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 1538 by chuckv, Tue Jan 11 18:58:12 2011 UTC vs.
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1588 by gezelter, Sat Jul 9 15:05:59 2011 UTC

# Line 1 | Line 1
1 < /**
2 < * @file ForceDecomposition.cpp
3 < * @author Charles Vardeman <cvardema.at.nd.edu>
4 < * @date 08/18/2010
5 < * @time 11:56am
6 < * @version 1.0
1 > /*
2 > * Copyright (c) 2005 The University of Notre Dame. All Rights Reserved.
3   *
8 – * @section LICENSE
9 – * Copyright (c) 2010 The University of Notre Dame. All Rights Reserved.
10 – *
4   * The University of Notre Dame grants you ("Licensee") a
5   * non-exclusive, royalty free, license to use, modify and
6   * redistribute this software in source and binary code form, provided
# Line 45 | Line 38
38   * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
39   * [4]  Vardeman & Gezelter, in progress (2009).                        
40   */
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 +  /**
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 < /*  -*- c++ -*-  */
69 < #include "config.h"
70 < #include <stdlib.h>
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 < #include <mpi.h>
77 < #endif
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 < #include <iostream>
84 < #include <vector>
85 < #include <algorithm>
86 < #include <cmath>
87 < #include "parallel/ForceDecomposition.hpp"
83 >    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
84 >    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
85 >    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
86 >    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
87 >    AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
88  
89 +    cgCommIntRow = new Communicator<Row,int>(nGroups_);
90 +    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
91 +    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
92 +    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
93  
94 < using namespace std;
95 < using namespace OpenMD;
94 >    nAtomsInRow_ = AtomCommIntRow->getSize();
95 >    nAtomsInCol_ = AtomCommIntColumn->getSize();
96 >    nGroupsInRow_ = cgCommIntRow->getSize();
97 >    nGroupsInCol_ = cgCommIntColumn->getSize();
98  
99 < //__static
100 < #ifdef IS_MPI
101 < static vector<MPI:Comm> communictors;
102 < #endif
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 >    vector<int>::iterator it;
116 >    for (it = AtomLocalToGlobal.begin(); it != AtomLocalToGlobal.end(); ++it) {
117 >      cerr << "my AtomLocalToGlobal = " << (*it) << "\n";
118 >    }
119 >    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
120 >    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
121 >    
122 >    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
123 >    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
124  
125 < //____ MPITypeTraits
126 < template<typename T>
75 < struct MPITypeTraits;
125 >    AtomCommRealRow->gather(massFactors, massFactorsRow);
126 >    AtomCommRealColumn->gather(massFactors, massFactorsCol);
127  
128 < #ifdef IS_MPI
129 < template<>
130 < struct MPITypeTraits<RealType> {
131 <  static const MPI::Datatype datatype;
132 < };
133 < const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL;
128 >    groupListRow_.clear();
129 >    groupListRow_.resize(nGroupsInRow_);
130 >    for (int i = 0; i < nGroupsInRow_; i++) {
131 >      int gid = cgRowToGlobal[i];
132 >      for (int j = 0; j < nAtomsInRow_; j++) {
133 >        int aid = AtomRowToGlobal[j];
134 >        if (globalGroupMembership[aid] == gid)
135 >          groupListRow_[i].push_back(j);
136 >      }      
137 >    }
138  
139 < template<>
140 < struct MPITypeTraits<int> {
141 <  static const MPI::Datatype datatype;
142 < };
143 < const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT;
144 < #endif
139 >    groupListCol_.clear();
140 >    groupListCol_.resize(nGroupsInCol_);
141 >    for (int i = 0; i < nGroupsInCol_; i++) {
142 >      int gid = cgColToGlobal[i];
143 >      for (int j = 0; j < nAtomsInCol_; j++) {
144 >        int aid = AtomColToGlobal[j];
145 >        if (globalGroupMembership[aid] == gid)
146 >          groupListCol_[i].push_back(j);
147 >      }      
148 >    }
149  
150 < /**
151 < * Constructor for ForceDecomposition Parallel Decomposition Method
152 < * Will try to construct a symmetric grid of processors. Ideally, the
153 < * number of processors will be a square ex: 4, 9, 16, 25.
154 < *
155 < */
150 >    excludesForAtom.clear();
151 >    excludesForAtom.resize(nAtomsInRow_);
152 >    toposForAtom.clear();
153 >    toposForAtom.resize(nAtomsInRow_);
154 >    topoDist.clear();
155 >    topoDist.resize(nAtomsInRow_);
156 >    for (int i = 0; i < nAtomsInRow_; i++) {
157 >      int iglob = AtomRowToGlobal[i];
158  
159 < ForceDecomposition::ForceDecomposition() {
159 >      for (int j = 0; j < nAtomsInCol_; j++) {
160 >        int jglob = AtomColToGlobal[j];
161  
162 < #ifdef IS_MPI
163 <  int nProcs = MPI::COMM_WORLD.Get_size();
164 <  int worldRank = MPI::COMM_WORLD.Get_rank();
162 >        if (excludes->hasPair(iglob, jglob))
163 >          excludesForAtom[i].push_back(j);      
164 >        
165 >        if (oneTwo->hasPair(iglob, jglob)) {
166 >          toposForAtom[i].push_back(j);
167 >          topoDist[i].push_back(1);
168 >        } else {
169 >          if (oneThree->hasPair(iglob, jglob)) {
170 >            toposForAtom[i].push_back(j);
171 >            topoDist[i].push_back(2);
172 >          } else {
173 >            if (oneFour->hasPair(iglob, jglob)) {
174 >              toposForAtom[i].push_back(j);
175 >              topoDist[i].push_back(3);
176 >            }
177 >          }
178 >        }
179 >      }      
180 >    }
181 >
182   #endif
183  
184 <  // First time through, construct column stride.
185 <  if (communicators.size() == 0)
186 <  {
187 <    int nColumnsMax = (int) round(sqrt((float) nProcs));
188 <    for (int i = 0; i < nProcs; ++i)
189 <    {
190 <      if (nProcs%i==0) nColumns=i;
184 >    groupList_.clear();
185 >    groupList_.resize(nGroups_);
186 >    for (int i = 0; i < nGroups_; i++) {
187 >      int gid = cgLocalToGlobal[i];
188 >      for (int j = 0; j < nLocal_; j++) {
189 >        int aid = AtomLocalToGlobal[j];
190 >        if (globalGroupMembership[aid] == gid) {
191 >          groupList_[i].push_back(j);
192 >        }
193 >      }      
194      }
195  
196 <    int nRows = nProcs/nColumns;    
197 <    myRank_ = (int) worldRank%nColumns;
198 <  }
199 <  else
200 <  {
201 <    myRank_ = myRank/nColumns;
120 <  }
121 <  MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0);
122 <  
123 <  isColumn_ = false;
124 <  
125 < }
196 >    excludesForAtom.clear();
197 >    excludesForAtom.resize(nLocal_);
198 >    toposForAtom.clear();
199 >    toposForAtom.resize(nLocal_);
200 >    topoDist.clear();
201 >    topoDist.resize(nLocal_);
202  
203 < ForceDecomposition::gather(sendbuf, receivebuf){
204 <  communicators(myIndex_).Allgatherv();
129 < }
203 >    for (int i = 0; i < nLocal_; i++) {
204 >      int iglob = AtomLocalToGlobal[i];
205  
206 +      for (int j = 0; j < nLocal_; j++) {
207 +        int jglob = AtomLocalToGlobal[j];
208  
209 +        if (excludes->hasPair(iglob, jglob))
210 +          excludesForAtom[i].push_back(j);              
211 +        
212 +        if (oneTwo->hasPair(iglob, jglob)) {
213 +          toposForAtom[i].push_back(j);
214 +          topoDist[i].push_back(1);
215 +        } else {
216 +          if (oneThree->hasPair(iglob, jglob)) {
217 +            toposForAtom[i].push_back(j);
218 +            topoDist[i].push_back(2);
219 +          } else {
220 +            if (oneFour->hasPair(iglob, jglob)) {
221 +              toposForAtom[i].push_back(j);
222 +              topoDist[i].push_back(3);
223 +            }
224 +          }
225 +        }
226 +      }      
227 +    }
228 +    
229 +    createGtypeCutoffMap();
230  
231 < ForceDecomposition::scatter(sbuffer, rbuffer){
232 <  communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM);
233 < }
231 >  }
232 >  
233 >  void ForceMatrixDecomposition::createGtypeCutoffMap() {
234 >    
235 >    RealType tol = 1e-6;
236 >    RealType rc;
237 >    int atid;
238 >    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
239 >    map<int, RealType> atypeCutoff;
240 >      
241 >    for (set<AtomType*>::iterator at = atypes.begin();
242 >         at != atypes.end(); ++at){
243 >      atid = (*at)->getIdent();
244 >      if (userChoseCutoff_)
245 >        atypeCutoff[atid] = userCutoff_;
246 >      else
247 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
248 >    }
249 >
250 >    vector<RealType> gTypeCutoffs;
251 >    // first we do a single loop over the cutoff groups to find the
252 >    // largest cutoff for any atypes present in this group.
253 > #ifdef IS_MPI
254 >    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
255 >    groupRowToGtype.resize(nGroupsInRow_);
256 >    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
257 >      vector<int> atomListRow = getAtomsInGroupRow(cg1);
258 >      for (vector<int>::iterator ia = atomListRow.begin();
259 >           ia != atomListRow.end(); ++ia) {            
260 >        int atom1 = (*ia);
261 >        atid = identsRow[atom1];
262 >        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
263 >          groupCutoffRow[cg1] = atypeCutoff[atid];
264 >        }
265 >      }
266 >
267 >      bool gTypeFound = false;
268 >      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
269 >        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
270 >          groupRowToGtype[cg1] = gt;
271 >          gTypeFound = true;
272 >        }
273 >      }
274 >      if (!gTypeFound) {
275 >        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
276 >        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
277 >      }
278 >      
279 >    }
280 >    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
281 >    groupColToGtype.resize(nGroupsInCol_);
282 >    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
283 >      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
284 >      for (vector<int>::iterator jb = atomListCol.begin();
285 >           jb != atomListCol.end(); ++jb) {            
286 >        int atom2 = (*jb);
287 >        atid = identsCol[atom2];
288 >        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
289 >          groupCutoffCol[cg2] = atypeCutoff[atid];
290 >        }
291 >      }
292 >      bool gTypeFound = false;
293 >      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
294 >        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
295 >          groupColToGtype[cg2] = gt;
296 >          gTypeFound = true;
297 >        }
298 >      }
299 >      if (!gTypeFound) {
300 >        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
301 >        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
302 >      }
303 >    }
304 > #else
305  
306 +    vector<RealType> groupCutoff(nGroups_, 0.0);
307 +    groupToGtype.resize(nGroups_);
308 +    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
309  
310 +      groupCutoff[cg1] = 0.0;
311 +      vector<int> atomList = getAtomsInGroupRow(cg1);
312 +
313 +      for (vector<int>::iterator ia = atomList.begin();
314 +           ia != atomList.end(); ++ia) {            
315 +        int atom1 = (*ia);
316 +        atid = idents[atom1];
317 +        if (atypeCutoff[atid] > groupCutoff[cg1]) {
318 +          groupCutoff[cg1] = atypeCutoff[atid];
319 +        }
320 +      }
321 +
322 +      bool gTypeFound = false;
323 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
324 +        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
325 +          groupToGtype[cg1] = gt;
326 +          gTypeFound = true;
327 +        }
328 +      }
329 +      if (!gTypeFound) {
330 +        gTypeCutoffs.push_back( groupCutoff[cg1] );
331 +        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
332 +      }      
333 +    }
334 + #endif
335 +
336 +    // Now we find the maximum group cutoff value present in the simulation
337 +
338 +    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
339 +
340 + #ifdef IS_MPI
341 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
342 + #endif
343 +    
344 +    RealType tradRcut = groupMax;
345 +
346 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
347 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
348 +        RealType thisRcut;
349 +        switch(cutoffPolicy_) {
350 +        case TRADITIONAL:
351 +          thisRcut = tradRcut;
352 +          break;
353 +        case MIX:
354 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
355 +          break;
356 +        case MAX:
357 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
358 +          break;
359 +        default:
360 +          sprintf(painCave.errMsg,
361 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
362 +                  "hit an unknown cutoff policy!\n");
363 +          painCave.severity = OPENMD_ERROR;
364 +          painCave.isFatal = 1;
365 +          simError();
366 +          break;
367 +        }
368 +
369 +        pair<int,int> key = make_pair(i,j);
370 +        gTypeCutoffMap[key].first = thisRcut;
371 +
372 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
373 +
374 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
375 +        
376 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
377 +
378 +        // sanity check
379 +        
380 +        if (userChoseCutoff_) {
381 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
382 +            sprintf(painCave.errMsg,
383 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
384 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
385 +            painCave.severity = OPENMD_ERROR;
386 +            painCave.isFatal = 1;
387 +            simError();            
388 +          }
389 +        }
390 +      }
391 +    }
392 +  }
393 +
394 +
395 +  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
396 +    int i, j;  
397 + #ifdef IS_MPI
398 +    i = groupRowToGtype[cg1];
399 +    j = groupColToGtype[cg2];
400 + #else
401 +    i = groupToGtype[cg1];
402 +    j = groupToGtype[cg2];
403 + #endif    
404 +    return gTypeCutoffMap[make_pair(i,j)];
405 +  }
406 +
407 +  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
408 +    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
409 +      if (toposForAtom[atom1][j] == atom2)
410 +        return topoDist[atom1][j];
411 +    }
412 +    return 0;
413 +  }
414 +
415 +  void ForceMatrixDecomposition::zeroWorkArrays() {
416 +    pairwisePot = 0.0;
417 +    embeddingPot = 0.0;
418 +
419 + #ifdef IS_MPI
420 +    if (storageLayout_ & DataStorage::dslForce) {
421 +      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
422 +      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
423 +    }
424 +
425 +    if (storageLayout_ & DataStorage::dslTorque) {
426 +      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
427 +      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
428 +    }
429 +    
430 +    fill(pot_row.begin(), pot_row.end(),
431 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
432 +
433 +    fill(pot_col.begin(), pot_col.end(),
434 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
435 +
436 +    if (storageLayout_ & DataStorage::dslParticlePot) {    
437 +      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
438 +      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
439 +    }
440 +
441 +    if (storageLayout_ & DataStorage::dslDensity) {      
442 +      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
443 +      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
444 +    }
445 +
446 +    if (storageLayout_ & DataStorage::dslFunctional) {  
447 +      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
448 +      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
449 +    }
450 +
451 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
452 +      fill(atomRowData.functionalDerivative.begin(),
453 +           atomRowData.functionalDerivative.end(), 0.0);
454 +      fill(atomColData.functionalDerivative.begin(),
455 +           atomColData.functionalDerivative.end(), 0.0);
456 +    }
457 +
458 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
459 +      fill(atomRowData.skippedCharge.begin(),
460 +           atomRowData.skippedCharge.end(), 0.0);
461 +      fill(atomColData.skippedCharge.begin(),
462 +           atomColData.skippedCharge.end(), 0.0);
463 +    }
464 +
465 + #else
466 +    
467 +    if (storageLayout_ & DataStorage::dslParticlePot) {      
468 +      fill(snap_->atomData.particlePot.begin(),
469 +           snap_->atomData.particlePot.end(), 0.0);
470 +    }
471 +    
472 +    if (storageLayout_ & DataStorage::dslDensity) {      
473 +      fill(snap_->atomData.density.begin(),
474 +           snap_->atomData.density.end(), 0.0);
475 +    }
476 +    if (storageLayout_ & DataStorage::dslFunctional) {
477 +      fill(snap_->atomData.functional.begin(),
478 +           snap_->atomData.functional.end(), 0.0);
479 +    }
480 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
481 +      fill(snap_->atomData.functionalDerivative.begin(),
482 +           snap_->atomData.functionalDerivative.end(), 0.0);
483 +    }
484 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
485 +      fill(snap_->atomData.skippedCharge.begin(),
486 +           snap_->atomData.skippedCharge.end(), 0.0);
487 +    }
488 + #endif
489 +    
490 +  }
491 +
492 +
493 +  void ForceMatrixDecomposition::distributeData()  {
494 +    snap_ = sman_->getCurrentSnapshot();
495 +    storageLayout_ = sman_->getStorageLayout();
496 + #ifdef IS_MPI
497 +    
498 +    // gather up the atomic positions
499 +    AtomCommVectorRow->gather(snap_->atomData.position,
500 +                              atomRowData.position);
501 +    AtomCommVectorColumn->gather(snap_->atomData.position,
502 +                                 atomColData.position);
503 +    
504 +    // gather up the cutoff group positions
505 +    cgCommVectorRow->gather(snap_->cgData.position,
506 +                            cgRowData.position);
507 +    cgCommVectorColumn->gather(snap_->cgData.position,
508 +                               cgColData.position);
509 +    
510 +    // if needed, gather the atomic rotation matrices
511 +    if (storageLayout_ & DataStorage::dslAmat) {
512 +      AtomCommMatrixRow->gather(snap_->atomData.aMat,
513 +                                atomRowData.aMat);
514 +      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
515 +                                   atomColData.aMat);
516 +    }
517 +    
518 +    // if needed, gather the atomic eletrostatic frames
519 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
520 +      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
521 +                                atomRowData.electroFrame);
522 +      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
523 +                                   atomColData.electroFrame);
524 +    }
525 + #endif      
526 +  }
527 +  
528 +  /* collects information obtained during the pre-pair loop onto local
529 +   * data structures.
530 +   */
531 +  void ForceMatrixDecomposition::collectIntermediateData() {
532 +    snap_ = sman_->getCurrentSnapshot();
533 +    storageLayout_ = sman_->getStorageLayout();
534 + #ifdef IS_MPI
535 +    
536 +    if (storageLayout_ & DataStorage::dslDensity) {
537 +      
538 +      AtomCommRealRow->scatter(atomRowData.density,
539 +                               snap_->atomData.density);
540 +      
541 +      int n = snap_->atomData.density.size();
542 +      vector<RealType> rho_tmp(n, 0.0);
543 +      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
544 +      for (int i = 0; i < n; i++)
545 +        snap_->atomData.density[i] += rho_tmp[i];
546 +    }
547 + #endif
548 +  }
549 +
550 +  /*
551 +   * redistributes information obtained during the pre-pair loop out to
552 +   * row and column-indexed data structures
553 +   */
554 +  void ForceMatrixDecomposition::distributeIntermediateData() {
555 +    snap_ = sman_->getCurrentSnapshot();
556 +    storageLayout_ = sman_->getStorageLayout();
557 + #ifdef IS_MPI
558 +    if (storageLayout_ & DataStorage::dslFunctional) {
559 +      AtomCommRealRow->gather(snap_->atomData.functional,
560 +                              atomRowData.functional);
561 +      AtomCommRealColumn->gather(snap_->atomData.functional,
562 +                                 atomColData.functional);
563 +    }
564 +    
565 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
566 +      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
567 +                              atomRowData.functionalDerivative);
568 +      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
569 +                                 atomColData.functionalDerivative);
570 +    }
571 + #endif
572 +  }
573 +  
574 +  
575 +  void ForceMatrixDecomposition::collectData() {
576 +    snap_ = sman_->getCurrentSnapshot();
577 +    storageLayout_ = sman_->getStorageLayout();
578 + #ifdef IS_MPI    
579 +    int n = snap_->atomData.force.size();
580 +    vector<Vector3d> frc_tmp(n, V3Zero);
581 +    
582 +    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
583 +    for (int i = 0; i < n; i++) {
584 +      snap_->atomData.force[i] += frc_tmp[i];
585 +      frc_tmp[i] = 0.0;
586 +    }
587 +    
588 +    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
589 +    for (int i = 0; i < n; i++)
590 +      snap_->atomData.force[i] += frc_tmp[i];
591 +    
592 +    
593 +    if (storageLayout_ & DataStorage::dslTorque) {
594 +
595 +      int nt = snap_->atomData.torque.size();
596 +      vector<Vector3d> trq_tmp(nt, V3Zero);
597 +
598 +      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
599 +      for (int i = 0; i < nt; i++) {
600 +        snap_->atomData.torque[i] += trq_tmp[i];
601 +        trq_tmp[i] = 0.0;
602 +      }
603 +      
604 +      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
605 +      for (int i = 0; i < nt; i++)
606 +        snap_->atomData.torque[i] += trq_tmp[i];
607 +    }
608 +
609 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
610 +
611 +      int ns = snap_->atomData.skippedCharge.size();
612 +      vector<RealType> skch_tmp(ns, 0.0);
613 +
614 +      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
615 +      for (int i = 0; i < ns; i++) {
616 +        snap_->atomData.skippedCharge[i] = skch_tmp[i];
617 +        skch_tmp[i] = 0.0;
618 +      }
619 +      
620 +      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
621 +      for (int i = 0; i < ns; i++)
622 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
623 +    }
624 +    
625 +    nLocal_ = snap_->getNumberOfAtoms();
626 +
627 +    vector<potVec> pot_temp(nLocal_,
628 +                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
629 +
630 +    // scatter/gather pot_row into the members of my column
631 +          
632 +    AtomCommPotRow->scatter(pot_row, pot_temp);
633 +
634 +    for (int ii = 0;  ii < pot_temp.size(); ii++ )
635 +      pairwisePot += pot_temp[ii];
636 +    
637 +    fill(pot_temp.begin(), pot_temp.end(),
638 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
639 +      
640 +    AtomCommPotColumn->scatter(pot_col, pot_temp);    
641 +    
642 +    for (int ii = 0;  ii < pot_temp.size(); ii++ )
643 +      pairwisePot += pot_temp[ii];    
644 + #endif
645 +
646 +  }
647 +
648 +  int ForceMatrixDecomposition::getNAtomsInRow() {  
649 + #ifdef IS_MPI
650 +    return nAtomsInRow_;
651 + #else
652 +    return nLocal_;
653 + #endif
654 +  }
655 +
656 +  /**
657 +   * returns the list of atoms belonging to this group.  
658 +   */
659 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
660 + #ifdef IS_MPI
661 +    return groupListRow_[cg1];
662 + #else
663 +    return groupList_[cg1];
664 + #endif
665 +  }
666 +
667 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
668 + #ifdef IS_MPI
669 +    return groupListCol_[cg2];
670 + #else
671 +    return groupList_[cg2];
672 + #endif
673 +  }
674 +  
675 +  Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
676 +    Vector3d d;
677 +    
678 + #ifdef IS_MPI
679 +    d = cgColData.position[cg2] - cgRowData.position[cg1];
680 + #else
681 +    d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
682 + #endif
683 +    
684 +    snap_->wrapVector(d);
685 +    return d;    
686 +  }
687 +
688 +
689 +  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
690 +
691 +    Vector3d d;
692 +    
693 + #ifdef IS_MPI
694 +    d = cgRowData.position[cg1] - atomRowData.position[atom1];
695 + #else
696 +    d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
697 + #endif
698 +
699 +    snap_->wrapVector(d);
700 +    return d;    
701 +  }
702 +  
703 +  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
704 +    Vector3d d;
705 +    
706 + #ifdef IS_MPI
707 +    d = cgColData.position[cg2] - atomColData.position[atom2];
708 + #else
709 +    d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
710 + #endif
711 +    
712 +    snap_->wrapVector(d);
713 +    return d;    
714 +  }
715 +
716 +  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
717 + #ifdef IS_MPI
718 +    return massFactorsRow[atom1];
719 + #else
720 +    return massFactors[atom1];
721 + #endif
722 +  }
723 +
724 +  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
725 + #ifdef IS_MPI
726 +    return massFactorsCol[atom2];
727 + #else
728 +    return massFactors[atom2];
729 + #endif
730 +
731 +  }
732 +    
733 +  Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
734 +    Vector3d d;
735 +    
736 + #ifdef IS_MPI
737 +    d = atomColData.position[atom2] - atomRowData.position[atom1];
738 + #else
739 +    d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
740 + #endif
741 +
742 +    snap_->wrapVector(d);
743 +    return d;    
744 +  }
745 +
746 +  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
747 +    return excludesForAtom[atom1];
748 +  }
749 +
750 +  /**
751 +   * We need to exclude some overcounted interactions that result from
752 +   * the parallel decomposition.
753 +   */
754 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
755 +    int unique_id_1, unique_id_2;
756 +
757 + #ifdef IS_MPI
758 +    // in MPI, we have to look up the unique IDs for each atom
759 +    unique_id_1 = AtomRowToGlobal[atom1];
760 +    unique_id_2 = AtomColToGlobal[atom2];
761 +
762 +    // this situation should only arise in MPI simulations
763 +    if (unique_id_1 == unique_id_2) return true;
764 +    
765 +    // this prevents us from doing the pair on multiple processors
766 +    if (unique_id_1 < unique_id_2) {
767 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
768 +    } else {
769 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
770 +    }
771 + #endif
772 +    return false;
773 +  }
774 +
775 +  /**
776 +   * We need to handle the interactions for atoms who are involved in
777 +   * the same rigid body as well as some short range interactions
778 +   * (bonds, bends, torsions) differently from other interactions.
779 +   * We'll still visit the pairwise routines, but with a flag that
780 +   * tells those routines to exclude the pair from direct long range
781 +   * interactions.  Some indirect interactions (notably reaction
782 +   * field) must still be handled for these pairs.
783 +   */
784 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
785 +    int unique_id_2;
786 +    
787 + #ifdef IS_MPI
788 +    // in MPI, we have to look up the unique IDs for the row atom.
789 +    unique_id_2 = AtomColToGlobal[atom2];
790 + #else
791 +    // in the normal loop, the atom numbers are unique
792 +    unique_id_2 = atom2;
793 + #endif
794 +    
795 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
796 +         i != excludesForAtom[atom1].end(); ++i) {
797 +      if ( (*i) == unique_id_2 ) return true;
798 +    }
799 +
800 +    return false;
801 +  }
802 +
803 +
804 +  void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
805 + #ifdef IS_MPI
806 +    atomRowData.force[atom1] += fg;
807 + #else
808 +    snap_->atomData.force[atom1] += fg;
809 + #endif
810 +  }
811 +
812 +  void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
813 + #ifdef IS_MPI
814 +    atomColData.force[atom2] += fg;
815 + #else
816 +    snap_->atomData.force[atom2] += fg;
817 + #endif
818 +  }
819 +
820 +    // filling interaction blocks with pointers
821 +  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
822 +                                                     int atom1, int atom2) {
823 +
824 +    idat.excluded = excludeAtomPair(atom1, atom2);
825 +  
826 + #ifdef IS_MPI
827 +    
828 +    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
829 +                             ff_->getAtomType(identsCol[atom2]) );
830 +    
831 +    if (storageLayout_ & DataStorage::dslAmat) {
832 +      idat.A1 = &(atomRowData.aMat[atom1]);
833 +      idat.A2 = &(atomColData.aMat[atom2]);
834 +    }
835 +    
836 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
837 +      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
838 +      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
839 +    }
840 +
841 +    if (storageLayout_ & DataStorage::dslTorque) {
842 +      idat.t1 = &(atomRowData.torque[atom1]);
843 +      idat.t2 = &(atomColData.torque[atom2]);
844 +    }
845 +
846 +    if (storageLayout_ & DataStorage::dslDensity) {
847 +      idat.rho1 = &(atomRowData.density[atom1]);
848 +      idat.rho2 = &(atomColData.density[atom2]);
849 +    }
850 +
851 +    if (storageLayout_ & DataStorage::dslFunctional) {
852 +      idat.frho1 = &(atomRowData.functional[atom1]);
853 +      idat.frho2 = &(atomColData.functional[atom2]);
854 +    }
855 +
856 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
857 +      idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
858 +      idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
859 +    }
860 +
861 +    if (storageLayout_ & DataStorage::dslParticlePot) {
862 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
863 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
864 +    }
865 +
866 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
867 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
868 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
869 +    }
870 +
871 + #else
872 +
873 +    idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
874 +                             ff_->getAtomType(idents[atom2]) );
875 +
876 +    if (storageLayout_ & DataStorage::dslAmat) {
877 +      idat.A1 = &(snap_->atomData.aMat[atom1]);
878 +      idat.A2 = &(snap_->atomData.aMat[atom2]);
879 +    }
880 +
881 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
882 +      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
883 +      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
884 +    }
885 +
886 +    if (storageLayout_ & DataStorage::dslTorque) {
887 +      idat.t1 = &(snap_->atomData.torque[atom1]);
888 +      idat.t2 = &(snap_->atomData.torque[atom2]);
889 +    }
890 +
891 +    if (storageLayout_ & DataStorage::dslDensity) {    
892 +      idat.rho1 = &(snap_->atomData.density[atom1]);
893 +      idat.rho2 = &(snap_->atomData.density[atom2]);
894 +    }
895 +
896 +    if (storageLayout_ & DataStorage::dslFunctional) {
897 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
898 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
899 +    }
900 +
901 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
902 +      idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
903 +      idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
904 +    }
905 +
906 +    if (storageLayout_ & DataStorage::dslParticlePot) {
907 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
908 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
909 +    }
910 +
911 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
912 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
913 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
914 +    }
915 + #endif
916 +  }
917 +
918 +  
919 +  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
920 + #ifdef IS_MPI
921 +    pot_row[atom1] += 0.5 *  *(idat.pot);
922 +    pot_col[atom2] += 0.5 *  *(idat.pot);
923 +
924 +    atomRowData.force[atom1] += *(idat.f1);
925 +    atomColData.force[atom2] -= *(idat.f1);
926 + #else
927 +    pairwisePot += *(idat.pot);
928 +
929 +    snap_->atomData.force[atom1] += *(idat.f1);
930 +    snap_->atomData.force[atom2] -= *(idat.f1);
931 + #endif
932 +    
933 +  }
934 +
935 +  /*
936 +   * buildNeighborList
937 +   *
938 +   * first element of pair is row-indexed CutoffGroup
939 +   * second element of pair is column-indexed CutoffGroup
940 +   */
941 +  vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
942 +      
943 +    vector<pair<int, int> > neighborList;
944 +    groupCutoffs cuts;
945 +    bool doAllPairs = false;
946 +
947 + #ifdef IS_MPI
948 +    cellListRow_.clear();
949 +    cellListCol_.clear();
950 + #else
951 +    cellList_.clear();
952 + #endif
953 +
954 +    RealType rList_ = (largestRcut_ + skinThickness_);
955 +    RealType rl2 = rList_ * rList_;
956 +    Snapshot* snap_ = sman_->getCurrentSnapshot();
957 +    Mat3x3d Hmat = snap_->getHmat();
958 +    Vector3d Hx = Hmat.getColumn(0);
959 +    Vector3d Hy = Hmat.getColumn(1);
960 +    Vector3d Hz = Hmat.getColumn(2);
961 +
962 +    nCells_.x() = (int) ( Hx.length() )/ rList_;
963 +    nCells_.y() = (int) ( Hy.length() )/ rList_;
964 +    nCells_.z() = (int) ( Hz.length() )/ rList_;
965 +
966 +    // handle small boxes where the cell offsets can end up repeating cells
967 +    
968 +    if (nCells_.x() < 3) doAllPairs = true;
969 +    if (nCells_.y() < 3) doAllPairs = true;
970 +    if (nCells_.z() < 3) doAllPairs = true;
971 +
972 +    Mat3x3d invHmat = snap_->getInvHmat();
973 +    Vector3d rs, scaled, dr;
974 +    Vector3i whichCell;
975 +    int cellIndex;
976 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
977 +
978 + #ifdef IS_MPI
979 +    cellListRow_.resize(nCtot);
980 +    cellListCol_.resize(nCtot);
981 + #else
982 +    cellList_.resize(nCtot);
983 + #endif
984 +
985 +    if (!doAllPairs) {
986 + #ifdef IS_MPI
987 +
988 +      for (int i = 0; i < nGroupsInRow_; i++) {
989 +        rs = cgRowData.position[i];
990 +        
991 +        // scaled positions relative to the box vectors
992 +        scaled = invHmat * rs;
993 +        
994 +        // wrap the vector back into the unit box by subtracting integer box
995 +        // numbers
996 +        for (int j = 0; j < 3; j++) {
997 +          scaled[j] -= roundMe(scaled[j]);
998 +          scaled[j] += 0.5;
999 +        }
1000 +        
1001 +        // find xyz-indices of cell that cutoffGroup is in.
1002 +        whichCell.x() = nCells_.x() * scaled.x();
1003 +        whichCell.y() = nCells_.y() * scaled.y();
1004 +        whichCell.z() = nCells_.z() * scaled.z();
1005 +        
1006 +        // find single index of this cell:
1007 +        cellIndex = Vlinear(whichCell, nCells_);
1008 +        
1009 +        // add this cutoff group to the list of groups in this cell;
1010 +        cellListRow_[cellIndex].push_back(i);
1011 +      }
1012 +      
1013 +      for (int i = 0; i < nGroupsInCol_; i++) {
1014 +        rs = cgColData.position[i];
1015 +        
1016 +        // scaled positions relative to the box vectors
1017 +        scaled = invHmat * rs;
1018 +        
1019 +        // wrap the vector back into the unit box by subtracting integer box
1020 +        // numbers
1021 +        for (int j = 0; j < 3; j++) {
1022 +          scaled[j] -= roundMe(scaled[j]);
1023 +          scaled[j] += 0.5;
1024 +        }
1025 +        
1026 +        // find xyz-indices of cell that cutoffGroup is in.
1027 +        whichCell.x() = nCells_.x() * scaled.x();
1028 +        whichCell.y() = nCells_.y() * scaled.y();
1029 +        whichCell.z() = nCells_.z() * scaled.z();
1030 +        
1031 +        // find single index of this cell:
1032 +        cellIndex = Vlinear(whichCell, nCells_);
1033 +        
1034 +        // add this cutoff group to the list of groups in this cell;
1035 +        cellListCol_[cellIndex].push_back(i);
1036 +      }
1037 + #else
1038 +      for (int i = 0; i < nGroups_; i++) {
1039 +        rs = snap_->cgData.position[i];
1040 +        
1041 +        // scaled positions relative to the box vectors
1042 +        scaled = invHmat * rs;
1043 +        
1044 +        // wrap the vector back into the unit box by subtracting integer box
1045 +        // numbers
1046 +        for (int j = 0; j < 3; j++) {
1047 +          scaled[j] -= roundMe(scaled[j]);
1048 +          scaled[j] += 0.5;
1049 +        }
1050 +        
1051 +        // find xyz-indices of cell that cutoffGroup is in.
1052 +        whichCell.x() = nCells_.x() * scaled.x();
1053 +        whichCell.y() = nCells_.y() * scaled.y();
1054 +        whichCell.z() = nCells_.z() * scaled.z();
1055 +        
1056 +        // find single index of this cell:
1057 +        cellIndex = Vlinear(whichCell, nCells_);      
1058 +        
1059 +        // add this cutoff group to the list of groups in this cell;
1060 +        cellList_[cellIndex].push_back(i);
1061 +      }
1062 + #endif
1063 +
1064 +      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1065 +        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1066 +          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1067 +            Vector3i m1v(m1x, m1y, m1z);
1068 +            int m1 = Vlinear(m1v, nCells_);
1069 +            
1070 +            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1071 +                 os != cellOffsets_.end(); ++os) {
1072 +              
1073 +              Vector3i m2v = m1v + (*os);
1074 +              
1075 +              if (m2v.x() >= nCells_.x()) {
1076 +                m2v.x() = 0;          
1077 +              } else if (m2v.x() < 0) {
1078 +                m2v.x() = nCells_.x() - 1;
1079 +              }
1080 +              
1081 +              if (m2v.y() >= nCells_.y()) {
1082 +                m2v.y() = 0;          
1083 +              } else if (m2v.y() < 0) {
1084 +                m2v.y() = nCells_.y() - 1;
1085 +              }
1086 +              
1087 +              if (m2v.z() >= nCells_.z()) {
1088 +                m2v.z() = 0;          
1089 +              } else if (m2v.z() < 0) {
1090 +                m2v.z() = nCells_.z() - 1;
1091 +              }
1092 +              
1093 +              int m2 = Vlinear (m2v, nCells_);
1094 +              
1095 + #ifdef IS_MPI
1096 +              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1097 +                   j1 != cellListRow_[m1].end(); ++j1) {
1098 +                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1099 +                     j2 != cellListCol_[m2].end(); ++j2) {
1100 +                  
1101 +                  // Always do this if we're in different cells or if
1102 +                  // we're in the same cell and the global index of the
1103 +                  // j2 cutoff group is less than the j1 cutoff group
1104 +                  
1105 +                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1106 +                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1107 +                    snap_->wrapVector(dr);
1108 +                    cuts = getGroupCutoffs( (*j1), (*j2) );
1109 +                    if (dr.lengthSquare() < cuts.third) {
1110 +                      neighborList.push_back(make_pair((*j1), (*j2)));
1111 +                    }
1112 +                  }
1113 +                }
1114 +              }
1115 + #else
1116 +              
1117 +              for (vector<int>::iterator j1 = cellList_[m1].begin();
1118 +                   j1 != cellList_[m1].end(); ++j1) {
1119 +                for (vector<int>::iterator j2 = cellList_[m2].begin();
1120 +                     j2 != cellList_[m2].end(); ++j2) {
1121 +                  
1122 +                  // Always do this if we're in different cells or if
1123 +                  // we're in the same cell and the global index of the
1124 +                  // j2 cutoff group is less than the j1 cutoff group
1125 +                  
1126 +                  if (m2 != m1 || (*j2) < (*j1)) {
1127 +                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1128 +                    snap_->wrapVector(dr);
1129 +                    cuts = getGroupCutoffs( (*j1), (*j2) );
1130 +                    if (dr.lengthSquare() < cuts.third) {
1131 +                      neighborList.push_back(make_pair((*j1), (*j2)));
1132 +                    }
1133 +                  }
1134 +                }
1135 +              }
1136 + #endif
1137 +            }
1138 +          }
1139 +        }
1140 +      }
1141 +    } else {
1142 +      // branch to do all cutoff group pairs
1143 + #ifdef IS_MPI
1144 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1145 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1146 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1147 +          snap_->wrapVector(dr);
1148 +          cuts = getGroupCutoffs( j1, j2 );
1149 +          if (dr.lengthSquare() < cuts.third) {
1150 +            neighborList.push_back(make_pair(j1, j2));
1151 +          }
1152 +        }
1153 +      }
1154 + #else
1155 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1156 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1157 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1158 +          snap_->wrapVector(dr);
1159 +          cuts = getGroupCutoffs( j1, j2 );
1160 +          if (dr.lengthSquare() < cuts.third) {
1161 +            neighborList.push_back(make_pair(j1, j2));
1162 +          }
1163 +        }
1164 +      }        
1165 + #endif
1166 +    }
1167 +      
1168 +    // save the local cutoff group positions for the check that is
1169 +    // done on each loop:
1170 +    saved_CG_positions_.clear();
1171 +    for (int i = 0; i < nGroups_; i++)
1172 +      saved_CG_positions_.push_back(snap_->cgData.position[i]);
1173 +    
1174 +    return neighborList;
1175 +  }
1176 + } //end namespace OpenMD

Diff Legend

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