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 1570 by gezelter, Thu May 26 21:56:04 2011 UTC vs.
Revision 1591 by gezelter, Tue Jul 12 15:25:07 2011 UTC

# Line 55 | 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();
66 –    vector<RealType> massFactorsLocal = info_->getMassFactors();
67 –    PairList excludes = info_->getExcludedInteractions();
68 –    PairList oneTwo = info_->getOneTwoInteractions();
69 –    PairList oneThree = info_->getOneThreeInteractions();
70 –    PairList oneFour = info_->getOneFourInteractions();
71 –    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 101 | 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));
107 <    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
108 <                                      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 134 | 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 144 | Line 161 | namespace OpenMD {
161        }      
162      }
163  
164 <    skipsForRowAtom.clear();
165 <    skipsForRowAtom.reserve(nAtomsInRow_);
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 = AtomColToGlobal[i];
151 <      for (int j = 0; j < nAtomsInCol_; j++) {
152 <        int jglob = AtomRowToGlobal[j];        
153 <        if (excludes.hasPair(iglob, jglob))
154 <          skipsForRowAtom[i].push_back(j);      
155 <      }      
156 <    }
171 >      int iglob = AtomRowToGlobal[i];
172  
158 –    toposForRowAtom.clear();
159 –    toposForRowAtom.reserve(nAtomsInRow_);
160 –    for (int i = 0; i < nAtomsInRow_; i++) {
161 –      int iglob = AtomColToGlobal[i];
162 –      int nTopos = 0;
173        for (int j = 0; j < nAtomsInCol_; j++) {
174 <        int jglob = AtomRowToGlobal[j];        
175 <        if (oneTwo.hasPair(iglob, jglob)) {
176 <          toposForRowAtom[i].push_back(j);
177 <          topoDistRow[i][nTopos] = 1;
178 <          nTopos++;
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          }
170 –        if (oneThree.hasPair(iglob, jglob)) {
171 –          toposForRowAtom[i].push_back(j);
172 –          topoDistRow[i][nTopos] = 2;
173 –          nTopos++;
174 –        }
175 –        if (oneFour.hasPair(iglob, jglob)) {
176 –          toposForRowAtom[i].push_back(j);
177 –          topoDistRow[i][nTopos] = 3;
178 –          nTopos++;
179 –        }
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 <    skipsForLocalAtom.clear();
217 <    skipsForLocalAtom.reserve(nLocal_);
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 <        if (excludes.hasPair(iglob, jglob))
229 <          skipsForLocalAtom[i].push_back(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 <    toposForLocalAtom.clear();
252 <    toposForLocalAtom.reserve(nLocal_);
253 <    for (int i = 0; i < nLocal_; i++) {
254 <      int iglob = AtomLocalToGlobal[i];
255 <      int nTopos = 0;
256 <      for (int j = 0; j < nLocal_; j++) {
257 <        int jglob = AtomLocalToGlobal[j];        
258 <        if (oneTwo.hasPair(iglob, jglob)) {
259 <          toposForLocalAtom[i].push_back(j);
260 <          topoDistLocal[i][nTopos] = 1;
261 <          nTopos++;
251 >  }
252 >  
253 >  void ForceMatrixDecomposition::createGtypeCutoffMap() {
254 >    
255 >    RealType tol = 1e-6;
256 >    RealType rc;
257 >    int atid;
258 >    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
259 >    map<int, RealType> atypeCutoff;
260 >      
261 >    for (set<AtomType*>::iterator at = atypes.begin();
262 >         at != atypes.end(); ++at){
263 >      atid = (*at)->getIdent();
264 >      if (userChoseCutoff_)
265 >        atypeCutoff[atid] = userCutoff_;
266 >      else
267 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
268 >    }
269 >
270 >    vector<RealType> gTypeCutoffs;
271 >    // first we do a single loop over the cutoff groups to find the
272 >    // largest cutoff for any atypes present in this group.
273 > #ifdef IS_MPI
274 >    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
275 >    groupRowToGtype.resize(nGroupsInRow_);
276 >    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
277 >      vector<int> atomListRow = getAtomsInGroupRow(cg1);
278 >      for (vector<int>::iterator ia = atomListRow.begin();
279 >           ia != atomListRow.end(); ++ia) {            
280 >        int atom1 = (*ia);
281 >        atid = identsRow[atom1];
282 >        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
283 >          groupCutoffRow[cg1] = atypeCutoff[atid];
284          }
285 <        if (oneThree.hasPair(iglob, jglob)) {
286 <          toposForLocalAtom[i].push_back(j);
287 <          topoDistLocal[i][nTopos] = 2;
288 <          nTopos++;
285 >      }
286 >
287 >      bool gTypeFound = false;
288 >      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
289 >        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
290 >          groupRowToGtype[cg1] = gt;
291 >          gTypeFound = true;
292 >        }
293 >      }
294 >      if (!gTypeFound) {
295 >        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
296 >        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
297 >      }
298 >      
299 >    }
300 >    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
301 >    groupColToGtype.resize(nGroupsInCol_);
302 >    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
303 >      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
304 >      for (vector<int>::iterator jb = atomListCol.begin();
305 >           jb != atomListCol.end(); ++jb) {            
306 >        int atom2 = (*jb);
307 >        atid = identsCol[atom2];
308 >        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
309 >          groupCutoffCol[cg2] = atypeCutoff[atid];
310          }
311 <        if (oneFour.hasPair(iglob, jglob)) {
312 <          toposForLocalAtom[i].push_back(j);
313 <          topoDistLocal[i][nTopos] = 3;
314 <          nTopos++;
311 >      }
312 >      bool gTypeFound = false;
313 >      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
314 >        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
315 >          groupColToGtype[cg2] = gt;
316 >          gTypeFound = true;
317 >        }
318 >      }
319 >      if (!gTypeFound) {
320 >        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
321 >        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
322 >      }
323 >    }
324 > #else
325 >
326 >    vector<RealType> groupCutoff(nGroups_, 0.0);
327 >    groupToGtype.resize(nGroups_);
328 >    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
329 >
330 >      groupCutoff[cg1] = 0.0;
331 >      vector<int> atomList = getAtomsInGroupRow(cg1);
332 >
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 +
342 +      bool gTypeFound = false;
343 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
344 +        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
345 +          groupToGtype[cg1] = gt;
346 +          gTypeFound = true;
347 +        }
348 +      }
349 +      if (!gTypeFound) {
350 +        gTypeCutoffs.push_back( groupCutoff[cg1] );
351 +        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
352        }      
353      }
354 + #endif
355 +
356 +    // Now we find the maximum group cutoff value present in the simulation
357 +
358 +    RealType groupMax = *max_element(gTypeCutoffs.begin(),
359 +                                     gTypeCutoffs.end());
360 +
361 + #ifdef IS_MPI
362 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
363 +                              MPI::MAX);
364 + #endif
365 +    
366 +    RealType tradRcut = groupMax;
367 +
368 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
369 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
370 +        RealType thisRcut;
371 +        switch(cutoffPolicy_) {
372 +        case TRADITIONAL:
373 +          thisRcut = tradRcut;
374 +          break;
375 +        case MIX:
376 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
377 +          break;
378 +        case MAX:
379 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
380 +          break;
381 +        default:
382 +          sprintf(painCave.errMsg,
383 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
384 +                  "hit an unknown cutoff policy!\n");
385 +          painCave.severity = OPENMD_ERROR;
386 +          painCave.isFatal = 1;
387 +          simError();
388 +          break;
389 +        }
390 +
391 +        pair<int,int> key = make_pair(i,j);
392 +        gTypeCutoffMap[key].first = thisRcut;
393 +
394 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
395 +
396 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
397 +        
398 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
399 +
400 +        // sanity check
401 +        
402 +        if (userChoseCutoff_) {
403 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
404 +            sprintf(painCave.errMsg,
405 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
406 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
407 +            painCave.severity = OPENMD_ERROR;
408 +            painCave.isFatal = 1;
