    AtomCommPotRow->scatter(pot_row, pot_temp);
757 >    AtomPlanPotRow->scatter(pot_row, pot_temp);
758  
759      for (int ii = 0;  ii < pot_temp.size(); ii++ )
760 <      pot_local += pot_temp[ii];
760 >      pairwisePot += pot_temp[ii];
761      
762      fill(pot_temp.begin(), pot_temp.end(),
763           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
764        
765 <    AtomCommPotColumn->scatter(pot_col, pot_temp);    
765 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
766      
767      for (int ii = 0;  ii < pot_temp.size(); ii++ )
768 <      pot_local += pot_temp[ii];
768 >      pairwisePot += pot_temp[ii];    
769      
770 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
771 +      RealType ploc1 = pairwisePot[ii];
772 +      RealType ploc2 = 0.0;
773 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
774 +      pairwisePot[ii] = ploc2;
775 +    }
776 +
777 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
778 +      RealType ploc1 = embeddingPot[ii];
779 +      RealType ploc2 = 0.0;
780 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
781 +      embeddingPot[ii] = ploc2;
782 +    }
783 +
784   #endif
785 +
786    }
787  
788    int ForceMatrixDecomposition::getNAtomsInRow() {  
# Line 717 | Line 883 | namespace OpenMD {
883      return d;    
884    }
885  
886 <  vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
887 <    return skipsForAtom[atom1];
886 >  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
887 >    return excludesForAtom[atom1];
888    }
889  
890    /**
891 <   * 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
891 >   * We need to exclude some overcounted interactions that result from
892     * the parallel decomposition.
893     */
894    bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
895      int unique_id_1, unique_id_2;
896 <
896 >        
897   #ifdef IS_MPI
898      // in MPI, we have to look up the unique IDs for each atom
899      unique_id_1 = AtomRowToGlobal[atom1];
900      unique_id_2 = AtomColToGlobal[atom2];
901 + #else
902 +    unique_id_1 = AtomLocalToGlobal[atom1];
903 +    unique_id_2 = AtomLocalToGlobal[atom2];
904 + #endif  
905  
739 –    // this situation should only arise in MPI simulations
906      if (unique_id_1 == unique_id_2) return true;
907 <    
907 >
908 > #ifdef IS_MPI
909      // this prevents us from doing the pair on multiple processors
910      if (unique_id_1 < unique_id_2) {
911        if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
912      } else {
913 <      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
913 >      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
914      }
748 – #else
749 –    // in the normal loop, the atom numbers are unique
750 –    unique_id_1 = atom1;
751 –    unique_id_2 = atom2;
915   #endif
916      
917 <    for (vector<int>::iterator i = skipsForAtom[atom1].begin();
918 <         i != skipsForAtom[atom1].end(); ++i) {
756 <      if ( (*i) == unique_id_2 ) return true;
757 <    }    
917 >    return false;
918 >  }
919  
920 +  /**
921 +   * We need to handle the interactions for atoms who are involved in
922 +   * the same rigid body as well as some short range interactions
923 +   * (bonds, bends, torsions) differently from other interactions.
924 +   * We'll still visit the pairwise routines, but with a flag that
925 +   * tells those routines to exclude the pair from direct long range
926 +   * interactions.  Some indirect interactions (notably reaction
927 +   * field) must still be handled for these pairs.
