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 1551 by gezelter, Thu Apr 28 18:38:21 2011 UTC vs.
Revision 1581 by gezelter, Mon Jun 13 22:13:12 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 {
# Line 54 | Line 55 | namespace OpenMD {
55    void ForceMatrixDecomposition::distributeInitialData() {
56      snap_ = sman_->getCurrentSnapshot();
57      storageLayout_ = sman_->getStorageLayout();
58 < #ifdef IS_MPI    
59 <    int nLocal = snap_->getNumberOfAtoms();
59 <    int nGroups = snap_->getNumberOfCutoffGroups();
60 <    
61 <    AtomCommIntRow = new Communicator<Row,int>(nLocal);
62 <    AtomCommRealRow = new Communicator<Row,RealType>(nLocal);
63 <    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal);
64 <    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal);
58 >    ff_ = info_->getForceField();
59 >    nLocal_ = snap_->getNumberOfAtoms();
60  
61 <    AtomCommIntColumn = new Communicator<Column,int>(nLocal);
62 <    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal);
63 <    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal);
64 <    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal);
61 >    nGroups_ = info_->getNLocalCutoffGroups();
62 >    cerr << "in dId, nGroups = " << nGroups_ << "\n";
63 >    // gather the information for atomtype IDs (atids):
64 >    identsLocal = info_->getIdentArray();
65 >    AtomLocalToGlobal = info_->getGlobalAtomIndices();
66 >    cgLocalToGlobal = info_->getGlobalGroupIndices();
67 >    vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
68 >    massFactors = info_->getMassFactors();
69 >    PairList excludes = info_->getExcludedInteractions();
70 >    PairList oneTwo = info_->getOneTwoInteractions();
71 >    PairList oneThree = info_->getOneThreeInteractions();
72 >    PairList oneFour = info_->getOneFourInteractions();
73  
74 <    cgCommIntRow = new Communicator<Row,int>(nGroups);
75 <    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups);
76 <    cgCommIntColumn = new Communicator<Column,int>(nGroups);
77 <    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups);
74 > #ifdef IS_MPI
75 >
76 >    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
77 >    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
78 >    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
79 >    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
80 >    AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
81  
82 <    int nAtomsInRow = AtomCommIntRow->getSize();
83 <    int nAtomsInCol = AtomCommIntColumn->getSize();
84 <    int nGroupsInRow = cgCommIntRow->getSize();
85 <    int nGroupsInCol = cgCommIntColumn->getSize();
82 >    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
83 >    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
84 >    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
85 >    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
86 >    AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
87  
88 +    cgCommIntRow = new Communicator<Row,int>(nGroups_);
89 +    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
90 +    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
91 +    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
92 +
93 +    nAtomsInRow_ = AtomCommIntRow->getSize();
94 +    nAtomsInCol_ = AtomCommIntColumn->getSize();
95 +    nGroupsInRow_ = cgCommIntRow->getSize();
96 +    nGroupsInCol_ = cgCommIntColumn->getSize();
97 +
98      // Modify the data storage objects with the correct layouts and sizes:
99 <    atomRowData.resize(nAtomsInRow);
99 >    atomRowData.resize(nAtomsInRow_);
100      atomRowData.setStorageLayout(storageLayout_);
101 <    atomColData.resize(nAtomsInCol);
101 >    atomColData.resize(nAtomsInCol_);
102      atomColData.setStorageLayout(storageLayout_);
103 <    cgRowData.resize(nGroupsInRow);
103 >    cgRowData.resize(nGroupsInRow_);
104      cgRowData.setStorageLayout(DataStorage::dslPosition);
105 <    cgColData.resize(nGroupsInCol);
105 >    cgColData.resize(nGroupsInCol_);
106      cgColData.setStorageLayout(DataStorage::dslPosition);
107 +        
108 +    identsRow.resize(nAtomsInRow_);
109 +    identsCol.resize(nAtomsInCol_);
110      
91 –    vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
92 –                                      vector<RealType> (nAtomsInRow, 0.0));
93 –    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
94 –                                      vector<RealType> (nAtomsInCol, 0.0));
95 –
96 –
97 –    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
98 –    
99 –    // gather the information for atomtype IDs (atids):
100 –    vector<int> identsLocal = info_->getIdentArray();
101 –    identsRow.reserve(nAtomsInRow);
102 –    identsCol.reserve(nAtomsInCol);
103 –    
111      AtomCommIntRow->gather(identsLocal, identsRow);
112      AtomCommIntColumn->gather(identsLocal, identsCol);
113      
107 –    AtomLocalToGlobal = info_->getGlobalAtomIndices();
114      AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
115      AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
116      
111 –    cgLocalToGlobal = info_->getGlobalGroupIndices();
117      cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
118      cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
119  
120 <    // still need:
121 <    // topoDist
122 <    // exclude
120 >    AtomCommRealRow->gather(massFactors, massFactorsRow);
121 >    AtomCommRealColumn->gather(massFactors, massFactorsCol);
122 >
123 >    groupListRow_.clear();
124 >    groupListRow_.resize(nGroupsInRow_);
125 >    for (int i = 0; i < nGroupsInRow_; i++) {
126 >      int gid = cgRowToGlobal[i];
127 >      for (int j = 0; j < nAtomsInRow_; j++) {
128 >        int aid = AtomRowToGlobal[j];
129 >        if (globalGroupMembership[aid] == gid)
130 >          groupListRow_[i].push_back(j);
131 >      }      
132 >    }
133 >
134 >    groupListCol_.clear();
135 >    groupListCol_.resize(nGroupsInCol_);
136 >    for (int i = 0; i < nGroupsInCol_; i++) {
137 >      int gid = cgColToGlobal[i];
138 >      for (int j = 0; j < nAtomsInCol_; j++) {
139 >        int aid = AtomColToGlobal[j];
140 >        if (globalGroupMembership[aid] == gid)
141 >          groupListCol_[i].push_back(j);
142 >      }      
143 >    }
144 >
145 >    skipsForAtom.clear();
146 >    skipsForAtom.resize(nAtomsInRow_);
147 >    toposForAtom.clear();
148 >    toposForAtom.resize(nAtomsInRow_);
149 >    topoDist.clear();
150 >    topoDist.resize(nAtomsInRow_);
151 >    for (int i = 0; i < nAtomsInRow_; i++) {
152 >      int iglob = AtomRowToGlobal[i];
153 >
154 >      for (int j = 0; j < nAtomsInCol_; j++) {
155 >        int jglob = AtomColToGlobal[j];
156 >
157 >        if (excludes.hasPair(iglob, jglob))
158 >          skipsForAtom[i].push_back(j);      
159 >        
160 >        if (oneTwo.hasPair(iglob, jglob)) {
161 >          toposForAtom[i].push_back(j);
162 >          topoDist[i].push_back(1);
163 >        } else {
164 >          if (oneThree.hasPair(iglob, jglob)) {
165 >            toposForAtom[i].push_back(j);
166 >            topoDist[i].push_back(2);
167 >          } else {
168 >            if (oneFour.hasPair(iglob, jglob)) {
169 >              toposForAtom[i].push_back(j);
170 >              topoDist[i].push_back(3);
171 >            }
172 >          }
173 >        }
174 >      }      
175 >    }
176 >
177   #endif
178 +
179 +    groupList_.clear();
180 +    groupList_.resize(nGroups_);
181 +    for (int i = 0; i < nGroups_; i++) {
182 +      int gid = cgLocalToGlobal[i];
183 +      for (int j = 0; j < nLocal_; j++) {
184 +        int aid = AtomLocalToGlobal[j];
185 +        if (globalGroupMembership[aid] == gid) {
186 +          groupList_[i].push_back(j);
187 +        }
188 +      }      
189 +    }
190 +
191 +    skipsForAtom.clear();
192 +    skipsForAtom.resize(nLocal_);
193 +    toposForAtom.clear();
194 +    toposForAtom.resize(nLocal_);
195 +    topoDist.clear();
196 +    topoDist.resize(nLocal_);
197 +
198 +    for (int i = 0; i < nLocal_; i++) {
199 +      int iglob = AtomLocalToGlobal[i];
200 +
201 +      for (int j = 0; j < nLocal_; j++) {
202 +        int jglob = AtomLocalToGlobal[j];
203 +
204 +        if (excludes.hasPair(iglob, jglob))
205 +          skipsForAtom[i].push_back(j);              
206 +        
207 +        if (oneTwo.hasPair(iglob, jglob)) {
208 +          toposForAtom[i].push_back(j);
209 +          topoDist[i].push_back(1);
210 +        } else {
211 +          if (oneThree.hasPair(iglob, jglob)) {
212 +            toposForAtom[i].push_back(j);
213 +            topoDist[i].push_back(2);
214 +          } else {
215 +            if (oneFour.hasPair(iglob, jglob)) {
216 +              toposForAtom[i].push_back(j);
217 +              topoDist[i].push_back(3);
218 +            }
219 +          }
220 +        }
221 +      }      
222 +    }
223 +    
224 +    createGtypeCutoffMap();
225    }
226 +  
227 +  void ForceMatrixDecomposition::createGtypeCutoffMap() {
228 +
229 +    RealType tol = 1e-6;
230 +    RealType rc;
231 +    int atid;
232 +    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
233 +    vector<RealType> atypeCutoff;
234 +    atypeCutoff.resize( atypes.size() );
235 +
236 +    for (set<AtomType*>::iterator at = atypes.begin();
237 +         at != atypes.end(); ++at){
238 +      rc = interactionMan_->getSuggestedCutoffRadius(*at);
239 +      atid = (*at)->getIdent();
240 +      atypeCutoff[atid] = rc;
241 +    }
242 +
243 +    vector<RealType> gTypeCutoffs;
244 +
245 +    // first we do a single loop over the cutoff groups to find the
246 +    // largest cutoff for any atypes present in this group.
247 + #ifdef IS_MPI
248 +    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
249 +    groupRowToGtype.resize(nGroupsInRow_);
250 +    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
251 +      vector<int> atomListRow = getAtomsInGroupRow(cg1);
252 +      for (vector<int>::iterator ia = atomListRow.begin();
253 +           ia != atomListRow.end(); ++ia) {            
254 +        int atom1 = (*ia);
255 +        atid = identsRow[atom1];
256 +        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
257 +          groupCutoffRow[cg1] = atypeCutoff[atid];
258 +        }
259 +      }
260 +
261 +      bool gTypeFound = false;
262 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
263 +        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
264 +          groupRowToGtype[cg1] = gt;
265 +          gTypeFound = true;
266 +        }
267 +      }
268 +      if (!gTypeFound) {
269 +        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
270 +        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
271 +      }
272 +      
273 +    }
274 +    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
275 +    groupColToGtype.resize(nGroupsInCol_);
276 +    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
277 +      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
278 +      for (vector<int>::iterator jb = atomListCol.begin();
279 +           jb != atomListCol.end(); ++jb) {            
280 +        int atom2 = (*jb);
281 +        atid = identsCol[atom2];
282 +        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
283 +          groupCutoffCol[cg2] = atypeCutoff[atid];
284 +        }
285 +      }
286 +      bool gTypeFound = false;
287 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
288 +        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
289 +          groupColToGtype[cg2] = gt;
290 +          gTypeFound = true;
291 +        }
292 +      }
293 +      if (!gTypeFound) {
294 +        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
295 +        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
296 +      }
297 +    }
298 + #else
299 +
300 +    vector<RealType> groupCutoff(nGroups_, 0.0);
301 +    groupToGtype.resize(nGroups_);
302 +
303 +    cerr << "nGroups = " << nGroups_ << "\n";
304 +    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
305 +
306 +      groupCutoff[cg1] = 0.0;
307 +      vector<int> atomList = getAtomsInGroupRow(cg1);
308 +
309 +      for (vector<int>::iterator ia = atomList.begin();
310 +           ia != atomList.end(); ++ia) {            
311 +        int atom1 = (*ia);
312 +        atid = identsLocal[atom1];
313 +        if (atypeCutoff[atid] > groupCutoff[cg1]) {
314 +          groupCutoff[cg1] = atypeCutoff[atid];
315 +        }
316 +      }
317 +
318 +      bool gTypeFound = false;
319 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
320 +        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
321 +          groupToGtype[cg1] = gt;
322 +          gTypeFound = true;
323 +        }
324 +      }
325 +      if (!gTypeFound) {
326 +        gTypeCutoffs.push_back( groupCutoff[cg1] );
327 +        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
328 +      }      
329 +    }
330 + #endif
331 +
332 +    cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
333 +    // Now we find the maximum group cutoff value present in the simulation
334 +
335 +    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
336 +
337 + #ifdef IS_MPI
338 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
339 + #endif
340      
341 +    RealType tradRcut = groupMax;
342  
343 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
344 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
345 +        RealType thisRcut;
346 +        switch(cutoffPolicy_) {
347 +        case TRADITIONAL:
348 +          thisRcut = tradRcut;
349 +          break;
350 +        case MIX:
351 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
352 +          break;
353 +        case MAX:
354 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
355 +          break;
356 +        default:
357 +          sprintf(painCave.errMsg,
358 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
359 +                  "hit an unknown cutoff policy!\n");
360 +          painCave.severity = OPENMD_ERROR;
361 +          painCave.isFatal = 1;
362 +          simError();
363 +          break;
364 +        }
365  
366 +        pair<int,int> key = make_pair(i,j);
367 +        gTypeCutoffMap[key].first = thisRcut;
368 +
369 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
370 +
371 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
372 +        
373 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
374 +
375 +        // sanity check
376 +        
377 +        if (userChoseCutoff_) {
378 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
379 +            sprintf(painCave.errMsg,
380 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
381 +                    "user-specified rCut does not match computed group Cutoff\n");
382 +            painCave.severity = OPENMD_ERROR;
383 +            painCave.isFatal = 1;
384 +            simError();            
385 +          }
386 +        }
387 +      }
388 +    }
389 +  }
390 +
391 +
392 +  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
393 +    int i, j;  
394 + #ifdef IS_MPI
395 +    i = groupRowToGtype[cg1];
396 +    j = groupColToGtype[cg2];
397 + #else
398 +    i = groupToGtype[cg1];
399 +    j = groupToGtype[cg2];
400 + #endif    
401 +    return gTypeCutoffMap[make_pair(i,j)];
402 +  }
403 +
404 +  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
405 +    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
406 +      if (toposForAtom[atom1][j] == atom2)
407 +        return topoDist[atom1][j];
408 +    }
409 +    return 0;
410 +  }
411 +
412 +  void ForceMatrixDecomposition::zeroWorkArrays() {
413 +
414 +    for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
415 +      longRangePot_[j] = 0.0;
416 +    }
417 +
418 + #ifdef IS_MPI
419 +    if (storageLayout_ & DataStorage::dslForce) {
420 +      fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
421 +      fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
422 +    }
423 +
424 +    if (storageLayout_ & DataStorage::dslTorque) {
425 +      fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
426 +      fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
427 +    }
428 +    
429 +    fill(pot_row.begin(), pot_row.end(),
430 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
431 +
432 +    fill(pot_col.begin(), pot_col.end(),
433 +         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
434 +    
435 +    pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
436 +
437 +    if (storageLayout_ & DataStorage::dslParticlePot) {    
438 +      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
439 +      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
440 +    }
441 +
442 +    if (storageLayout_ & DataStorage::dslDensity) {      
443 +      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
444 +      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
445 +    }
446 +
447 +    if (storageLayout_ & DataStorage::dslFunctional) {  
448 +      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
449 +      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
450 +    }
451 +
452 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
453 +      fill(atomRowData.functionalDerivative.begin(),
454 +           atomRowData.functionalDerivative.end(), 0.0);
455 +      fill(atomColData.functionalDerivative.begin(),
456 +           atomColData.functionalDerivative.end(), 0.0);
457 +    }
458 +
459 + #else
460 +    
461 +    if (storageLayout_ & DataStorage::dslParticlePot) {      
462 +      fill(snap_->atomData.particlePot.begin(),
463 +           snap_->atomData.particlePot.end(), 0.0);
464 +    }
465 +    
466 +    if (storageLayout_ & DataStorage::dslDensity) {      
467 +      fill(snap_->atomData.density.begin(),
468 +           snap_->atomData.density.end(), 0.0);
469 +    }
470 +    if (storageLayout_ & DataStorage::dslFunctional) {
471 +      fill(snap_->atomData.functional.begin(),
472 +           snap_->atomData.functional.end(), 0.0);
473 +    }
474 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
475 +      fill(snap_->atomData.functionalDerivative.begin(),
476 +           snap_->atomData.functionalDerivative.end(), 0.0);
477 +    }
478 + #endif
479 +    
480 +  }
481 +
482 +
483    void ForceMatrixDecomposition::distributeData()  {
484      snap_ = sman_->getCurrentSnapshot();
485      storageLayout_ = sman_->getStorageLayout();
# Line 155 | Line 515 | namespace OpenMD {
515   #endif      
516    }
517    
518 +  /* collects information obtained during the pre-pair loop onto local
519 +   * data structures.
520 +   */
521    void ForceMatrixDecomposition::collectIntermediateData() {
522      snap_ = sman_->getCurrentSnapshot();
523      storageLayout_ = sman_->getStorageLayout();
# Line 166 | Line 529 | namespace OpenMD {
529                                 snap_->atomData.density);
530        
531        int n = snap_->atomData.density.size();
532 <      std::vector<RealType> rho_tmp(n, 0.0);
532 >      vector<RealType> rho_tmp(n, 0.0);
533        AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
534        for (int i = 0; i < n; i++)
535          snap_->atomData.density[i] += rho_tmp[i];
536      }
537   #endif
538    }
539 <  
539 >
540 >  /*
541 >   * redistributes information obtained during the pre-pair loop out to
542 >   * row and column-indexed data structures
543 >   */
544    void ForceMatrixDecomposition::distributeIntermediateData() {
545      snap_ = sman_->getCurrentSnapshot();
546      storageLayout_ = sman_->getStorageLayout();
# Line 229 | Line 596 | namespace OpenMD {
596          snap_->atomData.torque[i] += trq_tmp[i];
597      }
598      
599 <    int nLocal = snap_->getNumberOfAtoms();
599 >    nLocal_ = snap_->getNumberOfAtoms();
600  
601 <    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
602 <                                       vector<RealType> (nLocal, 0.0));
603 <    
604 <    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
605 <      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
606 <      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
607 <        pot_local[i] += pot_temp[i][ii];
608 <      }
609 <    }
601 >    vector<potVec> pot_temp(nLocal_,
602 >                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
603 >
604 >    // scatter/gather pot_row into the members of my column
605 >          
606 >    AtomCommPotRow->scatter(pot_row, pot_temp);
607 >
608 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
609 >      pot_local += pot_temp[ii];
610 >    
611 >    fill(pot_temp.begin(), pot_temp.end(),
612 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
613 >      
614 >    AtomCommPotColumn->scatter(pot_col, pot_temp);    
615 >    
616 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
617 >      pot_local += pot_temp[ii];
618 >    
619   #endif
620    }
621  
622 +  int ForceMatrixDecomposition::getNAtomsInRow() {  
623 + #ifdef IS_MPI
624 +    return nAtomsInRow_;
625 + #else
626 +    return nLocal_;
627 + #endif
628 +  }
629 +
630 +  /**
631 +   * returns the list of atoms belonging to this group.  
632 +   */
633 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
634 + #ifdef IS_MPI
635 +    return groupListRow_[cg1];
636 + #else
637 +    return groupList_[cg1];
638 + #endif
639 +  }
640 +
641 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
642 + #ifdef IS_MPI
643 +    return groupListCol_[cg2];
644 + #else
645 +    return groupList_[cg2];
646 + #endif
647 +  }
648    
649    Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
650      Vector3d d;
# Line 284 | Line 686 | namespace OpenMD {
686      snap_->wrapVector(d);
687      return d;    
688    }
689 +
690 +  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
691 + #ifdef IS_MPI
692 +    return massFactorsRow[atom1];
693 + #else
694 +    return massFactors[atom1];
695 + #endif
696 +  }
697 +
698 +  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
699 + #ifdef IS_MPI
700 +    return massFactorsCol[atom2];
701 + #else
702 +    return massFactors[atom2];
703 + #endif
704 +
705 +  }
706      
707    Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
708      Vector3d d;
# Line 298 | Line 717 | namespace OpenMD {
717      return d;    
718    }
719  
720 +  vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
721 +    return skipsForAtom[atom1];
722 +  }
723 +
724 +  /**
725 +   * There are a number of reasons to skip a pair or a
726 +   * particle. Mostly we do this to exclude atoms who are involved in
727 +   * short range interactions (bonds, bends, torsions), but we also
728 +   * need to exclude some overcounted interactions that result from
729 +   * the parallel decomposition.
730 +   */
731 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
732 +    int unique_id_1, unique_id_2;
733 +
734 + #ifdef IS_MPI
735 +    // in MPI, we have to look up the unique IDs for each atom
736 +    unique_id_1 = AtomRowToGlobal[atom1];
737 +    unique_id_2 = AtomColToGlobal[atom2];
738 +
739 +    // this situation should only arise in MPI simulations
740 +    if (unique_id_1 == unique_id_2) return true;
741 +    
742 +    // this prevents us from doing the pair on multiple processors
743 +    if (unique_id_1 < unique_id_2) {
744 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
745 +    } else {
746 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
747 +    }
748 + #else
749 +    // in the normal loop, the atom numbers are unique
750 +    unique_id_1 = atom1;
751 +    unique_id_2 = atom2;
752 + #endif
753 +    
754 +    for (vector<int>::iterator i = skipsForAtom[atom1].begin();
755 +         i != skipsForAtom[atom1].end(); ++i) {
756 +      if ( (*i) == unique_id_2 ) return true;
757 +    }    
758 +
759 +  }
760 +
761 +
762    void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
763   #ifdef IS_MPI
764      atomRowData.force[atom1] += fg;
# Line 312 | Line 773 | namespace OpenMD {
773   #else
774      snap_->atomData.force[atom2] += fg;
775   #endif
315 –
776    }
777  
778      // filling interaction blocks with pointers
779 <  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
780 <
321 <    InteractionData idat;
779 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData idat,
780 >                                                     int atom1, int atom2) {    
781   #ifdef IS_MPI
782 +    
783 +    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
784 +                             ff_->getAtomType(identsCol[atom2]) );
785 +    
786      if (storageLayout_ & DataStorage::dslAmat) {
787 <      idat.A1 = atomRowData.aMat[atom1];
788 <      idat.A2 = atomColData.aMat[atom2];
787 >      idat.A1 = &(atomRowData.aMat[atom1]);
788 >      idat.A2 = &(atomColData.aMat[atom2]);
789      }
790 +    
791 +    if (storageLayout_ & DataStorage::dslElectroFrame) {
792 +      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
793 +      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
794 +    }
795  
796 +    if (storageLayout_ & DataStorage::dslTorque) {
797 +      idat.t1 = &(atomRowData.torque[atom1]);
798 +      idat.t2 = &(atomColData.torque[atom2]);
799 +    }
800 +
801 +    if (storageLayout_ & DataStorage::dslDensity) {
802 +      idat.rho1 = &(atomRowData.density[atom1]);
803 +      idat.rho2 = &(atomColData.density[atom2]);
804 +    }
805 +
806 +    if (storageLayout_ & DataStorage::dslFunctional) {
807 +      idat.frho1 = &(atomRowData.functional[atom1]);
808 +      idat.frho2 = &(atomColData.functional[atom2]);
809 +    }
810 +
811 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
812 +      idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
813 +      idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
814 +    }
815 +
816 +    if (storageLayout_ & DataStorage::dslParticlePot) {
817 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
818 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
819 +    }
820 +
821 + #else
822 +
823 +    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
824 +                             ff_->getAtomType(identsLocal[atom2]) );
825 +
826 +    if (storageLayout_ & DataStorage::dslAmat) {
827 +      idat.A1 = &(snap_->atomData.aMat[atom1]);
828 +      idat.A2 = &(snap_->atomData.aMat[atom2]);
829 +    }
830 +
831      if (storageLayout_ & DataStorage::dslElectroFrame) {
832 <      idat.eFrame1 = atomRowData.electroFrame[atom1];
833 <      idat.eFrame2 = atomColData.electroFrame[atom2];
832 >      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
833 >      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
834      }
835  
836      if (storageLayout_ & DataStorage::dslTorque) {
837 <      idat.t1 = atomRowData.torque[atom1];
838 <      idat.t2 = atomColData.torque[atom2];
837 >      idat.t1 = &(snap_->atomData.torque[atom1]);
838 >      idat.t2 = &(snap_->atomData.torque[atom2]);
839      }
840  
841      if (storageLayout_ & DataStorage::dslDensity) {
842 <      idat.rho1 = atomRowData.density[atom1];
843 <      idat.rho2 = atomColData.density[atom2];
842 >      idat.rho1 = &(snap_->atomData.density[atom1]);
843 >      idat.rho2 = &(snap_->atomData.density[atom2]);
844      }
845  
846 +    if (storageLayout_ & DataStorage::dslFunctional) {
847 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
848 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
849 +    }
850 +
851      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
852 <      idat.dfrho1 = atomRowData.functionalDerivative[atom1];
853 <      idat.dfrho2 = atomColData.functionalDerivative[atom2];
852 >      idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
853 >      idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
854      }
855 +
856 +    if (storageLayout_ & DataStorage::dslParticlePot) {
857 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
858 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
859 +    }
860 +
861   #endif
862 +  }
863 +
864 +  
865 +  void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {    
866 + #ifdef IS_MPI
867 +    pot_row[atom1] += 0.5 *  *(idat.pot);
868 +    pot_col[atom2] += 0.5 *  *(idat.pot);
869 +
870 +    atomRowData.force[atom1] += *(idat.f1);
871 +    atomColData.force[atom2] -= *(idat.f1);
872 + #else
873 +    longRangePot_ += *(idat.pot);
874      
875 +    snap_->atomData.force[atom1] += *(idat.f1);
876 +    snap_->atomData.force[atom2] -= *(idat.f1);
877 + #endif
878 +
879    }
880 <  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
881 <  }
882 <  SelfData ForceMatrixDecomposition::fillSelfData(int atom1) {
880 >
881 >
882 >  void ForceMatrixDecomposition::fillSkipData(InteractionData idat,
883 >                                              int atom1, int atom2) {
884 > #ifdef IS_MPI
885 >    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
886 >                             ff_->getAtomType(identsCol[atom2]) );
887 >
888 >    if (storageLayout_ & DataStorage::dslElectroFrame) {
889 >      idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
890 >      idat.eFrame2 = &(atomColData.electroFrame[atom2]);
891 >    }
892 >    if (storageLayout_ & DataStorage::dslTorque) {
893 >      idat.t1 = &(atomRowData.torque[atom1]);
894 >      idat.t2 = &(atomColData.torque[atom2]);
895 >    }
896 > #else
897 >    idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
898 >                             ff_->getAtomType(identsLocal[atom2]) );
899 >
900 >    if (storageLayout_ & DataStorage::dslElectroFrame) {
901 >      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
902 >      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
903 >    }
904 >    if (storageLayout_ & DataStorage::dslTorque) {
905 >      idat.t1 = &(snap_->atomData.torque[atom1]);
906 >      idat.t2 = &(snap_->atomData.torque[atom2]);
907 >    }
908 > #endif    
909    }
910  
911 <  
911 >  /*
912 >   * buildNeighborList
913 >   *
914 >   * first element of pair is row-indexed CutoffGroup
915 >   * second element of pair is column-indexed CutoffGroup
916 >   */
917 >  vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
918 >      
919 >    vector<pair<int, int> > neighborList;
920 >    groupCutoffs cuts;
921 > #ifdef IS_MPI
922 >    cellListRow_.clear();
923 >    cellListCol_.clear();
924 > #else
925 >    cellList_.clear();
926 > #endif
927 >
928 >    RealType rList_ = (largestRcut_ + skinThickness_);
929 >    RealType rl2 = rList_ * rList_;
930 >    Snapshot* snap_ = sman_->getCurrentSnapshot();
931 >    Mat3x3d Hmat = snap_->getHmat();
932 >    Vector3d Hx = Hmat.getColumn(0);
933 >    Vector3d Hy = Hmat.getColumn(1);
934 >    Vector3d Hz = Hmat.getColumn(2);
935 >
936 >    nCells_.x() = (int) ( Hx.length() )/ rList_;
937 >    nCells_.y() = (int) ( Hy.length() )/ rList_;
938 >    nCells_.z() = (int) ( Hz.length() )/ rList_;
939 >
940 >    Mat3x3d invHmat = snap_->getInvHmat();
941 >    Vector3d rs, scaled, dr;
942 >    Vector3i whichCell;
943 >    int cellIndex;
944 >    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
945 >
946 >    cerr << "flag1\n";
947 > #ifdef IS_MPI
948 >    cellListRow_.resize(nCtot);
949 >    cellListCol_.resize(nCtot);
950 > #else
951 >    cellList_.resize(nCtot);
952 > #endif
953 >    cerr << "flag2\n";
954 > #ifdef IS_MPI
955 >    for (int i = 0; i < nGroupsInRow_; i++) {
956 >      rs = cgRowData.position[i];
957 >
958 >      // scaled positions relative to the box vectors
959 >      scaled = invHmat * rs;
960 >
961 >      // wrap the vector back into the unit box by subtracting integer box
962 >      // numbers
963 >      for (int j = 0; j < 3; j++) {
964 >        scaled[j] -= roundMe(scaled[j]);
965 >        scaled[j] += 0.5;
966 >      }
967 >    
968 >      // find xyz-indices of cell that cutoffGroup is in.
969 >      whichCell.x() = nCells_.x() * scaled.x();
970 >      whichCell.y() = nCells_.y() * scaled.y();
971 >      whichCell.z() = nCells_.z() * scaled.z();
972 >
973 >      // find single index of this cell:
974 >      cellIndex = Vlinear(whichCell, nCells_);
975 >
976 >      // add this cutoff group to the list of groups in this cell;
977 >      cellListRow_[cellIndex].push_back(i);
978 >    }
979 >
980 >    for (int i = 0; i < nGroupsInCol_; i++) {
981 >      rs = cgColData.position[i];
982 >
983 >      // scaled positions relative to the box vectors
984 >      scaled = invHmat * rs;
985 >
986 >      // wrap the vector back into the unit box by subtracting integer box
987 >      // numbers
988 >      for (int j = 0; j < 3; j++) {
989 >        scaled[j] -= roundMe(scaled[j]);
990 >        scaled[j] += 0.5;
991 >      }
992 >
993 >      // find xyz-indices of cell that cutoffGroup is in.
994 >      whichCell.x() = nCells_.x() * scaled.x();
995 >      whichCell.y() = nCells_.y() * scaled.y();
996 >      whichCell.z() = nCells_.z() * scaled.z();
997 >
998 >      // find single index of this cell:
999 >      cellIndex = Vlinear(whichCell, nCells_);
1000 >
1001 >      // add this cutoff group to the list of groups in this cell;
1002 >      cellListCol_[cellIndex].push_back(i);
1003 >    }
1004 > #else
1005 >    for (int i = 0; i < nGroups_; i++) {
1006 >      rs = snap_->cgData.position[i];
1007 >
1008 >      // scaled positions relative to the box vectors
1009 >      scaled = invHmat * rs;
1010 >
1011 >      // wrap the vector back into the unit box by subtracting integer box
1012 >      // numbers
1013 >      for (int j = 0; j < 3; j++) {
1014 >        scaled[j] -= roundMe(scaled[j]);
1015 >        scaled[j] += 0.5;
1016 >      }
1017 >
1018 >      // find xyz-indices of cell that cutoffGroup is in.
1019 >      whichCell.x() = nCells_.x() * scaled.x();
1020 >      whichCell.y() = nCells_.y() * scaled.y();
1021 >      whichCell.z() = nCells_.z() * scaled.z();
1022 >
1023 >      // find single index of this cell:
1024 >      cellIndex = Vlinear(whichCell, nCells_);      
1025 >
1026 >      // add this cutoff group to the list of groups in this cell;
1027 >      cellList_[cellIndex].push_back(i);
1028 >    }
1029 > #endif
1030 >
1031 >    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1032 >      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1033 >        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1034 >          Vector3i m1v(m1x, m1y, m1z);
1035 >          int m1 = Vlinear(m1v, nCells_);
1036 >
1037 >          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1038 >               os != cellOffsets_.end(); ++os) {
1039 >            
1040 >            Vector3i m2v = m1v + (*os);
1041 >            
1042 >            if (m2v.x() >= nCells_.x()) {
1043 >              m2v.x() = 0;          
1044 >            } else if (m2v.x() < 0) {
1045 >              m2v.x() = nCells_.x() - 1;
1046 >            }
1047 >            
1048 >            if (m2v.y() >= nCells_.y()) {
1049 >              m2v.y() = 0;          
1050 >            } else if (m2v.y() < 0) {
1051 >              m2v.y() = nCells_.y() - 1;
1052 >            }
1053 >            
1054 >            if (m2v.z() >= nCells_.z()) {
1055 >              m2v.z() = 0;          
1056 >            } else if (m2v.z() < 0) {
1057 >              m2v.z() = nCells_.z() - 1;
1058 >            }
1059 >            
1060 >            int m2 = Vlinear (m2v, nCells_);
1061 >
1062 > #ifdef IS_MPI
1063 >            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1064 >                 j1 != cellListRow_[m1].end(); ++j1) {
1065 >              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1066 >                   j2 != cellListCol_[m2].end(); ++j2) {
1067 >                              
1068 >                // Always do this if we're in different cells or if
1069 >                // we're in the same cell and the global index of the
1070 >                // j2 cutoff group is less than the j1 cutoff group
1071 >
1072 >                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1073 >                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1074 >                  snap_->wrapVector(dr);
1075 >                  cuts = getGroupCutoffs( (*j1), (*j2) );
1076 >                  if (dr.lengthSquare() < cuts.third) {
1077 >                    neighborList.push_back(make_pair((*j1), (*j2)));
1078 >                  }
1079 >                }
1080 >              }
1081 >            }
1082 > #else
1083 >
1084 >            for (vector<int>::iterator j1 = cellList_[m1].begin();
1085 >                 j1 != cellList_[m1].end(); ++j1) {
1086 >              for (vector<int>::iterator j2 = cellList_[m2].begin();
1087 >                   j2 != cellList_[m2].end(); ++j2) {
1088 >
1089 >                // Always do this if we're in different cells or if
1090 >                // we're in the same cell and the global index of the
1091 >                // j2 cutoff group is less than the j1 cutoff group
1092 >
1093 >                if (m2 != m1 || (*j2) < (*j1)) {
1094 >                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1095 >                  snap_->wrapVector(dr);
1096 >                  cuts = getGroupCutoffs( (*j1), (*j2) );
1097 >                  if (dr.lengthSquare() < cuts.third) {
1098 >                    neighborList.push_back(make_pair((*j1), (*j2)));
1099 >                  }
1100 >                }
1101 >              }
1102 >            }
1103 > #endif
1104 >          }
1105 >        }
1106 >      }
1107 >    }
1108 >    
1109 >    // save the local cutoff group positions for the check that is
1110 >    // done on each loop:
1111 >    saved_CG_positions_.clear();
1112 >    for (int i = 0; i < nGroups_; i++)
1113 >      saved_CG_positions_.push_back(snap_->cgData.position[i]);
1114 >    
1115 >    return neighborList;
1116 >  }
1117   } //end namespace OpenMD

Diff Legend

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