409 +            simError();            
410 +          }
411 +        }
412 +      }
413 +    }
414    }
415 <  
415 >
416 >
417 >  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
418 >    int i, j;  
419 > #ifdef IS_MPI
420 >    i = groupRowToGtype[cg1];
421 >    j = groupColToGtype[cg2];
422 > #else
423 >    i = groupToGtype[cg1];
424 >    j = groupToGtype[cg2];
425 > #endif    
426 >    return gTypeCutoffMap[make_pair(i,j)];
427 >  }
428 >
429 >  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
430 >    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
431 >      if (toposForAtom[atom1][j] == atom2)
432 >        return topoDist[atom1][j];
433 >    }
434 >    return 0;
435 >  }
436 >
437 >  void ForceMatrixDecomposition::zeroWorkArrays() {
438 >    pairwisePot = 0.0;
439 >    embeddingPot = 0.0;
440 >
441 > #ifdef IS_MPI
442 >    if (storageLayout_ & DataStorage::dslForce) {
443 >      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
444 >      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
445 >    }
446 >
447 >    if (storageLayout_ & DataStorage::dslTorque) {
448 >      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
449 >      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
450 >    }
451 >    
452 >    fill(pot_row.begin(), pot_row.end(),
453 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
454 >
455 >    fill(pot_col.begin(), pot_col.end(),
456 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
457 >
458 >    if (storageLayout_ & DataStorage::dslParticlePot) {    
459 >      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
460 >           0.0);
461 >      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
462 >           0.0);
463 >    }
464 >
465 >    if (storageLayout_ & DataStorage::dslDensity) {      
466 >      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
467 >      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
468 >    }
469 >
470 >    if (storageLayout_ & DataStorage::dslFunctional) {  
471 >      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
472 >           0.0);
473 >      fill(atomColData.functional.begin(), atomColData.functional.end(),
474 >           0.0);
475 >    }
476 >
477 >    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
478 >      fill(atomRowData.functionalDerivative.begin(),
479 >           atomRowData.functionalDerivative.end(), 0.0);
480 >      fill(atomColData.functionalDerivative.begin(),
481 >           atomColData.functionalDerivative.end(), 0.0);
482 >    }
483 >
484 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
485 >      fill(atomRowData.skippedCharge.begin(),
486 >           atomRowData.skippedCharge.end(), 0.0);
487 >      fill(atomColData.skippedCharge.begin(),
488 >           atomColData.skippedCharge.end(), 0.0);
489 >    }
490 >
491 > #endif
492 >    // even in parallel, we need to zero out the local arrays:
493 >
494 >    if (storageLayout_ & DataStorage::dslParticlePot) {      
495 >      fill(snap_->atomData.particlePot.begin(),
496 >           snap_->atomData.particlePot.end(), 0.0);
497 >    }
498 >    
499 >    if (storageLayout_ & DataStorage::dslDensity) {      
500 >      fill(snap_->atomData.density.begin(),
501 >           snap_->atomData.density.end(), 0.0);
502 >    }
503 >    if (storageLayout_ & DataStorage::dslFunctional) {
504 >      fill(snap_->atomData.functional.begin(),
505 >           snap_->atomData.functional.end(), 0.0);
506 >    }
507 >    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
508 >      fill(snap_->atomData.functionalDerivative.begin(),
509 >           snap_->atomData.functionalDerivative.end(), 0.0);
510 >    }
511 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
512 >      fill(snap_->atomData.skippedCharge.begin(),
513 >           snap_->atomData.skippedCharge.end(), 0.0);
514 >    }
515 >    
516 >  }
517 >
518 >
519    void ForceMatrixDecomposition::distributeData()  {
520      snap_ = sman_->getCurrentSnapshot();
521      storageLayout_ = sman_->getStorageLayout();
# Line 263 | Line 548 | namespace OpenMD {
548        AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
549                                     atomColData.electroFrame);
550      }
551 +
552   #endif      
553    }
554    
555 +  /* collects information obtained during the pre-pair loop onto local
556 +   * data structures.
557 +   */
558    void ForceMatrixDecomposition::collectIntermediateData() {
559      snap_ = sman_->getCurrentSnapshot();
560      storageLayout_ = sman_->getStorageLayout();
# Line 277 | Line 566 | namespace OpenMD {
566                                 snap_->atomData.density);
567        
568        int n = snap_->atomData.density.size();
569 <      std::vector<RealType> rho_tmp(n, 0.0);
569 >      vector<RealType> rho_tmp(n, 0.0);
570        AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
571        for (int i = 0; i < n; i++)
572          snap_->atomData.density[i] += rho_tmp[i];
573      }
574   #endif
575    }
576 <  
576 >
577 >  /*
578 >   * redistributes information obtained during the pre-pair loop out to
579 >   * row and column-indexed data structures
580 >   */