928 +   */
929 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
930 +
931 +    // excludesForAtom was constructed to use row/column indices in the MPI
932 +    // version, and to use local IDs in the non-MPI version:
933 +    
934 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
935 +         i != excludesForAtom[atom1].end(); ++i) {
936 +      if ( (*i) == atom2 ) return true;
937 +    }
938 +
939 +    return false;
940    }
941  
942  
# Line 776 | Line 957 | namespace OpenMD {
957    }
958  
959      // filling interaction blocks with pointers
960 <  void ForceMatrixDecomposition::fillInteractionData(InteractionData idat,
961 <                                                     int atom1, int atom2) {    
960 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
961 >                                                     int atom1, int atom2) {
962 >
963 >    idat.excluded = excludeAtomPair(atom1, atom2);
964 >  
965   #ifdef IS_MPI
966 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
967 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
968 +    //                         ff_->getAtomType(identsCol[atom2]) );
969      
783 –    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
784 –                             ff_->getAtomType(identsCol[atom2]) );
785 –    
970      if (storageLayout_ & DataStorage::dslAmat) {
971        idat.A1 = &(atomRowData.aMat[atom1]);
972        idat.A2 = &(atomColData.aMat[atom2]);
# Line 818 | Line 1002 | namespace OpenMD {
1002        idat.particlePot2 = &(atomColData.particlePot[atom2]);
1003      }
1004  
1005 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
1006 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1007 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1008 +    }
1009 +
1010   #else
1011 +    
1012  
1013 <    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
1014 <                             ff_->getAtomType(identsLocal[atom2]) );
1013 >    // cerr << "atoms = " << atom1 << " " << atom2 << "\n";
1014 >    // cerr << "pos1 = " << snap_->atomData.position[atom1] << "\n";
1015 >    // cerr << "pos2 = " << snap_->atomData.position[atom2] << "\n";
1016  
1017 +    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1018 +    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1019 +    //                         ff_->getAtomType(idents[atom2]) );
1020 +
1021      if (storageLayout_ & DataStorage::dslAmat) {
1022        idat.A1 = &(snap_->atomData.aMat[atom1]);
1023        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 838 | Line 1033 | namespace OpenMD {
1033        idat.t2 = &(snap_->atomData.torque[atom2]);
1034      }
1035  
1036 <    if (storageLayout_ & DataStorage::dslDensity) {
1036 >    if (storageLayout_ & DataStorage::dslDensity) {    
1037        idat.rho1 = &(snap_->atomData.density[atom1]);
1038        idat.rho2 = &(snap_->atomData.density[atom2]);
1039      }
# Line 858 | Line 1053 | namespace OpenMD {
1053        idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1054      }
1055  
1056 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
1057 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1058 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1059 +    }
1060   #endif
1061    }
1062  
1063    
1064 <  void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {    
1064 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1065   #ifdef IS_MPI
1066 <    pot_row[atom1] += 0.5 *  *(idat.pot);
1067 <    pot_col[atom2] += 0.5 *  *(idat.pot);
1066 >    pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1067 >    pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1068  
1069      atomRowData.force[atom1] += *(idat.f1);
1070      atomColData.force[atom2] -= *(idat.f1);
1071 +
1072 +    // should particle pot be done here also?
1073   #else
1074 <    longRangePot_ += *(idat.pot);
1075 <    
1074 >    pairwisePot += *(idat.pot);
1075 >
1076      snap_->atomData.force[atom1] += *(idat.f1);
1077      snap_->atomData.force[atom2] -= *(idat.f1);
877 – #endif
1078  
1079 <  }
1080 <
1081 <
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]);
1079 >    if (idat.doParticlePot) {
1080 >      snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1081 >      snap_->atomData.particlePot[atom2] -= *(idat.vpair) * *(idat.sw);
1082      }
1083 <    if (storageLayout_ & DataStorage::dslTorque) {
1084 <      idat.t1 = &(atomRowData.torque[atom1]);
1085 <      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    
1083 >      
1084 > #endif
1085 >    
1086    }
1087  
1088    /*
# Line 918 | Line 1095 | namespace OpenMD {
1095        
1096      vector<pair<int, int> > neighborList;
1097      groupCutoffs cuts;
1098 +    bool doAllPairs = false;
1099 +
1100   #ifdef IS_MPI
1101      cellListRow_.clear();
1102      cellListCol_.clear();
# Line 937 | Line 1116 | namespace OpenMD {
1116      nCells_.y() = (int) ( Hy.length() )/ rList_;
1117      nCells_.z() = (int) ( Hz.length() )/ rList_;
1118  
1119 +    // handle small boxes where the cell offsets can end up repeating cells
1120 +    
1121 +    if (nCells_.x() < 3) doAllPairs = true;
1122 +    if (nCells_.y() < 3) doAllPairs = true;
1123 +    if (nCells_.z() < 3) doAllPairs = true;
1124 +
1125      Mat3x3d invHmat = snap_->getInvHmat();
1126      Vector3d rs, scaled, dr;
1127      Vector3i whichCell;
1128      int cellIndex;
1129      int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1130  
946 –    cerr << "flag1\n";
1131   #ifdef IS_MPI
1132      cellListRow_.resize(nCtot);
1133      cellListCol_.resize(nCtot);
1134   #else
1135      cellList_.resize(nCtot);
1136   #endif
1137 <    cerr << "flag2\n";
1137 >
1138 >    if (!doAllPairs) {
1139   #ifdef IS_MPI
955 –    for (int i = 0; i < nGroupsInRow_; i++) {
956 –      rs = cgRowData.position[i];
1140  
1141 <      // scaled positions relative to the box vectors
1142 <      scaled = invHmat * rs;
1143 <
1144 <      // wrap the vector back into the unit box by subtracting integer box
1145 <      // numbers
1146 <      for (int j = 0; j < 3; j++) {
1147 <        scaled[j] -= roundMe(scaled[j]);
1148 <        scaled[j] += 0.5;
1141 >      for (int i = 0; i < nGroupsInRow_; i++) {
1142 >        rs = cgRowData.position[i];
1143 >        
1144 >        // scaled positions relative to the box vectors
1145 >        scaled = invHmat * rs;
1146 >        
1147 >        // wrap the vector back into the unit box by subtracting integer box
1148 >        // numbers
1149 >        for (int j = 0; j < 3; j++) {
1150 >          scaled[j] -= roundMe(scaled[j]);
1151 >          scaled[j] += 0.5;
1152 >        }
1153 >        
1154 >        // find xyz-indices of cell that cutoffGroup is in.
1155 >        whichCell.x() = nCells_.x() * scaled.x();
1156 >        whichCell.y() = nCells_.y() * scaled.y();
1157 >        whichCell.z() = nCells_.z() * scaled.z();
1158 >        
1159 >        // find single index of this cell:
1160 >        cellIndex = Vlinear(whichCell, nCells_);
1161 >        
1162 >        // add this cutoff group to the list of groups in this cell;
1163 >        cellListRow_[cellIndex].push_back(i);
1164        }
1165 <    
1166 <      // find xyz-indices of cell that cutoffGroup is in.
1167 <      whichCell.x() = nCells_.x() * scaled.x();
1168 <      whichCell.y() = nCells_.y() * scaled.y();
1169 <      whichCell.z() = nCells_.z() * scaled.z();
1170 <
1171 <      // find single index of this cell:
1172 <      cellIndex = Vlinear(whichCell, nCells_);
1173 <
1174 <      // add this cutoff group to the list of groups in this cell;
1175 <      cellListRow_[cellIndex].push_back(i);
1176 <    }
1177 <
1178 <    for (int i = 0; i < nGroupsInCol_; i++) {
1179 <      rs = cgColData.position[i];
1180 <
1181 <      // scaled positions relative to the box vectors
1182 <      scaled = invHmat * rs;
1183 <
1184 <      // wrap the vector back into the unit box by subtracting integer box
1185 <      // numbers
1186 <      for (int j = 0; j < 3; j++) {
1187 <        scaled[j] -= roundMe(scaled[j]);
1188 <        scaled[j] += 0.5;
1189 <      }
992 <
993 <      // find xyz-indices of cell that cutoffGroup is in.
994 <      whichCell.x() = nCells_.x() * scaled.x();
995 <      whichCell.y() = nCells_.y() * scaled.y();
996 <      whichCell.z() = nCells_.z() * scaled.z();
997 <
998 <      // find single index of this cell:
999 <      cellIndex = Vlinear(whichCell, nCells_);
1000 <
1001 <      // add this cutoff group to the list of groups in this cell;
1002 <      cellListCol_[cellIndex].push_back(i);
1003 <    }
1165 >      for (int i = 0; i < nGroupsInCol_; i++) {
1166 >        rs = cgColData.position[i];
1167 >        
1168 >        // scaled positions relative to the box vectors
1169 >        scaled = invHmat * rs;
1170 >        
1171 >        // wrap the vector back into the unit box by subtracting integer box
1172 >        // numbers
1173 >        for (int j = 0; j < 3; j++) {
1174 >          scaled[j] -= roundMe(scaled[j]);
1175 >          scaled[j] += 0.5;
1176 >        }
1177 >        
1178 >        // find xyz-indices of cell that cutoffGroup is in.
1179 >        whichCell.x() = nCells_.x() * scaled.x();
1180 >        whichCell.y() = nCells_.y() * scaled.y();
1181 >        whichCell.z() = nCells_.z() * scaled.z();
1182 >        
1183 >        // find single index of this cell:
1184 >        cellIndex = Vlinear(whichCell, nCells_);
1185 >        
1186 >        // add this cutoff group to the list of groups in this cell;
1187 >        cellListCol_[cellIndex].push_back(i);
1188 >      }
1189 >    
1190   #else
1191 <    for (int i = 0; i < nGroups_; i++) {
1192 <      rs = snap_->cgData.position[i];
1193 <
1194 <      // scaled positions relative to the box vectors
1195 <      scaled = invHmat * rs;
1196 <
1197 <      // wrap the vector back into the unit box by subtracting integer box
1198 <      // numbers
1199 <      for (int j = 0; j < 3; j++) {
1200 <        scaled[j] -= roundMe(scaled[j]);
1201 <        scaled[j] += 0.5;
1191 >      for (int i = 0; i < nGroups_; i++) {
1192 >        rs = snap_->cgData.position[i];
1193 >        
1194 >        // scaled positions relative to the box vectors
1195 >        scaled = invHmat * rs;
1196 >        
1197 >        // wrap the vector back into the unit box by subtracting integer box
1198 >        // numbers
1199 >        for (int j = 0; j < 3; j++) {
1200 >          scaled[j] -= roundMe(scaled[j]);
1201 >          scaled[j] += 0.5;
1202 >        }
1203 >        
1204 >        // find xyz-indices of cell that cutoffGroup is in.
1205 >        whichCell.x() = nCells_.x() * scaled.x();
1206 >        whichCell.y() = nCells_.y() * scaled.y();
1207 >        whichCell.z() = nCells_.z() * scaled.z();
1208 >        
1209 >        // find single index of this cell:
1210 >        cellIndex = Vlinear(whichCell, nCells_);
1211 >        
1212 >        // add this cutoff group to the list of groups in this cell;
1213 >        cellList_[cellIndex].push_back(i);
1214        }
1215  
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 –    }
1216   #endif
1217  
1218 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1219 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1220 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1221 <          Vector3i m1v(m1x, m1y, m1z);
1222 <          int m1 = Vlinear(m1v, nCells_);
1036 <
1037 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1038 <               os != cellOffsets_.end(); ++os) {
1218 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1219 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1220 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1221 >            Vector3i m1v(m1x, m1y, m1z);
1222 >            int m1 = Vlinear(m1v, nCells_);
1223              
1224 <            Vector3i m2v = m1v + (*os);
1225 <            
1226 <            if (m2v.x() >= nCells_.x()) {
1227 <              m2v.x() = 0;          
1228 <            } else if (m2v.x() < 0) {
1045 <              m2v.x() = nCells_.x() - 1;
1046 <            }
1047 <            
1048 <            if (m2v.y() >= nCells_.y()) {
1049 <              m2v.y() = 0;          
1050 <            } else if (m2v.y() < 0) {
1051 <              m2v.y() = nCells_.y() - 1;
1052 <            }
1053 <            
1054 <            if (m2v.z() >= nCells_.z()) {
1055 <              m2v.z() = 0;          
1056 <            } else if (m2v.z() < 0) {
1057 <              m2v.z() = nCells_.z() - 1;
1058 <            }
1059 <            
1060 <            int m2 = Vlinear (m2v, nCells_);
1224 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1225 >                 os != cellOffsets_.end(); ++os) {
1226 >              
1227 >              Vector3i m2v = m1v + (*os);
1228 >            
1229  
1230 < #ifdef IS_MPI
1231 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1232 <                 j1 != cellListRow_[m1].end(); ++j1) {
1233 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1234 <                   j2 != cellListCol_[m2].end(); ++j2) {
1235 <                              
1236 <                // Always do this if we're in different cells or if
1237 <                // we're in the same cell and the global index of the
1238 <                // j2 cutoff group is less than the j1 cutoff group
1230 >              if (m2v.x() >= nCells_.x()) {
1231 >                m2v.x() = 0;          
1232 >              } else if (m2v.x() < 0) {
1233 >                m2v.x() = nCells_.x() - 1;
1234 >              }
1235 >              
1236 >              if (m2v.y() >= nCells_.y()) {
1237 >                m2v.y() = 0;          
1238 >              } else if (m2v.y() < 0) {
1239 >                m2v.y() = nCells_.y() - 1;
1240 >              }
1241 >              
1242 >              if (m2v.z() >= nCells_.z()) {
1243 >                m2v.z() = 0;          
1244 >              } else if (m2v.z() < 0) {
1245 >                m2v.z() = nCells_.z() - 1;
1246 >              }
1247  
1248 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1248 >              int m2 = Vlinear (m2v, nCells_);
1249 >              
1250 > #ifdef IS_MPI
1251 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1252 >                   j1 != cellListRow_[m1].end(); ++j1) {
1253 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1254 >                     j2 != cellListCol_[m2].end(); ++j2) {
1255 >                  
1256 >                  // In parallel, we need to visit *all* pairs of row
1257 >                  // & column indicies and will divide labor in the
1258 >                  // force evaluation later.
