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 1721 by gezelter, Thu May 24 14:17:42 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_);
# 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::dslFlucQForce) {      
529 >      fill(atomRowData.flucQFrc.begin(),
530 >           atomRowData.flucQFrc.end(), 0.0);
531 >      fill(atomColData.flucQFrc.begin(),
532 >           atomColData.flucQFrc.end(), 0.0);
533 >    }
534 >
535 >    if (storageLayout_ & DataStorage::dslElectricField) {    
536 >      fill(atomRowData.electricField.begin(),
537 >           atomRowData.electricField.end(), V3Zero);
538 >      fill(atomColData.electricField.begin(),
539 >           atomColData.electricField.end(), V3Zero);
540 >    }
541 >
542 >    if (storageLayout_ & DataStorage::dslFlucQForce) {    
543 >      fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(),
544 >           0.0);
545 >      fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.end(),
546 >           0.0);
547 >    }
548 >
549 > #endif
550 >    // even in parallel, we need to zero out the local arrays:
551 >
552      if (storageLayout_ & DataStorage::dslParticlePot) {      
553        fill(snap_->atomData.particlePot.begin(),
554             snap_->atomData.particlePot.end(), 0.0);
# Line 467 | Line 558 | namespace OpenMD {
558        fill(snap_->atomData.density.begin(),
559             snap_->atomData.density.end(), 0.0);
560      }
561 +
562      if (storageLayout_ & DataStorage::dslFunctional) {
563        fill(snap_->atomData.functional.begin(),
564             snap_->atomData.functional.end(), 0.0);
565      }
566 +
567      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
568        fill(snap_->atomData.functionalDerivative.begin(),
569             snap_->atomData.functionalDerivative.end(), 0.0);
570      }
571 < #endif
572 <    
571 >
572 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
573 >      fill(snap_->atomData.skippedCharge.begin(),
574 >           snap_->atomData.skippedCharge.end(), 0.0);
575 >    }
576 >
577 >    if (storageLayout_ & DataStorage::dslElectricField) {      
578 >      fill(snap_->atomData.electricField.begin(),
579 >           snap_->atomData.electricField.end(), V3Zero);
580 >    }
581    }
582  
583  
# Line 486 | Line 587 | namespace OpenMD {
587   #ifdef IS_MPI
588      
589      // gather up the atomic positions
590 <    AtomCommVectorRow->gather(snap_->atomData.position,
590 >    AtomPlanVectorRow->gather(snap_->atomData.position,
591                                atomRowData.position);
592 <    AtomCommVectorColumn->gather(snap_->atomData.position,
592 >    AtomPlanVectorColumn->gather(snap_->atomData.position,
593                                   atomColData.position);
594      
595      // gather up the cutoff group positions
596 <    cgCommVectorRow->gather(snap_->cgData.position,
596 >
597 >    cgPlanVectorRow->gather(snap_->cgData.position,
598                              cgRowData.position);
599 <    cgCommVectorColumn->gather(snap_->cgData.position,
599 >
600 >    cgPlanVectorColumn->gather(snap_->cgData.position,
601                                 cgColData.position);
602 +
603      
604      // if needed, gather the atomic rotation matrices
605      if (storageLayout_ & DataStorage::dslAmat) {
606 <      AtomCommMatrixRow->gather(snap_->atomData.aMat,
606 >      AtomPlanMatrixRow->gather(snap_->atomData.aMat,
607                                  atomRowData.aMat);
608 <      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
608 >      AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
609                                     atomColData.aMat);
610      }
611      
612      // if needed, gather the atomic eletrostatic frames
613      if (storageLayout_ & DataStorage::dslElectroFrame) {
614 <      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
614 >      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
615                                  atomRowData.electroFrame);
616 <      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
616 >      AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
617                                     atomColData.electroFrame);
618      }
619 +
620 +    // if needed, gather the atomic fluctuating charge values
621 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {
622 +      AtomPlanRealRow->gather(snap_->atomData.flucQPos,