581    void ForceMatrixDecomposition::distributeIntermediateData() {
582      snap_ = sman_->getCurrentSnapshot();
583      storageLayout_ = sman_->getStorageLayout();
# Line 322 | Line 615 | namespace OpenMD {
615      AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
616      for (int i = 0; i < n; i++)
617        snap_->atomData.force[i] += frc_tmp[i];
618 <    
326 <    
618 >        
619      if (storageLayout_ & DataStorage::dslTorque) {
620  
621 <      int nt = snap_->atomData.force.size();
621 >      int nt = snap_->atomData.torque.size();
622        vector<Vector3d> trq_tmp(nt, V3Zero);
623  
624        AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
625 <      for (int i = 0; i < n; i++) {
625 >      for (int i = 0; i < nt; i++) {
626          snap_->atomData.torque[i] += trq_tmp[i];
627          trq_tmp[i] = 0.0;
628        }
629        
630        AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
631 <      for (int i = 0; i < n; i++)
631 >      for (int i = 0; i < nt; i++)
632          snap_->atomData.torque[i] += trq_tmp[i];
633      }
342 –    
343 –    nLocal_ = snap_->getNumberOfAtoms();
634  
635 <    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
636 <                                       vector<RealType> (nLocal_, 0.0));
637 <    
638 <    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
639 <      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
640 <      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
641 <        pot_local[i] += pot_temp[i][ii];
635 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {
636 >
637 >      int ns = snap_->atomData.skippedCharge.size();
638 >      vector<RealType> skch_tmp(ns, 0.0);
639 >
640 >      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
641 >      for (int i = 0; i < ns; i++) {
642 >        snap_->atomData.skippedCharge[i] += skch_tmp[i];
643 >        skch_tmp[i] = 0.0;
644        }
645 +      
646 +      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
647 +      for (int i = 0; i < ns; i++)
648 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
649      }
650 +    
651 +    nLocal_ = snap_->getNumberOfAtoms();
652 +
653 +    vector<potVec> pot_temp(nLocal_,
654 +                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
655 +
656 +    // scatter/gather pot_row into the members of my column
657 +          
658 +    AtomCommPotRow->scatter(pot_row, pot_temp);
659 +
660 +    for (int ii = 0;  ii < pot_temp.size(); ii++ )
661 +      pairwisePot += pot_temp[ii];
662 +    
663 +    fill(pot_temp.begin(), pot_temp.end(),
664 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
665 +      
666 +    AtomCommPotColumn->scatter(pot_col, pot_temp);    
667 +    
668 +    for (int ii = 0;  ii < pot_temp.size(); ii++ )
669 +      pairwisePot += pot_temp[ii];    
670   #endif
671 +
672    }
673  
674    int ForceMatrixDecomposition::getNAtomsInRow() {  
# Line 426 | Line 743 | namespace OpenMD {
743   #ifdef IS_MPI
744      return massFactorsRow[atom1];
745   #else
746 <    return massFactorsLocal[atom1];
746 >    return massFactors[atom1];
747   #endif
748    }
749  
# Line 434 | Line 751 | namespace OpenMD {
751   #ifdef IS_MPI
752      return massFactorsCol[atom2];
753   #else
754 <    return massFactorsLocal[atom2];
754 >    return massFactors[atom2];
755   #endif
756  
757    }
# Line 452 | Line 769 | namespace OpenMD {
769      return d;    
770    }
771  
772 <  vector<int> ForceMatrixDecomposition::getSkipsForRowAtom(int atom1) {
773 < #ifdef IS_MPI
457 <    return skipsForRowAtom[atom1];
458 < #else
459 <    return skipsForLocalAtom[atom1];
460 < #endif
772 >  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
773 >    return excludesForAtom[atom1];
774    }
775  
776    /**
777 <   * there are a number of reasons to skip a pair or a particle mostly
778 <   * we do this to exclude atoms who are involved in short range
466 <   * interactions (bonds, bends, torsions), but we also need to
467 <   * exclude some overcounted interactions that result from the
468 <   * parallel decomposition.
777 >   * We need to exclude some overcounted interactions that result from
778 >   * the parallel decomposition.
779     */
780    bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
781      int unique_id_1, unique_id_2;
# Line 484 | Line 794 | namespace OpenMD {
794      } else {
795        if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
796      }
797 + #endif
798 +    return false;
799 +  }
800 +
801 +  /**
802 +   * We need to handle the interactions for atoms who are involved in
803 +   * the same rigid body as well as some short range interactions
804 +   * (bonds, bends, torsions) differently from other interactions.
805 +   * We'll still visit the pairwise routines, but with a flag that
806 +   * tells those routines to exclude the pair from direct long range
807 +   * interactions.  Some indirect interactions (notably reaction
808 +   * field) must still be handled for these pairs.
809 +   */
810 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
811 +    int unique_id_2;
812 +    
813 + #ifdef IS_MPI
814 +    // in MPI, we have to look up the unique IDs for the row atom.
