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 1569 by gezelter, Thu May 26 13:55:04 2011 UTC vs.
Revision 1665 by gezelter, Tue Nov 22 20:38:56 2011 UTC

# Line 36 | Line 36
36   * [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).            
37   * [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).          
38   * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).          
39 < * [4]  Vardeman & Gezelter, in progress (2009).                        
39 > * [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40 > * [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41   */
42   #include "parallel/ForceMatrixDecomposition.hpp"
43   #include "math/SquareMatrix3.hpp"
44   #include "nonbonded/NonBondedInteraction.hpp"
45   #include "brains/SnapshotManager.hpp"
46 + #include "brains/PairList.hpp"
47  
48   using namespace std;
49   namespace OpenMD {
50  
51 +  ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52 +
53 +    // In a parallel computation, row and colum scans must visit all
54 +    // surrounding cells (not just the 14 upper triangular blocks that
55 +    // are used when the processor can see all pairs)
56 + #ifdef IS_MPI
57 +    cellOffsets_.clear();
58 +    cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59 +    cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60 +    cellOffsets_.push_back( Vector3i( 1,-1,-1) );                          
61 +    cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62 +    cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63 +    cellOffsets_.push_back( Vector3i( 1, 0,-1) );
64 +    cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65 +    cellOffsets_.push_back( Vector3i( 0, 1,-1) );      
66 +    cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67 +    cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68 +    cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69 +    cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70 +    cellOffsets_.push_back( Vector3i(-1, 0, 0) );      
71 +    cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72 +    cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73 +    cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74 +    cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75 +    cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76 +    cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77 +    cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78 +    cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79 +    cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80 +    cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81 +    cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82 +    cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83 +    cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84 +    cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85 + #endif    
86 +  }
87 +
88 +
89    /**
90     * distributeInitialData is essentially a copy of the older fortran
91     * SimulationSetup
92     */
53 –  
93    void ForceMatrixDecomposition::distributeInitialData() {
94      snap_ = sman_->getCurrentSnapshot();
95      storageLayout_ = sman_->getStorageLayout();
96 +    ff_ = info_->getForceField();
97      nLocal_ = snap_->getNumberOfAtoms();
98 <    nGroups_ = snap_->getNumberOfCutoffGroups();
99 <
98 >    
99 >    nGroups_ = info_->getNLocalCutoffGroups();
100      // gather the information for atomtype IDs (atids):
101 <    vector<int> identsLocal = info_->getIdentArray();
101 >    idents = info_->getIdentArray();
102      AtomLocalToGlobal = info_->getGlobalAtomIndices();
103      cgLocalToGlobal = info_->getGlobalGroupIndices();
104      vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
65 –    vector<RealType> massFactorsLocal = info_->getMassFactors();
66 –    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
105  
106 +    massFactors = info_->getMassFactors();
107 +
108 +    PairList* excludes = info_->getExcludedInteractions();
109 +    PairList* oneTwo = info_->getOneTwoInteractions();
110 +    PairList* oneThree = info_->getOneThreeInteractions();
111 +    PairList* oneFour = info_->getOneFourInteractions();
112 +
113   #ifdef IS_MPI
114  
115 <    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
116 <    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
72 <    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
73 <    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
115 >    MPI::Intracomm row = rowComm.getComm();
116 >    MPI::Intracomm col = colComm.getComm();
117  
118 <    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
119 <    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
120 <    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
121 <    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
118 >    AtomPlanIntRow = new Plan<int>(row, nLocal_);
119 >    AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
120 >    AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
121 >    AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
122 >    AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
123  
124 <    cgCommIntRow = new Communicator<Row,int>(nGroups_);
125 <    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
126 <    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
127 <    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
124 >    AtomPlanIntColumn = new Plan<int>(col, nLocal_);
125 >    AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
126 >    AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
127 >    AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
128 >    AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
129  
130 <    nAtomsInRow_ = AtomCommIntRow->getSize();
131 <    nAtomsInCol_ = AtomCommIntColumn->getSize();
132 <    nGroupsInRow_ = cgCommIntRow->getSize();
133 <    nGroupsInCol_ = cgCommIntColumn->getSize();
130 >    cgPlanIntRow = new Plan<int>(row, nGroups_);
131 >    cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
132 >    cgPlanIntColumn = new Plan<int>(col, nGroups_);
133 >    cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
134  
135 +    nAtomsInRow_ = AtomPlanIntRow->getSize();
136 +    nAtomsInCol_ = AtomPlanIntColumn->getSize();
137 +    nGroupsInRow_ = cgPlanIntRow->getSize();
138 +    nGroupsInCol_ = cgPlanIntColumn->getSize();
139 +
140      // Modify the data storage objects with the correct layouts and sizes:
141      atomRowData.resize(nAtomsInRow_);
142      atomRowData.setStorageLayout(storageLayout_);
# Line 96 | Line 146 | namespace OpenMD {
146      cgRowData.setStorageLayout(DataStorage::dslPosition);
147      cgColData.resize(nGroupsInCol_);
148      cgColData.setStorageLayout(DataStorage::dslPosition);
149 +        
150 +    identsRow.resize(nAtomsInRow_);
151 +    identsCol.resize(nAtomsInCol_);
152      
153 <    vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
154 <                                      vector<RealType> (nAtomsInRow_, 0.0));
102 <    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
103 <                                      vector<RealType> (nAtomsInCol_, 0.0));
153 >    AtomPlanIntRow->gather(idents, identsRow);
154 >    AtomPlanIntColumn->gather(idents, identsCol);
155      
156 <    identsRow.reserve(nAtomsInRow_);
157 <    identsCol.reserve(nAtomsInCol_);
158 <    
108 <    AtomCommIntRow->gather(identsLocal, identsRow);
109 <    AtomCommIntColumn->gather(identsLocal, identsCol);
110 <    
111 <    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
112 <    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
113 <    
114 <    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
115 <    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
156 >    // allocate memory for the parallel objects
157 >    atypesRow.resize(nAtomsInRow_);
158 >    atypesCol.resize(nAtomsInCol_);
159  
160 <    AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
161 <    AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
160 >    for (int i = 0; i < nAtomsInRow_; i++)
161 >      atypesRow[i] = ff_->getAtomType(identsRow[i]);
162 >    for (int i = 0; i < nAtomsInCol_; i++)
163 >      atypesCol[i] = ff_->getAtomType(identsCol[i]);        
164  
165 +    pot_row.resize(nAtomsInRow_);
166 +    pot_col.resize(nAtomsInCol_);
167 +
168 +    AtomRowToGlobal.resize(nAtomsInRow_);
169 +    AtomColToGlobal.resize(nAtomsInCol_);
170 +    AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
171 +    AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
172 +
173 +    cgRowToGlobal.resize(nGroupsInRow_);
174 +    cgColToGlobal.resize(nGroupsInCol_);
175 +    cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
176 +    cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
177 +
178 +    massFactorsRow.resize(nAtomsInRow_);
179 +    massFactorsCol.resize(nAtomsInCol_);
180 +    AtomPlanRealRow->gather(massFactors, massFactorsRow);
181 +    AtomPlanRealColumn->gather(massFactors, massFactorsCol);
182 +
183      groupListRow_.clear();
184 <    groupListRow_.reserve(nGroupsInRow_);
184 >    groupListRow_.resize(nGroupsInRow_);
185      for (int i = 0; i < nGroupsInRow_; i++) {
186        int gid = cgRowToGlobal[i];
187        for (int j = 0; j < nAtomsInRow_; j++) {
# Line 129 | Line 192 | namespace OpenMD {
192      }
193  
194      groupListCol_.clear();
195 <    groupListCol_.reserve(nGroupsInCol_);
195 >    groupListCol_.resize(nGroupsInCol_);
196      for (int i = 0; i < nGroupsInCol_; i++) {
197        int gid = cgColToGlobal[i];
198        for (int j = 0; j < nAtomsInCol_; j++) {
# Line 139 | Line 202 | namespace OpenMD {
202        }      
203      }
204  
205 +    excludesForAtom.clear();
206 +    excludesForAtom.resize(nAtomsInRow_);
207 +    toposForAtom.clear();
208 +    toposForAtom.resize(nAtomsInRow_);
209 +    topoDist.clear();
210 +    topoDist.resize(nAtomsInRow_);
211 +    for (int i = 0; i < nAtomsInRow_; i++) {
212 +      int iglob = AtomRowToGlobal[i];
213 +
214 +      for (int j = 0; j < nAtomsInCol_; j++) {
215 +        int jglob = AtomColToGlobal[j];
216 +
217 +        if (excludes->hasPair(iglob, jglob))
218 +          excludesForAtom[i].push_back(j);      
219 +        
220 +        if (oneTwo->hasPair(iglob, jglob)) {
221 +          toposForAtom[i].push_back(j);
222 +          topoDist[i].push_back(1);
223 +        } else {
224 +          if (oneThree->hasPair(iglob, jglob)) {
225 +            toposForAtom[i].push_back(j);
226 +            topoDist[i].push_back(2);
227 +          } else {
228 +            if (oneFour->hasPair(iglob, jglob)) {
229 +              toposForAtom[i].push_back(j);
230 +              topoDist[i].push_back(3);
231 +            }
232 +          }
233 +        }
234 +      }      
235 +    }
236 +
237 + #else
238 +    excludesForAtom.clear();
239 +    excludesForAtom.resize(nLocal_);
240 +    toposForAtom.clear();
241 +    toposForAtom.resize(nLocal_);
242 +    topoDist.clear();
243 +    topoDist.resize(nLocal_);
244 +
245 +    for (int i = 0; i < nLocal_; i++) {
246 +      int iglob = AtomLocalToGlobal[i];
247 +
248 +      for (int j = 0; j < nLocal_; j++) {
249 +        int jglob = AtomLocalToGlobal[j];
250 +
251 +        if (excludes->hasPair(iglob, jglob))
252 +          excludesForAtom[i].push_back(j);              
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_.reserve(nGroups_);
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)
284 >        if (globalGroupMembership[aid] == gid) {
285            groupList_[i].push_back(j);
286 +        }
287        }      
288      }
289  
290 +
291 +    createGtypeCutoffMap();
292 +
293 +  }
294    
295 <    // still need:
296 <    // topoDist
297 <    // exclude
295 >  void ForceMatrixDecomposition::createGtypeCutoffMap() {
296 >    
297 >    RealType tol = 1e-6;
298 >    largestRcut_ = 0.0;
299 >    RealType rc;
300 >    int atid;
301 >    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
302 >    
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 >    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 +      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 203 | 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 243 | 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     */
# Line 348 | Line 800 | namespace OpenMD {
800   #ifdef IS_MPI
801      return massFactorsRow[atom1];
802   #else
803 <    return massFactorsLocal[atom1];
803 >    return massFactors[atom1];
804   #endif
805    }
806  
# Line 356 | Line 808 | namespace OpenMD {
808   #ifdef IS_MPI
809      return massFactorsCol[atom2];
810   #else
811 <    return massFactorsLocal[atom2];
811 >    return massFactors[atom2];
812   #endif
813  
814    }
# Line 374 | 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 + #else
845 +    unique_id_1 = AtomLocalToGlobal[atom1];
846 +    unique_id_2 = AtomLocalToGlobal[atom2];
847 + #endif  
848 +
849 +    if (unique_id_1 == unique_id_2) return true;
850 +
851 + #ifdef IS_MPI
852 +    // this prevents us from doing the pair on multiple processors
853 +    if (unique_id_1 < unique_id_2) {
854 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
855 +    } else {
856 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
857 +    }
858 + #endif
859 +    
860 +    return false;
861 +  }
862 +
863 +  /**
864 +   * We need to handle the interactions for atoms who are involved in
865 +   * the same rigid body as well as some short range interactions
866 +   * (bonds, bends, torsions) differently from other interactions.
867 +   * We'll still visit the pairwise routines, but with a flag that
868 +   * tells those routines to exclude the pair from direct long range
869 +   * interactions.  Some indirect interactions (notably reaction
870 +   * field) must still be handled for these pairs.
871 +   */
872 +  bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
873 +
874 +    // excludesForAtom was constructed to use row/column indices in the MPI
875 +    // version, and to use local IDs in the non-MPI version:
876 +    
877 +    for (vector<int>::iterator i = excludesForAtom[atom1].begin();
878 +         i != excludesForAtom[atom1].end(); ++i) {
879 +      if ( (*i) == atom2 ) return true;
880 +    }
881 +
882 +    return false;
883 +  }
884 +
885 +
886    void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
887   #ifdef IS_MPI
888      atomRowData.force[atom1] += fg;
# Line 391 | Line 900 | namespace OpenMD {
900    }
901  
902      // filling interaction blocks with pointers
903 <  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
904 <    InteractionData idat;
903 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
904 >                                                     int atom1, int atom2) {
905  
906 +    idat.excluded = excludeAtomPair(atom1, atom2);
907 +  
908   #ifdef IS_MPI
909 +    idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
910 +    //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
911 +    //                         ff_->getAtomType(identsCol[atom2]) );
912 +    
913      if (storageLayout_ & DataStorage::dslAmat) {
914        idat.A1 = &(atomRowData.aMat[atom1]);
915        idat.A2 = &(atomColData.aMat[atom2]);
# Line 415 | Line 930 | namespace OpenMD {
930        idat.rho2 = &(atomColData.density[atom2]);
931      }
932  
933 +    if (storageLayout_ & DataStorage::dslFunctional) {
934 +      idat.frho1 = &(atomRowData.functional[atom1]);
935 +      idat.frho2 = &(atomColData.functional[atom2]);
936 +    }
937 +
938      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
939        idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
940        idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
941      }
942 +
943 +    if (storageLayout_ & DataStorage::dslParticlePot) {
944 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
945 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
946 +    }
947 +
948 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {              
949 +      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
950 +      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
951 +    }
952 +
953   #else
954 +
955 +    idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
956 +    //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
957 +    //                         ff_->getAtomType(idents[atom2]) );
958 +
959      if (storageLayout_ & DataStorage::dslAmat) {
960        idat.A1 = &(snap_->atomData.aMat[atom1]);
961        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 435 | Line 971 | namespace OpenMD {
971        idat.t2 = &(snap_->atomData.torque[atom2]);
972      }
973  
974 <    if (storageLayout_ & DataStorage::dslDensity) {
974 >    if (storageLayout_ & DataStorage::dslDensity) {    
975        idat.rho1 = &(snap_->atomData.density[atom1]);
976        idat.rho2 = &(snap_->atomData.density[atom2]);
977      }
978  
979 +    if (storageLayout_ & DataStorage::dslFunctional) {
980 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
981 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
982 +    }
983 +
984      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
985        idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
986        idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
987      }
988 +
989 +    if (storageLayout_ & DataStorage::dslParticlePot) {
990 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
991 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
992 +    }
993 +
994 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {
995 +      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
996 +      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
997 +    }
998   #endif
448 –    return idat;
999    }
1000  
1001 <  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
1002 <
453 <    InteractionData idat;
1001 >  
1002 >  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
1003   #ifdef IS_MPI
1004 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
1005 <      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1006 <      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1007 <    }
1008 <    if (storageLayout_ & DataStorage::dslTorque) {
460 <      idat.t1 = &(atomRowData.torque[atom1]);
461 <      idat.t2 = &(atomColData.torque[atom2]);
462 <    }
463 <    if (storageLayout_ & DataStorage::dslForce) {
464 <      idat.t1 = &(atomRowData.force[atom1]);
465 <      idat.t2 = &(atomColData.force[atom2]);
466 <    }
1004 >    pot_row[atom1] += 0.5 *  *(idat.pot);
1005 >    pot_col[atom2] += 0.5 *  *(idat.pot);
1006 >
1007 >    atomRowData.force[atom1] += *(idat.f1);
1008 >    atomColData.force[atom2] -= *(idat.f1);
1009   #else
1010 <    if (storageLayout_ & DataStorage::dslElectroFrame) {
1011 <      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1012 <      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1013 <    }
472 <    if (storageLayout_ & DataStorage::dslTorque) {
473 <      idat.t1 = &(snap_->atomData.torque[atom1]);
474 <      idat.t2 = &(snap_->atomData.torque[atom2]);
475 <    }
476 <    if (storageLayout_ & DataStorage::dslForce) {
477 <      idat.t1 = &(snap_->atomData.force[atom1]);
478 <      idat.t2 = &(snap_->atomData.force[atom2]);
479 <    }
1010 >    pairwisePot += *(idat.pot);
1011 >
1012 >    snap_->atomData.force[atom1] += *(idat.f1);
1013 >    snap_->atomData.force[atom2] -= *(idat.f1);
1014   #endif
1015      
1016    }
1017  
484 –
485 –
486 –
1018    /*
1019     * buildNeighborList
1020     *
# Line 493 | Line 1024 | namespace OpenMD {
1024    vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1025        
1026      vector<pair<int, int> > neighborList;
1027 +    groupCutoffs cuts;
1028 +    bool doAllPairs = false;
1029 +
1030   #ifdef IS_MPI
1031      cellListRow_.clear();
1032      cellListCol_.clear();
# Line 500 | Line 1034 | namespace OpenMD {
1034      cellList_.clear();
1035   #endif
1036  
1037 <    // dangerous to not do error checking.
504 <    RealType rCut_;
505 <
506 <    RealType rList_ = (rCut_ + skinThickness_);
1037 >    RealType rList_ = (largestRcut_ + skinThickness_);
1038      RealType rl2 = rList_ * rList_;
1039      Snapshot* snap_ = sman_->getCurrentSnapshot();
1040      Mat3x3d Hmat = snap_->getHmat();
# Line 515 | Line 1046 | namespace OpenMD {
1046      nCells_.y() = (int) ( Hy.length() )/ rList_;
1047      nCells_.z() = (int) ( Hz.length() )/ rList_;
1048  
1049 +    // handle small boxes where the cell offsets can end up repeating cells
1050 +    
1051 +    if (nCells_.x() < 3) doAllPairs = true;
1052 +    if (nCells_.y() < 3) doAllPairs = true;
1053 +    if (nCells_.z() < 3) doAllPairs = true;
1054 +
1055      Mat3x3d invHmat = snap_->getInvHmat();
1056      Vector3d rs, scaled, dr;
1057      Vector3i whichCell;
1058      int cellIndex;
1059 +    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1060  
1061   #ifdef IS_MPI
1062 <    for (int i = 0; i < nGroupsInRow_; i++) {
1063 <      rs = cgRowData.position[i];
1064 <      // scaled positions relative to the box vectors
1065 <      scaled = invHmat * rs;
1066 <      // wrap the vector back into the unit box by subtracting integer box
529 <      // numbers
530 <      for (int j = 0; j < 3; j++)
531 <        scaled[j] -= roundMe(scaled[j]);
532 <    
533 <      // find xyz-indices of cell that cutoffGroup is in.
534 <      whichCell.x() = nCells_.x() * scaled.x();
535 <      whichCell.y() = nCells_.y() * scaled.y();
536 <      whichCell.z() = nCells_.z() * scaled.z();
1062 >    cellListRow_.resize(nCtot);
1063 >    cellListCol_.resize(nCtot);
1064 > #else
1065 >    cellList_.resize(nCtot);
1066 > #endif
1067  
1068 <      // find single index of this cell:
1069 <      cellIndex = Vlinear(whichCell, nCells_);
540 <      // add this cutoff group to the list of groups in this cell;
541 <      cellListRow_[cellIndex].push_back(i);
542 <    }
543 <
544 <    for (int i = 0; i < nGroupsInCol_; i++) {
545 <      rs = cgColData.position[i];
546 <      // scaled positions relative to the box vectors
547 <      scaled = invHmat * rs;
548 <      // wrap the vector back into the unit box by subtracting integer box
549 <      // numbers
550 <      for (int j = 0; j < 3; j++)
551 <        scaled[j] -= roundMe(scaled[j]);
1068 >    if (!doAllPairs) {
1069 > #ifdef IS_MPI
1070  
1071 <      // find xyz-indices of cell that cutoffGroup is in.
1072 <      whichCell.x() = nCells_.x() * scaled.x();
1073 <      whichCell.y() = nCells_.y() * scaled.y();
1074 <      whichCell.z() = nCells_.z() * scaled.z();
1075 <
1076 <      // find single index of this cell:
1077 <      cellIndex = Vlinear(whichCell, nCells_);
1078 <      // add this cutoff group to the list of groups in this cell;
1079 <      cellListCol_[cellIndex].push_back(i);
1080 <    }
1071 >      for (int i = 0; i < nGroupsInRow_; i++) {
1072 >        rs = cgRowData.position[i];
1073 >        
1074 >        // scaled positions relative to the box vectors
1075 >        scaled = invHmat * rs;
1076 >        
1077 >        // wrap the vector back into the unit box by subtracting integer box
1078 >        // numbers
1079 >        for (int j = 0; j < 3; j++) {
1080 >          scaled[j] -= roundMe(scaled[j]);
1081 >          scaled[j] += 0.5;
1082 >        }
1083 >        
1084 >        // find xyz-indices of cell that cutoffGroup is in.
1085 >        whichCell.x() = nCells_.x() * scaled.x();
1086 >        whichCell.y() = nCells_.y() * scaled.y();
1087 >        whichCell.z() = nCells_.z() * scaled.z();
1088 >        
1089 >        // find single index of this cell:
1090 >        cellIndex = Vlinear(whichCell, nCells_);
1091 >        
1092 >        // add this cutoff group to the list of groups in this cell;
1093 >        cellListRow_[cellIndex].push_back(i);
1094 >      }
1095 >      for (int i = 0; i < nGroupsInCol_; i++) {
1096 >        rs = cgColData.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 >        cellListCol_[cellIndex].push_back(i);
1118 >      }
1119 >    
1120   #else
1121 <    for (int i = 0; i < nGroups_; i++) {
1122 <      rs = snap_->cgData.position[i];
1123 <      // scaled positions relative to the box vectors
1124 <      scaled = invHmat * rs;
1125 <      // wrap the vector back into the unit box by subtracting integer box
1126 <      // numbers
1127 <      for (int j = 0; j < 3; j++)
1128 <        scaled[j] -= roundMe(scaled[j]);
1121 >      for (int i = 0; i < nGroups_; i++) {
1122 >        rs = snap_->cgData.position[i];
1123 >        
1124 >        // scaled positions relative to the box vectors
1125 >        scaled = invHmat * rs;
1126 >        
1127 >        // wrap the vector back into the unit box by subtracting integer box
1128 >        // numbers
1129 >        for (int j = 0; j < 3; j++) {
1130 >          scaled[j] -= roundMe(scaled[j]);
1131 >          scaled[j] += 0.5;
1132 >        }
1133 >        
1134 >        // find xyz-indices of cell that cutoffGroup is in.
1135 >        whichCell.x() = nCells_.x() * scaled.x();
1136 >        whichCell.y() = nCells_.y() * scaled.y();
1137 >        whichCell.z() = nCells_.z() * scaled.z();
1138 >        
1139 >        // find single index of this cell:
1140 >        cellIndex = Vlinear(whichCell, nCells_);
1141 >        
1142 >        // add this cutoff group to the list of groups in this cell;
1143 >        cellList_[cellIndex].push_back(i);
1144 >      }
1145  
573 –      // find xyz-indices of cell that cutoffGroup is in.
574 –      whichCell.x() = nCells_.x() * scaled.x();
575 –      whichCell.y() = nCells_.y() * scaled.y();
576 –      whichCell.z() = nCells_.z() * scaled.z();
577 –
578 –      // find single index of this cell:
579 –      cellIndex = Vlinear(whichCell, nCells_);
580 –      // add this cutoff group to the list of groups in this cell;
581 –      cellList_[cellIndex].push_back(i);
582 –    }
1146   #endif
1147  
1148 <
1149 <
1150 <    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1151 <      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1152 <        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
590 <          Vector3i m1v(m1x, m1y, m1z);
591 <          int m1 = Vlinear(m1v, nCells_);
592 <
593 <          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
594 <               os != cellOffsets_.end(); ++os) {
1148 >      for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1149 >        for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1150 >          for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1151 >            Vector3i m1v(m1x, m1y, m1z);
1152 >            int m1 = Vlinear(m1v, nCells_);
1153              
1154 <            Vector3i m2v = m1v + (*os);
1155 <            
1156 <            if (m2v.x() >= nCells_.x()) {
1157 <              m2v.x() = 0;          
1158 <            } else if (m2v.x() < 0) {
601 <              m2v.x() = nCells_.x() - 1;
602 <            }
603 <            
604 <            if (m2v.y() >= nCells_.y()) {
605 <              m2v.y() = 0;          
606 <            } else if (m2v.y() < 0) {
607 <              m2v.y() = nCells_.y() - 1;
608 <            }
609 <            
610 <            if (m2v.z() >= nCells_.z()) {
611 <              m2v.z() = 0;          
612 <            } else if (m2v.z() < 0) {
613 <              m2v.z() = nCells_.z() - 1;
614 <            }
615 <            
616 <            int m2 = Vlinear (m2v, nCells_);
1154 >            for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1155 >                 os != cellOffsets_.end(); ++os) {
1156 >              
1157 >              Vector3i m2v = m1v + (*os);
1158 >            
1159  
1160 < #ifdef IS_MPI
1161 <            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1162 <                 j1 != cellListRow_[m1].end(); ++j1) {
1163 <              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1164 <                   j2 != cellListCol_[m2].end(); ++j2) {
1165 <                              
1166 <                // Always do this if we're in different cells or if
1167 <                // we're in the same cell and the global index of the
1168 <                // j2 cutoff group is less than the j1 cutoff group
1160 >              if (m2v.x() >= nCells_.x()) {
1161 >                m2v.x() = 0;          
1162 >              } else if (m2v.x() < 0) {
1163 >                m2v.x() = nCells_.x() - 1;
1164 >              }
1165 >              
1166 >              if (m2v.y() >= nCells_.y()) {
1167 >                m2v.y() = 0;          
1168 >              } else if (m2v.y() < 0) {
1169 >                m2v.y() = nCells_.y() - 1;
1170 >              }
1171 >              
1172 >              if (m2v.z() >= nCells_.z()) {
1173 >                m2v.z() = 0;          
1174 >              } else if (m2v.z() < 0) {
1175 >                m2v.z() = nCells_.z() - 1;
1176 >              }
1177  
1178 <                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1178 >              int m2 = Vlinear (m2v, nCells_);
1179 >              
1180 > #ifdef IS_MPI
1181 >              for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1182 >                   j1 != cellListRow_[m1].end(); ++j1) {
1183 >                for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1184 >                     j2 != cellListCol_[m2].end(); ++j2) {
1185 >                  
1186 >                  // In parallel, we need to visit *all* pairs of row
1187 >                  // & column indicies and will divide labor in the
1188 >                  // force evaluation later.
1189                    dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1190                    snap_->wrapVector(dr);
1191 <                  if (dr.lengthSquare() < rl2) {
1191 >                  cuts = getGroupCutoffs( (*j1), (*j2) );
1192 >                  if (dr.lengthSquare() < cuts.third) {
1193                      neighborList.push_back(make_pair((*j1), (*j2)));
1194 <                  }
1194 >                  }                  
1195                  }
1196                }
636 –            }
1197   #else
1198 <            for (vector<int>::iterator j1 = cellList_[m1].begin();
1199 <                 j1 != cellList_[m1].end(); ++j1) {
1200 <              for (vector<int>::iterator j2 = cellList_[m2].begin();
1201 <                   j2 != cellList_[m2].end(); ++j2) {
1202 <                              
1203 <                // Always do this if we're in different cells or if
1204 <                // we're in the same cell and the global index of the
1205 <                // j2 cutoff group is less than the j1 cutoff group
1198 >              for (vector<int>::iterator j1 = cellList_[m1].begin();
1199 >                   j1 != cellList_[m1].end(); ++j1) {
1200 >                for (vector<int>::iterator j2 = cellList_[m2].begin();
1201 >                     j2 != cellList_[m2].end(); ++j2) {
1202 >    
1203 >                  // Always do this if we're in different cells or if
1204 >                  // we're in the same cell and the global index of
1205 >                  // the j2 cutoff group is greater than or equal to
1206 >                  // the j1 cutoff group.  Note that Rappaport's code
1207 >                  // has a "less than" conditional here, but that
1208 >                  // deals with atom-by-atom computation.  OpenMD
1209 >                  // allows atoms within a single cutoff group to
1210 >                  // interact with each other.
1211  
1212 <                if (m2 != m1 || (*j2) < (*j1)) {
1213 <                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1214 <                  snap_->wrapVector(dr);
1215 <                  if (dr.lengthSquare() < rl2) {
1216 <                    neighborList.push_back(make_pair((*j1), (*j2)));
1212 >
1213 >
1214 >                  if (m2 != m1 || (*j2) >= (*j1) ) {
1215 >
1216 >                    dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1217 >                    snap_->wrapVector(dr);
1218 >                    cuts = getGroupCutoffs( (*j1), (*j2) );
1219 >                    if (dr.lengthSquare() < cuts.third) {
1220 >                      neighborList.push_back(make_pair((*j1), (*j2)));
1221 >                    }
1222                    }
1223                  }
1224                }
655 –            }
1225   #endif
1226 +            }
1227            }
1228          }
1229        }
1230 +    } else {
1231 +      // branch to do all cutoff group pairs
1232 + #ifdef IS_MPI
1233 +      for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1234 +        for (int j2 = 0; j2 < nGroupsInCol_; j2++) {    
1235 +          dr = cgColData.position[j2] - cgRowData.position[j1];
1236 +          snap_->wrapVector(dr);
1237 +          cuts = getGroupCutoffs( j1, j2 );
1238 +          if (dr.lengthSquare() < cuts.third) {
1239 +            neighborList.push_back(make_pair(j1, j2));
1240 +          }
1241 +        }
1242 +      }      
1243 + #else
1244 +      // include all groups here.
1245 +      for (int j1 = 0; j1 < nGroups_; j1++) {
1246 +        // include self group interactions j2 == j1
1247 +        for (int j2 = j1; j2 < nGroups_; j2++) {
1248 +          dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1249 +          snap_->wrapVector(dr);
1250 +          cuts = getGroupCutoffs( j1, j2 );
1251 +          if (dr.lengthSquare() < cuts.third) {
1252 +            neighborList.push_back(make_pair(j1, j2));
1253 +          }
1254 +        }    
1255 +      }
1256 + #endif
1257      }
1258 <
1258 >      
1259      // save the local cutoff group positions for the check that is
1260      // done on each loop:
1261      saved_CG_positions_.clear();
1262      for (int i = 0; i < nGroups_; i++)
1263        saved_CG_positions_.push_back(snap_->cgData.position[i]);
1264 <
1264 >    
1265      return neighborList;
1266    }
1267   } //end namespace OpenMD

Diff Legend

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