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 1587 by gezelter, Fri Jul 8 20:25:32 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1896 by gezelter, Tue Jul 2 20:02:31 2013 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();
#	Line 71 | Line 109 | namespace OpenMD {
109		PairList* oneTwo = info_->getOneTwoInteractions();
110		PairList* oneThree = info_->getOneThreeInteractions();
111		PairList* oneFour = info_->getOneFourInteractions();
112	<
112	>
113	>	if (needVelocities_)
114	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
115	>	DataStorage::dslVelocity);
116	>	else
117	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
118	>
119		#ifdef IS_MPI
120
121	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
122	<	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_);
121	>	MPI::Intracomm row = rowComm.getComm();
122	>	MPI::Intracomm col = colComm.getComm();
123
124	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
125	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
126	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
127	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
128	<	AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
124	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
125	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
126	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
127	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
128	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
129
130	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
131	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
132	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
133	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
130	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
131	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
132	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
133	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
134	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
135
136	<	nAtomsInRow_ = AtomCommIntRow->getSize();
137	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
138	<	nGroupsInRow_ = cgCommIntRow->getSize();
139	<	nGroupsInCol_ = cgCommIntColumn->getSize();
136	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
137	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
138	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
139	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
140
141	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
142	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
143	+	nGroupsInRow_ = cgPlanIntRow->getSize();
144	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
145	+
146		// Modify the data storage objects with the correct layouts and sizes:
147		atomRowData.resize(nAtomsInRow_);
148		atomRowData.setStorageLayout(storageLayout_);
#	Line 104 | Line 151 | namespace OpenMD {
151		cgRowData.resize(nGroupsInRow_);
152		cgRowData.setStorageLayout(DataStorage::dslPosition);
153		cgColData.resize(nGroupsInCol_);
154	<	cgColData.setStorageLayout(DataStorage::dslPosition);
155	<
154	>	if (needVelocities_)
155	>	// we only need column velocities if we need them.
156	>	cgColData.setStorageLayout(DataStorage::dslPosition |
157	>	DataStorage::dslVelocity);
158	>	else
159	>	cgColData.setStorageLayout(DataStorage::dslPosition);
160	>
161		identsRow.resize(nAtomsInRow_);
162		identsCol.resize(nAtomsInCol_);
163
164	<	AtomCommIntRow->gather(idents, identsRow);
165	<	AtomCommIntColumn->gather(idents, identsCol);
164	>	AtomPlanIntRow->gather(idents, identsRow);
165	>	AtomPlanIntColumn->gather(idents, identsCol);
166
167	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
168	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
169	<
118	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
119	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
167	>	// allocate memory for the parallel objects
168	>	atypesRow.resize(nAtomsInRow_);
169	>	atypesCol.resize(nAtomsInCol_);
170
171	<	AtomCommRealRow->gather(massFactors, massFactorsRow);
172	<	AtomCommRealColumn->gather(massFactors, massFactorsCol);
171	>	for (int i = 0; i < nAtomsInRow_; i++)
172	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
173	>	for (int i = 0; i < nAtomsInCol_; i++)
174	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
175
176	+	pot_row.resize(nAtomsInRow_);
177	+	pot_col.resize(nAtomsInCol_);
178	+
179	+	expot_row.resize(nAtomsInRow_);
180	+	expot_col.resize(nAtomsInCol_);
181	+
182	+	AtomRowToGlobal.resize(nAtomsInRow_);
183	+	AtomColToGlobal.resize(nAtomsInCol_);
184	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186	+
187	+	cgRowToGlobal.resize(nGroupsInRow_);
188	+	cgColToGlobal.resize(nGroupsInCol_);
189	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
191	+
192	+	massFactorsRow.resize(nAtomsInRow_);
193	+	massFactorsCol.resize(nAtomsInCol_);
194	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
195	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
196	+
197		groupListRow_.clear();
198		groupListRow_.resize(nGroupsInRow_);
199		for (int i = 0; i < nGroupsInRow_; i++) {
#	Line 175 | Line 248 | namespace OpenMD {
248		}
249		}
250
251	<	#endif
179	<
180	<	groupList_.clear();
181	<	groupList_.resize(nGroups_);
182	<	for (int i = 0; i < nGroups_; i++) {
183	<	int gid = cgLocalToGlobal[i];
184	<	for (int j = 0; j < nLocal_; j++) {
185	<	int aid = AtomLocalToGlobal[j];
186	<	if (globalGroupMembership[aid] == gid) {
187	<	groupList_[i].push_back(j);
188	<	}
189	<	}
190	<	}
191	<
251	>	#else
252		excludesForAtom.clear();
253		excludesForAtom.resize(nLocal_);
254		toposForAtom.clear();
#	Line 221 | Line 281 | namespace OpenMD {
281		}
282		}
283		}
284	<
284	>	#endif
285	>
286	>	// allocate memory for the parallel objects
287	>	atypesLocal.resize(nLocal_);
288	>
289	>	for (int i = 0; i < nLocal_; i++)
290	>	atypesLocal[i] = ff_->getAtomType(idents[i]);
291	>
292	>	groupList_.clear();
293	>	groupList_.resize(nGroups_);
294	>	for (int i = 0; i < nGroups_; i++) {
295	>	int gid = cgLocalToGlobal[i];
296	>	for (int j = 0; j < nLocal_; j++) {
297	>	int aid = AtomLocalToGlobal[j];
298	>	if (globalGroupMembership[aid] == gid) {
299	>	groupList_[i].push_back(j);
300	>	}
301	>	}
302	>	}
303	>
304	>
305		createGtypeCutoffMap();
306
307		}
308
309		void ForceMatrixDecomposition::createGtypeCutoffMap() {
310
311	+	GrCut.clear();
312	+	GrCutSq.clear();
313	+	GrlistSq.clear();
314	+
315		RealType tol = 1e-6;
316	<	RealType rc;
316	>	largestRcut_ = 0.0;
317		int atid;
318		set<AtomType*> atypes = info_->getSimulatedAtomTypes();
319	+
320		map<int, RealType> atypeCutoff;
321
322		for (set<AtomType*>::iterator at = atypes.begin();
#	Line 239 | Line 324 | namespace OpenMD {
324		atid = (*at)->getIdent();
325		if (userChoseCutoff_)
326		atypeCutoff[atid] = userCutoff_;
327	<	else
327	>	else
328		atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
329		}
330	<
330	>
331		vector<RealType> gTypeCutoffs;
332		// first we do a single loop over the cutoff groups to find the
333		// largest cutoff for any atypes present in this group.
#	Line 302 | Line 387 | namespace OpenMD {
387		vector<RealType> groupCutoff(nGroups_, 0.0);
388		groupToGtype.resize(nGroups_);
389		for (int cg1 = 0; cg1 < nGroups_; cg1++) {
305	–
390		groupCutoff[cg1] = 0.0;
391		vector<int> atomList = getAtomsInGroupRow(cg1);
308	–
392		for (vector<int>::iterator ia = atomList.begin();
393		ia != atomList.end(); ++ia) {
394		int atom1 = (*ia);
395		atid = idents[atom1];
396	<	if (atypeCutoff[atid] > groupCutoff[cg1]) {
396	>	if (atypeCutoff[atid] > groupCutoff[cg1])
397		groupCutoff[cg1] = atypeCutoff[atid];
315	–	}
398		}
399	<
399	>
400		bool gTypeFound = false;
401	<	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
401	>	for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
402		if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
403		groupToGtype[cg1] = gt;
404		gTypeFound = true;
405		}
406		}
407	<	if (!gTypeFound) {
407	>	if (!gTypeFound) {
408		gTypeCutoffs.push_back( groupCutoff[cg1] );
409		groupToGtype[cg1] = gTypeCutoffs.size() - 1;
410		}
#	Line 331 | Line 413 | namespace OpenMD {
413
414		// Now we find the maximum group cutoff value present in the simulation
415
416	<	RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
416	>	RealType groupMax = *max_element(gTypeCutoffs.begin(),
417	>	gTypeCutoffs.end());
418
419		#ifdef IS_MPI
420	<	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
420	>	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
421	>	MPI::MAX);
422		#endif
423
424		RealType tradRcut = groupMax;
425
426	<	for (int i = 0; i < gTypeCutoffs.size();  i++) {
427	<	for (int j = 0; j < gTypeCutoffs.size();  j++) {
426	>	GrCut.resize( gTypeCutoffs.size() );
427	>	GrCutSq.resize( gTypeCutoffs.size() );
428	>	GrlistSq.resize( gTypeCutoffs.size() );
429	>
430	>
431	>	for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) {
432	>	GrCut[i].resize( gTypeCutoffs.size() , 0.0);
433	>	GrCutSq[i].resize( gTypeCutoffs.size(), 0.0 );
434	>	GrlistSq[i].resize( gTypeCutoffs.size(), 0.0 );
435	>
436	>	for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) {
437		RealType thisRcut;
438		switch(cutoffPolicy_) {
439		case TRADITIONAL:
#	Line 362 | Line 455 | namespace OpenMD {
455		break;
456		}
457
458	<	pair<int,int> key = make_pair(i,j);
366	<	gTypeCutoffMap[key].first = thisRcut;
367	<
458	>	GrCut[i][j] = thisRcut;
459		if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
460	+	GrCutSq[i][j] = thisRcut * thisRcut;
461	+	GrlistSq[i][j] = pow(thisRcut + skinThickness_, 2);
462
463	<	gTypeCutoffMap[key].second = thisRcut*thisRcut;
464	<
465	<	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
373	<
463	>	// pair<int,int> key = make_pair(i,j);
464	>	// gTypeCutoffMap[key].first = thisRcut;
465	>	// gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
466		// sanity check
467
468		if (userChoseCutoff_) {
469	<	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
469	>	if (abs(GrCut[i][j] - userCutoff_) > 0.0001) {
470		sprintf(painCave.errMsg,
471		"ForceMatrixDecomposition::createGtypeCutoffMap "
472		"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
#	Line 387 | Line 479 | namespace OpenMD {
479		}
480		}
481
482	<
391	<	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
482	>	void ForceMatrixDecomposition::getGroupCutoffs(int &cg1, int &cg2, RealType &rcut, RealType &rcutsq, RealType &rlistsq) {
483		int i, j;
484		#ifdef IS_MPI
485		i = groupRowToGtype[cg1];
#	Line 397 | Line 488 | namespace OpenMD {
488		i = groupToGtype[cg1];
489		j = groupToGtype[cg2];
490		#endif
491	<	return gTypeCutoffMap[make_pair(i,j)];
491	>	rcut = GrCut[i][j];
492	>	rcutsq = GrCutSq[i][j];
493	>	rlistsq = GrlistSq[i][j];
494	>	return;
495	>	//return gTypeCutoffMap[make_pair(i,j)];
496		}
497
498		int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
499	<	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
499	>	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
500		if (toposForAtom[atom1][j] == atom2)
501		return topoDist[atom1][j];
502	<	}
502	>	}
503		return 0;
504		}
505
506		void ForceMatrixDecomposition::zeroWorkArrays() {
507		pairwisePot = 0.0;
508		embeddingPot = 0.0;
509	+	excludedPot = 0.0;
510	+	excludedSelfPot = 0.0;
511
512		#ifdef IS_MPI
513		if (storageLayout_ & DataStorage::dslForce) {
#	Line 429 | Line 526 | namespace OpenMD {
526		fill(pot_col.begin(), pot_col.end(),
527		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
528
529	+	fill(expot_row.begin(), expot_row.end(),
530	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
531	+
532	+	fill(expot_col.begin(), expot_col.end(),
533	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
534	+
535		if (storageLayout_ & DataStorage::dslParticlePot) {
536	<	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
537	<	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
536	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
537	>	0.0);
538	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
539	>	0.0);
540		}
541
542		if (storageLayout_ & DataStorage::dslDensity) {
#	Line 440 | Line 545 | namespace OpenMD {
545		}
546
547		if (storageLayout_ & DataStorage::dslFunctional) {
548	<	fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
549	<	fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
548	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
549	>	0.0);
550	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
551	>	0.0);
552		}
553
554		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
#	Line 458 | Line 565 | namespace OpenMD {
565		atomColData.skippedCharge.end(), 0.0);
566		}
567
568	<	#else
569	<
568	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
569	>	fill(atomRowData.flucQFrc.begin(),
570	>	atomRowData.flucQFrc.end(), 0.0);
571	>	fill(atomColData.flucQFrc.begin(),
572	>	atomColData.flucQFrc.end(), 0.0);
573	>	}
574	>
575	>	if (storageLayout_ & DataStorage::dslElectricField) {
576	>	fill(atomRowData.electricField.begin(),
577	>	atomRowData.electricField.end(), V3Zero);
578	>	fill(atomColData.electricField.begin(),
579	>	atomColData.electricField.end(), V3Zero);
580	>	}
581	>
582	>	#endif
583	>	// even in parallel, we need to zero out the local arrays:
584	>
585		if (storageLayout_ & DataStorage::dslParticlePot) {
586		fill(snap_->atomData.particlePot.begin(),
587		snap_->atomData.particlePot.end(), 0.0);
#	Line 469 | Line 591 | namespace OpenMD {
591		fill(snap_->atomData.density.begin(),
592		snap_->atomData.density.end(), 0.0);
593		}
594	+
595		if (storageLayout_ & DataStorage::dslFunctional) {
596		fill(snap_->atomData.functional.begin(),
597		snap_->atomData.functional.end(), 0.0);
598		}
599	+
600		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
601		fill(snap_->atomData.functionalDerivative.begin(),
602		snap_->atomData.functionalDerivative.end(), 0.0);
603		}
604	+
605		if (storageLayout_ & DataStorage::dslSkippedCharge) {
606		fill(snap_->atomData.skippedCharge.begin(),
607		snap_->atomData.skippedCharge.end(), 0.0);
608		}
609	<	#endif
610	<
609	>
610	>	if (storageLayout_ & DataStorage::dslElectricField) {
611	>	fill(snap_->atomData.electricField.begin(),
612	>	snap_->atomData.electricField.end(), V3Zero);
613	>	}
614		}
615
616
#	Line 492 | Line 620 | namespace OpenMD {
620		#ifdef IS_MPI
621
622		// gather up the atomic positions
623	<	AtomCommVectorRow->gather(snap_->atomData.position,
623	>	AtomPlanVectorRow->gather(snap_->atomData.position,
624		atomRowData.position);
625	<	AtomCommVectorColumn->gather(snap_->atomData.position,
625	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
626		atomColData.position);
627
628		// gather up the cutoff group positions
629	<	cgCommVectorRow->gather(snap_->cgData.position,
629	>
630	>	cgPlanVectorRow->gather(snap_->cgData.position,
631		cgRowData.position);
632	<	cgCommVectorColumn->gather(snap_->cgData.position,
632	>
633	>	cgPlanVectorColumn->gather(snap_->cgData.position,
634		cgColData.position);
635	+
636	+
637	+
638	+	if (needVelocities_) {
639	+	// gather up the atomic velocities
640	+	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
641	+	atomColData.velocity);
642	+
643	+	cgPlanVectorColumn->gather(snap_->cgData.velocity,
644	+	cgColData.velocity);
645	+	}
646	+
647
648		// if needed, gather the atomic rotation matrices
649		if (storageLayout_ & DataStorage::dslAmat) {
650	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
650	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
651		atomRowData.aMat);
652	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
652	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
653		atomColData.aMat);
654		}
655	<
656	<	// if needed, gather the atomic eletrostatic frames
657	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
658	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
659	<	atomRowData.electroFrame);
660	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
661	<	atomColData.electroFrame);
655	>
656	>	// if needed, gather the atomic eletrostatic information
657	>	if (storageLayout_ & DataStorage::dslDipole) {
658	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
659	>	atomRowData.dipole);
660	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
661	>	atomColData.dipole);
662		}
663	+
664	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
665	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
666	+	atomRowData.quadrupole);
667	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
668	+	atomColData.quadrupole);
669	+	}
670	+
671	+	// if needed, gather the atomic fluctuating charge values
672	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
673	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
674	+	atomRowData.flucQPos);
675	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
676	+	atomColData.flucQPos);
677	+	}
678	+
679		#endif
680		}
681
#	Line 531 | Line 689 | namespace OpenMD {
689
690		if (storageLayout_ & DataStorage::dslDensity) {
691
692	<	AtomCommRealRow->scatter(atomRowData.density,
692	>	AtomPlanRealRow->scatter(atomRowData.density,
693		snap_->atomData.density);
694
695		int n = snap_->atomData.density.size();
696		vector<RealType> rho_tmp(n, 0.0);
697	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
697	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
698		for (int i = 0; i < n; i++)
699		snap_->atomData.density[i] += rho_tmp[i];
700		}
701	+
702	+	// this isn't necessary if we don't have polarizable atoms, but
703	+	// we'll leave it here for now.
704	+	if (storageLayout_ & DataStorage::dslElectricField) {
705	+
706	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
707	+	snap_->atomData.electricField);
708	+
709	+	int n = snap_->atomData.electricField.size();
710	+	vector<Vector3d> field_tmp(n, V3Zero);
711	+	AtomPlanVectorColumn->scatter(atomColData.electricField,
712	+	field_tmp);
713	+	for (int i = 0; i < n; i++)
714	+	snap_->atomData.electricField[i] += field_tmp[i];
715	+	}
716		#endif
717		}
718
#	Line 552 | Line 725 | namespace OpenMD {
725		storageLayout_ = sman_->getStorageLayout();
726		#ifdef IS_MPI
727		if (storageLayout_ & DataStorage::dslFunctional) {
728	<	AtomCommRealRow->gather(snap_->atomData.functional,
728	>	AtomPlanRealRow->gather(snap_->atomData.functional,
729		atomRowData.functional);
730	<	AtomCommRealColumn->gather(snap_->atomData.functional,
730	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
731		atomColData.functional);
732		}
733
734		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
735	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
735	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
736		atomRowData.functionalDerivative);
737	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
737	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
738		atomColData.functionalDerivative);
739		}
740		#endif
#	Line 575 | Line 748 | namespace OpenMD {
748		int n = snap_->atomData.force.size();
749		vector<Vector3d> frc_tmp(n, V3Zero);
750
751	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
751	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
752		for (int i = 0; i < n; i++) {
753		snap_->atomData.force[i] += frc_tmp[i];
754		frc_tmp[i] = 0.0;
755		}
756
757	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
758	<	for (int i = 0; i < n; i++)
757	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
758	>	for (int i = 0; i < n; i++) {
759		snap_->atomData.force[i] += frc_tmp[i];
760	<
761	<
760	>	}
761	>
762		if (storageLayout_ & DataStorage::dslTorque) {
763
764		int nt = snap_->atomData.torque.size();
765		vector<Vector3d> trq_tmp(nt, V3Zero);
766
767	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
767	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
768		for (int i = 0; i < nt; i++) {
769		snap_->atomData.torque[i] += trq_tmp[i];
770		trq_tmp[i] = 0.0;
771		}
772
773	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
773	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
774		for (int i = 0; i < nt; i++)
775		snap_->atomData.torque[i] += trq_tmp[i];
776		}
#	Line 607 | Line 780 | namespace OpenMD {
780		int ns = snap_->atomData.skippedCharge.size();
781		vector<RealType> skch_tmp(ns, 0.0);
782
783	<	AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
783	>	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
784		for (int i = 0; i < ns; i++) {
785	<	snap_->atomData.skippedCharge[i] = skch_tmp[i];
785	>	snap_->atomData.skippedCharge[i] += skch_tmp[i];
786		skch_tmp[i] = 0.0;
787		}
788
789	<	AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
790	<	for (int i = 0; i < ns; i++)
789	>	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
790	>	for (int i = 0; i < ns; i++)
791		snap_->atomData.skippedCharge[i] += skch_tmp[i];
792	+
793		}
794
795	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
796	+
797	+	int nq = snap_->atomData.flucQFrc.size();
798	+	vector<RealType> fqfrc_tmp(nq, 0.0);
799	+
800	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
801	+	for (int i = 0; i < nq; i++) {
802	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
803	+	fqfrc_tmp[i] = 0.0;
804	+	}
805	+
806	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
807	+	for (int i = 0; i < nq; i++)
808	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
809	+
810	+	}
811	+
812	+	if (storageLayout_ & DataStorage::dslElectricField) {
813	+
814	+	int nef = snap_->atomData.electricField.size();
815	+	vector<Vector3d> efield_tmp(nef, V3Zero);
816	+
817	+	AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
818	+	for (int i = 0; i < nef; i++) {
819	+	snap_->atomData.electricField[i] += efield_tmp[i];
820	+	efield_tmp[i] = 0.0;
821	+	}
822	+
823	+	AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
824	+	for (int i = 0; i < nef; i++)
825	+	snap_->atomData.electricField[i] += efield_tmp[i];
826	+	}
827	+
828	+
829		nLocal_ = snap_->getNumberOfAtoms();
830
831		vector<potVec> pot_temp(nLocal_,
832		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
833	+	vector<potVec> expot_temp(nLocal_,
834	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
835
836		// scatter/gather pot_row into the members of my column
837
838	<	AtomCommPotRow->scatter(pot_row, pot_temp);
838	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
839	>	AtomPlanPotRow->scatter(expot_row, expot_temp);
840
841	<	for (int ii = 0;  ii < pot_temp.size(); ii++ )
841	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
842		pairwisePot += pot_temp[ii];
843	<
843	>
844	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
845	>	excludedPot += expot_temp[ii];
846	>
847	>	if (storageLayout_ & DataStorage::dslParticlePot) {
848	>	// This is the pairwise contribution to the particle pot.  The
849	>	// embedding contribution is added in each of the low level
850	>	// non-bonded routines.  In single processor, this is done in
851	>	// unpackInteractionData, not in collectData.
852	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
853	>	for (int i = 0; i < nLocal_; i++) {
854	>	// factor of two is because the total potential terms are divided
855	>	// by 2 in parallel due to row/ column scatter
856	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
857	>	}
858	>	}
859	>	}
860	>
861		fill(pot_temp.begin(), pot_temp.end(),
862		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
863	+	fill(expot_temp.begin(), expot_temp.end(),
864	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
865
866	<	AtomCommPotColumn->scatter(pot_col, pot_temp);
866	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
867	>	AtomPlanPotColumn->scatter(expot_col, expot_temp);
868
869		for (int ii = 0;  ii < pot_temp.size(); ii++ )
870		pairwisePot += pot_temp[ii];
871	+
872	+	for (int ii = 0;  ii < expot_temp.size(); ii++ )
873	+	excludedPot += expot_temp[ii];
874	+
875	+	if (storageLayout_ & DataStorage::dslParticlePot) {
876	+	// This is the pairwise contribution to the particle pot.  The
877	+	// embedding contribution is added in each of the low level
878	+	// non-bonded routines.  In single processor, this is done in
879	+	// unpackInteractionData, not in collectData.
880	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
881	+	for (int i = 0; i < nLocal_; i++) {
882	+	// factor of two is because the total potential terms are divided
883	+	// by 2 in parallel due to row/ column scatter
884	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
885	+	}
886	+	}
887	+	}
888	+
889	+	if (storageLayout_ & DataStorage::dslParticlePot) {
890	+	int npp = snap_->atomData.particlePot.size();
891	+	vector<RealType> ppot_temp(npp, 0.0);
892	+
893	+	// This is the direct or embedding contribution to the particle
894	+	// pot.
895	+
896	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
897	+	for (int i = 0; i < npp; i++) {
898	+	snap_->atomData.particlePot[i] += ppot_temp[i];
899	+	}
900	+
901	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
902	+
903	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
904	+	for (int i = 0; i < npp; i++) {
905	+	snap_->atomData.particlePot[i] += ppot_temp[i];
906	+	}
907	+	}
908	+
909	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
910	+	RealType ploc1 = pairwisePot[ii];
911	+	RealType ploc2 = 0.0;
912	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
913	+	pairwisePot[ii] = ploc2;
914	+	}
915	+
916	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
917	+	RealType ploc1 = excludedPot[ii];
918	+	RealType ploc2 = 0.0;
919	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
920	+	excludedPot[ii] = ploc2;
921	+	}
922	+
923	+	// Here be dragons.
924	+	MPI::Intracomm col = colComm.getComm();
925	+
926	+	col.Allreduce(MPI::IN_PLACE,
927	+	&snap_->frameData.conductiveHeatFlux[0], 3,
928	+	MPI::REALTYPE, MPI::SUM);
929	+
930	+
931		#endif
932
933		}
934
935	<	int ForceMatrixDecomposition::getNAtomsInRow() {
935	>	/**
936	>	* Collects information obtained during the post-pair (and embedding
937	>	* functional) loops onto local data structures.
938	>	*/
939	>	void ForceMatrixDecomposition::collectSelfData() {
940	>	snap_ = sman_->getCurrentSnapshot();
941	>	storageLayout_ = sman_->getStorageLayout();
942	>
943		#ifdef IS_MPI
944	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
945	+	RealType ploc1 = embeddingPot[ii];
946	+	RealType ploc2 = 0.0;
947	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
948	+	embeddingPot[ii] = ploc2;
949	+	}
950	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
951	+	RealType ploc1 = excludedSelfPot[ii];
952	+	RealType ploc2 = 0.0;
953	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
954	+	excludedSelfPot[ii] = ploc2;
955	+	}
956	+	#endif
957	+
958	+	}
959	+
960	+
961	+
962	+	int& ForceMatrixDecomposition::getNAtomsInRow() {
963	+	#ifdef IS_MPI
964		return nAtomsInRow_;
965		#else
966		return nLocal_;
#	Line 652 | Line 970 | namespace OpenMD {
970		/**
971		* returns the list of atoms belonging to this group.
972		*/
973	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
973	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
974		#ifdef IS_MPI
975		return groupListRow_[cg1];
976		#else
#	Line 660 | Line 978 | namespace OpenMD {
978		#endif
979		}
980
981	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
981	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
982		#ifdef IS_MPI
983		return groupListCol_[cg2];
984		#else
#	Line 677 | Line 995 | namespace OpenMD {
995		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
996		#endif
997
998	<	snap_->wrapVector(d);
998	>	if (usePeriodicBoundaryConditions_) {
999	>	snap_->wrapVector(d);
1000	>	}
1001		return d;
1002		}
1003
1004	+	Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
1005	+	#ifdef IS_MPI
1006	+	return cgColData.velocity[cg2];
1007	+	#else
1008	+	return snap_->cgData.velocity[cg2];
1009	+	#endif
1010	+	}
1011
1012	+	Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
1013	+	#ifdef IS_MPI
1014	+	return atomColData.velocity[atom2];
1015	+	#else
1016	+	return snap_->atomData.velocity[atom2];
1017	+	#endif
1018	+	}
1019	+
1020	+
1021		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
1022
1023		Vector3d d;
#	Line 691 | Line 1027 | namespace OpenMD {
1027		#else
1028		d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
1029		#endif
1030	<
1031	<	snap_->wrapVector(d);
1030	>	if (usePeriodicBoundaryConditions_) {
1031	>	snap_->wrapVector(d);
1032	>	}
1033		return d;
1034		}
1035
#	Line 704 | Line 1041 | namespace OpenMD {
1041		#else
1042		d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1043		#endif
1044	<
1045	<	snap_->wrapVector(d);
1044	>	if (usePeriodicBoundaryConditions_) {
1045	>	snap_->wrapVector(d);
1046	>	}
1047		return d;
1048		}
1049
1050	<	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1050	>	RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1051		#ifdef IS_MPI
1052		return massFactorsRow[atom1];
1053		#else
#	Line 717 | Line 1055 | namespace OpenMD {
1055		#endif
1056		}
1057
1058	<	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1058	>	RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1059		#ifdef IS_MPI
1060		return massFactorsCol[atom2];
1061		#else
#	Line 734 | Line 1072 | namespace OpenMD {
1072		#else
1073		d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1074		#endif
1075	<
1076	<	snap_->wrapVector(d);
1075	>	if (usePeriodicBoundaryConditions_) {
1076	>	snap_->wrapVector(d);
1077	>	}
1078		return d;
1079		}
1080
1081	<	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1081	>	vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1082		return excludesForAtom[atom1];
1083		}
1084
#	Line 747 | Line 1086 | namespace OpenMD {
1086		* We need to exclude some overcounted interactions that result from
1087		* the parallel decomposition.
1088		*/
1089	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1089	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1090		int unique_id_1, unique_id_2;
1091	<
1091	>
1092		#ifdef IS_MPI
1093		// in MPI, we have to look up the unique IDs for each atom
1094		unique_id_1 = AtomRowToGlobal[atom1];
1095		unique_id_2 = AtomColToGlobal[atom2];
1096	+	// group1 = cgRowToGlobal[cg1];
1097	+	// group2 = cgColToGlobal[cg2];
1098	+	#else
1099	+	unique_id_1 = AtomLocalToGlobal[atom1];
1100	+	unique_id_2 = AtomLocalToGlobal[atom2];
1101	+	int group1 = cgLocalToGlobal[cg1];
1102	+	int group2 = cgLocalToGlobal[cg2];
1103	+	#endif
1104
758	–	// this situation should only arise in MPI simulations
1105		if (unique_id_1 == unique_id_2) return true;
1106	<
1106	>
1107	>	#ifdef IS_MPI
1108		// this prevents us from doing the pair on multiple processors
1109		if (unique_id_1 < unique_id_2) {
1110		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1111		} else {
1112	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1112	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1113		}
1114	+	#endif
1115	+
1116	+	#ifndef IS_MPI
1117	+	if (group1 == group2) {
1118	+	if (unique_id_1 < unique_id_2) return true;
1119	+	}
1120		#endif
1121	+
1122		return false;
1123		}
1124
#	Line 778 | Line 1132 | namespace OpenMD {
1132		* field) must still be handled for these pairs.
1133		*/
1134		bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1135	<	int unique_id_2;
1135	>
1136	>	// excludesForAtom was constructed to use row/column indices in the MPI
1137	>	// version, and to use local IDs in the non-MPI version:
1138
783	–	#ifdef IS_MPI
784	–	// in MPI, we have to look up the unique IDs for the row atom.
785	–	unique_id_2 = AtomColToGlobal[atom2];
786	–	#else
787	–	// in the normal loop, the atom numbers are unique
788	–	unique_id_2 = atom2;
789	–	#endif
790	–
1139		for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1140		i != excludesForAtom[atom1].end(); ++i) {
1141	<	if ( (*i) == unique_id_2 ) return true;
1141	>	if ( (*i) == atom2 ) return true;
1142		}
1143
1144		return false;
#	Line 820 | Line 1168 | namespace OpenMD {
1168		idat.excluded = excludeAtomPair(atom1, atom2);
1169
1170		#ifdef IS_MPI
1171	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1172	+	idat.atid1 = identsRow[atom1];
1173	+	idat.atid2 = identsCol[atom2];
1174	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1175	+	//                         ff_->getAtomType(identsCol[atom2]) );
1176
824	–	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
825	–	ff_->getAtomType(identsCol[atom2]) );
826	–
1177		if (storageLayout_ & DataStorage::dslAmat) {
1178		idat.A1 = &(atomRowData.aMat[atom1]);
1179		idat.A2 = &(atomColData.aMat[atom2]);
1180		}
1181
832	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
833	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
834	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
835	–	}
836	–
1182		if (storageLayout_ & DataStorage::dslTorque) {
1183		idat.t1 = &(atomRowData.torque[atom1]);
1184		idat.t2 = &(atomColData.torque[atom2]);
1185		}
1186
1187	+	if (storageLayout_ & DataStorage::dslDipole) {
1188	+	idat.dipole1 = &(atomRowData.dipole[atom1]);
1189	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1190	+	}
1191	+
1192	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1193	+	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1194	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1195	+	}
1196	+
1197		if (storageLayout_ & DataStorage::dslDensity) {
1198		idat.rho1 = &(atomRowData.density[atom1]);
1199		idat.rho2 = &(atomColData.density[atom2]);
#	Line 864 | Line 1219 | namespace OpenMD {
1219		idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1220		}
1221
1222	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1223	+	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1224	+	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1225	+	}
1226	+
1227		#else
1228	+
1229	+	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1230	+	idat.atid1 = idents[atom1];
1231	+	idat.atid2 = idents[atom2];
1232
869	–	idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
870	–	ff_->getAtomType(idents[atom2]) );
871	–
1233		if (storageLayout_ & DataStorage::dslAmat) {
1234		idat.A1 = &(snap_->atomData.aMat[atom1]);
1235		idat.A2 = &(snap_->atomData.aMat[atom2]);
1236		}
1237
877	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
878	–	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
879	–	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
880	–	}
881	–
1238		if (storageLayout_ & DataStorage::dslTorque) {
1239		idat.t1 = &(snap_->atomData.torque[atom1]);
1240		idat.t2 = &(snap_->atomData.torque[atom2]);
1241		}
1242
1243	+	if (storageLayout_ & DataStorage::dslDipole) {
1244	+	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1245	+	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1246	+	}
1247	+
1248	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1249	+	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1250	+	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1251	+	}
1252	+
1253		if (storageLayout_ & DataStorage::dslDensity) {
1254		idat.rho1 = &(snap_->atomData.density[atom1]);
1255		idat.rho2 = &(snap_->atomData.density[atom2]);
#	Line 908 | Line 1274 | namespace OpenMD {
1274		idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1275		idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1276		}
1277	+
1278	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1279	+	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1280	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1281	+	}
1282	+
1283		#endif
1284		}
1285
1286
1287		void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1288		#ifdef IS_MPI
1289	<	pot_row[atom1] += 0.5 *  *(idat.pot);
1290	<	pot_col[atom2] += 0.5 *  *(idat.pot);
1289	>	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1290	>	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1291	>	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1292	>	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1293
1294		atomRowData.force[atom1] += *(idat.f1);
1295		atomColData.force[atom2] -= *(idat.f1);
1296	+
1297	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1298	+	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1299	+	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1300	+	}
1301	+
1302	+	if (storageLayout_ & DataStorage::dslElectricField) {
1303	+	atomRowData.electricField[atom1] += *(idat.eField1);
1304	+	atomColData.electricField[atom2] += *(idat.eField2);
1305	+	}
1306	+
1307		#else
1308		pairwisePot += *(idat.pot);
1309	+	excludedPot += *(idat.excludedPot);
1310
1311		snap_->atomData.force[atom1] += *(idat.f1);
1312		snap_->atomData.force[atom2] -= *(idat.f1);
1313	+
1314	+	if (idat.doParticlePot) {
1315	+	// This is the pairwise contribution to the particle pot.  The
1316	+	// embedding contribution is added in each of the low level
1317	+	// non-bonded routines.  In parallel, this calculation is done
1318	+	// in collectData, not in unpackInteractionData.
1319	+	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1320	+	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1321	+	}
1322	+
1323	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1324	+	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1325	+	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1326	+	}
1327	+
1328	+	if (storageLayout_ & DataStorage::dslElectricField) {
1329	+	snap_->atomData.electricField[atom1] += *(idat.eField1);
1330	+	snap_->atomData.electricField[atom2] += *(idat.eField2);
1331	+	}
1332	+
1333		#endif
1334
1335		}
#	Line 934 | Line 1340 | namespace OpenMD {
1340		* first element of pair is row-indexed CutoffGroup
1341		* second element of pair is column-indexed CutoffGroup
1342		*/
1343	<	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1344	<
1345	<	vector<pair<int, int> > neighborList;
1343	>	void ForceMatrixDecomposition::buildNeighborList(vector<pair<int,int> >& neighborList) {
1344	>
1345	>	neighborList.clear();
1346		groupCutoffs cuts;
1347		bool doAllPairs = false;
1348
1349	+	RealType rList_ = (largestRcut_ + skinThickness_);
1350	+	RealType rcut, rcutsq, rlistsq;
1351	+	Snapshot* snap_ = sman_->getCurrentSnapshot();
1352	+	Mat3x3d box;
1353	+	Mat3x3d invBox;
1354	+
1355	+	Vector3d rs, scaled, dr;
1356	+	Vector3i whichCell;
1357	+	int cellIndex;
1358	+
1359		#ifdef IS_MPI
1360		cellListRow_.clear();
1361		cellListCol_.clear();
1362		#else
1363		cellList_.clear();
1364		#endif
1365	<
1366	<	RealType rList_ = (largestRcut_ + skinThickness_);
1367	<	RealType rl2 = rList_ * rList_;
1368	<	Snapshot* snap_ = sman_->getCurrentSnapshot();
1369	<	Mat3x3d Hmat = snap_->getHmat();
1370	<	Vector3d Hx = Hmat.getColumn(0);
1371	<	Vector3d Hy = Hmat.getColumn(1);
1372	<	Vector3d Hz = Hmat.getColumn(2);
1373	<
1374	<	nCells_.x() = (int) ( Hx.length() )/ rList_;
1375	<	nCells_.y() = (int) ( Hy.length() )/ rList_;
1376	<	nCells_.z() = (int) ( Hz.length() )/ rList_;
1377	<
1365	>
1366	>	if (!usePeriodicBoundaryConditions_) {
1367	>	box = snap_->getBoundingBox();
1368	>	invBox = snap_->getInvBoundingBox();
1369	>	} else {
1370	>	box = snap_->getHmat();
1371	>	invBox = snap_->getInvHmat();
1372	>	}
1373	>
1374	>	Vector3d boxX = box.getColumn(0);
1375	>	Vector3d boxY = box.getColumn(1);
1376	>	Vector3d boxZ = box.getColumn(2);
1377	>
1378	>	nCells_.x() = (int) ( boxX.length() )/ rList_;
1379	>	nCells_.y() = (int) ( boxY.length() )/ rList_;
1380	>	nCells_.z() = (int) ( boxZ.length() )/ rList_;
1381	>
1382		// handle small boxes where the cell offsets can end up repeating cells
1383
1384		if (nCells_.x() < 3) doAllPairs = true;
1385		if (nCells_.y() < 3) doAllPairs = true;
1386		if (nCells_.z() < 3) doAllPairs = true;
1387	<
968	<	Mat3x3d invHmat = snap_->getInvHmat();
969	<	Vector3d rs, scaled, dr;
970	<	Vector3i whichCell;
971	<	int cellIndex;
1387	>
1388		int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1389	<
1389	>
1390		#ifdef IS_MPI
1391		cellListRow_.resize(nCtot);
1392		cellListCol_.resize(nCtot);
1393		#else
1394		cellList_.resize(nCtot);
1395		#endif
1396	<
1396	>
1397		if (!doAllPairs) {
1398		#ifdef IS_MPI
1399	<
1399	>
1400		for (int i = 0; i < nGroupsInRow_; i++) {
1401		rs = cgRowData.position[i];
1402
1403		// scaled positions relative to the box vectors
1404	<	scaled = invHmat * rs;
1404	>	scaled = invBox * rs;
1405
1406		// wrap the vector back into the unit box by subtracting integer box
1407		// numbers
1408		for (int j = 0; j < 3; j++) {
1409		scaled[j] -= roundMe(scaled[j]);
1410		scaled[j] += 0.5;
1411	+	// Handle the special case when an object is exactly on the
1412	+	// boundary (a scaled coordinate of 1.0 is the same as
1413	+	// scaled coordinate of 0.0)
1414	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1415		}
1416
1417		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1005 | Line 1425 | namespace OpenMD {
1425		// add this cutoff group to the list of groups in this cell;
1426		cellListRow_[cellIndex].push_back(i);
1427		}
1008	–
1428		for (int i = 0; i < nGroupsInCol_; i++) {
1429		rs = cgColData.position[i];
1430
1431		// scaled positions relative to the box vectors
1432	<	scaled = invHmat * rs;
1432	>	scaled = invBox * rs;
1433
1434		// wrap the vector back into the unit box by subtracting integer box
1435		// numbers
1436		for (int j = 0; j < 3; j++) {
1437		scaled[j] -= roundMe(scaled[j]);
1438		scaled[j] += 0.5;
1439	+	// Handle the special case when an object is exactly on the
1440	+	// boundary (a scaled coordinate of 1.0 is the same as
1441	+	// scaled coordinate of 0.0)
1442	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1443		}
1444
1445		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1030 | Line 1453 | namespace OpenMD {
1453		// add this cutoff group to the list of groups in this cell;
1454		cellListCol_[cellIndex].push_back(i);
1455		}
1456	+
1457		#else
1458		for (int i = 0; i < nGroups_; i++) {
1459		rs = snap_->cgData.position[i];
1460
1461		// scaled positions relative to the box vectors
1462	<	scaled = invHmat * rs;
1462	>	scaled = invBox * rs;
1463
1464		// wrap the vector back into the unit box by subtracting integer box
1465		// numbers
1466		for (int j = 0; j < 3; j++) {
1467		scaled[j] -= roundMe(scaled[j]);
1468		scaled[j] += 0.5;
1469	+	// Handle the special case when an object is exactly on the
1470	+	// boundary (a scaled coordinate of 1.0 is the same as
1471	+	// scaled coordinate of 0.0)
1472	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1473		}
1474
1475		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1050 | Line 1478 | namespace OpenMD {
1478		whichCell.z() = nCells_.z() * scaled.z();
1479
1480		// find single index of this cell:
1481	<	cellIndex = Vlinear(whichCell, nCells_);
1481	>	cellIndex = Vlinear(whichCell, nCells_);
1482
1483		// add this cutoff group to the list of groups in this cell;
1484		cellList_[cellIndex].push_back(i);
1485		}
1486	+
1487		#endif
1488
1489		for (int m1z = 0; m1z < nCells_.z(); m1z++) {
#	Line 1067 | Line 1496 | namespace OpenMD {
1496		os != cellOffsets_.end(); ++os) {
1497
1498		Vector3i m2v = m1v + (*os);
1499	<
1499	>
1500	>
1501		if (m2v.x() >= nCells_.x()) {
1502		m2v.x() = 0;
1503		} else if (m2v.x() < 0) {
#	Line 1085 | Line 1515 | namespace OpenMD {
1515		} else if (m2v.z() < 0) {
1516		m2v.z() = nCells_.z() - 1;
1517		}
1518	<
1518	>
1519		int m2 = Vlinear (m2v, nCells_);
1520
1521		#ifdef IS_MPI
#	Line 1094 | Line 1524 | namespace OpenMD {
1524		for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1525		j2 != cellListCol_[m2].end(); ++j2) {
1526
1527	<	// Always do this if we're in different cells or if
1528	<	// we're in the same cell and the global index of the
1529	<	// j2 cutoff group is less than the j1 cutoff group
1530	<
1531	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1102	<	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1527	>	// In parallel, we need to visit *all* pairs of row
1528	>	// & column indicies and will divide labor in the
1529	>	// force evaluation later.
1530	>	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1531	>	if (usePeriodicBoundaryConditions_) {
1532		snap_->wrapVector(dr);
1104	–	cuts = getGroupCutoffs( (*j1), (*j2) );
1105	–	if (dr.lengthSquare() < cuts.third) {
1106	–	neighborList.push_back(make_pair((*j1), (*j2)));
1107	–	}
1533		}
1534	+	getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq );
1535	+	if (dr.lengthSquare() < rlistsq) {
1536	+	neighborList.push_back(make_pair((*j1), (*j2)));
1537	+	}
1538		}
1539		}
1540		#else
1112	–
1541		for (vector<int>::iterator j1 = cellList_[m1].begin();
1542		j1 != cellList_[m1].end(); ++j1) {
1543		for (vector<int>::iterator j2 = cellList_[m2].begin();
1544		j2 != cellList_[m2].end(); ++j2) {
1545	<
1545	>
1546		// Always do this if we're in different cells or if
1547	<	// we're in the same cell and the global index of the
1548	<	// j2 cutoff group is less than the j1 cutoff group
1549	<
1550	<	if (m2 != m1 || (*j2) < (*j1)) {
1547	>	// we're in the same cell and the global index of
1548	>	// the j2 cutoff group is greater than or equal to
1549	>	// the j1 cutoff group.  Note that Rappaport's code
1550	>	// has a "less than" conditional here, but that
1551	>	// deals with atom-by-atom computation.  OpenMD
1552	>	// allows atoms within a single cutoff group to
1553	>	// interact with each other.
1554	>
1555	>	if (m2 != m1 || (*j2) >= (*j1) ) {
1556	>
1557		dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1558	<	snap_->wrapVector(dr);
1559	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1560	<	if (dr.lengthSquare() < cuts.third) {
1558	>	if (usePeriodicBoundaryConditions_) {
1559	>	snap_->wrapVector(dr);
1560	>	}
1561	>	getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq );
1562	>	if (dr.lengthSquare() < rlistsq) {
1563		neighborList.push_back(make_pair((*j1), (*j2)));
1564		}
1565		}
#	Line 1138 | Line 1574 | namespace OpenMD {
1574		// branch to do all cutoff group pairs
1575		#ifdef IS_MPI
1576		for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1577	<	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1577	>	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1578		dr = cgColData.position[j2] - cgRowData.position[j1];
1579	<	snap_->wrapVector(dr);
1580	<	cuts = getGroupCutoffs( j1, j2 );
1581	<	if (dr.lengthSquare() < cuts.third) {
1579	>	if (usePeriodicBoundaryConditions_) {
1580	>	snap_->wrapVector(dr);
1581	>	}
1582	>	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq);
1583	>	if (dr.lengthSquare() < rlistsq) {
1584		neighborList.push_back(make_pair(j1, j2));
1585		}
1586		}
1587	<	}
1587	>	}
1588		#else
1589	<	for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1590	<	for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1589	>	// include all groups here.
1590	>	for (int j1 = 0; j1 < nGroups_; j1++) {
1591	>	// include self group interactions j2 == j1
1592	>	for (int j2 = j1; j2 < nGroups_; j2++) {
1593		dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1594	<	snap_->wrapVector(dr);
1595	<	cuts = getGroupCutoffs( j1, j2 );
1596	<	if (dr.lengthSquare() < cuts.third) {
1594	>	if (usePeriodicBoundaryConditions_) {
1595	>	snap_->wrapVector(dr);
1596	>	}
1597	>	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq );
1598	>	if (dr.lengthSquare() < rlistsq) {
1599		neighborList.push_back(make_pair(j1, j2));
1600		}
1601	<	}
1602	<	}
1601	>	}
1602	>	}
1603		#endif
1604		}
1605
#	Line 1166 | Line 1608 | namespace OpenMD {
1608		saved_CG_positions_.clear();
1609		for (int i = 0; i < nGroups_; i++)
1610		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1169	–
1170	–	return neighborList;
1611		}
1612		} //end namespace OpenMD

Diff Legend

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