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/ForceMatrixDecomposition.cpp (file contents):
Revision 1570 by gezelter, Thu May 26 21:56:04 2011 UTC vs.
Revision 1601 by gezelter, Thu Aug 4 20:04:35 2011 UTC

# Line 47 | Line 47 | namespace OpenMD {
47   using namespace std;
48   namespace OpenMD {
49  
50 +  ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
51 +
52 +    // In a parallel computation, row and colum scans must visit all
53 +    // surrounding cells (not just the 14 upper triangular blocks that
54 +    // are used when the processor can see all pairs)
55 + #ifdef IS_MPI
56 +    cellOffsets_.push_back( Vector3i(-1, 0, 0) );
57 +    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
58 +    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
59 +    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
60 +    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
61 +    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
62 +    cellOffsets_.push_back( Vector3i(-1,-1,-1) );
63 +    cellOffsets_.push_back( Vector3i( 0,-1,-1) );
64 +    cellOffsets_.push_back( Vector3i( 1,-1,-1) );
65 +    cellOffsets_.push_back( Vector3i( 1, 0,-1) );
66 +    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67 +    cellOffsets_.push_back( Vector3i( 0, 1,-1) );
68 +    cellOffsets_.push_back( Vector3i(-1, 1,-1) );
69 + #endif    
70 +  }
71 +
72 +
73    /**
74     * distributeInitialData is essentially a copy of the older fortran
75     * SimulationSetup
76     */
54  
77    void ForceMatrixDecomposition::distributeInitialData() {
78      snap_ = sman_->getCurrentSnapshot();
79      storageLayout_ = sman_->getStorageLayout();
80 +    ff_ = info_->getForceField();
81      nLocal_ = snap_->getNumberOfAtoms();
82 <    nGroups_ = snap_->getNumberOfCutoffGroups();
83 <
82 >    
83 >    nGroups_ = info_->getNLocalCutoffGroups();
84      // gather the information for atomtype IDs (atids):
85 <    vector<int> identsLocal = info_->getIdentArray();
85 >    idents = info_->getIdentArray();
86      AtomLocalToGlobal = info_->getGlobalAtomIndices();
87      cgLocalToGlobal = info_->getGlobalGroupIndices();
88      vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
66    vector<RealType> massFactorsLocal = info_->getMassFactors();
67    PairList excludes = info_->getExcludedInteractions();
68    PairList oneTwo = info_->getOneTwoInteractions();
69    PairList oneThree = info_->getOneThreeInteractions();
70    PairList oneFour = info_->getOneFourInteractions();
71    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
89  
90 +    massFactors = info_->getMassFactors();
91 +
92 +    PairList* excludes = info_->getExcludedInteractions();
93 +    PairList* oneTwo = info_->getOneTwoInteractions();
94 +    PairList* oneThree = info_->getOneThreeInteractions();
95 +    PairList* oneFour = info_->getOneFourInteractions();
96 +
97   #ifdef IS_MPI
98  
99 <    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
100 <    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
77 <    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
78 <    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
99 >    MPI::Intracomm row = rowComm.getComm();
100 >    MPI::Intracomm col = colComm.getComm();
101  
102 <    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
103 <    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
104 <    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
105 <    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
102 >    AtomPlanIntRow = new Plan<int>(row, nLocal_);
103 >    AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
104 >    AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
105 >    AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
106 >    AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
107  
108 <    cgCommIntRow = new Communicator<Row,int>(nGroups_);
109 <    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
110 <    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
111 <    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
108 >    AtomPlanIntColumn = new Plan<int>(col, nLocal_);
109 >    AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
110 >    AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
111 >    AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
112 >    AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
113  
114 <    nAtomsInRow_ = AtomCommIntRow->getSize();
115 <    nAtomsInCol_ = AtomCommIntColumn->getSize();
116 <    nGroupsInRow_ = cgCommIntRow->getSize();
117 <    nGroupsInCol_ = cgCommIntColumn->getSize();
114 >    cgPlanIntRow = new Plan<int>(row, nGroups_);
115 >    cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
116 >    cgPlanIntColumn = new Plan<int>(col, nGroups_);
117 >    cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
118  
119 +    nAtomsInRow_ = AtomPlanIntRow->getSize();
120 +    nAtomsInCol_ = AtomPlanIntColumn->getSize();
121 +    nGroupsInRow_ = cgPlanIntRow->getSize();
122 +    nGroupsInCol_ = cgPlanIntColumn->getSize();
123 +
124      // Modify the data storage objects with the correct layouts and sizes:
125      atomRowData.resize(nAtomsInRow_);
126      atomRowData.setStorageLayout(storageLayout_);
# Line 101 | Line 130 | namespace OpenMD {
130      cgRowData.setStorageLayout(DataStorage::dslPosition);
131      cgColData.resize(nGroupsInCol_);
132      cgColData.setStorageLayout(DataStorage::dslPosition);
133 +        
134 +    identsRow.resize(nAtomsInRow_);
135 +    identsCol.resize(nAtomsInCol_);
136      
137 <    vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
138 <                                      vector<RealType> (nAtomsInRow_, 0.0));
107 <    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
108 <                                      vector<RealType> (nAtomsInCol_, 0.0));
137 >    AtomPlanIntRow->gather(idents, identsRow);
138 >    AtomPlanIntColumn->gather(idents, identsCol);
139      
140 <    identsRow.reserve(nAtomsInRow_);
141 <    identsCol.reserve(nAtomsInCol_);
142 <    
113 <    AtomCommIntRow->gather(identsLocal, identsRow);
114 <    AtomCommIntColumn->gather(identsLocal, identsCol);
115 <    
116 <    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
117 <    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
118 <    
119 <    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
