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 1562 by gezelter, Thu May 12 17:00:14 2011 UTC vs.
Revision 1586 by gezelter, Tue Jun 21 06:34:35 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 >    // gather the information for atomtype IDs (atids):
63 >    idents = info_->getIdentArray();
64 >    AtomLocalToGlobal = info_->getGlobalAtomIndices();
65 >    cgLocalToGlobal = info_->getGlobalGroupIndices();
66 >    vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
67  
68 <    cgCommIntRow = new Communicator<Row,int>(nGroups);
72 <    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups);
73 <    cgCommIntColumn = new Communicator<Column,int>(nGroups);
74 <    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups);
68 >    massFactors = info_->getMassFactors();
69  
70 <    int nAtomsInRow = AtomCommIntRow->getSize();
71 <    int nAtomsInCol = AtomCommIntColumn->getSize();
72 <    int nGroupsInRow = cgCommIntRow->getSize();
73 <    int nGroupsInCol = cgCommIntColumn->getSize();
70 >    PairList excludes = info_->getExcludedInteractions();
71 >    PairList oneTwo = info_->getOneTwoInteractions();
72 >    PairList oneThree = info_->getOneThreeInteractions();
73 >    PairList oneFour = info_->getOneFourInteractions();
74  
75 + #ifdef IS_MPI
76 +
77 +    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
78 +    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
79 +    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
80 +    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
81 +    AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
82 +
83 +    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
84 +    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
85 +    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
86 +    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
87 +    AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
88 +
89 +    cgCommIntRow = new Communicator<Row,int>(nGroups_);
90 +    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
91 +    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
92 +    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
93 +
94 +    nAtomsInRow_ = AtomCommIntRow->getSize();
95 +    nAtomsInCol_ = AtomCommIntColumn->getSize();
96 +    nGroupsInRow_ = cgCommIntRow->getSize();
97 +    nGroupsInCol_ = cgCommIntColumn->getSize();
98 +
99      // Modify the data storage objects with the correct layouts and sizes:
100 <    atomRowData.resize(nAtomsInRow);
100 >    atomRowData.resize(nAtomsInRow_);
101      atomRowData.setStorageLayout(storageLayout_);
102 <    atomColData.resize(nAtomsInCol);
102 >    atomColData.resize(nAtomsInCol_);
103      atomColData.setStorageLayout(storageLayout_);
104 <    cgRowData.resize(nGroupsInRow);
104 >    cgRowData.resize(nGroupsInRow_);
105      cgRowData.setStorageLayout(DataStorage::dslPosition);
106 <    cgColData.resize(nGroupsInCol);
106 >    cgColData.resize(nGroupsInCol_);
107      cgColData.setStorageLayout(DataStorage::dslPosition);
108 +        
109 +    identsRow.resize(nAtomsInRow_);
110 +    identsCol.resize(nAtomsInCol_);
111      
112 <    vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
113 <                                      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);
112 >    AtomCommIntRow->gather(idents, identsRow);
113 >    AtomCommIntColumn->gather(idents, identsCol);
114      
99 –    // gather the information for atomtype IDs (atids):
100 –    vector<int> identsLocal = info_->getIdentArray();
101 –    identsRow.reserve(nAtomsInRow);
102 –    identsCol.reserve(nAtomsInCol);
103 –    
104 –    AtomCommIntRow->gather(identsLocal, identsRow);
105 –    AtomCommIntColumn->gather(identsLocal, identsCol);
106 –    
107 –    AtomLocalToGlobal = info_->getGlobalAtomIndices();
115      AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
116      AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
117      
111 –    cgLocalToGlobal = info_->getGlobalGroupIndices();
118      cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
119      cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
120  
121 <    // still need:
122 <    // topoDist
123 <    // exclude
121 >    AtomCommRealRow->gather(massFactors, massFactorsRow);
122 >    AtomCommRealColumn->gather(massFactors, massFactorsCol);
123 >
124 >    groupListRow_.clear();
125 >    groupListRow_.resize(nGroupsInRow_);
126 >    for (int i = 0; i < nGroupsInRow_; i++) {
127 >      int gid = cgRowToGlobal[i];
128 >      for (int j = 0; j < nAtomsInRow_; j++) {
129 >        int aid = AtomRowToGlobal[j];
130 >        if (globalGroupMembership[aid] == gid)
131 >          groupListRow_[i].push_back(j);
132 >      }      
133 >    }
134 >
135 >    groupListCol_.clear();
136 >    groupListCol_.resize(nGroupsInCol_);
137 >    for (int i = 0; i < nGroupsInCol_; i++) {
138 >      int gid = cgColToGlobal[i];
139 >      for (int j = 0; j < nAtomsInCol_; j++) {
140 >        int aid = AtomColToGlobal[j];
141 >        if (globalGroupMembership[aid] == gid)
142 >          groupListCol_[i].push_back(j);
143 >      }      
144 >    }
145 >
146 >    skipsForAtom.clear();
147 >    skipsForAtom.resize(nAtomsInRow_);
148 >    toposForAtom.clear();
149 >    toposForAtom.resize(nAtomsInRow_);
150 >    topoDist.clear();
151 >    topoDist.resize(nAtomsInRow_);
152 >    for (int i = 0; i < nAtomsInRow_; i++) {
153 >      int iglob = AtomRowToGlobal[i];
154 >
155 >      for (int j = 0; j < nAtomsInCol_; j++) {
156 >        int jglob = AtomColToGlobal[j];
157 >
158 >        if (excludes.hasPair(iglob, jglob))
159 >          skipsForAtom[i].push_back(j);      
160 >        
161 >        if (oneTwo.hasPair(iglob, jglob)) {
162 >          toposForAtom[i].push_back(j);
163 >          topoDist[i].push_back(1);
164 >        } else {
165 >          if (oneThree.hasPair(iglob, jglob)) {
166 >            toposForAtom[i].push_back(j);
167 >            topoDist[i].push_back(2);
168 >          } else {
169 >            if (oneFour.hasPair(iglob, jglob)) {
170 >              toposForAtom[i].push_back(j);
171 >              topoDist[i].push_back(3);
172 >            }
173 >          }
174 >        }
175 >      }      
176 >    }
177 >
178   #endif
179 +
180 +    groupList_.clear();
181 +    groupList_.resize(nGroups_);
182 +    for (int i = 0; i < nGroups_; i++) {
183 +      int gid = cgLocalToGlobal[i];
184 +      for (int j = 0; j < nLocal_; j++) {
185 +        int aid = AtomLocalToGlobal[j];
186 +        if (globalGroupMembership[aid] == gid) {
187 +          groupList_[i].push_back(j);
188 +        }
189 +      }      
190 +    }
191 +
192 +    skipsForAtom.clear();
193 +    skipsForAtom.resize(nLocal_);
194 +    toposForAtom.clear();
195 +    toposForAtom.resize(nLocal_);
196 +    topoDist.clear();
197 +    topoDist.resize(nLocal_);
198 +
199 +    for (int i = 0; i < nLocal_; i++) {
200 +      int iglob = AtomLocalToGlobal[i];
201 +
202 +      for (int j = 0; j < nLocal_; j++) {
203 +        int jglob = AtomLocalToGlobal[j];
204 +
205 +        if (excludes.hasPair(iglob, jglob))
206 +          skipsForAtom[i].push_back(j);              
207 +        
208 +        if (oneTwo.hasPair(iglob, jglob)) {
209 +          toposForAtom[i].push_back(j);
210 +          topoDist[i].push_back(1);
211 +        } else {
212 +          if (oneThree.hasPair(iglob, jglob)) {
213 +            toposForAtom[i].push_back(j);
214 +            topoDist[i].push_back(2);
215 +          } else {
216 +            if (oneFour.hasPair(iglob, jglob)) {
217 +              toposForAtom[i].push_back(j);
218 +              topoDist[i].push_back(3);
219 +            }
220 +          }
221 +        }
222 +      }      
223 +    }
224 +    
225 +    createGtypeCutoffMap();
226    }
227 +  
228 +  void ForceMatrixDecomposition::createGtypeCutoffMap() {
229      
230 +    RealType tol = 1e-6;
231 +    RealType rc;
232 +    int atid;
233 +    set<AtomType*> atypes = info_->getSimulatedAtomTypes();
234 +    vector<RealType> atypeCutoff;
235 +    atypeCutoff.resize( atypes.size() );
236 +      
237 +    for (set<AtomType*>::iterator at = atypes.begin();
238 +         at != atypes.end(); ++at){
239 +      atid = (*at)->getIdent();
240  
241 +      if (userChoseCutoff_)
242 +        atypeCutoff[atid] = userCutoff_;
243 +      else
244 +        atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
245 +    }
246  
247 +    vector<RealType> gTypeCutoffs;
248 +
249 +    // first we do a single loop over the cutoff groups to find the
250 +    // largest cutoff for any atypes present in this group.
251 + #ifdef IS_MPI
252 +    vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
253 +    groupRowToGtype.resize(nGroupsInRow_);
254 +    for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
255 +      vector<int> atomListRow = getAtomsInGroupRow(cg1);
256 +      for (vector<int>::iterator ia = atomListRow.begin();
257 +           ia != atomListRow.end(); ++ia) {            
258 +        int atom1 = (*ia);
259 +        atid = identsRow[atom1];
260 +        if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
261 +          groupCutoffRow[cg1] = atypeCutoff[atid];
262 +        }
263 +      }
264 +
265 +      bool gTypeFound = false;
266 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
267 +        if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
268 +          groupRowToGtype[cg1] = gt;
269 +          gTypeFound = true;
270 +        }
271 +      }
272 +      if (!gTypeFound) {
273 +        gTypeCutoffs.push_back( groupCutoffRow[cg1] );
274 +        groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
275 +      }
276 +      
277 +    }
278 +    vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
279 +    groupColToGtype.resize(nGroupsInCol_);
280 +    for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
281 +      vector<int> atomListCol = getAtomsInGroupColumn(cg2);
282 +      for (vector<int>::iterator jb = atomListCol.begin();
283 +           jb != atomListCol.end(); ++jb) {            
284 +        int atom2 = (*jb);
285 +        atid = identsCol[atom2];
286 +        if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
287 +          groupCutoffCol[cg2] = atypeCutoff[atid];
288 +        }
289 +      }
290 +      bool gTypeFound = false;
291 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
292 +        if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
293 +          groupColToGtype[cg2] = gt;
294 +          gTypeFound = true;
295 +        }
296 +      }
297 +      if (!gTypeFound) {
298 +        gTypeCutoffs.push_back( groupCutoffCol[cg2] );
299 +        groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
300 +      }
301 +    }
302 + #else
303 +
304 +    vector<RealType> groupCutoff(nGroups_, 0.0);
305 +    groupToGtype.resize(nGroups_);
306 +
307 +    for (int cg1 = 0; cg1 < nGroups_; cg1++) {
308 +
309 +      groupCutoff[cg1] = 0.0;
310 +      vector<int> atomList = getAtomsInGroupRow(cg1);
311 +
312 +      for (vector<int>::iterator ia = atomList.begin();
313 +           ia != atomList.end(); ++ia) {            
314 +        int atom1 = (*ia);
315 +        atid = idents[atom1];
316 +        if (atypeCutoff[atid] > groupCutoff[cg1]) {
317 +          groupCutoff[cg1] = atypeCutoff[atid];
318 +        }
319 +      }
320 +
321 +      bool gTypeFound = false;
322 +      for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
323 +        if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
324 +          groupToGtype[cg1] = gt;
325 +          gTypeFound = true;
326 +        }
327 +      }
328 +      if (!gTypeFound) {
329 +        gTypeCutoffs.push_back( groupCutoff[cg1] );
330 +        groupToGtype[cg1] = gTypeCutoffs.size() - 1;
331 +      }      
332 +    }
333 + #endif
334 +
335 +    // Now we find the maximum group cutoff value present in the simulation
336 +
337 +    RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
338 +
339 + #ifdef IS_MPI
340 +    MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
341 + #endif
342 +    
343 +    RealType tradRcut = groupMax;
344 +
345 +    for (int i = 0; i < gTypeCutoffs.size();  i++) {
346 +      for (int j = 0; j < gTypeCutoffs.size();  j++) {      
347 +        RealType thisRcut;
348 +        switch(cutoffPolicy_) {
349 +        case TRADITIONAL:
350 +          thisRcut = tradRcut;
351 +          break;
352 +        case MIX:
353 +          thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
354 +          break;
355 +        case MAX:
356 +          thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
357 +          break;
358 +        default:
359 +          sprintf(painCave.errMsg,
360 +                  "ForceMatrixDecomposition::createGtypeCutoffMap "
361 +                  "hit an unknown cutoff policy!\n");
362 +          painCave.severity = OPENMD_ERROR;
363 +          painCave.isFatal = 1;
364 +          simError();
365 +          break;
366 +        }
367 +
368 +        pair<int,int> key = make_pair(i,j);
369 +        gTypeCutoffMap[key].first = thisRcut;
370 +
371 +        if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
372 +
373 +        gTypeCutoffMap[key].second = thisRcut*thisRcut;
374 +        
375 +        gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
376 +
377 +        // sanity check
378 +        
379 +        if (userChoseCutoff_) {
380 +          if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
381 +            sprintf(painCave.errMsg,
382 +                    "ForceMatrixDecomposition::createGtypeCutoffMap "
383 +                    "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
384 +            painCave.severity = OPENMD_ERROR;
385 +            painCave.isFatal = 1;
386 +            simError();            
387 +          }
388 +        }
389 +      }
390 +    }
391 +  }
392 +
393 +
394 +  groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
395 +    int i, j;  
396 + #ifdef IS_MPI
397 +    i = groupRowToGtype[cg1];
398 +    j = groupColToGtype[cg2];
399 + #else
400 +    i = groupToGtype[cg1];
401 +    j = groupToGtype[cg2];
402 + #endif    
403 +    return gTypeCutoffMap[make_pair(i,j)];
404 +  }
405 +
406 +  int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
407 +    for (int j = 0; j < toposForAtom[atom1].size(); j++) {
408 +      if (toposForAtom[atom1][j] == atom2)
409 +        return topoDist[atom1][j];
410 +    }
411 +    return 0;
412 +  }
413 +
414 +  void ForceMatrixDecomposition::zeroWorkArrays() {
415 +    pairwisePot = 0.0;
416 +    embeddingPot = 0.0;
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 +    if (storageLayout_ & DataStorage::dslParticlePot) {    
436 +      fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
437 +      fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
438 +    }
439 +
440 +    if (storageLayout_ & DataStorage::dslDensity) {      
441 +      fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
442 +      fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
443 +    }
444 +
445 +    if (storageLayout_ & DataStorage::dslFunctional) {  
446 +      fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
447 +      fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
448 +    }
449 +
450 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
451 +      fill(atomRowData.functionalDerivative.begin(),
452 +           atomRowData.functionalDerivative.end(), 0.0);
453 +      fill(atomColData.functionalDerivative.begin(),
454 +           atomColData.functionalDerivative.end(), 0.0);
455 +    }
456 +
457 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
458 +      fill(atomRowData.skippedCharge.begin(), atomRowData.skippedCharge.end(), 0.0);
459 +      fill(atomColData.skippedCharge.begin(), atomColData.skippedCharge.end(), 0.0);
460 +    }
461 +
462 + #else
463 +    
464 +    if (storageLayout_ & DataStorage::dslParticlePot) {      
465 +      fill(snap_->atomData.particlePot.begin(),
466 +           snap_->atomData.particlePot.end(), 0.0);
467 +    }
468 +    
469 +    if (storageLayout_ & DataStorage::dslDensity) {      
470 +      fill(snap_->atomData.density.begin(),
471 +           snap_->atomData.density.end(), 0.0);
472 +    }
473 +    if (storageLayout_ & DataStorage::dslFunctional) {
474 +      fill(snap_->atomData.functional.begin(),
475 +           snap_->atomData.functional.end(), 0.0);
476 +    }
477 +    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {      
478 +      fill(snap_->atomData.functionalDerivative.begin(),
479 +           snap_->atomData.functionalDerivative.end(), 0.0);
480 +    }
481 +    if (storageLayout_ & DataStorage::dslSkippedCharge) {      
482 +      fill(snap_->atomData.skippedCharge.begin(),
483 +           snap_->atomData.skippedCharge.end(), 0.0);
484 +    }
485 + #endif
486 +    
487 +  }
488 +
489 +
490    void ForceMatrixDecomposition::distributeData()  {
491      snap_ = sman_->getCurrentSnapshot();
492      storageLayout_ = sman_->getStorageLayout();
# Line 155 | Line 522 | namespace OpenMD {
522   #endif      
523    }
524    
525 +  /* collects information obtained during the pre-pair loop onto local
526 +   * data structures.
527 +   */
528    void ForceMatrixDecomposition::collectIntermediateData() {
529      snap_ = sman_->getCurrentSnapshot();
530      storageLayout_ = sman_->getStorageLayout();
# Line 166 | Line 536 | namespace OpenMD {
536                                 snap_->atomData.density);
537        
538        int n = snap_->atomData.density.size();
539 <      std::vector<RealType> rho_tmp(n, 0.0);
539 >      vector<RealType> rho_tmp(n, 0.0);
540        AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
541        for (int i = 0; i < n; i++)
542          snap_->atomData.density[i] += rho_tmp[i];
543      }
544   #endif
545    }
546 <  
546 >
547 >  /*
548 >   * redistributes information obtained during the pre-pair loop out to
549 >   * row and column-indexed data structures
550 >   */
551    void ForceMatrixDecomposition::distributeIntermediateData() {
552      snap_ = sman_->getCurrentSnapshot();
553      storageLayout_ = sman_->getStorageLayout();
# Line 229 | Line 603 | namespace OpenMD {
603          snap_->atomData.torque[i] += trq_tmp[i];
604      }
605      
606 <    int nLocal = snap_->getNumberOfAtoms();
606 >    nLocal_ = snap_->getNumberOfAtoms();
607  
608 <    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
609 <                                       vector<RealType> (nLocal, 0.0));
608 >    vector<potVec> pot_temp(nLocal_,
609 >                            Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
610 >
611 >    // scatter/gather pot_row into the members of my column
612 >          
613 >    AtomCommPotRow->scatter(pot_row, pot_temp);
614 >
615 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
616 >      pairwisePot += pot_temp[ii];
617      
618 <    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
619 <      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
620 <      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
621 <        pot_local[i] += pot_temp[i][ii];
622 <      }
623 <    }
618 >    fill(pot_temp.begin(), pot_temp.end(),
619 >         Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
620 >      
621 >    AtomCommPotColumn->scatter(pot_col, pot_temp);    
622 >    
623 >    for (int ii = 0;  ii < pot_temp.size(); ii++ )
624 >      pairwisePot += pot_temp[ii];    
625   #endif
626 +
627    }
628  
629 +  int ForceMatrixDecomposition::getNAtomsInRow() {  
630 + #ifdef IS_MPI
631 +    return nAtomsInRow_;
632 + #else
633 +    return nLocal_;
634 + #endif
635 +  }
636 +
637 +  /**
638 +   * returns the list of atoms belonging to this group.  
639 +   */
640 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
641 + #ifdef IS_MPI
642 +    return groupListRow_[cg1];
643 + #else
644 +    return groupList_[cg1];
645 + #endif
646 +  }
647 +
648 +  vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
649 + #ifdef IS_MPI
650 +    return groupListCol_[cg2];
651 + #else
652 +    return groupList_[cg2];
653 + #endif
654 +  }
655    
656    Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
657      Vector3d d;
# Line 284 | Line 693 | namespace OpenMD {
693      snap_->wrapVector(d);
694      return d;    
695    }
696 +
697 +  RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
698 + #ifdef IS_MPI
699 +    return massFactorsRow[atom1];
700 + #else
701 +    return massFactors[atom1];
702 + #endif
703 +  }
704 +
705 +  RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
706 + #ifdef IS_MPI
707 +    return massFactorsCol[atom2];
708 + #else
709 +    return massFactors[atom2];
710 + #endif
711 +
712 +  }
713      
714    Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
715      Vector3d d;
# Line 298 | Line 724 | namespace OpenMD {
724      return d;    
725    }
726  
727 +  vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
728 +    return skipsForAtom[atom1];
729 +  }
730 +
731 +  /**
732 +   * There are a number of reasons to skip a pair or a
733 +   * particle. Mostly we do this to exclude atoms who are involved in
734 +   * short range interactions (bonds, bends, torsions), but we also
735 +   * need to exclude some overcounted interactions that result from
736 +   * the parallel decomposition.
737 +   */
738 +  bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
739 +    int unique_id_1, unique_id_2;
740 +
741 + #ifdef IS_MPI
742 +    // in MPI, we have to look up the unique IDs for each atom
743 +    unique_id_1 = AtomRowToGlobal[atom1];
744 +    unique_id_2 = AtomColToGlobal[atom2];
745 +
746 +    // this situation should only arise in MPI simulations
747 +    if (unique_id_1 == unique_id_2) return true;
748 +    
749 +    // this prevents us from doing the pair on multiple processors
750 +    if (unique_id_1 < unique_id_2) {
751 +      if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
752 +    } else {
753 +      if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
754 +    }
755 + #else
756 +    // in the normal loop, the atom numbers are unique
757 +    unique_id_1 = atom1;
758 +    unique_id_2 = atom2;
759 + #endif
760 +    
761 +    for (vector<int>::iterator i = skipsForAtom[atom1].begin();
762 +         i != skipsForAtom[atom1].end(); ++i) {
763 +      if ( (*i) == unique_id_2 ) return true;
764 +    }
765 +
766 +    return false;
767 +  }
768 +
769 +
770    void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
771   #ifdef IS_MPI
772      atomRowData.force[atom1] += fg;
# Line 312 | Line 781 | namespace OpenMD {
781   #else
782      snap_->atomData.force[atom2] += fg;
783   #endif
315 –
784    }
785  
786      // filling interaction blocks with pointers
787 <  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
788 <
321 <    InteractionData idat;
787 >  void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
788 >                                                     int atom1, int atom2) {    
789   #ifdef IS_MPI
790 +    
791 +    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
792 +                             ff_->getAtomType(identsCol[atom2]) );
793 +    
794      if (storageLayout_ & DataStorage::dslAmat) {
795        idat.A1 = &(atomRowData.aMat[atom1]);
796        idat.A2 = &(atomColData.aMat[atom2]);
797      }
798 <
798 >    
799      if (storageLayout_ & DataStorage::dslElectroFrame) {
800        idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
801        idat.eFrame2 = &(atomColData.electroFrame[atom2]);
# Line 338 | Line 809 | namespace OpenMD {
809      if (storageLayout_ & DataStorage::dslDensity) {
810        idat.rho1 = &(atomRowData.density[atom1]);
811        idat.rho2 = &(atomColData.density[atom2]);
812 +    }
813 +
814 +    if (storageLayout_ & DataStorage::dslFunctional) {
815 +      idat.frho1 = &(atomRowData.functional[atom1]);
816 +      idat.frho2 = &(atomColData.functional[atom2]);
817      }
818  
819      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
820        idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
821        idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
822      }
823 +
824 +    if (storageLayout_ & DataStorage::dslParticlePot) {
825 +      idat.particlePot1 = &(atomRowData.particlePot[atom1]);
826 +      idat.particlePot2 = &(atomColData.particlePot[atom2]);
827 +    }
828 +
829   #else
830 +
831 +    idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
832 +                             ff_->getAtomType(idents[atom2]) );
833 +
834      if (storageLayout_ & DataStorage::dslAmat) {
835        idat.A1 = &(snap_->atomData.aMat[atom1]);
836        idat.A2 = &(snap_->atomData.aMat[atom2]);
# Line 360 | Line 846 | namespace OpenMD {
846        idat.t2 = &(snap_->atomData.torque[atom2]);
847      }
848  
849 <    if (storageLayout_ & DataStorage::dslDensity) {
849 >    if (storageLayout_ & DataStorage::dslDensity) {    
850        idat.rho1 = &(snap_->atomData.density[atom1]);
851        idat.rho2 = &(snap_->atomData.density[atom2]);
852      }
853  
854 +    if (storageLayout_ & DataStorage::dslFunctional) {
855 +      idat.frho1 = &(snap_->atomData.functional[atom1]);
856 +      idat.frho2 = &(snap_->atomData.functional[atom2]);
857 +    }
858 +
859      if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
860        idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
861        idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
862      }
863 +
864 +    if (storageLayout_ & DataStorage::dslParticlePot) {
865 +      idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
866 +      idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
867 +    }
868 +
869   #endif
870 +  }
871 +
872 +  
873 +  void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {    
874 + #ifdef IS_MPI
875 +    pot_row[atom1] += 0.5 *  *(idat.pot);
876 +    pot_col[atom2] += 0.5 *  *(idat.pot);
877 +
878 +    atomRowData.force[atom1] += *(idat.f1);
879 +    atomColData.force[atom2] -= *(idat.f1);
880 + #else
881 +    pairwisePot += *(idat.pot);
882 +
883 +    snap_->atomData.force[atom1] += *(idat.f1);
884 +    snap_->atomData.force[atom2] -= *(idat.f1);
885 + #endif
886      
887    }
888 <  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
889 <    InteractionData idat;
890 <    skippedCharge1
891 <      skippedCharge2
379 <      rij
380 <      d
381 <    electroMult
382 <    sw
383 <    f
888 >
889 >
890 >  void ForceMatrixDecomposition::fillSkipData(InteractionData &idat,
891 >                                              int atom1, int atom2) {
892   #ifdef IS_MPI
893 +    idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
894 +                             ff_->getAtomType(identsCol[atom2]) );
895  
896      if (storageLayout_ & DataStorage::dslElectroFrame) {
897        idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
898        idat.eFrame2 = &(atomColData.electroFrame[atom2]);
899      }
900 +
901      if (storageLayout_ & DataStorage::dslTorque) {
902        idat.t1 = &(atomRowData.torque[atom1]);
903        idat.t2 = &(atomColData.torque[atom2]);
904      }
905  
906 <    
906 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {
907 >      idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
908 >      idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
909 >    }
910 > #else
911 >    idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
912 >                             ff_->getAtomType(idents[atom2]) );
913 >
914 >    if (storageLayout_ & DataStorage::dslElectroFrame) {
915 >      idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
916 >      idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
917 >    }
918 >
919 >    if (storageLayout_ & DataStorage::dslTorque) {
920 >      idat.t1 = &(snap_->atomData.torque[atom1]);
921 >      idat.t2 = &(snap_->atomData.torque[atom2]);
922 >    }
923 >
924 >    if (storageLayout_ & DataStorage::dslSkippedCharge) {
925 >      idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
926 >      idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
927 >    }
928 > #endif    
929    }
930 <  SelfData ForceMatrixDecomposition::fillSelfData(int atom1) {
930 >
931 >
932 >  void ForceMatrixDecomposition::unpackSkipData(InteractionData &idat, int atom1, int atom2) {    
933 > #ifdef IS_MPI
934 >    pot_row[atom1] += 0.5 *  *(idat.pot);
935 >    pot_col[atom2] += 0.5 *  *(idat.pot);
936 > #else
937 >    pairwisePot += *(idat.pot);  
938 > #endif
939 >
940    }
941  
942  
# Line 404 | Line 946 | namespace OpenMD {
946     * first element of pair is row-indexed CutoffGroup
947     * second element of pair is column-indexed CutoffGroup
948     */
949 <  vector<pair<int, int> >  buildNeighborList() {
950 <    Vector3d dr, invWid, rs, shift;
951 <    Vector3i cc, m1v, m2s;
952 <    RealType rrNebr;
953 <    int c, j1, j2, m1, m1x, m1y, m1z, m2, n, offset;
954 <
955 <
956 <    vector<pair<int, int> > neighborList;  
957 <    Vector3i nCells;
958 <    Vector3d invWid, r;
417 <
418 <    rList_ = (rCut_ + skinThickness_);
419 <    rl2 = rList_ * rList_;
949 >  vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
950 >      
951 >    vector<pair<int, int> > neighborList;
952 >    groupCutoffs cuts;
953 > #ifdef IS_MPI
954 >    cellListRow_.clear();
955 >    cellListCol_.clear();
956 > #else
957 >    cellList_.clear();
958 > #endif
959  
960 <    snap_ = sman_->getCurrentSnapshot();
960 >    RealType rList_ = (largestRcut_ + skinThickness_);
961 >    RealType rl2 = rList_ * rList_;
962 >    Snapshot* snap_ = sman_->getCurrentSnapshot();
963      Mat3x3d Hmat = snap_->getHmat();
964      Vector3d Hx = Hmat.getColumn(0);
965      Vector3d Hy = Hmat.getColumn(1);
966      Vector3d Hz = Hmat.getColumn(2);
967  
968 <    nCells.x() = (int) ( Hx.length() )/ rList_;
969 <    nCells.y() = (int) ( Hy.length() )/ rList_;
970 <    nCells.z() = (int) ( Hz.length() )/ rList_;
968 >    nCells_.x() = (int) ( Hx.length() )/ rList_;
969 >    nCells_.y() = (int) ( Hy.length() )/ rList_;
970 >    nCells_.z() = (int) ( Hz.length() )/ rList_;
971  
972 <    for (i = 0; i < nGroupsInRow; i++) {
972 >    Mat3x3d invHmat = snap_->getInvHmat();
973 >    Vector3d rs, scaled, dr;
974 >    Vector3i whichCell;
975 >    int cellIndex;
976 >    int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
977 >
978 > #ifdef IS_MPI
979 >    cellListRow_.resize(nCtot);
980 >    cellListCol_.resize(nCtot);
981 > #else
982 >    cellList_.resize(nCtot);
983 > #endif
984 >
985 > #ifdef IS_MPI
986 >    for (int i = 0; i < nGroupsInRow_; i++) {
987        rs = cgRowData.position[i];
988 <      snap_->scaleVector(rs);    
988 >
989 >      // scaled positions relative to the box vectors
990 >      scaled = invHmat * rs;
991 >
992 >      // wrap the vector back into the unit box by subtracting integer box
993 >      // numbers
994 >      for (int j = 0; j < 3; j++) {
995 >        scaled[j] -= roundMe(scaled[j]);
996 >        scaled[j] += 0.5;
997 >      }
998 >    
999 >      // find xyz-indices of cell that cutoffGroup is in.
1000 >      whichCell.x() = nCells_.x() * scaled.x();
1001 >      whichCell.y() = nCells_.y() * scaled.y();
1002 >      whichCell.z() = nCells_.z() * scaled.z();
1003 >
1004 >      // find single index of this cell:
1005 >      cellIndex = Vlinear(whichCell, nCells_);
1006 >
1007 >      // add this cutoff group to the list of groups in this cell;
1008 >      cellListRow_[cellIndex].push_back(i);
1009      }
435 –    
1010  
1011 <    VDiv (invWid, cells, region);
1012 <    for (n = nMol; n < nMol + cells.componentProduct(); n ++) cellList[n] = -1;
1013 <    for (n = 0; n < nMol; n ++) {
1014 <      VSAdd (rs, mol[n].r, 0.5, region);
1015 <      VMul (cc, rs, invWid);
1016 <      c = VLinear (cc, cells) + nMol;
1017 <      cellList[n] = cellList[c];
1018 <      cellList[c] = n;
1011 >    for (int i = 0; i < nGroupsInCol_; i++) {
1012 >      rs = cgColData.position[i];
1013 >
1014 >      // scaled positions relative to the box vectors
1015 >      scaled = invHmat * rs;
1016 >
1017 >      // wrap the vector back into the unit box by subtracting integer box
1018 >      // numbers
1019 >      for (int j = 0; j < 3; j++) {
1020 >        scaled[j] -= roundMe(scaled[j]);
1021 >        scaled[j] += 0.5;
1022 >      }
1023 >
1024 >      // find xyz-indices of cell that cutoffGroup is in.
1025 >      whichCell.x() = nCells_.x() * scaled.x();
1026 >      whichCell.y() = nCells_.y() * scaled.y();
1027 >      whichCell.z() = nCells_.z() * scaled.z();
1028 >
1029 >      // find single index of this cell:
1030 >      cellIndex = Vlinear(whichCell, nCells_);
1031 >
1032 >      // add this cutoff group to the list of groups in this cell;
1033 >      cellListCol_[cellIndex].push_back(i);
1034      }
1035 <    nebrTabLen = 0;
1036 <    for (m1z = 0; m1z < cells.z(); m1z++) {
1037 <      for (m1y = 0; m1y < cells.y(); m1y++) {
1038 <        for (m1x = 0; m1x < cells.x(); m1x++) {
1035 > #else
1036 >    for (int i = 0; i < nGroups_; i++) {
1037 >      rs = snap_->cgData.position[i];
1038 >
1039 >      // scaled positions relative to the box vectors
1040 >      scaled = invHmat * rs;
1041 >
1042 >      // wrap the vector back into the unit box by subtracting integer box
1043 >      // numbers
1044 >      for (int j = 0; j < 3; j++) {
1045 >        scaled[j] -= roundMe(scaled[j]);
1046 >        scaled[j] += 0.5;
1047 >      }
1048 >
1049 >      // find xyz-indices of cell that cutoffGroup is in.
1050 >      whichCell.x() = nCells_.x() * scaled.x();
1051 >      whichCell.y() = nCells_.y() * scaled.y();
1052 >      whichCell.z() = nCells_.z() * scaled.z();
1053 >
1054 >      // find single index of this cell:
1055 >      cellIndex = Vlinear(whichCell, nCells_);      
1056 >
1057 >      // add this cutoff group to the list of groups in this cell;
1058 >      cellList_[cellIndex].push_back(i);
1059 >    }
1060 > #endif
1061 >
1062 >    for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1063 >      for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1064 >        for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1065            Vector3i m1v(m1x, m1y, m1z);
1066 <          m1 = VLinear(m1v, cells) + nMol;
452 <          for (offset = 0; offset < nOffset_; offset++) {
453 <            m2v = m1v + cellOffsets_[offset];
454 <            shift = V3Zero();
1066 >          int m1 = Vlinear(m1v, nCells_);
1067  
1068 <            if (m2v.x() >= cells.x) {
1068 >          for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1069 >               os != cellOffsets_.end(); ++os) {
1070 >            
1071 >            Vector3i m2v = m1v + (*os);
1072 >            
1073 >            if (m2v.x() >= nCells_.x()) {
1074                m2v.x() = 0;          
458 –              shift.x() = region.x();  
1075              } else if (m2v.x() < 0) {
1076 <              m2v.x() = cells.x() - 1;
461 <              shift.x() = - region.x();
1076 >              m2v.x() = nCells_.x() - 1;
1077              }
1078 <
1079 <            if (m2v.y() >= cells.y()) {
1078 >            
1079 >            if (m2v.y() >= nCells_.y()) {
1080                m2v.y() = 0;          
466 –              shift.y() = region.y();  
1081              } else if (m2v.y() < 0) {
1082 <              m2v.y() = cells.y() - 1;
469 <              shift.y() = - region.y();
1082 >              m2v.y() = nCells_.y() - 1;
1083              }
1084 +            
1085 +            if (m2v.z() >= nCells_.z()) {
1086 +              m2v.z() = 0;          
1087 +            } else if (m2v.z() < 0) {
1088 +              m2v.z() = nCells_.z() - 1;
1089 +            }
1090 +            
1091 +            int m2 = Vlinear (m2v, nCells_);
1092  
1093 <            m2 = VLinear (m2v, cells) + nMol;
1094 <            for (j1 = cellList[m1]; j1 >= 0; j1 = cellList[j1]) {
1095 <              for (j2 = cellList[m2]; j2 >= 0; j2 = cellList[j2]) {
1096 <                if (m1 != m2 || j2 < j1) {
1097 <                  dr = mol[j1].r - mol[j2].r;
1098 <                  VSub (dr, mol[j1].r, mol[j2].r);
1099 <                  VVSub (dr, shift);
1100 <                  if (VLenSq (dr) < rrNebr) {
1101 <                    neighborList.push_back(make_pair(j1, j2));
1093 > #ifdef IS_MPI
1094 >            for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1095 >                 j1 != cellListRow_[m1].end(); ++j1) {
1096 >              for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1097 >                   j2 != cellListCol_[m2].end(); ++j2) {
1098 >                              
1099 >                // Always do this if we're in different cells or if
1100 >                // we're in the same cell and the global index of the
1101 >                // j2 cutoff group is less than the j1 cutoff group
1102 >
1103 >                if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1104 >                  dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1105 >                  snap_->wrapVector(dr);
1106 >                  cuts = getGroupCutoffs( (*j1), (*j2) );
1107 >                  if (dr.lengthSquare() < cuts.third) {
1108 >                    neighborList.push_back(make_pair((*j1), (*j2)));
1109                    }
1110                  }
1111                }
1112              }
1113 + #else
1114 +
1115 +            for (vector<int>::iterator j1 = cellList_[m1].begin();
1116 +                 j1 != cellList_[m1].end(); ++j1) {
1117 +              for (vector<int>::iterator j2 = cellList_[m2].begin();
1118 +                   j2 != cellList_[m2].end(); ++j2) {
1119 +
1120 +                // Always do this if we're in different cells or if
1121 +                // we're in the same cell and the global index of the
1122 +                // j2 cutoff group is less than the j1 cutoff group
1123 +
1124 +                if (m2 != m1 || (*j2) < (*j1)) {
1125 +                  dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1126 +                  snap_->wrapVector(dr);
1127 +                  cuts = getGroupCutoffs( (*j1), (*j2) );
1128 +                  if (dr.lengthSquare() < cuts.third) {
1129 +                    neighborList.push_back(make_pair((*j1), (*j2)));
1130 +                  }
1131 +                }
1132 +              }
1133 +            }
1134 + #endif
1135            }
1136          }
1137        }
1138      }
1139 +    
1140 +    // save the local cutoff group positions for the check that is
1141 +    // done on each loop:
1142 +    saved_CG_positions_.clear();
1143 +    for (int i = 0; i < nGroups_; i++)
1144 +      saved_CG_positions_.push_back(snap_->cgData.position[i]);
1145 +  
1146 +    return neighborList;
1147    }
490 –
491 –  
1148   } //end namespace OpenMD

Diff Legend

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