623 +                              atomRowData.flucQPos);
624 +      AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
625 +                                 atomColData.flucQPos);
626 +    }
627 +
628   #endif      
629    }
630    
# Line 525 | Line 638 | namespace OpenMD {
638      
639      if (storageLayout_ & DataStorage::dslDensity) {
640        
641 <      AtomCommRealRow->scatter(atomRowData.density,
641 >      AtomPlanRealRow->scatter(atomRowData.density,
642                                 snap_->atomData.density);
643        
644        int n = snap_->atomData.density.size();
645        vector<RealType> rho_tmp(n, 0.0);
646 <      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
646 >      AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
647        for (int i = 0; i < n; i++)
648          snap_->atomData.density[i] += rho_tmp[i];
649 +    }
650 +
651 +    if (storageLayout_ & DataStorage::dslElectricField) {
652 +      
653 +      AtomPlanVectorRow->scatter(atomRowData.electricField,
654 +                                 snap_->atomData.electricField);
655 +      
656 +      int n = snap_->atomData.electricField.size();
657 +      vector<Vector3d> field_tmp(n, V3Zero);
658 +      AtomPlanVectorColumn->scatter(atomColData.electricField, field_tmp);
659 +      for (int i = 0; i < n; i++)
660 +        snap_->atomData.electricField[i] += field_tmp[i];
661      }
662   #endif
663    }
# Line 546 | Line 671 | namespace OpenMD {
671      storageLayout_ = sman_->getStorageLayout();
672   #ifdef IS_MPI
673      if (storageLayout_ & DataStorage::dslFunctional) {
674 <      AtomCommRealRow->gather(snap_->atomData.functional,
674 >      AtomPlanRealRow->gather(snap_->atomData.functional,
675                                atomRowData.functional);
676 <      AtomCommRealColumn->gather(snap_->atomData.functional,
676 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
677                                   atomColData.functional);
678      }
679      
680      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
681 <      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
681 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
682                                atomRowData.functionalDerivative);
683 <      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
683 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
684                                   atomColData.functionalDerivative);
685      }
686   #endif
# Line 569 | Line 694 | namespace OpenMD {
694      int n = snap_->atomData.force.size();
695      vector<Vector3d> frc_tmp(n, V3Zero);
696      
697 <    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
697 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
698      for (int i = 0; i < n; i++) {
699        snap_->atomData.force[i] += frc_tmp[i];
700        frc_tmp[i] = 0.0;
701      }
702      
703 <    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
704 <    for (int i = 0; i < n; i++)
703 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
704 >    for (int i = 0; i < n; i++) {
705        snap_->atomData.force[i] += frc_tmp[i];
706 <    
707 <    
706 >    }
707 >        
708      if (storageLayout_ & DataStorage::dslTorque) {
709  
710 <      int nt = snap_->atomData.force.size();
710 >      int nt = snap_->atomData.torque.size();
711        vector<Vector3d> trq_tmp(nt, V3Zero);
712  
713 <      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
714 <      for (int i = 0; i < n; i++) {
713 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
714 >      for (int i = 0; i < nt; i++) {
715          snap_->atomData.torque[i] += trq_tmp[i];
716          trq_tmp[i] = 0.0;
717        }
718        
719 <      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
720 <      for (int i = 0; i < n; i++)
719 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
720 >      for (int i = 0; i < nt; i++)
721          snap_->atomData.torque[i] += trq_tmp[i];
722      }
723 +
724 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
725 +
726 +      int ns = snap_->atomData.skippedCharge.size();
727 +      vector<RealType> skch_tmp(ns, 0.0);
728 +
729 +      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
730 +      for (int i = 0; i < ns; i++) {
731 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
732 +        skch_tmp[i] = 0.0;
733 +      }
734 +      
735 +      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
736 +      for (int i = 0; i < ns; i++)
737 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
738 +            
739 +    }
740      
741 +    if (storageLayout_ & DataStorage::dslFlucQForce) {
742 +
743 +      int nq = snap_->atomData.flucQFrc.size();
744 +      vector<RealType> fqfrc_tmp(nq, 0.0);
745 +
746 +      AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
747 +      for (int i = 0; i < nq; i++) {
748 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
749 +        fqfrc_tmp[i] = 0.0;
750 +      }
751 +      
752 +      AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
753 +      for (int i = 0; i < nq; i++)
754 +        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
755 +            
756 +    }
757 +
758      nLocal_ = snap_->getNumberOfAtoms();
759  
760      vector<potVec> pot_temp(nLocal_,
# Line 603 | Line 762 | namespace OpenMD {
762  
763      // scatter/gather pot_row into the members of my column
764            
765 <    AtomCommPotRow->scatter(pot_row, pot_temp);
765 >    AtomPlanPotRow->scatter(pot_row, 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      fill(pot_temp.begin(), pot_temp.end(),
771           Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
772        
773 <    AtomCommPotColumn->scatter(pot_col, pot_temp);    
773 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
774      
775      for (int ii = 0;  ii < pot_temp.size(); ii++ )
776 <      pot_local += pot_temp[ii];
776 >      pairwisePot += pot_temp[ii];    
777      
778 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
779 +      RealType ploc1 = pairwisePot[ii];
780 +      RealType ploc2 = 0.0;
781 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
782 +      pairwisePot[ii] = ploc2;
783 +    }
784 +
785 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
786 +      RealType ploc1 = embeddingPot[ii];
787 +      RealType ploc2 = 0.0;
788 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
789 +      embeddingPot[ii] = ploc2;
790 +    }
791 +
792   #endif
793 +
794    }
795  
796    int ForceMatrixDecomposition::getNAtomsInRow() {  
# Line 717 | Line 891 | namespace OpenMD {
891      return d;    
892    }
893  
894 <  vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
895 <    return skipsForAtom[atom1];
894 >  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
895 >    return excludesForAtom[atom1];
896    }
897  
898    /**
899 <   * 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
899 >   * We need to exclude some overcounted interactions that result from
900     * the parallel decomposition.
901     */
902    bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
903      int unique_id_1, unique_id_2;
904 <
904 >        
905   #ifdef IS_MPI
906      // in MPI, we have to look up the unique IDs for each atom
907      unique_id_1 = AtomRowToGlobal[atom1];
908      unique_id_2 = AtomColToGlobal[atom2];
909 + #else
910 +    unique_id_1 = AtomLocalToGlobal[atom1];
911 +    unique_id_2 = AtomLocalToGlobal[atom2];
912 + #endif  
913  
739 –    // this situation should only arise in MPI simulations
914      if (unique_id_1 == unique_id_2) return true;
915 <    
915 >
916 > #ifdef IS_MPI
917      // this prevents us from doing the pair on multiple processors
918      if (unique_id_1 < unique_id_2) {
919        if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
920      } else {
921 <      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
921 >      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
922      }
748 – #else
749 –    // in the normal loop, the atom numbers are unique
750 –    unique_id_1 = atom1;
751 –    unique_id_2 = atom2;
923   #endif
924      
925 <    for (vector<int>::iterator i = skipsForAtom[atom1].begin();
926 <         i != skipsForAtom[atom1].end(); ++i) {
756 <      if ( (*i) == unique_id_2 ) return true;
757 <    }    
925 >    return false;
926 >  }
927  
928 +  /**
929 +   * We need to handle the interactions for atoms who are involved in
930 +   * the same rigid body as well as some short range interactions
931 +   * (bonds, bends, torsions) differently from other interactions.
932 +   * We'll still visit the pairwise routines, but with a flag that
933 +   * tells those routines to exclude the pair from direct long range
934 +   * interactions.  Some indirect interactions (notably reaction
935 +   * field) must still be handled for these pairs.
936 +   */
937 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
938 +
939 +    // excludesForAtom was constructed to use row/column indices in the MPI
940 +    // version, and to use local IDs in the non-MPI version:
941 +    
942 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
943 +         i != excludesForAtom[atom1].end(); ++i) {
944 +      if ( (*i) == atom2 ) return true;
945 +    }
946 +
947 +    return false;
948    }
949  
950  
# Line 776 | Line 965 | namespace OpenMD {
965    }
966  
967      // filling interaction blocks with pointers
968 <  void ForceMatrixDecomposition::fillInteractionData(InteractionData idat,
969 <                                                     int atom1, int atom2) {    
968 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
969 >                                                     int atom1, int atom2) {
970 >
971 >    idat.excluded = excludeAtomPair(atom1, atom2);
972 >  
973   #ifdef IS_MPI
974 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
975 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
976 +    //                         ff_->getAtomType(identsCol[atom2]) );
977      
783 –    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
784 –                             ff_->getAtomType(identsCol[atom2]) );
785 –    
978      if (storageLayout_ & DataStorage::dslAmat) {
979        idat.A1 = &(atomRowData.aMat[atom1]);
980        idat.A2 = &(atomColData.aMat[atom2]);
# Line 818 | Line 1010 | namespace OpenMD {
1010        idat.particlePot2 = &(atomColData.particlePot[atom2]);
1011      }
1012  
1013 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
1014 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1015 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1016 +    }
1017 +
1018 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1019 +      idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1020 +      idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1021 +    }
1022 +
1023   #else
1024 +    
1025  
1026 <    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
1027 <                             ff_->getAtomType(identsLocal[atom2]) );
1026 >    // cerr << "atoms = " << atom1 << " " << atom2 << "\n";
1027 >    // cerr << "pos1 = " << snap_->atomData.position[atom1] << "\n";
1028 >    // cerr << "pos2 = " << snap_->atomData.position[atom2] << "\n";
1029  
1030 +    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1031 +    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1032 +    //                         ff_->getAtomType(idents[atom2]) );
1033 +
1034      if (storageLayout_ & DataStorage::dslAmat) {
1035        idat.A1 = &(snap_->atomData.aMat[atom1]);
1036        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 838 | Line 1046 | namespace OpenMD {
1046        idat.t2 = &(snap_->atomData.torque[atom2]);
1047      }
1048  
1049 <    if (storageLayout_ & DataStorage::dslDensity) {
1049 >    if (storageLayout_ & DataStorage::dslDensity) {    
1050        idat.rho1 = &(snap_->atomData.density[atom1]);
1051        idat.rho2 = &(snap_->atomData.density[atom2]);
1052      }
# Line 858 | Line 1066 | namespace OpenMD {
1066        idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1067      }
1068  
1069 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
1070 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1071 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1072 +    }
1073 +
1074 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1075 +      idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1076 +      idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1077 +    }
1078 +
1079   #endif
1080    }
1081  
1082    
1083 <  void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {    
1083 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1084   #ifdef IS_MPI
1085 <    pot_row[atom1] += 0.5 *  *(idat.pot);
1086 <    pot_col[atom2] += 0.5 *  *(idat.pot);
1085 >    pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1086 >    pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1087  
1088      atomRowData.force[atom1] += *(idat.f1);
1089      atomColData.force[atom2] -= *(idat.f1);
1090 +
1091 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1092 +      atomRowData.flucQFrc[atom1] += *(idat.dVdFQ1);
1093 +      atomColData.flucQFrc[atom2] += *(idat.dVdFQ2);
1094 +    }
1095 +
1096 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1097 +      atomRowData.electricField[atom1] += *(idat.eField1);
1098 +      atomColData.electricField[atom2] += *(idat.eField2);
1099 +    }
1100 +
1101 +    // should particle pot be done here also?
1102   #else
1103 <    longRangePot_ += *(idat.pot);
1104 <    
1103 >    pairwisePot += *(idat.pot);
1104 >
1105      snap_->atomData.force[atom1] += *(idat.f1);
1106      snap_->atomData.force[atom2] -= *(idat.f1);
877 – #endif
1107  
1108 <  }
1108 >    if (idat.doParticlePot) {
1109 >      snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1110 >      snap_->atomData.particlePot[atom2] -= *(idat.vpair) * *(idat.sw);
1111 >    }
1112 >    
1113 >    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1114 >      snap_->atomData.flucQFrc[atom1] += *(idat.dVdFQ1);
1115 >      snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1116 >    }
1117  
1118 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1119 +      snap_->atomData.electricField[atom1] += *(idat.eField1);
1120 +      snap_->atomData.electricField[atom2] += *(idat.eField2);
1121 +    }
1122  
1123 <  void ForceMatrixDecomposition::fillSkipData(InteractionData idat,
1124 <                                              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    
1123 > #endif
1124 >    
1125    }
1126  
1127    /*
# Line 918 | Line 1134 | namespace OpenMD {
1134        
1135      vector<pair<int, int> > neighborList;
1136      groupCutoffs cuts;
1137 +    bool doAllPairs = false;
1138 +
1139   #ifdef IS_MPI
1140      cellListRow_.clear();
1141      cellListCol_.clear();
# Line 937 | Line 1155 | namespace OpenMD {
1155      nCells_.y() = (int) ( Hy.length() )/ rList_;
1156      nCells_.z() = (int) ( Hz.length() )/ rList_;
1157  
1158 +    // handle small boxes where the cell offsets can end up repeating cells
1159 +    
1160 +    if (nCells_.x() < 3) doAllPairs = true;
1161 +    if (nCells_.y() < 3) doAllPairs = true;
1162 +    if (nCells_.z() < 3) doAllPairs = true;
1163 +
1164      Mat3x3d invHmat = snap_->getInvHmat();
1165      Vector3d rs, scaled, dr;
1166      Vector3i whichCell;
1167      int cellIndex;
1168      int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1169  
946 –    cerr << "flag1\n";
1170   #ifdef IS_MPI
1171      cellListRow_.resize(nCtot);
1172      cellListCol_.resize(nCtot);
1173   #else
1174      cellList_.resize(nCtot);
1175   #endif
1176 <    cerr << "flag2\n";
1176 >
1177 >    if (!doAllPairs) {
1178   #ifdef IS_MPI
955 –    for (int i = 0; i < nGroupsInRow_; i++) {
956 –      rs = cgRowData.position[i];
1179  
1180 <      // scaled positions relative to the box vectors
1181 <      scaled = invHmat * rs;
1182 <
1183 <      // wrap the vector back into the unit box by subtracting integer box
1184 <      // numbers
1185 <      for (int j = 0; j < 3; j++) {
1186 <        scaled[j] -= roundMe(scaled[j]);
1187 <        scaled[j] += 0.5;
1180 >      for (int i = 0; i < nGroupsInRow_; i++) {
1181 >        rs = cgRowData.position[i];
1182 >        
1183 >        // scaled positions relative to the box vectors
1184 >        scaled = invHmat * rs;
1185 >        
1186 >        // wrap the vector back into the unit box by subtracting integer box
1187 >        // numbers
1188 >        for (int j = 0; j < 3; j++) {
1189 >          scaled[j] -= roundMe(scaled[j]);
1190 >          scaled[j] += 0.5;
1191 >        }
1192 >        
1193 >        // find xyz-indices of cell that cutoffGroup is in.
1194 >        whichCell.x() = nCells_.x() * scaled.x();
1195 >        whichCell.y() = nCells_.y() * scaled.y();
1196 >        whichCell.z() = nCells_.z() * scaled.z();
1197 >        
1198 >        // find single index of this cell:
1199 >        cellIndex = Vlinear(whichCell, nCells_);
1200 >        
1201 >        // add this cutoff group to the list of groups in this cell;
1202 >        cellListRow_[cellIndex].push_back(i);
1203        }
1204 <    
1205 <      // find xyz-indices of cell that cutoffGroup is in.
1206 <      whichCell.x() = nCells_.x() * scaled.x();
1207 <      whichCell.y() = nCells_.y() * scaled.y();
1208 <      whichCell.z() = nCells_.z() * scaled.z();
1209 <
1210 <      // find single index of this cell:
1211 <      cellIndex = Vlinear(whichCell, nCells_);
1212 <
1213 <      // add this cutoff group to the list of groups in this cell;
1214 <      cellListRow_[cellIndex].push_back(i);
1215 <    }
1216 <
1217 <    for (int i = 0; i < nGroupsInCol_; i++) {
1218 <      rs = cgColData.position[i];
1219 <
1220 <      // scaled positions relative to the box vectors
1221 <      scaled = invHmat * rs;
1222 <
1223 <      // wrap the vector back into the unit box by subtracting integer box
1224 <      // numbers
1225 <      for (int j = 0; j < 3; j++) {
1226 <        scaled[j] -= roundMe(scaled[j]);
1227 <        scaled[j] += 0.5;
1228 <      }
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 <    }
1204 >      for (int i = 0; i < nGroupsInCol_; i++) {
1205 >        rs = cgColData.position[i];
1206 >        
1207 >        // scaled positions relative to the box vectors
1208 >        scaled = invHmat * rs;
1209 >        
1210 >        // wrap the vector back into the unit box by subtracting integer box
1211 >        // numbers
1212 >        for (int j = 0; j < 3; j++) {
1213 >          scaled[j] -= roundMe(scaled[j]);
1214 >          scaled[j] += 0.5;
1215 >        }
1216 >        
1217 >        // find xyz-indices of cell that cutoffGroup is in.
1218 >        whichCell.x() = nCells_.x() * scaled.x();
1219 >        whichCell.y() = nCells_.y() * scaled.y();
1220 >        whichCell.z() = nCells_.z() * scaled.z();
1221 >        
1222 >        // find single index of this cell:
1223 >        cellIndex = Vlinear(whichCell, nCells_);
1224 >        
1225 >        // add this cutoff group to the list of groups in this cell;
1226 >        cellListCol_[cellIndex].push_back(i);
1227 >      }
1228 >    
1229   #else
1230 <    for (int i = 0; i < nGroups_; i++) {
1231 <      rs = snap_->cgData.position[i];
1232 <
1233 <      // scaled positions relative to the box vectors
1234 <      scaled = invHmat * rs;
1235 <
1236 <      // wrap the vector back into the unit box by subtracting integer box
1237 <      // numbers
1238 <      for (int j = 0; j < 3; j++) {
1239 <        scaled[j] -= roundMe(scaled[j]);
1240 <        scaled[j] += 0.5;
1230 >      for (int i = 0; i < nGroups_; i++) {
1231 >        rs = snap_->cgData.position[i];
1232 >        
1233 >        // scaled positions relative to the box vectors
1234 >        scaled = invHmat * rs;
1235 >        
1236 >        // wrap the vector back into the unit box by subtracting integer box
1237 >        // numbers
1238 >        for (int j = 0; j < 3; j++) {
1239 >          scaled[j] -= roundMe(scaled[j]);
1240 >          scaled[j] += 0.5;
1241 >        }
1242 >        
1243 >        // find xyz-indices of cell that cutoffGroup is in.
1244 >        whichCell.x() = nCells_.x() * scaled.x();
1245 >        whichCell.y() = nCells_.y() * scaled.y();
1246 >        whichCell.z() = nCells_.z() * scaled.z();
1247 >        
1248 >        // find single index of this cell:
1249 >        cellIndex = Vlinear(whichCell, nCells_);
1250 >        
1251 >        // add this cutoff group to the list of groups in this cell;
1252 >        cellList_[cellIndex].push_back(i);
1253        }
1254  
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 –    }
1255   #endif
1256  
1257 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1258 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1259 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1260 <          Vector3i m1v(m1x, m1y, m1z);
1261 <          int m1 = Vlinear(m1v, nCells_);
1036 <
1037 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1038 <               os != cellOffsets_.end(); ++os) {
1257 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1258 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1259 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1260 >            Vector3i m1v(m1x, m1y, m1z);
1261 >            int m1 = Vlinear(m1v, nCells_);
1262              
1263 <            Vector3i m2v = m1v + (*os);
1264 <            
1265 <            if (m2v.x() >= nCells_.x()) {
1266 <              m2v.x() = 0;          
1267 <            } 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_);
1263 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1264 >                 os != cellOffsets_.end(); ++os) {
1265 >              
1266 >              Vector3i m2v = m1v + (*os);
1267 >            
1268  
1269 < #ifdef IS_MPI
1270 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1271 <                 j1 != cellListRow_[m1].end(); ++j1) {
1272 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1273 <                   j2 != cellListCol_[m2].end(); ++j2) {
1274 <                              
1275 <                // Always do this if we're in different cells or if
1276 <                // we're in the same cell and the global index of the
1277 <                // j2 cutoff group is less than the j1 cutoff group
1269 >              if (m2v.x() >= nCells_.x()) {
1270 >                m2v.x() = 0;          
1271 >              } else if (m2v.x() < 0) {
1272 >                m2v.x() = nCells_.x() - 1;
1273 >              }
1274 >              
1275 >              if (m2v.y() >= nCells_.y()) {
1276 >                m2v.y() = 0;          
1277 >              } else if (m2v.y() < 0) {
1278 >                m2v.y() = nCells_.y() - 1;
1279 >              }
1280 >              
1281 >              if (m2v.z() >= nCells_.z()) {
1282 >                m2v.z() = 0;          
1283 >              } else if (m2v.z() < 0) {
1284 >                m2v.z() = nCells_.z() - 1;
1285 >              }
1286  
1287 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1287 >              int m2 = Vlinear (m2v, nCells_);
1288 >              
1289 > #ifdef IS_MPI
1290 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1291 >                   j1 != cellListRow_[m1].end(); ++j1) {
1292 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1293 >                     j2 != cellListCol_[m2].end(); ++j2) {
1294 >                  
1295 >                  // In parallel, we need to visit *all* pairs of row
1296 >                  // & column indicies and will divide labor in the
1297 >                  // force evaluation later.
1298                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1299                    snap_->wrapVector(dr);
1300                    cuts = getGroupCutoffs( (*j1), (*j2) );
1301                    if (dr.lengthSquare() < cuts.third) {
1302                      neighborList.push_back(make_pair((*j1), (*j2)));
1303 <                  }
1303 >                  }                  
1304                  }
1305                }
1081 –            }
1306   #else
1307 +              for (vector<int>::iterator j1 = cellList_[m1].begin();
1308 +                   j1 != cellList_[m1].end(); ++j1) {
1309 +                for (vector<int>::iterator j2 = cellList_[m2].begin();
1310 +                     j2 != cellList_[m2].end(); ++j2) {
1311 +    
1312 +                  // Always do this if we're in different cells or if
1313 +                  // we're in the same cell and the global index of
1314 +                  // the j2 cutoff group is greater than or equal to
1315 +                  // the j1 cutoff group.  Note that Rappaport's code
1316 +                  // has a "less than" conditional here, but that
1317 +                  // deals with atom-by-atom computation.  OpenMD
1318 +                  // allows atoms within a single cutoff group to
1319 +                  // interact with each other.
1320  
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) {
1321  
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
1322  
1323 <                if (m2 != m1 || (*j2) < (*j1)) {
1324 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1325 <                  snap_->wrapVector(dr);
1326 <                  cuts = getGroupCutoffs( (*j1), (*j2) );
1327 <                  if (dr.lengthSquare() < cuts.third) {
1328 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1323 >                  if (m2 != m1 || (*j2) >= (*j1) ) {
1324 >
1325 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1326 >                    snap_->wrapVector(dr);
1327 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1328 >                    if (dr.lengthSquare() < cuts.third) {
1329 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1330 >                    }
1331                    }
1332                  }
1333                }
1102 –            }
1334   #endif
1335 +            }
1336            }
1337          }
1338        }
1339 +    } else {
1340 +      // branch to do all cutoff group pairs
1341 + #ifdef IS_MPI
1342 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1343 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
1344 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1345 +          snap_->wrapVector(dr);
1346 +          cuts = getGroupCutoffs( j1, j2 );
1347 +          if (dr.lengthSquare() < cuts.third) {
1348 +            neighborList.push_back(make_pair(j1, j2));
1349 +          }
1350 +        }
1351 +      }      
1352 + #else
1353 +      // include all groups here.
1354 +      for (int j1 = 0; j1 < nGroups_; j1++) {
1355 +        // include self group interactions j2 == j1
1356 +        for (int j2 = j1; j2 < nGroups_; j2++) {
1357 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1358 +          snap_->wrapVector(dr);
1359 +          cuts = getGroupCutoffs( j1, j2 );
1360 +          if (dr.lengthSquare() < cuts.third) {
1361 +            neighborList.push_back(make_pair(j1, j2));
1362 +          }
1363 +        }    
1364 +      }
1365 + #endif
1366      }
1367 <    
1367 >      
1368      // save the local cutoff group positions for the check that is
1369      // done on each loop:
1370      saved_CG_positions_.clear();

Diff Legend

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