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/ForceDecomposition.cpp (file contents), Revision 1539 by gezelter, Fri Jan 14 22:31:31 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1895 by gezelter, Mon Jul 1 21:09:37 2013 UTC

# Line 35 | Line 35
35   *                                                                      
36   * [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).            
37   * [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).          
38 < * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
39 < * [4]  Vardeman & Gezelter, in progress (2009).                        
38 > * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008).          
39 > * [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40 > * [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41   */
42 < #include "parallel/ForceDecomposition.hpp"
42 < #include "parallel/Communicator.hpp"
42 > #include "parallel/ForceMatrixDecomposition.hpp"
43   #include "math/SquareMatrix3.hpp"
44 + #include "nonbonded/NonBondedInteraction.hpp"
45 + #include "brains/SnapshotManager.hpp"
46 + #include "brains/PairList.hpp"
47  
48 + using namespace std;
49   namespace OpenMD {
50  
51 <  void ForceDecomposition::distributeInitialData() {
51 >  ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52 >
53 >    // In a parallel computation, row and colum scans must visit all
54 >    // surrounding cells (not just the 14 upper triangular blocks that
55 >    // are used when the processor can see all pairs)
56   #ifdef IS_MPI
57 +    cellOffsets_.clear();
58 +    cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59 +    cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60 +    cellOffsets_.push_back( Vector3i( 1,-1,-1) );                          
61 +    cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62 +    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63 +    cellOffsets_.push_back( Vector3i( 1, 0,-1) );
64 +    cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65 +    cellOffsets_.push_back( Vector3i( 0, 1,-1) );      
66 +    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67 +    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68 +    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69 +    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70 +    cellOffsets_.push_back( Vector3i(-1, 0, 0) );      
71 +    cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72 +    cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73 +    cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74 +    cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75 +    cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76 +    cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77 +    cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78 +    cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79 +    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80 +    cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81 +    cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82 +    cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83 +    cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84 +    cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85 + #endif    
86 +  }
87  
50    int nAtoms;
51    int nGroups;
88  
89 <    AtomCommRealI = new Comm<I,RealType>(nAtoms);
90 <    AtomCommVectorI = new Comm<I,Vector3d>(nAtoms);
91 <    AtomCommMatrixI = new Comm<I,Mat3x3d>(nAtoms);
89 >  /**
90 >   * distributeInitialData is essentially a copy of the older fortran
91 >   * SimulationSetup
92 >   */
93 >  void ForceMatrixDecomposition::distributeInitialData() {
94 >    snap_ = sman_->getCurrentSnapshot();
95 >    storageLayout_ = sman_->getStorageLayout();
96 >    ff_ = info_->getForceField();
97 >    nLocal_ = snap_->getNumberOfAtoms();
98 >  
99 >    nGroups_ = info_->getNLocalCutoffGroups();
100 >    // gather the information for atomtype IDs (atids):
101 >    idents = info_->getIdentArray();
102 >    AtomLocalToGlobal = info_->getGlobalAtomIndices();
103 >    cgLocalToGlobal = info_->getGlobalGroupIndices();
104 >    vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
105  
106 <    AtomCommRealJ = new Comm<J,RealType>(nAtoms);
58 <    AtomCommVectorJ = new Comm<J,Vector3d>(nAtoms);
59 <    AtomCommMatrixJ = new Comm<J,Mat3x3d>(nAtoms);
106 >    massFactors = info_->getMassFactors();
107  
108 <    cgCommVectorI = new Comm<I,Vector3d>(nGroups);
109 <    cgCommVectorJ = new Comm<J,Vector3d>(nGroups);
110 <    // more to come
108 >    PairList* excludes = info_->getExcludedInteractions();
109 >    PairList* oneTwo = info_->getOneTwoInteractions();
110 >    PairList* oneThree = info_->getOneThreeInteractions();
111 >    PairList* oneFour = info_->getOneFourInteractions();
112 >    
113 >    if (needVelocities_)
114 >      snap_->cgData.setStorageLayout(DataStorage::dslPosition |
115 >                                     DataStorage::dslVelocity);
116 >    else
117 >      snap_->cgData.setStorageLayout(DataStorage::dslPosition);
118 >    
119 > #ifdef IS_MPI
120 >
121 >    MPI::Intracomm row = rowComm.getComm();
122 >    MPI::Intracomm col = colComm.getComm();
123 >
124 >    AtomPlanIntRow = new Plan<int>(row, nLocal_);
125 >    AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
126 >    AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
127 >    AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
128 >    AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
129 >
130 >    AtomPlanIntColumn = new Plan<int>(col, nLocal_);
131 >    AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
132 >    AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
133 >    AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
134 >    AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
135 >
136 >    cgPlanIntRow = new Plan<int>(row, nGroups_);
137 >    cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
138 >    cgPlanIntColumn = new Plan<int>(col, nGroups_);
139 >    cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
140 >
141 >    nAtomsInRow_ = AtomPlanIntRow->getSize();
142 >    nAtomsInCol_ = AtomPlanIntColumn->getSize();
143 >    nGroupsInRow_ = cgPlanIntRow->getSize();
144 >    nGroupsInCol_ = cgPlanIntColumn->getSize();
145 >
146 >    // Modify the data storage objects with the correct layouts and sizes:
147 >    atomRowData.resize(nAtomsInRow_);
148 >    atomRowData.setStorageLayout(storageLayout_);
149 >    atomColData.resize(nAtomsInCol_);
150 >    atomColData.setStorageLayout(storageLayout_);
151 >    cgRowData.resize(nGroupsInRow_);
152 >    cgRowData.setStorageLayout(DataStorage::dslPosition);
153 >    cgColData.resize(nGroupsInCol_);
154 >    if (needVelocities_)
155 >      // we only need column velocities if we need them.
156 >      cgColData.setStorageLayout(DataStorage::dslPosition |
157 >                                 DataStorage::dslVelocity);
158 >    else    
159 >      cgColData.setStorageLayout(DataStorage::dslPosition);
160 >      
161 >    identsRow.resize(nAtomsInRow_);
162 >    identsCol.resize(nAtomsInCol_);
163 >    
164 >    AtomPlanIntRow->gather(idents, identsRow);
165 >    AtomPlanIntColumn->gather(idents, identsCol);
166 >    
167 >    // allocate memory for the parallel objects
168 >    atypesRow.resize(nAtomsInRow_);
169 >    atypesCol.resize(nAtomsInCol_);
170 >
171 >    for (int i = 0; i < nAtomsInRow_; i++)
172 >      atypesRow[i] = ff_->getAtomType(identsRow[i]);
173 >    for (int i = 0; i < nAtomsInCol_; i++)
174 >      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
175 >
176 >    pot_row.resize(nAtomsInRow_);
177 >    pot_col.resize(nAtomsInCol_);
178 >
179 >    expot_row.resize(nAtomsInRow_);
180 >    expot_col.resize(nAtomsInCol_);
181 >
182 >    AtomRowToGlobal.resize(nAtomsInRow_);
183 >    AtomColToGlobal.resize(nAtomsInCol_);
184 >    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185 >    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186 >
187 >    cgRowToGlobal.resize(nGroupsInRow_);
188 >    cgColToGlobal.resize(nGroupsInCol_);
189 >    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190 >    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
191 >
192 >    massFactorsRow.resize(nAtomsInRow_);
193 >    massFactorsCol.resize(nAtomsInCol_);
194 >    AtomPlanRealRow->gather(massFactors, massFactorsRow);
195 >    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
196 >
197 >    groupListRow_.clear();
198 >    groupListRow_.resize(nGroupsInRow_);
199 >    for (int i = 0; i < nGroupsInRow_; i++) {
200 >      int gid = cgRowToGlobal[i];
201 >      for (int j = 0; j < nAtomsInRow_; j++) {
202 >        int aid = AtomRowToGlobal[j];
203 >        if (globalGroupMembership[aid] == gid)
204 >          groupListRow_[i].push_back(j);
205 >      }      
206 >    }
207 >
208 >    groupListCol_.clear();
209 >    groupListCol_.resize(nGroupsInCol_);
210 >    for (int i = 0; i < nGroupsInCol_; i++) {
211 >      int gid = cgColToGlobal[i];
212 >      for (int j = 0; j < nAtomsInCol_; j++) {
213 >        int aid = AtomColToGlobal[j];
214 >        if (globalGroupMembership[aid] == gid)
215 >          groupListCol_[i].push_back(j);
216 >      }      
217 >    }
218 >
219 >    excludesForAtom.clear();
220 >    excludesForAtom.resize(nAtomsInRow_);
221 >    toposForAtom.clear();
222 >    toposForAtom.resize(nAtomsInRow_);
223 >    topoDist.clear();
224 >    topoDist.resize(nAtomsInRow_);
225 >    for (int i = 0; i < nAtomsInRow_; i++) {
226 >      int iglob = AtomRowToGlobal[i];
227 >
228 >      for (int j = 0; j < nAtomsInCol_; j++) {
229 >        int jglob = AtomColToGlobal[j];
230 >
231 >        if (excludes->hasPair(iglob, jglob))
232 >          excludesForAtom[i].push_back(j);      
233 >        
234 >        if (oneTwo->hasPair(iglob, jglob)) {
235 >          toposForAtom[i].push_back(j);
236 >          topoDist[i].push_back(1);
237 >        } else {
238 >          if (oneThree->hasPair(iglob, jglob)) {
239 >            toposForAtom[i].push_back(j);
240 >            topoDist[i].push_back(2);
241 >          } else {
242 >            if (oneFour->hasPair(iglob, jglob)) {
243 >              toposForAtom[i].push_back(j);
244 >              topoDist[i].push_back(3);
245 >            }
246 >          }
247 >        }
248 >      }      
249 >    }
250 >
251 > #else
252 >    excludesForAtom.clear();
253 >    excludesForAtom.resize(nLocal_);
254 >    toposForAtom.clear();
255 >    toposForAtom.resize(nLocal_);
256 >    topoDist.clear();
257 >    topoDist.resize(nLocal_);
258 >
259 >    for (int i = 0; i < nLocal_; i++) {
260 >      int iglob = AtomLocalToGlobal[i];
261 >
262 >      for (int j = 0; j < nLocal_; j++) {
263 >        int jglob = AtomLocalToGlobal[j];
264 >
265 >        if (excludes->hasPair(iglob, jglob))
266 >          excludesForAtom[i].push_back(j);              
267 >        
268 >        if (oneTwo->hasPair(iglob, jglob)) {
269 >          toposForAtom[i].push_back(j);
270 >          topoDist[i].push_back(1);
271 >        } else {
272 >          if (oneThree->hasPair(iglob, jglob)) {
273 >            toposForAtom[i].push_back(j);
274 >            topoDist[i].push_back(2);
275 >          } else {
276 >            if (oneFour->hasPair(iglob, jglob)) {
277 >              toposForAtom[i].push_back(j);
278 >              topoDist[i].push_back(3);
279 >            }
280 >          }
281 >        }
282 >      }      
283 >    }
284   #endif
285 +
286 +    // allocate memory for the parallel objects
287 +    atypesLocal.resize(nLocal_);
288 +
289 +    for (int i = 0; i < nLocal_; i++)
290 +      atypesLocal[i] = ff_->getAtomType(idents[i]);
291 +
292 +    groupList_.clear();
293 +    groupList_.resize(nGroups_);
294 +    for (int i = 0; i < nGroups_; i++) {
295 +      int gid = cgLocalToGlobal[i];
296 +      for (int j = 0; j < nLocal_; j++) {
297 +        int aid = AtomLocalToGlobal[j];
298 +        if (globalGroupMembership[aid] == gid) {
299 +          groupList_[i].push_back(j);
300 +        }
301 +      }      
302 +    }
303 +
304 +
305 +    createGtypeCutoffMap();
306 +
307    }
308 +  
309 +  void ForceMatrixDecomposition::createGtypeCutoffMap() {
310      
311 +    RealType tol = 1e-6;
312 +    largestRcut_ = 0.0;
313 +    int atid;
314 +    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
315 +    
316 +    map<int, RealType> atypeCutoff;
317 +      
318 +    for (set<AtomType*>::iterator at = atypes.begin();
319 +         at != atypes.end(); ++at){
320 +      atid = (*at)->getIdent();
321 +      if (userChoseCutoff_)
322 +        atypeCutoff[atid] = userCutoff_;
323 +      else
324 +        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
325 +    }
326 +    
327 +    vector<RealType> gTypeCutoffs;
328 +    // first we do a single loop over the cutoff groups to find the
329 +    // largest cutoff for any atypes present in this group.
330 + #ifdef IS_MPI
331 +    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
332 +    groupRowToGtype.resize(nGroupsInRow_);
333 +    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
334 +      vector<int> atomListRow = getAtomsInGroupRow(cg1);
335 +      for (vector<int>::iterator ia = atomListRow.begin();
336 +           ia != atomListRow.end(); ++ia) {            
337 +        int atom1 = (*ia);
338 +        atid = identsRow[atom1];
339 +        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
340 +          groupCutoffRow[cg1] = atypeCutoff[atid];
341 +        }
342 +      }
343  
344 +      bool gTypeFound = false;
345 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
346 +        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
347 +          groupRowToGtype[cg1] = gt;
348 +          gTypeFound = true;
349 +        }
350 +      }
351 +      if (!gTypeFound) {
352 +        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
353 +        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
354 +      }
355 +      
356 +    }
357 +    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
358 +    groupColToGtype.resize(nGroupsInCol_);
359 +    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
360 +      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
361 +      for (vector<int>::iterator jb = atomListCol.begin();
362 +           jb != atomListCol.end(); ++jb) {            
363 +        int atom2 = (*jb);
364 +        atid = identsCol[atom2];
365 +        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
366 +          groupCutoffCol[cg2] = atypeCutoff[atid];
367 +        }
368 +      }
369 +      bool gTypeFound = false;
370 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
371 +        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
372 +          groupColToGtype[cg2] = gt;
373 +          gTypeFound = true;
374 +        }
375 +      }
376 +      if (!gTypeFound) {
377 +        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
378 +        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
379 +      }
380 +    }
381 + #else
382  
383 <  void ForceDecomposition::distributeData()  {
383 >    vector<RealType> groupCutoff(nGroups_, 0.0);
384 >    groupToGtype.resize(nGroups_);
385 >    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
386 >      groupCutoff[cg1] = 0.0;
387 >      vector<int> atomList = getAtomsInGroupRow(cg1);
388 >      for (vector<int>::iterator ia = atomList.begin();
389 >           ia != atomList.end(); ++ia) {            
390 >        int atom1 = (*ia);
391 >        atid = idents[atom1];
392 >        if (atypeCutoff[atid] > groupCutoff[cg1])
393 >          groupCutoff[cg1] = atypeCutoff[atid];
394 >      }
395 >      
396 >      bool gTypeFound = false;
397 >      for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
398 >        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
399 >          groupToGtype[cg1] = gt;
400 >          gTypeFound = true;
401 >        }
402 >      }
403 >      if (!gTypeFound) {      
404 >        gTypeCutoffs.push_back( groupCutoff[cg1] );
405 >        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
406 >      }      
407 >    }
408 > #endif
409 >
410 >    // Now we find the maximum group cutoff value present in the simulation
411 >
412 >    RealType groupMax = *max_element(gTypeCutoffs.begin(),
413 >                                     gTypeCutoffs.end());
414 >
415   #ifdef IS_MPI
416 <    Snapshot* snap = sman_->getCurrentSnapshot();
416 >    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
417 >                              MPI::MAX);
418 > #endif
419 >    
420 >    RealType tradRcut = groupMax;
421  
422 +    for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) {
423 +      for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) {      
424 +        RealType thisRcut;
425 +        switch(cutoffPolicy_) {
426 +        case TRADITIONAL:
427 +          thisRcut = tradRcut;
428 +          break;
429 +        case MIX:
430 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
431 +          break;
432 +        case MAX:
433 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
434 +          break;
435 +        default:
436 +          sprintf(painCave.errMsg,
437 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
438 +                  "hit an unknown cutoff policy!\n");
439 +          painCave.severity = OPENMD_ERROR;
440 +          painCave.isFatal = 1;
441 +          simError();
442 +          break;
443 +        }
444 +
445 +        pair<int,int> key = make_pair(i,j);
446 +        gTypeCutoffMap[key].first = thisRcut;
447 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
448 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
449 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
450 +        // sanity check
451 +        
452 +        if (userChoseCutoff_) {
453 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
454 +            sprintf(painCave.errMsg,
455 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
456 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
457 +            painCave.severity = OPENMD_ERROR;
458 +            painCave.isFatal = 1;
459 +            simError();            
460 +          }
461 +        }
462 +      }
463 +    }
464 +  }
465 +
466 +  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
467 +    int i, j;  
468 + #ifdef IS_MPI
469 +    i = groupRowToGtype[cg1];
470 +    j = groupColToGtype[cg2];
471 + #else
472 +    i = groupToGtype[cg1];
473 +    j = groupToGtype[cg2];
474 + #endif    
475 +    return gTypeCutoffMap[make_pair(i,j)];
476 +  }
477 +
478 +  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
479 +    for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
480 +      if (toposForAtom[atom1][j] == atom2)
481 +        return topoDist[atom1][j];
482 +    }                                          
483 +    return 0;
484 +  }
485 +
486 +  void ForceMatrixDecomposition::zeroWorkArrays() {
487 +    pairwisePot = 0.0;
488 +    embeddingPot = 0.0;
489 +    excludedPot = 0.0;
490 +    excludedSelfPot = 0.0;
491 +
492 + #ifdef IS_MPI
493 +    if (storageLayout_ & DataStorage::dslForce) {
494 +      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
495 +      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
496 +    }
497 +
498 +    if (storageLayout_ & DataStorage::dslTorque) {
499 +      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
500 +      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
501 +    }
502 +    
503 +    fill(pot_row.begin(), pot_row.end(),
504 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
505 +
506 +    fill(pot_col.begin(), pot_col.end(),
507 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
508 +
509 +    fill(expot_row.begin(), expot_row.end(),
510 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
511 +
512 +    fill(expot_col.begin(), expot_col.end(),
513 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
514 +
515 +    if (storageLayout_ & DataStorage::dslParticlePot) {    
516 +      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
517 +           0.0);
518 +      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
519 +           0.0);
520 +    }
521 +
522 +    if (storageLayout_ & DataStorage::dslDensity) {      
523 +      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
524 +      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
525 +    }
526 +
527 +    if (storageLayout_ & DataStorage::dslFunctional) {  
528 +      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
529 +           0.0);
530 +      fill(atomColData.functional.begin(), atomColData.functional.end(),
531 +           0.0);
532 +    }
533 +
534 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
535 +      fill(atomRowData.functionalDerivative.begin(),
536 +           atomRowData.functionalDerivative.end(), 0.0);
537 +      fill(atomColData.functionalDerivative.begin(),
538 +           atomColData.functionalDerivative.end(), 0.0);
539 +    }
540 +
541 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
542 +      fill(atomRowData.skippedCharge.begin(),
543 +           atomRowData.skippedCharge.end(), 0.0);
544 +      fill(atomColData.skippedCharge.begin(),
545 +           atomColData.skippedCharge.end(), 0.0);
546 +    }
547 +
548 +    if (storageLayout_ & DataStorage::dslFlucQForce) {      
549 +      fill(atomRowData.flucQFrc.begin(),
550 +           atomRowData.flucQFrc.end(), 0.0);
551 +      fill(atomColData.flucQFrc.begin(),
552 +           atomColData.flucQFrc.end(), 0.0);
553 +    }
554 +
555 +    if (storageLayout_ & DataStorage::dslElectricField) {    
556 +      fill(atomRowData.electricField.begin(),
557 +           atomRowData.electricField.end(), V3Zero);
558 +      fill(atomColData.electricField.begin(),
559 +           atomColData.electricField.end(), V3Zero);
560 +    }
561 +
562 + #endif
563 +    // even in parallel, we need to zero out the local arrays:
564 +
565 +    if (storageLayout_ & DataStorage::dslParticlePot) {      
566 +      fill(snap_->atomData.particlePot.begin(),
567 +           snap_->atomData.particlePot.end(), 0.0);
568 +    }
569 +    
570 +    if (storageLayout_ & DataStorage::dslDensity) {      
571 +      fill(snap_->atomData.density.begin(),
572 +           snap_->atomData.density.end(), 0.0);
573 +    }
574 +
575 +    if (storageLayout_ & DataStorage::dslFunctional) {
576 +      fill(snap_->atomData.functional.begin(),
577 +           snap_->atomData.functional.end(), 0.0);
578 +    }
579 +
580 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
581 +      fill(snap_->atomData.functionalDerivative.begin(),
582 +           snap_->atomData.functionalDerivative.end(), 0.0);
583 +    }
584 +
585 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
586 +      fill(snap_->atomData.skippedCharge.begin(),
587 +           snap_->atomData.skippedCharge.end(), 0.0);
588 +    }
589 +
590 +    if (storageLayout_ & DataStorage::dslElectricField) {      
591 +      fill(snap_->atomData.electricField.begin(),
592 +           snap_->atomData.electricField.end(), V3Zero);
593 +    }
594 +  }
595 +
596 +
597 +  void ForceMatrixDecomposition::distributeData()  {
598 +    snap_ = sman_->getCurrentSnapshot();
599 +    storageLayout_ = sman_->getStorageLayout();
600 + #ifdef IS_MPI
601 +    
602      // gather up the atomic positions
603 <    AtomCommVectorI->gather(snap->atomData.position,
604 <                            snap->atomIData.position);
605 <    AtomCommVectorJ->gather(snap->atomData.position,
606 <                           snap->atomJData.position);
603 >    AtomPlanVectorRow->gather(snap_->atomData.position,
604 >                              atomRowData.position);
605 >    AtomPlanVectorColumn->gather(snap_->atomData.position,
606 >                                 atomColData.position);
607      
608      // gather up the cutoff group positions
609 <    cgCommVectorI->gather(snap->cgData.position,
610 <                         snap->cgIData.position);
611 <    cgCommVectorJ->gather(snap->cgData.position,
612 <                         snap->cgJData.position);
609 >
610 >    cgPlanVectorRow->gather(snap_->cgData.position,
611 >                            cgRowData.position);
612 >
613 >    cgPlanVectorColumn->gather(snap_->cgData.position,
614 >                               cgColData.position);
615 >
616 >
617 >
618 >    if (needVelocities_) {
619 >      // gather up the atomic velocities
620 >      AtomPlanVectorColumn->gather(snap_->atomData.velocity,
621 >                                   atomColData.velocity);
622 >      
623 >      cgPlanVectorColumn->gather(snap_->cgData.velocity,
624 >                                 cgColData.velocity);
625 >    }
626 >
627      
628      // if needed, gather the atomic rotation matrices
629 <    if (snap->atomData.getStorageLayout() & DataStorage::dslAmat) {
630 <      AtomCommMatrixI->gather(snap->atomData.aMat,
631 <                             snap->atomIData.aMat);
632 <      AtomCommMatrixJ->gather(snap->atomData.aMat,
633 <                             snap->atomJData.aMat);
629 >    if (storageLayout_ & DataStorage::dslAmat) {
630 >      AtomPlanMatrixRow->gather(snap_->atomData.aMat,
631 >                                atomRowData.aMat);
632 >      AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
633 >                                   atomColData.aMat);
634      }
635 <    
636 <    // if needed, gather the atomic eletrostatic frames
637 <    if (snap->atomData.getStorageLayout() & DataStorage::dslElectroFrame) {
638 <      AtomCommMatrixI->gather(snap->atomData.electroFrame,
639 <                             snap->atomIData.electroFrame);
640 <      AtomCommMatrixJ->gather(snap->atomData.electroFrame,
641 <                             snap->atomJData.electroFrame);
635 >
636 >    // if needed, gather the atomic eletrostatic information
637 >    if (storageLayout_ & DataStorage::dslDipole) {
638 >      AtomPlanVectorRow->gather(snap_->atomData.dipole,
639 >                                atomRowData.dipole);
640 >      AtomPlanVectorColumn->gather(snap_->atomData.dipole,
641 >                                   atomColData.dipole);
642      }
643 +
644 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
645 +      AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
646 +                                atomRowData.quadrupole);
647 +      AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
648 +                                   atomColData.quadrupole);
649 +    }
650 +        
651 +    // if needed, gather the atomic fluctuating charge values
652 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {
653 +      AtomPlanRealRow->gather(snap_->atomData.flucQPos,
654 +                              atomRowData.flucQPos);
655 +      AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
656 +                                 atomColData.flucQPos);
657 +    }
658 +
659   #endif      
660    }
661    
662 <  void ForceDecomposition::collectIntermediateData() {
662 >  /* collects information obtained during the pre-pair loop onto local
663 >   * data structures.
664 >   */
665 >  void ForceMatrixDecomposition::collectIntermediateData() {
666 >    snap_ = sman_->getCurrentSnapshot();
667 >    storageLayout_ = sman_->getStorageLayout();
668   #ifdef IS_MPI
105    Snapshot* snap = sman_->getCurrentSnapshot();
106    // gather up the atomic positions
669      
670 <    if (snap->atomData.getStorageLayout() & DataStorage::dslDensity) {
671 <      AtomCommRealI->scatter(snap->atomIData.density,
672 <                            snap->atomData.density);
673 <      std::vector<RealType> rho_tmp;
674 <      int n = snap->getNumberOfAtoms();
675 <      rho_tmp.reserve( n );
676 <      AtomCommRealJ->scatter(snap->atomJData.density, rho_tmp);
670 >    if (storageLayout_ & DataStorage::dslDensity) {
671 >      
672 >      AtomPlanRealRow->scatter(atomRowData.density,
673 >                               snap_->atomData.density);
674 >      
675 >      int n = snap_->atomData.density.size();
676 >      vector<RealType> rho_tmp(n, 0.0);
677 >      AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
678        for (int i = 0; i < n; i++)
679 <        snap->atomData.density[i] += rho_tmp[i];
679 >        snap_->atomData.density[i] += rho_tmp[i];
680      }
681 +
682 +    // this isn't necessary if we don't have polarizable atoms, but
683 +    // we'll leave it here for now.
684 +    if (storageLayout_ & DataStorage::dslElectricField) {
685 +      
686 +      AtomPlanVectorRow->scatter(atomRowData.electricField,
687 +                                 snap_->atomData.electricField);
688 +      
689 +      int n = snap_->atomData.electricField.size();
690 +      vector<Vector3d> field_tmp(n, V3Zero);
691 +      AtomPlanVectorColumn->scatter(atomColData.electricField,
692 +                                    field_tmp);
693 +      for (int i = 0; i < n; i++)
694 +        snap_->atomData.electricField[i] += field_tmp[i];
695 +    }
696   #endif
697    }
698 <  
699 <  void ForceDecomposition::distributeIntermediateData() {
698 >
699 >  /*
700 >   * redistributes information obtained during the pre-pair loop out to
701 >   * row and column-indexed data structures
702 >   */
703 >  void ForceMatrixDecomposition::distributeIntermediateData() {
704 >    snap_ = sman_->getCurrentSnapshot();
705 >    storageLayout_ = sman_->getStorageLayout();
706   #ifdef IS_MPI
707 <    Snapshot* snap = sman_->getCurrentSnapshot();
708 <    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctional) {
709 <      AtomCommRealI->gather(snap->atomData.functional,
710 <                           snap->atomIData.functional);
711 <      AtomCommRealJ->gather(snap->atomData.functional,
128 <                           snap->atomJData.functional);
707 >    if (storageLayout_ & DataStorage::dslFunctional) {
708 >      AtomPlanRealRow->gather(snap_->atomData.functional,
709 >                              atomRowData.functional);
710 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
711 >                                 atomColData.functional);
712      }
713      
714 <    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctionalDerivative) {
715 <      AtomCommRealI->gather(snap->atomData.functionalDerivative,
716 <                           snap->atomIData.functionalDerivative);
717 <      AtomCommRealJ->gather(snap->atomData.functionalDerivative,
718 <                           snap->atomJData.functionalDerivative);
714 >    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
715 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
716 >                              atomRowData.functionalDerivative);
717 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
718 >                                 atomColData.functionalDerivative);
719      }
720   #endif
721    }
722    
723    
724 <  void ForceDecomposition::collectData() {
724 >  void ForceMatrixDecomposition::collectData() {
725 >    snap_ = sman_->getCurrentSnapshot();
726 >    storageLayout_ = sman_->getStorageLayout();
727 > #ifdef IS_MPI    
728 >    int n = snap_->atomData.force.size();
729 >    vector<Vector3d> frc_tmp(n, V3Zero);
730 >    
731 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
732 >    for (int i = 0; i < n; i++) {
733 >      snap_->atomData.force[i] += frc_tmp[i];
734 >      frc_tmp[i] = 0.0;
735 >    }
736 >    
737 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
738 >    for (int i = 0; i < n; i++) {
739 >      snap_->atomData.force[i] += frc_tmp[i];
740 >    }
741 >        
742 >    if (storageLayout_ & DataStorage::dslTorque) {
743 >
744 >      int nt = snap_->atomData.torque.size();
745 >      vector<Vector3d> trq_tmp(nt, V3Zero);
746 >
747 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
748 >      for (int i = 0; i < nt; i++) {
749 >        snap_->atomData.torque[i] += trq_tmp[i];
750 >        trq_tmp[i] = 0.0;
751 >      }
752 >      
753 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
754 >      for (int i = 0; i < nt; i++)
755 >        snap_->atomData.torque[i] += trq_tmp[i];
756 >    }
757 >
758 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {
759 >
760 >      int ns = snap_->atomData.skippedCharge.size();
761 >      vector<RealType> skch_tmp(ns, 0.0);
762 >
763 >      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
764 >      for (int i = 0; i < ns; i++) {
765 >        snap_->atomData.skippedCharge[i] += skch_tmp[i];
766 >        skch_tmp[i] = 0.0;
767 >      }
768 >      
769 >      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
770 >      for (int i = 0; i < ns; i++)
771 >        snap_->atomData.skippedCharge[i] += skch_tmp[i];
772 >            
773 >    }
774 >    
775 >    if (storageLayout_ & DataStorage::dslFlucQForce) {
776 >
777 >      int nq = snap_->atomData.flucQFrc.size();
778 >      vector<RealType> fqfrc_tmp(nq, 0.0);
779 >
780 >      AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
781 >      for (int i = 0; i < nq; i++) {
782 >        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
783 >        fqfrc_tmp[i] = 0.0;
784 >      }
785 >      
786 >      AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
787 >      for (int i = 0; i < nq; i++)
788 >        snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
789 >            
790 >    }
791 >
792 >    if (storageLayout_ & DataStorage::dslElectricField) {
793 >
794 >      int nef = snap_->atomData.electricField.size();
795 >      vector<Vector3d> efield_tmp(nef, V3Zero);
796 >
797 >      AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
798 >      for (int i = 0; i < nef; i++) {
799 >        snap_->atomData.electricField[i] += efield_tmp[i];
800 >        efield_tmp[i] = 0.0;
801 >      }
802 >      
803 >      AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
804 >      for (int i = 0; i < nef; i++)
805 >        snap_->atomData.electricField[i] += efield_tmp[i];
806 >    }
807 >
808 >
809 >    nLocal_ = snap_->getNumberOfAtoms();
810 >
811 >    vector<potVec> pot_temp(nLocal_,
812 >                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
813 >    vector<potVec> expot_temp(nLocal_,
814 >                              Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
815 >
816 >    // scatter/gather pot_row into the members of my column
817 >          
818 >    AtomPlanPotRow->scatter(pot_row, pot_temp);
819 >    AtomPlanPotRow->scatter(expot_row, expot_temp);
820 >
821 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
822 >      pairwisePot += pot_temp[ii];
823 >
824 >    for (int ii = 0;  ii < expot_temp.size(); ii++ )
825 >      excludedPot += expot_temp[ii];
826 >        
827 >    if (storageLayout_ & DataStorage::dslParticlePot) {
828 >      // This is the pairwise contribution to the particle pot.  The
829 >      // embedding contribution is added in each of the low level
830 >      // non-bonded routines.  In single processor, this is done in
831 >      // unpackInteractionData, not in collectData.
832 >      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
833 >        for (int i = 0; i < nLocal_; i++) {
834 >          // factor of two is because the total potential terms are divided
835 >          // by 2 in parallel due to row/ column scatter      
836 >          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
837 >        }
838 >      }
839 >    }
840 >
841 >    fill(pot_temp.begin(), pot_temp.end(),
842 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
843 >    fill(expot_temp.begin(), expot_temp.end(),
844 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
845 >      
846 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
847 >    AtomPlanPotColumn->scatter(expot_col, expot_temp);    
848 >    
849 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
850 >      pairwisePot += pot_temp[ii];    
851 >
852 >    for (int ii = 0;  ii < expot_temp.size(); ii++ )
853 >      excludedPot += expot_temp[ii];    
854 >
855 >    if (storageLayout_ & DataStorage::dslParticlePot) {
856 >      // This is the pairwise contribution to the particle pot.  The
857 >      // embedding contribution is added in each of the low level
858 >      // non-bonded routines.  In single processor, this is done in
859 >      // unpackInteractionData, not in collectData.
860 >      for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
861 >        for (int i = 0; i < nLocal_; i++) {
862 >          // factor of two is because the total potential terms are divided
863 >          // by 2 in parallel due to row/ column scatter      
864 >          snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
865 >        }
866 >      }
867 >    }
868 >    
869 >    if (storageLayout_ & DataStorage::dslParticlePot) {
870 >      int npp = snap_->atomData.particlePot.size();
871 >      vector<RealType> ppot_temp(npp, 0.0);
872 >
873 >      // This is the direct or embedding contribution to the particle
874 >      // pot.
875 >      
876 >      AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
877 >      for (int i = 0; i < npp; i++) {
878 >        snap_->atomData.particlePot[i] += ppot_temp[i];
879 >      }
880 >
881 >      fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
882 >      
883 >      AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
884 >      for (int i = 0; i < npp; i++) {
885 >        snap_->atomData.particlePot[i] += ppot_temp[i];
886 >      }
887 >    }
888 >
889 >    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
890 >      RealType ploc1 = pairwisePot[ii];
891 >      RealType ploc2 = 0.0;
892 >      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
893 >      pairwisePot[ii] = ploc2;
894 >    }
895 >
896 >    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
897 >      RealType ploc1 = excludedPot[ii];
898 >      RealType ploc2 = 0.0;
899 >      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
900 >      excludedPot[ii] = ploc2;
901 >    }
902 >
903 >    // Here be dragons.
904 >    MPI::Intracomm col = colComm.getComm();
905 >
906 >    col.Allreduce(MPI::IN_PLACE,
907 >                  &snap_->frameData.conductiveHeatFlux[0], 3,
908 >                  MPI::REALTYPE, MPI::SUM);
909 >
910 >
911 > #endif
912 >
913 >  }
914 >
915 >  /**
916 >   * Collects information obtained during the post-pair (and embedding
917 >   * functional) loops onto local data structures.
918 >   */
919 >  void ForceMatrixDecomposition::collectSelfData() {
920 >    snap_ = sman_->getCurrentSnapshot();
921 >    storageLayout_ = sman_->getStorageLayout();
922 >
923   #ifdef IS_MPI
924 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
925 +      RealType ploc1 = embeddingPot[ii];
926 +      RealType ploc2 = 0.0;
927 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
928 +      embeddingPot[ii] = ploc2;
929 +    }    
930 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
931 +      RealType ploc1 = excludedSelfPot[ii];
932 +      RealType ploc2 = 0.0;
933 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
934 +      excludedSelfPot[ii] = ploc2;
935 +    }    
936   #endif
937 +    
938    }
939 +
940 +
941 +
942 +  int& ForceMatrixDecomposition::getNAtomsInRow() {  
943 + #ifdef IS_MPI
944 +    return nAtomsInRow_;
945 + #else
946 +    return nLocal_;
947 + #endif
948 +  }
949 +
950 +  /**
951 +   * returns the list of atoms belonging to this group.  
952 +   */
953 +  vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
954 + #ifdef IS_MPI
955 +    return groupListRow_[cg1];
956 + #else
957 +    return groupList_[cg1];
958 + #endif
959 +  }
960 +
961 +  vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
962 + #ifdef IS_MPI
963 +    return groupListCol_[cg2];
964 + #else
965 +    return groupList_[cg2];
966 + #endif
967 +  }
968    
969 +  Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
970 +    Vector3d d;
971 +    
972 + #ifdef IS_MPI
973 +    d = cgColData.position[cg2] - cgRowData.position[cg1];
974 + #else
975 +    d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
976 + #endif
977 +    
978 +    if (usePeriodicBoundaryConditions_) {
979 +      snap_->wrapVector(d);
980 +    }
981 +    return d;    
982 +  }
983 +
984 +  Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
985 + #ifdef IS_MPI
986 +    return cgColData.velocity[cg2];
987 + #else
988 +    return snap_->cgData.velocity[cg2];
989 + #endif
990 +  }
991 +
992 +  Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
993 + #ifdef IS_MPI
994 +    return atomColData.velocity[atom2];
995 + #else
996 +    return snap_->atomData.velocity[atom2];
997 + #endif
998 +  }
999 +
1000 +
1001 +  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
1002 +
1003 +    Vector3d d;
1004 +    
1005 + #ifdef IS_MPI
1006 +    d = cgRowData.position[cg1] - atomRowData.position[atom1];
1007 + #else
1008 +    d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
1009 + #endif
1010 +    if (usePeriodicBoundaryConditions_) {
1011 +      snap_->wrapVector(d);
1012 +    }
1013 +    return d;    
1014 +  }
1015 +  
1016 +  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
1017 +    Vector3d d;
1018 +    
1019 + #ifdef IS_MPI
1020 +    d = cgColData.position[cg2] - atomColData.position[atom2];
1021 + #else
1022 +    d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1023 + #endif
1024 +    if (usePeriodicBoundaryConditions_) {
1025 +      snap_->wrapVector(d);
1026 +    }
1027 +    return d;    
1028 +  }
1029 +
1030 +  RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1031 + #ifdef IS_MPI
1032 +    return massFactorsRow[atom1];
1033 + #else
1034 +    return massFactors[atom1];
1035 + #endif
1036 +  }
1037 +
1038 +  RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1039 + #ifdef IS_MPI
1040 +    return massFactorsCol[atom2];
1041 + #else
1042 +    return massFactors[atom2];
1043 + #endif
1044 +
1045 +  }
1046 +    
1047 +  Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
1048 +    Vector3d d;
1049 +    
1050 + #ifdef IS_MPI
1051 +    d = atomColData.position[atom2] - atomRowData.position[atom1];
1052 + #else
1053 +    d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1054 + #endif
1055 +    if (usePeriodicBoundaryConditions_) {
1056 +      snap_->wrapVector(d);
1057 +    }
1058 +    return d;    
1059 +  }
1060 +
1061 +  vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1062 +    return excludesForAtom[atom1];
1063 +  }
1064 +
1065 +  /**
1066 +   * We need to exclude some overcounted interactions that result from
1067 +   * the parallel decomposition.
1068 +   */
1069 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1070 +    int unique_id_1, unique_id_2;
1071 +        
1072 + #ifdef IS_MPI
1073 +    // in MPI, we have to look up the unique IDs for each atom
1074 +    unique_id_1 = AtomRowToGlobal[atom1];
1075 +    unique_id_2 = AtomColToGlobal[atom2];
1076 +    // group1 = cgRowToGlobal[cg1];
1077 +    // group2 = cgColToGlobal[cg2];
1078 + #else
1079 +    unique_id_1 = AtomLocalToGlobal[atom1];
1080 +    unique_id_2 = AtomLocalToGlobal[atom2];
1081 +    int group1 = cgLocalToGlobal[cg1];
1082 +    int group2 = cgLocalToGlobal[cg2];
1083 + #endif  
1084 +
1085 +    if (unique_id_1 == unique_id_2) return true;
1086 +
1087 + #ifdef IS_MPI
1088 +    // this prevents us from doing the pair on multiple processors
1089 +    if (unique_id_1 < unique_id_2) {
1090 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1091 +    } else {
1092 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1093 +    }
1094 + #endif    
1095 +
1096 + #ifndef IS_MPI
1097 +    if (group1 == group2) {
1098 +      if (unique_id_1 < unique_id_2) return true;
1099 +    }
1100 + #endif
1101 +    
1102 +    return false;
1103 +  }
1104 +
1105 +  /**
1106 +   * We need to handle the interactions for atoms who are involved in
1107 +   * the same rigid body as well as some short range interactions
1108 +   * (bonds, bends, torsions) differently from other interactions.
1109 +   * We'll still visit the pairwise routines, but with a flag that
1110 +   * tells those routines to exclude the pair from direct long range
1111 +   * interactions.  Some indirect interactions (notably reaction
1112 +   * field) must still be handled for these pairs.
1113 +   */
1114 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1115 +
1116 +    // excludesForAtom was constructed to use row/column indices in the MPI
1117 +    // version, and to use local IDs in the non-MPI version:
1118 +    
1119 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1120 +         i != excludesForAtom[atom1].end(); ++i) {
1121 +      if ( (*i) == atom2 ) return true;
1122 +    }
1123 +
1124 +    return false;
1125 +  }
1126 +
1127 +
1128 +  void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
1129 + #ifdef IS_MPI
1130 +    atomRowData.force[atom1] += fg;
1131 + #else
1132 +    snap_->atomData.force[atom1] += fg;
1133 + #endif
1134 +  }
1135 +
1136 +  void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
1137 + #ifdef IS_MPI
1138 +    atomColData.force[atom2] += fg;
1139 + #else
1140 +    snap_->atomData.force[atom2] += fg;
1141 + #endif
1142 +  }
1143 +
1144 +    // filling interaction blocks with pointers
1145 +  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1146 +                                                     int atom1, int atom2) {
1147 +
1148 +    idat.excluded = excludeAtomPair(atom1, atom2);
1149 +  
1150 + #ifdef IS_MPI
1151 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1152 +    idat.atid1 = identsRow[atom1];
1153 +    idat.atid2 = identsCol[atom2];
1154 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1155 +    //                         ff_->getAtomType(identsCol[atom2]) );
1156 +    
1157 +    if (storageLayout_ & DataStorage::dslAmat) {
1158 +      idat.A1 = &(atomRowData.aMat[atom1]);
1159 +      idat.A2 = &(atomColData.aMat[atom2]);
1160 +    }
1161 +    
1162 +    if (storageLayout_ & DataStorage::dslTorque) {
1163 +      idat.t1 = &(atomRowData.torque[atom1]);
1164 +      idat.t2 = &(atomColData.torque[atom2]);
1165 +    }
1166 +
1167 +    if (storageLayout_ & DataStorage::dslDipole) {
1168 +      idat.dipole1 = &(atomRowData.dipole[atom1]);
1169 +      idat.dipole2 = &(atomColData.dipole[atom2]);
1170 +    }
1171 +
1172 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1173 +      idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1174 +      idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1175 +    }
1176 +
1177 +    if (storageLayout_ & DataStorage::dslDensity) {
1178 +      idat.rho1 = &(atomRowData.density[atom1]);
1179 +      idat.rho2 = &(atomColData.density[atom2]);
1180 +    }
1181 +
1182 +    if (storageLayout_ & DataStorage::dslFunctional) {
1183 +      idat.frho1 = &(atomRowData.functional[atom1]);
1184 +      idat.frho2 = &(atomColData.functional[atom2]);
1185 +    }
1186 +
1187 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1188 +      idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1189 +      idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1190 +    }
1191 +
1192 +    if (storageLayout_ & DataStorage::dslParticlePot) {
1193 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1194 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
1195 +    }
1196 +
1197 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
1198 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1199 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1200 +    }
1201 +
1202 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1203 +      idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1204 +      idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1205 +    }
1206 +
1207 + #else
1208 +    
1209 +    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1210 +    idat.atid1 = idents[atom1];
1211 +    idat.atid2 = idents[atom2];
1212 +
1213 +    if (storageLayout_ & DataStorage::dslAmat) {
1214 +      idat.A1 = &(snap_->atomData.aMat[atom1]);
1215 +      idat.A2 = &(snap_->atomData.aMat[atom2]);
1216 +    }
1217 +
1218 +    if (storageLayout_ & DataStorage::dslTorque) {
1219 +      idat.t1 = &(snap_->atomData.torque[atom1]);
1220 +      idat.t2 = &(snap_->atomData.torque[atom2]);
1221 +    }
1222 +
1223 +    if (storageLayout_ & DataStorage::dslDipole) {
1224 +      idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1225 +      idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1226 +    }
1227 +
1228 +    if (storageLayout_ & DataStorage::dslQuadrupole) {
1229 +      idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1230 +      idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1231 +    }
1232 +
1233 +    if (storageLayout_ & DataStorage::dslDensity) {    
1234 +      idat.rho1 = &(snap_->atomData.density[atom1]);
1235 +      idat.rho2 = &(snap_->atomData.density[atom2]);
1236 +    }
1237 +
1238 +    if (storageLayout_ & DataStorage::dslFunctional) {
1239 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
1240 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
1241 +    }
1242 +
1243 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1244 +      idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1245 +      idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1246 +    }
1247 +
1248 +    if (storageLayout_ & DataStorage::dslParticlePot) {
1249 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1250 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1251 +    }
1252 +
1253 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
1254 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1255 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1256 +    }
1257 +
1258 +    if (storageLayout_ & DataStorage::dslFlucQPosition) {              
1259 +      idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1260 +      idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1261 +    }
1262 +
1263 + #endif
1264 +  }
1265 +
1266 +  
1267 +  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1268 + #ifdef IS_MPI
1269 +    pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1270 +    pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1271 +    expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1272 +    expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1273 +
1274 +    atomRowData.force[atom1] += *(idat.f1);
1275 +    atomColData.force[atom2] -= *(idat.f1);
1276 +
1277 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1278 +      atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1279 +      atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1280 +    }
1281 +
1282 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1283 +      atomRowData.electricField[atom1] += *(idat.eField1);
1284 +      atomColData.electricField[atom2] += *(idat.eField2);
1285 +    }
1286 +
1287 + #else
1288 +    pairwisePot += *(idat.pot);
1289 +    excludedPot += *(idat.excludedPot);
1290 +
1291 +    snap_->atomData.force[atom1] += *(idat.f1);
1292 +    snap_->atomData.force[atom2] -= *(idat.f1);
1293 +
1294 +    if (idat.doParticlePot) {
1295 +      // This is the pairwise contribution to the particle pot.  The
1296 +      // embedding contribution is added in each of the low level
1297 +      // non-bonded routines.  In parallel, this calculation is done
1298 +      // in collectData, not in unpackInteractionData.
1299 +      snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1300 +      snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1301 +    }
1302 +    
1303 +    if (storageLayout_ & DataStorage::dslFlucQForce) {              
1304 +      snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1305 +      snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1306 +    }
1307 +
1308 +    if (storageLayout_ & DataStorage::dslElectricField) {              
1309 +      snap_->atomData.electricField[atom1] += *(idat.eField1);
1310 +      snap_->atomData.electricField[atom2] += *(idat.eField2);
1311 +    }
1312 +
1313 + #endif
1314 +    
1315 +  }
1316 +
1317 +  /*
1318 +   * buildNeighborList
1319 +   *
1320 +   * first element of pair is row-indexed CutoffGroup
1321 +   * second element of pair is column-indexed CutoffGroup
1322 +   */
1323 +  void ForceMatrixDecomposition::buildNeighborList(vector<pair<int,int> >& neighborList) {
1324 +    
1325 +    neighborList.clear();
1326 +    groupCutoffs cuts;
1327 +    bool doAllPairs = false;
1328 +
1329 +    RealType rList_ = (largestRcut_ + skinThickness_);
1330 +    Snapshot* snap_ = sman_->getCurrentSnapshot();
1331 +    Mat3x3d box;
1332 +    Mat3x3d invBox;
1333 +
1334 +    Vector3d rs, scaled, dr;
1335 +    Vector3i whichCell;
1336 +    int cellIndex;
1337 +
1338 + #ifdef IS_MPI
1339 +    cellListRow_.clear();
1340 +    cellListCol_.clear();
1341 + #else
1342 +    cellList_.clear();
1343 + #endif
1344 +    
1345 +    if (!usePeriodicBoundaryConditions_) {
1346 +      box = snap_->getBoundingBox();
1347 +      invBox = snap_->getInvBoundingBox();
1348 +    } else {
1349 +      box = snap_->getHmat();
1350 +      invBox = snap_->getInvHmat();
1351 +    }
1352 +    
1353 +    Vector3d boxX = box.getColumn(0);
1354 +    Vector3d boxY = box.getColumn(1);
1355 +    Vector3d boxZ = box.getColumn(2);
1356 +    
1357 +    nCells_.x() = (int) ( boxX.length() )/ rList_;
1358 +    nCells_.y() = (int) ( boxY.length() )/ rList_;
1359 +    nCells_.z() = (int) ( boxZ.length() )/ rList_;
1360 +    
1361 +    // handle small boxes where the cell offsets can end up repeating cells
1362 +    
1363 +    if (nCells_.x() < 3) doAllPairs = true;
1364 +    if (nCells_.y() < 3) doAllPairs = true;
1365 +    if (nCells_.z() < 3) doAllPairs = true;
1366 +    
1367 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1368 +    
1369 + #ifdef IS_MPI
1370 +    cellListRow_.resize(nCtot);
1371 +    cellListCol_.resize(nCtot);
1372 + #else
1373 +    cellList_.resize(nCtot);
1374 + #endif
1375 +    
1376 +    if (!doAllPairs) {
1377 + #ifdef IS_MPI
1378 +      
1379 +      for (int i = 0; i < nGroupsInRow_; i++) {
1380 +        rs = cgRowData.position[i];
1381 +        
1382 +        // scaled positions relative to the box vectors
1383 +        scaled = invBox * rs;
1384 +        
1385 +        // wrap the vector back into the unit box by subtracting integer box
1386 +        // numbers
1387 +        for (int j = 0; j < 3; j++) {
1388 +          scaled[j] -= roundMe(scaled[j]);
1389 +          scaled[j] += 0.5;
1390 +          // Handle the special case when an object is exactly on the
1391 +          // boundary (a scaled coordinate of 1.0 is the same as
1392 +          // scaled coordinate of 0.0)
1393 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1394 +        }
1395 +        
1396 +        // find xyz-indices of cell that cutoffGroup is in.
1397 +        whichCell.x() = nCells_.x() * scaled.x();
1398 +        whichCell.y() = nCells_.y() * scaled.y();
1399 +        whichCell.z() = nCells_.z() * scaled.z();
1400 +        
1401 +        // find single index of this cell:
1402 +        cellIndex = Vlinear(whichCell, nCells_);
1403 +        
1404 +        // add this cutoff group to the list of groups in this cell;
1405 +        cellListRow_[cellIndex].push_back(i);
1406 +      }
1407 +      for (int i = 0; i < nGroupsInCol_; i++) {
1408 +        rs = cgColData.position[i];
1409 +        
1410 +        // scaled positions relative to the box vectors
1411 +        scaled = invBox * rs;
1412 +        
1413 +        // wrap the vector back into the unit box by subtracting integer box
1414 +        // numbers
1415 +        for (int j = 0; j < 3; j++) {
1416 +          scaled[j] -= roundMe(scaled[j]);
1417 +          scaled[j] += 0.5;
1418 +          // Handle the special case when an object is exactly on the
1419 +          // boundary (a scaled coordinate of 1.0 is the same as
1420 +          // scaled coordinate of 0.0)
1421 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1422 +        }
1423 +        
1424 +        // find xyz-indices of cell that cutoffGroup is in.
1425 +        whichCell.x() = nCells_.x() * scaled.x();
1426 +        whichCell.y() = nCells_.y() * scaled.y();
1427 +        whichCell.z() = nCells_.z() * scaled.z();
1428 +        
1429 +        // find single index of this cell:
1430 +        cellIndex = Vlinear(whichCell, nCells_);
1431 +        
1432 +        // add this cutoff group to the list of groups in this cell;
1433 +        cellListCol_[cellIndex].push_back(i);
1434 +      }
1435 +      
1436 + #else
1437 +      for (int i = 0; i < nGroups_; i++) {
1438 +        rs = snap_->cgData.position[i];
1439 +        
1440 +        // scaled positions relative to the box vectors
1441 +        scaled = invBox * rs;
1442 +        
1443 +        // wrap the vector back into the unit box by subtracting integer box
1444 +        // numbers
1445 +        for (int j = 0; j < 3; j++) {
1446 +          scaled[j] -= roundMe(scaled[j]);
1447 +          scaled[j] += 0.5;
1448 +          // Handle the special case when an object is exactly on the
1449 +          // boundary (a scaled coordinate of 1.0 is the same as
1450 +          // scaled coordinate of 0.0)
1451 +          if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1452 +        }
1453 +        
1454 +        // find xyz-indices of cell that cutoffGroup is in.
1455 +        whichCell.x() = nCells_.x() * scaled.x();
1456 +        whichCell.y() = nCells_.y() * scaled.y();
1457 +        whichCell.z() = nCells_.z() * scaled.z();
1458 +        
1459 +        // find single index of this cell:
1460 +        cellIndex = Vlinear(whichCell, nCells_);
1461 +        
1462 +        // add this cutoff group to the list of groups in this cell;
1463 +        cellList_[cellIndex].push_back(i);
1464 +      }
1465 +
1466 + #endif
1467 +
1468 +      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1469 +        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1470 +          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1471 +            Vector3i m1v(m1x, m1y, m1z);
1472 +            int m1 = Vlinear(m1v, nCells_);
1473 +            
1474 +            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1475 +                 os != cellOffsets_.end(); ++os) {
1476 +              
1477 +              Vector3i m2v = m1v + (*os);
1478 +            
1479 +
1480 +              if (m2v.x() >= nCells_.x()) {
1481 +                m2v.x() = 0;          
1482 +              } else if (m2v.x() < 0) {
1483 +                m2v.x() = nCells_.x() - 1;
1484 +              }
1485 +              
1486 +              if (m2v.y() >= nCells_.y()) {
1487 +                m2v.y() = 0;          
1488 +              } else if (m2v.y() < 0) {
1489 +                m2v.y() = nCells_.y() - 1;
1490 +              }
1491 +              
1492 +              if (m2v.z() >= nCells_.z()) {
1493 +                m2v.z() = 0;          
1494 +              } else if (m2v.z() < 0) {
1495 +                m2v.z() = nCells_.z() - 1;
1496 +              }
1497 +
1498 +              int m2 = Vlinear (m2v, nCells_);
1499 +              
1500 + #ifdef IS_MPI
1501 +              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1502 +                   j1 != cellListRow_[m1].end(); ++j1) {
1503 +                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1504 +                     j2 != cellListCol_[m2].end(); ++j2) {
1505 +                  
1506 +                  // In parallel, we need to visit *all* pairs of row
1507 +                  // & column indicies and will divide labor in the
1508 +                  // force evaluation later.
1509 +                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1510 +                  if (usePeriodicBoundaryConditions_) {
1511 +                    snap_->wrapVector(dr);
1512 +                  }
1513 +                  cuts = getGroupCutoffs( (*j1), (*j2) );
1514 +                  if (dr.lengthSquare() < cuts.third) {
1515 +                    neighborList.push_back(make_pair((*j1), (*j2)));
1516 +                  }                  
1517 +                }
1518 +              }
1519 + #else
1520 +              for (vector<int>::iterator j1 = cellList_[m1].begin();
1521 +                   j1 != cellList_[m1].end(); ++j1) {
1522 +                for (vector<int>::iterator j2 = cellList_[m2].begin();
1523 +                     j2 != cellList_[m2].end(); ++j2) {
1524 +    
1525 +                  // Always do this if we're in different cells or if
1526 +                  // we're in the same cell and the global index of
1527 +                  // the j2 cutoff group is greater than or equal to
1528 +                  // the j1 cutoff group.  Note that Rappaport's code
1529 +                  // has a "less than" conditional here, but that
1530 +                  // deals with atom-by-atom computation.  OpenMD
1531 +                  // allows atoms within a single cutoff group to
1532 +                  // interact with each other.
1533 +
1534 +                  if (m2 != m1 || (*j2) >= (*j1) ) {
1535 +
1536 +                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1537 +                    if (usePeriodicBoundaryConditions_) {
1538 +                      snap_->wrapVector(dr);
1539 +                    }
1540 +                    cuts = getGroupCutoffs( (*j1), (*j2) );
1541 +                    if (dr.lengthSquare() < cuts.third) {
1542 +                      neighborList.push_back(make_pair((*j1), (*j2)));
1543 +                    }
1544 +                  }
1545 +                }
1546 +              }
1547 + #endif
1548 +            }
1549 +          }
1550 +        }
1551 +      }
1552 +    } else {
1553 +      // branch to do all cutoff group pairs
1554 + #ifdef IS_MPI
1555 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1556 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
1557 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1558 +          if (usePeriodicBoundaryConditions_) {
1559 +            snap_->wrapVector(dr);
1560 +          }
1561 +          cuts = getGroupCutoffs( j1, j2 );
1562 +          if (dr.lengthSquare() < cuts.third) {
1563 +            neighborList.push_back(make_pair(j1, j2));
1564 +          }
1565 +        }
1566 +      }      
1567 + #else
1568 +      // include all groups here.
1569 +      for (int j1 = 0; j1 < nGroups_; j1++) {
1570 +        // include self group interactions j2 == j1
1571 +        for (int j2 = j1; j2 < nGroups_; j2++) {
1572 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1573 +          if (usePeriodicBoundaryConditions_) {
1574 +            snap_->wrapVector(dr);
1575 +          }
1576 +          cuts = getGroupCutoffs( j1, j2 );
1577 +          if (dr.lengthSquare() < cuts.third) {
1578 +            neighborList.push_back(make_pair(j1, j2));
1579 +          }
1580 +        }    
1581 +      }
1582 + #endif
1583 +    }
1584 +      
1585 +    // save the local cutoff group positions for the check that is
1586 +    // done on each loop:
1587 +    saved_CG_positions_.clear();
1588 +    for (int i = 0; i < nGroups_; i++)
1589 +      saved_CG_positions_.push_back(snap_->cgData.position[i]);
1590 +  }
1591   } //end namespace OpenMD

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines