ViewVC Help

View File | Revision Log | Show Annotations | View Changeset | Root Listing

root/OpenMD/trunk/src/parallel/ForceMatrixDecomposition.cpp

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1588 by gezelter, Sat Jul 9 15:05:59 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1971 by gezelter, Fri Feb 28 13:25:13 2014 UTC

#	Line 35 | Line 35
35		*
36		* [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37		* [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38	<	* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
39	<	* [4]  Vardeman & Gezelter, in progress (2009).
38	>	* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008).
39	>	* [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40	>	* [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41		*/
42		#include "parallel/ForceMatrixDecomposition.hpp"
43		#include "math/SquareMatrix3.hpp"
#	Line 47 | Line 48 | namespace OpenMD {
48		using namespace std;
49		namespace OpenMD {
50
51	+	ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52	+
53	+	// In a parallel computation, row and colum scans must visit all
54	+	// surrounding cells (not just the 14 upper triangular blocks that
55	+	// are used when the processor can see all pairs)
56	+	#ifdef IS_MPI
57	+	cellOffsets_.clear();
58	+	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59	+	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60	+	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
61	+	cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62	+	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63	+	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
64	+	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65	+	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
66	+	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67	+	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68	+	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69	+	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70	+	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
71	+	cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72	+	cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73	+	cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74	+	cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75	+	cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76	+	cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77	+	cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78	+	cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79	+	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80	+	cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81	+	cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82	+	cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83	+	cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84	+	cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85	+	#endif
86	+	}
87	+
88	+
89		/**
90		* distributeInitialData is essentially a copy of the older fortran
91		* SimulationSetup
92		*/
54	–
93		void ForceMatrixDecomposition::distributeInitialData() {
94		snap_ = sman_->getCurrentSnapshot();
95		storageLayout_ = sman_->getStorageLayout();
96		ff_ = info_->getForceField();
97		nLocal_ = snap_->getNumberOfAtoms();
98	<
98	>
99		nGroups_ = info_->getNLocalCutoffGroups();
100		// gather the information for atomtype IDs (atids):
101		idents = info_->getIdentArray();
102	+	regions = info_->getRegions();
103		AtomLocalToGlobal = info_->getGlobalAtomIndices();
104		cgLocalToGlobal = info_->getGlobalGroupIndices();
105		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
#	Line 71 | Line 110 | namespace OpenMD {
110		PairList* oneTwo = info_->getOneTwoInteractions();
111		PairList* oneThree = info_->getOneThreeInteractions();
112		PairList* oneFour = info_->getOneFourInteractions();
113	<
113	>
114	>	if (needVelocities_)
115	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
116	>	DataStorage::dslVelocity);
117	>	else
118	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
119	>
120		#ifdef IS_MPI
121
122	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
123	<	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_);
122	>	MPI_Comm row = rowComm.getComm();
123	>	MPI_Comm col = colComm.getComm();
124
125	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
126	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
127	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
128	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
129	<	AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
125	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
126	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
127	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
128	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
129	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
130
131	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
132	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
133	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
134	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
131	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
132	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
133	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
134	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
135	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
136
137	<	nAtomsInRow_ = AtomCommIntRow->getSize();
138	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
139	<	nGroupsInRow_ = cgCommIntRow->getSize();
140	<	nGroupsInCol_ = cgCommIntColumn->getSize();
137	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
138	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
139	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
140	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
141
142	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
143	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
144	+	nGroupsInRow_ = cgPlanIntRow->getSize();
145	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
146	+
147		// Modify the data storage objects with the correct layouts and sizes:
148		atomRowData.resize(nAtomsInRow_);
149		atomRowData.setStorageLayout(storageLayout_);
#	Line 104 | Line 152 | namespace OpenMD {
152		cgRowData.resize(nGroupsInRow_);
153		cgRowData.setStorageLayout(DataStorage::dslPosition);
154		cgColData.resize(nGroupsInCol_);
155	<	cgColData.setStorageLayout(DataStorage::dslPosition);
156	<
155	>	if (needVelocities_)
156	>	// we only need column velocities if we need them.
157	>	cgColData.setStorageLayout(DataStorage::dslPosition |
158	>	DataStorage::dslVelocity);
159	>	else
160	>	cgColData.setStorageLayout(DataStorage::dslPosition);
161	>
162		identsRow.resize(nAtomsInRow_);
163		identsCol.resize(nAtomsInCol_);
164
165	<	AtomCommIntRow->gather(idents, identsRow);
166	<	AtomCommIntColumn->gather(idents, identsCol);
165	>	AtomPlanIntRow->gather(idents, identsRow);
166	>	AtomPlanIntColumn->gather(idents, identsCol);
167	>
168	>	regionsRow.resize(nAtomsInRow_);
169	>	regionsCol.resize(nAtomsInCol_);
170
171	<	vector<int>::iterator it;
172	<	for (it = AtomLocalToGlobal.begin(); it != AtomLocalToGlobal.end(); ++it) {
117	<	cerr << "my AtomLocalToGlobal = " << (*it) << "\n";
118	<	}
119	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
120	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
171	>	AtomPlanIntRow->gather(regions, regionsRow);
172	>	AtomPlanIntColumn->gather(regions, regionsCol);
173
174	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
175	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
174	>	// allocate memory for the parallel objects
175	>	atypesRow.resize(nAtomsInRow_);
176	>	atypesCol.resize(nAtomsInCol_);
177
178	<	AtomCommRealRow->gather(massFactors, massFactorsRow);
179	<	AtomCommRealColumn->gather(massFactors, massFactorsCol);
178	>	for (int i = 0; i < nAtomsInRow_; i++)
179	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
180	>	for (int i = 0; i < nAtomsInCol_; i++)
181	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
182	>
183	>	pot_row.resize(nAtomsInRow_);
184	>	pot_col.resize(nAtomsInCol_);
185
186	+	expot_row.resize(nAtomsInRow_);
187	+	expot_col.resize(nAtomsInCol_);
188	+
189	+	AtomRowToGlobal.resize(nAtomsInRow_);
190	+	AtomColToGlobal.resize(nAtomsInCol_);
191	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
192	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
193	+
194	+	cgRowToGlobal.resize(nGroupsInRow_);
195	+	cgColToGlobal.resize(nGroupsInCol_);
196	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
197	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
198	+
199	+	massFactorsRow.resize(nAtomsInRow_);
200	+	massFactorsCol.resize(nAtomsInCol_);
201	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
202	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
203	+
204		groupListRow_.clear();
205		groupListRow_.resize(nGroupsInRow_);
206		for (int i = 0; i < nGroupsInRow_; i++) {
#	Line 179 | Line 255 | namespace OpenMD {
255		}
256		}
257
258	<	#endif
183	<
184	<	groupList_.clear();
185	<	groupList_.resize(nGroups_);
186	<	for (int i = 0; i < nGroups_; i++) {
187	<	int gid = cgLocalToGlobal[i];
188	<	for (int j = 0; j < nLocal_; j++) {
189	<	int aid = AtomLocalToGlobal[j];
190	<	if (globalGroupMembership[aid] == gid) {
191	<	groupList_[i].push_back(j);
192	<	}
193	<	}
194	<	}
195	<
258	>	#else
259		excludesForAtom.clear();
260		excludesForAtom.resize(nLocal_);
261		toposForAtom.clear();
#	Line 225 | Line 288 | namespace OpenMD {
288		}
289		}
290		}
291	<
291	>	#endif
292	>
293	>	// allocate memory for the parallel objects
294	>	atypesLocal.resize(nLocal_);
295	>
296	>	for (int i = 0; i < nLocal_; i++)
297	>	atypesLocal[i] = ff_->getAtomType(idents[i]);
298	>
299	>	groupList_.clear();
300	>	groupList_.resize(nGroups_);
301	>	for (int i = 0; i < nGroups_; i++) {
302	>	int gid = cgLocalToGlobal[i];
303	>	for (int j = 0; j < nLocal_; j++) {
304	>	int aid = AtomLocalToGlobal[j];
305	>	if (globalGroupMembership[aid] == gid) {
306	>	groupList_[i].push_back(j);
307	>	}
308	>	}
309	>	}
310	>
311	>
312		createGtypeCutoffMap();
313
314		}
315
316		void ForceMatrixDecomposition::createGtypeCutoffMap() {
317
318	+	GrCut.clear();
319	+	GrCutSq.clear();
320	+	GrlistSq.clear();
321	+
322		RealType tol = 1e-6;
323	<	RealType rc;
323	>	largestRcut_ = 0.0;
324		int atid;
325		set<AtomType*> atypes = info_->getSimulatedAtomTypes();
326	+
327		map<int, RealType> atypeCutoff;
328
329		for (set<AtomType*>::iterator at = atypes.begin();
#	Line 243 | Line 331 | namespace OpenMD {
331		atid = (*at)->getIdent();
332		if (userChoseCutoff_)
333		atypeCutoff[atid] = userCutoff_;
334	<	else
334	>	else
335		atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
336		}
337	<
337	>
338		vector<RealType> gTypeCutoffs;
339		// first we do a single loop over the cutoff groups to find the
340		// largest cutoff for any atypes present in this group.
#	Line 306 | Line 394 | namespace OpenMD {
394		vector<RealType> groupCutoff(nGroups_, 0.0);
395		groupToGtype.resize(nGroups_);
396		for (int cg1 = 0; cg1 < nGroups_; cg1++) {
309	–
397		groupCutoff[cg1] = 0.0;
398		vector<int> atomList = getAtomsInGroupRow(cg1);
312	–
399		for (vector<int>::iterator ia = atomList.begin();
400		ia != atomList.end(); ++ia) {
401		int atom1 = (*ia);
402		atid = idents[atom1];
403	<	if (atypeCutoff[atid] > groupCutoff[cg1]) {
403	>	if (atypeCutoff[atid] > groupCutoff[cg1])
404		groupCutoff[cg1] = atypeCutoff[atid];
319	–	}
405		}
406	<
406	>
407		bool gTypeFound = false;
408	<	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
408	>	for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
409		if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
410		groupToGtype[cg1] = gt;
411		gTypeFound = true;
412		}
413		}
414	<	if (!gTypeFound) {
414	>	if (!gTypeFound) {
415		gTypeCutoffs.push_back( groupCutoff[cg1] );
416		groupToGtype[cg1] = gTypeCutoffs.size() - 1;
417		}
#	Line 335 | Line 420 | namespace OpenMD {
420
421		// Now we find the maximum group cutoff value present in the simulation
422
423	<	RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
423	>	RealType groupMax = *max_element(gTypeCutoffs.begin(),
424	>	gTypeCutoffs.end());
425
426		#ifdef IS_MPI
427	<	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
427	>	MPI_Allreduce(MPI_IN_PLACE, &groupMax, 1, MPI_REALTYPE,
428	>	MPI_MAX, MPI_COMM_WORLD);
429		#endif
430
431		RealType tradRcut = groupMax;
432
433	<	for (int i = 0; i < gTypeCutoffs.size();  i++) {
434	<	for (int j = 0; j < gTypeCutoffs.size();  j++) {
433	>	GrCut.resize( gTypeCutoffs.size() );
434	>	GrCutSq.resize( gTypeCutoffs.size() );
435	>	GrlistSq.resize( gTypeCutoffs.size() );
436	>
437	>
438	>	for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) {
439	>	GrCut[i].resize( gTypeCutoffs.size() , 0.0);
440	>	GrCutSq[i].resize( gTypeCutoffs.size(), 0.0 );
441	>	GrlistSq[i].resize( gTypeCutoffs.size(), 0.0 );
442	>
443	>	for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) {
444		RealType thisRcut;
445		switch(cutoffPolicy_) {
446		case TRADITIONAL:
#	Line 366 | Line 462 | namespace OpenMD {
462		break;
463		}
464
465	<	pair<int,int> key = make_pair(i,j);
370	<	gTypeCutoffMap[key].first = thisRcut;
371	<
465	>	GrCut[i][j] = thisRcut;
466		if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
467	+	GrCutSq[i][j] = thisRcut * thisRcut;
468	+	GrlistSq[i][j] = pow(thisRcut + skinThickness_, 2);
469
470	<	gTypeCutoffMap[key].second = thisRcut*thisRcut;
471	<
472	<	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
377	<
470	>	// pair<int,int> key = make_pair(i,j);
471	>	// gTypeCutoffMap[key].first = thisRcut;
472	>	// gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
473		// sanity check
474
475		if (userChoseCutoff_) {
476	<	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
476	>	if (abs(GrCut[i][j] - userCutoff_) > 0.0001) {
477		sprintf(painCave.errMsg,
478		"ForceMatrixDecomposition::createGtypeCutoffMap "
479		"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
#	Line 391 | Line 486 | namespace OpenMD {
486		}
487		}
488
489	<
395	<	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
489	>	void ForceMatrixDecomposition::getGroupCutoffs(int &cg1, int &cg2, RealType &rcut, RealType &rcutsq, RealType &rlistsq) {
490		int i, j;
491		#ifdef IS_MPI
492		i = groupRowToGtype[cg1];
#	Line 401 | Line 495 | namespace OpenMD {
495		i = groupToGtype[cg1];
496		j = groupToGtype[cg2];
497		#endif
498	<	return gTypeCutoffMap[make_pair(i,j)];
498	>	rcut = GrCut[i][j];
499	>	rcutsq = GrCutSq[i][j];
500	>	rlistsq = GrlistSq[i][j];
501	>	return;
502	>	//return gTypeCutoffMap[make_pair(i,j)];
503		}
504
505		int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
506	<	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
506	>	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
507		if (toposForAtom[atom1][j] == atom2)
508		return topoDist[atom1][j];
509	<	}
509	>	}
510		return 0;
511		}
512
513		void ForceMatrixDecomposition::zeroWorkArrays() {
514		pairwisePot = 0.0;
515		embeddingPot = 0.0;
516	+	excludedPot = 0.0;
517	+	excludedSelfPot = 0.0;
518
519		#ifdef IS_MPI
520		if (storageLayout_ & DataStorage::dslForce) {
#	Line 433 | Line 533 | namespace OpenMD {
533		fill(pot_col.begin(), pot_col.end(),
534		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
535
536	+	fill(expot_row.begin(), expot_row.end(),
537	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
538	+
539	+	fill(expot_col.begin(), expot_col.end(),
540	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
541	+
542		if (storageLayout_ & DataStorage::dslParticlePot) {
543	<	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
544	<	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
543	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
544	>	0.0);
545	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
546	>	0.0);
547		}
548
549		if (storageLayout_ & DataStorage::dslDensity) {
#	Line 444 | Line 552 | namespace OpenMD {
552		}
553
554		if (storageLayout_ & DataStorage::dslFunctional) {
555	<	fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
556	<	fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
555	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
556	>	0.0);
557	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
558	>	0.0);
559		}
560
561		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
#	Line 462 | Line 572 | namespace OpenMD {
572		atomColData.skippedCharge.end(), 0.0);
573		}
574
575	<	#else
576	<
575	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
576	>	fill(atomRowData.flucQFrc.begin(),
577	>	atomRowData.flucQFrc.end(), 0.0);
578	>	fill(atomColData.flucQFrc.begin(),
579	>	atomColData.flucQFrc.end(), 0.0);
580	>	}
581	>
582	>	if (storageLayout_ & DataStorage::dslElectricField) {
583	>	fill(atomRowData.electricField.begin(),
584	>	atomRowData.electricField.end(), V3Zero);
585	>	fill(atomColData.electricField.begin(),
586	>	atomColData.electricField.end(), V3Zero);
587	>	}
588	>
589	>	#endif
590	>	// even in parallel, we need to zero out the local arrays:
591	>
592		if (storageLayout_ & DataStorage::dslParticlePot) {
593		fill(snap_->atomData.particlePot.begin(),
594		snap_->atomData.particlePot.end(), 0.0);
#	Line 473 | Line 598 | namespace OpenMD {
598		fill(snap_->atomData.density.begin(),
599		snap_->atomData.density.end(), 0.0);
600		}
601	+
602		if (storageLayout_ & DataStorage::dslFunctional) {
603		fill(snap_->atomData.functional.begin(),
604		snap_->atomData.functional.end(), 0.0);
605		}
606	+
607		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
608		fill(snap_->atomData.functionalDerivative.begin(),
609		snap_->atomData.functionalDerivative.end(), 0.0);
610		}
611	+
612		if (storageLayout_ & DataStorage::dslSkippedCharge) {
613		fill(snap_->atomData.skippedCharge.begin(),
614		snap_->atomData.skippedCharge.end(), 0.0);
615		}
616	<	#endif
617	<
616	>
617	>	if (storageLayout_ & DataStorage::dslElectricField) {
618	>	fill(snap_->atomData.electricField.begin(),
619	>	snap_->atomData.electricField.end(), V3Zero);
620	>	}
621		}
622
623
#	Line 496 | Line 627 | namespace OpenMD {
627		#ifdef IS_MPI
628
629		// gather up the atomic positions
630	<	AtomCommVectorRow->gather(snap_->atomData.position,
630	>	AtomPlanVectorRow->gather(snap_->atomData.position,
631		atomRowData.position);
632	<	AtomCommVectorColumn->gather(snap_->atomData.position,
632	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
633		atomColData.position);
634
635		// gather up the cutoff group positions
636	<	cgCommVectorRow->gather(snap_->cgData.position,
636	>
637	>	cgPlanVectorRow->gather(snap_->cgData.position,
638		cgRowData.position);
639	<	cgCommVectorColumn->gather(snap_->cgData.position,
639	>
640	>	cgPlanVectorColumn->gather(snap_->cgData.position,
641		cgColData.position);
642	+
643	+
644	+
645	+	if (needVelocities_) {
646	+	// gather up the atomic velocities
647	+	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
648	+	atomColData.velocity);
649	+
650	+	cgPlanVectorColumn->gather(snap_->cgData.velocity,
651	+	cgColData.velocity);
652	+	}
653	+
654
655		// if needed, gather the atomic rotation matrices
656		if (storageLayout_ & DataStorage::dslAmat) {
657	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
657	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
658		atomRowData.aMat);
659	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
659	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
660		atomColData.aMat);
661		}
662	<
663	<	// if needed, gather the atomic eletrostatic frames
664	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
665	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
666	<	atomRowData.electroFrame);
667	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
668	<	atomColData.electroFrame);
662	>
663	>	// if needed, gather the atomic eletrostatic information
664	>	if (storageLayout_ & DataStorage::dslDipole) {
665	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
666	>	atomRowData.dipole);
667	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
668	>	atomColData.dipole);
669		}
670	+
671	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
672	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
673	+	atomRowData.quadrupole);
674	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
675	+	atomColData.quadrupole);
676	+	}
677	+
678	+	// if needed, gather the atomic fluctuating charge values
679	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
680	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
681	+	atomRowData.flucQPos);
682	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
683	+	atomColData.flucQPos);
684	+	}
685	+
686		#endif
687		}
688
#	Line 535 | Line 696 | namespace OpenMD {
696
697		if (storageLayout_ & DataStorage::dslDensity) {
698
699	<	AtomCommRealRow->scatter(atomRowData.density,
699	>	AtomPlanRealRow->scatter(atomRowData.density,
700		snap_->atomData.density);
701
702		int n = snap_->atomData.density.size();
703		vector<RealType> rho_tmp(n, 0.0);
704	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
704	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
705		for (int i = 0; i < n; i++)
706		snap_->atomData.density[i] += rho_tmp[i];
707	+	}
708	+
709	+	// this isn't necessary if we don't have polarizable atoms, but
710	+	// we'll leave it here for now.
711	+	if (storageLayout_ & DataStorage::dslElectricField) {
712	+
713	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
714	+	snap_->atomData.electricField);
715	+
716	+	int n = snap_->atomData.electricField.size();
717	+	vector<Vector3d> field_tmp(n, V3Zero);
718	+	AtomPlanVectorColumn->scatter(atomColData.electricField,
719	+	field_tmp);
720	+	for (int i = 0; i < n; i++)
721	+	snap_->atomData.electricField[i] += field_tmp[i];
722		}
723		#endif
724		}
#	Line 556 | Line 732 | namespace OpenMD {
732		storageLayout_ = sman_->getStorageLayout();
733		#ifdef IS_MPI
734		if (storageLayout_ & DataStorage::dslFunctional) {
735	<	AtomCommRealRow->gather(snap_->atomData.functional,
735	>	AtomPlanRealRow->gather(snap_->atomData.functional,
736		atomRowData.functional);
737	<	AtomCommRealColumn->gather(snap_->atomData.functional,
737	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
738		atomColData.functional);
739		}
740
741		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
742	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
742	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
743		atomRowData.functionalDerivative);
744	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
744	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
745		atomColData.functionalDerivative);
746		}
747		#endif
#	Line 579 | Line 755 | namespace OpenMD {
755		int n = snap_->atomData.force.size();
756		vector<Vector3d> frc_tmp(n, V3Zero);
757
758	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
758	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
759		for (int i = 0; i < n; i++) {
760		snap_->atomData.force[i] += frc_tmp[i];
761		frc_tmp[i] = 0.0;
762		}
763
764	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
765	<	for (int i = 0; i < n; i++)
764	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
765	>	for (int i = 0; i < n; i++) {
766		snap_->atomData.force[i] += frc_tmp[i];
767	<
768	<
767	>	}
768	>
769		if (storageLayout_ & DataStorage::dslTorque) {
770
771		int nt = snap_->atomData.torque.size();
772		vector<Vector3d> trq_tmp(nt, V3Zero);
773
774	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
774	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
775		for (int i = 0; i < nt; i++) {
776		snap_->atomData.torque[i] += trq_tmp[i];
777		trq_tmp[i] = 0.0;
778		}
779
780	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
780	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
781		for (int i = 0; i < nt; i++)
782		snap_->atomData.torque[i] += trq_tmp[i];
783		}
#	Line 611 | Line 787 | namespace OpenMD {
787		int ns = snap_->atomData.skippedCharge.size();
788		vector<RealType> skch_tmp(ns, 0.0);
789
790	<	AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
790	>	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
791		for (int i = 0; i < ns; i++) {
792	<	snap_->atomData.skippedCharge[i] = skch_tmp[i];
792	>	snap_->atomData.skippedCharge[i] += skch_tmp[i];
793		skch_tmp[i] = 0.0;
794		}
795
796	<	AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
797	<	for (int i = 0; i < ns; i++)
796	>	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
797	>	for (int i = 0; i < ns; i++)
798		snap_->atomData.skippedCharge[i] += skch_tmp[i];
799	+
800		}
801
802	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
803	+
804	+	int nq = snap_->atomData.flucQFrc.size();
805	+	vector<RealType> fqfrc_tmp(nq, 0.0);
806	+
807	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
808	+	for (int i = 0; i < nq; i++) {
809	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
810	+	fqfrc_tmp[i] = 0.0;
811	+	}
812	+
813	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
814	+	for (int i = 0; i < nq; i++)
815	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
816	+
817	+	}
818	+
819	+	if (storageLayout_ & DataStorage::dslElectricField) {
820	+
821	+	int nef = snap_->atomData.electricField.size();
822	+	vector<Vector3d> efield_tmp(nef, V3Zero);
823	+
824	+	AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
825	+	for (int i = 0; i < nef; i++) {
826	+	snap_->atomData.electricField[i] += efield_tmp[i];
827	+	efield_tmp[i] = 0.0;
828	+	}
829	+
830	+	AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
831	+	for (int i = 0; i < nef; i++)
832	+	snap_->atomData.electricField[i] += efield_tmp[i];
833	+	}
834	+
835	+
836		nLocal_ = snap_->getNumberOfAtoms();
837
838		vector<potVec> pot_temp(nLocal_,
839		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
840	+	vector<potVec> expot_temp(nLocal_,
841	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
842
843		// scatter/gather pot_row into the members of my column
844
845	<	AtomCommPotRow->scatter(pot_row, pot_temp);
845	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
846	>	AtomPlanPotRow->scatter(expot_row, expot_temp);
847
848	<	for (int ii = 0;  ii < pot_temp.size(); ii++ )
848	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
849		pairwisePot += pot_temp[ii];
850	<
850	>
851	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
852	>	excludedPot += expot_temp[ii];
853	>
854	>	if (storageLayout_ & DataStorage::dslParticlePot) {
855	>	// This is the pairwise contribution to the particle pot.  The
856	>	// embedding contribution is added in each of the low level
857	>	// non-bonded routines.  In single processor, this is done in
858	>	// unpackInteractionData, not in collectData.
859	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
860	>	for (int i = 0; i < nLocal_; i++) {
861	>	// factor of two is because the total potential terms are divided
862	>	// by 2 in parallel due to row/ column scatter
863	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
864	>	}
865	>	}
866	>	}
867	>
868		fill(pot_temp.begin(), pot_temp.end(),
869		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
870	+	fill(expot_temp.begin(), expot_temp.end(),
871	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
872
873	<	AtomCommPotColumn->scatter(pot_col, pot_temp);
873	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
874	>	AtomPlanPotColumn->scatter(expot_col, expot_temp);
875
876		for (int ii = 0;  ii < pot_temp.size(); ii++ )
877		pairwisePot += pot_temp[ii];
878	+
879	+	for (int ii = 0;  ii < expot_temp.size(); ii++ )
880	+	excludedPot += expot_temp[ii];
881	+
882	+	if (storageLayout_ & DataStorage::dslParticlePot) {
883	+	// This is the pairwise contribution to the particle pot.  The
884	+	// embedding contribution is added in each of the low level
885	+	// non-bonded routines.  In single processor, this is done in
886	+	// unpackInteractionData, not in collectData.
887	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
888	+	for (int i = 0; i < nLocal_; i++) {
889	+	// factor of two is because the total potential terms are divided
890	+	// by 2 in parallel due to row/ column scatter
891	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
892	+	}
893	+	}
894	+	}
895	+
896	+	if (storageLayout_ & DataStorage::dslParticlePot) {
897	+	int npp = snap_->atomData.particlePot.size();
898	+	vector<RealType> ppot_temp(npp, 0.0);
899	+
900	+	// This is the direct or embedding contribution to the particle
901	+	// pot.
902	+
903	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
904	+	for (int i = 0; i < npp; i++) {
905	+	snap_->atomData.particlePot[i] += ppot_temp[i];
906	+	}
907	+
908	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
909	+
910	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
911	+	for (int i = 0; i < npp; i++) {
912	+	snap_->atomData.particlePot[i] += ppot_temp[i];
913	+	}
914	+	}
915	+
916	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
917	+	RealType ploc1 = pairwisePot[ii];
918	+	RealType ploc2 = 0.0;
919	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
920	+	pairwisePot[ii] = ploc2;
921	+	}
922	+
923	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
924	+	RealType ploc1 = excludedPot[ii];
925	+	RealType ploc2 = 0.0;
926	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
927	+	excludedPot[ii] = ploc2;
928	+	}
929	+
930	+	// Here be dragons.
931	+	MPI_Comm col = colComm.getComm();
932	+
933	+	MPI_Allreduce(MPI_IN_PLACE,
934	+	&snap_->frameData.conductiveHeatFlux[0], 3,
935	+	MPI_REALTYPE, MPI_SUM, col);
936	+
937	+
938		#endif
939
940		}
941
942	<	int ForceMatrixDecomposition::getNAtomsInRow() {
942	>	/**
943	>	* Collects information obtained during the post-pair (and embedding
944	>	* functional) loops onto local data structures.
945	>	*/
946	>	void ForceMatrixDecomposition::collectSelfData() {
947	>	snap_ = sman_->getCurrentSnapshot();
948	>	storageLayout_ = sman_->getStorageLayout();
949	>
950		#ifdef IS_MPI
951	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
952	+	RealType ploc1 = embeddingPot[ii];
953	+	RealType ploc2 = 0.0;
954	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
955	+	embeddingPot[ii] = ploc2;
956	+	}
957	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
958	+	RealType ploc1 = excludedSelfPot[ii];
959	+	RealType ploc2 = 0.0;
960	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
961	+	excludedSelfPot[ii] = ploc2;
962	+	}
963	+	#endif
964	+
965	+	}
966	+
967	+
968	+
969	+	int& ForceMatrixDecomposition::getNAtomsInRow() {
970	+	#ifdef IS_MPI
971		return nAtomsInRow_;
972		#else
973		return nLocal_;
#	Line 656 | Line 977 | namespace OpenMD {
977		/**
978		* returns the list of atoms belonging to this group.
979		*/
980	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
980	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
981		#ifdef IS_MPI
982		return groupListRow_[cg1];
983		#else
#	Line 664 | Line 985 | namespace OpenMD {
985		#endif
986		}
987
988	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
988	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
989		#ifdef IS_MPI
990		return groupListCol_[cg2];
991		#else
#	Line 681 | Line 1002 | namespace OpenMD {
1002		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
1003		#endif
1004
1005	<	snap_->wrapVector(d);
1005	>	if (usePeriodicBoundaryConditions_) {
1006	>	snap_->wrapVector(d);
1007	>	}
1008		return d;
1009		}
1010
1011	+	Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
1012	+	#ifdef IS_MPI
1013	+	return cgColData.velocity[cg2];
1014	+	#else
1015	+	return snap_->cgData.velocity[cg2];
1016	+	#endif
1017	+	}
1018
1019	+	Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
1020	+	#ifdef IS_MPI
1021	+	return atomColData.velocity[atom2];
1022	+	#else
1023	+	return snap_->atomData.velocity[atom2];
1024	+	#endif
1025	+	}
1026	+
1027	+
1028		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
1029
1030		Vector3d d;
#	Line 695 | Line 1034 | namespace OpenMD {
1034		#else
1035		d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
1036		#endif
1037	<
1038	<	snap_->wrapVector(d);
1037	>	if (usePeriodicBoundaryConditions_) {
1038	>	snap_->wrapVector(d);
1039	>	}
1040		return d;
1041		}
1042
#	Line 708 | Line 1048 | namespace OpenMD {
1048		#else
1049		d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1050		#endif
1051	<
1052	<	snap_->wrapVector(d);
1051	>	if (usePeriodicBoundaryConditions_) {
1052	>	snap_->wrapVector(d);
1053	>	}
1054		return d;
1055		}
1056
1057	<	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1057	>	RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1058		#ifdef IS_MPI
1059		return massFactorsRow[atom1];
1060		#else
#	Line 721 | Line 1062 | namespace OpenMD {
1062		#endif
1063		}
1064
1065	<	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1065	>	RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1066		#ifdef IS_MPI
1067		return massFactorsCol[atom2];
1068		#else
#	Line 738 | Line 1079 | namespace OpenMD {
1079		#else
1080		d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1081		#endif
1082	<
1083	<	snap_->wrapVector(d);
1082	>	if (usePeriodicBoundaryConditions_) {
1083	>	snap_->wrapVector(d);
1084	>	}
1085		return d;
1086		}
1087
1088	<	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1088	>	vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1089		return excludesForAtom[atom1];
1090		}
1091
#	Line 751 | Line 1093 | namespace OpenMD {
1093		* We need to exclude some overcounted interactions that result from
1094		* the parallel decomposition.
1095		*/
1096	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1096	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1097		int unique_id_1, unique_id_2;
1098	<
1098	>
1099		#ifdef IS_MPI
1100		// in MPI, we have to look up the unique IDs for each atom
1101		unique_id_1 = AtomRowToGlobal[atom1];
1102		unique_id_2 = AtomColToGlobal[atom2];
1103	+	// group1 = cgRowToGlobal[cg1];
1104	+	// group2 = cgColToGlobal[cg2];
1105	+	#else
1106	+	unique_id_1 = AtomLocalToGlobal[atom1];
1107	+	unique_id_2 = AtomLocalToGlobal[atom2];
1108	+	int group1 = cgLocalToGlobal[cg1];
1109	+	int group2 = cgLocalToGlobal[cg2];
1110	+	#endif
1111
762	–	// this situation should only arise in MPI simulations
1112		if (unique_id_1 == unique_id_2) return true;
1113	<
1113	>
1114	>	#ifdef IS_MPI
1115		// this prevents us from doing the pair on multiple processors
1116		if (unique_id_1 < unique_id_2) {
1117		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1118		} else {
1119	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1119	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1120		}
1121	+	#endif
1122	+
1123	+	#ifndef IS_MPI
1124	+	if (group1 == group2) {
1125	+	if (unique_id_1 < unique_id_2) return true;
1126	+	}
1127		#endif
1128	+
1129		return false;
1130		}
1131
#	Line 782 | Line 1139 | namespace OpenMD {
1139		* field) must still be handled for these pairs.
1140		*/
1141		bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1142	<	int unique_id_2;
1142	>
1143	>	// excludesForAtom was constructed to use row/column indices in the MPI
1144	>	// version, and to use local IDs in the non-MPI version:
1145
787	–	#ifdef IS_MPI
788	–	// in MPI, we have to look up the unique IDs for the row atom.
789	–	unique_id_2 = AtomColToGlobal[atom2];
790	–	#else
791	–	// in the normal loop, the atom numbers are unique
792	–	unique_id_2 = atom2;
793	–	#endif
794	–
1146		for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1147		i != excludesForAtom[atom1].end(); ++i) {
1148	<	if ( (*i) == unique_id_2 ) return true;
1148	>	if ( (*i) == atom2 ) return true;
1149		}
1150
1151		return false;
#	Line 824 | Line 1175 | namespace OpenMD {
1175		idat.excluded = excludeAtomPair(atom1, atom2);
1176
1177		#ifdef IS_MPI
1178	<
1179	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1180	<	ff_->getAtomType(identsCol[atom2]) );
1181	<
1178	>	//idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1179	>	idat.atid1 = identsRow[atom1];
1180	>	idat.atid2 = identsCol[atom2];
1181	>
1182	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1183	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1184	>	} else {
1185	>	idat.sameRegion = false;
1186	>	}
1187	>
1188		if (storageLayout_ & DataStorage::dslAmat) {
1189		idat.A1 = &(atomRowData.aMat[atom1]);
1190		idat.A2 = &(atomColData.aMat[atom2]);
1191		}
1192
836	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
837	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
838	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
839	–	}
840	–
1193		if (storageLayout_ & DataStorage::dslTorque) {
1194		idat.t1 = &(atomRowData.torque[atom1]);
1195		idat.t2 = &(atomColData.torque[atom2]);
1196		}
1197
1198	+	if (storageLayout_ & DataStorage::dslDipole) {
1199	+	idat.dipole1 = &(atomRowData.dipole[atom1]);
1200	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1201	+	}
1202	+
1203	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1204	+	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1205	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1206	+	}
1207	+
1208		if (storageLayout_ & DataStorage::dslDensity) {
1209		idat.rho1 = &(atomRowData.density[atom1]);
1210		idat.rho2 = &(atomColData.density[atom2]);
#	Line 868 | Line 1230 | namespace OpenMD {
1230		idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1231		}
1232
1233	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1234	+	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1235	+	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1236	+	}
1237	+
1238		#else
1239	+
1240	+	//idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1241	+	idat.atid1 = idents[atom1];
1242	+	idat.atid2 = idents[atom2];
1243
1244	<	idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1245	<	ff_->getAtomType(idents[atom2]) );
1244	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1245	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1246	>	} else {
1247	>	idat.sameRegion = false;
1248	>	}
1249
1250		if (storageLayout_ & DataStorage::dslAmat) {
1251		idat.A1 = &(snap_->atomData.aMat[atom1]);
1252		idat.A2 = &(snap_->atomData.aMat[atom2]);
1253		}
1254
881	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
882	–	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
883	–	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
884	–	}
885	–
1255		if (storageLayout_ & DataStorage::dslTorque) {
1256		idat.t1 = &(snap_->atomData.torque[atom1]);
1257		idat.t2 = &(snap_->atomData.torque[atom2]);
1258	+	}
1259	+
1260	+	if (storageLayout_ & DataStorage::dslDipole) {
1261	+	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1262	+	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1263	+	}
1264	+
1265	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1266	+	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1267	+	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1268		}
1269
1270		if (storageLayout_ & DataStorage::dslDensity) {
#	Line 912 | Line 1291 | namespace OpenMD {
1291		idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1292		idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1293		}
1294	+
1295	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1296	+	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1297	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1298	+	}
1299	+
1300		#endif
1301		}
1302
1303
1304		void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1305		#ifdef IS_MPI
1306	<	pot_row[atom1] += 0.5 *  *(idat.pot);
1307	<	pot_col[atom2] += 0.5 *  *(idat.pot);
1306	>	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1307	>	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1308	>	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1309	>	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1310
1311		atomRowData.force[atom1] += *(idat.f1);
1312		atomColData.force[atom2] -= *(idat.f1);
1313	+
1314	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1315	+	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1316	+	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1317	+	}
1318	+
1319	+	if (storageLayout_ & DataStorage::dslElectricField) {
1320	+	atomRowData.electricField[atom1] += *(idat.eField1);
1321	+	atomColData.electricField[atom2] += *(idat.eField2);
1322	+	}
1323	+
1324		#else
1325		pairwisePot += *(idat.pot);
1326	+	excludedPot += *(idat.excludedPot);
1327
1328		snap_->atomData.force[atom1] += *(idat.f1);
1329		snap_->atomData.force[atom2] -= *(idat.f1);
1330	+
1331	+	if (idat.doParticlePot) {
1332	+	// This is the pairwise contribution to the particle pot.  The
1333	+	// embedding contribution is added in each of the low level
1334	+	// non-bonded routines.  In parallel, this calculation is done
1335	+	// in collectData, not in unpackInteractionData.
1336	+	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1337	+	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1338	+	}
1339	+
1340	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1341	+	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1342	+	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1343	+	}
1344	+
1345	+	if (storageLayout_ & DataStorage::dslElectricField) {
1346	+	snap_->atomData.electricField[atom1] += *(idat.eField1);
1347	+	snap_->atomData.electricField[atom2] += *(idat.eField2);
1348	+	}
1349	+
1350		#endif
1351
1352		}
#	Line 938 | Line 1357 | namespace OpenMD {
1357		* first element of pair is row-indexed CutoffGroup
1358		* second element of pair is column-indexed CutoffGroup
1359		*/
1360	<	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1361	<
1362	<	vector<pair<int, int> > neighborList;
1360	>	void ForceMatrixDecomposition::buildNeighborList(vector<pair<int,int> >& neighborList) {
1361	>
1362	>	neighborList.clear();
1363		groupCutoffs cuts;
1364		bool doAllPairs = false;
1365
1366	+	RealType rList_ = (largestRcut_ + skinThickness_);
1367	+	RealType rcut, rcutsq, rlistsq;
1368	+	Snapshot* snap_ = sman_->getCurrentSnapshot();
1369	+	Mat3x3d box;
1370	+	Mat3x3d invBox;
1371	+
1372	+	Vector3d rs, scaled, dr;
1373	+	Vector3i whichCell;
1374	+	int cellIndex;
1375	+
1376		#ifdef IS_MPI
1377		cellListRow_.clear();
1378		cellListCol_.clear();
1379		#else
1380		cellList_.clear();
1381		#endif
1382	<
1383	<	RealType rList_ = (largestRcut_ + skinThickness_);
1384	<	RealType rl2 = rList_ * rList_;
1385	<	Snapshot* snap_ = sman_->getCurrentSnapshot();
1386	<	Mat3x3d Hmat = snap_->getHmat();
1387	<	Vector3d Hx = Hmat.getColumn(0);
1388	<	Vector3d Hy = Hmat.getColumn(1);
1389	<	Vector3d Hz = Hmat.getColumn(2);
1390	<
1391	<	nCells_.x() = (int) ( Hx.length() )/ rList_;
1392	<	nCells_.y() = (int) ( Hy.length() )/ rList_;
1393	<	nCells_.z() = (int) ( Hz.length() )/ rList_;
1394	<
1382	>
1383	>	if (!usePeriodicBoundaryConditions_) {
1384	>	box = snap_->getBoundingBox();
1385	>	invBox = snap_->getInvBoundingBox();
1386	>	} else {
1387	>	box = snap_->getHmat();
1388	>	invBox = snap_->getInvHmat();
1389	>	}
1390	>
1391	>	Vector3d boxX = box.getColumn(0);
1392	>	Vector3d boxY = box.getColumn(1);
1393	>	Vector3d boxZ = box.getColumn(2);
1394	>
1395	>	nCells_.x() = int( boxX.length() / rList_ );
1396	>	nCells_.y() = int( boxY.length() / rList_ );
1397	>	nCells_.z() = int( boxZ.length() / rList_ );
1398	>
1399		// handle small boxes where the cell offsets can end up repeating cells
1400
1401		if (nCells_.x() < 3) doAllPairs = true;
1402		if (nCells_.y() < 3) doAllPairs = true;
1403		if (nCells_.z() < 3) doAllPairs = true;
1404	<
972	<	Mat3x3d invHmat = snap_->getInvHmat();
973	<	Vector3d rs, scaled, dr;
974	<	Vector3i whichCell;
975	<	int cellIndex;
1404	>
1405		int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1406	<
1406	>
1407		#ifdef IS_MPI
1408		cellListRow_.resize(nCtot);
1409		cellListCol_.resize(nCtot);
1410		#else
1411		cellList_.resize(nCtot);
1412		#endif
1413	<
1413	>
1414		if (!doAllPairs) {
1415		#ifdef IS_MPI
1416	<
1416	>
1417		for (int i = 0; i < nGroupsInRow_; i++) {
1418		rs = cgRowData.position[i];
1419
1420		// scaled positions relative to the box vectors
1421	<	scaled = invHmat * rs;
1421	>	scaled = invBox * rs;
1422
1423		// wrap the vector back into the unit box by subtracting integer box
1424		// numbers
1425		for (int j = 0; j < 3; j++) {
1426		scaled[j] -= roundMe(scaled[j]);
1427		scaled[j] += 0.5;
1428	+	// Handle the special case when an object is exactly on the
1429	+	// boundary (a scaled coordinate of 1.0 is the same as
1430	+	// scaled coordinate of 0.0)
1431	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1432		}
1433
1434		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1009 | Line 1442 | namespace OpenMD {
1442		// add this cutoff group to the list of groups in this cell;
1443		cellListRow_[cellIndex].push_back(i);
1444		}
1012	–
1445		for (int i = 0; i < nGroupsInCol_; i++) {
1446		rs = cgColData.position[i];
1447
1448		// scaled positions relative to the box vectors
1449	<	scaled = invHmat * rs;
1449	>	scaled = invBox * rs;
1450
1451		// wrap the vector back into the unit box by subtracting integer box
1452		// numbers
1453		for (int j = 0; j < 3; j++) {
1454		scaled[j] -= roundMe(scaled[j]);
1455		scaled[j] += 0.5;
1456	+	// Handle the special case when an object is exactly on the
1457	+	// boundary (a scaled coordinate of 1.0 is the same as
1458	+	// scaled coordinate of 0.0)
1459	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1460		}
1461
1462		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1034 | Line 1470 | namespace OpenMD {
1470		// add this cutoff group to the list of groups in this cell;
1471		cellListCol_[cellIndex].push_back(i);
1472		}
1473	+
1474		#else
1475		for (int i = 0; i < nGroups_; i++) {
1476		rs = snap_->cgData.position[i];
1477
1478		// scaled positions relative to the box vectors
1479	<	scaled = invHmat * rs;
1479	>	scaled = invBox * rs;
1480
1481		// wrap the vector back into the unit box by subtracting integer box
1482		// numbers
1483		for (int j = 0; j < 3; j++) {
1484		scaled[j] -= roundMe(scaled[j]);
1485		scaled[j] += 0.5;
1486	+	// Handle the special case when an object is exactly on the
1487	+	// boundary (a scaled coordinate of 1.0 is the same as
1488	+	// scaled coordinate of 0.0)
1489	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1490		}
1491
1492		// find xyz-indices of cell that cutoffGroup is in.
1493	<	whichCell.x() = nCells_.x() * scaled.x();
1494	<	whichCell.y() = nCells_.y() * scaled.y();
1495	<	whichCell.z() = nCells_.z() * scaled.z();
1493	>	whichCell.x() = int(nCells_.x() * scaled.x());
1494	>	whichCell.y() = int(nCells_.y() * scaled.y());
1495	>	whichCell.z() = int(nCells_.z() * scaled.z());
1496
1497		// find single index of this cell:
1498	<	cellIndex = Vlinear(whichCell, nCells_);
1498	>	cellIndex = Vlinear(whichCell, nCells_);
1499
1500		// add this cutoff group to the list of groups in this cell;
1501		cellList_[cellIndex].push_back(i);
1502		}
1503	+
1504		#endif
1505
1506		for (int m1z = 0; m1z < nCells_.z(); m1z++) {
#	Line 1071 | Line 1513 | namespace OpenMD {
1513		os != cellOffsets_.end(); ++os) {
1514
1515		Vector3i m2v = m1v + (*os);
1516	<
1516	>
1517	>
1518		if (m2v.x() >= nCells_.x()) {
1519		m2v.x() = 0;
1520		} else if (m2v.x() < 0) {
#	Line 1089 | Line 1532 | namespace OpenMD {
1532		} else if (m2v.z() < 0) {
1533		m2v.z() = nCells_.z() - 1;
1534		}
1535	<
1535	>
1536		int m2 = Vlinear (m2v, nCells_);
1537
1538		#ifdef IS_MPI
#	Line 1098 | Line 1541 | namespace OpenMD {
1541		for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1542		j2 != cellListCol_[m2].end(); ++j2) {
1543
1544	<	// Always do this if we're in different cells or if
1545	<	// we're in the same cell and the global index of the
1546	<	// j2 cutoff group is less than the j1 cutoff group
1547	<
1548	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1106	<	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1544	>	// In parallel, we need to visit *all* pairs of row
1545	>	// & column indicies and will divide labor in the
1546	>	// force evaluation later.
1547	>	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1548	>	if (usePeriodicBoundaryConditions_) {
1549		snap_->wrapVector(dr);
1108	–	cuts = getGroupCutoffs( (*j1), (*j2) );
1109	–	if (dr.lengthSquare() < cuts.third) {
1110	–	neighborList.push_back(make_pair((*j1), (*j2)));
1111	–	}
1550		}
1551	+	getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq );
1552	+	if (dr.lengthSquare() < rlistsq) {
1553	+	neighborList.push_back(make_pair((*j1), (*j2)));
1554	+	}
1555		}
1556		}
1557		#else
1116	–
1558		for (vector<int>::iterator j1 = cellList_[m1].begin();
1559		j1 != cellList_[m1].end(); ++j1) {
1560		for (vector<int>::iterator j2 = cellList_[m2].begin();
1561		j2 != cellList_[m2].end(); ++j2) {
1562	<
1562	>
1563		// Always do this if we're in different cells or if
1564	<	// we're in the same cell and the global index of the
1565	<	// j2 cutoff group is less than the j1 cutoff group
1566	<
1567	<	if (m2 != m1 || (*j2) < (*j1)) {
1564	>	// we're in the same cell and the global index of
1565	>	// the j2 cutoff group is greater than or equal to
1566	>	// the j1 cutoff group.  Note that Rappaport's code
1567	>	// has a "less than" conditional here, but that
1568	>	// deals with atom-by-atom computation.  OpenMD
1569	>	// allows atoms within a single cutoff group to
1570	>	// interact with each other.
1571	>
1572	>	if (m2 != m1 || (*j2) >= (*j1) ) {
1573	>
1574		dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1575	<	snap_->wrapVector(dr);
1576	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1577	<	if (dr.lengthSquare() < cuts.third) {
1575	>	if (usePeriodicBoundaryConditions_) {
1576	>	snap_->wrapVector(dr);
1577	>	}
1578	>	getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq );
1579	>	if (dr.lengthSquare() < rlistsq) {
1580		neighborList.push_back(make_pair((*j1), (*j2)));
1581		}
1582		}
#	Line 1142 | Line 1591 | namespace OpenMD {
1591		// branch to do all cutoff group pairs
1592		#ifdef IS_MPI
1593		for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1594	<	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1594	>	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1595		dr = cgColData.position[j2] - cgRowData.position[j1];
1596	<	snap_->wrapVector(dr);
1597	<	cuts = getGroupCutoffs( j1, j2 );
1598	<	if (dr.lengthSquare() < cuts.third) {
1596	>	if (usePeriodicBoundaryConditions_) {
1597	>	snap_->wrapVector(dr);
1598	>	}
1599	>	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq);
1600	>	if (dr.lengthSquare() < rlistsq) {
1601		neighborList.push_back(make_pair(j1, j2));
1602		}
1603		}
1604	<	}
1604	>	}
1605		#else
1606	<	for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1607	<	for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1606	>	// include all groups here.
1607	>	for (int j1 = 0; j1 < nGroups_; j1++) {
1608	>	// include self group interactions j2 == j1
1609	>	for (int j2 = j1; j2 < nGroups_; j2++) {
1610		dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1611	<	snap_->wrapVector(dr);
1612	<	cuts = getGroupCutoffs( j1, j2 );
1613	<	if (dr.lengthSquare() < cuts.third) {
1611	>	if (usePeriodicBoundaryConditions_) {
1612	>	snap_->wrapVector(dr);
1613	>	}
1614	>	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq );
1615	>	if (dr.lengthSquare() < rlistsq) {
1616		neighborList.push_back(make_pair(j1, j2));
1617		}
1618	<	}
1619	<	}
1618	>	}
1619	>	}
1620		#endif
1621		}
1622
#	Line 1170 | Line 1625 | namespace OpenMD {
1625		saved_CG_positions_.clear();
1626		for (int i = 0; i < nGroups_; i++)
1627		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1173	–
1174	–	return neighborList;
1628		}
1629		} //end namespace OpenMD

Diff Legend

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