120 <    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
140 >    // allocate memory for the parallel objects
141 >    atypesRow.resize(nAtomsInRow_);
142 >    atypesCol.resize(nAtomsInCol_);
143  
144 <    AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
145 <    AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
144 >    for (int i = 0; i < nAtomsInRow_; i++)
145 >      atypesRow[i] = ff_->getAtomType(identsRow[i]);
146 >    for (int i = 0; i < nAtomsInCol_; i++)
147 >      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
148  
149 +    pot_row.resize(nAtomsInRow_);
150 +    pot_col.resize(nAtomsInCol_);
151 +
152 +    AtomRowToGlobal.resize(nAtomsInRow_);
153 +    AtomColToGlobal.resize(nAtomsInCol_);
154 +    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
155 +    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
156 +
157 +    cgRowToGlobal.resize(nGroupsInRow_);
158 +    cgColToGlobal.resize(nGroupsInCol_);
159 +    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
160 +    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
161 +
162 +    massFactorsRow.resize(nAtomsInRow_);
163 +    massFactorsCol.resize(nAtomsInCol_);
164 +    AtomPlanRealRow->gather(massFactors, massFactorsRow);
165 +    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
166 +
167      groupListRow_.clear();
168 <    groupListRow_.reserve(nGroupsInRow_);
168 >    groupListRow_.resize(nGroupsInRow_);
169      for (int i = 0; i < nGroupsInRow_; i++) {
170        int gid = cgRowToGlobal[i];
171        for (int j = 0; j < nAtomsInRow_; j++) {
# Line 134 | Line 176 | namespace OpenMD {
176      }
177  
178      groupListCol_.clear();
179 <    groupListCol_.reserve(nGroupsInCol_);
179 >    groupListCol_.resize(nGroupsInCol_);
180      for (int i = 0; i < nGroupsInCol_; i++) {
181        int gid = cgColToGlobal[i];
182        for (int j = 0; j < nAtomsInCol_; j++) {
# Line 144 | Line 186 | namespace OpenMD {
186        }      
187      }
188  
189 <    skipsForRowAtom.clear();
190 <    skipsForRowAtom.reserve(nAtomsInRow_);
189 >    excludesForAtom.clear();
190 >    excludesForAtom.resize(nAtomsInRow_);
191 >    toposForAtom.clear();
192 >    toposForAtom.resize(nAtomsInRow_);
193 >    topoDist.clear();
194 >    topoDist.resize(nAtomsInRow_);
195      for (int i = 0; i < nAtomsInRow_; i++) {
196 <      int iglob = AtomColToGlobal[i];
196 >      int iglob = AtomRowToGlobal[i];
197 >
198        for (int j = 0; j < nAtomsInCol_; j++) {
199 <        int jglob = AtomRowToGlobal[j];        
153 <        if (excludes.hasPair(iglob, jglob))
154 <          skipsForRowAtom[i].push_back(j);      
155 <      }      
156 <    }
199 >        int jglob = AtomColToGlobal[j];
200  
201 <    toposForRowAtom.clear();
202 <    toposForRowAtom.reserve(nAtomsInRow_);
203 <    for (int i = 0; i < nAtomsInRow_; i++) {
204 <      int iglob = AtomColToGlobal[i];
205 <      int nTopos = 0;
206 <      for (int j = 0; j < nAtomsInCol_; j++) {
207 <        int jglob = AtomRowToGlobal[j];        
208 <        if (oneTwo.hasPair(iglob, jglob)) {
209 <          toposForRowAtom[i].push_back(j);
210 <          topoDistRow[i][nTopos] = 1;
211 <          nTopos++;
201 >        if (excludes->hasPair(iglob, jglob))
202 >          excludesForAtom[i].push_back(j);      
203 >        
204 >        if (oneTwo->hasPair(iglob, jglob)) {
205 >          toposForAtom[i].push_back(j);
206 >          topoDist[i].push_back(1);
207 >        } else {
208 >          if (oneThree->hasPair(iglob, jglob)) {
209 >            toposForAtom[i].push_back(j);
210 >            topoDist[i].push_back(2);
211 >          } else {
212 >            if (oneFour->hasPair(iglob, jglob)) {
213 >              toposForAtom[i].push_back(j);
214 >              topoDist[i].push_back(3);
215 >            }
216 >          }
217          }
170        if (oneThree.hasPair(iglob, jglob)) {
171          toposForRowAtom[i].push_back(j);
172          topoDistRow[i][nTopos] = 2;
173          nTopos++;
174        }
175        if (oneFour.hasPair(iglob, jglob)) {
176          toposForRowAtom[i].push_back(j);
177          topoDistRow[i][nTopos] = 3;
178          nTopos++;
179        }
218        }      
219      }
220  
221   #endif
222  
223 +    // allocate memory for the parallel objects
224 +    atypesLocal.resize(nLocal_);
225 +
226 +    for (int i = 0; i < nLocal_; i++)
227 +      atypesLocal[i] = ff_->getAtomType(idents[i]);
228 +
229      groupList_.clear();
230 <    groupList_.reserve(nGroups_);
230 >    groupList_.resize(nGroups_);
231      for (int i = 0; i < nGroups_; i++) {
232        int gid = cgLocalToGlobal[i];
233        for (int j = 0; j < nLocal_; j++) {
234          int aid = AtomLocalToGlobal[j];
235 <        if (globalGroupMembership[aid] == gid)
235 >        if (globalGroupMembership[aid] == gid) {
236            groupList_[i].push_back(j);
237 +        }
238        }      
239      }
240  
241 <    skipsForLocalAtom.clear();
242 <    skipsForLocalAtom.reserve(nLocal_);
241 >    excludesForAtom.clear();
242 >    excludesForAtom.resize(nLocal_);
243 >    toposForAtom.clear();
244 >    toposForAtom.resize(nLocal_);
245 >    topoDist.clear();
246 >    topoDist.resize(nLocal_);
247  
248      for (int i = 0; i < nLocal_; i++) {
249        int iglob = AtomLocalToGlobal[i];
250 +
251        for (int j = 0; j < nLocal_; j++) {
252 <        int jglob = AtomLocalToGlobal[j];        
253 <        if (excludes.hasPair(iglob, jglob))
254 <          skipsForLocalAtom[i].push_back(j);      
252 >        int jglob = AtomLocalToGlobal[j];
253 >
254 >        if (excludes->hasPair(iglob, jglob))
255 >          excludesForAtom[i].push_back(j);              
256 >        
257 >        if (oneTwo->hasPair(iglob, jglob)) {
258 >          toposForAtom[i].push_back(j);
259 >          topoDist[i].push_back(1);
260 >        } else {
261 >          if (oneThree->hasPair(iglob, jglob)) {
262 >            toposForAtom[i].push_back(j);
263 >            topoDist[i].push_back(2);
264 >          } else {
265 >            if (oneFour->hasPair(iglob, jglob)) {
266 >              toposForAtom[i].push_back(j);
267 >              topoDist[i].push_back(3);
268 >            }
269 >          }
270 >        }
271        }      
272      }
273 +    
274 +    createGtypeCutoffMap();
275  
276 <    toposForLocalAtom.clear();
277 <    toposForLocalAtom.reserve(nLocal_);
278 <    for (int i = 0; i < nLocal_; i++) {
279 <      int iglob = AtomLocalToGlobal[i];
280 <      int nTopos = 0;
281 <      for (int j = 0; j < nLocal_; j++) {
282 <        int jglob = AtomLocalToGlobal[j];        
283 <        if (oneTwo.hasPair(iglob, jglob)) {
284 <          toposForLocalAtom[i].push_back(j);
285 <          topoDistLocal[i][nTopos] = 1;
286 <          nTopos++;
287 <        }
288 <        if (oneThree.hasPair(iglob, jglob)) {
289 <          toposForLocalAtom[i].push_back(j);
290 <          topoDistLocal[i][nTopos] = 2;
291 <          nTopos++;
292 <        }
293 <        if (oneFour.hasPair(iglob, jglob)) {
294 <          toposForLocalAtom[i].push_back(j);
295 <          topoDistLocal[i][nTopos] = 3;
296 <          nTopos++;
276 >  }
277 >  
278 >  void ForceMatrixDecomposition::createGtypeCutoffMap() {
279 >    
280 >    RealType tol = 1e-6;
281 >    largestRcut_ = 0.0;
282 >    RealType rc;
283 >    int atid;
284 >    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
285 >    
286 >    map<int, RealType> atypeCutoff;
287 >      
288 >    for (set<AtomType*>::iterator at = atypes.begin();
289 >         at != atypes.end(); ++at){
290 >      atid = (*at)->getIdent();
291 >      if (userChoseCutoff_)
292 >        atypeCutoff[atid] = userCutoff_;
293 >      else
294 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
295 >    }
296 >    
297 >    vector<RealType> gTypeCutoffs;
298 >    // first we do a single loop over the cutoff groups to find the
299 >    // largest cutoff for any atypes present in this group.
300 > #ifdef IS_MPI
301 >    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
302 >    groupRowToGtype.resize(nGroupsInRow_);
303 >    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
304 >      vector<int> atomListRow = getAtomsInGroupRow(cg1);
305 >      for (vector<int>::iterator ia = atomListRow.begin();
306 >           ia != atomListRow.end(); ++ia) {            
307 >        int atom1 = (*ia);
308 >        atid = identsRow[atom1];
309 >        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
310 >          groupCutoffRow[cg1] = atypeCutoff[atid];
311          }
312 +      }
313 +
314 +      bool gTypeFound = false;
315 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
316 +        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
317 +          groupRowToGtype[cg1] = gt;
318 +          gTypeFound = true;
319 +        }
320 +      }
321 +      if (!gTypeFound) {
322 +        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
323 +        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
324 +      }
325 +      
326 +    }
327 +    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
328 +    groupColToGtype.resize(nGroupsInCol_);
329 +    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
330 +      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
331 +      for (vector<int>::iterator jb = atomListCol.begin();
332 +           jb != atomListCol.end(); ++jb) {            
333 +        int atom2 = (*jb);
334 +        atid = identsCol[atom2];
335 +        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
336 +          groupCutoffCol[cg2] = atypeCutoff[atid];
337 +        }
338 +      }
339 +      bool gTypeFound = false;
340 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
341 +        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
342 +          groupColToGtype[cg2] = gt;
343 +          gTypeFound = true;
344 +        }
345 +      }
346 +      if (!gTypeFound) {
347 +        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
348 +        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
349 +      }
350 +    }
351 + #else
352 +
353 +    vector<RealType> groupCutoff(nGroups_, 0.0);
354 +    groupToGtype.resize(nGroups_);
355 +    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
356 +      groupCutoff[cg1] = 0.0;
357 +      vector<int> atomList = getAtomsInGroupRow(cg1);
358 +      for (vector<int>::iterator ia = atomList.begin();
359 +           ia != atomList.end(); ++ia) {            
360 +        int atom1 = (*ia);
361 +        atid = idents[atom1];
362 +        if (atypeCutoff[atid] > groupCutoff[cg1])
363 +          groupCutoff[cg1] = atypeCutoff[atid];
364 +      }
365 +      
366 +      bool gTypeFound = false;
367 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
368 +        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
369 +          groupToGtype[cg1] = gt;
370 +          gTypeFound = true;
371 +        }
372 +      }
373 +      if (!gTypeFound) {      
374 +        gTypeCutoffs.push_back( groupCutoff[cg1] );
375 +        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
376        }      
377      }
378 + #endif
379 +
380 +    // Now we find the maximum group cutoff value present in the simulation
381 +
382 +    RealType groupMax = *max_element(gTypeCutoffs.begin(),
383 +                                     gTypeCutoffs.end());
384 +
385 + #ifdef IS_MPI
386 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
387 +                              MPI::MAX);
388 + #endif
389 +    
390 +    RealType tradRcut = groupMax;
391 +
392 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
393 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
394 +        RealType thisRcut;
395 +        switch(cutoffPolicy_) {
396 +        case TRADITIONAL:
397 +          thisRcut = tradRcut;
398 +          break;
399 +        case MIX:
400 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
401 +          break;
402 +        case MAX:
403 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
404 +          break;
405 +        default:
406 +          sprintf(painCave.errMsg,
407 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
408 +                  "hit an unknown cutoff policy!\n");
409 +          painCave.severity = OPENMD_ERROR;
410 +          painCave.isFatal = 1;
411 +          simError();
412 +          break;
413 +        }
414 +
415 +        pair<int,int> key = make_pair(i,j);
416 +        gTypeCutoffMap[key].first = thisRcut;
417 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
418 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
419 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
420 +        // sanity check
421 +        
422 +        if (userChoseCutoff_) {
423 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
424 +            sprintf(painCave.errMsg,
425 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
426 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
427 +            painCave.severity = OPENMD_ERROR;
428 +            painCave.isFatal = 1;
429 +            simError();            
430 +          }
431 +        }
432 +      }
433 +    }
434    }
435 <  
435 >
436 >
437 >  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
438 >    int i, j;  
439 > #ifdef IS_MPI
440 >    i = groupRowToGtype[cg1];
441 >    j = groupColToGtype[cg2];
442 > #else
443 >    i = groupToGtype[cg1];
444 >    j = groupToGtype[cg2];
445 > #endif    
446 >    return gTypeCutoffMap[make_pair(i,j)];
447 >  }
448 >
449 >  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
450 >    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
451 >      if (toposForAtom[atom1][j] == atom2)
452 >        return topoDist[atom1][j];
453 >    }
454 >    return 0;
455 >  }
456 >
457 >  void ForceMatrixDecomposition::zeroWorkArrays() {
458 >    pairwisePot = 0.0;
459 >    embeddingPot = 0.0;
460 >
461 > #ifdef IS_MPI
462 >    if (storageLayout_ & DataStorage::dslForce) {
463 >      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
464 >      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
465 >    }
466 >
467 >    if (storageLayout_ & DataStorage::dslTorque) {
468 >      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
469 >      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
470 >    }
471 >    
472 >    fill(pot_row.begin(), pot_row.end(),
473 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
474 >
475 >    fill(pot_col.begin(), pot_col.end(),
476 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
477 >
478 >    if (storageLayout_ & DataStorage::dslParticlePot) {    
479 >      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
480 >           0.0);
481 >      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
482 >           0.0);
483 >    }
484 >
485 >    if (storageLayout_ & DataStorage::dslDensity) {      
486 >      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
487 >      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
488 >    }
489 >
490 >    if (storageLayout_ & DataStorage::dslFunctional) {  
491 >      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
492 >           0.0);
493 >      fill(atomColData.functional.begin(), atomColData.functional.end(),
494 >           0.0);
495 >    }
496 >
497 >    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
498 >      fill(atomRowData.functionalDerivative.begin(),
499 >           atomRowData.functionalDerivative.end(), 0.0);
500 >      fill(atomColData.functionalDerivative.begin(),
501 >           atomColData.functionalDerivative.end(), 0.0);
502 >    }
503 >
504 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
505 >      fill(atomRowData.skippedCharge.begin(),
506 >           atomRowData.skippedCharge.end(), 0.0);
507 >      fill(atomColData.skippedCharge.begin(),
508 >           atomColData.skippedCharge.end(), 0.0);
509 >    }
510 >
511 > #endif
512 >    // even in parallel, we need to zero out the local arrays:
513 >
514 >    if (storageLayout_ & DataStorage::dslParticlePot) {      
515 >      fill(snap_->atomData.particlePot.begin(),
516 >           snap_->atomData.particlePot.end(), 0.0);
517 >    }
518 >    
519 >    if (storageLayout_ & DataStorage::dslDensity) {      
520 >      fill(snap_->atomData.density.begin(),
521 >           snap_->atomData.density.end(), 0.0);
522 >    }
523 >    if (storageLayout_ & DataStorage::dslFunctional) {
524 >      fill(snap_->atomData.functional.begin(),
525 >           snap_->atomData.functional.end(), 0.0);
526 >    }
527 >    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
528 >      fill(snap_->atomData.functionalDerivative.begin(),
529 >           snap_->atomData.functionalDerivative.end(), 0.0);
530 >    }
531 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
532 >      fill(snap_->atomData.skippedCharge.begin(),
533 >           snap_->atomData.skippedCharge.end(), 0.0);
534 >    }
535 >    
536 >  }
537 >
538 >
539    void ForceMatrixDecomposition::distributeData()  {
540      snap_ = sman_->getCurrentSnapshot();
541      storageLayout_ = sman_->getStorageLayout();
542   #ifdef IS_MPI
543      
544      // gather up the atomic positions
545 <    AtomCommVectorRow->gather(snap_->atomData.position,
545 >    AtomPlanVectorRow->gather(snap_->atomData.position,
546                                atomRowData.position);
547 <    AtomCommVectorColumn->gather(snap_->atomData.position,
547 >    AtomPlanVectorColumn->gather(snap_->atomData.position,
548                                   atomColData.position);
549      
550      // gather up the cutoff group positions
551 <    cgCommVectorRow->gather(snap_->cgData.position,
551 >
552 >    cgPlanVectorRow->gather(snap_->cgData.position,
553                              cgRowData.position);
554 <    cgCommVectorColumn->gather(snap_->cgData.position,
554 >
555 >    cgPlanVectorColumn->gather(snap_->cgData.position,
556                                 cgColData.position);
557 +
558      
559      // if needed, gather the atomic rotation matrices
560      if (storageLayout_ & DataStorage::dslAmat) {
561 <      AtomCommMatrixRow->gather(snap_->atomData.aMat,
561 >      AtomPlanMatrixRow->gather(snap_->atomData.aMat,
562                                  atomRowData.aMat);
563 <      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
563 >      AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
564                                     atomColData.aMat);
565      }
566      
567      // if needed, gather the atomic eletrostatic frames
568      if (storageLayout_ & DataStorage::dslElectroFrame) {
569 <      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
569 >      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
570                                  atomRowData.electroFrame);
571 <      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
571 >      AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
572                                     atomColData.electroFrame);
573      }
574 +
575   #endif      
576    }
577    
578 +  /* collects information obtained during the pre-pair loop onto local
579 +   * data structures.
580 +   */
581    void ForceMatrixDecomposition::collectIntermediateData() {
582      snap_ = sman_->getCurrentSnapshot();
583      storageLayout_ = sman_->getStorageLayout();
# Line 273 | Line 585 | namespace OpenMD {
585      
586      if (storageLayout_ & DataStorage::dslDensity) {
587        
588 <      AtomCommRealRow->scatter(atomRowData.density,
588 >      AtomPlanRealRow->scatter(atomRowData.density,
589                                 snap_->atomData.density);
590        
591        int n = snap_->atomData.density.size();
592 <      std::vector<RealType> rho_tmp(n, 0.0);
593 <      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
592 >      vector<RealType> rho_tmp(n, 0.0);
593 >      AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
594        for (int i = 0; i < n; i++)
595          snap_->atomData.density[i] += rho_tmp[i];
596      }
597   #endif
598    }
599 <  
599 >
600 >  /*
601 >   * redistributes information obtained during the pre-pair loop out to
602 >   * row and column-indexed data structures
603 >   */
604    void ForceMatrixDecomposition::distributeIntermediateData() {
605      snap_ = sman_->getCurrentSnapshot();
606      storageLayout_ = sman_->getStorageLayout();
607   #ifdef IS_MPI
608      if (storageLayout_ & DataStorage::dslFunctional) {
609 <      AtomCommRealRow->gather(snap_->atomData.functional,
609 >      AtomPlanRealRow->gather(snap_->atomData.functional,
610                                atomRowData.functional);
611 <      AtomCommRealColumn->gather(snap_->atomData.functional,
611 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
612                                   atomColData.functional);
613      }
614      
615      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
616 <      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
616 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
617                                atomRowData.functionalDerivative);
618 <      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
618 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
619                                   atomColData.functionalDerivative);
620      }
621   #endif
# Line 313 | Line 629 | namespace OpenMD {
629      int n = snap_->atomData.force.size();
630      vector<Vector3d> frc_tmp(n, V3Zero);
631      
632 <    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
632 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
633      for (int i = 0; i < n; i++) {
634        snap_->atomData.force[i] += frc_tmp[i];
635        frc_tmp[i] = 0.0;
636      }
637      
638 <    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
639 <    for (int i = 0; i < n; i++)
638 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
639 >    for (int i = 0; i < n; i++) {
640        snap_->atomData.force[i] += frc_tmp[i];
641 <    
642 <    
641 >    }
642 >        
643      if (storageLayout_ & DataStorage::dslTorque) {
644  
645 <      int nt = snap_->atomData.force.size();
645 >      int nt = snap_->atomData.torque.size();
646        vector<Vector3d> trq_tmp(nt, V3Zero);
647  
648 <      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
649 <      for (int i = 0; i < n; i++) {
648 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
649 >      for (int i = 0; i < nt; i++) {
650          snap_->atomData.torque[i] += trq_tmp[i];
651          trq_tmp[i] = 0.0;
652        }
653        
654 <      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
655 <      for (int i = 0; i < n; i++)
654 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
655 >      for (int i = 0; i < nt; i++)
656          snap_->atomData.torque[i] += trq_tmp[i];
657      }
658 +
659 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
660 +
661 +      int ns = snap_->atomData.skippedCharge.size();
662 +      vector<RealType> skch_tmp(ns, 0.0);
663 +
664 +      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
665 +      for (int i = 0; i < ns; i++) {
666 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
667 +        skch_tmp[i] = 0.0;
668 +      }
669 +      
670 +      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
671 +      for (int i = 0; i < ns; i++)
672 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
673 +    }
674      
675      nLocal_ = snap_->getNumberOfAtoms();
676  
677 <    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
678 <                                       vector<RealType> (nLocal_, 0.0));
677 >    vector<potVec> pot_temp(nLocal_,
678 >                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
679 >
680 >    // scatter/gather pot_row into the members of my column
681 >          
682 >    AtomPlanPotRow->scatter(pot_row, pot_temp);
683 >
684 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
685 >      pairwisePot += pot_temp[ii];
686      
687 <    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
688 <      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
689 <      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
690 <        pot_local[i] += pot_temp[i][ii];
691 <      }
687 >    fill(pot_temp.begin(), pot_temp.end(),
688 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
689 >      
690 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
691 >    
692 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
693 >      pairwisePot += pot_temp[ii];    
694 >    
695 >    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
696 >      RealType ploc1 = pairwisePot[ii];
697 >      RealType ploc2 = 0.0;
698 >      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
699 >      pairwisePot[ii] = ploc2;
700      }
701 +
702   #endif
703 +
704    }
705  
706    int ForceMatrixDecomposition::getNAtomsInRow() {  
# Line 426 | Line 775 | namespace OpenMD {
775   #ifdef IS_MPI
776      return massFactorsRow[atom1];
777   #else
778 <    return massFactorsLocal[atom1];
778 >    return massFactors[atom1];
779   #endif
780    }
781  
# Line 434 | Line 783 | namespace OpenMD {
783   #ifdef IS_MPI
784      return massFactorsCol[atom2];
785   #else
786 <    return massFactorsLocal[atom2];
786 >    return massFactors[atom2];
787   #endif
788  
789    }
# Line 452 | Line 801 | namespace OpenMD {
801      return d;    
802    }
803  
804 <  vector<int> ForceMatrixDecomposition::getSkipsForRowAtom(int atom1) {
805 < #ifdef IS_MPI
457 <    return skipsForRowAtom[atom1];
458 < #else
459 <    return skipsForLocalAtom[atom1];
460 < #endif
804 >  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
805 >    return excludesForAtom[atom1];
806    }
807  
808    /**
809 <   * there are a number of reasons to skip a pair or a particle mostly
810 <   * we do this to exclude atoms who are involved in short range
466 <   * interactions (bonds, bends, torsions), but we also need to
467 <   * exclude some overcounted interactions that result from the
468 <   * parallel decomposition.
809 >   * We need to exclude some overcounted interactions that result from
810 >   * the parallel decomposition.
811     */
812    bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
813      int unique_id_1, unique_id_2;
814 <
814 >    
815   #ifdef IS_MPI
816      // in MPI, we have to look up the unique IDs for each atom
817      unique_id_1 = AtomRowToGlobal[atom1];
# Line 484 | Line 826 | namespace OpenMD {
826      } else {
827        if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
828      }
829 + #endif
830 +    return false;
831 +  }
832 +
833 +  /**
834 +   * We need to handle the interactions for atoms who are involved in
835 +   * the same rigid body as well as some short range interactions
836 +   * (bonds, bends, torsions) differently from other interactions.
837 +   * We'll still visit the pairwise routines, but with a flag that
838 +   * tells those routines to exclude the pair from direct long range
839 +   * interactions.  Some indirect interactions (notably reaction
840 +   * field) must still be handled for these pairs.
841 +   */
842 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
843 +    int unique_id_2;
844 + #ifdef IS_MPI
845 +    // in MPI, we have to look up the unique IDs for the row atom.
846 +    unique_id_2 = AtomColToGlobal[atom2];
847   #else
848      // in the normal loop, the atom numbers are unique
489    unique_id_1 = atom1;
849      unique_id_2 = atom2;
850   #endif
851      
852 < #ifdef IS_MPI
853 <    for (vector<int>::iterator i = skipsForRowAtom[atom1].begin();
495 <         i != skipsForRowAtom[atom1].end(); ++i) {
852 >    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
853 >         i != excludesForAtom[atom1].end(); ++i) {
854        if ( (*i) == unique_id_2 ) return true;
497    }    
498 #else
499    for (vector<int>::iterator i = skipsForLocalAtom[atom1].begin();
500         i != skipsForLocalAtom[atom1].end(); ++i) {
501      if ( (*i) == unique_id_2 ) return true;
502    }    
503 #endif
504  }
505
506  int ForceMatrixDecomposition::getTopoDistance(int atom1, int atom2) {
507    
508 #ifdef IS_MPI
509    for (int i = 0; i < toposForRowAtom[atom1].size(); i++) {
510      if ( toposForRowAtom[atom1][i] == atom2 ) return topoDistRow[atom1][i];
855      }
512 #else
513    for (int i = 0; i < toposForLocalAtom[atom1].size(); i++) {
514      if ( toposForLocalAtom[atom1][i] == atom2 ) return topoDistLocal[atom1][i];
515    }
516 #endif
856  
857 <    // zero is default for unconnected (i.e. normal) pair interactions
519 <    return 0;
857 >    return false;
858    }
859  
860 +
861    void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
862   #ifdef IS_MPI
863      atomRowData.force[atom1] += fg;
# Line 536 | Line 875 | namespace OpenMD {
875    }
876  
877      // filling interaction blocks with pointers
878 <  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
879 <    InteractionData idat;
878 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
879 >                                                     int atom1, int atom2) {
880  
881 +    idat.excluded = excludeAtomPair(atom1, atom2);
882 +  
883   #ifdef IS_MPI
884 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
885 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
886 +    //                         ff_->getAtomType(identsCol[atom2]) );
887 +    
888      if (storageLayout_ & DataStorage::dslAmat) {
889        idat.A1 = &(atomRowData.aMat[atom1]);
890        idat.A2 = &(atomColData.aMat[atom2]);
# Line 560 | Line 905 | namespace OpenMD {
905        idat.rho2 = &(atomColData.density[atom2]);
906      }
907  
908 +    if (storageLayout_ & DataStorage::dslFunctional) {
909 +      idat.frho1 = &(atomRowData.functional[atom1]);
910 +      idat.frho2 = &(atomColData.functional[atom2]);
911 +    }
912 +
913      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
914        idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
915        idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
916      }
917  
918 +    if (storageLayout_ & DataStorage::dslParticlePot) {
919 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
920 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
921 +    }
922 +
923 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
924 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
925 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
926 +    }
927 +
928   #else
929 +
930 +    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
931 +    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
932 +    //                         ff_->getAtomType(idents[atom2]) );
933 +
934      if (storageLayout_ & DataStorage::dslAmat) {
935        idat.A1 = &(snap_->atomData.aMat[atom1]);
936        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 581 | Line 946 | namespace OpenMD {
946        idat.t2 = &(snap_->atomData.torque[atom2]);
947      }
948  
949 <    if (storageLayout_ & DataStorage::dslDensity) {
949 >    if (storageLayout_ & DataStorage::dslDensity) {    
950        idat.rho1 = &(snap_->atomData.density[atom1]);
951        idat.rho2 = &(snap_->atomData.density[atom2]);
952      }
953  
954 +    if (storageLayout_ & DataStorage::dslFunctional) {
955 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
956 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
957 +    }
958 +
959      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
960        idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
961        idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
962      }
963 +
964 +    if (storageLayout_ & DataStorage::dslParticlePot) {
965 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
966 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
967 +    }
968 +
969 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
970 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
971 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
972 +    }
973   #endif
594    return idat;
974    }
975  
976 <  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
977 <
599 <    InteractionData idat;
976 >  
977 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
978   #ifdef IS_MPI
979 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
980 <      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
981 <      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
982 <    }
983 <    if (storageLayout_ & DataStorage::dslTorque) {
606 <      idat.t1 = &(atomRowData.torque[atom1]);
607 <      idat.t2 = &(atomColData.torque[atom2]);
608 <    }
609 <    if (storageLayout_ & DataStorage::dslForce) {
610 <      idat.t1 = &(atomRowData.force[atom1]);
611 <      idat.t2 = &(atomColData.force[atom2]);
612 <    }
979 >    pot_row[atom1] += 0.5 *  *(idat.pot);
980 >    pot_col[atom2] += 0.5 *  *(idat.pot);
981 >
982 >    atomRowData.force[atom1] += *(idat.f1);
983 >    atomColData.force[atom2] -= *(idat.f1);
984   #else
985 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
986 <      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
987 <      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
988 <    }
618 <    if (storageLayout_ & DataStorage::dslTorque) {
619 <      idat.t1 = &(snap_->atomData.torque[atom1]);
620 <      idat.t2 = &(snap_->atomData.torque[atom2]);
621 <    }
622 <    if (storageLayout_ & DataStorage::dslForce) {
623 <      idat.t1 = &(snap_->atomData.force[atom1]);
624 <      idat.t2 = &(snap_->atomData.force[atom2]);
625 <    }
985 >    pairwisePot += *(idat.pot);
986 >
987 >    snap_->atomData.force[atom1] += *(idat.f1);
988 >    snap_->atomData.force[atom2] -= *(idat.f1);
989   #endif
990      
991    }
992  
630
631
632
993    /*
994     * buildNeighborList
995     *
# Line 639 | Line 999 | namespace OpenMD {
999    vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1000        
1001      vector<pair<int, int> > neighborList;
1002 +    groupCutoffs cuts;
1003 +    bool doAllPairs = false;
1004 +
1005   #ifdef IS_MPI
1006      cellListRow_.clear();
1007      cellListCol_.clear();
# Line 646 | Line 1009 | namespace OpenMD {
1009      cellList_.clear();
1010   #endif
1011  
1012 <    // dangerous to not do error checking.
650 <    RealType rCut_;
651 <
652 <    RealType rList_ = (rCut_ + skinThickness_);
1012 >    RealType rList_ = (largestRcut_ + skinThickness_);
1013      RealType rl2 = rList_ * rList_;
1014      Snapshot* snap_ = sman_->getCurrentSnapshot();
1015      Mat3x3d Hmat = snap_->getHmat();
# Line 661 | Line 1021 | namespace OpenMD {
1021      nCells_.y() = (int) ( Hy.length() )/ rList_;
1022      nCells_.z() = (int) ( Hz.length() )/ rList_;
1023  
1024 +    // handle small boxes where the cell offsets can end up repeating cells
1025 +    
1026 +    if (nCells_.x() < 3) doAllPairs = true;
1027 +    if (nCells_.y() < 3) doAllPairs = true;
1028 +    if (nCells_.z() < 3) doAllPairs = true;
1029 +
1030      Mat3x3d invHmat = snap_->getInvHmat();
1031      Vector3d rs, scaled, dr;
1032      Vector3i whichCell;
1033      int cellIndex;
1034 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1035  
1036   #ifdef IS_MPI
1037 <    for (int i = 0; i < nGroupsInRow_; i++) {
1038 <      rs = cgRowData.position[i];
1039 <      // scaled positions relative to the box vectors
1040 <      scaled = invHmat * rs;
1041 <      // wrap the vector back into the unit box by subtracting integer box
675 <      // numbers
676 <      for (int j = 0; j < 3; j++)
677 <        scaled[j] -= roundMe(scaled[j]);
678 <    
679 <      // find xyz-indices of cell that cutoffGroup is in.
680 <      whichCell.x() = nCells_.x() * scaled.x();
681 <      whichCell.y() = nCells_.y() * scaled.y();
682 <      whichCell.z() = nCells_.z() * scaled.z();
1037 >    cellListRow_.resize(nCtot);
1038 >    cellListCol_.resize(nCtot);
1039 > #else
1040 >    cellList_.resize(nCtot);
1041 > #endif
1042  
1043 <      // find single index of this cell:
1044 <      cellIndex = Vlinear(whichCell, nCells_);
686 <      // add this cutoff group to the list of groups in this cell;
687 <      cellListRow_[cellIndex].push_back(i);
688 <    }
1043 >    if (!doAllPairs) {
1044 > #ifdef IS_MPI
1045  
1046 <    for (int i = 0; i < nGroupsInCol_; i++) {
1047 <      rs = cgColData.position[i];
1048 <      // scaled positions relative to the box vectors
1049 <      scaled = invHmat * rs;
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 <
1055 <      // find xyz-indices of cell that cutoffGroup is in.
1056 <      whichCell.x() = nCells_.x() * scaled.x();
1057 <      whichCell.y() = nCells_.y() * scaled.y();
1058 <      whichCell.z() = nCells_.z() * scaled.z();
1059 <
1060 <      // find single index of this cell:
1061 <      cellIndex = Vlinear(whichCell, nCells_);
1062 <      // add this cutoff group to the list of groups in this cell;
1063 <      cellListCol_[cellIndex].push_back(i);
1064 <    }
1046 >      for (int i = 0; i < nGroupsInRow_; i++) {
1047 >        rs = cgRowData.position[i];
1048 >        
1049 >        // scaled positions relative to the box vectors
1050 >        scaled = invHmat * rs;
1051 >        
1052 >        // wrap the vector back into the unit box by subtracting integer box
1053 >        // numbers
1054 >        for (int j = 0; j < 3; j++) {
1055 >          scaled[j] -= roundMe(scaled[j]);
1056 >          scaled[j] += 0.5;
1057 >        }
1058 >        
1059 >        // find xyz-indices of cell that cutoffGroup is in.
1060 >        whichCell.x() = nCells_.x() * scaled.x();
1061 >        whichCell.y() = nCells_.y() * scaled.y();
1062 >        whichCell.z() = nCells_.z() * scaled.z();
1063 >        
1064 >        // find single index of this cell:
1065 >        cellIndex = Vlinear(whichCell, nCells_);
1066 >        
1067 >        // add this cutoff group to the list of groups in this cell;
1068 >        cellListRow_[cellIndex].push_back(i);
1069 >      }
1070 >      for (int i = 0; i < nGroupsInCol_; i++) {
1071 >        rs = cgColData.position[i];
1072 >        
1073 >        // scaled positions relative to the box vectors
1074 >        scaled = invHmat * rs;
1075 >        
1076 >        // wrap the vector back into the unit box by subtracting integer box
1077 >        // numbers
1078 >        for (int j = 0; j < 3; j++) {
1079 >          scaled[j] -= roundMe(scaled[j]);
1080 >          scaled[j] += 0.5;
1081 >        }
1082 >        
1083 >        // find xyz-indices of cell that cutoffGroup is in.
1084 >        whichCell.x() = nCells_.x() * scaled.x();
1085 >        whichCell.y() = nCells_.y() * scaled.y();
1086 >        whichCell.z() = nCells_.z() * scaled.z();
1087 >        
1088 >        // find single index of this cell:
1089 >        cellIndex = Vlinear(whichCell, nCells_);
1090 >        
1091 >        // add this cutoff group to the list of groups in this cell;
1092 >        cellListCol_[cellIndex].push_back(i);
1093 >      }
1094   #else
1095 <    for (int i = 0; i < nGroups_; i++) {
1096 <      rs = snap_->cgData.position[i];
1097 <      // scaled positions relative to the box vectors
1098 <      scaled = invHmat * rs;
1099 <      // wrap the vector back into the unit box by subtracting integer box
1100 <      // numbers
1101 <      for (int j = 0; j < 3; j++)
1102 <        scaled[j] -= roundMe(scaled[j]);
1103 <
1104 <      // find xyz-indices of cell that cutoffGroup is in.
1105 <      whichCell.x() = nCells_.x() * scaled.x();
1106 <      whichCell.y() = nCells_.y() * scaled.y();
1107 <      whichCell.z() = nCells_.z() * scaled.z();
1108 <
1109 <      // find single index of this cell:
1110 <      cellIndex = Vlinear(whichCell, nCells_);
1111 <      // add this cutoff group to the list of groups in this cell;
1112 <      cellList_[cellIndex].push_back(i);
1113 <    }
1095 >      for (int i = 0; i < nGroups_; i++) {
1096 >        rs = snap_->cgData.position[i];
1097 >        
1098 >        // scaled positions relative to the box vectors
1099 >        scaled = invHmat * rs;
1100 >        
1101 >        // wrap the vector back into the unit box by subtracting integer box
1102 >        // numbers
1103 >        for (int j = 0; j < 3; j++) {
1104 >          scaled[j] -= roundMe(scaled[j]);
1105 >          scaled[j] += 0.5;
1106 >        }
1107 >        
1108 >        // find xyz-indices of cell that cutoffGroup is in.
1109 >        whichCell.x() = nCells_.x() * scaled.x();
1110 >        whichCell.y() = nCells_.y() * scaled.y();
1111 >        whichCell.z() = nCells_.z() * scaled.z();
1112 >        
1113 >        // find single index of this cell:
1114 >        cellIndex = Vlinear(whichCell, nCells_);
1115 >        
1116 >        // add this cutoff group to the list of groups in this cell;
1117 >        cellList_[cellIndex].push_back(i);
1118 >      }
1119   #endif
1120  
1121 <
1122 <
1123 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1124 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1125 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
736 <          Vector3i m1v(m1x, m1y, m1z);
737 <          int m1 = Vlinear(m1v, nCells_);
738 <
739 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
740 <               os != cellOffsets_.end(); ++os) {
1121 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1122 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1123 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1124 >            Vector3i m1v(m1x, m1y, m1z);
1125 >            int m1 = Vlinear(m1v, nCells_);
1126              
1127 <            Vector3i m2v = m1v + (*os);
1128 <            
1129 <            if (m2v.x() >= nCells_.x()) {
1130 <              m2v.x() = 0;          
1131 <            } else if (m2v.x() < 0) {
1132 <              m2v.x() = nCells_.x() - 1;
1133 <            }
1134 <            
1135 <            if (m2v.y() >= nCells_.y()) {
1136 <              m2v.y() = 0;          
1137 <            } else if (m2v.y() < 0) {
1138 <              m2v.y() = nCells_.y() - 1;
1139 <            }
1140 <            
1141 <            if (m2v.z() >= nCells_.z()) {
1142 <              m2v.z() = 0;          
1143 <            } else if (m2v.z() < 0) {
1144 <              m2v.z() = nCells_.z() - 1;
1145 <            }
1146 <            
1147 <            int m2 = Vlinear (m2v, nCells_);
1148 <
1127 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1128 >                 os != cellOffsets_.end(); ++os) {
1129 >              
1130 >              Vector3i m2v = m1v + (*os);
1131 >              
1132 >              if (m2v.x() >= nCells_.x()) {
1133 >                m2v.x() = 0;          
1134 >              } else if (m2v.x() < 0) {
1135 >                m2v.x() = nCells_.x() - 1;
1136 >              }
1137 >              
1138 >              if (m2v.y() >= nCells_.y()) {
1139 >                m2v.y() = 0;          
1140 >              } else if (m2v.y() < 0) {
1141 >                m2v.y() = nCells_.y() - 1;
1142 >              }
1143 >              
1144 >              if (m2v.z() >= nCells_.z()) {
1145 >                m2v.z() = 0;          
1146 >              } else if (m2v.z() < 0) {
1147 >                m2v.z() = nCells_.z() - 1;
1148 >              }
1149 >              
1150 >              int m2 = Vlinear (m2v, nCells_);
1151 >              
1152   #ifdef IS_MPI
1153 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1154 <                 j1 != cellListRow_[m1].end(); ++j1) {
1155 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1156 <                   j2 != cellListCol_[m2].end(); ++j2) {
1157 <                              
1158 <                // Always do this if we're in different cells or if
1159 <                // we're in the same cell and the global index of the
772 <                // j2 cutoff group is less than the j1 cutoff group
773 <
774 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1153 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1154 >                   j1 != cellListRow_[m1].end(); ++j1) {
1155 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1156 >                     j2 != cellListCol_[m2].end(); ++j2) {
1157 >                  
1158 >                  // In parallel, we need to visit *all* pairs of row &
1159 >                  // column indicies and will truncate later on.
1160                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1161                    snap_->wrapVector(dr);
1162 <                  if (dr.lengthSquare() < rl2) {
1162 >                  cuts = getGroupCutoffs( (*j1), (*j2) );
1163 >                  if (dr.lengthSquare() < cuts.third) {
1164                      neighborList.push_back(make_pair((*j1), (*j2)));
1165 <                  }
1165 >                  }                  
1166                  }
1167                }
782            }
1168   #else
1169 <            for (vector<int>::iterator j1 = cellList_[m1].begin();
1170 <                 j1 != cellList_[m1].end(); ++j1) {
1171 <              for (vector<int>::iterator j2 = cellList_[m2].begin();
1172 <                   j2 != cellList_[m2].end(); ++j2) {
1173 <                              
1174 <                // Always do this if we're in different cells or if
1175 <                // we're in the same cell and the global index of the
1176 <                // j2 cutoff group is less than the j1 cutoff group
1177 <
1178 <                if (m2 != m1 || (*j2) < (*j1)) {
1179 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1180 <                  snap_->wrapVector(dr);
1181 <                  if (dr.lengthSquare() < rl2) {
1182 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1169 >              
1170 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1171 >                   j1 != cellList_[m1].end(); ++j1) {
1172 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1173 >                     j2 != cellList_[m2].end(); ++j2) {
1174 >                  
1175 >                  // Always do this if we're in different cells or if
1176 >                  // we're in the same cell and the global index of the
1177 >                  // j2 cutoff group is less than the j1 cutoff group
1178 >                  
1179 >                  if (m2 != m1 || (*j2) < (*j1)) {
1180 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1181 >                    snap_->wrapVector(dr);
1182 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1183 >                    if (dr.lengthSquare() < cuts.third) {
1184 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1185 >                    }
1186                    }
1187                  }
1188                }
801            }
1189   #endif
1190 +            }
1191            }
1192          }
1193        }
1194 +    } else {
1195 +      // branch to do all cutoff group pairs
1196 + #ifdef IS_MPI
1197 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1198 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1199 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1200 +          snap_->wrapVector(dr);
1201 +          cuts = getGroupCutoffs( j1, j2 );
1202 +          if (dr.lengthSquare() < cuts.third) {
1203 +            neighborList.push_back(make_pair(j1, j2));
1204 +          }
1205 +        }
1206 +      }
1207 + #else
1208 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1209 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1210 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1211 +          snap_->wrapVector(dr);
1212 +          cuts = getGroupCutoffs( j1, j2 );
1213 +          if (dr.lengthSquare() < cuts.third) {
1214 +            neighborList.push_back(make_pair(j1, j2));
1215 +          }
1216 +        }
1217 +      }        
1218 + #endif
1219      }
1220 <
1220 >      
1221      // save the local cutoff group positions for the check that is
1222      // done on each loop:
1223      saved_CG_positions_.clear();
1224      for (int i = 0; i < nGroups_; i++)
1225        saved_CG_positions_.push_back(snap_->cgData.position[i]);
1226 <
1226 >    
1227      return neighborList;
1228    }
1229   } //end namespace OpenMD

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines