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 1589 by gezelter, Sun Jul 10 16:05:34 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 >    // 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 < //____ MPITypeTraits
126 < template<typename T>
127 < struct MPITypeTraits;
125 >    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
126 >    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
127 >    
128 >    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
129 >    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
130  
131 < #ifdef IS_MPI
132 < template<>
79 < struct MPITypeTraits<RealType> {
80 <  static const MPI::Datatype datatype;
81 < };
82 < const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL;
131 >    AtomCommRealRow->gather(massFactors, massFactorsRow);
132 >    AtomCommRealColumn->gather(massFactors, massFactorsCol);
133  
134 < template<>
135 < struct MPITypeTraits<int> {
136 <  static const MPI::Datatype datatype;
137 < };
138 < const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT;
139 < #endif
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 < /**
146 < * Constructor for ForceDecomposition Parallel Decomposition Method
147 < * Will try to construct a symmetric grid of processors. Ideally, the
148 < * number of processors will be a square ex: 4, 9, 16, 25.
149 < *
150 < */
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 < ForceDecomposition::ForceDecomposition() {
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 < #ifdef IS_MPI
166 <  int nProcs = MPI::COMM_WORLD.Get_size();
167 <  int worldRank = MPI::COMM_WORLD.Get_rank();
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 <  // First time through, construct column stride.
191 <  if (communicators.size() == 0)
192 <  {
193 <    int nColumnsMax = (int) round(sqrt((float) nProcs));
194 <    for (int i = 0; i < nProcs; ++i)
195 <    {
196 <      if (nProcs%i==0) nColumns=i;
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 <    int nRows = nProcs/nColumns;    
203 <    myRank_ = (int) worldRank%nColumns;
204 <  }
205 <  else
206 <  {
207 <    myRank_ = myRank/nColumns;
120 <  }
121 <  MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0);
122 <  
123 <  isColumn_ = false;
124 <  
125 < }
202 >    excludesForAtom.clear();
203 >    excludesForAtom.resize(nLocal_);
204 >    toposForAtom.clear();
205 >    toposForAtom.resize(nLocal_);
206 >    topoDist.clear();
207 >    topoDist.resize(nLocal_);
208  
209 < ForceDecomposition::gather(sendbuf, receivebuf){
210 <  communicators(myIndex_).Allgatherv();
129 < }
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 < ForceDecomposition::scatter(sbuffer, rbuffer){
238 <  communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM);
239 < }
237 >  }
238 >  
239 >  void ForceMatrixDecomposition::createGtypeCutoffMap() {
240 >    
241 >    RealType tol = 1e-6;
242 >    RealType rc;
243 >    int atid;
244 >    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
245 >    map<int, RealType> atypeCutoff;
246 >      
247 >    for (set<AtomType*>::iterator at = atypes.begin();
248 >         at != atypes.end(); ++at){
249 >      atid = (*at)->getIdent();
250 >      if (userChoseCutoff_)
251 >        atypeCutoff[atid] = userCutoff_;
252 >      else
253 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
254 >    }
255  
256 +    vector<RealType> gTypeCutoffs;
257 +    // first we do a single loop over the cutoff groups to find the
258 +    // largest cutoff for any atypes present in this group.
