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 1587 by gezelter, Fri Jul 8 20:25:32 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 >    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
116 >    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
117 >    
118 >    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
119 >    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
120  
121 < //____ MPITypeTraits
122 < template<typename T>
75 < struct MPITypeTraits;
121 >    AtomCommRealRow->gather(massFactors, massFactorsRow);
122 >    AtomCommRealColumn->gather(massFactors, massFactorsCol);
123  
124 < #ifdef IS_MPI
125 < template<>
126 < struct MPITypeTraits<RealType> {
127 <  static const MPI::Datatype datatype;
128 < };
129 < const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL;
124 >    groupListRow_.clear();
125 >    groupListRow_.resize(nGroupsInRow_);
126 >    for (int i = 0; i < nGroupsInRow_; i++) {
127 >      int gid = cgRowToGlobal[i];
128 >      for (int j = 0; j < nAtomsInRow_; j++) {
129 >        int aid = AtomRowToGlobal[j];
130 >        if (globalGroupMembership[aid] == gid)
131 >          groupListRow_[i].push_back(j);
132 >      }      
133 >    }
134  
135 < template<>
136 < struct MPITypeTraits<int> {
137 <  static const MPI::Datatype datatype;
138 < };
139 < const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT;
140 < #endif
135 >    groupListCol_.clear();
136 >    groupListCol_.resize(nGroupsInCol_);
137 >    for (int i = 0; i < nGroupsInCol_; i++) {
138 >      int gid = cgColToGlobal[i];
139 >      for (int j = 0; j < nAtomsInCol_; j++) {
140 >        int aid = AtomColToGlobal[j];
141 >        if (globalGroupMembership[aid] == gid)
142 >          groupListCol_[i].push_back(j);
143 >      }      
144 >    }
145  
146 < /**
147 < * Constructor for ForceDecomposition Parallel Decomposition Method
148 < * Will try to construct a symmetric grid of processors. Ideally, the
149 < * number of processors will be a square ex: 4, 9, 16, 25.
150 < *
151 < */
146 >    excludesForAtom.clear();
147 >    excludesForAtom.resize(nAtomsInRow_);
148 >    toposForAtom.clear();
149 >    toposForAtom.resize(nAtomsInRow_);
150 >    topoDist.clear();
151 >    topoDist.resize(nAtomsInRow_);
152 >    for (int i = 0; i < nAtomsInRow_; i++) {
153 >      int iglob = AtomRowToGlobal[i];
154  
155 < ForceDecomposition::ForceDecomposition() {
155 >      for (int j = 0; j < nAtomsInCol_; j++) {
156 >        int jglob = AtomColToGlobal[j];
157  
158 < #ifdef IS_MPI
159 <  int nProcs = MPI::COMM_WORLD.Get_size();
160 <  int worldRank = MPI::COMM_WORLD.Get_rank();
158 >        if (excludes->hasPair(iglob, jglob))
159 >          excludesForAtom[i].push_back(j);      
160 >        
161 >        if (oneTwo->hasPair(iglob, jglob)) {
162 >          toposForAtom[i].push_back(j);
163 >          topoDist[i].push_back(1);
164 >        } else {
165 >          if (oneThree->hasPair(iglob, jglob)) {
166 >            toposForAtom[i].push_back(j);
167 >            topoDist[i].push_back(2);
168 >          } else {
169 >            if (oneFour->hasPair(iglob, jglob)) {
170 >              toposForAtom[i].push_back(j);
171 >              topoDist[i].push_back(3);
172 >            }
173 >          }
174 >        }
175 >      }      
176 >    }
177 >
178   #endif
179  
180 <  // First time through, construct column stride.
181 <  if (communicators.size() == 0)
182 <  {
183 <    int nColumnsMax = (int) round(sqrt((float) nProcs));
184 <    for (int i = 0; i < nProcs; ++i)
185 <    {
186 <      if (nProcs%i==0) nColumns=i;
180 >    groupList_.clear();
181 >    groupList_.resize(nGroups_);
182 >    for (int i = 0; i < nGroups_; i++) {
183 >      int gid = cgLocalToGlobal[i];
184 >      for (int j = 0; j < nLocal_; j++) {
185 >        int aid = AtomLocalToGlobal[j];
186 >        if (globalGroupMembership[aid] == gid) {
187 >          groupList_[i].push_back(j);
188 >        }
189 >      }      
190      }
191  
192 <    int nRows = nProcs/nColumns;    
193 <    myRank_ = (int) worldRank%nColumns;
194 <  }
195 <  else
196 <  {
197 <    myRank_ = myRank/nColumns;
120 <  }
121 <  MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0);
122 <  
123 <  isColumn_ = false;
124 <  
125 < }
192 >    excludesForAtom.clear();
193 >    excludesForAtom.resize(nLocal_);
194 >    toposForAtom.clear();
195 >    toposForAtom.resize(nLocal_);
196 >    topoDist.clear();
197 >    topoDist.resize(nLocal_);
198  
199 < ForceDecomposition::gather(sendbuf, receivebuf){
200 <  communicators(myIndex_).Allgatherv();
129 < }
199 >    for (int i = 0; i < nLocal_; i++) {
200 >      int iglob = AtomLocalToGlobal[i];
201  
202 +      for (int j = 0; j < nLocal_; j++) {
203 +        int jglob = AtomLocalToGlobal[j];
204  
205 +        if (excludes->hasPair(iglob, jglob))
206 +          excludesForAtom[i].push_back(j);              
207 +        
208 +        if (oneTwo->hasPair(iglob, jglob)) {
209 +          toposForAtom[i].push_back(j);
210 +          topoDist[i].push_back(1);
211 +        } else {
212 +          if (oneThree->hasPair(iglob, jglob)) {
213 +            toposForAtom[i].push_back(j);
214 +            topoDist[i].push_back(2);
215 +          } else {
216 +            if (oneFour->hasPair(iglob, jglob)) {
217 +              toposForAtom[i].push_back(j);
218 +              topoDist[i].push_back(3);
219 +            }
220 +          }
221 +        }
222 +      }      
223 +    }
224 +    
225 +    createGtypeCutoffMap();
226  
227 < ForceDecomposition::scatter(sbuffer, rbuffer){
228 <  communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM);
229 < }
227 >  }
228 >  
229 >  void ForceMatrixDecomposition::createGtypeCutoffMap() {
230 >    
231 >    RealType tol = 1e-6;
232 >    RealType rc;
233 >    int atid;
234 >    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
235 >    map<int, RealType> atypeCutoff;
236 >      
237 >    for (set<AtomType*>::iterator at = atypes.begin();
238 >         at != atypes.end(); ++at){
239 >      atid = (*at)->getIdent();
240 >      if (userChoseCutoff_)
241 >        atypeCutoff[atid] = userCutoff_;
242 >      else
243 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
244 >    }
245 >
246 >    vector<RealType> gTypeCutoffs;
247 >    // first we do a single loop over the cutoff groups to find the
248 >    // largest cutoff for any atypes present in this group.
249 > #ifdef IS_MPI
250 >    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
251 >    groupRowToGtype.resize(nGroupsInRow_);
252 >    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
253 >      vector<int> atomListRow = getAtomsInGroupRow(cg1);
254 >      for (vector<int>::iterator ia = atomListRow.begin();
255 >           ia != atomListRow.end(); ++ia) {            
256 >        int atom1 = (*ia);
257 >        atid = identsRow[atom1];
258 >        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
259 >          groupCutoffRow[cg1] = atypeCutoff[atid];
260 >        }
261 >      }
262 >
263 >      bool gTypeFound = false;
264 >      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
265 >        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
266 >          groupRowToGtype[cg1] = gt;
267 >          gTypeFound = true;
268 >        }
269 >      }
270 >      if (!gTypeFound) {
271 >        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
272 >        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
273 >      }
274 >      
275 >    }
276 >    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
277 >    groupColToGtype.resize(nGroupsInCol_);
278 >    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
279 >      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
280 >      for (vector<int>::iterator jb = atomListCol.begin();
281 >           jb != atomListCol.end(); ++jb) {            
282 >        int atom2 = (*jb);
283 >        atid = identsCol[atom2];
284 >        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
285 >          groupCutoffCol[cg2] = atypeCutoff[atid];
286 >        }
287 >      }
288 >      bool gTypeFound = false;
289 >      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
290 >        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
291 >          groupColToGtype[cg2] = gt;
292 >          gTypeFound = true;
293 >        }
294 >      }
295 >      if (!gTypeFound) {
296 >        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
297 >        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
298 >      }
299 >    }
300 > #else
301  
302 +    vector<RealType> groupCutoff(nGroups_, 0.0);
303 +    groupToGtype.resize(nGroups_);
304 +    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
305  
306 +      groupCutoff[cg1] = 0.0;
307 +      vector<int> atomList = getAtomsInGroupRow(cg1);
308 +
309 +      for (vector<int>::iterator ia = atomList.begin();
310 +           ia != atomList.end(); ++ia) {            
311 +        int atom1 = (*ia);
312 +        atid = idents[atom1];
313 +        if (atypeCutoff[atid] > groupCutoff[cg1]) {
314 +          groupCutoff[cg1] = atypeCutoff[atid];
315 +        }
316 +      }
317 +
318 +      bool gTypeFound = false;
319 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
320 +        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
321 +          groupToGtype[cg1] = gt;
322 +          gTypeFound = true;
323 +        }
324 +      }
325 +      if (!gTypeFound) {
326 +        gTypeCutoffs.push_back( groupCutoff[cg1] );
327 +        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
328 +      }      
329 +    }
330 + #endif
331 +
332 +    // Now we find the maximum group cutoff value present in the simulation
333 +
334 +    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
335 +
336 + #ifdef IS_MPI
337 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
338 + #endif
339 +    
340 +    RealType tradRcut = groupMax;
341 +
342 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
343 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
344 +        RealType thisRcut;
345 +        switch(cutoffPolicy_) {
346 +        case TRADITIONAL:
347 +          thisRcut = tradRcut;
348 +          break;
349 +        case MIX:
350 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
351 +          break;
352 +        case MAX:
353 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
354 +          break;
355 +        default:
356 +          sprintf(painCave.errMsg,
357 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
358 +                  "hit an unknown cutoff policy!\n");
359 +          painCave.severity = OPENMD_ERROR;
360 +          painCave.isFatal = 1;
361 +          simError();
362 +          break;
363 +        }
364 +
365 +        pair<int,int> key = make_pair(i,j);
366 +        gTypeCutoffMap[key].first = thisRcut;
367 +
368 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
369 +
370 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
371 +        
372 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
373 +
374 +        // sanity check
375 +        
376 +        if (userChoseCutoff_) {
377 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
378 +            sprintf(painCave.errMsg,
379 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
380 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
381 +            painCave.severity = OPENMD_ERROR;
382 +            painCave.isFatal = 1;
383 +            simError();            
384 +          }
385 +        }
386 +      }
387 +    }
388 +  }
389 +
390 +
391 +  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
392 +    int i, j;  
393 + #ifdef IS_MPI
394 +    i = groupRowToGtype[cg1];
395 +    j = groupColToGtype[cg2];
396 + #else
397 +    i = groupToGtype[cg1];
398 +    j = groupToGtype[cg2];
399 + #endif    
400 +    return gTypeCutoffMap[make_pair(i,j)];
401 +  }
402 +
403 +  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
404 +    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
405 +      if (toposForAtom[atom1][j] == atom2)
406 +        return topoDist[atom1][j];
407 +    }
408 +    return 0;
409 +  }
410 +
411 +  void ForceMatrixDecomposition::zeroWorkArrays() {
412 +    pairwisePot = 0.0;
413 +    embeddingPot = 0.0;
414 +
415 + #ifdef IS_MPI
416 +    if (storageLayout_ & DataStorage::dslForce) {
417 +      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
418 +      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
419 +    }
420 +
421 +    if (storageLayout_ & DataStorage::dslTorque) {
422 +      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
423 +      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
424 +    }
425 +    
426 +    fill(pot_row.begin(), pot_row.end(),
427 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
428 +
429 +    fill(pot_col.begin(), pot_col.end(),
430 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
431 +
432 +    if (storageLayout_ & DataStorage::dslParticlePot) {    
433 +      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
434 +      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
435 +    }
436 +
437 +    if (storageLayout_ & DataStorage::dslDensity) {      
438 +      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
439 +      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
440 +    }
441 +
442 +    if (storageLayout_ & DataStorage::dslFunctional) {  
443 +      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
444 +      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
445 +    }
446 +
447 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
448 +      fill(atomRowData.functionalDerivative.begin(),
449 +           atomRowData.functionalDerivative.end(), 0.0);
450 +      fill(atomColData.functionalDerivative.begin(),
451 +           atomColData.functionalDerivative.end(), 0.0);
452 +    }
453 +
454 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
455 +      fill(atomRowData.skippedCharge.begin(),
456 +           atomRowData.skippedCharge.end(), 0.0);
457 +      fill(atomColData.skippedCharge.begin(),
458 +           atomColData.skippedCharge.end(), 0.0);
459 +    }
460 +
461 + #else
462 +    
463 +    if (storageLayout_ & DataStorage::dslParticlePot) {      
464 +      fill(snap_->atomData.particlePot.begin(),
465 +           snap_->atomData.particlePot.end(), 0.0);
466 +    }
467 +    
468 +    if (storageLayout_ & DataStorage::dslDensity) {      
469 +      fill(snap_->atomData.density.begin(),
470 +           snap_->atomData.density.end(), 0.0);
471 +    }
472 +    if (storageLayout_ & DataStorage::dslFunctional) {
473 +      fill(snap_->atomData.functional.begin(),
474 +           snap_->atomData.functional.end(), 0.0);
475 +    }
476 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
477 +      fill(snap_->atomData.functionalDerivative.begin(),
478 +           snap_->atomData.functionalDerivative.end(), 0.0);
479 +    }
480 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
481 +      fill(snap_->atomData.skippedCharge.begin(),
482 +           snap_->atomData.skippedCharge.end(), 0.0);
483 +    }
484 + #endif
485 +    
486 +  }
487 +
488 +
489 +  void ForceMatrixDecomposition::distributeData()  {
490 +    snap_ = sman_->getCurrentSnapshot();
491 +    storageLayout_ = sman_->getStorageLayout();
492 + #ifdef IS_MPI
493 +    
494 +    // gather up the atomic positions
