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

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1593 by gezelter, Fri Jul 15 21:35:14 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1993 by gezelter, Tue Apr 29 17:32:31 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 53 \| Line 54 \| namespace OpenMD {
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_.push_back( Vector3i(-1, 0, 0) );
57	<	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
58	<	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
59	<	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
60	<	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
61	<	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
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) );
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) );
66	–	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67	–	cellOffsets_.push_back( Vector3i( 0, 1,-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
#	Line 79 \| Line 95 \| namespace OpenMD {
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 93 \| 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	<	MPI::Intracomm row = rowComm.getComm();
123	<	MPI::Intracomm col = colComm.getComm();
122	>	MPI_Comm row = rowComm.getComm();
123	>	MPI_Comm col = colComm.getComm();
124
125		AtomPlanIntRow = new Plan<int>(row, nLocal_);
126		AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
#	Line 129 \| 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		AtomPlanIntRow->gather(idents, identsRow);
166		AtomPlanIntColumn->gather(idents, identsCol);
167	+
168	+	regionsRow.resize(nAtomsInRow_);
169	+	regionsCol.resize(nAtomsInCol_);
170
171	+	AtomPlanIntRow->gather(regions, regionsRow);
172	+	AtomPlanIntColumn->gather(regions, regionsCol);
173	+
174		// allocate memory for the parallel objects
175		atypesRow.resize(nAtomsInRow_);
176		atypesCol.resize(nAtomsInCol_);
#	Line 149 \| Line 183 \| namespace OpenMD {
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
157	–	cerr << "Atoms in Local:\n";
158	–	for (int i = 0; i < AtomLocalToGlobal.size(); i++) {
159	–	cerr << "i =\t" << i << "\t localAt =\t" << AtomLocalToGlobal[i] << "\n";
160	–	}
161	–	cerr << "Atoms in Row:\n";
162	–	for (int i = 0; i < AtomRowToGlobal.size(); i++) {
163	–	cerr << "i =\t" << i << "\t rowAt =\t" << AtomRowToGlobal[i] << "\n";
164	–	}
165	–	cerr << "Atoms in Col:\n";
166	–	for (int i = 0; i < AtomColToGlobal.size(); i++) {
167	–	cerr << "i =\t" << i << "\t colAt =\t" << AtomColToGlobal[i] << "\n";
168	–	}
169	–
194		cgRowToGlobal.resize(nGroupsInRow_);
195		cgColToGlobal.resize(nGroupsInCol_);
196		cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
197		cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
198
175	–	cerr << "Gruops in Local:\n";
176	–	for (int i = 0; i < cgLocalToGlobal.size(); i++) {
177	–	cerr << "i =\t" << i << "\t localCG =\t" << cgLocalToGlobal[i] << "\n";
178	–	}
179	–	cerr << "Groups in Row:\n";
180	–	for (int i = 0; i < cgRowToGlobal.size(); i++) {
181	–	cerr << "i =\t" << i << "\t rowCG =\t" << cgRowToGlobal[i] << "\n";
182	–	}
183	–	cerr << "Groups in Col:\n";
184	–	for (int i = 0; i < cgColToGlobal.size(); i++) {
185	–	cerr << "i =\t" << i << "\t colCG =\t" << cgColToGlobal[i] << "\n";
186	–	}
187	–
188	–
199		massFactorsRow.resize(nAtomsInRow_);
200		massFactorsCol.resize(nAtomsInCol_);
201		AtomPlanRealRow->gather(massFactors, massFactorsRow);
#	Line 245 \| Line 255 \| namespace OpenMD {
255		}
256		}
257
258	<	#endif
249	<
250	<	// allocate memory for the parallel objects
251	<	atypesLocal.resize(nLocal_);
252	<
253	<	for (int i = 0; i < nLocal_; i++)
254	<	atypesLocal[i] = ff_->getAtomType(idents[i]);
255	<
256	<	groupList_.clear();
257	<	groupList_.resize(nGroups_);
258	<	for (int i = 0; i < nGroups_; i++) {
259	<	int gid = cgLocalToGlobal[i];
260	<	for (int j = 0; j < nLocal_; j++) {
261	<	int aid = AtomLocalToGlobal[j];
262	<	if (globalGroupMembership[aid] == gid) {
263	<	groupList_[i].push_back(j);
264	<	}
265	<	}
266	<	}
267	<
258	>	#else
259		excludesForAtom.clear();
260		excludesForAtom.resize(nLocal_);
261		toposForAtom.clear();
#	Line 297 \| 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		largestRcut_ = 0.0;
309	–	RealType rc;
324		int atid;
325		set<AtomType*> atypes = info_->getSimulatedAtomTypes();
326
#	Line 391 \| Line 405 \| namespace OpenMD {
405		}
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;
#	Line 410 \| Line 424 \| namespace OpenMD {
424		gTypeCutoffs.end());
425
426		#ifdef IS_MPI
427	<	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
428	<	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 439 \| Line 462 \| namespace OpenMD {
462		break;
463		}
464
465	<	pair<int,int> key = make_pair(i,j);
443	<	gTypeCutoffMap[key].first = thisRcut;
465	>	GrCut[i][j] = thisRcut;
466		if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
467	<	gTypeCutoffMap[key].second = thisRcut*thisRcut;
468	<	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
467	>	GrCutSq[i][j] = thisRcut * thisRcut;
468	>	GrlistSq[i][j] = pow(thisRcut + skinThickness_, 2);
469	>
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 460 \| Line 486 \| namespace OpenMD {
486		}
487		}
488
489	<
464	<	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 470 \| 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 502 \| 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(),
544		0.0);
#	Line 533 \| Line 570 \| namespace OpenMD {
570		atomRowData.skippedCharge.end(), 0.0);
571		fill(atomColData.skippedCharge.begin(),
572		atomColData.skippedCharge.end(), 0.0);
573	+	}
574	+
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	+	if (storageLayout_ & DataStorage::dslSitePotential) {
590	+	fill(atomRowData.sitePotential.begin(),
591	+	atomRowData.sitePotential.end(), 0.0);
592	+	fill(atomColData.sitePotential.begin(),
593	+	atomColData.sitePotential.end(), 0.0);
594		}
595
596		#endif
#	Line 547 \| Line 605 \| namespace OpenMD {
605		fill(snap_->atomData.density.begin(),
606		snap_->atomData.density.end(), 0.0);
607		}
608	+
609		if (storageLayout_ & DataStorage::dslFunctional) {
610		fill(snap_->atomData.functional.begin(),
611		snap_->atomData.functional.end(), 0.0);
612		}
613	+
614		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
615		fill(snap_->atomData.functionalDerivative.begin(),
616		snap_->atomData.functionalDerivative.end(), 0.0);
617		}
618	+
619		if (storageLayout_ & DataStorage::dslSkippedCharge) {
620		fill(snap_->atomData.skippedCharge.begin(),
621		snap_->atomData.skippedCharge.end(), 0.0);
622		}
623	<
623	>
624	>	if (storageLayout_ & DataStorage::dslElectricField) {
625	>	fill(snap_->atomData.electricField.begin(),
626	>	snap_->atomData.electricField.end(), V3Zero);
627	>	}
628	>	if (storageLayout_ & DataStorage::dslSitePotential) {
629	>	fill(snap_->atomData.sitePotential.begin(),
630	>	snap_->atomData.sitePotential.end(), 0.0);
631	>	}
632		}
633
634
#	Line 576 \| Line 645 \| namespace OpenMD {
645
646		// gather up the cutoff group positions
647
579	–	cerr << "before gather\n";
580	–	for (int i = 0; i < snap_->cgData.position.size(); i++) {
581	–	cerr << "cgpos = " << snap_->cgData.position[i] << "\n";
582	–	}
583	–
648		cgPlanVectorRow->gather(snap_->cgData.position,
649		cgRowData.position);
650
587	–	cerr << "after gather\n";
588	–	for (int i = 0; i < cgRowData.position.size(); i++) {
589	–	cerr << "cgRpos = " << cgRowData.position[i] << "\n";
590	–	}
591	–
651		cgPlanVectorColumn->gather(snap_->cgData.position,
652		cgColData.position);
653	<	for (int i = 0; i < cgColData.position.size(); i++) {
654	<	cerr << "cgCpos = " << cgColData.position[i] << "\n";
653	>
654	>
655	>
656	>	if (needVelocities_) {
657	>	// gather up the atomic velocities
658	>	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
659	>	atomColData.velocity);
660	>
661	>	cgPlanVectorColumn->gather(snap_->cgData.velocity,
662	>	cgColData.velocity);
663		}
664
665
#	Line 603 \| Line 670 \| namespace OpenMD {
670		AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
671		atomColData.aMat);
672		}
673	<
674	<	// if needed, gather the atomic eletrostatic frames
675	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
676	<	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
677	<	atomRowData.electroFrame);
678	<	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
679	<	atomColData.electroFrame);
673	>
674	>	// if needed, gather the atomic eletrostatic information
675	>	if (storageLayout_ & DataStorage::dslDipole) {
676	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
677	>	atomRowData.dipole);
678	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
679	>	atomColData.dipole);
680		}
681
682	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
683	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
684	+	atomRowData.quadrupole);
685	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
686	+	atomColData.quadrupole);
687	+	}
688	+
689	+	// if needed, gather the atomic fluctuating charge values
690	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
691	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
692	+	atomRowData.flucQPos);
693	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
694	+	atomColData.flucQPos);
695	+	}
696	+
697		#endif
698		}
699
#	Line 634 \| Line 716 \| namespace OpenMD {
716		for (int i = 0; i < n; i++)
717		snap_->atomData.density[i] += rho_tmp[i];
718		}
719	+
720	+	// this isn't necessary if we don't have polarizable atoms, but
721	+	// we'll leave it here for now.
722	+	if (storageLayout_ & DataStorage::dslElectricField) {
723	+
724	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
725	+	snap_->atomData.electricField);
726	+
727	+	int n = snap_->atomData.electricField.size();
728	+	vector<Vector3d> field_tmp(n, V3Zero);
729	+	AtomPlanVectorColumn->scatter(atomColData.electricField,
730	+	field_tmp);
731	+	for (int i = 0; i < n; i++)
732	+	snap_->atomData.electricField[i] += field_tmp[i];
733	+	}
734		#endif
735		}
736
#	Line 708 \| Line 805 \| namespace OpenMD {
805		}
806
807		AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
808	<	for (int i = 0; i < ns; i++)
808	>	for (int i = 0; i < ns; i++)
809		snap_->atomData.skippedCharge[i] += skch_tmp[i];
810	+
811		}
812
813	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
814	+
815	+	int nq = snap_->atomData.flucQFrc.size();
816	+	vector<RealType> fqfrc_tmp(nq, 0.0);
817	+
818	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
819	+	for (int i = 0; i < nq; i++) {
820	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
821	+	fqfrc_tmp[i] = 0.0;
822	+	}
823	+
824	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
825	+	for (int i = 0; i < nq; i++)
826	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
827	+
828	+	}
829	+
830	+	if (storageLayout_ & DataStorage::dslElectricField) {
831	+
832	+	int nef = snap_->atomData.electricField.size();
833	+	vector<Vector3d> efield_tmp(nef, V3Zero);
834	+
835	+	AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
836	+	for (int i = 0; i < nef; i++) {
837	+	snap_->atomData.electricField[i] += efield_tmp[i];
838	+	efield_tmp[i] = 0.0;
839	+	}
840	+
841	+	AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
842	+	for (int i = 0; i < nef; i++)
843	+	snap_->atomData.electricField[i] += efield_tmp[i];
844	+	}
845	+
846	+	if (storageLayout_ & DataStorage::dslSitePotential) {
847	+
848	+	int nsp = snap_->atomData.sitePotential.size();
849	+	vector<RealType> sp_tmp(nsp, 0.0);
850	+
851	+	AtomPlanRealRow->scatter(atomRowData.sitePotential, sp_tmp);
852	+	for (int i = 0; i < nsp; i++) {
853	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
854	+	sp_tmp[i] = 0.0;
855	+	}
856	+
857	+	AtomPlanRealColumn->scatter(atomColData.sitePotential, sp_tmp);
858	+	for (int i = 0; i < nsp; i++)
859	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
860	+	}
861	+
862		nLocal_ = snap_->getNumberOfAtoms();
863
864		vector<potVec> pot_temp(nLocal_,
865		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
866	+	vector<potVec> expot_temp(nLocal_,
867	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
868
869		// scatter/gather pot_row into the members of my column
870
871		AtomPlanPotRow->scatter(pot_row, pot_temp);
872	+	AtomPlanPotRow->scatter(expot_row, expot_temp);
873
874	<	for (int ii = 0; ii < pot_temp.size(); ii++ )
874	>	for (int ii = 0; ii < pot_temp.size(); ii++ )
875		pairwisePot += pot_temp[ii];
876	<
876	>
877	>	for (int ii = 0; ii < expot_temp.size(); ii++ )
878	>	excludedPot += expot_temp[ii];
879	>
880	>	if (storageLayout_ & DataStorage::dslParticlePot) {
881	>	// This is the pairwise contribution to the particle pot. The
882	>	// embedding contribution is added in each of the low level
883	>	// non-bonded routines. In single processor, this is done in
884	>	// unpackInteractionData, not in collectData.
885	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
886	>	for (int i = 0; i < nLocal_; i++) {
887	>	// factor of two is because the total potential terms are divided
888	>	// by 2 in parallel due to row/ column scatter
889	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
890	>	}
891	>	}
892	>	}
893	>
894		fill(pot_temp.begin(), pot_temp.end(),
895		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
896	+	fill(expot_temp.begin(), expot_temp.end(),
897	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
898
899		AtomPlanPotColumn->scatter(pot_col, pot_temp);
900	+	AtomPlanPotColumn->scatter(expot_col, expot_temp);
901
902		for (int ii = 0; ii < pot_temp.size(); ii++ )
903		pairwisePot += pot_temp[ii];
904	+
905	+	for (int ii = 0; ii < expot_temp.size(); ii++ )
906	+	excludedPot += expot_temp[ii];
907	+
908	+	if (storageLayout_ & DataStorage::dslParticlePot) {
909	+	// This is the pairwise contribution to the particle pot. The
910	+	// embedding contribution is added in each of the low level
911	+	// non-bonded routines. In single processor, this is done in
912	+	// unpackInteractionData, not in collectData.
913	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
914	+	for (int i = 0; i < nLocal_; i++) {
915	+	// factor of two is because the total potential terms are divided
916	+	// by 2 in parallel due to row/ column scatter
917	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
918	+	}
919	+	}
920	+	}
921	+
922	+	if (storageLayout_ & DataStorage::dslParticlePot) {
923	+	int npp = snap_->atomData.particlePot.size();
924	+	vector<RealType> ppot_temp(npp, 0.0);
925	+
926	+	// This is the direct or embedding contribution to the particle
927	+	// pot.
928	+
929	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
930	+	for (int i = 0; i < npp; i++) {
931	+	snap_->atomData.particlePot[i] += ppot_temp[i];
932	+	}
933	+
934	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
935	+
936	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
937	+	for (int i = 0; i < npp; i++) {
938	+	snap_->atomData.particlePot[i] += ppot_temp[i];
939	+	}
940	+	}
941	+
942	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
943	+	RealType ploc1 = pairwisePot[ii];
944	+	RealType ploc2 = 0.0;
945	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
946	+	pairwisePot[ii] = ploc2;
947	+	}
948	+
949	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
950	+	RealType ploc1 = excludedPot[ii];
951	+	RealType ploc2 = 0.0;
952	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
953	+	excludedPot[ii] = ploc2;
954	+	}
955	+
956	+	// Here be dragons.
957	+	MPI_Comm col = colComm.getComm();
958	+
959	+	MPI_Allreduce(MPI_IN_PLACE,
960	+	&snap_->frameData.conductiveHeatFlux[0], 3,
961	+	MPI_REALTYPE, MPI_SUM, col);
962	+
963	+
964		#endif
965
736	–	cerr << "pairwisePot = " << pairwisePot << "\n";
966		}
967
968	<	int ForceMatrixDecomposition::getNAtomsInRow() {
968	>	/**
969	>	* Collects information obtained during the post-pair (and embedding
970	>	* functional) loops onto local data structures.
971	>	*/
972	>	void ForceMatrixDecomposition::collectSelfData() {
973	>	snap_ = sman_->getCurrentSnapshot();
974	>	storageLayout_ = sman_->getStorageLayout();
975	>
976		#ifdef IS_MPI
977	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
978	+	RealType ploc1 = embeddingPot[ii];
979	+	RealType ploc2 = 0.0;
980	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
981	+	embeddingPot[ii] = ploc2;
982	+	}
983	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
984	+	RealType ploc1 = excludedSelfPot[ii];
985	+	RealType ploc2 = 0.0;
986	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
987	+	excludedSelfPot[ii] = ploc2;
988	+	}
989	+	#endif
990	+
991	+	}
992	+
993	+
994	+
995	+	int& ForceMatrixDecomposition::getNAtomsInRow() {
996	+	#ifdef IS_MPI
997		return nAtomsInRow_;
998		#else
999		return nLocal_;
#	Line 747 \| Line 1003 \| namespace OpenMD {
1003		/**
1004		* returns the list of atoms belonging to this group.
1005		*/
1006	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
1006	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
1007		#ifdef IS_MPI
1008		return groupListRow_[cg1];
1009		#else
#	Line 755 \| Line 1011 \| namespace OpenMD {
1011		#endif
1012		}
1013
1014	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
1014	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
1015		#ifdef IS_MPI
1016		return groupListCol_[cg2];
1017		#else
#	Line 768 \| Line 1024 \| namespace OpenMD {
1024
1025		#ifdef IS_MPI
1026		d = cgColData.position[cg2] - cgRowData.position[cg1];
771	–	cerr << "cg1 = " << cg1 << "\tcg1p = " << cgRowData.position[cg1] << "\n";
772	–	cerr << "cg2 = " << cg2 << "\tcg2p = " << cgColData.position[cg2] << "\n";
1027		#else
1028		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
775	–	cerr << "cg1 = " << cg1 << "\tcg1p = " << snap_->cgData.position[cg1] << "\n";
776	–	cerr << "cg2 = " << cg2 << "\tcg2p = " << snap_->cgData.position[cg2] << "\n";
1029		#endif
1030
1031	<	snap_->wrapVector(d);
1031	>	if (usePeriodicBoundaryConditions_) {
1032	>	snap_->wrapVector(d);
1033	>	}
1034		return d;
1035		}
1036
1037	+	Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
1038	+	#ifdef IS_MPI
1039	+	return cgColData.velocity[cg2];
1040	+	#else
1041	+	return snap_->cgData.velocity[cg2];
1042	+	#endif
1043	+	}
1044
1045	+	Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
1046	+	#ifdef IS_MPI
1047	+	return atomColData.velocity[atom2];
1048	+	#else
1049	+	return snap_->atomData.velocity[atom2];
1050	+	#endif
1051	+	}
1052	+
1053	+
1054		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
1055
1056		Vector3d d;
#	Line 790 \| Line 1060 \| namespace OpenMD {
1060		#else
1061		d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
1062		#endif
1063	<
1064	<	snap_->wrapVector(d);
1063	>	if (usePeriodicBoundaryConditions_) {
1064	>	snap_->wrapVector(d);
1065	>	}
1066		return d;
1067		}
1068
#	Line 803 \| Line 1074 \| namespace OpenMD {
1074		#else
1075		d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1076		#endif
1077	<
1078	<	snap_->wrapVector(d);
1077	>	if (usePeriodicBoundaryConditions_) {
1078	>	snap_->wrapVector(d);
1079	>	}
1080		return d;
1081		}
1082
1083	<	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1083	>	RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1084		#ifdef IS_MPI
1085		return massFactorsRow[atom1];
1086		#else
#	Line 816 \| Line 1088 \| namespace OpenMD {
1088		#endif
1089		}
1090
1091	<	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1091	>	RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1092		#ifdef IS_MPI
1093		return massFactorsCol[atom2];
1094		#else
#	Line 833 \| Line 1105 \| namespace OpenMD {
1105		#else
1106		d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1107		#endif
1108	<
1109	<	snap_->wrapVector(d);
1108	>	if (usePeriodicBoundaryConditions_) {
1109	>	snap_->wrapVector(d);
1110	>	}
1111		return d;
1112		}
1113
1114	<	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1114	>	vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1115		return excludesForAtom[atom1];
1116		}
1117
#	Line 846 \| Line 1119 \| namespace OpenMD {
1119		* We need to exclude some overcounted interactions that result from
1120		* the parallel decomposition.
1121		*/
1122	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1122	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1123		int unique_id_1, unique_id_2;
1124	<
852	<
853	<	cerr << "sap with atom1, atom2 =\t" << atom1 << "\t" << atom2 << "\n";
1124	>
1125		#ifdef IS_MPI
1126		// in MPI, we have to look up the unique IDs for each atom
1127		unique_id_1 = AtomRowToGlobal[atom1];
1128		unique_id_2 = AtomColToGlobal[atom2];
1129	+	// group1 = cgRowToGlobal[cg1];
1130	+	// group2 = cgColToGlobal[cg2];
1131	+	#else
1132	+	unique_id_1 = AtomLocalToGlobal[atom1];
1133	+	unique_id_2 = AtomLocalToGlobal[atom2];
1134	+	int group1 = cgLocalToGlobal[cg1];
1135	+	int group2 = cgLocalToGlobal[cg2];
1136	+	#endif
1137
859	–	cerr << "sap with uid1, uid2 =\t" << unique_id_1 << "\t" << unique_id_2 << "\n";
860	–	// this situation should only arise in MPI simulations
1138		if (unique_id_1 == unique_id_2) return true;
1139	<
1139	>
1140	>	#ifdef IS_MPI
1141		// this prevents us from doing the pair on multiple processors
1142		if (unique_id_1 < unique_id_2) {
1143		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1144		} else {
1145	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1145	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1146		}
1147	+	#endif
1148	+
1149	+	#ifndef IS_MPI
1150	+	if (group1 == group2) {
1151	+	if (unique_id_1 < unique_id_2) return true;
1152	+	}
1153		#endif
1154	+
1155		return false;
1156		}
1157
#	Line 880 \| Line 1165 \| namespace OpenMD {
1165		* field) must still be handled for these pairs.
1166		*/
1167		bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1168	<	int unique_id_2;
1169	<	#ifdef IS_MPI
1170	<	// in MPI, we have to look up the unique IDs for the row atom.
886	<	unique_id_2 = AtomColToGlobal[atom2];
887	<	#else
888	<	// in the normal loop, the atom numbers are unique
889	<	unique_id_2 = atom2;
890	<	#endif
1168	>
1169	>	// excludesForAtom was constructed to use row/column indices in the MPI
1170	>	// version, and to use local IDs in the non-MPI version:
1171
1172		for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1173		i != excludesForAtom[atom1].end(); ++i) {
1174	<	if ( (*i) == unique_id_2 ) return true;
1174	>	if ( (*i) == atom2 ) return true;
1175		}
1176
1177		return false;
#	Line 921 \| Line 1201 \| namespace OpenMD {
1201		idat.excluded = excludeAtomPair(atom1, atom2);
1202
1203		#ifdef IS_MPI
1204	<	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1205	<	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1206	<	// ff_->getAtomType(identsCol[atom2]) );
1207	<
1204	>	//idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1205	>	idat.atid1 = identsRow[atom1];
1206	>	idat.atid2 = identsCol[atom2];
1207	>
1208	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1209	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1210	>	} else {
1211	>	idat.sameRegion = false;
1212	>	}
1213	>
1214		if (storageLayout_ & DataStorage::dslAmat) {
1215		idat.A1 = &(atomRowData.aMat[atom1]);
1216		idat.A2 = &(atomColData.aMat[atom2]);
1217		}
1218
933	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
934	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
935	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
936	–	}
937	–
1219		if (storageLayout_ & DataStorage::dslTorque) {
1220		idat.t1 = &(atomRowData.torque[atom1]);
1221		idat.t2 = &(atomColData.torque[atom2]);
1222		}
1223
1224	+	if (storageLayout_ & DataStorage::dslDipole) {
1225	+	idat.dipole1 = &(atomRowData.dipole[atom1]);
1226	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1227	+	}
1228	+
1229	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1230	+	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1231	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1232	+	}
1233	+
1234		if (storageLayout_ & DataStorage::dslDensity) {
1235		idat.rho1 = &(atomRowData.density[atom1]);
1236		idat.rho2 = &(atomColData.density[atom2]);
#	Line 965 \| Line 1256 \| namespace OpenMD {
1256		idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1257		}
1258
1259	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1260	+	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1261	+	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1262	+	}
1263	+
1264		#else
1265	+
1266	+	//idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1267	+	idat.atid1 = idents[atom1];
1268	+	idat.atid2 = idents[atom2];
1269
1270	<	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1271	<	//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1272	<	// ff_->getAtomType(idents[atom2]) );
1270	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1271	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1272	>	} else {
1273	>	idat.sameRegion = false;
1274	>	}
1275
1276		if (storageLayout_ & DataStorage::dslAmat) {
1277		idat.A1 = &(snap_->atomData.aMat[atom1]);
1278		idat.A2 = &(snap_->atomData.aMat[atom2]);
1279		}
1280
979	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
980	–	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
981	–	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
982	–	}
983	–
1281		if (storageLayout_ & DataStorage::dslTorque) {
1282		idat.t1 = &(snap_->atomData.torque[atom1]);
1283		idat.t2 = &(snap_->atomData.torque[atom2]);
1284		}
1285
1286	+	if (storageLayout_ & DataStorage::dslDipole) {
1287	+	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1288	+	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1289	+	}
1290	+
1291	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1292	+	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1293	+	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1294	+	}
1295	+
1296		if (storageLayout_ & DataStorage::dslDensity) {
1297		idat.rho1 = &(snap_->atomData.density[atom1]);
1298		idat.rho2 = &(snap_->atomData.density[atom2]);
#	Line 1010 \| Line 1317 \| namespace OpenMD {
1317		idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1318		idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1319		}
1320	+
1321	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1322	+	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1323	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1324	+	}
1325	+
1326		#endif
1327		}
1328
1329
1330		void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1331		#ifdef IS_MPI
1332	<	pot_row[atom1] += 0.5 * *(idat.pot);
1333	<	pot_col[atom2] += 0.5 * *(idat.pot);
1332	>	pot_row[atom1] += RealType(0.5) * *(idat.pot);
1333	>	pot_col[atom2] += RealType(0.5) * *(idat.pot);
1334	>	expot_row[atom1] += RealType(0.5) * *(idat.excludedPot);
1335	>	expot_col[atom2] += RealType(0.5) * *(idat.excludedPot);
1336
1337		atomRowData.force[atom1] += *(idat.f1);
1338		atomColData.force[atom2] -= *(idat.f1);
1339	+
1340	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1341	+	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1342	+	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1343	+	}
1344	+
1345	+	if (storageLayout_ & DataStorage::dslElectricField) {
1346	+	atomRowData.electricField[atom1] += *(idat.eField1);
1347	+	atomColData.electricField[atom2] += *(idat.eField2);
1348	+	}
1349	+
1350	+	if (storageLayout_ & DataStorage::dslSitePotential) {
1351	+	atomRowData.sitePotential[atom1] += *(idat.sPot1);
1352	+	atomColData.sitePotential[atom2] += *(idat.sPot2);
1353	+	}
1354	+
1355		#else
1356		pairwisePot += *(idat.pot);
1357	+	excludedPot += *(idat.excludedPot);
1358
1359		snap_->atomData.force[atom1] += *(idat.f1);
1360		snap_->atomData.force[atom2] -= *(idat.f1);
1361	+
1362	+	if (idat.doParticlePot) {
1363	+	// This is the pairwise contribution to the particle pot. The
1364	+	// embedding contribution is added in each of the low level
1365	+	// non-bonded routines. In parallel, this calculation is done
1366	+	// in collectData, not in unpackInteractionData.
1367	+	snap_->atomData.particlePot[atom1] += (idat.vpair) *(idat.sw);
1368	+	snap_->atomData.particlePot[atom2] += (idat.vpair) *(idat.sw);
1369	+	}
1370	+
1371	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1372	+	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1373	+	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1374	+	}
1375	+
1376	+	if (storageLayout_ & DataStorage::dslElectricField) {
1377	+	snap_->atomData.electricField[atom1] += *(idat.eField1);
1378	+	snap_->atomData.electricField[atom2] += *(idat.eField2);
1379	+	}
1380	+
1381	+	if (storageLayout_ & DataStorage::dslSitePotential) {
1382	+	snap_->atomData.sitePotential[atom1] += *(idat.sPot1);
1383	+	snap_->atomData.sitePotential[atom2] += *(idat.sPot2);
1384	+	}
1385	+
1386		#endif
1387
1388		}
#	Line 1036 \| Line 1393 \| namespace OpenMD {
1393		* first element of pair is row-indexed CutoffGroup
1394		* second element of pair is column-indexed CutoffGroup
1395		*/
1396	<	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1397	<
1398	<	vector<pair<int, int> > neighborList;
1396	>	void ForceMatrixDecomposition::buildNeighborList(vector<pair<int,int> >& neighborList) {
1397	>
1398	>	neighborList.clear();
1399		groupCutoffs cuts;
1400		bool doAllPairs = false;
1401
1402	+	RealType rList_ = (largestRcut_ + skinThickness_);
1403	+	RealType rcut, rcutsq, rlistsq;
1404	+	Snapshot* snap_ = sman_->getCurrentSnapshot();
1405	+	Mat3x3d box;
1406	+	Mat3x3d invBox;
1407	+
1408	+	Vector3d rs, scaled, dr;
1409	+	Vector3i whichCell;
1410	+	int cellIndex;
1411	+
1412		#ifdef IS_MPI
1413		cellListRow_.clear();
1414		cellListCol_.clear();
1415		#else
1416		cellList_.clear();
1417		#endif
1418	<
1419	<	RealType rList_ = (largestRcut_ + skinThickness_);
1420	<	RealType rl2 = rList_ * rList_;
1421	<	Snapshot* snap_ = sman_->getCurrentSnapshot();
1422	<	Mat3x3d Hmat = snap_->getHmat();
1423	<	Vector3d Hx = Hmat.getColumn(0);
1424	<	Vector3d Hy = Hmat.getColumn(1);
1425	<	Vector3d Hz = Hmat.getColumn(2);
1426	<
1427	<	nCells_.x() = (int) ( Hx.length() )/ rList_;
1428	<	nCells_.y() = (int) ( Hy.length() )/ rList_;
1429	<	nCells_.z() = (int) ( Hz.length() )/ rList_;
1430	<
1418	>
1419	>	if (!usePeriodicBoundaryConditions_) {
1420	>	box = snap_->getBoundingBox();
1421	>	invBox = snap_->getInvBoundingBox();
1422	>	} else {
1423	>	box = snap_->getHmat();
1424	>	invBox = snap_->getInvHmat();
1425	>	}
1426	>
1427	>	Vector3d boxX = box.getColumn(0);
1428	>	Vector3d boxY = box.getColumn(1);
1429	>	Vector3d boxZ = box.getColumn(2);
1430	>
1431	>	nCells_.x() = int( boxX.length() / rList_ );
1432	>	nCells_.y() = int( boxY.length() / rList_ );
1433	>	nCells_.z() = int( boxZ.length() / rList_ );
1434	>
1435		// handle small boxes where the cell offsets can end up repeating cells
1436
1437		if (nCells_.x() < 3) doAllPairs = true;
1438		if (nCells_.y() < 3) doAllPairs = true;
1439		if (nCells_.z() < 3) doAllPairs = true;
1440	<
1070	<	Mat3x3d invHmat = snap_->getInvHmat();
1071	<	Vector3d rs, scaled, dr;
1072	<	Vector3i whichCell;
1073	<	int cellIndex;
1440	>
1441		int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1442	<
1442	>
1443		#ifdef IS_MPI
1444		cellListRow_.resize(nCtot);
1445		cellListCol_.resize(nCtot);
1446		#else
1447		cellList_.resize(nCtot);
1448		#endif
1449	<
1449	>
1450		if (!doAllPairs) {
1451		#ifdef IS_MPI
1452	<
1452	>
1453		for (int i = 0; i < nGroupsInRow_; i++) {
1454		rs = cgRowData.position[i];
1455
1456		// scaled positions relative to the box vectors
1457	<	scaled = invHmat * rs;
1457	>	scaled = invBox * rs;
1458
1459		// wrap the vector back into the unit box by subtracting integer box
1460		// numbers
1461		for (int j = 0; j < 3; j++) {
1462		scaled[j] -= roundMe(scaled[j]);
1463		scaled[j] += 0.5;
1464	+	// Handle the special case when an object is exactly on the
1465	+	// boundary (a scaled coordinate of 1.0 is the same as
1466	+	// scaled coordinate of 0.0)
1467	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1468		}
1469
1470		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1111 \| Line 1482 \| namespace OpenMD {
1482		rs = cgColData.position[i];
1483
1484		// scaled positions relative to the box vectors
1485	<	scaled = invHmat * rs;
1485	>	scaled = invBox * rs;
1486
1487		// wrap the vector back into the unit box by subtracting integer box
1488		// numbers
1489		for (int j = 0; j < 3; j++) {
1490		scaled[j] -= roundMe(scaled[j]);
1491		scaled[j] += 0.5;
1492	+	// Handle the special case when an object is exactly on the
1493	+	// boundary (a scaled coordinate of 1.0 is the same as
1494	+	// scaled coordinate of 0.0)
1495	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1496		}
1497
1498		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1131 \| Line 1506 \| namespace OpenMD {
1506		// add this cutoff group to the list of groups in this cell;
1507		cellListCol_[cellIndex].push_back(i);
1508		}
1509	+
1510		#else
1511		for (int i = 0; i < nGroups_; i++) {
1512		rs = snap_->cgData.position[i];
1513
1514		// scaled positions relative to the box vectors
1515	<	scaled = invHmat * rs;
1515	>	scaled = invBox * rs;
1516
1517		// wrap the vector back into the unit box by subtracting integer box
1518		// numbers
1519		for (int j = 0; j < 3; j++) {
1520		scaled[j] -= roundMe(scaled[j]);
1521		scaled[j] += 0.5;
1522	+	// Handle the special case when an object is exactly on the
1523	+	// boundary (a scaled coordinate of 1.0 is the same as
1524	+	// scaled coordinate of 0.0)
1525	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1526		}
1527
1528		// find xyz-indices of cell that cutoffGroup is in.
1529	<	whichCell.x() = nCells_.x() * scaled.x();
1530	<	whichCell.y() = nCells_.y() * scaled.y();
1531	<	whichCell.z() = nCells_.z() * scaled.z();
1529	>	whichCell.x() = int(nCells_.x() * scaled.x());
1530	>	whichCell.y() = int(nCells_.y() * scaled.y());
1531	>	whichCell.z() = int(nCells_.z() * scaled.z());
1532
1533		// find single index of this cell:
1534		cellIndex = Vlinear(whichCell, nCells_);
#	Line 1156 \| Line 1536 \| namespace OpenMD {
1536		// add this cutoff group to the list of groups in this cell;
1537		cellList_[cellIndex].push_back(i);
1538		}
1539	+
1540		#endif
1541
1542		for (int m1z = 0; m1z < nCells_.z(); m1z++) {
#	Line 1168 \| Line 1549 \| namespace OpenMD {
1549		os != cellOffsets_.end(); ++os) {
1550
1551		Vector3i m2v = m1v + (*os);
1552	<
1552	>
1553	>
1554		if (m2v.x() >= nCells_.x()) {
1555		m2v.x() = 0;
1556		} else if (m2v.x() < 0) {
#	Line 1186 \| Line 1568 \| namespace OpenMD {
1568		} else if (m2v.z() < 0) {
1569		m2v.z() = nCells_.z() - 1;
1570		}
1571	<
1571	>
1572		int m2 = Vlinear (m2v, nCells_);
1573
1574		#ifdef IS_MPI
#	Line 1195 \| Line 1577 \| namespace OpenMD {
1577		for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1578		j2 != cellListCol_[m2].end(); ++j2) {
1579
1580	<	// In parallel, we need to visit all pairs of row &
1581	<	// column indicies and will truncate later on.
1580	>	// In parallel, we need to visit all pairs of row
1581	>	// & column indicies and will divide labor in the
1582	>	// force evaluation later.
1583		dr = cgColData.position[(j2)] - cgRowData.position[(j1)];
1584	<	snap_->wrapVector(dr);
1585	<	cuts = getGroupCutoffs( (j1), (j2) );
1586	<	if (dr.lengthSquare() < cuts.third) {
1584	>	if (usePeriodicBoundaryConditions_) {
1585	>	snap_->wrapVector(dr);
1586	>	}
1587	>	getGroupCutoffs( (j1), (j2), rcut, rcutsq, rlistsq );
1588	>	if (dr.lengthSquare() < rlistsq) {
1589		neighborList.push_back(make_pair((j1), (j2)));
1590		}
1591		}
1592		}
1593		#else
1209	–
1594		for (vector<int>::iterator j1 = cellList_[m1].begin();
1595		j1 != cellList_[m1].end(); ++j1) {
1596		for (vector<int>::iterator j2 = cellList_[m2].begin();
1597		j2 != cellList_[m2].end(); ++j2) {
1598	<
1598	>
1599		// Always do this if we're in different cells or if
1600	<	// we're in the same cell and the global index of the
1601	<	// j2 cutoff group is less than the j1 cutoff group
1602	<
1603	<	if (m2 != m1 \|\| (j2) < (j1)) {
1600	>	// we're in the same cell and the global index of
1601	>	// the j2 cutoff group is greater than or equal to
1602	>	// the j1 cutoff group. Note that Rappaport's code
1603	>	// has a "less than" conditional here, but that
1604	>	// deals with atom-by-atom computation. OpenMD
1605	>	// allows atoms within a single cutoff group to
1606	>	// interact with each other.
1607	>
1608	>	if (m2 != m1 \|\| (j2) >= (j1) ) {
1609	>
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		}
#	Line 1235 \| Line 1627 \| namespace OpenMD {
1627		// branch to do all cutoff group pairs
1628		#ifdef IS_MPI
1629		for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1630	<	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1630	>	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1631		dr = cgColData.position[j2] - cgRowData.position[j1];
1632	<	snap_->wrapVector(dr);
1633	<	cuts = getGroupCutoffs( j1, j2 );
1634	<	if (dr.lengthSquare() < cuts.third) {
1632	>	if (usePeriodicBoundaryConditions_) {
1633	>	snap_->wrapVector(dr);
1634	>	}
1635	>	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq);
1636	>	if (dr.lengthSquare() < rlistsq) {
1637		neighborList.push_back(make_pair(j1, j2));
1638		}
1639		}
1640	<	}
1640	>	}
1641		#else
1642	<	for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1643	<	for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1642	>	// include all groups here.
1643	>	for (int j1 = 0; j1 < nGroups_; j1++) {
1644	>	// include self group interactions j2 == j1
1645	>	for (int j2 = j1; j2 < nGroups_; j2++) {
1646		dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1647	<	snap_->wrapVector(dr);
1648	<	cuts = getGroupCutoffs( j1, j2 );
1649	<	if (dr.lengthSquare() < cuts.third) {
1647	>	if (usePeriodicBoundaryConditions_) {
1648	>	snap_->wrapVector(dr);
1649	>	}
1650	>	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq );
1651	>	if (dr.lengthSquare() < rlistsq) {
1652		neighborList.push_back(make_pair(j1, j2));
1653		}
1654	<	}
1655	<	}
1654	>	}
1655	>	}
1656		#endif
1657		}
1658
#	Line 1263 \| Line 1661 \| namespace OpenMD {
1661		saved_CG_positions_.clear();
1662		for (int i = 0; i < nGroups_; i++)
1663		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1266	–
1267	–	return neighborList;
1664		}
1665		} //end namespace OpenMD

Diff Legend

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

Comparing: branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1593 by gezelter, Fri Jul 15 21:35:14 2011 UTC vs. trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1993 by gezelter, Tue Apr 29 17:32:31 2014 UTC

Diff Legend

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1593 by gezelter, Fri Jul 15 21:35:14 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1993 by gezelter, Tue Apr 29 17:32:31 2014 UTC