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 1569 by gezelter, Thu May 26 13:55:04 2011 UTC vs.
Revision 1592 by gezelter, Tue Jul 12 20:33:14 2011 UTC

# Line 42 | Line 42
42   #include "math/SquareMatrix3.hpp"
43   #include "nonbonded/NonBondedInteraction.hpp"
44   #include "brains/SnapshotManager.hpp"
45 + #include "brains/PairList.hpp"
46  
47   using namespace std;
48   namespace OpenMD {
# Line 54 | Line 55 | namespace OpenMD {
55    void ForceMatrixDecomposition::distributeInitialData() {
56      snap_ = sman_->getCurrentSnapshot();
57      storageLayout_ = sman_->getStorageLayout();
58 +    ff_ = info_->getForceField();
59      nLocal_ = snap_->getNumberOfAtoms();
60 <    nGroups_ = snap_->getNumberOfCutoffGroups();
61 <
60 >    
61 >    nGroups_ = info_->getNLocalCutoffGroups();
62      // gather the information for atomtype IDs (atids):
63 <    vector<int> identsLocal = info_->getIdentArray();
63 >    idents = info_->getIdentArray();
64      AtomLocalToGlobal = info_->getGlobalAtomIndices();
65      cgLocalToGlobal = info_->getGlobalGroupIndices();
66      vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
65 –    vector<RealType> massFactorsLocal = info_->getMassFactors();
66 –    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
67  
68 +    massFactors = info_->getMassFactors();
69 +
70 +    PairList* excludes = info_->getExcludedInteractions();
71 +    PairList* oneTwo = info_->getOneTwoInteractions();
72 +    PairList* oneThree = info_->getOneThreeInteractions();
73 +    PairList* oneFour = info_->getOneFourInteractions();
74 +
75   #ifdef IS_MPI
76  
77      AtomCommIntRow = new Communicator<Row,int>(nLocal_);
78      AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
79      AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
80      AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
81 +    AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
82  
83      AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
84      AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
85      AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
86      AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
87 +    AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
88  
89      cgCommIntRow = new Communicator<Row,int>(nGroups_);
90      cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
# Line 96 | Line 105 | namespace OpenMD {
105      cgRowData.setStorageLayout(DataStorage::dslPosition);
106      cgColData.resize(nGroupsInCol_);
107      cgColData.setStorageLayout(DataStorage::dslPosition);
108 +        
109 +    identsRow.resize(nAtomsInRow_);
110 +    identsCol.resize(nAtomsInCol_);
111      
112 <    vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
113 <                                      vector<RealType> (nAtomsInRow_, 0.0));
102 <    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
103 <                                      vector<RealType> (nAtomsInCol_, 0.0));
112 >    AtomCommIntRow->gather(idents, identsRow);
113 >    AtomCommIntColumn->gather(idents, identsCol);
114      
115 <    identsRow.reserve(nAtomsInRow_);
116 <    identsCol.reserve(nAtomsInCol_);
117 <    
118 <    AtomCommIntRow->gather(identsLocal, identsRow);
119 <    AtomCommIntColumn->gather(identsLocal, identsCol);
120 <    
115 >    // allocate memory for the parallel objects
116 >    atypesRow.resize(nAtomsInRow_);
117 >    atypesCol.resize(nAtomsInCol_);
118 >
119 >    for (int i = 0; i < nAtomsInRow_; i++)
120 >      atypesRow[i] = ff_->getAtomType(identsRow[i]);
121 >    for (int i = 0; i < nAtomsInCol_; i++)
122 >      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
123 >
124 >    pot_row.resize(nAtomsInRow_);
125 >    pot_col.resize(nAtomsInCol_);
126 >
127 >    AtomRowToGlobal.resize(nAtomsInRow_);
128 >    AtomColToGlobal.resize(nAtomsInCol_);
129      AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
130      AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
131      
132 +    cgRowToGlobal.resize(nGroupsInRow_);
133 +    cgColToGlobal.resize(nGroupsInCol_);
134      cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
135      cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
136  
137 <    AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
138 <    AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
137 >    massFactorsRow.resize(nAtomsInRow_);
138 >    massFactorsCol.resize(nAtomsInCol_);
139 >    AtomCommRealRow->gather(massFactors, massFactorsRow);
140 >    AtomCommRealColumn->gather(massFactors, massFactorsCol);
141  
142      groupListRow_.clear();
143 <    groupListRow_.reserve(nGroupsInRow_);
143 >    groupListRow_.resize(nGroupsInRow_);
144      for (int i = 0; i < nGroupsInRow_; i++) {
145        int gid = cgRowToGlobal[i];
146        for (int j = 0; j < nAtomsInRow_; j++) {
# Line 129 | Line 151 | namespace OpenMD {
151      }
152  
153      groupListCol_.clear();
154 <    groupListCol_.reserve(nGroupsInCol_);
154 >    groupListCol_.resize(nGroupsInCol_);
155      for (int i = 0; i < nGroupsInCol_; i++) {
156        int gid = cgColToGlobal[i];
157        for (int j = 0; j < nAtomsInCol_; j++) {
# Line 139 | Line 161 | namespace OpenMD {
161        }      
162      }
163  
164 +    excludesForAtom.clear();
165 +    excludesForAtom.resize(nAtomsInRow_);
166 +    toposForAtom.clear();
167 +    toposForAtom.resize(nAtomsInRow_);
168 +    topoDist.clear();
169 +    topoDist.resize(nAtomsInRow_);
170 +    for (int i = 0; i < nAtomsInRow_; i++) {
171 +      int iglob = AtomRowToGlobal[i];
172 +
173 +      for (int j = 0; j < nAtomsInCol_; j++) {
174 +        int jglob = AtomColToGlobal[j];
175 +
176 +        if (excludes->hasPair(iglob, jglob))
177 +          excludesForAtom[i].push_back(j);      
178 +        
179 +        if (oneTwo->hasPair(iglob, jglob)) {
180 +          toposForAtom[i].push_back(j);
181 +          topoDist[i].push_back(1);
182 +        } else {
183 +          if (oneThree->hasPair(iglob, jglob)) {
184 +            toposForAtom[i].push_back(j);
185 +            topoDist[i].push_back(2);
186 +          } else {
187 +            if (oneFour->hasPair(iglob, jglob)) {
188 +              toposForAtom[i].push_back(j);
189 +              topoDist[i].push_back(3);
190 +            }
191 +          }
192 +        }
193 +      }      
194 +    }
195 +
196   #endif
197  
198 +    // allocate memory for the parallel objects
199 +    atypesLocal.resize(nLocal_);
200 +
201 +    for (int i = 0; i < nLocal_; i++)
202 +      atypesLocal[i] = ff_->getAtomType(idents[i]);
203 +
204      groupList_.clear();
205 <    groupList_.reserve(nGroups_);
205 >    groupList_.resize(nGroups_);
206      for (int i = 0; i < nGroups_; i++) {
207        int gid = cgLocalToGlobal[i];
208        for (int j = 0; j < nLocal_; j++) {
209          int aid = AtomLocalToGlobal[j];
210 <        if (globalGroupMembership[aid] == gid)
210 >        if (globalGroupMembership[aid] == gid) {
211            groupList_[i].push_back(j);
212 +        }
213        }      
214      }
215  
216 +    excludesForAtom.clear();
217 +    excludesForAtom.resize(nLocal_);
218 +    toposForAtom.clear();
219 +    toposForAtom.resize(nLocal_);
220 +    topoDist.clear();
221 +    topoDist.resize(nLocal_);
222 +
223 +    for (int i = 0; i < nLocal_; i++) {
224 +      int iglob = AtomLocalToGlobal[i];
225 +
226 +      for (int j = 0; j < nLocal_; j++) {
227 +        int jglob = AtomLocalToGlobal[j];
228 +
229 +        if (excludes->hasPair(iglob, jglob))
230 +          excludesForAtom[i].push_back(j);              
231 +        
232 +        if (oneTwo->hasPair(iglob, jglob)) {
233 +          toposForAtom[i].push_back(j);
234 +          topoDist[i].push_back(1);
235 +        } else {
236 +          if (oneThree->hasPair(iglob, jglob)) {
237 +            toposForAtom[i].push_back(j);
238 +            topoDist[i].push_back(2);
239 +          } else {
240 +            if (oneFour->hasPair(iglob, jglob)) {
241 +              toposForAtom[i].push_back(j);
242 +              topoDist[i].push_back(3);
243 +            }
244 +          }
245 +        }
246 +      }      
247 +    }
248 +    
249 +    createGtypeCutoffMap();
250 +
251 +  }
252    
253 <    // still need:
254 <    // topoDist
255 <    // exclude
253 >  void ForceMatrixDecomposition::createGtypeCutoffMap() {
254 >    
255 >    RealType tol = 1e-6;
256 >    largestRcut_ = 0.0;
257 >    RealType rc;
258 >    int atid;
259 >    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
260 >    
261 >    map<int, RealType> atypeCutoff;
262 >      
263 >    for (set<AtomType*>::iterator at = atypes.begin();
264 >         at != atypes.end(); ++at){
265 >      atid = (*at)->getIdent();
266 >      if (userChoseCutoff_)
267 >        atypeCutoff[atid] = userCutoff_;
268 >      else
269 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
270 >    }
271 >    
272 >    vector<RealType> gTypeCutoffs;
273 >    // first we do a single loop over the cutoff groups to find the
274 >    // largest cutoff for any atypes present in this group.
275 > #ifdef IS_MPI
276 >    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
277 >    groupRowToGtype.resize(nGroupsInRow_);
278 >    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
279 >      vector<int> atomListRow = getAtomsInGroupRow(cg1);
280 >      for (vector<int>::iterator ia = atomListRow.begin();
281 >           ia != atomListRow.end(); ++ia) {            
282 >        int atom1 = (*ia);
283 >        atid = identsRow[atom1];
284 >        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
285 >          groupCutoffRow[cg1] = atypeCutoff[atid];
286 >        }
287 >      }
288  
289 +      bool gTypeFound = false;
290 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
291 +        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
292 +          groupRowToGtype[cg1] = gt;
293 +          gTypeFound = true;
294 +        }
295 +      }
296 +      if (!gTypeFound) {
297 +        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
298 +        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
299 +      }
300 +      
301 +    }
302 +    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
303 +    groupColToGtype.resize(nGroupsInCol_);
304 +    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
305 +      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
306 +      for (vector<int>::iterator jb = atomListCol.begin();
307 +           jb != atomListCol.end(); ++jb) {            
308 +        int atom2 = (*jb);
309 +        atid = identsCol[atom2];
310 +        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
311 +          groupCutoffCol[cg2] = atypeCutoff[atid];
312 +        }
313 +      }
314 +      bool gTypeFound = false;
315 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
316 +        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
317 +          groupColToGtype[cg2] = gt;
318 +          gTypeFound = true;
319 +        }
320 +      }
321 +      if (!gTypeFound) {
322 +        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
323 +        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
324 +      }
325 +    }
326 + #else
327 +
328 +    vector<RealType> groupCutoff(nGroups_, 0.0);
329 +    groupToGtype.resize(nGroups_);
330 +    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
331 +      groupCutoff[cg1] = 0.0;
332 +      vector<int> atomList = getAtomsInGroupRow(cg1);
333 +      for (vector<int>::iterator ia = atomList.begin();
334 +           ia != atomList.end(); ++ia) {            
335 +        int atom1 = (*ia);
336 +        atid = idents[atom1];
337 +        if (atypeCutoff[atid] > groupCutoff[cg1])
338 +          groupCutoff[cg1] = atypeCutoff[atid];
339 +      }
340 +      
341 +      bool gTypeFound = false;
342 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
343 +        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
344 +          groupToGtype[cg1] = gt;
345 +          gTypeFound = true;
346 +        }
347 +      }
348 +      if (!gTypeFound) {      
349 +        gTypeCutoffs.push_back( groupCutoff[cg1] );
350 +        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
351 +      }      
352 +    }
353 + #endif
354 +
355 +    // Now we find the maximum group cutoff value present in the simulation
356 +
357 +    RealType groupMax = *max_element(gTypeCutoffs.begin(),
358 +                                     gTypeCutoffs.end());
359 +
360 + #ifdef IS_MPI
361 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
362 +                              MPI::MAX);
363 + #endif
364 +    
365 +    RealType tradRcut = groupMax;
366 +
367 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
368 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
369 +        RealType thisRcut;
370 +        switch(cutoffPolicy_) {
371 +        case TRADITIONAL:
372 +          thisRcut = tradRcut;
373 +          break;
374 +        case MIX:
375 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
376 +          break;
377 +        case MAX:
378 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
379 +          break;
380 +        default:
381 +          sprintf(painCave.errMsg,
382 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
383 +                  "hit an unknown cutoff policy!\n");
384 +          painCave.severity = OPENMD_ERROR;
385 +          painCave.isFatal = 1;
386 +          simError();
387 +          break;
388 +        }
389 +
390 +        pair<int,int> key = make_pair(i,j);
391 +        gTypeCutoffMap[key].first = thisRcut;
392 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
393 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
394 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
395 +        // sanity check
396 +        
397 +        if (userChoseCutoff_) {
398 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
399 +            sprintf(painCave.errMsg,
400 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
401 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
402 +            painCave.severity = OPENMD_ERROR;
403 +            painCave.isFatal = 1;
404 +            simError();            
405 +          }
406 +        }
407 +      }
408 +    }
409    }
410 +
411 +
412 +  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
413 +    int i, j;  
414 + #ifdef IS_MPI
415 +    i = groupRowToGtype[cg1];
416 +    j = groupColToGtype[cg2];
417 + #else
418 +    i = groupToGtype[cg1];
419 +    j = groupToGtype[cg2];
420 + #endif    
421 +    return gTypeCutoffMap[make_pair(i,j)];
422 +  }
423 +
424 +  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
425 +    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
426 +      if (toposForAtom[atom1][j] == atom2)
427 +        return topoDist[atom1][j];
428 +    }
429 +    return 0;
430 +  }
431 +
432 +  void ForceMatrixDecomposition::zeroWorkArrays() {
433 +    pairwisePot = 0.0;
434 +    embeddingPot = 0.0;
435 +
436 + #ifdef IS_MPI
437 +    if (storageLayout_ & DataStorage::dslForce) {
438 +      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
439 +      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
440 +    }
441 +
442 +    if (storageLayout_ & DataStorage::dslTorque) {
443 +      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
444 +      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
445 +    }
446      
447 +    fill(pot_row.begin(), pot_row.end(),
448 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
449  
450 +    fill(pot_col.begin(), pot_col.end(),
451 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
452  
453 +    if (storageLayout_ & DataStorage::dslParticlePot) {    
454 +      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
455 +           0.0);
456 +      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
457 +           0.0);
458 +    }
459 +
460 +    if (storageLayout_ & DataStorage::dslDensity) {      
461 +      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
462 +      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
463 +    }
464 +
465 +    if (storageLayout_ & DataStorage::dslFunctional) {  
466 +      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
467 +           0.0);
468 +      fill(atomColData.functional.begin(), atomColData.functional.end(),
469 +           0.0);
470 +    }
471 +
472 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
473 +      fill(atomRowData.functionalDerivative.begin(),
474 +           atomRowData.functionalDerivative.end(), 0.0);
475 +      fill(atomColData.functionalDerivative.begin(),
476 +           atomColData.functionalDerivative.end(), 0.0);
477 +    }
478 +
479 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
480 +      fill(atomRowData.skippedCharge.begin(),
481 +           atomRowData.skippedCharge.end(), 0.0);
482 +      fill(atomColData.skippedCharge.begin(),
483 +           atomColData.skippedCharge.end(), 0.0);
484 +    }
485 +
486 + #endif
487 +    // even in parallel, we need to zero out the local arrays:
488 +
489 +    if (storageLayout_ & DataStorage::dslParticlePot) {      
490 +      fill(snap_->atomData.particlePot.begin(),
491 +           snap_->atomData.particlePot.end(), 0.0);
492 +    }
493 +    
494 +    if (storageLayout_ & DataStorage::dslDensity) {      
495 +      fill(snap_->atomData.density.begin(),
496 +           snap_->atomData.density.end(), 0.0);
497 +    }
498 +    if (storageLayout_ & DataStorage::dslFunctional) {
499 +      fill(snap_->atomData.functional.begin(),
500 +           snap_->atomData.functional.end(), 0.0);
501 +    }
502 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
503 +      fill(snap_->atomData.functionalDerivative.begin(),
504 +           snap_->atomData.functionalDerivative.end(), 0.0);
505 +    }
506 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
507 +      fill(snap_->atomData.skippedCharge.begin(),
508 +           snap_->atomData.skippedCharge.end(), 0.0);
509 +    }
510 +    
511 +  }
512 +
513 +
514    void ForceMatrixDecomposition::distributeData()  {
515      snap_ = sman_->getCurrentSnapshot();
516      storageLayout_ = sman_->getStorageLayout();
# Line 193 | Line 543 | namespace OpenMD {
543        AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
544                                     atomColData.electroFrame);
545      }
546 +
547   #endif      
548    }
549    
550 +  /* collects information obtained during the pre-pair loop onto local
551 +   * data structures.
552 +   */
553    void ForceMatrixDecomposition::collectIntermediateData() {
554      snap_ = sman_->getCurrentSnapshot();
555      storageLayout_ = sman_->getStorageLayout();
# Line 207 | Line 561 | namespace OpenMD {
561                                 snap_->atomData.density);
562        
563        int n = snap_->atomData.density.size();
564 <      std::vector<RealType> rho_tmp(n, 0.0);
564 >      vector<RealType> rho_tmp(n, 0.0);
565        AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
566        for (int i = 0; i < n; i++)
567          snap_->atomData.density[i] += rho_tmp[i];
568      }
569   #endif
570    }
571 <  
571 >
572 >  /*
573 >   * redistributes information obtained during the pre-pair loop out to
574 >   * row and column-indexed data structures
575 >   */
576    void ForceMatrixDecomposition::distributeIntermediateData() {
577      snap_ = sman_->getCurrentSnapshot();
578      storageLayout_ = sman_->getStorageLayout();
# Line 252 | Line 610 | namespace OpenMD {
610      AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
611      for (int i = 0; i < n; i++)
612        snap_->atomData.force[i] += frc_tmp[i];
613 <    
256 <    
613 >        
614      if (storageLayout_ & DataStorage::dslTorque) {
615  
616 <      int nt = snap_->atomData.force.size();
616 >      int nt = snap_->atomData.torque.size();
617        vector<Vector3d> trq_tmp(nt, V3Zero);
618  
619        AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
620 <      for (int i = 0; i < n; i++) {
620 >      for (int i = 0; i < nt; i++) {
621          snap_->atomData.torque[i] += trq_tmp[i];
622          trq_tmp[i] = 0.0;
623        }
624        
625        AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
626 <      for (int i = 0; i < n; i++)
626 >      for (int i = 0; i < nt; i++)
627          snap_->atomData.torque[i] += trq_tmp[i];
628      }
629 +
630 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
631 +
632 +      int ns = snap_->atomData.skippedCharge.size();
633 +      vector<RealType> skch_tmp(ns, 0.0);
634 +
635 +      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
636 +      for (int i = 0; i < ns; i++) {
637 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
638 +        skch_tmp[i] = 0.0;
639 +      }
640 +      
641 +      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
642 +      for (int i = 0; i < ns; i++)
643 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
644 +    }
645      
646      nLocal_ = snap_->getNumberOfAtoms();
647  
648 <    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
649 <                                       vector<RealType> (nLocal_, 0.0));
648 >    vector<potVec> pot_temp(nLocal_,
649 >                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
650 >
651 >    // scatter/gather pot_row into the members of my column
652 >          
653 >    AtomCommPotRow->scatter(pot_row, pot_temp);
654 >
655 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
656 >      pairwisePot += pot_temp[ii];
657      
658 <    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
659 <      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
660 <      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
661 <        pot_local[i] += pot_temp[i][ii];
662 <      }
663 <    }
658 >    fill(pot_temp.begin(), pot_temp.end(),
659 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
660 >      
661 >    AtomCommPotColumn->scatter(pot_col, pot_temp);    
662 >    
663 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
664 >      pairwisePot += pot_temp[ii];    
665   #endif
666 +
667    }
668  
669 +  int ForceMatrixDecomposition::getNAtomsInRow() {  
670 + #ifdef IS_MPI
671 +    return nAtomsInRow_;
672 + #else
673 +    return nLocal_;
674 + #endif
675 +  }
676 +
677    /**
678     * returns the list of atoms belonging to this group.  
679     */
# Line 348 | Line 738 | namespace OpenMD {
738   #ifdef IS_MPI
739      return massFactorsRow[atom1];
740   #else
741 <    return massFactorsLocal[atom1];
741 >    return massFactors[atom1];
742   #endif
743    }
744  
# Line 356 | Line 746 | namespace OpenMD {
746   #ifdef IS_MPI
747      return massFactorsCol[atom2];
748   #else
749 <    return massFactorsLocal[atom2];
749 >    return massFactors[atom2];
750   #endif
751  
752    }
# Line 374 | Line 764 | namespace OpenMD {
764      return d;    
765    }
766  
767 +  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
768 +    return excludesForAtom[atom1];
769 +  }
770 +
771 +  /**
772 +   * We need to exclude some overcounted interactions that result from
773 +   * the parallel decomposition.
774 +   */
775 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
776 +    int unique_id_1, unique_id_2;
777 +
778 + #ifdef IS_MPI
779 +    // in MPI, we have to look up the unique IDs for each atom
780 +    unique_id_1 = AtomRowToGlobal[atom1];
781 +    unique_id_2 = AtomColToGlobal[atom2];
782 +
783 +    // this situation should only arise in MPI simulations
784 +    if (unique_id_1 == unique_id_2) return true;
785 +    
786 +    // this prevents us from doing the pair on multiple processors
787 +    if (unique_id_1 < unique_id_2) {
788 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
789 +    } else {
790 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
791 +    }
792 + #endif
793 +    return false;
794 +  }
795 +
796 +  /**
797 +   * We need to handle the interactions for atoms who are involved in
798 +   * the same rigid body as well as some short range interactions
799 +   * (bonds, bends, torsions) differently from other interactions.
800 +   * We'll still visit the pairwise routines, but with a flag that
801 +   * tells those routines to exclude the pair from direct long range
802 +   * interactions.  Some indirect interactions (notably reaction
803 +   * field) must still be handled for these pairs.
804 +   */
805 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
806 +    int unique_id_2;
807 +    
808 + #ifdef IS_MPI
809 +    // in MPI, we have to look up the unique IDs for the row atom.
810 +    unique_id_2 = AtomColToGlobal[atom2];
811 + #else
812 +    // in the normal loop, the atom numbers are unique
813 +    unique_id_2 = atom2;
814 + #endif
815 +    
816 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
817 +         i != excludesForAtom[atom1].end(); ++i) {
818 +      if ( (*i) == unique_id_2 ) return true;
819 +    }
820 +
821 +    return false;
822 +  }
823 +
824 +
825    void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
826   #ifdef IS_MPI
827      atomRowData.force[atom1] += fg;
# Line 391 | Line 839 | namespace OpenMD {
839    }
840  
841      // filling interaction blocks with pointers
842 <  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
843 <    InteractionData idat;
842 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
843 >                                                     int atom1, int atom2) {
844  
845 +    idat.excluded = excludeAtomPair(atom1, atom2);
846 +  
847   #ifdef IS_MPI
848 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
849 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
850 +    //                         ff_->getAtomType(identsCol[atom2]) );
851 +    
852      if (storageLayout_ & DataStorage::dslAmat) {
853        idat.A1 = &(atomRowData.aMat[atom1]);
854        idat.A2 = &(atomColData.aMat[atom2]);
# Line 415 | Line 869 | namespace OpenMD {
869        idat.rho2 = &(atomColData.density[atom2]);
870      }
871  
872 +    if (storageLayout_ & DataStorage::dslFunctional) {
873 +      idat.frho1 = &(atomRowData.functional[atom1]);
874 +      idat.frho2 = &(atomColData.functional[atom2]);
875 +    }
876 +
877      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
878        idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
879        idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
880      }
881 +
882 +    if (storageLayout_ & DataStorage::dslParticlePot) {
883 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
884 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
885 +    }
886 +
887 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
888 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
889 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
890 +    }
891 +
892   #else
893 +
894 +    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
895 +    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
896 +    //                         ff_->getAtomType(idents[atom2]) );
897 +
898      if (storageLayout_ & DataStorage::dslAmat) {
899        idat.A1 = &(snap_->atomData.aMat[atom1]);
900        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 435 | Line 910 | namespace OpenMD {
910        idat.t2 = &(snap_->atomData.torque[atom2]);
911      }
912  
913 <    if (storageLayout_ & DataStorage::dslDensity) {
913 >    if (storageLayout_ & DataStorage::dslDensity) {    
914        idat.rho1 = &(snap_->atomData.density[atom1]);
915        idat.rho2 = &(snap_->atomData.density[atom2]);
916      }
917  
918 +    if (storageLayout_ & DataStorage::dslFunctional) {
919 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
920 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
921 +    }
922 +
923      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
924        idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
925        idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
926      }
927 +
928 +    if (storageLayout_ & DataStorage::dslParticlePot) {
929 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
930 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
931 +    }
932 +
933 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
934 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
935 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
936 +    }
937   #endif
448 –    return idat;
938    }
939  
940 <  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
941 <
453 <    InteractionData idat;
940 >  
941 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
942   #ifdef IS_MPI
943 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
944 <      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
945 <      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
946 <    }
947 <    if (storageLayout_ & DataStorage::dslTorque) {
460 <      idat.t1 = &(atomRowData.torque[atom1]);
461 <      idat.t2 = &(atomColData.torque[atom2]);
462 <    }
463 <    if (storageLayout_ & DataStorage::dslForce) {
464 <      idat.t1 = &(atomRowData.force[atom1]);
465 <      idat.t2 = &(atomColData.force[atom2]);
466 <    }
943 >    pot_row[atom1] += 0.5 *  *(idat.pot);
944 >    pot_col[atom2] += 0.5 *  *(idat.pot);
945 >
946 >    atomRowData.force[atom1] += *(idat.f1);
947 >    atomColData.force[atom2] -= *(idat.f1);
948   #else
949 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
950 <      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
951 <      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
952 <    }
472 <    if (storageLayout_ & DataStorage::dslTorque) {
473 <      idat.t1 = &(snap_->atomData.torque[atom1]);
474 <      idat.t2 = &(snap_->atomData.torque[atom2]);
475 <    }
476 <    if (storageLayout_ & DataStorage::dslForce) {
477 <      idat.t1 = &(snap_->atomData.force[atom1]);
478 <      idat.t2 = &(snap_->atomData.force[atom2]);
479 <    }
949 >    pairwisePot += *(idat.pot);
950 >
951 >    snap_->atomData.force[atom1] += *(idat.f1);
952 >    snap_->atomData.force[atom2] -= *(idat.f1);
953   #endif
954      
955    }
956  
484 –
485 –
486 –
957    /*
958     * buildNeighborList
959     *
# Line 493 | Line 963 | namespace OpenMD {
963    vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
964        
965      vector<pair<int, int> > neighborList;
966 +    groupCutoffs cuts;
967 +    bool doAllPairs = false;
968 +
969   #ifdef IS_MPI
970      cellListRow_.clear();
971      cellListCol_.clear();
# Line 500 | Line 973 | namespace OpenMD {
973      cellList_.clear();
974   #endif
975  
976 <    // dangerous to not do error checking.
504 <    RealType rCut_;
505 <
506 <    RealType rList_ = (rCut_ + skinThickness_);
976 >    RealType rList_ = (largestRcut_ + skinThickness_);
977      RealType rl2 = rList_ * rList_;
978      Snapshot* snap_ = sman_->getCurrentSnapshot();
979      Mat3x3d Hmat = snap_->getHmat();
# Line 515 | Line 985 | namespace OpenMD {
985      nCells_.y() = (int) ( Hy.length() )/ rList_;
986      nCells_.z() = (int) ( Hz.length() )/ rList_;
987  
988 +    // handle small boxes where the cell offsets can end up repeating cells
989 +    
990 +    if (nCells_.x() < 3) doAllPairs = true;
991 +    if (nCells_.y() < 3) doAllPairs = true;
992 +    if (nCells_.z() < 3) doAllPairs = true;
993 +
994      Mat3x3d invHmat = snap_->getInvHmat();
995      Vector3d rs, scaled, dr;
996      Vector3i whichCell;
997      int cellIndex;
998 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
999  
1000   #ifdef IS_MPI
1001 <    for (int i = 0; i < nGroupsInRow_; i++) {
1002 <      rs = cgRowData.position[i];
526 <      // scaled positions relative to the box vectors
527 <      scaled = invHmat * rs;
528 <      // wrap the vector back into the unit box by subtracting integer box
529 <      // numbers
530 <      for (int j = 0; j < 3; j++)
531 <        scaled[j] -= roundMe(scaled[j]);
532 <    
533 <      // find xyz-indices of cell that cutoffGroup is in.
534 <      whichCell.x() = nCells_.x() * scaled.x();
535 <      whichCell.y() = nCells_.y() * scaled.y();
536 <      whichCell.z() = nCells_.z() * scaled.z();
537 <
538 <      // find single index of this cell:
539 <      cellIndex = Vlinear(whichCell, nCells_);
540 <      // add this cutoff group to the list of groups in this cell;
541 <      cellListRow_[cellIndex].push_back(i);
542 <    }
543 <
544 <    for (int i = 0; i < nGroupsInCol_; i++) {
545 <      rs = cgColData.position[i];
546 <      // scaled positions relative to the box vectors
547 <      scaled = invHmat * rs;
548 <      // wrap the vector back into the unit box by subtracting integer box
549 <      // numbers
550 <      for (int j = 0; j < 3; j++)
551 <        scaled[j] -= roundMe(scaled[j]);
552 <
553 <      // find xyz-indices of cell that cutoffGroup is in.
554 <      whichCell.x() = nCells_.x() * scaled.x();
555 <      whichCell.y() = nCells_.y() * scaled.y();
556 <      whichCell.z() = nCells_.z() * scaled.z();
557 <
558 <      // find single index of this cell:
559 <      cellIndex = Vlinear(whichCell, nCells_);
560 <      // add this cutoff group to the list of groups in this cell;
561 <      cellListCol_[cellIndex].push_back(i);
562 <    }
1001 >    cellListRow_.resize(nCtot);
1002 >    cellListCol_.resize(nCtot);
1003   #else
1004 <    for (int i = 0; i < nGroups_; i++) {
565 <      rs = snap_->cgData.position[i];
566 <      // scaled positions relative to the box vectors
567 <      scaled = invHmat * rs;
568 <      // wrap the vector back into the unit box by subtracting integer box
569 <      // numbers
570 <      for (int j = 0; j < 3; j++)
571 <        scaled[j] -= roundMe(scaled[j]);
572 <
573 <      // find xyz-indices of cell that cutoffGroup is in.
574 <      whichCell.x() = nCells_.x() * scaled.x();
575 <      whichCell.y() = nCells_.y() * scaled.y();
576 <      whichCell.z() = nCells_.z() * scaled.z();
577 <
578 <      // find single index of this cell:
579 <      cellIndex = Vlinear(whichCell, nCells_);
580 <      // add this cutoff group to the list of groups in this cell;
581 <      cellList_[cellIndex].push_back(i);
582 <    }
1004 >    cellList_.resize(nCtot);
1005   #endif
1006  
1007 +    if (!doAllPairs) {
1008 + #ifdef IS_MPI
1009  
1010 +      for (int i = 0; i < nGroupsInRow_; i++) {
1011 +        rs = cgRowData.position[i];
1012 +        
1013 +        // scaled positions relative to the box vectors
1014 +        scaled = invHmat * rs;
1015 +        
1016 +        // wrap the vector back into the unit box by subtracting integer box
1017 +        // numbers
1018 +        for (int j = 0; j < 3; j++) {
1019 +          scaled[j] -= roundMe(scaled[j]);
1020 +          scaled[j] += 0.5;
1021 +        }
1022 +        
1023 +        // find xyz-indices of cell that cutoffGroup is in.
1024 +        whichCell.x() = nCells_.x() * scaled.x();
1025 +        whichCell.y() = nCells_.y() * scaled.y();
1026 +        whichCell.z() = nCells_.z() * scaled.z();
1027 +        
1028 +        // find single index of this cell:
1029 +        cellIndex = Vlinear(whichCell, nCells_);
1030 +        
1031 +        // add this cutoff group to the list of groups in this cell;
1032 +        cellListRow_[cellIndex].push_back(i);
1033 +      }
1034 +      
1035 +      for (int i = 0; i < nGroupsInCol_; i++) {
1036 +        rs = cgColData.position[i];
1037 +        
1038 +        // scaled positions relative to the box vectors
1039 +        scaled = invHmat * rs;
1040 +        
1041 +        // wrap the vector back into the unit box by subtracting integer box
1042 +        // numbers
1043 +        for (int j = 0; j < 3; j++) {
1044 +          scaled[j] -= roundMe(scaled[j]);
1045 +          scaled[j] += 0.5;
1046 +        }
1047 +        
1048 +        // find xyz-indices of cell that cutoffGroup is in.
1049 +        whichCell.x() = nCells_.x() * scaled.x();
1050 +        whichCell.y() = nCells_.y() * scaled.y();
1051 +        whichCell.z() = nCells_.z() * scaled.z();
1052 +        
1053 +        // find single index of this cell:
1054 +        cellIndex = Vlinear(whichCell, nCells_);
1055 +        
1056 +        // add this cutoff group to the list of groups in this cell;
1057 +        cellListCol_[cellIndex].push_back(i);
1058 +      }
1059 + #else
1060 +      for (int i = 0; i < nGroups_; i++) {
1061 +        rs = snap_->cgData.position[i];
1062 +        
1063 +        // scaled positions relative to the box vectors
1064 +        scaled = invHmat * rs;
1065 +        
1066 +        // wrap the vector back into the unit box by subtracting integer box
1067 +        // numbers
1068 +        for (int j = 0; j < 3; j++) {
1069 +          scaled[j] -= roundMe(scaled[j]);
1070 +          scaled[j] += 0.5;
1071 +        }
1072 +        
1073 +        // find xyz-indices of cell that cutoffGroup is in.
1074 +        whichCell.x() = nCells_.x() * scaled.x();
1075 +        whichCell.y() = nCells_.y() * scaled.y();
1076 +        whichCell.z() = nCells_.z() * scaled.z();
1077 +        
1078 +        // find single index of this cell:
1079 +        cellIndex = Vlinear(whichCell, nCells_);      
1080 +        
1081 +        // add this cutoff group to the list of groups in this cell;
1082 +        cellList_[cellIndex].push_back(i);
1083 +      }
1084 + #endif
1085  
1086 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1087 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1088 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1089 <          Vector3i m1v(m1x, m1y, m1z);
1090 <          int m1 = Vlinear(m1v, nCells_);
592 <
593 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
594 <               os != cellOffsets_.end(); ++os) {
1086 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1087 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1088 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1089 >            Vector3i m1v(m1x, m1y, m1z);
1090 >            int m1 = Vlinear(m1v, nCells_);
1091              
1092 <            Vector3i m2v = m1v + (*os);
1093 <            
1094 <            if (m2v.x() >= nCells_.x()) {
1095 <              m2v.x() = 0;          
1096 <            } else if (m2v.x() < 0) {
1097 <              m2v.x() = nCells_.x() - 1;
1098 <            }
1099 <            
1100 <            if (m2v.y() >= nCells_.y()) {
1101 <              m2v.y() = 0;          
1102 <            } else if (m2v.y() < 0) {
1103 <              m2v.y() = nCells_.y() - 1;
1104 <            }
1105 <            
1106 <            if (m2v.z() >= nCells_.z()) {
1107 <              m2v.z() = 0;          
1108 <            } else if (m2v.z() < 0) {
1109 <              m2v.z() = nCells_.z() - 1;
1110 <            }
1111 <            
1112 <            int m2 = Vlinear (m2v, nCells_);
1113 <
1092 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1093 >                 os != cellOffsets_.end(); ++os) {
1094 >              
1095 >              Vector3i m2v = m1v + (*os);
1096 >              
1097 >              if (m2v.x() >= nCells_.x()) {
1098 >                m2v.x() = 0;          
1099 >              } else if (m2v.x() < 0) {
1100 >                m2v.x() = nCells_.x() - 1;
1101 >              }
1102 >              
1103 >              if (m2v.y() >= nCells_.y()) {
1104 >                m2v.y() = 0;          
1105 >              } else if (m2v.y() < 0) {
1106 >                m2v.y() = nCells_.y() - 1;
1107 >              }
1108 >              
1109 >              if (m2v.z() >= nCells_.z()) {
1110 >                m2v.z() = 0;          
1111 >              } else if (m2v.z() < 0) {
1112 >                m2v.z() = nCells_.z() - 1;
1113 >              }
1114 >              
1115 >              int m2 = Vlinear (m2v, nCells_);
1116 >              
1117   #ifdef IS_MPI
1118 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1119 <                 j1 != cellListRow_[m1].end(); ++j1) {
1120 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1121 <                   j2 != cellListCol_[m2].end(); ++j2) {
1122 <                              
1123 <                // Always do this if we're in different cells or if
1124 <                // we're in the same cell and the global index of the
1125 <                // j2 cutoff group is less than the j1 cutoff group
1126 <
1127 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1128 <                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1129 <                  snap_->wrapVector(dr);
1130 <                  if (dr.lengthSquare() < rl2) {
1131 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1118 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1119 >                   j1 != cellListRow_[m1].end(); ++j1) {
1120 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1121 >                     j2 != cellListCol_[m2].end(); ++j2) {
1122 >                  
1123 >                  // Always do this if we're in different cells or if
1124 >                  // we're in the same cell and the global index of the
1125 >                  // j2 cutoff group is less than the j1 cutoff group
1126 >                  
1127 >                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1128 >                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1129 >                    snap_->wrapVector(dr);
1130 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1131 >                    if (dr.lengthSquare() < cuts.third) {
1132 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1133 >                    }
1134                    }
1135                  }
1136                }
636 –            }
1137   #else
1138 <            for (vector<int>::iterator j1 = cellList_[m1].begin();
1139 <                 j1 != cellList_[m1].end(); ++j1) {
1140 <              for (vector<int>::iterator j2 = cellList_[m2].begin();
1141 <                   j2 != cellList_[m2].end(); ++j2) {
1142 <                              
1143 <                // Always do this if we're in different cells or if
1144 <                // we're in the same cell and the global index of the
1145 <                // j2 cutoff group is less than the j1 cutoff group
1146 <
1147 <                if (m2 != m1 || (*j2) < (*j1)) {
1148 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1149 <                  snap_->wrapVector(dr);
1150 <                  if (dr.lengthSquare() < rl2) {
1151 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1138 >              
1139 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1140 >                   j1 != cellList_[m1].end(); ++j1) {
1141 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1142 >                     j2 != cellList_[m2].end(); ++j2) {
1143 >                  
1144 >                  // Always do this if we're in different cells or if
1145 >                  // we're in the same cell and the global index of the
1146 >                  // j2 cutoff group is less than the j1 cutoff group
1147 >                  
1148 >                  if (m2 != m1 || (*j2) < (*j1)) {
1149 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1150 >                    snap_->wrapVector(dr);
1151 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1152 >                    if (dr.lengthSquare() < cuts.third) {
1153 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1154 >                    }
1155                    }
1156                  }
1157                }
655 –            }
1158   #endif
1159 +            }
1160            }
1161          }
1162        }
1163 +    } else {
1164 +      // branch to do all cutoff group pairs
1165 + #ifdef IS_MPI
1166 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1167 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1168 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1169 +          snap_->wrapVector(dr);
1170 +          cuts = getGroupCutoffs( j1, j2 );
1171 +          if (dr.lengthSquare() < cuts.third) {
1172 +            neighborList.push_back(make_pair(j1, j2));
1173 +          }
1174 +        }
1175 +      }
1176 + #else
1177 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1178 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1179 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1180 +          snap_->wrapVector(dr);
1181 +          cuts = getGroupCutoffs( j1, j2 );
1182 +          if (dr.lengthSquare() < cuts.third) {
1183 +            neighborList.push_back(make_pair(j1, j2));
1184 +          }
1185 +        }
1186 +      }        
1187 + #endif
1188      }
1189 <
1189 >      
1190      // save the local cutoff group positions for the check that is
1191      // done on each loop:
1192      saved_CG_positions_.clear();
1193      for (int i = 0; i < nGroups_; i++)
1194        saved_CG_positions_.push_back(snap_->cgData.position[i]);
1195 <
1195 >    
1196      return neighborList;
1197    }
1198   } //end namespace OpenMD

Diff Legend

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