815 +    unique_id_2 = AtomColToGlobal[atom2];
816   #else
817      // in the normal loop, the atom numbers are unique
489 –    unique_id_1 = atom1;
818      unique_id_2 = atom2;
819   #endif
820      
821 < #ifdef IS_MPI
822 <    for (vector<int>::iterator i = skipsForRowAtom[atom1].begin();
495 <         i != skipsForRowAtom[atom1].end(); ++i) {
496 <      if ( (*i) == unique_id_2 ) return true;
497 <    }    
498 < #else
499 <    for (vector<int>::iterator i = skipsForLocalAtom[atom1].begin();
500 <         i != skipsForLocalAtom[atom1].end(); ++i) {
821 >    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
822 >         i != excludesForAtom[atom1].end(); ++i) {
823        if ( (*i) == unique_id_2 ) return true;
502 –    }    
503 – #endif
504 –  }
505 –
506 –  int ForceMatrixDecomposition::getTopoDistance(int atom1, int atom2) {
507 –    
508 – #ifdef IS_MPI
509 –    for (int i = 0; i < toposForRowAtom[atom1].size(); i++) {
510 –      if ( toposForRowAtom[atom1][i] == atom2 ) return topoDistRow[atom1][i];
824      }
512 – #else
513 –    for (int i = 0; i < toposForLocalAtom[atom1].size(); i++) {
514 –      if ( toposForLocalAtom[atom1][i] == atom2 ) return topoDistLocal[atom1][i];
515 –    }
516 – #endif
825  
826 <    // zero is default for unconnected (i.e. normal) pair interactions
519 <    return 0;
826 >    return false;
827    }
828  
829 +
830    void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
831   #ifdef IS_MPI
832      atomRowData.force[atom1] += fg;
# Line 536 | Line 844 | namespace OpenMD {
844    }
845  
846      // filling interaction blocks with pointers
847 <  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
848 <    InteractionData idat;
847 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
848 >                                                     int atom1, int atom2) {
849  
850 +    idat.excluded = excludeAtomPair(atom1, atom2);
851 +  
852   #ifdef IS_MPI
853 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
854 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
855 +    //                         ff_->getAtomType(identsCol[atom2]) );
856 +    
857      if (storageLayout_ & DataStorage::dslAmat) {
858        idat.A1 = &(atomRowData.aMat[atom1]);
859        idat.A2 = &(atomColData.aMat[atom2]);
# Line 560 | Line 874 | namespace OpenMD {
874        idat.rho2 = &(atomColData.density[atom2]);
875      }
876  
877 +    if (storageLayout_ & DataStorage::dslFunctional) {
878 +      idat.frho1 = &(atomRowData.functional[atom1]);
879 +      idat.frho2 = &(atomColData.functional[atom2]);
880 +    }
881 +
882      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
883        idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
884        idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
885      }
886  
887 +    if (storageLayout_ & DataStorage::dslParticlePot) {
888 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
889 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
890 +    }
891 +
892 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
893 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
894 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
895 +    }
896 +
897   #else
898 +
899 +    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
900 +    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
901 +    //                         ff_->getAtomType(idents[atom2]) );
902 +
903      if (storageLayout_ & DataStorage::dslAmat) {
904        idat.A1 = &(snap_->atomData.aMat[atom1]);
905        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 581 | Line 915 | namespace OpenMD {
915        idat.t2 = &(snap_->atomData.torque[atom2]);
916      }
917  
918 <    if (storageLayout_ & DataStorage::dslDensity) {
918 >    if (storageLayout_ & DataStorage::dslDensity) {    
919        idat.rho1 = &(snap_->atomData.density[atom1]);
920        idat.rho2 = &(snap_->atomData.density[atom2]);
921      }
922  
923 +    if (storageLayout_ & DataStorage::dslFunctional) {
924 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
925 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
926 +    }
927 +
928      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
929        idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
930        idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
931      }
593 – #endif
594 –    return idat;
595 –  }
932  
933 <  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
933 >    if (storageLayout_ & DataStorage::dslParticlePot) {
934 >      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
935 >      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
936 >    }
937  
938 <    InteractionData idat;
939 < #ifdef IS_MPI
940 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
602 <      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
603 <      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
938 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {
939 >      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
940 >      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
941      }
605 –    if (storageLayout_ & DataStorage::dslTorque) {
606 –      idat.t1 = &(atomRowData.torque[atom1]);
607 –      idat.t2 = &(atomColData.torque[atom2]);
608 –    }
609 –    if (storageLayout_ & DataStorage::dslForce) {
610 –      idat.t1 = &(atomRowData.force[atom1]);
611 –      idat.t2 = &(atomColData.force[atom2]);
612 –    }
613 – #else
614 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
615 –      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
616 –      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
617 –    }
618 –    if (storageLayout_ & DataStorage::dslTorque) {
619 –      idat.t1 = &(snap_->atomData.torque[atom1]);
620 –      idat.t2 = &(snap_->atomData.torque[atom2]);
621 –    }
622 –    if (storageLayout_ & DataStorage::dslForce) {
623 –      idat.t1 = &(snap_->atomData.force[atom1]);
624 –      idat.t2 = &(snap_->atomData.force[atom2]);
625 –    }
942   #endif
627 –    
943    }
944  
945 +  
946 +  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
947 + #ifdef IS_MPI
948 +    pot_row[atom1] += 0.5 *  *(idat.pot);
949 +    pot_col[atom2] += 0.5 *  *(idat.pot);
950  
951 +    atomRowData.force[atom1] += *(idat.f1);
952 +    atomColData.force[atom2] -= *(idat.f1);
953 + #else
954 +    pairwisePot += *(idat.pot);
955  
956 +    snap_->atomData.force[atom1] += *(idat.f1);
957 +    snap_->atomData.force[atom2] -= *(idat.f1);
958 + #endif
959 +    
960 +  }
961  
962    /*
963     * buildNeighborList
# Line 639 | Line 968 | namespace OpenMD {
968    vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
969        
970      vector<pair<int, int> > neighborList;
971 +    groupCutoffs cuts;
972 +    bool doAllPairs = false;
973 +
974   #ifdef IS_MPI
975      cellListRow_.clear();
976      cellListCol_.clear();
# Line 646 | Line 978 | namespace OpenMD {
978      cellList_.clear();
979   #endif
980  
981 <    // dangerous to not do error checking.
650 <    RealType rCut_;
651 <
652 <    RealType rList_ = (rCut_ + skinThickness_);
981 >    RealType rList_ = (largestRcut_ + skinThickness_);
982      RealType rl2 = rList_ * rList_;
983      Snapshot* snap_ = sman_->getCurrentSnapshot();
984      Mat3x3d Hmat = snap_->getHmat();
# Line 661 | Line 990 | namespace OpenMD {
990      nCells_.y() = (int) ( Hy.length() )/ rList_;
991      nCells_.z() = (int) ( Hz.length() )/ rList_;
992  
993 +    // handle small boxes where the cell offsets can end up repeating cells
994 +    
995 +    if (nCells_.x() < 3) doAllPairs = true;
996 +    if (nCells_.y() < 3) doAllPairs = true;
997 +    if (nCells_.z() < 3) doAllPairs = true;
998 +
999      Mat3x3d invHmat = snap_->getInvHmat();
1000      Vector3d rs, scaled, dr;
1001      Vector3i whichCell;
1002      int cellIndex;
1003 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1004  
1005   #ifdef IS_MPI
1006 <    for (int i = 0; i < nGroupsInRow_; i++) {
1007 <      rs = cgRowData.position[i];
672 <      // scaled positions relative to the box vectors
673 <      scaled = invHmat * rs;
674 <      // wrap the vector back into the unit box by subtracting integer box
675 <      // numbers
676 <      for (int j = 0; j < 3; j++)
677 <        scaled[j] -= roundMe(scaled[j]);
678 <    
679 <      // find xyz-indices of cell that cutoffGroup is in.
680 <      whichCell.x() = nCells_.x() * scaled.x();
681 <      whichCell.y() = nCells_.y() * scaled.y();
682 <      whichCell.z() = nCells_.z() * scaled.z();
683 <
684 <      // find single index of this cell:
685 <      cellIndex = Vlinear(whichCell, nCells_);
686 <      // add this cutoff group to the list of groups in this cell;
687 <      cellListRow_[cellIndex].push_back(i);
688 <    }
689 <
690 <    for (int i = 0; i < nGroupsInCol_; i++) {
691 <      rs = cgColData.position[i];
692 <      // scaled positions relative to the box vectors
693 <      scaled = invHmat * rs;
694 <      // wrap the vector back into the unit box by subtracting integer box
695 <      // numbers
696 <      for (int j = 0; j < 3; j++)
697 <        scaled[j] -= roundMe(scaled[j]);
698 <
699 <      // find xyz-indices of cell that cutoffGroup is in.
700 <      whichCell.x() = nCells_.x() * scaled.x();
701 <      whichCell.y() = nCells_.y() * scaled.y();
702 <      whichCell.z() = nCells_.z() * scaled.z();
703 <
704 <      // find single index of this cell:
705 <      cellIndex = Vlinear(whichCell, nCells_);
706 <      // add this cutoff group to the list of groups in this cell;
707 <      cellListCol_[cellIndex].push_back(i);
708 <    }
1006 >    cellListRow_.resize(nCtot);
1007 >    cellListCol_.resize(nCtot);
1008   #else
1009 <    for (int i = 0; i < nGroups_; i++) {
711 <      rs = snap_->cgData.position[i];
712 <      // scaled positions relative to the box vectors
713 <      scaled = invHmat * rs;
714 <      // wrap the vector back into the unit box by subtracting integer box
715 <      // numbers
716 <      for (int j = 0; j < 3; j++)
717 <        scaled[j] -= roundMe(scaled[j]);
718 <
719 <      // find xyz-indices of cell that cutoffGroup is in.
720 <      whichCell.x() = nCells_.x() * scaled.x();
721 <      whichCell.y() = nCells_.y() * scaled.y();
722 <      whichCell.z() = nCells_.z() * scaled.z();
723 <
724 <      // find single index of this cell:
725 <      cellIndex = Vlinear(whichCell, nCells_);
726 <      // add this cutoff group to the list of groups in this cell;
727 <      cellList_[cellIndex].push_back(i);
728 <    }
1009 >    cellList_.resize(nCtot);
1010   #endif
1011  
1012 +    if (!doAllPairs) {
1013 + #ifdef IS_MPI
1014  
1015 +      for (int i = 0; i < nGroupsInRow_; i++) {
1016 +        rs = cgRowData.position[i];
1017 +        
1018 +        // scaled positions relative to the box vectors
1019 +        scaled = invHmat * rs;
1020 +        
1021 +        // wrap the vector back into the unit box by subtracting integer box
1022 +        // numbers
1023 +        for (int j = 0; j < 3; j++) {
1024 +          scaled[j] -= roundMe(scaled[j]);
1025 +          scaled[j] += 0.5;
1026 +        }
1027 +        
1028 +        // find xyz-indices of cell that cutoffGroup is in.
1029 +        whichCell.x() = nCells_.x() * scaled.x();
1030 +        whichCell.y() = nCells_.y() * scaled.y();
1031 +        whichCell.z() = nCells_.z() * scaled.z();
1032 +        
1033 +        // find single index of this cell:
1034 +        cellIndex = Vlinear(whichCell, nCells_);
1035 +        
1036 +        // add this cutoff group to the list of groups in this cell;
1037 +        cellListRow_[cellIndex].push_back(i);
1038 +      }
1039 +      
1040 +      for (int i = 0; i < nGroupsInCol_; i++) {
1041 +        rs = cgColData.position[i];
1042 +        
1043 +        // scaled positions relative to the box vectors
1044 +        scaled = invHmat * rs;
1045 +        
1046 +        // wrap the vector back into the unit box by subtracting integer box
1047 +        // numbers
1048 +        for (int j = 0; j < 3; j++) {
1049 +          scaled[j] -= roundMe(scaled[j]);
1050 +          scaled[j] += 0.5;
1051 +        }
1052 +        
1053 +        // find xyz-indices of cell that cutoffGroup is in.
1054 +        whichCell.x() = nCells_.x() * scaled.x();
1055 +        whichCell.y() = nCells_.y() * scaled.y();
1056 +        whichCell.z() = nCells_.z() * scaled.z();
1057 +        
1058 +        // find single index of this cell:
1059 +        cellIndex = Vlinear(whichCell, nCells_);
1060 +        
1061 +        // add this cutoff group to the list of groups in this cell;
1062 +        cellListCol_[cellIndex].push_back(i);
1063 +      }
1064 + #else
1065 +      for (int i = 0; i < nGroups_; i++) {
1066 +        rs = snap_->cgData.position[i];
1067 +        
1068 +        // scaled positions relative to the box vectors
1069 +        scaled = invHmat * rs;
1070 +        
1071 +        // wrap the vector back into the unit box by subtracting integer box
1072 +        // numbers
1073 +        for (int j = 0; j < 3; j++) {
1074 +          scaled[j] -= roundMe(scaled[j]);
1075 +          scaled[j] += 0.5;
1076 +        }
1077 +        
1078 +        // find xyz-indices of cell that cutoffGroup is in.
1079 +        whichCell.x() = nCells_.x() * scaled.x();
1080 +        whichCell.y() = nCells_.y() * scaled.y();
1081 +        whichCell.z() = nCells_.z() * scaled.z();
1082 +        
1083 +        // find single index of this cell:
1084 +        cellIndex = Vlinear(whichCell, nCells_);      
1085 +        
1086 +        // add this cutoff group to the list of groups in this cell;
1087 +        cellList_[cellIndex].push_back(i);
1088 +      }
1089 + #endif
1090  
1091 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1092 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1093 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1094 <          Vector3i m1v(m1x, m1y, m1z);
1095 <          int m1 = Vlinear(m1v, nCells_);
738 <
739 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
740 <               os != cellOffsets_.end(); ++os) {
1091 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1092 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1093 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1094 >            Vector3i m1v(m1x, m1y, m1z);
1095 >            int m1 = Vlinear(m1v, nCells_);
1096              
1097 <            Vector3i m2v = m1v + (*os);
1098 <            
1099 <            if (m2v.x() >= nCells_.x()) {
1100 <              m2v.x() = 0;          
1101 <            } else if (m2v.x() < 0) {
1102 <              m2v.x() = nCells_.x() - 1;
1103 <            }
1104 <            
1105 <            if (m2v.y() >= nCells_.y()) {
1106 <              m2v.y() = 0;          
1107 <            } else if (m2v.y() < 0) {
1108 <              m2v.y() = nCells_.y() - 1;
1109 <            }
1110 <            
1111 <            if (m2v.z() >= nCells_.z()) {
1112 <              m2v.z() = 0;          
1113 <            } else if (m2v.z() < 0) {
1114 <              m2v.z() = nCells_.z() - 1;
1115 <            }
1116 <            
1117 <            int m2 = Vlinear (m2v, nCells_);
1118 <
1097 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1098 >                 os != cellOffsets_.end(); ++os) {
1099 >              
1100 >              Vector3i m2v = m1v + (*os);
1101 >              
1102 >              if (m2v.x() >= nCells_.x()) {
1103 >                m2v.x() = 0;          
1104 >              } else if (m2v.x() < 0) {
1105 >                m2v.x() = nCells_.x() - 1;
1106 >              }
1107 >              
1108 >              if (m2v.y() >= nCells_.y()) {
1109 >                m2v.y() = 0;          
1110 >              } else if (m2v.y() < 0) {
1111 >                m2v.y() = nCells_.y() - 1;
1112 >              }
1113 >              
1114 >              if (m2v.z() >= nCells_.z()) {
1115 >                m2v.z() = 0;          
1116 >              } else if (m2v.z() < 0) {
1117 >                m2v.z() = nCells_.z() - 1;
1118 >              }
1119 >              
1120 >              int m2 = Vlinear (m2v, nCells_);
1121 >              
1122   #ifdef IS_MPI
1123 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1124 <                 j1 != cellListRow_[m1].end(); ++j1) {
1125 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1126 <                   j2 != cellListCol_[m2].end(); ++j2) {
1127 <                              
1128 <                // Always do this if we're in different cells or if
1129 <                // we're in the same cell and the global index of the
1130 <                // j2 cutoff group is less than the j1 cutoff group
1131 <
1132 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1133 <                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1134 <                  snap_->wrapVector(dr);
1135 <                  if (dr.lengthSquare() < rl2) {
1136 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1123 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1124 >                   j1 != cellListRow_[m1].end(); ++j1) {
1125 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1126 >                     j2 != cellListCol_[m2].end(); ++j2) {
1127 >                  
1128 >                  // Always do this if we're in different cells or if
1129 >                  // we're in the same cell and the global index of the
1130 >                  // j2 cutoff group is less than the j1 cutoff group
1131 >                  
1132 >                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1133 >                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1134 >                    snap_->wrapVector(dr);
1135 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1136 >                    if (dr.lengthSquare() < cuts.third) {
1137 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1138 >                    }
1139                    }
1140                  }
1141                }
782 –            }
1142   #else
1143 <            for (vector<int>::iterator j1 = cellList_[m1].begin();
1144 <                 j1 != cellList_[m1].end(); ++j1) {
1145 <              for (vector<int>::iterator j2 = cellList_[m2].begin();
1146 <                   j2 != cellList_[m2].end(); ++j2) {
1147 <                              
1148 <                // Always do this if we're in different cells or if
1149 <                // we're in the same cell and the global index of the
1150 <                // j2 cutoff group is less than the j1 cutoff group
1151 <
1152 <                if (m2 != m1 || (*j2) < (*j1)) {
1153 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1154 <                  snap_->wrapVector(dr);
1155 <                  if (dr.lengthSquare() < rl2) {
1156 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1143 >              
1144 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1145 >                   j1 != cellList_[m1].end(); ++j1) {
1146 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1147 >                     j2 != cellList_[m2].end(); ++j2) {
1148 >                  
1149 >                  // Always do this if we're in different cells or if
1150 >                  // we're in the same cell and the global index of the
1151 >                  // j2 cutoff group is less than the j1 cutoff group
1152 >                  
1153 >                  if (m2 != m1 || (*j2) < (*j1)) {
1154 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1155 >                    snap_->wrapVector(dr);
1156 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1157 >                    if (dr.lengthSquare() < cuts.third) {
1158 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1159 >                    }
1160                    }
1161                  }
1162                }
801 –            }
1163   #endif
1164 +            }
1165            }
1166          }
1167        }
1168 +    } else {
1169 +      // branch to do all cutoff group pairs
1170 + #ifdef IS_MPI
1171 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1172 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1173 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1174 +          snap_->wrapVector(dr);
1175 +          cuts = getGroupCutoffs( j1, j2 );
1176 +          if (dr.lengthSquare() < cuts.third) {
1177 +            neighborList.push_back(make_pair(j1, j2));
1178 +          }
1179 +        }
1180 +      }
1181 + #else
1182 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1183 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1184 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1185 +          snap_->wrapVector(dr);
1186 +          cuts = getGroupCutoffs( j1, j2 );
1187 +          if (dr.lengthSquare() < cuts.third) {
1188 +            neighborList.push_back(make_pair(j1, j2));
1189 +          }
1190 +        }
1191 +      }        
1192 + #endif
1193      }
1194 <
1194 >      
1195      // save the local cutoff group positions for the check that is
1196      // done on each loop:
1197      saved_CG_positions_.clear();
1198      for (int i = 0; i < nGroups_; i++)
1199        saved_CG_positions_.push_back(snap_->cgData.position[i]);
1200 <
1200 >    
1201      return neighborList;
1202    }
1203   } //end namespace OpenMD

Diff Legend

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