1259                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1260                    snap_->wrapVector(dr);
1261                    cuts = getGroupCutoffs( (*j1), (*j2) );
1262                    if (dr.lengthSquare() < cuts.third) {
1263                      neighborList.push_back(make_pair((*j1), (*j2)));
1264 <                  }
1264 >                  }                  
1265                  }
1266                }
1081 –            }
1267   #else
1268 +              for (vector<int>::iterator j1 = cellList_[m1].begin();
1269 +                   j1 != cellList_[m1].end(); ++j1) {
1270 +                for (vector<int>::iterator j2 = cellList_[m2].begin();
1271 +                     j2 != cellList_[m2].end(); ++j2) {
1272 +    
1273 +                  // Always do this if we're in different cells or if
1274 +                  // we're in the same cell and the global index of
1275 +                  // the j2 cutoff group is greater than or equal to
1276 +                  // the j1 cutoff group.  Note that Rappaport's code
1277 +                  // has a "less than" conditional here, but that
1278 +                  // deals with atom-by-atom computation.  OpenMD
1279 +                  // allows atoms within a single cutoff group to
1280 +                  // interact with each other.
1281  
1084 –            for (vector<int>::iterator j1 = cellList_[m1].begin();
1085 –                 j1 != cellList_[m1].end(); ++j1) {
1086 –              for (vector<int>::iterator j2 = cellList_[m2].begin();
1087 –                   j2 != cellList_[m2].end(); ++j2) {
1282  
1089 –                // Always do this if we're in different cells or if
1090 –                // we're in the same cell and the global index of the
1091 –                // j2 cutoff group is less than the j1 cutoff group
1283  
1284 <                if (m2 != m1 || (*j2) < (*j1)) {
1285 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1286 <                  snap_->wrapVector(dr);
1287 <                  cuts = getGroupCutoffs( (*j1), (*j2) );
1288 <                  if (dr.lengthSquare() < cuts.third) {
1289 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1284 >                  if (m2 != m1 || (*j2) >= (*j1) ) {
1285 >
1286 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1287 >                    snap_->wrapVector(dr);
1288 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1289 >                    if (dr.lengthSquare() < cuts.third) {
1290 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1291 >                    }
1292                    }
1293                  }
1294                }
1102 –            }
1295   #endif
1296 +            }
1297            }
1298          }
1299        }
1300 +    } else {
1301 +      // branch to do all cutoff group pairs
1302 + #ifdef IS_MPI
1303 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1304 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
1305 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1306 +          snap_->wrapVector(dr);
1307 +          cuts = getGroupCutoffs( j1, j2 );
1308 +          if (dr.lengthSquare() < cuts.third) {
1309 +            neighborList.push_back(make_pair(j1, j2));
1310 +          }
1311 +        }
1312 +      }      
1313 + #else
1314 +      // include all groups here.
1315 +      for (int j1 = 0; j1 < nGroups_; j1++) {
1316 +        // include self group interactions j2 == j1
1317 +        for (int j2 = j1; j2 < nGroups_; j2++) {
1318 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1319 +          snap_->wrapVector(dr);
1320 +          cuts = getGroupCutoffs( j1, j2 );
1321 +          if (dr.lengthSquare() < cuts.third) {
1322 +            neighborList.push_back(make_pair(j1, j2));
1323 +          }
1324 +        }    
1325 +      }
1326 + #endif
1327      }
1328 <    
1328 >      
1329      // save the local cutoff group positions for the check that is
1330      // done on each loop:
1331      saved_CG_positions_.clear();

Diff Legend

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