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 1541 by gezelter, Fri Feb 4 20:04:56 2011 UTC vs.
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1590 by gezelter, Mon Jul 11 01:39:49 2011 UTC

# Line 38 | Line 38
38   * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
39   * [4]  Vardeman & Gezelter, in progress (2009).                        
40   */
41 < #include "parallel/ForceDecomposition.hpp"
42 < #include "parallel/Communicator.hpp"
41 > #include "parallel/ForceMatrixDecomposition.hpp"
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 {
49  
50 <  void ForceDecomposition::distributeInitialData() {
50 >  /**
51 >   * distributeInitialData is essentially a copy of the older fortran
52 >   * SimulationSetup
53 >   */
54 >  
55 >  void ForceMatrixDecomposition::distributeInitialData() {
56 >    snap_ = sman_->getCurrentSnapshot();
57 >    storageLayout_ = sman_->getStorageLayout();
58 >    ff_ = info_->getForceField();
59 >    nLocal_ = snap_->getNumberOfAtoms();
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 <    Snapshot* snap = sman_->getCurrentSnapshot();
77 <    int nAtoms = snap->getNumberOfAtoms();
78 <    int nGroups = snap->getNumberOfCutoffGroups();
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 <    AtomCommRealI = new Communicator<Row,RealType>(nAtoms);
84 <    AtomCommVectorI = new Communicator<Row,Vector3d>(nAtoms);
85 <    AtomCommMatrixI = new Communicator<Row,Mat3x3d>(nAtoms);
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 <    AtomCommRealJ = new Communicator<Column,RealType>(nAtoms);
90 <    AtomCommVectorJ = new Communicator<Column,Vector3d>(nAtoms);
91 <    AtomCommMatrixJ = new Communicator<Column,Mat3x3d>(nAtoms);
89 >    cgCommIntRow = new Communicator<Row,int>(nGroups_);
90 >    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
91 >    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
92 >    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
93  
94 <    cgCommVectorI = new Communicator<Row,Vector3d>(nGroups);
95 <    cgCommVectorJ = new Communicator<Column,Vector3d>(nGroups);
94 >    nAtomsInRow_ = AtomCommIntRow->getSize();
95 >    nAtomsInCol_ = AtomCommIntColumn->getSize();
96 >    nGroupsInRow_ = cgCommIntRow->getSize();
97 >    nGroupsInCol_ = cgCommIntColumn->getSize();
98  
99 <    int nInRow = AtomCommRealI.getSize();
100 <    int nInCol = AtomCommRealJ.getSize();
99 >    // Modify the data storage objects with the correct layouts and sizes:
100 >    atomRowData.resize(nAtomsInRow_);
101 >    atomRowData.setStorageLayout(storageLayout_);
102 >    atomColData.resize(nAtomsInCol_);
103 >    atomColData.setStorageLayout(storageLayout_);
104 >    cgRowData.resize(nGroupsInRow_);
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 >    AtomCommIntRow->gather(idents, identsRow);
113 >    AtomCommIntColumn->gather(idents, identsCol);
114 >    
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 <    vector<vector<RealType> > pot_row(LR_POT_TYPES,
126 <                                      vector<RealType> (nInRow, 0.0));
127 <    vector<vector<RealType> > pot_col(LR_POT_TYPES,
128 <                                      vector<RealType> (nInCol, 0.0));
125 >    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
126 >    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
127 >    
128 >    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
129 >    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
130  
131 <    vector<vector<RealType> > pot_local(LR_POT_TYPES,
132 <                                        vector<RealType> (nAtoms, 0.0));
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 <  void ForceDecomposition::distributeData()  {
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 <    Snapshot* snap = sman_->getCurrentSnapshot();
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();
508 + #ifdef IS_MPI
509 +    
510      // gather up the atomic positions
511 <    AtomCommVectorI->gather(snap->atomData.position,
512 <                            snap->atomIData.position);
513 <    AtomCommVectorJ->gather(snap->atomData.position,
514 <                            snap->atomJData.position);
511 >    AtomCommVectorRow->gather(snap_->atomData.position,
512 >                              atomRowData.position);
513 >    AtomCommVectorColumn->gather(snap_->atomData.position,
514 >                                 atomColData.position);
515      
516      // gather up the cutoff group positions
517 <    cgCommVectorI->gather(snap->cgData.position,
518 <                          snap->cgIData.position);
519 <    cgCommVectorJ->gather(snap->cgData.position,
520 <                          snap->cgJData.position);
517 >    cgCommVectorRow->gather(snap_->cgData.position,
518 >                            cgRowData.position);
519 >    cgCommVectorColumn->gather(snap_->cgData.position,
520 >                               cgColData.position);
521      
522      // if needed, gather the atomic rotation matrices
523 <    if (snap->atomData.getStorageLayout() & DataStorage::dslAmat) {
524 <      AtomCommMatrixI->gather(snap->atomData.aMat,
525 <                              snap->atomIData.aMat);
526 <      AtomCommMatrixJ->gather(snap->atomData.aMat,
527 <                              snap->atomJData.aMat);
523 >    if (storageLayout_ & DataStorage::dslAmat) {
524 >      AtomCommMatrixRow->gather(snap_->atomData.aMat,
525 >                                atomRowData.aMat);
526 >      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
527 >                                   atomColData.aMat);
528      }
529      
530      // if needed, gather the atomic eletrostatic frames
531 <    if (snap->atomData.getStorageLayout() & DataStorage::dslElectroFrame) {
532 <      AtomCommMatrixI->gather(snap->atomData.electroFrame,
533 <                              snap->atomIData.electroFrame);
534 <      AtomCommMatrixJ->gather(snap->atomData.electroFrame,
535 <                              snap->atomJData.electroFrame);
531 >    if (storageLayout_ & DataStorage::dslElectroFrame) {
532 >      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
533 >                                atomRowData.electroFrame);
534 >      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
535 >                                   atomColData.electroFrame);
536      }
537 +
538   #endif      
539    }
540    
541 <  void ForceDecomposition::collectIntermediateData() {
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();
547   #ifdef IS_MPI
117 –    Snapshot* snap = sman_->getCurrentSnapshot();
548      
549 <    if (snap->atomData.getStorageLayout() & DataStorage::dslDensity) {
550 <
551 <      AtomCommRealI->scatter(snap->atomIData.density,
552 <                             snap->atomData.density);
553 <
554 <      int n = snap->atomData.density.size();
555 <      std::vector<RealType> rho_tmp(n, 0.0);
556 <      AtomCommRealJ->scatter(snap->atomJData.density, rho_tmp);
549 >    if (storageLayout_ & DataStorage::dslDensity) {
550 >      
551 >      AtomCommRealRow->scatter(atomRowData.density,
552 >                               snap_->atomData.density);
553 >      
554 >      int n = snap_->atomData.density.size();
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];
558 >        snap_->atomData.density[i] += rho_tmp[i];
559      }
560   #endif
561    }
562 <  
563 <  void ForceDecomposition::distributeIntermediateData() {
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();
570   #ifdef IS_MPI
571 <    Snapshot* snap = sman_->getCurrentSnapshot();
572 <    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctional) {
573 <      AtomCommRealI->gather(snap->atomData.functional,
574 <                            snap->atomIData.functional);
575 <      AtomCommRealJ->gather(snap->atomData.functional,
140 <                            snap->atomJData.functional);
571 >    if (storageLayout_ & DataStorage::dslFunctional) {
572 >      AtomCommRealRow->gather(snap_->atomData.functional,
573 >                              atomRowData.functional);
574 >      AtomCommRealColumn->gather(snap_->atomData.functional,
575 >                                 atomColData.functional);
576      }
577      
578 <    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctionalDerivative) {
579 <      AtomCommRealI->gather(snap->atomData.functionalDerivative,
580 <                            snap->atomIData.functionalDerivative);
581 <      AtomCommRealJ->gather(snap->atomData.functionalDerivative,
582 <                            snap->atomJData.functionalDerivative);
578 >    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
579 >      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
580 >                              atomRowData.functionalDerivative);
581 >      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
582 >                                 atomColData.functionalDerivative);
583      }
584   #endif
585    }
586    
587    
588 <  void ForceDecomposition::collectData() {
589 < #ifdef IS_MPI
590 <    Snapshot* snap = sman_->getCurrentSnapshot();
588 >  void ForceMatrixDecomposition::collectData() {
589 >    snap_ = sman_->getCurrentSnapshot();
590 >    storageLayout_ = sman_->getStorageLayout();
591 > #ifdef IS_MPI    
592 >    int n = snap_->atomData.force.size();
593 >    vector<Vector3d> frc_tmp(n, V3Zero);
594      
595 <    int n = snap->atomData.force.size();
158 <    std::vector<Vector3d> frc_tmp(n, 0.0);
159 <    
160 <    AtomCommVectorI->scatter(snap->atomIData.force, frc_tmp);
595 >    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
596      for (int i = 0; i < n; i++) {
597 <      snap->atomData.force[i] += frc_tmp[i];
597 >      snap_->atomData.force[i] += frc_tmp[i];
598        frc_tmp[i] = 0.0;
599      }
600      
601 <    AtomCommVectorJ->scatter(snap->atomJData.force, frc_tmp);
601 >    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
602      for (int i = 0; i < n; i++)
603 <      snap->atomData.force[i] += frc_tmp[i];
604 <    
605 <    
171 <    if (snap->atomData.getStorageLayout() & DataStorage::dslTorque) {
603 >      snap_->atomData.force[i] += frc_tmp[i];
604 >        
605 >    if (storageLayout_ & DataStorage::dslTorque) {
606  
607 <      int nt = snap->atomData.force.size();
608 <      std::vector<Vector3d> trq_tmp(nt, 0.0);
607 >      int nt = snap_->atomData.torque.size();
608 >      vector<Vector3d> trq_tmp(nt, V3Zero);
609  
610 <      AtomCommVectorI->scatter(snap->atomIData.torque, trq_tmp);
611 <      for (int i = 0; i < n; i++) {
612 <        snap->atomData.torque[i] += trq_tmp[i];
610 >      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
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 <      AtomCommVectorJ->scatter(snap->atomJData.torque, trq_tmp);
617 <      for (int i = 0; i < n; i++)
618 <        snap->atomData.torque[i] += trq_tmp[i];
616 >      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
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<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 <    vector<vector<RealType> > pot_temp(LR_POT_TYPES,
650 <                                       vector<RealType> (nAtoms, 0.0));
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 i = 0; i < LR_POT_TYPES; i++) {
655 <      AtomCommRealI->scatter(pot_row[i], pot_temp[i]);
656 <      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
657 <        pot_local[i] += pot_temp[i][ii];
658 <      }
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;
689 >    
690 > #ifdef IS_MPI
691 >    d = cgColData.position[cg2] - cgRowData.position[cg1];
692 > #else
693 >    d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
694 > #endif
695 >    
696 >    snap_->wrapVector(d);
697 >    return d;    
698 >  }
699 >
700 >
701 >  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
702 >
703 >    Vector3d d;
704 >    
705 > #ifdef IS_MPI
706 >    d = cgRowData.position[cg1] - atomRowData.position[atom1];
707 > #else
708 >    d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
709 > #endif
710 >
711 >    snap_->wrapVector(d);
712 >    return d;    
713 >  }
714 >  
715 >  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
716 >    Vector3d d;
717 >    
718 > #ifdef IS_MPI
719 >    d = cgColData.position[cg2] - atomColData.position[atom2];
720 > #else
721 >    d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
722 > #endif
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){
746 >    Vector3d d;
747 >    
748 > #ifdef IS_MPI
749 >    d = atomColData.position[atom2] - atomRowData.position[atom1];
750 > #else
751 >    d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
752 > #endif
753 >
754 >    snap_->wrapVector(d);
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;
819 + #else
820 +    snap_->atomData.force[atom1] += fg;
821 + #endif
822 +  }
823 +
824 +  void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
825 + #ifdef IS_MPI
826 +    atomColData.force[atom2] += fg;
827 + #else
828 +    snap_->atomData.force[atom2] += fg;
829 + #endif
830 +  }
831 +
832 +    // filling interaction blocks with pointers
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]);
846 +    }
847 +    
848 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
849 +      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
850 +      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
851 +    }
852 +
853 +    if (storageLayout_ & DataStorage::dslTorque) {
854 +      idat.t1 = &(atomRowData.torque[atom1]);
855 +      idat.t2 = &(atomColData.torque[atom2]);
856 +    }
857 +
858 +    if (storageLayout_ & DataStorage::dslDensity) {
859 +      idat.rho1 = &(atomRowData.density[atom1]);
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]);
891 +    }
892 +
893 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
894 +      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
895 +      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
896 +    }
897  
898 +    if (storageLayout_ & DataStorage::dslTorque) {
899 +      idat.t1 = &(snap_->atomData.torque[atom1]);
900 +      idat.t2 = &(snap_->atomData.torque[atom2]);
901 +    }
902  
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
928    }
929 +
930    
931 +  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
932 + #ifdef IS_MPI
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 +    pairwisePot += *(idat.pot);
940 +
941 +    snap_->atomData.force[atom1] += *(idat.f1);
942 +    snap_->atomData.force[atom2] -= *(idat.f1);
943 + #endif
944 +    
945 +  }
946 +
947 +  /*
948 +   * buildNeighborList
949 +   *
950 +   * first element of pair is row-indexed CutoffGroup
951 +   * second element of pair is column-indexed CutoffGroup
952 +   */
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();
962 + #else
963 +    cellList_.clear();
964 + #endif
965 +
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);
973 +
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 +    cellListRow_.resize(nCtot);
992 +    cellListCol_.resize(nCtot);
993 + #else
994 +    cellList_.resize(nCtot);
995 + #endif
996 +
997 +    if (!doAllPairs) {
998 + #ifdef IS_MPI
999 +
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 +      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 +                    cuts = getGroupCutoffs( (*j1), (*j2) );
1121 +                    if (dr.lengthSquare() < cuts.third) {
1122 +                      neighborList.push_back(make_pair((*j1), (*j2)));
1123 +                    }
1124 +                  }
1125 +                }
1126 +              }
1127 + #else
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 +              }
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