259 + #ifdef IS_MPI
260 +    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
261 +    groupRowToGtype.resize(nGroupsInRow_);
262 +    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
263 +      vector<int> atomListRow = getAtomsInGroupRow(cg1);
264 +      for (vector<int>::iterator ia = atomListRow.begin();
265 +           ia != atomListRow.end(); ++ia) {            
266 +        int atom1 = (*ia);
267 +        atid = identsRow[atom1];
268 +        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
269 +          groupCutoffRow[cg1] = atypeCutoff[atid];
270 +        }
271 +      }
272  
273 +      bool gTypeFound = false;
274 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
275 +        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
276 +          groupRowToGtype[cg1] = gt;
277 +          gTypeFound = true;
278 +        }
279 +      }
280 +      if (!gTypeFound) {
281 +        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
282 +        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
283 +      }
284 +      
285 +    }
286 +    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
287 +    groupColToGtype.resize(nGroupsInCol_);
288 +    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
289 +      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
290 +      for (vector<int>::iterator jb = atomListCol.begin();
291 +           jb != atomListCol.end(); ++jb) {            
292 +        int atom2 = (*jb);
293 +        atid = identsCol[atom2];
294 +        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
295 +          groupCutoffCol[cg2] = atypeCutoff[atid];
296 +        }
297 +      }
298 +      bool gTypeFound = false;
299 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
300 +        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
301 +          groupColToGtype[cg2] = gt;
302 +          gTypeFound = true;
303 +        }
304 +      }
305 +      if (!gTypeFound) {
306 +        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
307 +        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
308 +      }
309 +    }
310 + #else
311 +
312 +    vector<RealType> groupCutoff(nGroups_, 0.0);
313 +    groupToGtype.resize(nGroups_);
314 +    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
315 +
316 +      groupCutoff[cg1] = 0.0;
317 +      vector<int> atomList = getAtomsInGroupRow(cg1);
318 +
319 +      for (vector<int>::iterator ia = atomList.begin();
320 +           ia != atomList.end(); ++ia) {            
321 +        int atom1 = (*ia);
322 +        atid = idents[atom1];
323 +        if (atypeCutoff[atid] > groupCutoff[cg1]) {
324 +          groupCutoff[cg1] = atypeCutoff[atid];
325 +        }
326 +      }
327 +
328 +      bool gTypeFound = false;
329 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
330 +        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
331 +          groupToGtype[cg1] = gt;
332 +          gTypeFound = true;
333 +        }
334 +      }
335 +      if (!gTypeFound) {
336 +        gTypeCutoffs.push_back( groupCutoff[cg1] );
337 +        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
338 +      }      
339 +    }
340 + #endif
341 +
342 +    // Now we find the maximum group cutoff value present in the simulation
343 +
344 +    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
345 +
346 + #ifdef IS_MPI
347 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
348 + #endif
349 +    
350 +    RealType tradRcut = groupMax;
351 +
352 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
353 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
354 +        RealType thisRcut;
355 +        switch(cutoffPolicy_) {
356 +        case TRADITIONAL:
357 +          thisRcut = tradRcut;
358 +          break;
359 +        case MIX:
360 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
361 +          break;
362 +        case MAX:
363 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
364 +          break;
365 +        default:
366 +          sprintf(painCave.errMsg,
367 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
368 +                  "hit an unknown cutoff policy!\n");
369 +          painCave.severity = OPENMD_ERROR;
370 +          painCave.isFatal = 1;
371 +          simError();
372 +          break;
373 +        }
374 +
375 +        pair<int,int> key = make_pair(i,j);
376 +        gTypeCutoffMap[key].first = thisRcut;
377 +
378 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
379 +
380 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
381 +        
382 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
383 +
384 +        // sanity check
385 +        
386 +        if (userChoseCutoff_) {
387 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
388 +            sprintf(painCave.errMsg,
389 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
390 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
391 +            painCave.severity = OPENMD_ERROR;
392 +            painCave.isFatal = 1;
393 +            simError();            
394 +          }
395 +        }
396 +      }
397 +    }
398 +  }
399 +
400 +
401 +  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
402 +    int i, j;  
403 + #ifdef IS_MPI
404 +    i = groupRowToGtype[cg1];
405 +    j = groupColToGtype[cg2];
406 + #else
407 +    i = groupToGtype[cg1];
408 +    j = groupToGtype[cg2];
409 + #endif    
410 +    return gTypeCutoffMap[make_pair(i,j)];
411 +  }
412 +
413 +  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
414 +    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
415 +      if (toposForAtom[atom1][j] == atom2)
416 +        return topoDist[atom1][j];
417 +    }
418 +    return 0;
419 +  }
420 +
421 +  void ForceMatrixDecomposition::zeroWorkArrays() {
422 +    pairwisePot = 0.0;
423 +    embeddingPot = 0.0;
424 +
425 + #ifdef IS_MPI
426 +    if (storageLayout_ & DataStorage::dslForce) {
427 +      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
428 +      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
429 +    }
430 +
431 +    if (storageLayout_ & DataStorage::dslTorque) {
432 +      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
433 +      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
434 +    }
435 +    
436 +    fill(pot_row.begin(), pot_row.end(),
437 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
438 +
439 +    fill(pot_col.begin(), pot_col.end(),
440 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
441 +
442 +    if (storageLayout_ & DataStorage::dslParticlePot) {    
443 +      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
444 +      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
445 +    }
446 +
447 +    if (storageLayout_ & DataStorage::dslDensity) {      
448 +      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
449 +      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
450 +    }
451 +
452 +    if (storageLayout_ & DataStorage::dslFunctional) {  
453 +      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
454 +      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
455 +    }
456 +
457 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
458 +      fill(atomRowData.functionalDerivative.begin(),
459 +           atomRowData.functionalDerivative.end(), 0.0);
460 +      fill(atomColData.functionalDerivative.begin(),
461 +           atomColData.functionalDerivative.end(), 0.0);
462 +    }
463 +
464 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
465 +      fill(atomRowData.skippedCharge.begin(),
466 +           atomRowData.skippedCharge.end(), 0.0);
467 +      fill(atomColData.skippedCharge.begin(),
468 +           atomColData.skippedCharge.end(), 0.0);
469 +    }
470 +
471 + #else
472 +    
473 +    if (storageLayout_ & DataStorage::dslParticlePot) {      
474 +      fill(snap_->atomData.particlePot.begin(),
475 +           snap_->atomData.particlePot.end(), 0.0);
476 +    }
477 +    
478 +    if (storageLayout_ & DataStorage::dslDensity) {      
479 +      fill(snap_->atomData.density.begin(),
480 +           snap_->atomData.density.end(), 0.0);
481 +    }
482 +    if (storageLayout_ & DataStorage::dslFunctional) {
483 +      fill(snap_->atomData.functional.begin(),
484 +           snap_->atomData.functional.end(), 0.0);
485 +    }
486 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
487 +      fill(snap_->atomData.functionalDerivative.begin(),
488 +           snap_->atomData.functionalDerivative.end(), 0.0);
489 +    }
490 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
491 +      fill(snap_->atomData.skippedCharge.begin(),
492 +           snap_->atomData.skippedCharge.end(), 0.0);
493 +    }
494 + #endif
495 +    
496 +  }
497 +
498 +
499 +  void ForceMatrixDecomposition::distributeData()  {
500 +    snap_ = sman_->getCurrentSnapshot();
501 +    storageLayout_ = sman_->getStorageLayout();
502 + #ifdef IS_MPI
503 +    
504 +    // gather up the atomic positions
505 +    AtomCommVectorRow->gather(snap_->atomData.position,
506 +                              atomRowData.position);
507 +    AtomCommVectorColumn->gather(snap_->atomData.position,
508 +                                 atomColData.position);
509 +    
510 +    // gather up the cutoff group positions
511 +    cgCommVectorRow->gather(snap_->cgData.position,
512 +                            cgRowData.position);
513 +    cgCommVectorColumn->gather(snap_->cgData.position,
514 +                               cgColData.position);
515 +    
516 +    // if needed, gather the atomic rotation matrices
517 +    if (storageLayout_ & DataStorage::dslAmat) {
518 +      AtomCommMatrixRow->gather(snap_->atomData.aMat,
519 +                                atomRowData.aMat);
520 +      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
521 +                                   atomColData.aMat);
522 +    }
523 +    
524 +    // if needed, gather the atomic eletrostatic frames
525 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
526 +      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
527 +                                atomRowData.electroFrame);
528 +      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
529 +                                   atomColData.electroFrame);
530 +    }
531 + #endif      
532 +  }
533 +  
534 +  /* collects information obtained during the pre-pair loop onto local
535 +   * data structures.
536 +   */
537 +  void ForceMatrixDecomposition::collectIntermediateData() {
538 +    snap_ = sman_->getCurrentSnapshot();
539 +    storageLayout_ = sman_->getStorageLayout();
540 + #ifdef IS_MPI
541 +    
542 +    if (storageLayout_ & DataStorage::dslDensity) {
543 +      
544 +      AtomCommRealRow->scatter(atomRowData.density,
545 +                               snap_->atomData.density);
546 +      
547 +      int n = snap_->atomData.density.size();
548 +      vector<RealType> rho_tmp(n, 0.0);
549 +      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
550 +      for (int i = 0; i < n; i++)
551 +        snap_->atomData.density[i] += rho_tmp[i];
552 +    }
553 + #endif
554 +  }
555 +
556 +  /*
557 +   * redistributes information obtained during the pre-pair loop out to
558 +   * row and column-indexed data structures
559 +   */
560 +  void ForceMatrixDecomposition::distributeIntermediateData() {
561 +    snap_ = sman_->getCurrentSnapshot();
562 +    storageLayout_ = sman_->getStorageLayout();
563 + #ifdef IS_MPI
564 +    if (storageLayout_ & DataStorage::dslFunctional) {
565 +      AtomCommRealRow->gather(snap_->atomData.functional,
566 +                              atomRowData.functional);
567 +      AtomCommRealColumn->gather(snap_->atomData.functional,
568 +                                 atomColData.functional);
569 +    }
570 +    
571 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
572 +      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
573 +                              atomRowData.functionalDerivative);
574 +      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
575 +                                 atomColData.functionalDerivative);
576 +    }
577 + #endif
578 +  }
579 +  
580 +  
581 +  void ForceMatrixDecomposition::collectData() {
582 +    snap_ = sman_->getCurrentSnapshot();
583 +    storageLayout_ = sman_->getStorageLayout();
584 + #ifdef IS_MPI    
585 +    int n = snap_->atomData.force.size();
586 +    vector<Vector3d> frc_tmp(n, V3Zero);
587 +    
588 +    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
589 +    for (int i = 0; i < n; i++) {
590 +      snap_->atomData.force[i] += frc_tmp[i];
591 +      frc_tmp[i] = 0.0;
592 +    }
593 +    
594 +    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
595 +    for (int i = 0; i < n; i++)
596 +      snap_->atomData.force[i] += frc_tmp[i];
597 +    
598 +    
599 +    if (storageLayout_ & DataStorage::dslTorque) {
600 +
601 +      int nt = snap_->atomData.torque.size();
602 +      vector<Vector3d> trq_tmp(nt, V3Zero);
603 +
604 +      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
605 +      for (int i = 0; i < nt; i++) {
606 +        snap_->atomData.torque[i] += trq_tmp[i];
607 +        trq_tmp[i] = 0.0;
608 +      }
609 +      
610 +      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
611 +      for (int i = 0; i < nt; i++)
612 +        snap_->atomData.torque[i] += trq_tmp[i];
613 +    }
614 +
615 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
616 +
617 +      int ns = snap_->atomData.skippedCharge.size();
618 +      vector<RealType> skch_tmp(ns, 0.0);
619 +
620 +      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
621 +      for (int i = 0; i < ns; i++) {
622 +        snap_->atomData.skippedCharge[i] = skch_tmp[i];
623 +        skch_tmp[i] = 0.0;
624 +      }
625 +      
626 +      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
627 +      for (int i = 0; i < ns; i++)
628 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
629 +    }
630 +    
631 +    nLocal_ = snap_->getNumberOfAtoms();
632 +
633 +    vector<potVec> pot_temp(nLocal_,
634 +                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
635 +
636 +    // scatter/gather pot_row into the members of my column
637 +          
638 +    AtomCommPotRow->scatter(pot_row, pot_temp);
639 +
640 +    for (int ii = 0;  ii < pot_temp.size(); ii++ )
641 +      pairwisePot += pot_temp[ii];
642 +    
643 +    fill(pot_temp.begin(), pot_temp.end(),
644 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
645 +      
646 +    AtomCommPotColumn->scatter(pot_col, pot_temp);    
647 +    
648 +    for (int ii = 0;  ii < pot_temp.size(); ii++ )
649 +      pairwisePot += pot_temp[ii];    
650 + #endif
651 +
652 +  }
653 +
654 +  int ForceMatrixDecomposition::getNAtomsInRow() {  
655 + #ifdef IS_MPI
656 +    return nAtomsInRow_;
657 + #else
658 +    return nLocal_;
659 + #endif
660 +  }
661 +
662 +  /**
663 +   * returns the list of atoms belonging to this group.  
664 +   */
665 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
666 + #ifdef IS_MPI
667 +    return groupListRow_[cg1];
668 + #else
669 +    return groupList_[cg1];
670 + #endif
671 +  }
672 +
673 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
674 + #ifdef IS_MPI
675 +    return groupListCol_[cg2];
676 + #else
677 +    return groupList_[cg2];
678 + #endif
679 +  }
680 +  
681 +  Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
682 +    Vector3d d;
683 +    
684 + #ifdef IS_MPI
685 +    d = cgColData.position[cg2] - cgRowData.position[cg1];
686 + #else
687 +    d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
688 + #endif
689 +    
690 +    snap_->wrapVector(d);
691 +    return d;    
692 +  }
693 +
694 +
695 +  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
696 +
697 +    Vector3d d;
698 +    
699 + #ifdef IS_MPI
700 +    d = cgRowData.position[cg1] - atomRowData.position[atom1];
701 + #else
702 +    d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
703 + #endif
704 +
705 +    snap_->wrapVector(d);
706 +    return d;    
707 +  }
708 +  
709 +  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
710 +    Vector3d d;
711 +    
712 + #ifdef IS_MPI
713 +    d = cgColData.position[cg2] - atomColData.position[atom2];
714 + #else
715 +    d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
716 + #endif
717 +    
718 +    snap_->wrapVector(d);
719 +    return d;    
720 +  }
721 +
722 +  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
723 + #ifdef IS_MPI
724 +    return massFactorsRow[atom1];
725 + #else
726 +    return massFactors[atom1];
727 + #endif
728 +  }
729 +
730 +  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
731 + #ifdef IS_MPI
732 +    return massFactorsCol[atom2];
733 + #else
734 +    return massFactors[atom2];
735 + #endif
736 +
737 +  }
738 +    
739 +  Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
740 +    Vector3d d;
741 +    
742 + #ifdef IS_MPI
743 +    d = atomColData.position[atom2] - atomRowData.position[atom1];
744 + #else
745 +    d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
746 + #endif
747 +
748 +    snap_->wrapVector(d);
749 +    return d;    
750 +  }
751 +
752 +  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
753 +    return excludesForAtom[atom1];
754 +  }
755 +
756 +  /**
757 +   * We need to exclude some overcounted interactions that result from
758 +   * the parallel decomposition.
759 +   */
760 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
761 +    int unique_id_1, unique_id_2;
762 +
763 + #ifdef IS_MPI
764 +    // in MPI, we have to look up the unique IDs for each atom
765 +    unique_id_1 = AtomRowToGlobal[atom1];
766 +    unique_id_2 = AtomColToGlobal[atom2];
767 +
768 +    // this situation should only arise in MPI simulations
769 +    if (unique_id_1 == unique_id_2) return true;
770 +    
771 +    // this prevents us from doing the pair on multiple processors
772 +    if (unique_id_1 < unique_id_2) {
773 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
774 +    } else {
775 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
776 +    }
777 + #endif
778 +    return false;
779 +  }
780 +
781 +  /**
782 +   * We need to handle the interactions for atoms who are involved in
783 +   * the same rigid body as well as some short range interactions
784 +   * (bonds, bends, torsions) differently from other interactions.
785 +   * We'll still visit the pairwise routines, but with a flag that
786 +   * tells those routines to exclude the pair from direct long range
787 +   * interactions.  Some indirect interactions (notably reaction
788 +   * field) must still be handled for these pairs.
789 +   */
790 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
791 +    int unique_id_2;
792 +    
793 + #ifdef IS_MPI
794 +    // in MPI, we have to look up the unique IDs for the row atom.
795 +    unique_id_2 = AtomColToGlobal[atom2];
796 + #else
797 +    // in the normal loop, the atom numbers are unique
798 +    unique_id_2 = atom2;
799 + #endif
800 +    
801 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
802 +         i != excludesForAtom[atom1].end(); ++i) {
803 +      if ( (*i) == unique_id_2 ) return true;
804 +    }
805 +
806 +    return false;
807 +  }
808 +
809 +
810 +  void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
811 + #ifdef IS_MPI
812 +    atomRowData.force[atom1] += fg;
813 + #else
814 +    snap_->atomData.force[atom1] += fg;
815 + #endif
816 +  }
817 +
818 +  void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
819 + #ifdef IS_MPI
820 +    atomColData.force[atom2] += fg;
821 + #else
822 +    snap_->atomData.force[atom2] += fg;
823 + #endif
824 +  }
825 +
826 +    // filling interaction blocks with pointers
827 +  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
828 +                                                     int atom1, int atom2) {
829 +
830 +    idat.excluded = excludeAtomPair(atom1, atom2);
831 +  
832 + #ifdef IS_MPI
833 +    
834 +    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
835 +                             ff_->getAtomType(identsCol[atom2]) );
836 +    
837 +    if (storageLayout_ & DataStorage::dslAmat) {
838 +      idat.A1 = &(atomRowData.aMat[atom1]);
839 +      idat.A2 = &(atomColData.aMat[atom2]);
840 +    }
841 +    
842 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
843 +      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
844 +      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
845 +    }
846 +
847 +    if (storageLayout_ & DataStorage::dslTorque) {
848 +      idat.t1 = &(atomRowData.torque[atom1]);
849 +      idat.t2 = &(atomColData.torque[atom2]);
850 +    }
851 +
852 +    if (storageLayout_ & DataStorage::dslDensity) {
853 +      idat.rho1 = &(atomRowData.density[atom1]);
854 +      idat.rho2 = &(atomColData.density[atom2]);
855 +    }
856 +
857 +    if (storageLayout_ & DataStorage::dslFunctional) {
858 +      idat.frho1 = &(atomRowData.functional[atom1]);
859 +      idat.frho2 = &(atomColData.functional[atom2]);
860 +    }
861 +
862 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
863 +      idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
864 +      idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
865 +    }
866 +
867 +    if (storageLayout_ & DataStorage::dslParticlePot) {
868 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
869 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
870 +    }
871 +
872 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
873 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
874 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
875 +    }
876 +
877 + #else
878 +
879 +    idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
880 +                             ff_->getAtomType(idents[atom2]) );
881 +
882 +    if (storageLayout_ & DataStorage::dslAmat) {
883 +      idat.A1 = &(snap_->atomData.aMat[atom1]);
884 +      idat.A2 = &(snap_->atomData.aMat[atom2]);
885 +    }
886 +
887 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
888 +      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
889 +      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
890 +    }
891 +
892 +    if (storageLayout_ & DataStorage::dslTorque) {
893 +      idat.t1 = &(snap_->atomData.torque[atom1]);
894 +      idat.t2 = &(snap_->atomData.torque[atom2]);
895 +    }
896 +
897 +    if (storageLayout_ & DataStorage::dslDensity) {    
898 +      idat.rho1 = &(snap_->atomData.density[atom1]);
899 +      idat.rho2 = &(snap_->atomData.density[atom2]);
900 +    }
901 +
902 +    if (storageLayout_ & DataStorage::dslFunctional) {
903 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
904 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
905 +    }
906 +
907 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
908 +      idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
909 +      idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
910 +    }
911 +
912 +    if (storageLayout_ & DataStorage::dslParticlePot) {
913 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
914 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
915 +    }
916 +
917 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
918 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
919 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
920 +    }
921 + #endif
922 +  }
923 +
924 +  
925 +  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
926 + #ifdef IS_MPI
927 +    pot_row[atom1] += 0.5 *  *(idat.pot);
928 +    pot_col[atom2] += 0.5 *  *(idat.pot);
929 +
930 +    atomRowData.force[atom1] += *(idat.f1);
931 +    atomColData.force[atom2] -= *(idat.f1);
932 + #else
933 +    pairwisePot += *(idat.pot);
934 +
935 +    snap_->atomData.force[atom1] += *(idat.f1);
936 +    snap_->atomData.force[atom2] -= *(idat.f1);
937 + #endif
938 +    
939 +  }
940 +
941 +  /*
942 +   * buildNeighborList
943 +   *
944 +   * first element of pair is row-indexed CutoffGroup
945 +   * second element of pair is column-indexed CutoffGroup
946 +   */
947 +  vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
948 +      
949 +    vector<pair<int, int> > neighborList;
950 +    groupCutoffs cuts;
951 +    bool doAllPairs = false;
952 +
953 + #ifdef IS_MPI
954 +    cellListRow_.clear();
955 +    cellListCol_.clear();
956 + #else
957 +    cellList_.clear();
958 + #endif
959 +
960 +    RealType rList_ = (largestRcut_ + skinThickness_);
961 +    RealType rl2 = rList_ * rList_;
962 +    Snapshot* snap_ = sman_->getCurrentSnapshot();
963 +    Mat3x3d Hmat = snap_->getHmat();
964 +    Vector3d Hx = Hmat.getColumn(0);
965 +    Vector3d Hy = Hmat.getColumn(1);
966 +    Vector3d Hz = Hmat.getColumn(2);
967 +
968 +    nCells_.x() = (int) ( Hx.length() )/ rList_;
969 +    nCells_.y() = (int) ( Hy.length() )/ rList_;
970 +    nCells_.z() = (int) ( Hz.length() )/ rList_;
971 +
972 +    // handle small boxes where the cell offsets can end up repeating cells
973 +    
974 +    if (nCells_.x() < 3) doAllPairs = true;
975 +    if (nCells_.y() < 3) doAllPairs = true;
976 +    if (nCells_.z() < 3) doAllPairs = true;
977 +
978 +    Mat3x3d invHmat = snap_->getInvHmat();
979 +    Vector3d rs, scaled, dr;
980 +    Vector3i whichCell;
981 +    int cellIndex;
982 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
983 +
984 + #ifdef IS_MPI
985 +    cellListRow_.resize(nCtot);
986 +    cellListCol_.resize(nCtot);
987 + #else
988 +    cellList_.resize(nCtot);
989 + #endif
990 +
991 +    if (!doAllPairs) {
992 + #ifdef IS_MPI
993 +
994 +      for (int i = 0; i < nGroupsInRow_; i++) {
995 +        rs = cgRowData.position[i];
996 +        
997 +        // scaled positions relative to the box vectors
998 +        scaled = invHmat * rs;
999 +        
1000 +        // wrap the vector back into the unit box by subtracting integer box
1001 +        // numbers
1002 +        for (int j = 0; j < 3; j++) {
1003 +          scaled[j] -= roundMe(scaled[j]);
1004 +          scaled[j] += 0.5;
1005 +        }
1006 +        
1007 +        // find xyz-indices of cell that cutoffGroup is in.
1008 +        whichCell.x() = nCells_.x() * scaled.x();
1009 +        whichCell.y() = nCells_.y() * scaled.y();
1010 +        whichCell.z() = nCells_.z() * scaled.z();
1011 +        
1012 +        // find single index of this cell:
1013 +        cellIndex = Vlinear(whichCell, nCells_);
1014 +        
1015 +        // add this cutoff group to the list of groups in this cell;
1016 +        cellListRow_[cellIndex].push_back(i);
1017 +      }
1018 +      
1019 +      for (int i = 0; i < nGroupsInCol_; i++) {
1020 +        rs = cgColData.position[i];
1021 +        
1022 +        // scaled positions relative to the box vectors
1023 +        scaled = invHmat * rs;
1024 +        
1025 +        // wrap the vector back into the unit box by subtracting integer box
1026 +        // numbers
1027 +        for (int j = 0; j < 3; j++) {
1028 +          scaled[j] -= roundMe(scaled[j]);
1029 +          scaled[j] += 0.5;
1030 +        }
1031 +        
1032 +        // find xyz-indices of cell that cutoffGroup is in.
1033 +        whichCell.x() = nCells_.x() * scaled.x();
1034 +        whichCell.y() = nCells_.y() * scaled.y();
1035 +        whichCell.z() = nCells_.z() * scaled.z();
1036 +        
1037 +        // find single index of this cell:
1038 +        cellIndex = Vlinear(whichCell, nCells_);
1039 +        
1040 +        // add this cutoff group to the list of groups in this cell;
1041 +        cellListCol_[cellIndex].push_back(i);
1042 +      }
1043 + #else
1044 +      for (int i = 0; i < nGroups_; i++) {
1045 +        rs = snap_->cgData.position[i];
1046 +        
1047 +        // scaled positions relative to the box vectors
1048 +        scaled = invHmat * rs;
1049 +        
1050 +        // wrap the vector back into the unit box by subtracting integer box
1051 +        // numbers
1052 +        for (int j = 0; j < 3; j++) {
1053 +          scaled[j] -= roundMe(scaled[j]);
1054 +          scaled[j] += 0.5;
1055 +        }
1056 +        
1057 +        // find xyz-indices of cell that cutoffGroup is in.
1058 +        whichCell.x() = nCells_.x() * scaled.x();
1059 +        whichCell.y() = nCells_.y() * scaled.y();
1060 +        whichCell.z() = nCells_.z() * scaled.z();
1061 +        
1062 +        // find single index of this cell:
1063 +        cellIndex = Vlinear(whichCell, nCells_);      
1064 +        
1065 +        // add this cutoff group to the list of groups in this cell;
1066 +        cellList_[cellIndex].push_back(i);
1067 +      }
1068 + #endif
1069 +
1070 +      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1071 +        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1072 +          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1073 +            Vector3i m1v(m1x, m1y, m1z);
1074 +            int m1 = Vlinear(m1v, nCells_);
1075 +            
1076 +            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1077 +                 os != cellOffsets_.end(); ++os) {
1078 +              
1079 +              Vector3i m2v = m1v + (*os);
1080 +              
1081 +              if (m2v.x() >= nCells_.x()) {
1082 +                m2v.x() = 0;          
1083 +              } else if (m2v.x() < 0) {
1084 +                m2v.x() = nCells_.x() - 1;
1085 +              }
1086 +              
1087 +              if (m2v.y() >= nCells_.y()) {
1088 +                m2v.y() = 0;          
1089 +              } else if (m2v.y() < 0) {
1090 +                m2v.y() = nCells_.y() - 1;
1091 +              }
1092 +              
1093 +              if (m2v.z() >= nCells_.z()) {
1094 +                m2v.z() = 0;          
1095 +              } else if (m2v.z() < 0) {
1096 +                m2v.z() = nCells_.z() - 1;
1097 +              }
1098 +              
1099 +              int m2 = Vlinear (m2v, nCells_);
1100 +              
1101 + #ifdef IS_MPI
1102 +              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1103 +                   j1 != cellListRow_[m1].end(); ++j1) {
1104 +                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1105 +                     j2 != cellListCol_[m2].end(); ++j2) {
1106 +                  
1107 +                  // Always do this if we're in different cells or if
1108 +                  // we're in the same cell and the global index of the
1109 +                  // j2 cutoff group is less than the j1 cutoff group
1110 +                  
1111 +                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1112 +                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1113 +                    snap_->wrapVector(dr);
1114 +                    cuts = getGroupCutoffs( (*j1), (*j2) );
1115 +                    if (dr.lengthSquare() < cuts.third) {
1116 +                      neighborList.push_back(make_pair((*j1), (*j2)));
1117 +                    }
1118 +                  }
1119 +                }
1120 +              }
1121 + #else
1122 +              
1123 +              for (vector<int>::iterator j1 = cellList_[m1].begin();
1124 +                   j1 != cellList_[m1].end(); ++j1) {
1125 +                for (vector<int>::iterator j2 = cellList_[m2].begin();
1126 +                     j2 != cellList_[m2].end(); ++j2) {
1127 +                  
1128 +                  // Always do this if we're in different cells or if
1129 +                  // we're in the same cell and the global index of the
1130 +                  // j2 cutoff group is less than the j1 cutoff group
1131 +                  
1132 +                  if (m2 != m1 || (*j2) < (*j1)) {
1133 +                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1134 +                    snap_->wrapVector(dr);
1135 +                    cuts = getGroupCutoffs( (*j1), (*j2) );
1136 +                    if (dr.lengthSquare() < cuts.third) {
1137 +                      neighborList.push_back(make_pair((*j1), (*j2)));
1138 +                    }
1139 +                  }
1140 +                }
1141 +              }
1142 + #endif
1143 +            }
1144 +          }
1145 +        }
1146 +      }
1147 +    } else {
1148 +      // branch to do all cutoff group pairs
1149 + #ifdef IS_MPI
1150 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1151 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1152 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1153 +          snap_->wrapVector(dr);
1154 +          cuts = getGroupCutoffs( j1, j2 );
1155 +          if (dr.lengthSquare() < cuts.third) {
1156 +            neighborList.push_back(make_pair(j1, j2));
1157 +          }
1158 +        }
1159 +      }
1160 + #else
1161 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1162 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1163 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1164 +          snap_->wrapVector(dr);
1165 +          cuts = getGroupCutoffs( j1, j2 );
1166 +          if (dr.lengthSquare() < cuts.third) {
1167 +            neighborList.push_back(make_pair(j1, j2));
1168 +          }
1169 +        }
1170 +      }        
1171 + #endif
1172 +    }
1173 +      
1174 +    // save the local cutoff group positions for the check that is
1175 +    // done on each loop:
1176 +    saved_CG_positions_.clear();
1177 +    for (int i = 0; i < nGroups_; i++)
1178 +      saved_CG_positions_.push_back(snap_->cgData.position[i]);
1179 +    
1180 +    return neighborList;
1181 +  }
1182 + } //end namespace OpenMD

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines