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 1568 by gezelter, Wed May 25 16:20:37 2011 UTC vs.
Revision 1613 by gezelter, Thu Aug 18 20:18:19 2011 UTC

# Line 42 | Line 42
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 +  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_.clear();
57 +    cellOffsets_.push_back( Vector3i(-1,-1,-1) );
58 +    cellOffsets_.push_back( Vector3i( 0,-1,-1) );
59 +    cellOffsets_.push_back( Vector3i( 1,-1,-1) );                          
60 +    cellOffsets_.push_back( Vector3i(-1, 0,-1) );
61 +    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
62 +    cellOffsets_.push_back( Vector3i( 1, 0,-1) );
63 +    cellOffsets_.push_back( Vector3i(-1, 1,-1) );
64 +    cellOffsets_.push_back( Vector3i( 0, 1,-1) );      
65 +    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
66 +    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
67 +    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
68 +    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
69 +    cellOffsets_.push_back( Vector3i(-1, 0, 0) );      
70 +    cellOffsets_.push_back( Vector3i( 0, 0, 0) );
71 +    cellOffsets_.push_back( Vector3i( 1, 0, 0) );
72 +    cellOffsets_.push_back( Vector3i(-1, 1, 0) );
73 +    cellOffsets_.push_back( Vector3i( 0, 1, 0) );
74 +    cellOffsets_.push_back( Vector3i( 1, 1, 0) );
75 +    cellOffsets_.push_back( Vector3i(-1,-1, 1) );
76 +    cellOffsets_.push_back( Vector3i( 0,-1, 1) );
77 +    cellOffsets_.push_back( Vector3i( 1,-1, 1) );
78 +    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
79 +    cellOffsets_.push_back( Vector3i( 0, 0, 1) );
80 +    cellOffsets_.push_back( Vector3i( 1, 0, 1) );
81 +    cellOffsets_.push_back( Vector3i(-1, 1, 1) );
82 +    cellOffsets_.push_back( Vector3i( 0, 1, 1) );
83 +    cellOffsets_.push_back( Vector3i( 1, 1, 1) );
84 + #endif    
85 +  }
86 +
87 +
88    /**
89     * distributeInitialData is essentially a copy of the older fortran
90     * SimulationSetup
91     */
53 –  
92    void ForceMatrixDecomposition::distributeInitialData() {
93      snap_ = sman_->getCurrentSnapshot();
94      storageLayout_ = sman_->getStorageLayout();
95 +    ff_ = info_->getForceField();
96      nLocal_ = snap_->getNumberOfAtoms();
97 <    nGroups_ = snap_->getNumberOfCutoffGroups();
97 >    
98 >    nGroups_ = info_->getNLocalCutoffGroups();
99 >    // gather the information for atomtype IDs (atids):
100 >    idents = info_->getIdentArray();
101 >    AtomLocalToGlobal = info_->getGlobalAtomIndices();
102 >    cgLocalToGlobal = info_->getGlobalGroupIndices();
103 >    vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
104  
105 +    massFactors = info_->getMassFactors();
106 +
107 +    PairList* excludes = info_->getExcludedInteractions();
108 +    PairList* oneTwo = info_->getOneTwoInteractions();
109 +    PairList* oneThree = info_->getOneThreeInteractions();
110 +    PairList* oneFour = info_->getOneFourInteractions();
111 +
112   #ifdef IS_MPI
113  
114 <    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
115 <    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
64 <    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
65 <    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
114 >    MPI::Intracomm row = rowComm.getComm();
115 >    MPI::Intracomm col = colComm.getComm();
116  
117 <    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
118 <    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
119 <    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
120 <    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
117 >    AtomPlanIntRow = new Plan<int>(row, nLocal_);
118 >    AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
119 >    AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
120 >    AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
121 >    AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
122  
123 <    cgCommIntRow = new Communicator<Row,int>(nGroups_);
124 <    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
125 <    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
126 <    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
123 >    AtomPlanIntColumn = new Plan<int>(col, nLocal_);
124 >    AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
125 >    AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
126 >    AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
127 >    AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
128  
129 <    nAtomsInRow_ = AtomCommIntRow->getSize();
130 <    nAtomsInCol_ = AtomCommIntColumn->getSize();
131 <    nGroupsInRow_ = cgCommIntRow->getSize();
132 <    nGroupsInCol_ = cgCommIntColumn->getSize();
129 >    cgPlanIntRow = new Plan<int>(row, nGroups_);
130 >    cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
131 >    cgPlanIntColumn = new Plan<int>(col, nGroups_);
132 >    cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
133  
134 +    nAtomsInRow_ = AtomPlanIntRow->getSize();
135 +    nAtomsInCol_ = AtomPlanIntColumn->getSize();
136 +    nGroupsInRow_ = cgPlanIntRow->getSize();
137 +    nGroupsInCol_ = cgPlanIntColumn->getSize();
138 +
139      // Modify the data storage objects with the correct layouts and sizes:
140      atomRowData.resize(nAtomsInRow_);
141      atomRowData.setStorageLayout(storageLayout_);
# Line 88 | Line 145 | namespace OpenMD {
145      cgRowData.setStorageLayout(DataStorage::dslPosition);
146      cgColData.resize(nGroupsInCol_);
147      cgColData.setStorageLayout(DataStorage::dslPosition);
148 +        
149 +    identsRow.resize(nAtomsInRow_);
150 +    identsCol.resize(nAtomsInCol_);
151      
152 <    vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
153 <                                      vector<RealType> (nAtomsInRow_, 0.0));
154 <    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
155 <                                      vector<RealType> (nAtomsInCol_, 0.0));
152 >    AtomPlanIntRow->gather(idents, identsRow);
153 >    AtomPlanIntColumn->gather(idents, identsCol);
154 >    
155 >    // allocate memory for the parallel objects
156 >    atypesRow.resize(nAtomsInRow_);
157 >    atypesCol.resize(nAtomsInCol_);
158  
159 +    for (int i = 0; i < nAtomsInRow_; i++)
160 +      atypesRow[i] = ff_->getAtomType(identsRow[i]);
161 +    for (int i = 0; i < nAtomsInCol_; i++)
162 +      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
163  
164 <    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
164 >    pot_row.resize(nAtomsInRow_);
165 >    pot_col.resize(nAtomsInCol_);
166 >
167 >    AtomRowToGlobal.resize(nAtomsInRow_);
168 >    AtomColToGlobal.resize(nAtomsInCol_);
169 >    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
170 >    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
171 >
172 >    cgRowToGlobal.resize(nGroupsInRow_);
173 >    cgColToGlobal.resize(nGroupsInCol_);
174 >    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
175 >    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
176 >
177 >    massFactorsRow.resize(nAtomsInRow_);
178 >    massFactorsCol.resize(nAtomsInCol_);
179 >    AtomPlanRealRow->gather(massFactors, massFactorsRow);
180 >    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
181 >
182 >    groupListRow_.clear();
183 >    groupListRow_.resize(nGroupsInRow_);
184 >    for (int i = 0; i < nGroupsInRow_; i++) {
185 >      int gid = cgRowToGlobal[i];
186 >      for (int j = 0; j < nAtomsInRow_; j++) {
187 >        int aid = AtomRowToGlobal[j];
188 >        if (globalGroupMembership[aid] == gid)
189 >          groupListRow_[i].push_back(j);
190 >      }      
191 >    }
192 >
193 >    groupListCol_.clear();
194 >    groupListCol_.resize(nGroupsInCol_);
195 >    for (int i = 0; i < nGroupsInCol_; i++) {
196 >      int gid = cgColToGlobal[i];
197 >      for (int j = 0; j < nAtomsInCol_; j++) {
198 >        int aid = AtomColToGlobal[j];
199 >        if (globalGroupMembership[aid] == gid)
200 >          groupListCol_[i].push_back(j);
201 >      }      
202 >    }
203 >
204 >    excludesForAtom.clear();
205 >    excludesForAtom.resize(nAtomsInRow_);
206 >    toposForAtom.clear();
207 >    toposForAtom.resize(nAtomsInRow_);
208 >    topoDist.clear();
209 >    topoDist.resize(nAtomsInRow_);
210 >    for (int i = 0; i < nAtomsInRow_; i++) {
211 >      int iglob = AtomRowToGlobal[i];
212 >
213 >      for (int j = 0; j < nAtomsInCol_; j++) {
214 >        int jglob = AtomColToGlobal[j];
215 >
216 >        if (excludes->hasPair(iglob, jglob))
217 >          excludesForAtom[i].push_back(j);      
218 >        
219 >        if (oneTwo->hasPair(iglob, jglob)) {
220 >          toposForAtom[i].push_back(j);
221 >          topoDist[i].push_back(1);
222 >        } else {
223 >          if (oneThree->hasPair(iglob, jglob)) {
224 >            toposForAtom[i].push_back(j);
225 >            topoDist[i].push_back(2);
226 >          } else {
227 >            if (oneFour->hasPair(iglob, jglob)) {
228 >              toposForAtom[i].push_back(j);
229 >              topoDist[i].push_back(3);
230 >            }
231 >          }
232 >        }
233 >      }      
234 >    }
235 >
236 > #else
237 >    excludesForAtom.clear();
238 >    excludesForAtom.resize(nLocal_);
239 >    toposForAtom.clear();
240 >    toposForAtom.resize(nLocal_);
241 >    topoDist.clear();
242 >    topoDist.resize(nLocal_);
243 >
244 >    for (int i = 0; i < nLocal_; i++) {
245 >      int iglob = AtomLocalToGlobal[i];
246 >
247 >      for (int j = 0; j < nLocal_; j++) {
248 >        int jglob = AtomLocalToGlobal[j];
249 >
250 >        if (excludes->hasPair(iglob, jglob))          
251 >          excludesForAtom[i].push_back(j);              
252 >        
253 >        
254 >        if (oneTwo->hasPair(iglob, jglob)) {
255 >          toposForAtom[i].push_back(j);
256 >          topoDist[i].push_back(1);
257 >        } else {
258 >          if (oneThree->hasPair(iglob, jglob)) {
259 >            toposForAtom[i].push_back(j);
260 >            topoDist[i].push_back(2);
261 >          } else {
262 >            if (oneFour->hasPair(iglob, jglob)) {
263 >              toposForAtom[i].push_back(j);
264 >              topoDist[i].push_back(3);
265 >            }
266 >          }
267 >        }
268 >      }      
269 >    }
270 > #endif
271 >
272 >    // allocate memory for the parallel objects
273 >    atypesLocal.resize(nLocal_);
274 >
275 >    for (int i = 0; i < nLocal_; i++)
276 >      atypesLocal[i] = ff_->getAtomType(idents[i]);
277 >
278 >    groupList_.clear();
279 >    groupList_.resize(nGroups_);
280 >    for (int i = 0; i < nGroups_; i++) {
281 >      int gid = cgLocalToGlobal[i];
282 >      for (int j = 0; j < nLocal_; j++) {
283 >        int aid = AtomLocalToGlobal[j];
284 >        if (globalGroupMembership[aid] == gid) {
285 >          groupList_[i].push_back(j);
286 >        }
287 >      }      
288 >    }
289 >
290 >
291 >    createGtypeCutoffMap();
292 >
293 >  }
294 >  
295 >  void ForceMatrixDecomposition::createGtypeCutoffMap() {
296      
297 <    // gather the information for atomtype IDs (atids):
298 <    vector<int> identsLocal = info_->getIdentArray();
299 <    identsRow.reserve(nAtomsInRow_);
300 <    identsCol.reserve(nAtomsInCol_);
297 >    RealType tol = 1e-6;
298 >    largestRcut_ = 0.0;
299 >    RealType rc;
300 >    int atid;
301 >    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
302      
303 <    AtomCommIntRow->gather(identsLocal, identsRow);
304 <    AtomCommIntColumn->gather(identsLocal, identsCol);
303 >    map<int, RealType> atypeCutoff;
304 >      
305 >    for (set<AtomType*>::iterator at = atypes.begin();
306 >         at != atypes.end(); ++at){
307 >      atid = (*at)->getIdent();
308 >      if (userChoseCutoff_)
309 >        atypeCutoff[atid] = userCutoff_;
310 >      else
311 >        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
312 >    }
313      
314 <    AtomLocalToGlobal = info_->getGlobalAtomIndices();
315 <    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
316 <    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
317 <    
318 <    cgLocalToGlobal = info_->getGlobalGroupIndices();
319 <    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
320 <    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
314 >    vector<RealType> gTypeCutoffs;
315 >    // first we do a single loop over the cutoff groups to find the
316 >    // largest cutoff for any atypes present in this group.
317 > #ifdef IS_MPI
318 >    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
319 >    groupRowToGtype.resize(nGroupsInRow_);
320 >    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
321 >      vector<int> atomListRow = getAtomsInGroupRow(cg1);
322 >      for (vector<int>::iterator ia = atomListRow.begin();
323 >           ia != atomListRow.end(); ++ia) {            
324 >        int atom1 = (*ia);
325 >        atid = identsRow[atom1];
326 >        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
327 >          groupCutoffRow[cg1] = atypeCutoff[atid];
328 >        }
329 >      }
330  
331 <    // still need:
332 <    // topoDist
333 <    // exclude
331 >      bool gTypeFound = false;
332 >      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
333 >        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
334 >          groupRowToGtype[cg1] = gt;
335 >          gTypeFound = true;
336 >        }
337 >      }
338 >      if (!gTypeFound) {
339 >        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
340 >        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
341 >      }
342 >      
343 >    }
344 >    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
345 >    groupColToGtype.resize(nGroupsInCol_);
346 >    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
347 >      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
348 >      for (vector<int>::iterator jb = atomListCol.begin();
349 >           jb != atomListCol.end(); ++jb) {            
350 >        int atom2 = (*jb);
351 >        atid = identsCol[atom2];
352 >        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
353 >          groupCutoffCol[cg2] = atypeCutoff[atid];
354 >        }
355 >      }
356 >      bool gTypeFound = false;
357 >      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
358 >        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
359 >          groupColToGtype[cg2] = gt;
360 >          gTypeFound = true;
361 >        }
362 >      }
363 >      if (!gTypeFound) {
364 >        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
365 >        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
366 >      }
367 >    }
368 > #else
369 >
370 >    vector<RealType> groupCutoff(nGroups_, 0.0);
371 >    groupToGtype.resize(nGroups_);
372 >    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
373 >      groupCutoff[cg1] = 0.0;
374 >      vector<int> atomList = getAtomsInGroupRow(cg1);
375 >      for (vector<int>::iterator ia = atomList.begin();
376 >           ia != atomList.end(); ++ia) {            
377 >        int atom1 = (*ia);
378 >        atid = idents[atom1];
379 >        if (atypeCutoff[atid] > groupCutoff[cg1])
380 >          groupCutoff[cg1] = atypeCutoff[atid];
381 >      }
382 >      
383 >      bool gTypeFound = false;
384 >      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
385 >        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
386 >          groupToGtype[cg1] = gt;
387 >          gTypeFound = true;
388 >        }
389 >      }
390 >      if (!gTypeFound) {      
391 >        gTypeCutoffs.push_back( groupCutoff[cg1] );
392 >        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
393 >      }      
394 >    }
395   #endif
396 +
397 +    // Now we find the maximum group cutoff value present in the simulation
398 +
399 +    RealType groupMax = *max_element(gTypeCutoffs.begin(),
400 +                                     gTypeCutoffs.end());
401 +
402 + #ifdef IS_MPI
403 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
404 +                              MPI::MAX);
405 + #endif
406 +    
407 +    RealType tradRcut = groupMax;
408 +
409 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
410 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
411 +        RealType thisRcut;
412 +        switch(cutoffPolicy_) {
413 +        case TRADITIONAL:
414 +          thisRcut = tradRcut;
415 +          break;
416 +        case MIX:
417 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
418 +          break;
419 +        case MAX:
420 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
421 +          break;
422 +        default:
423 +          sprintf(painCave.errMsg,
424 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
425 +                  "hit an unknown cutoff policy!\n");
426 +          painCave.severity = OPENMD_ERROR;
427 +          painCave.isFatal = 1;
428 +          simError();
429 +          break;
430 +        }
431 +
432 +        pair<int,int> key = make_pair(i,j);
433 +        gTypeCutoffMap[key].first = thisRcut;
434 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
435 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
436 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
437 +        // sanity check
438 +        
439 +        if (userChoseCutoff_) {
440 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
441 +            sprintf(painCave.errMsg,
442 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
443 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
444 +            painCave.severity = OPENMD_ERROR;
445 +            painCave.isFatal = 1;
446 +            simError();            
447 +          }
448 +        }
449 +      }
450 +    }
451    }
452 +
453 +
454 +  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
455 +    int i, j;  
456 + #ifdef IS_MPI
457 +    i = groupRowToGtype[cg1];
458 +    j = groupColToGtype[cg2];
459 + #else
460 +    i = groupToGtype[cg1];
461 +    j = groupToGtype[cg2];
462 + #endif    
463 +    return gTypeCutoffMap[make_pair(i,j)];
464 +  }
465 +
466 +  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
467 +    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
468 +      if (toposForAtom[atom1][j] == atom2)
469 +        return topoDist[atom1][j];
470 +    }
471 +    return 0;
472 +  }
473 +
474 +  void ForceMatrixDecomposition::zeroWorkArrays() {
475 +    pairwisePot = 0.0;
476 +    embeddingPot = 0.0;
477 +
478 + #ifdef IS_MPI
479 +    if (storageLayout_ & DataStorage::dslForce) {
480 +      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
481 +      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
482 +    }
483 +
484 +    if (storageLayout_ & DataStorage::dslTorque) {
485 +      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
486 +      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
487 +    }
488      
489 +    fill(pot_row.begin(), pot_row.end(),
490 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
491  
492 +    fill(pot_col.begin(), pot_col.end(),
493 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));  
494  
495 +    if (storageLayout_ & DataStorage::dslParticlePot) {    
496 +      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
497 +           0.0);
498 +      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
499 +           0.0);
500 +    }
501 +
502 +    if (storageLayout_ & DataStorage::dslDensity) {      
503 +      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
504 +      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
505 +    }
506 +
507 +    if (storageLayout_ & DataStorage::dslFunctional) {  
508 +      fill(atomRowData.functional.begin(), atomRowData.functional.end(),
509 +           0.0);
510 +      fill(atomColData.functional.begin(), atomColData.functional.end(),
511 +           0.0);
512 +    }
513 +
514 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
515 +      fill(atomRowData.functionalDerivative.begin(),
516 +           atomRowData.functionalDerivative.end(), 0.0);
517 +      fill(atomColData.functionalDerivative.begin(),
518 +           atomColData.functionalDerivative.end(), 0.0);
519 +    }
520 +
521 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
522 +      fill(atomRowData.skippedCharge.begin(),
523 +           atomRowData.skippedCharge.end(), 0.0);
524 +      fill(atomColData.skippedCharge.begin(),
525 +           atomColData.skippedCharge.end(), 0.0);
526 +    }
527 +
528 + #endif
529 +    // even in parallel, we need to zero out the local arrays:
530 +
531 +    if (storageLayout_ & DataStorage::dslParticlePot) {      
532 +      fill(snap_->atomData.particlePot.begin(),
533 +           snap_->atomData.particlePot.end(), 0.0);
534 +    }
535 +    
536 +    if (storageLayout_ & DataStorage::dslDensity) {      
537 +      fill(snap_->atomData.density.begin(),
538 +           snap_->atomData.density.end(), 0.0);
539 +    }
540 +    if (storageLayout_ & DataStorage::dslFunctional) {
541 +      fill(snap_->atomData.functional.begin(),
542 +           snap_->atomData.functional.end(), 0.0);
543 +    }
544 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
545 +      fill(snap_->atomData.functionalDerivative.begin(),
546 +           snap_->atomData.functionalDerivative.end(), 0.0);
547 +    }
548 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
549 +      fill(snap_->atomData.skippedCharge.begin(),
550 +           snap_->atomData.skippedCharge.end(), 0.0);
551 +    }
552 +    
553 +  }
554 +
555 +
556    void ForceMatrixDecomposition::distributeData()  {
557      snap_ = sman_->getCurrentSnapshot();
558      storageLayout_ = sman_->getStorageLayout();
559   #ifdef IS_MPI
560      
561      // gather up the atomic positions
562 <    AtomCommVectorRow->gather(snap_->atomData.position,
562 >    AtomPlanVectorRow->gather(snap_->atomData.position,
563                                atomRowData.position);
564 <    AtomCommVectorColumn->gather(snap_->atomData.position,
564 >    AtomPlanVectorColumn->gather(snap_->atomData.position,
565                                   atomColData.position);
566      
567      // gather up the cutoff group positions
568 <    cgCommVectorRow->gather(snap_->cgData.position,
568 >
569 >    cgPlanVectorRow->gather(snap_->cgData.position,
570                              cgRowData.position);
571 <    cgCommVectorColumn->gather(snap_->cgData.position,
571 >
572 >    cgPlanVectorColumn->gather(snap_->cgData.position,
573                                 cgColData.position);
574 +
575      
576      // if needed, gather the atomic rotation matrices
577      if (storageLayout_ & DataStorage::dslAmat) {
578 <      AtomCommMatrixRow->gather(snap_->atomData.aMat,
578 >      AtomPlanMatrixRow->gather(snap_->atomData.aMat,
579                                  atomRowData.aMat);
580 <      AtomCommMatrixColumn->gather(snap_->atomData.aMat,
580 >      AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
581                                     atomColData.aMat);
582      }
583      
584      // if needed, gather the atomic eletrostatic frames
585      if (storageLayout_ & DataStorage::dslElectroFrame) {
586 <      AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
586 >      AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
587                                  atomRowData.electroFrame);
588 <      AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
588 >      AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
589                                     atomColData.electroFrame);
590      }
591 +
592   #endif      
593    }
594    
595 +  /* collects information obtained during the pre-pair loop onto local
596 +   * data structures.
597 +   */
598    void ForceMatrixDecomposition::collectIntermediateData() {
599      snap_ = sman_->getCurrentSnapshot();
600      storageLayout_ = sman_->getStorageLayout();
# Line 163 | Line 602 | namespace OpenMD {
602      
603      if (storageLayout_ & DataStorage::dslDensity) {
604        
605 <      AtomCommRealRow->scatter(atomRowData.density,
605 >      AtomPlanRealRow->scatter(atomRowData.density,
606                                 snap_->atomData.density);
607        
608        int n = snap_->atomData.density.size();
609 <      std::vector<RealType> rho_tmp(n, 0.0);
610 <      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
609 >      vector<RealType> rho_tmp(n, 0.0);
610 >      AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
611        for (int i = 0; i < n; i++)
612          snap_->atomData.density[i] += rho_tmp[i];
613      }
614   #endif
615    }
616 <  
616 >
617 >  /*
618 >   * redistributes information obtained during the pre-pair loop out to
619 >   * row and column-indexed data structures
620 >   */
621    void ForceMatrixDecomposition::distributeIntermediateData() {
622      snap_ = sman_->getCurrentSnapshot();
623      storageLayout_ = sman_->getStorageLayout();
624   #ifdef IS_MPI
625      if (storageLayout_ & DataStorage::dslFunctional) {
626 <      AtomCommRealRow->gather(snap_->atomData.functional,
626 >      AtomPlanRealRow->gather(snap_->atomData.functional,
627                                atomRowData.functional);
628 <      AtomCommRealColumn->gather(snap_->atomData.functional,
628 >      AtomPlanRealColumn->gather(snap_->atomData.functional,
629                                   atomColData.functional);
630      }
631      
632      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
633 <      AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
633 >      AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
634                                atomRowData.functionalDerivative);
635 <      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
635 >      AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
636                                   atomColData.functionalDerivative);
637      }
638   #endif
# Line 203 | Line 646 | namespace OpenMD {
646      int n = snap_->atomData.force.size();
647      vector<Vector3d> frc_tmp(n, V3Zero);
648      
649 <    AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
649 >    AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
650      for (int i = 0; i < n; i++) {
651        snap_->atomData.force[i] += frc_tmp[i];
652        frc_tmp[i] = 0.0;
653      }
654      
655 <    AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
656 <    for (int i = 0; i < n; i++)
655 >    AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
656 >    for (int i = 0; i < n; i++) {
657        snap_->atomData.force[i] += frc_tmp[i];
658 <    
659 <    
658 >    }
659 >        
660      if (storageLayout_ & DataStorage::dslTorque) {
661  
662 <      int nt = snap_->atomData.force.size();
662 >      int nt = snap_->atomData.torque.size();
663        vector<Vector3d> trq_tmp(nt, V3Zero);
664  
665 <      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
666 <      for (int i = 0; i < n; i++) {
665 >      AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
666 >      for (int i = 0; i < nt; i++) {
667          snap_->atomData.torque[i] += trq_tmp[i];
668          trq_tmp[i] = 0.0;
669        }
670        
671 <      AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
672 <      for (int i = 0; i < n; i++)
671 >      AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
672 >      for (int i = 0; i < nt; i++)
673          snap_->atomData.torque[i] += trq_tmp[i];
674      }
675 +
676 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
677 +
678 +      int ns = snap_->atomData.skippedCharge.size();
679 +      vector<RealType> skch_tmp(ns, 0.0);
680 +
681 +      AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
682 +      for (int i = 0; i < ns; i++) {
683 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
684 +        skch_tmp[i] = 0.0;
685 +      }
686 +      
687 +      AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
688 +      for (int i = 0; i < ns; i++)
689 +        snap_->atomData.skippedCharge[i] += skch_tmp[i];
690 +            
691 +    }
692      
693      nLocal_ = snap_->getNumberOfAtoms();
694  
695 <    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
696 <                                       vector<RealType> (nLocal_, 0.0));
695 >    vector<potVec> pot_temp(nLocal_,
696 >                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
697 >
698 >    // scatter/gather pot_row into the members of my column
699 >          
700 >    AtomPlanPotRow->scatter(pot_row, pot_temp);
701 >
702 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
703 >      pairwisePot += pot_temp[ii];
704      
705 <    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
706 <      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
707 <      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
708 <        pot_local[i] += pot_temp[i][ii];
709 <      }
705 >    fill(pot_temp.begin(), pot_temp.end(),
706 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
707 >      
708 >    AtomPlanPotColumn->scatter(pot_col, pot_temp);    
709 >    
710 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
711 >      pairwisePot += pot_temp[ii];    
712 >    
713 >    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
714 >      RealType ploc1 = pairwisePot[ii];
715 >      RealType ploc2 = 0.0;
716 >      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
717 >      pairwisePot[ii] = ploc2;
718      }
719 +
720 +    for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
721 +      RealType ploc1 = embeddingPot[ii];
722 +      RealType ploc2 = 0.0;
723 +      MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
724 +      embeddingPot[ii] = ploc2;
725 +    }
726 +
727   #endif
728 +
729    }
730  
731 +  int ForceMatrixDecomposition::getNAtomsInRow() {  
732 + #ifdef IS_MPI
733 +    return nAtomsInRow_;
734 + #else
735 +    return nLocal_;
736 + #endif
737 +  }
738 +
739 +  /**
740 +   * returns the list of atoms belonging to this group.  
741 +   */
742 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
743 + #ifdef IS_MPI
744 +    return groupListRow_[cg1];
745 + #else
746 +    return groupList_[cg1];
747 + #endif
748 +  }
749 +
750 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
751 + #ifdef IS_MPI
752 +    return groupListCol_[cg2];
753 + #else
754 +    return groupList_[cg2];
755 + #endif
756 +  }
757    
758    Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
759      Vector3d d;
# Line 285 | Line 795 | namespace OpenMD {
795      snap_->wrapVector(d);
796      return d;    
797    }
798 +
799 +  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
800 + #ifdef IS_MPI
801 +    return massFactorsRow[atom1];
802 + #else
803 +    return massFactors[atom1];
804 + #endif
805 +  }
806 +
807 +  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
808 + #ifdef IS_MPI
809 +    return massFactorsCol[atom2];
810 + #else
811 +    return massFactors[atom2];
812 + #endif
813 +
814 +  }
815      
816    Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
817      Vector3d d;
# Line 299 | Line 826 | namespace OpenMD {
826      return d;    
827    }
828  
829 +  vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
830 +    return excludesForAtom[atom1];
831 +  }
832 +
833 +  /**
834 +   * We need to exclude some overcounted interactions that result from
835 +   * the parallel decomposition.
836 +   */
837 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
838 +    int unique_id_1, unique_id_2;
839 +    
840 + #ifdef IS_MPI
841 +    // in MPI, we have to look up the unique IDs for each atom
842 +    unique_id_1 = AtomRowToGlobal[atom1];
843 +    unique_id_2 = AtomColToGlobal[atom2];
844 +
845 +    // this situation should only arise in MPI simulations
846 +    if (unique_id_1 == unique_id_2) return true;
847 +    
848 +    // this prevents us from doing the pair on multiple processors
849 +    if (unique_id_1 < unique_id_2) {
850 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
851 +    } else {
852 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
853 +    }
854 + #endif
855 +    return false;
856 +  }
857 +
858 +  /**
859 +   * We need to handle the interactions for atoms who are involved in
860 +   * the same rigid body as well as some short range interactions
861 +   * (bonds, bends, torsions) differently from other interactions.
862 +   * We'll still visit the pairwise routines, but with a flag that
863 +   * tells those routines to exclude the pair from direct long range
864 +   * interactions.  Some indirect interactions (notably reaction
865 +   * field) must still be handled for these pairs.
866 +   */
867 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
868 +
869 +    // excludesForAtom was constructed to use row/column indices in the MPI
870 +    // version, and to use local IDs in the non-MPI version:
871 +    
872 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
873 +         i != excludesForAtom[atom1].end(); ++i) {
874 +      if ( (*i) == atom2 )  return true;
875 +    }
876 +
877 +    return false;
878 +  }
879 +
880 +
881    void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
882   #ifdef IS_MPI
883      atomRowData.force[atom1] += fg;
# Line 316 | Line 895 | namespace OpenMD {
895    }
896  
897      // filling interaction blocks with pointers
898 <  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
899 <    InteractionData idat;
898 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
899 >                                                     int atom1, int atom2) {
900  
901 +    idat.excluded = excludeAtomPair(atom1, atom2);
902 +  
903   #ifdef IS_MPI
904 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
905 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
906 +    //                         ff_->getAtomType(identsCol[atom2]) );
907 +    
908      if (storageLayout_ & DataStorage::dslAmat) {
909        idat.A1 = &(atomRowData.aMat[atom1]);
910        idat.A2 = &(atomColData.aMat[atom2]);
# Line 340 | Line 925 | namespace OpenMD {
925        idat.rho2 = &(atomColData.density[atom2]);
926      }
927  
928 +    if (storageLayout_ & DataStorage::dslFunctional) {
929 +      idat.frho1 = &(atomRowData.functional[atom1]);
930 +      idat.frho2 = &(atomColData.functional[atom2]);
931 +    }
932 +
933      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
934        idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
935        idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
936      }
937 +
938 +    if (storageLayout_ & DataStorage::dslParticlePot) {
939 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
940 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
941 +    }
942 +
943 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
944 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
945 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
946 +    }
947 +
948   #else
949 +
950 +    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
951 +    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
952 +    //                         ff_->getAtomType(idents[atom2]) );
953 +
954      if (storageLayout_ & DataStorage::dslAmat) {
955        idat.A1 = &(snap_->atomData.aMat[atom1]);
956        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 360 | Line 966 | namespace OpenMD {
966        idat.t2 = &(snap_->atomData.torque[atom2]);
967      }
968  
969 <    if (storageLayout_ & DataStorage::dslDensity) {
969 >    if (storageLayout_ & DataStorage::dslDensity) {    
970        idat.rho1 = &(snap_->atomData.density[atom1]);
971        idat.rho2 = &(snap_->atomData.density[atom2]);
972      }
973  
974 +    if (storageLayout_ & DataStorage::dslFunctional) {
975 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
976 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
977 +    }
978 +
979      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
980        idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
981        idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
982      }
983 +
984 +    if (storageLayout_ & DataStorage::dslParticlePot) {
985 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
986 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
987 +    }
988 +
989 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
990 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
991 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
992 +    }
993   #endif
373 –    return idat;
994    }
995  
996 <  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
997 <
378 <    InteractionData idat;
996 >  
997 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
998   #ifdef IS_MPI
999 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
1000 <      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1001 <      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1002 <    }
1003 <    if (storageLayout_ & DataStorage::dslTorque) {
385 <      idat.t1 = &(atomRowData.torque[atom1]);
386 <      idat.t2 = &(atomColData.torque[atom2]);
387 <    }
388 <    if (storageLayout_ & DataStorage::dslForce) {
389 <      idat.t1 = &(atomRowData.force[atom1]);
390 <      idat.t2 = &(atomColData.force[atom2]);
391 <    }
999 >    pot_row[atom1] += 0.5 *  *(idat.pot);
1000 >    pot_col[atom2] += 0.5 *  *(idat.pot);
1001 >
1002 >    atomRowData.force[atom1] += *(idat.f1);
1003 >    atomColData.force[atom2] -= *(idat.f1);
1004   #else
1005 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
1006 <      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1007 <      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1008 <    }
397 <    if (storageLayout_ & DataStorage::dslTorque) {
398 <      idat.t1 = &(snap_->atomData.torque[atom1]);
399 <      idat.t2 = &(snap_->atomData.torque[atom2]);
400 <    }
401 <    if (storageLayout_ & DataStorage::dslForce) {
402 <      idat.t1 = &(snap_->atomData.force[atom1]);
403 <      idat.t2 = &(snap_->atomData.force[atom2]);
404 <    }
1005 >    pairwisePot += *(idat.pot);
1006 >
1007 >    snap_->atomData.force[atom1] += *(idat.f1);
1008 >    snap_->atomData.force[atom2] -= *(idat.f1);
1009   #endif
1010      
1011    }
1012  
409 –
410 –
411 –
1013    /*
1014     * buildNeighborList
1015     *
# Line 418 | Line 1019 | namespace OpenMD {
1019    vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1020        
1021      vector<pair<int, int> > neighborList;
1022 +    groupCutoffs cuts;
1023 +    bool doAllPairs = false;
1024 +
1025   #ifdef IS_MPI
1026      cellListRow_.clear();
1027      cellListCol_.clear();
# Line 425 | Line 1029 | namespace OpenMD {
1029      cellList_.clear();
1030   #endif
1031  
1032 <    // dangerous to not do error checking.
429 <    RealType rCut_;
430 <
431 <    RealType rList_ = (rCut_ + skinThickness_);
1032 >    RealType rList_ = (largestRcut_ + skinThickness_);
1033      RealType rl2 = rList_ * rList_;
1034      Snapshot* snap_ = sman_->getCurrentSnapshot();
1035      Mat3x3d Hmat = snap_->getHmat();
# Line 440 | Line 1041 | namespace OpenMD {
1041      nCells_.y() = (int) ( Hy.length() )/ rList_;
1042      nCells_.z() = (int) ( Hz.length() )/ rList_;
1043  
1044 +    // handle small boxes where the cell offsets can end up repeating cells
1045 +    
1046 +    if (nCells_.x() < 3) doAllPairs = true;
1047 +    if (nCells_.y() < 3) doAllPairs = true;
1048 +    if (nCells_.z() < 3) doAllPairs = true;
1049 +
1050      Mat3x3d invHmat = snap_->getInvHmat();
1051      Vector3d rs, scaled, dr;
1052      Vector3i whichCell;
1053      int cellIndex;
1054 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1055  
1056   #ifdef IS_MPI
1057 <    for (int i = 0; i < nGroupsInRow_; i++) {
1058 <      rs = cgRowData.position[i];
1059 <      // scaled positions relative to the box vectors
1060 <      scaled = invHmat * rs;
1061 <      // wrap the vector back into the unit box by subtracting integer box
454 <      // numbers
455 <      for (int j = 0; j < 3; j++)
456 <        scaled[j] -= roundMe(scaled[j]);
457 <    
458 <      // find xyz-indices of cell that cutoffGroup is in.
459 <      whichCell.x() = nCells_.x() * scaled.x();
460 <      whichCell.y() = nCells_.y() * scaled.y();
461 <      whichCell.z() = nCells_.z() * scaled.z();
1057 >    cellListRow_.resize(nCtot);
1058 >    cellListCol_.resize(nCtot);
1059 > #else
1060 >    cellList_.resize(nCtot);
1061 > #endif
1062  
1063 <      // find single index of this cell:
1064 <      cellIndex = Vlinear(whichCell, nCells_);
465 <      // add this cutoff group to the list of groups in this cell;
466 <      cellListRow_[cellIndex].push_back(i);
467 <    }
1063 >    if (!doAllPairs) {
1064 > #ifdef IS_MPI
1065  
1066 <    for (int i = 0; i < nGroupsInCol_; i++) {
1067 <      rs = cgColData.position[i];
1068 <      // scaled positions relative to the box vectors
1069 <      scaled = invHmat * rs;
1070 <      // wrap the vector back into the unit box by subtracting integer box
1071 <      // numbers
1072 <      for (int j = 0; j < 3; j++)
1073 <        scaled[j] -= roundMe(scaled[j]);
1074 <
1075 <      // find xyz-indices of cell that cutoffGroup is in.
1076 <      whichCell.x() = nCells_.x() * scaled.x();
1077 <      whichCell.y() = nCells_.y() * scaled.y();
1078 <      whichCell.z() = nCells_.z() * scaled.z();
1079 <
1080 <      // find single index of this cell:
1081 <      cellIndex = Vlinear(whichCell, nCells_);
1082 <      // add this cutoff group to the list of groups in this cell;
1083 <      cellListCol_[cellIndex].push_back(i);
1084 <    }
1066 >      for (int i = 0; i < nGroupsInRow_; i++) {
1067 >        rs = cgRowData.position[i];
1068 >        
1069 >        // scaled positions relative to the box vectors
1070 >        scaled = invHmat * rs;
1071 >        
1072 >        // wrap the vector back into the unit box by subtracting integer box
1073 >        // numbers
1074 >        for (int j = 0; j < 3; j++) {
1075 >          scaled[j] -= roundMe(scaled[j]);
1076 >          scaled[j] += 0.5;
1077 >        }
1078 >        
1079 >        // find xyz-indices of cell that cutoffGroup is in.
1080 >        whichCell.x() = nCells_.x() * scaled.x();
1081 >        whichCell.y() = nCells_.y() * scaled.y();
1082 >        whichCell.z() = nCells_.z() * scaled.z();
1083 >        
1084 >        // find single index of this cell:
1085 >        cellIndex = Vlinear(whichCell, nCells_);
1086 >        
1087 >        // add this cutoff group to the list of groups in this cell;
1088 >        cellListRow_[cellIndex].push_back(i);
1089 >      }
1090 >      for (int i = 0; i < nGroupsInCol_; i++) {
1091 >        rs = cgColData.position[i];
1092 >        
1093 >        // scaled positions relative to the box vectors
1094 >        scaled = invHmat * rs;
1095 >        
1096 >        // wrap the vector back into the unit box by subtracting integer box
1097 >        // numbers
1098 >        for (int j = 0; j < 3; j++) {
1099 >          scaled[j] -= roundMe(scaled[j]);
1100 >          scaled[j] += 0.5;
1101 >        }
1102 >        
1103 >        // find xyz-indices of cell that cutoffGroup is in.
1104 >        whichCell.x() = nCells_.x() * scaled.x();
1105 >        whichCell.y() = nCells_.y() * scaled.y();
1106 >        whichCell.z() = nCells_.z() * scaled.z();
1107 >        
1108 >        // find single index of this cell:
1109 >        cellIndex = Vlinear(whichCell, nCells_);
1110 >        
1111 >        // add this cutoff group to the list of groups in this cell;
1112 >        cellListCol_[cellIndex].push_back(i);
1113 >      }
1114 >    
1115   #else
1116 <    for (int i = 0; i < nGroups_; i++) {
1117 <      rs = snap_->cgData.position[i];
1118 <      // scaled positions relative to the box vectors
1119 <      scaled = invHmat * rs;
1120 <      // wrap the vector back into the unit box by subtracting integer box
1121 <      // numbers
1122 <      for (int j = 0; j < 3; j++)
1123 <        scaled[j] -= roundMe(scaled[j]);
1116 >      for (int i = 0; i < nGroups_; i++) {
1117 >        rs = snap_->cgData.position[i];
1118 >        
1119 >        // scaled positions relative to the box vectors
1120 >        scaled = invHmat * rs;
1121 >        
1122 >        // wrap the vector back into the unit box by subtracting integer box
1123 >        // numbers
1124 >        for (int j = 0; j < 3; j++) {
1125 >          scaled[j] -= roundMe(scaled[j]);
1126 >          scaled[j] += 0.5;
1127 >        }
1128 >        
1129 >        // find xyz-indices of cell that cutoffGroup is in.
1130 >        whichCell.x() = nCells_.x() * scaled.x();
1131 >        whichCell.y() = nCells_.y() * scaled.y();
1132 >        whichCell.z() = nCells_.z() * scaled.z();
1133 >        
1134 >        // find single index of this cell:
1135 >        cellIndex = Vlinear(whichCell, nCells_);
1136 >        
1137 >        // add this cutoff group to the list of groups in this cell;
1138 >        cellList_[cellIndex].push_back(i);
1139 >      }
1140  
498 –      // find xyz-indices of cell that cutoffGroup is in.
499 –      whichCell.x() = nCells_.x() * scaled.x();
500 –      whichCell.y() = nCells_.y() * scaled.y();
501 –      whichCell.z() = nCells_.z() * scaled.z();
502 –
503 –      // find single index of this cell:
504 –      cellIndex = Vlinear(whichCell, nCells_);
505 –      // add this cutoff group to the list of groups in this cell;
506 –      cellList_[cellIndex].push_back(i);
507 –    }
1141   #endif
1142  
1143 <
1144 <
1145 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1146 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1147 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
515 <          Vector3i m1v(m1x, m1y, m1z);
516 <          int m1 = Vlinear(m1v, nCells_);
517 <
518 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
519 <               os != cellOffsets_.end(); ++os) {
1143 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1144 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1145 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1146 >            Vector3i m1v(m1x, m1y, m1z);
1147 >            int m1 = Vlinear(m1v, nCells_);
1148              
1149 <            Vector3i m2v = m1v + (*os);
1150 <            
1151 <            if (m2v.x() >= nCells_.x()) {
1152 <              m2v.x() = 0;          
1153 <            } else if (m2v.x() < 0) {
526 <              m2v.x() = nCells_.x() - 1;
527 <            }
528 <            
529 <            if (m2v.y() >= nCells_.y()) {
530 <              m2v.y() = 0;          
531 <            } else if (m2v.y() < 0) {
532 <              m2v.y() = nCells_.y() - 1;
533 <            }
534 <            
535 <            if (m2v.z() >= nCells_.z()) {
536 <              m2v.z() = 0;          
537 <            } else if (m2v.z() < 0) {
538 <              m2v.z() = nCells_.z() - 1;
539 <            }
540 <            
541 <            int m2 = Vlinear (m2v, nCells_);
1149 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1150 >                 os != cellOffsets_.end(); ++os) {
1151 >              
1152 >              Vector3i m2v = m1v + (*os);
1153 >            
1154  
1155 < #ifdef IS_MPI
1156 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1157 <                 j1 != cellListRow_[m1].end(); ++j1) {
1158 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1159 <                   j2 != cellListCol_[m2].end(); ++j2) {
1160 <                              
1161 <                // Always do this if we're in different cells or if
1162 <                // we're in the same cell and the global index of the
1163 <                // j2 cutoff group is less than the j1 cutoff group
1155 >              if (m2v.x() >= nCells_.x()) {
1156 >                m2v.x() = 0;          
1157 >              } else if (m2v.x() < 0) {
1158 >                m2v.x() = nCells_.x() - 1;
1159 >              }
1160 >              
1161 >              if (m2v.y() >= nCells_.y()) {
1162 >                m2v.y() = 0;          
1163 >              } else if (m2v.y() < 0) {
1164 >                m2v.y() = nCells_.y() - 1;
1165 >              }
1166 >              
1167 >              if (m2v.z() >= nCells_.z()) {
1168 >                m2v.z() = 0;          
1169 >              } else if (m2v.z() < 0) {
1170 >                m2v.z() = nCells_.z() - 1;
1171 >              }
1172  
1173 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1173 >              int m2 = Vlinear (m2v, nCells_);
1174 >              
1175 > #ifdef IS_MPI
1176 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1177 >                   j1 != cellListRow_[m1].end(); ++j1) {
1178 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1179 >                     j2 != cellListCol_[m2].end(); ++j2) {
1180 >                  
1181 >                  // In parallel, we need to visit *all* pairs of row
1182 >                  // & column indicies and will divide labor in the
1183 >                  // force evaluation later.
1184                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1185                    snap_->wrapVector(dr);
1186 <                  if (dr.lengthSquare() < rl2) {
1186 >                  cuts = getGroupCutoffs( (*j1), (*j2) );
1187 >                  if (dr.lengthSquare() < cuts.third) {
1188                      neighborList.push_back(make_pair((*j1), (*j2)));
1189 <                  }
1189 >                  }                  
1190                  }
1191                }
561 –            }
1192   #else
1193 <            for (vector<int>::iterator j1 = cellList_[m1].begin();
1194 <                 j1 != cellList_[m1].end(); ++j1) {
1195 <              for (vector<int>::iterator j2 = cellList_[m2].begin();
1196 <                   j2 != cellList_[m2].end(); ++j2) {
1197 <                              
1198 <                // Always do this if we're in different cells or if
1199 <                // we're in the same cell and the global index of the
1200 <                // j2 cutoff group is less than the j1 cutoff group
1201 <
1202 <                if (m2 != m1 || (*j2) < (*j1)) {
1203 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1204 <                  snap_->wrapVector(dr);
1205 <                  if (dr.lengthSquare() < rl2) {
1206 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1193 >              
1194 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1195 >                   j1 != cellList_[m1].end(); ++j1) {
1196 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1197 >                     j2 != cellList_[m2].end(); ++j2) {
1198 >                  
1199 >                  // Always do this if we're in different cells or if
1200 >                  // we're in the same cell and the global index of the
1201 >                  // j2 cutoff group is less than the j1 cutoff group
1202 >                  
1203 >                  if (m2 != m1 || (*j2) < (*j1)) {
1204 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1205 >                    snap_->wrapVector(dr);
1206 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1207 >                    if (dr.lengthSquare() < cuts.third) {
1208 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1209 >                    }
1210                    }
1211                  }
1212                }
580 –            }
1213   #endif
1214 +            }
1215            }
1216          }
1217        }
1218 +    } else {
1219 +      // branch to do all cutoff group pairs
1220 + #ifdef IS_MPI
1221 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1222 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {      
1223 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1224 +          snap_->wrapVector(dr);
1225 +          cuts = getGroupCutoffs( j1, j2 );
1226 +          if (dr.lengthSquare() < cuts.third) {
1227 +            neighborList.push_back(make_pair(j1, j2));
1228 +          }
1229 +        }
1230 +      }
1231 + #else
1232 +      for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1233 +        for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1234 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1235 +          snap_->wrapVector(dr);
1236 +          cuts = getGroupCutoffs( j1, j2 );
1237 +          if (dr.lengthSquare() < cuts.third) {
1238 +            neighborList.push_back(make_pair(j1, j2));
1239 +          }
1240 +        }
1241 +      }        
1242 + #endif
1243      }
1244 <
1244 >      
1245      // save the local cutoff group positions for the check that is
1246      // done on each loop:
1247      saved_CG_positions_.clear();
1248      for (int i = 0; i < nGroups_; i++)
1249        saved_CG_positions_.push_back(snap_->cgData.position[i]);
1250 <
1250 >    
1251      return neighborList;
1252    }
1253   } //end namespace OpenMD

Diff Legend

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