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 1567 by gezelter, Tue May 24 21:24:45 2011 UTC vs.
Revision 1590 by gezelter, Mon Jul 11 01:39:49 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();
60 >    
61 >    nGroups_ = info_->getNLocalCutoffGroups();
62 >    // gather the information for atomtype IDs (atids):
63 >    idents = info_->getIdentArray();
64 >    AtomLocalToGlobal = info_->getGlobalAtomIndices();
65 >    cgLocalToGlobal = info_->getGlobalGroupIndices();
66 >    vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
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 88 | 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));
94 <    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
95 <                                      vector<RealType> (nAtomsInCol_, 0.0));
96 <
97 <
98 <    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
112 >    AtomCommIntRow->gather(idents, identsRow);
113 >    AtomCommIntColumn->gather(idents, identsCol);
114      
115 <    // gather the information for atomtype IDs (atids):
116 <    vector<int> identsLocal = info_->getIdentArray();
117 <    identsRow.reserve(nAtomsInRow_);
118 <    identsCol.reserve(nAtomsInCol_);
119 <    
120 <    AtomCommIntRow->gather(identsLocal, identsRow);
121 <    AtomCommIntColumn->gather(identsLocal, identsCol);
122 <    
123 <    AtomLocalToGlobal = info_->getGlobalAtomIndices();
115 >    // allocate memory for the parallel objects
116 >    AtomRowToGlobal.resize(nAtomsInRow_);
117 >    AtomColToGlobal.resize(nAtomsInCol_);
118 >    cgRowToGlobal.resize(nGroupsInRow_);
119 >    cgColToGlobal.resize(nGroupsInCol_);
120 >    massFactorsRow.resize(nAtomsInRow_);
121 >    massFactorsCol.resize(nAtomsInCol_);
122 >    pot_row.resize(nAtomsInRow_);
123 >    pot_col.resize(nAtomsInCol_);
124 >
125      AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
126      AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
127      
112 –    cgLocalToGlobal = info_->getGlobalGroupIndices();
128      cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
129      cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
130  
131 <    // still need:
132 <    // topoDist
133 <    // exclude
131 >    AtomCommRealRow->gather(massFactors, massFactorsRow);
132 >    AtomCommRealColumn->gather(massFactors, massFactorsCol);
133 >
134 >    groupListRow_.clear();
135 >    groupListRow_.resize(nGroupsInRow_);
136 >    for (int i = 0; i < nGroupsInRow_; i++) {
137 >      int gid = cgRowToGlobal[i];
138 >      for (int j = 0; j < nAtomsInRow_; j++) {
139 >        int aid = AtomRowToGlobal[j];
140 >        if (globalGroupMembership[aid] == gid)
141 >          groupListRow_[i].push_back(j);
142 >      }      
143 >    }
144 >
145 >    groupListCol_.clear();
146 >    groupListCol_.resize(nGroupsInCol_);
147 >    for (int i = 0; i < nGroupsInCol_; i++) {
148 >      int gid = cgColToGlobal[i];
149 >      for (int j = 0; j < nAtomsInCol_; j++) {
150 >        int aid = AtomColToGlobal[j];
151 >        if (globalGroupMembership[aid] == gid)
152 >          groupListCol_[i].push_back(j);
153 >      }      
154 >    }
155 >
156 >    excludesForAtom.clear();
157 >    excludesForAtom.resize(nAtomsInRow_);
158 >    toposForAtom.clear();
159 >    toposForAtom.resize(nAtomsInRow_);
160 >    topoDist.clear();
161 >    topoDist.resize(nAtomsInRow_);
162 >    for (int i = 0; i < nAtomsInRow_; i++) {
163 >      int iglob = AtomRowToGlobal[i];
164 >
165 >      for (int j = 0; j < nAtomsInCol_; j++) {
166 >        int jglob = AtomColToGlobal[j];
167 >
168 >        if (excludes->hasPair(iglob, jglob))
169 >          excludesForAtom[i].push_back(j);      
170 >        
171 >        if (oneTwo->hasPair(iglob, jglob)) {
172 >          toposForAtom[i].push_back(j);
173 >          topoDist[i].push_back(1);
174 >        } else {
175 >          if (oneThree->hasPair(iglob, jglob)) {
176 >            toposForAtom[i].push_back(j);
177 >            topoDist[i].push_back(2);
178 >          } else {
179 >            if (oneFour->hasPair(iglob, jglob)) {
180 >              toposForAtom[i].push_back(j);
181 >              topoDist[i].push_back(3);
182 >            }
183 >          }
184 >        }
185 >      }      
186 >    }
187 >
188   #endif
189 +
190 +    groupList_.clear();
191 +    groupList_.resize(nGroups_);
192 +    for (int i = 0; i < nGroups_; i++) {
193 +      int gid = cgLocalToGlobal[i];
194 +      for (int j = 0; j < nLocal_; j++) {
195 +        int aid = AtomLocalToGlobal[j];
196 +        if (globalGroupMembership[aid] == gid) {
197 +          groupList_[i].push_back(j);
198 +        }
199 +      }      
200 +    }
201 +
202 +    excludesForAtom.clear();
203 +    excludesForAtom.resize(nLocal_);
204 +    toposForAtom.clear();
205 +    toposForAtom.resize(nLocal_);
206 +    topoDist.clear();
207 +    topoDist.resize(nLocal_);
208 +
209 +    for (int i = 0; i < nLocal_; i++) {
210 +      int iglob = AtomLocalToGlobal[i];
211 +
212 +      for (int j = 0; j < nLocal_; j++) {
213 +        int jglob = AtomLocalToGlobal[j];
214 +
215 +        if (excludes->hasPair(iglob, jglob))
216 +          excludesForAtom[i].push_back(j);              
217 +        
218 +        if (oneTwo->hasPair(iglob, jglob)) {
219 +          toposForAtom[i].push_back(j);
220 +          topoDist[i].push_back(1);
221 +        } else {
222 +          if (oneThree->hasPair(iglob, jglob)) {
223 +            toposForAtom[i].push_back(j);
224 +            topoDist[i].push_back(2);
225 +          } else {
226 +            if (oneFour->hasPair(iglob, jglob)) {
227 +              toposForAtom[i].push_back(j);
228 +              topoDist[i].push_back(3);
229 +            }
230 +          }
231 +        }
232 +      }      
233 +    }
234 +    
235 +    createGtypeCutoffMap();
236 +
237    }
238 +  
239 +  void ForceMatrixDecomposition::createGtypeCutoffMap() {
240      
241 +    RealType tol = 1e-6;
242 +    RealType rc;
243 +    int atid;
244 +    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
245 +    map<int, RealType> atypeCutoff;
246 +      
247 +    for (set<AtomType*>::iterator at = atypes.begin();
248 +         at != atypes.end(); ++at){
249 +      atid = (*at)->getIdent();
250 +      if (userChoseCutoff_)
251 +        atypeCutoff[atid] = userCutoff_;
252 +      else
253 +        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
254 +    }
255  
256 +    vector<RealType> gTypeCutoffs;
257 +    // first we do a single loop over the cutoff groups to find the
258 +    // largest cutoff for any atypes present in this group.
259 + #ifdef IS_MPI
260 +    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
261 +    groupRowToGtype.resize(nGroupsInRow_);
262 +    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
263 +      vector<int> atomListRow = getAtomsInGroupRow(cg1);
264 +      for (vector<int>::iterator ia = atomListRow.begin();
265 +           ia != atomListRow.end(); ++ia) {            
266 +        int atom1 = (*ia);
267 +        atid = identsRow[atom1];
268 +        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
269 +          groupCutoffRow[cg1] = atypeCutoff[atid];
270 +        }
271 +      }
272  
273 +      bool gTypeFound = false;
274 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
275 +        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
276 +          groupRowToGtype[cg1] = gt;
277 +          gTypeFound = true;
278 +        }
279 +      }
280 +      if (!gTypeFound) {
281 +        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
282 +        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
283 +      }
284 +      
285 +    }
286 +    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
287 +    groupColToGtype.resize(nGroupsInCol_);
288 +    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
289 +      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
290 +      for (vector<int>::iterator jb = atomListCol.begin();
291 +           jb != atomListCol.end(); ++jb) {            
292 +        int atom2 = (*jb);
293 +        atid = identsCol[atom2];
294 +        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
295 +          groupCutoffCol[cg2] = atypeCutoff[atid];
296 +        }
297 +      }
298 +      bool gTypeFound = false;
299 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
300 +        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
301 +          groupColToGtype[cg2] = gt;
302 +          gTypeFound = true;
303 +        }
304 +      }
305 +      if (!gTypeFound) {
306 +        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
307 +        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
308 +      }
309 +    }
310 + #else
311 +
312 +    vector<RealType> groupCutoff(nGroups_, 0.0);
313 +    groupToGtype.resize(nGroups_);
314 +    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
315 +
316 +      groupCutoff[cg1] = 0.0;
317 +      vector<int> atomList = getAtomsInGroupRow(cg1);
318 +
319 +      for (vector<int>::iterator ia = atomList.begin();
320 +           ia != atomList.end(); ++ia) {            
321 +        int atom1 = (*ia);
322 +        atid = idents[atom1];
323 +        if (atypeCutoff[atid] > groupCutoff[cg1]) {
324 +          groupCutoff[cg1] = atypeCutoff[atid];
325 +        }
326 +      }
327 +
328 +      bool gTypeFound = false;
329 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
330 +        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
331 +          groupToGtype[cg1] = gt;
332 +          gTypeFound = true;
333 +        }
334 +      }
335 +      if (!gTypeFound) {
336 +        gTypeCutoffs.push_back( groupCutoff[cg1] );
337 +        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
338 +      }      
339 +    }
340 + #endif
341 +
342 +    // Now we find the maximum group cutoff value present in the simulation
343 +
344 +    RealType groupMax = *max_element(gTypeCutoffs.begin(),
345 +                                     gTypeCutoffs.end());
346 +
347 + #ifdef IS_MPI
348 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
349 +                              MPI::MAX);
350 + #endif
351 +    
352 +    RealType tradRcut = groupMax;
353 +
354 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
355 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
356 +        RealType thisRcut;
357 +        switch(cutoffPolicy_) {
358 +        case TRADITIONAL:
359 +          thisRcut = tradRcut;
360 +          break;
361 +        case MIX:
362 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
363 +          break;
364 +        case MAX:
365 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
366 +          break;
367 +        default:
