[ViewVC] Diff of: OpenMD/trunk/src/parallel/ForceMatrixDecomposition.cpp

Comparing branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents):
Revision 1582 by gezelter, Tue Jun 14 20:41:44 2011 UTC vs.
Revision 1616 by gezelter, Tue Aug 30 15:45:35 2011 UTC

#	Line 47 \| Line 47 \| namespace OpenMD {
47		using namespace std;
48		namespace OpenMD {
49
50	+	ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
51	+
52	+	// In a parallel computation, row and colum scans must visit all
53	+	// surrounding cells (not just the 14 upper triangular blocks that
54	+	// are used when the processor can see all pairs)
55	+	#ifdef IS_MPI
56	+	cellOffsets_.clear();
57	+	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
58	+	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
59	+	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
60	+	cellOffsets_.push_back( Vector3i(-1, 0,-1) );
61	+	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
62	+	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
63	+	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
64	+	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
65	+	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
66	+	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
67	+	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
68	+	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
69	+	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
70	+	cellOffsets_.push_back( Vector3i( 0, 0, 0) );
71	+	cellOffsets_.push_back( Vector3i( 1, 0, 0) );
72	+	cellOffsets_.push_back( Vector3i(-1, 1, 0) );
73	+	cellOffsets_.push_back( Vector3i( 0, 1, 0) );
74	+	cellOffsets_.push_back( Vector3i( 1, 1, 0) );
75	+	cellOffsets_.push_back( Vector3i(-1,-1, 1) );
76	+	cellOffsets_.push_back( Vector3i( 0,-1, 1) );
77	+	cellOffsets_.push_back( Vector3i( 1,-1, 1) );
78	+	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
79	+	cellOffsets_.push_back( Vector3i( 0, 0, 1) );
80	+	cellOffsets_.push_back( Vector3i( 1, 0, 1) );
81	+	cellOffsets_.push_back( Vector3i(-1, 1, 1) );
82	+	cellOffsets_.push_back( Vector3i( 0, 1, 1) );
83	+	cellOffsets_.push_back( Vector3i( 1, 1, 1) );
84	+	#endif
85	+	}
86	+
87	+
88		/**
89		* distributeInitialData is essentially a copy of the older fortran
90		* SimulationSetup
91		*/
54	–
92		void ForceMatrixDecomposition::distributeInitialData() {
93		snap_ = sman_->getCurrentSnapshot();
94		storageLayout_ = sman_->getStorageLayout();
95		ff_ = info_->getForceField();
96		nLocal_ = snap_->getNumberOfAtoms();
97	<
97	>
98		nGroups_ = info_->getNLocalCutoffGroups();
62	–	cerr << "in dId, nGroups = " << nGroups_ << "\n";
99		// gather the information for atomtype IDs (atids):
100	<	identsLocal = info_->getIdentArray();
100	>	idents = info_->getIdentArray();
101		AtomLocalToGlobal = info_->getGlobalAtomIndices();
102		cgLocalToGlobal = info_->getGlobalGroupIndices();
103		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
104	+
105		massFactors = info_->getMassFactors();
69	–	PairList excludes = info_->getExcludedInteractions();
70	–	PairList oneTwo = info_->getOneTwoInteractions();
71	–	PairList oneThree = info_->getOneThreeInteractions();
72	–	PairList oneFour = info_->getOneFourInteractions();
106
107	+	PairList* excludes = info_->getExcludedInteractions();
108	+	PairList* oneTwo = info_->getOneTwoInteractions();
109	+	PairList* oneThree = info_->getOneThreeInteractions();
110	+	PairList* oneFour = info_->getOneFourInteractions();
111	+
112		#ifdef IS_MPI
113
114	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
115	<	AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
78	<	AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
79	<	AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
80	<	AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
114	>	MPI::Intracomm row = rowComm.getComm();
115	>	MPI::Intracomm col = colComm.getComm();
116
117	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
118	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
119	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
120	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
121	<	AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
117	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
118	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
119	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
120	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
121	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
122
123	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
124	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
125	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
126	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
123	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
124	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
125	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
126	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
127	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
128
129	<	nAtomsInRow_ = AtomCommIntRow->getSize();
130	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
131	<	nGroupsInRow_ = cgCommIntRow->getSize();
132	<	nGroupsInCol_ = cgCommIntColumn->getSize();
129	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
130	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
131	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
132	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
133
134	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
135	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
136	+	nGroupsInRow_ = cgPlanIntRow->getSize();
137	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
138	+
139		// Modify the data storage objects with the correct layouts and sizes:
140		atomRowData.resize(nAtomsInRow_);
141		atomRowData.setStorageLayout(storageLayout_);
#	Line 108 \| Line 149 \| namespace OpenMD {
149		identsRow.resize(nAtomsInRow_);
150		identsCol.resize(nAtomsInCol_);
151
152	<	AtomCommIntRow->gather(identsLocal, identsRow);
153	<	AtomCommIntColumn->gather(identsLocal, identsCol);
152	>	AtomPlanIntRow->gather(idents, identsRow);
153	>	AtomPlanIntColumn->gather(idents, identsCol);
154
155	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
156	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
157	<
117	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
118	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
155	>	// allocate memory for the parallel objects
156	>	atypesRow.resize(nAtomsInRow_);
157	>	atypesCol.resize(nAtomsInCol_);
158
159	<	AtomCommRealRow->gather(massFactors, massFactorsRow);
160	<	AtomCommRealColumn->gather(massFactors, massFactorsCol);
159	>	for (int i = 0; i < nAtomsInRow_; i++)
160	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
161	>	for (int i = 0; i < nAtomsInCol_; i++)
162	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
163	>
164	>	pot_row.resize(nAtomsInRow_);
165	>	pot_col.resize(nAtomsInCol_);
166
167	+	AtomRowToGlobal.resize(nAtomsInRow_);
168	+	AtomColToGlobal.resize(nAtomsInCol_);
169	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
170	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
171	+
172	+	cgRowToGlobal.resize(nGroupsInRow_);
173	+	cgColToGlobal.resize(nGroupsInCol_);
174	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
175	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
176	+
177	+	massFactorsRow.resize(nAtomsInRow_);
178	+	massFactorsCol.resize(nAtomsInCol_);
179	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
180	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
181	+
182		groupListRow_.clear();
183		groupListRow_.resize(nGroupsInRow_);
184		for (int i = 0; i < nGroupsInRow_; i++) {
#	Line 142 \| Line 201 \| namespace OpenMD {
201		}
202		}
203
204	<	skipsForAtom.clear();
205	<	skipsForAtom.resize(nAtomsInRow_);
204	>	excludesForAtom.clear();
205	>	excludesForAtom.resize(nAtomsInRow_);
206		toposForAtom.clear();
207		toposForAtom.resize(nAtomsInRow_);
208		topoDist.clear();
#	Line 154 \| Line 213 \| namespace OpenMD {
213		for (int j = 0; j < nAtomsInCol_; j++) {
214		int jglob = AtomColToGlobal[j];
215
216	<	if (excludes.hasPair(iglob, jglob))
217	<	skipsForAtom[i].push_back(j);
216	>	if (excludes->hasPair(iglob, jglob))
217	>	excludesForAtom[i].push_back(j);
218
219	<	if (oneTwo.hasPair(iglob, jglob)) {
219	>	if (oneTwo->hasPair(iglob, jglob)) {
220		toposForAtom[i].push_back(j);
221		topoDist[i].push_back(1);
222		} else {
223	<	if (oneThree.hasPair(iglob, jglob)) {
223	>	if (oneThree->hasPair(iglob, jglob)) {
224		toposForAtom[i].push_back(j);
225		topoDist[i].push_back(2);
226		} else {
227	<	if (oneFour.hasPair(iglob, jglob)) {
227	>	if (oneFour->hasPair(iglob, jglob)) {
228		toposForAtom[i].push_back(j);
229		topoDist[i].push_back(3);
230		}
#	Line 174 \| Line 233 \| namespace OpenMD {
233		}
234		}
235
236	<	#endif
237	<
238	<	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_);
236	>	#else
237	>	excludesForAtom.clear();
238	>	excludesForAtom.resize(nLocal_);
239		toposForAtom.clear();
240		toposForAtom.resize(nLocal_);
241		topoDist.clear();
#	Line 201 \| Line 247 \| namespace OpenMD {
247		for (int j = 0; j < nLocal_; j++) {
248		int jglob = AtomLocalToGlobal[j];
249
250	<	if (excludes.hasPair(iglob, jglob))
251	<	skipsForAtom[i].push_back(j);
250	>	if (excludes->hasPair(iglob, jglob))
251	>	excludesForAtom[i].push_back(j);
252
253	<	if (oneTwo.hasPair(iglob, jglob)) {
253	>	if (oneTwo->hasPair(iglob, jglob)) {
254		toposForAtom[i].push_back(j);
255		topoDist[i].push_back(1);
256		} else {
257	<	if (oneThree.hasPair(iglob, jglob)) {
257	>	if (oneThree->hasPair(iglob, jglob)) {
258		toposForAtom[i].push_back(j);
259		topoDist[i].push_back(2);
260		} else {
261	<	if (oneFour.hasPair(iglob, jglob)) {
261	>	if (oneFour->hasPair(iglob, jglob)) {
262		toposForAtom[i].push_back(j);
263		topoDist[i].push_back(3);
264		}
#	Line 220 \| Line 266 \| namespace OpenMD {
266		}
267		}
268		}
269	<
269	>	#endif
270	>
271	>	// allocate memory for the parallel objects
272	>	atypesLocal.resize(nLocal_);
273	>
274	>	for (int i = 0; i < nLocal_; i++)
275	>	atypesLocal[i] = ff_->getAtomType(idents[i]);
276	>
277	>	groupList_.clear();
278	>	groupList_.resize(nGroups_);
279	>	for (int i = 0; i < nGroups_; i++) {
280	>	int gid = cgLocalToGlobal[i];
281	>	for (int j = 0; j < nLocal_; j++) {
282	>	int aid = AtomLocalToGlobal[j];
283	>	if (globalGroupMembership[aid] == gid) {
284	>	groupList_[i].push_back(j);
285	>	}
286	>	}
287	>	}
288	>
289	>
290		createGtypeCutoffMap();
291	+
292		}
293
294		void ForceMatrixDecomposition::createGtypeCutoffMap() {
295	<
295	>
296		RealType tol = 1e-6;
297	+	largestRcut_ = 0.0;
298		RealType rc;
299		int atid;
300		set<AtomType*> atypes = info_->getSimulatedAtomTypes();
301	<	vector<RealType> atypeCutoff;
302	<	atypeCutoff.resize( atypes.size() );
303	<
301	>
302	>	map<int, RealType> atypeCutoff;
303	>
304		for (set<AtomType*>::iterator at = atypes.begin();
305		at != atypes.end(); ++at){
238	–	rc = interactionMan_->getSuggestedCutoffRadius(*at);
306		atid = (*at)->getIdent();
307	<	atypeCutoff[atid] = rc;
307	>	if (userChoseCutoff_)
308	>	atypeCutoff[atid] = userCutoff_;
309	>	else
310	>	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
311		}
312	<
312	>
313		vector<RealType> gTypeCutoffs;
244	–
314		// first we do a single loop over the cutoff groups to find the
315		// largest cutoff for any atypes present in this group.
316		#ifdef IS_MPI
#	Line 299 \| Line 368 \| namespace OpenMD {
368
369		vector<RealType> groupCutoff(nGroups_, 0.0);
370		groupToGtype.resize(nGroups_);
302	–
303	–	cerr << "nGroups = " << nGroups_ << "\n";
371		for (int cg1 = 0; cg1 < nGroups_; cg1++) {
305	–
372		groupCutoff[cg1] = 0.0;
373		vector<int> atomList = getAtomsInGroupRow(cg1);
308	–
374		for (vector<int>::iterator ia = atomList.begin();
375		ia != atomList.end(); ++ia) {
376		int atom1 = (*ia);
377	<	atid = identsLocal[atom1];
378	<	if (atypeCutoff[atid] > groupCutoff[cg1]) {
377	>	atid = idents[atom1];
378	>	if (atypeCutoff[atid] > groupCutoff[cg1])
379		groupCutoff[cg1] = atypeCutoff[atid];
315	–	}
380		}
381	<
381	>
382		bool gTypeFound = false;
383		for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
384		if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
#	Line 322 \| Line 386 \| namespace OpenMD {
386		gTypeFound = true;
387		}
388		}
389	<	if (!gTypeFound) {
389	>	if (!gTypeFound) {
390		gTypeCutoffs.push_back( groupCutoff[cg1] );
391		groupToGtype[cg1] = gTypeCutoffs.size() - 1;
392		}
393		}
394		#endif
395
332	–	cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
396		// Now we find the maximum group cutoff value present in the simulation
397
398	<	RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
398	>	RealType groupMax = *max_element(gTypeCutoffs.begin(),
399	>	gTypeCutoffs.end());
400
401		#ifdef IS_MPI
402	<	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
402	>	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
403	>	MPI::MAX);
404		#endif
405
406		RealType tradRcut = groupMax;
#	Line 365 \| Line 430 \| namespace OpenMD {
430
431		pair<int,int> key = make_pair(i,j);
432		gTypeCutoffMap[key].first = thisRcut;
368	–
433		if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
370	–
434		gTypeCutoffMap[key].second = thisRcut*thisRcut;
372	–
435		gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
374	–
436		// sanity check
437
438		if (userChoseCutoff_) {
439		if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
440		sprintf(painCave.errMsg,
441		"ForceMatrixDecomposition::createGtypeCutoffMap "
442	<	"user-specified rCut does not match computed group Cutoff\n");
442	>	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
443		painCave.severity = OPENMD_ERROR;
444		painCave.isFatal = 1;
445		simError();
#	Line 410 \| Line 471 \| namespace OpenMD {
471		}
472
473		void ForceMatrixDecomposition::zeroWorkArrays() {
474	+	pairwisePot = 0.0;
475	+	embeddingPot = 0.0;
476
414	–	for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
415	–	longRangePot_[j] = 0.0;
416	–	}
417	–
477		#ifdef IS_MPI
478		if (storageLayout_ & DataStorage::dslForce) {
479		fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
#	Line 430 \| Line 489 \| namespace OpenMD {
489		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
490
491		fill(pot_col.begin(), pot_col.end(),
492	<	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
434	<
435	<	pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
492	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
493
494		if (storageLayout_ & DataStorage::dslParticlePot) {
495	<	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
496	<	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
495	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
496	>	0.0);
497	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
498	>	0.0);
499		}
500
501		if (storageLayout_ & DataStorage::dslDensity) {
#	Line 445 \| Line 504 \| namespace OpenMD {
504		}
505
506		if (storageLayout_ & DataStorage::dslFunctional) {
507	<	fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
508	<	fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
507	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
508	>	0.0);
509	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
510	>	0.0);
511		}
512
513		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
#	Line 456 \| Line 517 \| namespace OpenMD {
517		atomColData.functionalDerivative.end(), 0.0);
518		}
519
520	<	#else
521	<
520	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
521	>	fill(atomRowData.skippedCharge.begin(),
522	>	atomRowData.skippedCharge.end(), 0.0);
523	>	fill(atomColData.skippedCharge.begin(),
524	>	atomColData.skippedCharge.end(), 0.0);
525	>	}
526	>
527	>	#endif
528	>	// even in parallel, we need to zero out the local arrays:
529	>
530		if (storageLayout_ & DataStorage::dslParticlePot) {
531		fill(snap_->atomData.particlePot.begin(),
532		snap_->atomData.particlePot.end(), 0.0);
#	Line 475 \| Line 544 \| namespace OpenMD {
544		fill(snap_->atomData.functionalDerivative.begin(),
545		snap_->atomData.functionalDerivative.end(), 0.0);
546		}
547	<	#endif
547	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
548	>	fill(snap_->atomData.skippedCharge.begin(),
549	>	snap_->atomData.skippedCharge.end(), 0.0);
550	>	}
551
552		}
553
#	Line 486 \| Line 558 \| namespace OpenMD {
558		#ifdef IS_MPI
559
560		// gather up the atomic positions
561	<	AtomCommVectorRow->gather(snap_->atomData.position,
561	>	AtomPlanVectorRow->gather(snap_->atomData.position,
562		atomRowData.position);
563	<	AtomCommVectorColumn->gather(snap_->atomData.position,
563	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
564		atomColData.position);
565
566		// gather up the cutoff group positions
567	<	cgCommVectorRow->gather(snap_->cgData.position,
567	>
568	>	cgPlanVectorRow->gather(snap_->cgData.position,
569		cgRowData.position);
570	<	cgCommVectorColumn->gather(snap_->cgData.position,
570	>
571	>	cgPlanVectorColumn->gather(snap_->cgData.position,
572		cgColData.position);
573	+
574
575		// if needed, gather the atomic rotation matrices
576		if (storageLayout_ & DataStorage::dslAmat) {
577	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
577	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
578		atomRowData.aMat);
579	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
579	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
580		atomColData.aMat);
581		}
582
583		// if needed, gather the atomic eletrostatic frames
584		if (storageLayout_ & DataStorage::dslElectroFrame) {
585	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
585	>	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
586		atomRowData.electroFrame);
587	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
587	>	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
588		atomColData.electroFrame);
589		}
590	+
591		#endif
592		}
593
#	Line 525 \| Line 601 \| namespace OpenMD {
601
602		if (storageLayout_ & DataStorage::dslDensity) {
603
604	<	AtomCommRealRow->scatter(atomRowData.density,
604	>	AtomPlanRealRow->scatter(atomRowData.density,
605		snap_->atomData.density);
606
607		int n = snap_->atomData.density.size();
608		vector<RealType> rho_tmp(n, 0.0);
609	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
609	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
610		for (int i = 0; i < n; i++)
611		snap_->atomData.density[i] += rho_tmp[i];
612		}
#	Line 546 \| Line 622 \| namespace OpenMD {
622		storageLayout_ = sman_->getStorageLayout();
623		#ifdef IS_MPI
624		if (storageLayout_ & DataStorage::dslFunctional) {
625	<	AtomCommRealRow->gather(snap_->atomData.functional,
625	>	AtomPlanRealRow->gather(snap_->atomData.functional,
626		atomRowData.functional);
627	<	AtomCommRealColumn->gather(snap_->atomData.functional,
627	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
628		atomColData.functional);
629		}
630
631		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
632	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
632	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
633		atomRowData.functionalDerivative);
634	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
634	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
635		atomColData.functionalDerivative);
636		}
637		#endif
#	Line 569 \| Line 645 \| namespace OpenMD {
645		int n = snap_->atomData.force.size();
646		vector<Vector3d> frc_tmp(n, V3Zero);
647
648	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
648	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
649		for (int i = 0; i < n; i++) {
650		snap_->atomData.force[i] += frc_tmp[i];
651		frc_tmp[i] = 0.0;
652		}
653
654	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
655	<	for (int i = 0; i < n; i++)
654	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
655	>	for (int i = 0; i < n; i++) {
656		snap_->atomData.force[i] += frc_tmp[i];
657	<
658	<
657	>	}
658	>
659		if (storageLayout_ & DataStorage::dslTorque) {
660
661	<	int nt = snap_->atomData.force.size();
661	>	int nt = snap_->atomData.torque.size();
662		vector<Vector3d> trq_tmp(nt, V3Zero);
663
664	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
665	<	for (int i = 0; i < n; i++) {
664	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
665	>	for (int i = 0; i < nt; i++) {
666		snap_->atomData.torque[i] += trq_tmp[i];
667		trq_tmp[i] = 0.0;
668		}
669
670	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
671	<	for (int i = 0; i < n; i++)
670	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
671	>	for (int i = 0; i < nt; i++)
672		snap_->atomData.torque[i] += trq_tmp[i];
673		}
674	+
675	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
676	+
677	+	int ns = snap_->atomData.skippedCharge.size();
678	+	vector<RealType> skch_tmp(ns, 0.0);
679	+
680	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
681	+	for (int i = 0; i < ns; i++) {
682	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
683	+	skch_tmp[i] = 0.0;
684	+	}
685	+
686	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
687	+	for (int i = 0; i < ns; i++)
688	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
689	+
690	+	}
691
692		nLocal_ = snap_->getNumberOfAtoms();
693
#	Line 603 \| Line 696 \| namespace OpenMD {
696
697		// scatter/gather pot_row into the members of my column
698
699	<	AtomCommPotRow->scatter(pot_row, pot_temp);
699	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
700
701		for (int ii = 0; ii < pot_temp.size(); ii++ )
702	<	pot_local += pot_temp[ii];
702	>	pairwisePot += pot_temp[ii];
703
704		fill(pot_temp.begin(), pot_temp.end(),
705		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
706
707	<	AtomCommPotColumn->scatter(pot_col, pot_temp);
707	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
708
709		for (int ii = 0; ii < pot_temp.size(); ii++ )
710	<	pot_local += pot_temp[ii];
710	>	pairwisePot += pot_temp[ii];
711
712	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
713	+	RealType ploc1 = pairwisePot[ii];
714	+	RealType ploc2 = 0.0;
715	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
716	+	pairwisePot[ii] = ploc2;
717	+	}
718	+
719	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
720	+	RealType ploc1 = embeddingPot[ii];
721	+	RealType ploc2 = 0.0;
722	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
723	+	embeddingPot[ii] = ploc2;
724	+	}
725	+
726		#endif
727	+
728		}
729
730		int ForceMatrixDecomposition::getNAtomsInRow() {
#	Line 717 \| Line 825 \| namespace OpenMD {
825		return d;
826		}
827
828	<	vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
829	<	return skipsForAtom[atom1];
828	>	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
829	>	return excludesForAtom[atom1];
830		}
831
832		/**
833	<	* 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
833	>	* We need to exclude some overcounted interactions that result from
834		* the parallel decomposition.
835		*/
836		bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
837		int unique_id_1, unique_id_2;
838	<
838	>
839		#ifdef IS_MPI
840		// in MPI, we have to look up the unique IDs for each atom
841		unique_id_1 = AtomRowToGlobal[atom1];
842		unique_id_2 = AtomColToGlobal[atom2];
843	+	#else
844	+	unique_id_1 = AtomLocalToGlobal[atom1];
845	+	unique_id_2 = AtomLocalToGlobal[atom2];
846	+	#endif
847
739	–	// this situation should only arise in MPI simulations
848		if (unique_id_1 == unique_id_2) return true;
849	<
849	>
850	>	#ifdef IS_MPI
851		// this prevents us from doing the pair on multiple processors
852		if (unique_id_1 < unique_id_2) {
853		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
854		} else {
855	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
855	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
856		}
748	–	#else
749	–	// in the normal loop, the atom numbers are unique
750	–	unique_id_1 = atom1;
751	–	unique_id_2 = atom2;
857		#endif
858
859	<	for (vector<int>::iterator i = skipsForAtom[atom1].begin();
860	<	i != skipsForAtom[atom1].end(); ++i) {
756	<	if ( (*i) == unique_id_2 ) return true;
757	<	}
859	>	return false;
860	>	}
861
862	+	/**
863	+	* We need to handle the interactions for atoms who are involved in
864	+	* the same rigid body as well as some short range interactions
865	+	* (bonds, bends, torsions) differently from other interactions.
866	+	* We'll still visit the pairwise routines, but with a flag that
867	+	* tells those routines to exclude the pair from direct long range
868	+	* interactions. Some indirect interactions (notably reaction
869	+	* field) must still be handled for these pairs.
870	+	*/
871	+	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
872	+
873	+	// excludesForAtom was constructed to use row/column indices in the MPI
874	+	// version, and to use local IDs in the non-MPI version:
875	+
876	+	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
877	+	i != excludesForAtom[atom1].end(); ++i) {
878	+	if ( (*i) == atom2 ) return true;
879	+	}
880	+
881	+	return false;
882		}
883
884
#	Line 777 \| Line 900 \| namespace OpenMD {
900
901		// filling interaction blocks with pointers
902		void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
903	<	int atom1, int atom2) {
903	>	int atom1, int atom2) {
904	>
905	>	idat.excluded = excludeAtomPair(atom1, atom2);
906	>
907		#ifdef IS_MPI
908	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
909	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
910	+	// ff_->getAtomType(identsCol[atom2]) );
911
783	–	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
784	–	ff_->getAtomType(identsCol[atom2]) );
785	–
912		if (storageLayout_ & DataStorage::dslAmat) {
913		idat.A1 = &(atomRowData.aMat[atom1]);
914		idat.A2 = &(atomColData.aMat[atom2]);
#	Line 818 \| Line 944 \| namespace OpenMD {
944		idat.particlePot2 = &(atomColData.particlePot[atom2]);
945		}
946
947	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
948	+	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
949	+	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
950	+	}
951	+
952		#else
953
954	<	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
955	<	ff_->getAtomType(identsLocal[atom2]) );
954	>	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
955	>	//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
956	>	// ff_->getAtomType(idents[atom2]) );
957
958		if (storageLayout_ & DataStorage::dslAmat) {
959		idat.A1 = &(snap_->atomData.aMat[atom1]);
#	Line 838 \| Line 970 \| namespace OpenMD {
970		idat.t2 = &(snap_->atomData.torque[atom2]);
971		}
972
973	<	if (storageLayout_ & DataStorage::dslDensity) {
973	>	if (storageLayout_ & DataStorage::dslDensity) {
974		idat.rho1 = &(snap_->atomData.density[atom1]);
975		idat.rho2 = &(snap_->atomData.density[atom2]);
976		}
#	Line 858 \| Line 990 \| namespace OpenMD {
990		idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
991		}
992
993	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
994	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
995	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
996	+	}
997		#endif
998		}
999
#	Line 870 \| Line 1006 \| namespace OpenMD {
1006		atomRowData.force[atom1] += *(idat.f1);
1007		atomColData.force[atom2] -= *(idat.f1);
1008		#else
1009	<	longRangePot_ += *(idat.pot);
1010	<
1009	>	pairwisePot += *(idat.pot);
1010	>
1011		snap_->atomData.force[atom1] += *(idat.f1);
1012		snap_->atomData.force[atom2] -= *(idat.f1);
1013		#endif
1014	<
1014	>
1015		}
1016
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	–
1017		/*
1018		* buildNeighborList
1019		*
#	Line 918 \| Line 1024 \| namespace OpenMD {
1024
1025		vector<pair<int, int> > neighborList;
1026		groupCutoffs cuts;
1027	+	bool doAllPairs = false;
1028	+
1029		#ifdef IS_MPI
1030		cellListRow_.clear();
1031		cellListCol_.clear();
#	Line 937 \| Line 1045 \| namespace OpenMD {
1045		nCells_.y() = (int) ( Hy.length() )/ rList_;
1046		nCells_.z() = (int) ( Hz.length() )/ rList_;
1047
1048	+	// handle small boxes where the cell offsets can end up repeating cells
1049	+
1050	+	if (nCells_.x() < 3) doAllPairs = true;
1051	+	if (nCells_.y() < 3) doAllPairs = true;
1052	+	if (nCells_.z() < 3) doAllPairs = true;
1053	+
1054		Mat3x3d invHmat = snap_->getInvHmat();
1055		Vector3d rs, scaled, dr;
1056		Vector3i whichCell;
#	Line 950 \| Line 1064 \| namespace OpenMD {
1064		cellList_.resize(nCtot);
1065		#endif
1066
1067	+	if (!doAllPairs) {
1068		#ifdef IS_MPI
954	–	for (int i = 0; i < nGroupsInRow_; i++) {
955	–	rs = cgRowData.position[i];
1069
1070	<	// scaled positions relative to the box vectors
1071	<	scaled = invHmat * rs;
1072	<
1073	<	// wrap the vector back into the unit box by subtracting integer box
1074	<	// numbers
1075	<	for (int j = 0; j < 3; j++) {
1076	<	scaled[j] -= roundMe(scaled[j]);
1077	<	scaled[j] += 0.5;
1070	>	for (int i = 0; i < nGroupsInRow_; i++) {
1071	>	rs = cgRowData.position[i];
1072	>
1073	>	// scaled positions relative to the box vectors
1074	>	scaled = invHmat * rs;
1075	>
1076	>	// wrap the vector back into the unit box by subtracting integer box
1077	>	// numbers
1078	>	for (int j = 0; j < 3; j++) {
1079	>	scaled[j] -= roundMe(scaled[j]);
1080	>	scaled[j] += 0.5;
1081	>	}
1082	>
1083	>	// find xyz-indices of cell that cutoffGroup is in.
1084	>	whichCell.x() = nCells_.x() * scaled.x();
1085	>	whichCell.y() = nCells_.y() * scaled.y();
1086	>	whichCell.z() = nCells_.z() * scaled.z();
1087	>
1088	>	// find single index of this cell:
1089	>	cellIndex = Vlinear(whichCell, nCells_);
1090	>
1091	>	// add this cutoff group to the list of groups in this cell;
1092	>	cellListRow_[cellIndex].push_back(i);
1093		}
1094	<
1095	<	// find xyz-indices of cell that cutoffGroup is in.
1096	<	whichCell.x() = nCells_.x() * scaled.x();
1097	<	whichCell.y() = nCells_.y() * scaled.y();
1098	<	whichCell.z() = nCells_.z() * scaled.z();
1099	<
1100	<	// find single index of this cell:
1101	<	cellIndex = Vlinear(whichCell, nCells_);
1102	<
1103	<	// add this cutoff group to the list of groups in this cell;
1104	<	cellListRow_[cellIndex].push_back(i);
1105	<	}
1106	<
1107	<	for (int i = 0; i < nGroupsInCol_; i++) {
1108	<	rs = cgColData.position[i];
1109	<
1110	<	// scaled positions relative to the box vectors
1111	<	scaled = invHmat * rs;
1112	<
1113	<	// wrap the vector back into the unit box by subtracting integer box
1114	<	// numbers
1115	<	for (int j = 0; j < 3; j++) {
1116	<	scaled[j] -= roundMe(scaled[j]);
989	<	scaled[j] += 0.5;
1094	>	for (int i = 0; i < nGroupsInCol_; i++) {
1095	>	rs = cgColData.position[i];
1096	>
1097	>	// scaled positions relative to the box vectors
1098	>	scaled = invHmat * rs;
1099	>
1100	>	// wrap the vector back into the unit box by subtracting integer box
1101	>	// numbers
1102	>	for (int j = 0; j < 3; j++) {
1103	>	scaled[j] -= roundMe(scaled[j]);
1104	>	scaled[j] += 0.5;
1105	>	}
1106	>
1107	>	// find xyz-indices of cell that cutoffGroup is in.
1108	>	whichCell.x() = nCells_.x() * scaled.x();
1109	>	whichCell.y() = nCells_.y() * scaled.y();
1110	>	whichCell.z() = nCells_.z() * scaled.z();
1111	>
1112	>	// find single index of this cell:
1113	>	cellIndex = Vlinear(whichCell, nCells_);
1114	>
1115	>	// add this cutoff group to the list of groups in this cell;
1116	>	cellListCol_[cellIndex].push_back(i);
1117		}
1118	<
992	<	// find xyz-indices of cell that cutoffGroup is in.
993	<	whichCell.x() = nCells_.x() * scaled.x();
994	<	whichCell.y() = nCells_.y() * scaled.y();
995	<	whichCell.z() = nCells_.z() * scaled.z();
996	<
997	<	// find single index of this cell:
998	<	cellIndex = Vlinear(whichCell, nCells_);
999	<
1000	<	// add this cutoff group to the list of groups in this cell;
1001	<	cellListCol_[cellIndex].push_back(i);
1002	<	}
1118	>
1119		#else
1120	<	for (int i = 0; i < nGroups_; i++) {
1121	<	rs = snap_->cgData.position[i];
1122	<
1123	<	// scaled positions relative to the box vectors
1124	<	scaled = invHmat * rs;
1125	<
1126	<	// wrap the vector back into the unit box by subtracting integer box
1127	<	// numbers
1128	<	for (int j = 0; j < 3; j++) {
1129	<	scaled[j] -= roundMe(scaled[j]);
1130	<	scaled[j] += 0.5;
1120	>	for (int i = 0; i < nGroups_; i++) {
1121	>	rs = snap_->cgData.position[i];
1122	>
1123	>	// scaled positions relative to the box vectors
1124	>	scaled = invHmat * rs;
1125	>
1126	>	// wrap the vector back into the unit box by subtracting integer box
1127	>	// numbers
1128	>	for (int j = 0; j < 3; j++) {
1129	>	scaled[j] -= roundMe(scaled[j]);
1130	>	scaled[j] += 0.5;
1131	>	}
1132	>
1133	>	// find xyz-indices of cell that cutoffGroup is in.
1134	>	whichCell.x() = nCells_.x() * scaled.x();
1135	>	whichCell.y() = nCells_.y() * scaled.y();
1136	>	whichCell.z() = nCells_.z() * scaled.z();
1137	>
1138	>	// find single index of this cell:
1139	>	cellIndex = Vlinear(whichCell, nCells_);
1140	>
1141	>	// add this cutoff group to the list of groups in this cell;
1142	>	cellList_[cellIndex].push_back(i);
1143		}
1144
1017	–	// find xyz-indices of cell that cutoffGroup is in.
1018	–	whichCell.x() = nCells_.x() * scaled.x();
1019	–	whichCell.y() = nCells_.y() * scaled.y();
1020	–	whichCell.z() = nCells_.z() * scaled.z();
1021	–
1022	–	// find single index of this cell:
1023	–	cellIndex = Vlinear(whichCell, nCells_);
1024	–
1025	–	// add this cutoff group to the list of groups in this cell;
1026	–	cellList_[cellIndex].push_back(i);
1027	–	}
1145		#endif
1146
1147	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1148	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1149	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1150	<	Vector3i m1v(m1x, m1y, m1z);
1151	<	int m1 = Vlinear(m1v, nCells_);
1035	<
1036	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1037	<	os != cellOffsets_.end(); ++os) {
1147	>	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1148	>	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1149	>	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1150	>	Vector3i m1v(m1x, m1y, m1z);
1151	>	int m1 = Vlinear(m1v, nCells_);
1152
1153	<	Vector3i m2v = m1v + (*os);
1154	<
1155	<	if (m2v.x() >= nCells_.x()) {
1156	<	m2v.x() = 0;
1157	<	} else if (m2v.x() < 0) {
1044	<	m2v.x() = nCells_.x() - 1;
1045	<	}
1046	<
1047	<	if (m2v.y() >= nCells_.y()) {
1048	<	m2v.y() = 0;
1049	<	} else if (m2v.y() < 0) {
1050	<	m2v.y() = nCells_.y() - 1;
1051	<	}
1052	<
1053	<	if (m2v.z() >= nCells_.z()) {
1054	<	m2v.z() = 0;
1055	<	} else if (m2v.z() < 0) {
1056	<	m2v.z() = nCells_.z() - 1;
1057	<	}
1058	<
1059	<	int m2 = Vlinear (m2v, nCells_);
1153	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1154	>	os != cellOffsets_.end(); ++os) {
1155	>
1156	>	Vector3i m2v = m1v + (*os);
1157	>
1158
1159	<	#ifdef IS_MPI
1160	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1161	<	j1 != cellListRow_[m1].end(); ++j1) {
1162	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1163	<	j2 != cellListCol_[m2].end(); ++j2) {
1164	<
1165	<	// Always do this if we're in different cells or if
1166	<	// we're in the same cell and the global index of the
1167	<	// j2 cutoff group is less than the j1 cutoff group
1159	>	if (m2v.x() >= nCells_.x()) {
1160	>	m2v.x() = 0;
1161	>	} else if (m2v.x() < 0) {
1162	>	m2v.x() = nCells_.x() - 1;
1163	>	}
1164	>
1165	>	if (m2v.y() >= nCells_.y()) {
1166	>	m2v.y() = 0;
1167	>	} else if (m2v.y() < 0) {
1168	>	m2v.y() = nCells_.y() - 1;
1169	>	}
1170	>
1171	>	if (m2v.z() >= nCells_.z()) {
1172	>	m2v.z() = 0;
1173	>	} else if (m2v.z() < 0) {
1174	>	m2v.z() = nCells_.z() - 1;
1175	>	}
1176
1177	<	if (m2 != m1 \|\| cgColToGlobal[(j2)] < cgRowToGlobal[(j1)]) {
1177	>	int m2 = Vlinear (m2v, nCells_);
1178	>
1179	>	#ifdef IS_MPI
1180	>	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1181	>	j1 != cellListRow_[m1].end(); ++j1) {
1182	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1183	>	j2 != cellListCol_[m2].end(); ++j2) {
1184	>
1185	>	// In parallel, we need to visit all pairs of row
1186	>	// & column indicies and will divide labor in the
1187	>	// force evaluation later.
1188		dr = cgColData.position[(j2)] - cgRowData.position[(j1)];
1189		snap_->wrapVector(dr);
1190		cuts = getGroupCutoffs( (j1), (j2) );
1191		if (dr.lengthSquare() < cuts.third) {
1192		neighborList.push_back(make_pair((j1), (j2)));
1193	<	}
1193	>	}
1194		}
1195		}
1080	–	}
1196		#else
1197	+	for (vector<int>::iterator j1 = cellList_[m1].begin();
1198	+	j1 != cellList_[m1].end(); ++j1) {
1199	+	for (vector<int>::iterator j2 = cellList_[m2].begin();
1200	+	j2 != cellList_[m2].end(); ++j2) {
1201	+
1202	+	// Always do this if we're in different cells or if
1203	+	// we're in the same cell and the global index of
1204	+	// the j2 cutoff group is greater than or equal to
1205	+	// the j1 cutoff group. Note that Rappaport's code
1206	+	// has a "less than" conditional here, but that
1207	+	// deals with atom-by-atom computation. OpenMD
1208	+	// allows atoms within a single cutoff group to
1209	+	// interact with each other.
1210
1083	–	for (vector<int>::iterator j1 = cellList_[m1].begin();
1084	–	j1 != cellList_[m1].end(); ++j1) {
1085	–	for (vector<int>::iterator j2 = cellList_[m2].begin();
1086	–	j2 != cellList_[m2].end(); ++j2) {
1211
1088	–	// Always do this if we're in different cells or if
1089	–	// we're in the same cell and the global index of the
1090	–	// j2 cutoff group is less than the j1 cutoff group
1212
1213	<	if (m2 != m1 \|\| (j2) < (j1)) {
1214	<	dr = snap_->cgData.position[(j2)] - snap_->cgData.position[(j1)];
1215	<	snap_->wrapVector(dr);
1216	<	cuts = getGroupCutoffs( (j1), (j2) );
1217	<	if (dr.lengthSquare() < cuts.third) {
1218	<	neighborList.push_back(make_pair((j1), (j2)));
1213	>	if (m2 != m1 \|\| (j2) >= (j1) ) {
1214	>
1215	>	dr = snap_->cgData.position[(j2)] - snap_->cgData.position[(j1)];
1216	>	snap_->wrapVector(dr);
1217	>	cuts = getGroupCutoffs( (j1), (j2) );
1218	>	if (dr.lengthSquare() < cuts.third) {
1219	>	neighborList.push_back(make_pair((j1), (j2)));
1220	>	}
1221		}
1222		}
1223		}
1101	–	}
1224		#endif
1225	+	}
1226		}
1227		}
1228		}
1229	+	} else {
1230	+	// branch to do all cutoff group pairs
1231	+	#ifdef IS_MPI
1232	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1233	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1234	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1235	+	snap_->wrapVector(dr);
1236	+	cuts = getGroupCutoffs( j1, j2 );
1237	+	if (dr.lengthSquare() < cuts.third) {
1238	+	neighborList.push_back(make_pair(j1, j2));
1239	+	}
1240	+	}
1241	+	}
1242	+	#else
1243	+	// include all groups here.
1244	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1245	+	// include self group interactions j2 == j1
1246	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1247	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1248	+	snap_->wrapVector(dr);
1249	+	cuts = getGroupCutoffs( j1, j2 );
1250	+	if (dr.lengthSquare() < cuts.third) {
1251	+	neighborList.push_back(make_pair(j1, j2));
1252	+	}
1253	+	}
1254	+	}
1255	+	#endif
1256		}
1257	<
1257	>
1258		// save the local cutoff group positions for the check that is
1259		// done on each loop:
1260		saved_CG_positions_.clear();

Diff Legend

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

Comparing branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents): Revision 1582 by gezelter, Tue Jun 14 20:41:44 2011 UTC vs. Revision 1616 by gezelter, Tue Aug 30 15:45:35 2011 UTC

Diff Legend

Comparing branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents):
Revision 1582 by gezelter, Tue Jun 14 20:41:44 2011 UTC vs.
Revision 1616 by gezelter, Tue Aug 30 15:45:35 2011 UTC