495 +    AtomCommVectorRow->gather(snap_->atomData.position,
496 +                              atomRowData.position);
497 +    AtomCommVectorColumn->gather(snap_->atomData.position,
498 +                                 atomColData.position);
499 +    
500 +    // gather up the cutoff group positions
501 +    cgCommVectorRow->gather(snap_->cgData.position,
502 +                            cgRowData.position);
503 +    cgCommVectorColumn->gather(snap_->cgData.position,
504 +                               cgColData.position);
505 +    
506 +    // if needed, gather the atomic rotation matrices
507 +    if (storageLayout_ & DataStorage::dslAmat) {
508 +      AtomCommMatrixRow->gather(snap_->atomData.aMat,
509 +                                atomRowData.aMat);
510 +      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
511 +                                   atomColData.aMat);
512 +    }
513 +    
514 +    // if needed, gather the atomic eletrostatic frames
515 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
516 +      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
517 +                                atomRowData.electroFrame);
518 +      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
519 +                                   atomColData.electroFrame);
520 +    }
521 + #endif      
522 +  }
523 +  
524 +  /* collects information obtained during the pre-pair loop onto local
525 +   * data structures.
526 +   */
527 +  void ForceMatrixDecomposition::collectIntermediateData() {
528 +    snap_ = sman_->getCurrentSnapshot();
529 +    storageLayout_ = sman_->getStorageLayout();
530 + #ifdef IS_MPI
531 +    
532 +    if (storageLayout_ & DataStorage::dslDensity) {
533 +      
534 +      AtomCommRealRow->scatter(atomRowData.density,
535 +                               snap_->atomData.density);
536 +      
537 +      int n = snap_->atomData.density.size();
538 +      vector<RealType> rho_tmp(n, 0.0);
539 +      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
540 +      for (int i = 0; i < n; i++)
541 +        snap_->atomData.density[i] += rho_tmp[i];
542 +    }
543 + #endif
544 +  }
545 +
546 +  /*
547 +   * redistributes information obtained during the pre-pair loop out to
548 +   * row and column-indexed data structures
549 +   */
550 +  void ForceMatrixDecomposition::distributeIntermediateData() {
551 +    snap_ = sman_->getCurrentSnapshot();
552 +    storageLayout_ = sman_->getStorageLayout();
553 + #ifdef IS_MPI
554 +    if (storageLayout_ & DataStorage::dslFunctional) {
555 +      AtomCommRealRow->gather(snap_->atomData.functional,
556 +                              atomRowData.functional);
557 +      AtomCommRealColumn->gather(snap_->atomData.functional,
558 +                                 atomColData.functional);
559 +    }
560 +    
561 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
562 +      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
563 +                              atomRowData.functionalDerivative);
564 +      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
565 +                                 atomColData.functionalDerivative);
566 +    }
567 + #endif
568 +  }
569 +  
570 +  
571 +  void ForceMatrixDecomposition::collectData() {
572 +    snap_ = sman_->getCurrentSnapshot();
573 +    storageLayout_ = sman_->getStorageLayout();
574 + #ifdef IS_MPI    
575 +    int n = snap_->atomData.force.size();
576 +    vector<Vector3d> frc_tmp(n, V3Zero);
577 +    
578 +    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
579 +    for (int i = 0; i < n; i++) {
580 +      snap_->atomData.force[i] += frc_tmp[i];
581 +      frc_tmp[i] = 0.0;
582 +    }
583 +    
584 +    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
585 +    for (int i = 0; i < n; i++)
586 +      snap_->atomData.force[i] += frc_tmp[i];
587 +    
588 +    
589 +    if (storageLayout_ & DataStorage::dslTorque) {
590 +
591 +      int nt = snap_->atomData.torque.size();
592 +      vector<Vector3d> trq_tmp(nt, V3Zero);
593 +
594 +      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
595 +      for (int i = 0; i < nt; i++) {
596 +        snap_->atomData.torque[i] += trq_tmp[i];
597 +        trq_tmp[i] = 0.0;
598 +      }
599 +      
600 +      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
601 +      for (int i = 0; i < nt; i++)
602 +        snap_->atomData.torque[i] += trq_tmp[i];
603 +    }
604 +
605 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
606 +
607 +      int ns = snap_->atomData.skippedCharge.size();
608 +      vector<RealType> skch_tmp(ns, 0.0);
609 +
610 +      AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
611 +      for (int i = 0; i < ns; i++) {
612 +        snap_->atomData.skippedCharge[i] = skch_tmp[i];
613 +        skch_tmp[i] = 0.0;
614 +      }
615 +      
616 +      AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
617 +      for (int i = 0; i < ns; i++)
618 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
619 +    }
620 +    
621 +    nLocal_ = snap_->getNumberOfAtoms();
622 +
623 +    vector<potVec> pot_temp(nLocal_,
624 +                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
625 +
626 +    // scatter/gather pot_row into the members of my column
627 +          
628 +    AtomCommPotRow->scatter(pot_row, pot_temp);
629 +
630 +    for (int ii = 0;  ii < pot_temp.size(); ii++ )
631 +      pairwisePot += pot_temp[ii];
632 +    
633 +    fill(pot_temp.begin(), pot_temp.end(),
634 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
635 +      
636 +    AtomCommPotColumn->scatter(pot_col, pot_temp);    
637 +    
638 +    for (int ii = 0;  ii < pot_temp.size(); ii++ )
639 +      pairwisePot += pot_temp[ii];    
640 + #endif
641 +
642 +  }
643 +
644 +  int ForceMatrixDecomposition::getNAtomsInRow() {  
645 + #ifdef IS_MPI
646 +    return nAtomsInRow_;
647 + #else
648 +    return nLocal_;
649 + #endif
650 +  }
651 +
652 +  /**
653 +   * returns the list of atoms belonging to this group.  
654 +   */
655 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
656 + #ifdef IS_MPI
657 +    return groupListRow_[cg1];
658 + #else
659 +    return groupList_[cg1];
660 + #endif
661 +  }
662 +
663 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
664 + #ifdef IS_MPI
665 +    return groupListCol_[cg2];
666 + #else
667 +    return groupList_[cg2];
668 + #endif
669 +  }
670 +  
671 +  Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
672 +    Vector3d d;
673 +    
674 + #ifdef IS_MPI
675 +    d = cgColData.position[cg2] - cgRowData.position[cg1];
676 + #else
677 +    d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
678 + #endif
679 +    
680 +    snap_->wrapVector(d);
681 +    return d;    
682 +  }
683 +
684 +
685 +  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
686 +
687 +    Vector3d d;
688 +    
689 + #ifdef IS_MPI
690 +    d = cgRowData.position[cg1] - atomRowData.position[atom1];
691 + #else
692 +    d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
693 + #endif
694 +
695 +    snap_->wrapVector(d);
696 +    return d;    
697 +  }
698 +  
699 +  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
700 +    Vector3d d;
701 +    
702 + #ifdef IS_MPI
703 +    d = cgColData.position[cg2] - atomColData.position[atom2];
704 + #else
705 +    d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
706 + #endif
707 +    
708 +    snap_->wrapVector(d);
709 +    return d;    
710 +  }
711 +
712 +  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
713 + #ifdef IS_MPI
714 +    return massFactorsRow[atom1];
715 + #else
716 +    return massFactors[atom1];
717 + #endif
718 +  }
719 +
720 +  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
721 + #ifdef IS_MPI
722 +    return massFactorsCol[atom2];
723 + #else
724 +    return massFactors[atom2];
725 + #endif
726 +
727 +  }
728 +    
729 +  Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
730 +    Vector3d d;
731 +    
732 + #ifdef IS_MPI
733 +    d = atomColData.position[atom2] - atomRowData.position[atom1];
734 + #else
735 +    d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
736 + #endif
737 +
738 +    snap_->wrapVector(d);
739 +    return d;    
740 +  }
741 +
742 +  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
743 +    return excludesForAtom[atom1];
744 +  }
745 +
746 +  /**
747 +   * We need to exclude some overcounted interactions that result from
748 +   * the parallel decomposition.
749 +   */
750 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
751 +    int unique_id_1, unique_id_2;
752 +
753 + #ifdef IS_MPI
754 +    // in MPI, we have to look up the unique IDs for each atom
755 +    unique_id_1 = AtomRowToGlobal[atom1];
756 +    unique_id_2 = AtomColToGlobal[atom2];
757 +
758 +    // this situation should only arise in MPI simulations
759 +    if (unique_id_1 == unique_id_2) return true;
760 +    
761 +    // this prevents us from doing the pair on multiple processors
762 +    if (unique_id_1 < unique_id_2) {
763 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
764 +    } else {
765 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
766 +    }
767 + #endif
768 +    return false;
769 +  }
770 +
771 +  /**
772 +   * We need to handle the interactions for atoms who are involved in
773 +   * the same rigid body as well as some short range interactions
774 +   * (bonds, bends, torsions) differently from other interactions.
775 +   * We'll still visit the pairwise routines, but with a flag that
776 +   * tells those routines to exclude the pair from direct long range
777 +   * interactions.  Some indirect interactions (notably reaction
778 +   * field) must still be handled for these pairs.
779 +   */
780 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
781 +    int unique_id_2;
782 +    
783 + #ifdef IS_MPI
784 +    // in MPI, we have to look up the unique IDs for the row atom.
785 +    unique_id_2 = AtomColToGlobal[atom2];
786 + #else
787 +    // in the normal loop, the atom numbers are unique
788 +    unique_id_2 = atom2;
789 + #endif
790 +    
791 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
792 +         i != excludesForAtom[atom1].end(); ++i) {
793 +      if ( (*i) == unique_id_2 ) return true;
794 +    }
795 +
796 +    return false;
797 +  }
798 +
799 +
800 +  void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
801 + #ifdef IS_MPI
802 +    atomRowData.force[atom1] += fg;
803 + #else
804 +    snap_->atomData.force[atom1] += fg;
805 + #endif
806 +  }
807 +
808 +  void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
809 + #ifdef IS_MPI
810 +    atomColData.force[atom2] += fg;
811 + #else
812 +    snap_->atomData.force[atom2] += fg;
813 + #endif
814 +  }
815 +
816 +    // filling interaction blocks with pointers
817 +  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
818 +                                                     int atom1, int atom2) {
819 +
820 +    idat.excluded = excludeAtomPair(atom1, atom2);
821 +  
822 + #ifdef IS_MPI
823 +    
824 +    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
825 +                             ff_->getAtomType(identsCol[atom2]) );
826 +    
827 +    if (storageLayout_ & DataStorage::dslAmat) {
828 +      idat.A1 = &(atomRowData.aMat[atom1]);
829 +      idat.A2 = &(atomColData.aMat[atom2]);
830 +    }
831 +    
832 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
833 +      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
834 +      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
835 +    }
836 +
837 +    if (storageLayout_ & DataStorage::dslTorque) {
838 +      idat.t1 = &(atomRowData.torque[atom1]);
839 +      idat.t2 = &(atomColData.torque[atom2]);
840 +    }
841 +
842 +    if (storageLayout_ & DataStorage::dslDensity) {
843 +      idat.rho1 = &(atomRowData.density[atom1]);
844 +      idat.rho2 = &(atomColData.density[atom2]);
845 +    }
846 +
847 +    if (storageLayout_ & DataStorage::dslFunctional) {
848 +      idat.frho1 = &(atomRowData.functional[atom1]);
849 +      idat.frho2 = &(atomColData.functional[atom2]);
850 +    }
851 +
852 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
853 +      idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
854 +      idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
855 +    }
856 +
857 +    if (storageLayout_ & DataStorage::dslParticlePot) {
858 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
859 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
860 +    }
861 +
862 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
863 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
864 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
865 +    }
866 +
867 + #else
868 +
869 +    idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
870 +                             ff_->getAtomType(idents[atom2]) );
871 +
872 +    if (storageLayout_ & DataStorage::dslAmat) {
873 +      idat.A1 = &(snap_->atomData.aMat[atom1]);
874 +      idat.A2 = &(snap_->atomData.aMat[atom2]);
875 +    }
876 +
877 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
878 +      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
879 +      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
880 +    }
881 +
882 +    if (storageLayout_ & DataStorage::dslTorque) {
883 +      idat.t1 = &(snap_->atomData.torque[atom1]);
884 +      idat.t2 = &(snap_->atomData.torque[atom2]);
885 +    }
886 +
887 +    if (storageLayout_ & DataStorage::dslDensity) {    
888 +      idat.rho1 = &(snap_->atomData.density[atom1]);
889 +      idat.rho2 = &(snap_->atomData.density[atom2]);
890 +    }
891 +
892 +    if (storageLayout_ & DataStorage::dslFunctional) {
893 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
894 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
895 +    }
896 +
897 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
898 +      idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
899 +      idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
900 +    }
901 +
902 +    if (storageLayout_ & DataStorage::dslParticlePot) {
903 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
904 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
905 +    }
906 +
907 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
908 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
909 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
910 +    }
911 + #endif
912 +  }
913 +
914 +  
915 +  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
916 + #ifdef IS_MPI
917 +    pot_row[atom1] += 0.5 *  *(idat.pot);
918 +    pot_col[atom2] += 0.5 *  *(idat.pot);
919 +
920 +    atomRowData.force[atom1] += *(idat.f1);
921 +    atomColData.force[atom2] -= *(idat.f1);
922 + #else
923 +    pairwisePot += *(idat.pot);
924 +
925 +    snap_->atomData.force[atom1] += *(idat.f1);
926 +    snap_->atomData.force[atom2] -= *(idat.f1);
927 + #endif
928 +    
929 +  }
930 +
931 +  /*
932 +   * buildNeighborList
933 +   *
934 +   * first element of pair is row-indexed CutoffGroup
935 +   * second element of pair is column-indexed CutoffGroup
936 +   */
937 +  vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
938 +      
939 +    vector<pair<int, int> > neighborList;
940 +    groupCutoffs cuts;
941 +    bool doAllPairs = false;
942 +
943 + #ifdef IS_MPI
944 +    cellListRow_.clear();
945 +    cellListCol_.clear();
946 + #else
947 +    cellList_.clear();
948 + #endif
949 +
950 +    RealType rList_ = (largestRcut_ + skinThickness_);
951 +    RealType rl2 = rList_ * rList_;
952 +    Snapshot* snap_ = sman_->getCurrentSnapshot();
953 +    Mat3x3d Hmat = snap_->getHmat();
954 +    Vector3d Hx = Hmat.getColumn(0);
955 +    Vector3d Hy = Hmat.getColumn(1);
956 +    Vector3d Hz = Hmat.getColumn(2);
957 +
958 +    nCells_.x() = (int) ( Hx.length() )/ rList_;
959 +    nCells_.y() = (int) ( Hy.length() )/ rList_;
960 +    nCells_.z() = (int) ( Hz.length() )/ rList_;
961 +
962 +    // handle small boxes where the cell offsets can end up repeating cells
963 +    
964 +    if (nCells_.x() < 3) doAllPairs = true;
965 +    if (nCells_.y() < 3) doAllPairs = true;
966 +    if (nCells_.z() < 3) doAllPairs = true;
967 +
968 +    Mat3x3d invHmat = snap_->getInvHmat();
969 +    Vector3d rs, scaled, dr;
970 +    Vector3i whichCell;
971 +    int cellIndex;
972 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
973 +
974 + #ifdef IS_MPI
975 +    cellListRow_.resize(nCtot);
976 +    cellListCol_.resize(nCtot);
977 + #else
978 +    cellList_.resize(nCtot);
979 + #endif
980 +
981 +    if (!doAllPairs) {
982 + #ifdef IS_MPI
983 +
984 +      for (int i = 0; i < nGroupsInRow_; i++) {
985 +        rs = cgRowData.position[i];
986 +        
987 +        // scaled positions relative to the box vectors
988 +        scaled = invHmat * rs;
989 +        
990 +        // wrap the vector back into the unit box by subtracting integer box
991 +        // numbers
992 +        for (int j = 0; j < 3; j++) {
993 +          scaled[j] -= roundMe(scaled[j]);
994 +          scaled[j] += 0.5;
995 +        }
996 +        
997 +        // find xyz-indices of cell that cutoffGroup is in.
998 +        whichCell.x() = nCells_.x() * scaled.x();
999 +        whichCell.y() = nCells_.y() * scaled.y();
1000 +        whichCell.z() = nCells_.z() * scaled.z();
1001 +        
1002 +        // find single index of this cell:
1003 +        cellIndex = Vlinear(whichCell, nCells_);
1004 +        
1005 +        // add this cutoff group to the list of groups in this cell;
1006 +        cellListRow_[cellIndex].push_back(i);
1007 +      }
1008 +      
1009 +      for (int i = 0; i < nGroupsInCol_; i++) {
1010 +        rs = cgColData.position[i];
1011 +        
1012 +        // scaled positions relative to the box vectors
1013 +        scaled = invHmat * rs;
1014 +        
1015 +        // wrap the vector back into the unit box by subtracting integer box
1016 +        // numbers
1017 +        for (int j = 0; j < 3; j++) {
1018 +          scaled[j] -= roundMe(scaled[j]);
1019 +          scaled[j] += 0.5;
1020 +        }
1021 +        
1022 +        // find xyz-indices of cell that cutoffGroup is in.
1023 +        whichCell.x() = nCells_.x() * scaled.x();
1024 +        whichCell.y() = nCells_.y() * scaled.y();
1025 +        whichCell.z() = nCells_.z() * scaled.z();
1026 +        
1027 +        // find single index of this cell:
1028 +        cellIndex = Vlinear(whichCell, nCells_);
1029 +        
1030 +        // add this cutoff group to the list of groups in this cell;
1031 +        cellListCol_[cellIndex].push_back(i);
1032 +      }
1033 + #else
1034 +      for (int i = 0; i < nGroups_; i++) {
1035 +        rs = snap_->cgData.position[i];
1036 +        
1037 +        // scaled positions relative to the box vectors
1038 +        scaled = invHmat * rs;
1039 +        
1040 +        // wrap the vector back into the unit box by subtracting integer box
1041 +        // numbers
1042 +        for (int j = 0; j < 3; j++) {
1043 +          scaled[j] -= roundMe(scaled[j]);
1044 +          scaled[j] += 0.5;
1045 +        }
1046 +        
1047 +        // find xyz-indices of cell that cutoffGroup is in.
1048 +        whichCell.x() = nCells_.x() * scaled.x();
1049 +        whichCell.y() = nCells_.y() * scaled.y();
1050 +        whichCell.z() = nCells_.z() * scaled.z();
1051 +        
1052 +        // find single index of this cell:
1053 +        cellIndex = Vlinear(whichCell, nCells_);      
1054 +        
1055 +        // add this cutoff group to the list of groups in this cell;
1056 +        cellList_[cellIndex].push_back(i);
1057 +      }
1058 + #endif
1059 +
1060 +      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1061 +        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1062 +          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1063 +            Vector3i m1v(m1x, m1y, m1z);
1064 +            int m1 = Vlinear(m1v, nCells_);
1065 +            
1066 +            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1067 +                 os != cellOffsets_.end(); ++os) {
1068 +              
1069 +              Vector3i m2v = m1v + (*os);
1070 +              
1071 +              if (m2v.x() >= nCells_.x()) {
1072 +                m2v.x() = 0;          
1073 +              } else if (m2v.x() < 0) {
1074 +                m2v.x() = nCells_.x() - 1;
1075 +              }
1076 +              
1077 +              if (m2v.y() >= nCells_.y()) {
1078 +                m2v.y() = 0;          
1079 +              } else if (m2v.y() < 0) {
1080 +                m2v.y() = nCells_.y() - 1;
1081 +              }
1082 +              
1083 +              if (m2v.z() >= nCells_.z()) {
1084 +                m2v.z() = 0;          
1085 +              } else if (m2v.z() < 0) {
1086 +                m2v.z() = nCells_.z() - 1;
1087 +              }
1088 +              
1089 +              int m2 = Vlinear (m2v, nCells_);
1090 +              
1091 + #ifdef IS_MPI
1092 +              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1093 +                   j1 != cellListRow_[m1].end(); ++j1) {
1094 +                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1095 +                     j2 != cellListCol_[m2].end(); ++j2) {
1096 +                  
1097 +                  // Always do this if we're in different cells or if
1098 +                  // we're in the same cell and the global index of the
1099 +                  // j2 cutoff group is less than the j1 cutoff group
1100 +                  
1101 +                  if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1102 +                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1103 +                    snap_->wrapVector(dr);
1104 +                    cuts = getGroupCutoffs( (*j1), (*j2) );
1105 +                    if (dr.lengthSquare() < cuts.third) {
1106 +                      neighborList.push_back(make_pair((*j1), (*j2)));
1107 +                    }
1108 +                  }
1109 +                }
1110 +              }
1111 + #else
1112 +              
1113 +              for (vector<int>::iterator j1 = cellList_[m1].begin();
1114 +                   j1 != cellList_[m1].end(); ++j1) {
1115 +                for (vector<int>::iterator j2 = cellList_[m2].begin();
1116 +                     j2 != cellList_[m2].end(); ++j2) {
1117 +                  
1118 +                  // Always do this if we're in different cells or if
1119 +                  // we're in the same cell and the global index of the
1120 +                  // j2 cutoff group is less than the j1 cutoff group
1121 +                  
1122 +                  if (m2 != m1 || (*j2) < (*j1)) {
1123 +                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1124 +                    snap_->wrapVector(dr);
1125 +                    cuts = getGroupCutoffs( (*j1), (*j2) );
1126 +                    if (dr.lengthSquare() < cuts.third) {
1127 +                      neighborList.push_back(make_pair((*j1), (*j2)));
1128 +                    }
1129 +                  }
1130 +                }
1131 +              }
1132 + #endif
1133 +            }
1134 +          }
1135 +        }
1136 +      }
1137 +    } else {
1138 +      // branch to do all cutoff group pairs
1139 + #ifdef IS_MPI
1140 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1141 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1142 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1143 +          snap_->wrapVector(dr);
1144 +          cuts = getGroupCutoffs( j1, j2 );
1145 +          if (dr.lengthSquare() < cuts.third) {
1146 +            neighborList.push_back(make_pair(j1, j2));
1147 +          }
1148 +        }
1149 +      }
1150 + #else
1151 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1152 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1153 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1154 +          snap_->wrapVector(dr);
1155 +          cuts = getGroupCutoffs( j1, j2 );
1156 +          if (dr.lengthSquare() < cuts.third) {
1157 +            neighborList.push_back(make_pair(j1, j2));
1158 +          }
1159 +        }
1160 +      }        
1161 + #endif
1162 +    }
1163 +      
1164 +    // save the local cutoff group positions for the check that is
1165 +    // done on each loop:
1166 +    saved_CG_positions_.clear();
1167 +    for (int i = 0; i < nGroups_; i++)
1168 +      saved_CG_positions_.push_back(snap_->cgData.position[i]);
1169 +    
1170 +    return neighborList;
1171 +  }
1172 + } //end namespace OpenMD

Diff Legend

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