368 +          sprintf(painCave.errMsg,
369 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
370 +                  "hit an unknown cutoff policy!\n");
371 +          painCave.severity = OPENMD_ERROR;
372 +          painCave.isFatal = 1;
373 +          simError();
374 +          break;
375 +        }
376 +
377 +        pair<int,int> key = make_pair(i,j);
378 +        gTypeCutoffMap[key].first = thisRcut;
379 +
380 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
381 +
382 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
383 +        
384 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
385 +
386 +        // sanity check
387 +        
388 +        if (userChoseCutoff_) {
389 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
390 +            sprintf(painCave.errMsg,
391 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
392 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
393 +            painCave.severity = OPENMD_ERROR;
394 +            painCave.isFatal = 1;
395 +            simError();            
396 +          }
397 +        }
398 +      }
399 +    }
400 +  }
401 +
402 +
403 +  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
404 +    int i, j;  
405 + #ifdef IS_MPI
406 +    i = groupRowToGtype[cg1];
407 +    j = groupColToGtype[cg2];
408 + #else
409 +    i = groupToGtype[cg1];
410 +    j = groupToGtype[cg2];
411 + #endif    
412 +    return gTypeCutoffMap[make_pair(i,j)];
413 +  }
414 +
415 +  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
416 +    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
417 +      if (toposForAtom[atom1][j] == atom2)
418 +        return topoDist[atom1][j];
419 +    }
420 +    return 0;
421 +  }
422 +
423 +  void ForceMatrixDecomposition::zeroWorkArrays() {
424 +    pairwisePot = 0.0;
425 +    embeddingPot = 0.0;
426 +
427 + #ifdef IS_MPI
428 +    if (storageLayout_ & DataStorage::dslForce) {
429 +      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
430 +      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
431 +    }
432 +
433 +    if (storageLayout_ & DataStorage::dslTorque) {
434 +      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
435 +      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
436 +    }
437 +    
438 +    fill(pot_row.begin(), pot_row.end(),
439 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
440 +
441 +    fill(pot_col.begin(), pot_col.end(),
442 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
443 +
444 +    if (storageLayout_ & DataStorage::dslParticlePot) {    
445 +      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
446 +           0.0);
447 +      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
448 +           0.0);
449 +    }
450 +
451 +    if (storageLayout_ & DataStorage::dslDensity) {      
452 +      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
453 +      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
454 +    }
455 +
456 +    if (storageLayout_ & DataStorage::dslFunctional) {  
457 +      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
458 +           0.0);
459 +      fill(atomColData.functional.begin(), atomColData.functional.end(),
460 +           0.0);
461 +    }
462 +
463 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
464 +      fill(atomRowData.functionalDerivative.begin(),
465 +           atomRowData.functionalDerivative.end(), 0.0);
466 +      fill(atomColData.functionalDerivative.begin(),
467 +           atomColData.functionalDerivative.end(), 0.0);
468 +    }
469 +
470 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
471 +      fill(atomRowData.skippedCharge.begin(),
472 +           atomRowData.skippedCharge.end(), 0.0);
473 +      fill(atomColData.skippedCharge.begin(),
474 +           atomColData.skippedCharge.end(), 0.0);
475 +    }
476 +
477 + #endif
478 +    // even in parallel, we need to zero out the local arrays:
479 +
480 +    if (storageLayout_ & DataStorage::dslParticlePot) {      
481 +      fill(snap_->atomData.particlePot.begin(),
482 +           snap_->atomData.particlePot.end(), 0.0);
483 +    }
484 +    
485 +    if (storageLayout_ & DataStorage::dslDensity) {      
486 +      fill(snap_->atomData.density.begin(),
487 +           snap_->atomData.density.end(), 0.0);
488 +    }
489 +    if (storageLayout_ & DataStorage::dslFunctional) {
490 +      fill(snap_->atomData.functional.begin(),
491 +           snap_->atomData.functional.end(), 0.0);
492 +    }
493 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
494 +      fill(snap_->atomData.functionalDerivative.begin(),
495 +           snap_->atomData.functionalDerivative.end(), 0.0);
496 +    }
497 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
498 +      fill(snap_->atomData.skippedCharge.begin(),
499 +           snap_->atomData.skippedCharge.end(), 0.0);
500 +    }
501 +    
502 +  }
503 +
504 +
505    void ForceMatrixDecomposition::distributeData()  {
506      snap_ = sman_->getCurrentSnapshot();
507      storageLayout_ = sman_->getStorageLayout();
# Line 153 | Line 534 | namespace OpenMD {
534        AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
535                                     atomColData.electroFrame);
536      }
537 +
538   #endif      
539    }
540    
541 +  /* collects information obtained during the pre-pair loop onto local
542 +   * data structures.
543 +   */
544    void ForceMatrixDecomposition::collectIntermediateData() {
545      snap_ = sman_->getCurrentSnapshot();
546      storageLayout_ = sman_->getStorageLayout();
# Line 167 | Line 552 | namespace OpenMD {
552                                 snap_->atomData.density);
553        
554        int n = snap_->atomData.density.size();
555 <      std::vector<RealType> rho_tmp(n, 0.0);
555 >      vector<RealType> rho_tmp(n, 0.0);
556        AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
557        for (int i = 0; i < n; i++)
558          snap_->atomData.density[i] += rho_tmp[i];
559      }
560   #endif
561    }
562 <  
562 >
563 >  /*
564 >   * redistributes information obtained during the pre-pair loop out to
565 >   * row and column-indexed data structures
566 >   */
567    void ForceMatrixDecomposition::distributeIntermediateData() {
568      snap_ = sman_->getCurrentSnapshot();
569      storageLayout_ = sman_->getStorageLayout();
# Line 212 | Line 601 | namespace OpenMD {
601      AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
602      for (int i = 0; i < n; i++)
603        snap_->atomData.force[i] += frc_tmp[i];
604 <    
216 <    
604 >        
605      if (storageLayout_ & DataStorage::dslTorque) {
606  
607 <      int nt = snap_->atomData.force.size();
607 >      int nt = snap_->atomData.torque.size();
608        vector<Vector3d> trq_tmp(nt, V3Zero);
609  
610        AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
611 <      for (int i = 0; i < n; i++) {
611 >      for (int i = 0; i < nt; i++) {
612          snap_->atomData.torque[i] += trq_tmp[i];
613          trq_tmp[i] = 0.0;
614        }
615        
616        AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
617 <      for (int i = 0; i < n; i++)
617 >      for (int i = 0; i < nt; i++)
618          snap_->atomData.torque[i] += trq_tmp[i];
619      }
620 +
621 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
622 +
623 +      int ns = snap_->atomData.skippedCharge.size();
624 +      vector<RealType> skch_tmp(ns, 0.0);
625 +
626 +      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
627 +      for (int i = 0; i < ns; i++) {
628 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
629 +        skch_tmp[i] = 0.0;
630 +      }
631 +      
632 +      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
633 +      for (int i = 0; i < ns; i++)
634 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
635 +    }
636      
637      nLocal_ = snap_->getNumberOfAtoms();
638  
639 <    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
640 <                                       vector<RealType> (nLocal_, 0.0));
639 >    vector<potVec> pot_temp(nLocal_,
640 >                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
641 >
642 >    // scatter/gather pot_row into the members of my column
643 >          
644 >    AtomCommPotRow->scatter(pot_row, pot_temp);
645 >
646 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
647 >      pairwisePot += pot_temp[ii];
648      
649 <    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
650 <      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
651 <      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
652 <        pot_local[i] += pot_temp[i][ii];
653 <      }
654 <    }
649 >    fill(pot_temp.begin(), pot_temp.end(),
650 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
651 >      
652 >    AtomCommPotColumn->scatter(pot_col, pot_temp);    
653 >    
654 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
655 >      pairwisePot += pot_temp[ii];    
656   #endif
657 +
658    }
659  
660 +  int ForceMatrixDecomposition::getNAtomsInRow() {  
661 + #ifdef IS_MPI
662 +    return nAtomsInRow_;
663 + #else
664 +    return nLocal_;
665 + #endif
666 +  }
667 +
668 +  /**
669 +   * returns the list of atoms belonging to this group.  
670 +   */
671 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
672 + #ifdef IS_MPI
673 +    return groupListRow_[cg1];
674 + #else
675 +    return groupList_[cg1];
676 + #endif
677 +  }
678 +
679 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
680 + #ifdef IS_MPI
681 +    return groupListCol_[cg2];
682 + #else
683 +    return groupList_[cg2];
684 + #endif
685 +  }
686    
687    Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
688      Vector3d d;
# Line 284 | Line 723 | namespace OpenMD {
723      
724      snap_->wrapVector(d);
725      return d;    
726 +  }
727 +
728 +  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
729 + #ifdef IS_MPI
730 +    return massFactorsRow[atom1];
731 + #else
732 +    return massFactors[atom1];
733 + #endif
734 +  }
735 +
736 +  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
737 + #ifdef IS_MPI
738 +    return massFactorsCol[atom2];
739 + #else
740 +    return massFactors[atom2];
741 + #endif
742 +
743    }
744      
745    Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
# Line 299 | Line 755 | namespace OpenMD {
755      return d;    
756    }
757  
758 +  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
759 +    return excludesForAtom[atom1];
760 +  }
761 +
762 +  /**
763 +   * We need to exclude some overcounted interactions that result from
764 +   * the parallel decomposition.
765 +   */
766 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
767 +    int unique_id_1, unique_id_2;
768 +
769 + #ifdef IS_MPI
770 +    // in MPI, we have to look up the unique IDs for each atom
771 +    unique_id_1 = AtomRowToGlobal[atom1];
772 +    unique_id_2 = AtomColToGlobal[atom2];
773 +
774 +    // this situation should only arise in MPI simulations
775 +    if (unique_id_1 == unique_id_2) return true;
776 +    
777 +    // this prevents us from doing the pair on multiple processors
778 +    if (unique_id_1 < unique_id_2) {
779 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
780 +    } else {
781 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
782 +    }
783 + #endif
784 +    return false;
785 +  }
786 +
787 +  /**
788 +   * We need to handle the interactions for atoms who are involved in
789 +   * the same rigid body as well as some short range interactions
790 +   * (bonds, bends, torsions) differently from other interactions.
791 +   * We'll still visit the pairwise routines, but with a flag that
792 +   * tells those routines to exclude the pair from direct long range
793 +   * interactions.  Some indirect interactions (notably reaction
794 +   * field) must still be handled for these pairs.
795 +   */
796 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
797 +    int unique_id_2;
798 +    
799 + #ifdef IS_MPI
800 +    // in MPI, we have to look up the unique IDs for the row atom.
801 +    unique_id_2 = AtomColToGlobal[atom2];
802 + #else
803 +    // in the normal loop, the atom numbers are unique
804 +    unique_id_2 = atom2;
805 + #endif
806 +    
807 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
808 +         i != excludesForAtom[atom1].end(); ++i) {
809 +      if ( (*i) == unique_id_2 ) return true;
810 +    }
811 +
812 +    return false;
813 +  }
814 +
815 +
816    void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
817   #ifdef IS_MPI
818      atomRowData.force[atom1] += fg;
# Line 316 | Line 830 | namespace OpenMD {
830    }
831  
832      // filling interaction blocks with pointers
833 <  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
834 <    InteractionData idat;
833 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
834 >                                                     int atom1, int atom2) {
835  
836 +    idat.excluded = excludeAtomPair(atom1, atom2);
837 +  
838   #ifdef IS_MPI
839 +    
840 +    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
841 +                             ff_->getAtomType(identsCol[atom2]) );
842 +    
843      if (storageLayout_ & DataStorage::dslAmat) {
844        idat.A1 = &(atomRowData.aMat[atom1]);
845        idat.A2 = &(atomColData.aMat[atom2]);
# Line 340 | Line 860 | namespace OpenMD {
860        idat.rho2 = &(atomColData.density[atom2]);
861      }
862  
863 +    if (storageLayout_ & DataStorage::dslFunctional) {
864 +      idat.frho1 = &(atomRowData.functional[atom1]);
865 +      idat.frho2 = &(atomColData.functional[atom2]);
866 +    }
867 +
868      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
869        idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
870        idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
871      }
872 +
873 +    if (storageLayout_ & DataStorage::dslParticlePot) {
874 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
875 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
876 +    }
877 +
878 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
879 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
880 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
881 +    }
882 +
883   #else
884 +
885 +    idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
886 +                             ff_->getAtomType(idents[atom2]) );
887 +
888      if (storageLayout_ & DataStorage::dslAmat) {
889        idat.A1 = &(snap_->atomData.aMat[atom1]);
890        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 360 | Line 900 | namespace OpenMD {
900        idat.t2 = &(snap_->atomData.torque[atom2]);
901      }
902  
903 <    if (storageLayout_ & DataStorage::dslDensity) {
903 >    if (storageLayout_ & DataStorage::dslDensity) {    
904        idat.rho1 = &(snap_->atomData.density[atom1]);
905        idat.rho2 = &(snap_->atomData.density[atom2]);
906      }
907  
908 +    if (storageLayout_ & DataStorage::dslFunctional) {
909 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
910 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
911 +    }
912 +
913      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
914        idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
915        idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
916      }
917 +
918 +    if (storageLayout_ & DataStorage::dslParticlePot) {
919 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
920 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
921 +    }
922 +
923 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
924 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
925 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
926 +    }
927   #endif
373 –    return idat;
928    }
929  
930 <  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
931 <
378 <    InteractionData idat;
930 >  
931 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
932   #ifdef IS_MPI
933 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
934 <      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
935 <      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
936 <    }
937 <    if (storageLayout_ & DataStorage::dslTorque) {
385 <      idat.t1 = &(atomRowData.torque[atom1]);
386 <      idat.t2 = &(atomColData.torque[atom2]);
387 <    }
388 <    if (storageLayout_ & DataStorage::dslForce) {
389 <      idat.t1 = &(atomRowData.force[atom1]);
390 <      idat.t2 = &(atomColData.force[atom2]);
391 <    }
933 >    pot_row[atom1] += 0.5 *  *(idat.pot);
934 >    pot_col[atom2] += 0.5 *  *(idat.pot);
935 >
936 >    atomRowData.force[atom1] += *(idat.f1);
937 >    atomColData.force[atom2] -= *(idat.f1);
938   #else
939 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
940 <      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
941 <      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
942 <    }
397 <    if (storageLayout_ & DataStorage::dslTorque) {
398 <      idat.t1 = &(snap_->atomData.torque[atom1]);
399 <      idat.t2 = &(snap_->atomData.torque[atom2]);
400 <    }
401 <    if (storageLayout_ & DataStorage::dslForce) {
402 <      idat.t1 = &(snap_->atomData.force[atom1]);
403 <      idat.t2 = &(snap_->atomData.force[atom2]);
404 <    }
939 >    pairwisePot += *(idat.pot);
940 >
941 >    snap_->atomData.force[atom1] += *(idat.f1);
942 >    snap_->atomData.force[atom2] -= *(idat.f1);
943   #endif
944      
945    }
946  
409 –  SelfData ForceMatrixDecomposition::fillSelfData(int atom1) {
410 –    SelfData sdat;
411 –    // Still Missing atype, skippedCharge, potVec pot,
412 –    if (storageLayout_ & DataStorage::dslElectroFrame) {
413 –      sdat.eFrame = &(snap_->atomData.electroFrame[atom1]);
414 –    }
415 –    
416 –    if (storageLayout_ & DataStorage::dslTorque) {
417 –      sdat.t = &(snap_->atomData.torque[atom1]);
418 –    }
419 –    
420 –    if (storageLayout_ & DataStorage::dslDensity) {
421 –      sdat.rho = &(snap_->atomData.density[atom1]);
422 –    }
423 –    
424 –    if (storageLayout_ & DataStorage::dslFunctional) {
425 –      sdat.frho = &(snap_->atomData.functional[atom1]);
426 –    }
427 –    
428 –    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
429 –      sdat.dfrhodrho = &(snap_->atomData.functionalDerivative[atom1]);
430 –    }
431 –
432 –    return sdat;    
433 –  }
434 –
435 –
436 –
947    /*
948     * buildNeighborList
949     *
# Line 443 | Line 953 | namespace OpenMD {
953    vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
954        
955      vector<pair<int, int> > neighborList;
956 +    groupCutoffs cuts;
957 +    bool doAllPairs = false;
958 +
959   #ifdef IS_MPI
960 <    CellListRow.clear();
961 <    CellListCol.clear();
960 >    cellListRow_.clear();
961 >    cellListCol_.clear();
962   #else
963 <    CellList.clear();
963 >    cellList_.clear();
964   #endif
965  
966 <    // dangerous to not do error checking.
454 <    RealType skinThickness_ = info_->getSimParams()->getSkinThickness();
455 <    RealType rCut_;
456 <
457 <    RealType rList_ = (rCut_ + skinThickness_);
966 >    RealType rList_ = (largestRcut_ + skinThickness_);
967      RealType rl2 = rList_ * rList_;
968      Snapshot* snap_ = sman_->getCurrentSnapshot();
969      Mat3x3d Hmat = snap_->getHmat();
970      Vector3d Hx = Hmat.getColumn(0);
971      Vector3d Hy = Hmat.getColumn(1);
972      Vector3d Hz = Hmat.getColumn(2);
464 –    Vector3i nCells;
973  
974 <    nCells.x() = (int) ( Hx.length() )/ rList_;
975 <    nCells.y() = (int) ( Hy.length() )/ rList_;
976 <    nCells.z() = (int) ( Hz.length() )/ rList_;
974 >    nCells_.x() = (int) ( Hx.length() )/ rList_;
975 >    nCells_.y() = (int) ( Hy.length() )/ rList_;
976 >    nCells_.z() = (int) ( Hz.length() )/ rList_;
977  
978 +    // handle small boxes where the cell offsets can end up repeating cells
979 +    
980 +    if (nCells_.x() < 3) doAllPairs = true;
981 +    if (nCells_.y() < 3) doAllPairs = true;
982 +    if (nCells_.z() < 3) doAllPairs = true;
983 +
984      Mat3x3d invHmat = snap_->getInvHmat();
985      Vector3d rs, scaled, dr;
986      Vector3i whichCell;
987      int cellIndex;
988 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
989  
990   #ifdef IS_MPI
991 <    for (int i = 0; i < nGroupsInRow_; i++) {
992 <      rs = cgRowData.position[i];
993 <      // scaled positions relative to the box vectors
994 <      scaled = invHmat * rs;
995 <      // wrap the vector back into the unit box by subtracting integer box
481 <      // numbers
482 <      for (int j = 0; j < 3; j++)
483 <        scaled[j] -= roundMe(scaled[j]);
484 <    
485 <      // find xyz-indices of cell that cutoffGroup is in.
486 <      whichCell.x() = nCells.x() * scaled.x();
487 <      whichCell.y() = nCells.y() * scaled.y();
488 <      whichCell.z() = nCells.z() * scaled.z();
991 >    cellListRow_.resize(nCtot);
992 >    cellListCol_.resize(nCtot);
993 > #else
994 >    cellList_.resize(nCtot);
995 > #endif
996  
997 <      // find single index of this cell:
998 <      cellIndex = Vlinear(whichCell, nCells);
492 <      // add this cutoff group to the list of groups in this cell;
493 <      CellListRow[cellIndex].push_back(i);
494 <    }
495 <
496 <    for (int i = 0; i < nGroupsInCol_; i++) {
497 <      rs = cgColData.position[i];
498 <      // scaled positions relative to the box vectors
499 <      scaled = invHmat * rs;
500 <      // wrap the vector back into the unit box by subtracting integer box
501 <      // numbers
502 <      for (int j = 0; j < 3; j++)
503 <        scaled[j] -= roundMe(scaled[j]);
504 <
505 <      // find xyz-indices of cell that cutoffGroup is in.
506 <      whichCell.x() = nCells.x() * scaled.x();
507 <      whichCell.y() = nCells.y() * scaled.y();
508 <      whichCell.z() = nCells.z() * scaled.z();
509 <
510 <      // find single index of this cell:
511 <      cellIndex = Vlinear(whichCell, nCells);
512 <      // add this cutoff group to the list of groups in this cell;
513 <      CellListCol[cellIndex].push_back(i);
514 <    }
515 < #else
516 <    for (int i = 0; i < nGroups_; i++) {
517 <      rs = snap_->cgData.position[i];
518 <      // scaled positions relative to the box vectors
519 <      scaled = invHmat * rs;
520 <      // wrap the vector back into the unit box by subtracting integer box
521 <      // numbers
522 <      for (int j = 0; j < 3; j++)
523 <        scaled[j] -= roundMe(scaled[j]);
997 >    if (!doAllPairs) {
998 > #ifdef IS_MPI
999  
1000 <      // find xyz-indices of cell that cutoffGroup is in.
1001 <      whichCell.x() = nCells.x() * scaled.x();
1002 <      whichCell.y() = nCells.y() * scaled.y();
1003 <      whichCell.z() = nCells.z() * scaled.z();
1004 <
1005 <      // find single index of this cell:
1006 <      cellIndex = Vlinear(whichCell, nCells);
1007 <      // add this cutoff group to the list of groups in this cell;
1008 <      CellList[cellIndex].push_back(i);
1009 <    }
1000 >      for (int i = 0; i < nGroupsInRow_; i++) {
1001 >        rs = cgRowData.position[i];
1002 >        
1003 >        // scaled positions relative to the box vectors
1004 >        scaled = invHmat * rs;
1005 >        
1006 >        // wrap the vector back into the unit box by subtracting integer box
1007 >        // numbers
1008 >        for (int j = 0; j < 3; j++) {
1009 >          scaled[j] -= roundMe(scaled[j]);
1010 >          scaled[j] += 0.5;
1011 >        }
1012 >        
1013 >        // find xyz-indices of cell that cutoffGroup is in.
1014 >        whichCell.x() = nCells_.x() * scaled.x();
1015 >        whichCell.y() = nCells_.y() * scaled.y();
1016 >        whichCell.z() = nCells_.z() * scaled.z();
1017 >        
1018 >        // find single index of this cell:
1019 >        cellIndex = Vlinear(whichCell, nCells_);
1020 >        
1021 >        // add this cutoff group to the list of groups in this cell;
1022 >        cellListRow_[cellIndex].push_back(i);
1023 >      }
1024 >      
1025 >      for (int i = 0; i < nGroupsInCol_; i++) {
1026 >        rs = cgColData.position[i];
1027 >        
1028 >        // scaled positions relative to the box vectors
1029 >        scaled = invHmat * rs;
1030 >        
1031 >        // wrap the vector back into the unit box by subtracting integer box
1032 >        // numbers
1033 >        for (int j = 0; j < 3; j++) {
1034 >          scaled[j] -= roundMe(scaled[j]);
1035 >          scaled[j] += 0.5;
1036 >        }
1037 >        
1038 >        // find xyz-indices of cell that cutoffGroup is in.
1039 >        whichCell.x() = nCells_.x() * scaled.x();
1040 >        whichCell.y() = nCells_.y() * scaled.y();
1041 >        whichCell.z() = nCells_.z() * scaled.z();
1042 >        
1043 >        // find single index of this cell:
1044 >        cellIndex = Vlinear(whichCell, nCells_);
1045 >        
1046 >        // add this cutoff group to the list of groups in this cell;
1047 >        cellListCol_[cellIndex].push_back(i);
1048 >      }
1049 > #else
1050 >      for (int i = 0; i < nGroups_; i++) {
1051 >        rs = snap_->cgData.position[i];
1052 >        
1053 >        // scaled positions relative to the box vectors
1054 >        scaled = invHmat * rs;
1055 >        
1056 >        // wrap the vector back into the unit box by subtracting integer box
1057 >        // numbers
1058 >        for (int j = 0; j < 3; j++) {
1059 >          scaled[j] -= roundMe(scaled[j]);
1060 >          scaled[j] += 0.5;
1061 >        }
1062 >        
1063 >        // find xyz-indices of cell that cutoffGroup is in.
1064 >        whichCell.x() = nCells_.x() * scaled.x();
1065 >        whichCell.y() = nCells_.y() * scaled.y();
1066 >        whichCell.z() = nCells_.z() * scaled.z();
1067 >        
1068 >        // find single index of this cell:
1069 >        cellIndex = Vlinear(whichCell, nCells_);      
1070 >        
1071 >        // add this cutoff group to the list of groups in this cell;
1072 >        cellList_[cellIndex].push_back(i);
1073 >      }
1074   #endif
1075  
1076 <
1077 <
1078 <    for (int m1z = 0; m1z < nCells.z(); m1z++) {
1079 <      for (int m1y = 0; m1y < nCells.y(); m1y++) {
1080 <        for (int m1x = 0; m1x < nCells.x(); m1x++) {
1081 <          Vector3i m1v(m1x, m1y, m1z);
1082 <          int m1 = Vlinear(m1v, nCells);
1083 <          for (int offset = 0; offset < nOffset_; offset++) {
1084 <            Vector3i m2v = m1v + cellOffsets_[offset];
1085 <
1086 <            if (m2v.x() >= nCells.x()) {
1087 <              m2v.x() = 0;          
1088 <            } else if (m2v.x() < 0) {
1089 <              m2v.x() = nCells.x() - 1;
1090 <            }
1091 <
1092 <            if (m2v.y() >= nCells.y()) {
1093 <              m2v.y() = 0;          
1094 <            } else if (m2v.y() < 0) {
1095 <              m2v.y() = nCells.y() - 1;
1096 <            }
1097 <
1098 <            if (m2v.z() >= nCells.z()) {
1099 <              m2v.z() = 0;          
1100 <            } else if (m2v.z() < 0) {
1101 <              m2v.z() = nCells.z() - 1;
1102 <            }
1103 <
1104 <            int m2 = Vlinear (m2v, nCells);
1105 <
1076 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1077 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1078 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1079 >            Vector3i m1v(m1x, m1y, m1z);
1080 >            int m1 = Vlinear(m1v, nCells_);
1081 >            
1082 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1083 >                 os != cellOffsets_.end(); ++os) {
1084 >              
1085 >              Vector3i m2v = m1v + (*os);
1086 >              
1087 >              if (m2v.x() >= nCells_.x()) {
1088 >                m2v.x() = 0;          
1089 >              } else if (m2v.x() < 0) {
1090 >                m2v.x() = nCells_.x() - 1;
1091 >              }
1092 >              
1093 >              if (m2v.y() >= nCells_.y()) {
1094 >                m2v.y() = 0;          
1095 >              } else if (m2v.y() < 0) {
1096 >                m2v.y() = nCells_.y() - 1;
1097 >              }
1098 >              
1099 >              if (m2v.z() >= nCells_.z()) {
1100 >                m2v.z() = 0;          
1101 >              } else if (m2v.z() < 0) {
1102 >                m2v.z() = nCells_.z() - 1;
1103 >              }
1104 >              
1105 >              int m2 = Vlinear (m2v, nCells_);
1106 >              
1107   #ifdef IS_MPI
1108 <            for (vector<int>::iterator j1 = CellListRow[m1].begin();
1109 <                 j1 != CellListRow[m1].end(); ++j1) {
1110 <              for (vector<int>::iterator j2 = CellListCol[m2].begin();
1111 <                   j2 != CellListCol[m2].end(); ++j2) {
1112 <                              
1113 <                // Always do this if we're in different cells or if
1114 <                // we're in the same cell and the global index of the
1115 <                // j2 cutoff group is less than the j1 cutoff group
1116 <
1117 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1118 <                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1119 <                  snap_->wrapVector(dr);
1120 <                  if (dr.lengthSquare() < rl2) {
1121 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1108 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1109 >                   j1 != cellListRow_[m1].end(); ++j1) {
1110 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1111 >                     j2 != cellListCol_[m2].end(); ++j2) {
1112 >                  
1113 >                  // Always do this if we're in different cells or if
1114 >                  // we're in the same cell and the global index of the
1115 >                  // j2 cutoff group is less than the j1 cutoff group
1116 >                  
1117 >                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1118 >                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1119 >                    snap_->wrapVector(dr);
1120 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1121 >                    if (dr.lengthSquare() < cuts.third) {
1122 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1123 >                    }
1124                    }
1125                  }
1126                }
585 –            }
1127   #else
1128 <            for (vector<int>::iterator j1 = CellList[m1].begin();
1129 <                 j1 != CellList[m1].end(); ++j1) {
1130 <              for (vector<int>::iterator j2 = CellList[m2].begin();
1131 <                   j2 != CellList[m2].end(); ++j2) {
1132 <                              
1133 <                // Always do this if we're in different cells or if
1134 <                // we're in the same cell and the global index of the
1135 <                // j2 cutoff group is less than the j1 cutoff group
1136 <
1137 <                if (m2 != m1 || (*j2) < (*j1)) {
1138 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1139 <                  snap_->wrapVector(dr);
1140 <                  if (dr.lengthSquare() < rl2) {
1141 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1128 >              
1129 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1130 >                   j1 != cellList_[m1].end(); ++j1) {
1131 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1132 >                     j2 != cellList_[m2].end(); ++j2) {
1133 >                  
1134 >                  // Always do this if we're in different cells or if
1135 >                  // we're in the same cell and the global index of the
1136 >                  // j2 cutoff group is less than the j1 cutoff group
1137 >                  
1138 >                  if (m2 != m1 || (*j2) < (*j1)) {
1139 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1140 >                    snap_->wrapVector(dr);
1141 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1142 >                    if (dr.lengthSquare() < cuts.third) {
1143 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1144 >                    }
1145                    }
1146                  }
1147                }
604 –            }
1148   #endif
1149 +            }
1150            }
1151          }
1152        }
1153 +    } else {
1154 +      // branch to do all cutoff group pairs
1155 + #ifdef IS_MPI
1156 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1157 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1158 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1159 +          snap_->wrapVector(dr);
1160 +          cuts = getGroupCutoffs( j1, j2 );
1161 +          if (dr.lengthSquare() < cuts.third) {
1162 +            neighborList.push_back(make_pair(j1, j2));
1163 +          }
1164 +        }
1165 +      }
1166 + #else
1167 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1168 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1169 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1170 +          snap_->wrapVector(dr);
1171 +          cuts = getGroupCutoffs( j1, j2 );
1172 +          if (dr.lengthSquare() < cuts.third) {
1173 +            neighborList.push_back(make_pair(j1, j2));
1174 +          }
1175 +        }
1176 +      }        
1177 + #endif
1178      }
1179 +      
1180 +    // save the local cutoff group positions for the check that is
1181 +    // done on each loop:
1182 +    saved_CG_positions_.clear();
1183 +    for (int i = 0; i < nGroups_; i++)
1184 +      saved_CG_positions_.push_back(snap_->cgData.position[i]);
1185 +    
1186      return neighborList;
1187    }
1188   } //end namespace OpenMD

Diff Legend

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