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

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1723 by gezelter, Thu May 24 20:59:54 2012 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 2057 by gezelter, Tue Mar 3 15:22:26 2015 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).
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		*/
#	Line 50 \| Line 50 \| 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
53	>	// Row and colum scans must visit all surrounding cells
54		cellOffsets_.clear();
55		cellOffsets_.push_back( Vector3i(-1,-1,-1) );
56		cellOffsets_.push_back( Vector3i( 0,-1,-1) );
#	Line 82 \| Line 79 \| namespace OpenMD {
79		cellOffsets_.push_back( Vector3i(-1, 1, 1) );
80		cellOffsets_.push_back( Vector3i( 0, 1, 1) );
81		cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85	–	#endif
82		}
83
84
#	Line 99 \| Line 95 \| namespace OpenMD {
95		nGroups_ = info_->getNLocalCutoffGroups();
96		// gather the information for atomtype IDs (atids):
97		idents = info_->getIdentArray();
98	+	regions = info_->getRegions();
99		AtomLocalToGlobal = info_->getGlobalAtomIndices();
100		cgLocalToGlobal = info_->getGlobalGroupIndices();
101		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
#	Line 118 \| Line 115 \| namespace OpenMD {
115
116		#ifdef IS_MPI
117
118	<	MPI::Intracomm row = rowComm.getComm();
119	<	MPI::Intracomm col = colComm.getComm();
118	>	MPI_Comm row = rowComm.getComm();
119	>	MPI_Comm col = colComm.getComm();
120
121		AtomPlanIntRow = new Plan<int>(row, nLocal_);
122		AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
#	Line 163 \| Line 160 \| namespace OpenMD {
160
161		AtomPlanIntRow->gather(idents, identsRow);
162		AtomPlanIntColumn->gather(idents, identsCol);
163	+
164	+	regionsRow.resize(nAtomsInRow_);
165	+	regionsCol.resize(nAtomsInCol_);
166	+
167	+	AtomPlanIntRow->gather(regions, regionsRow);
168	+	AtomPlanIntColumn->gather(regions, regionsCol);
169
170		// allocate memory for the parallel objects
171		atypesRow.resize(nAtomsInRow_);
#	Line 176 \| Line 179 \| namespace OpenMD {
179		pot_row.resize(nAtomsInRow_);
180		pot_col.resize(nAtomsInCol_);
181
182	+	expot_row.resize(nAtomsInRow_);
183	+	expot_col.resize(nAtomsInCol_);
184	+
185		AtomRowToGlobal.resize(nAtomsInRow_);
186		AtomColToGlobal.resize(nAtomsInCol_);
187		AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
#	Line 296 \| Line 302 \| namespace OpenMD {
302		groupList_[i].push_back(j);
303		}
304		}
305	<	}
300	<
301	<
302	<	createGtypeCutoffMap();
303	<
305	>	}
306		}
305	–
306	–	void ForceMatrixDecomposition::createGtypeCutoffMap() {
307
308	<	RealType tol = 1e-6;
309	<	largestRcut_ = 0.0;
310	<	RealType rc;
311	<	int atid;
312	<	set<AtomType*> atypes = info_->getSimulatedAtomTypes();
313	<
314	<	map<int, RealType> atypeCutoff;
315	<
316	<	for (set<AtomType*>::iterator at = atypes.begin();
317	<	at != atypes.end(); ++at){
318	<	atid = (*at)->getIdent();
319	<	if (userChoseCutoff_)
320	<	atypeCutoff[atid] = userCutoff_;
321	<	else
322	<	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
323	<	}
324	<
325	<	vector<RealType> gTypeCutoffs;
326	<	// first we do a single loop over the cutoff groups to find the
327	<	// largest cutoff for any atypes present in this group.
328	<	#ifdef IS_MPI
329	<	vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
330	<	groupRowToGtype.resize(nGroupsInRow_);
331	<	for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
332	<	vector<int> atomListRow = getAtomsInGroupRow(cg1);
333	<	for (vector<int>::iterator ia = atomListRow.begin();
334	<	ia != atomListRow.end(); ++ia) {
335	<	int atom1 = (*ia);
336	<	atid = identsRow[atom1];
337	<	if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
338	<	groupCutoffRow[cg1] = atypeCutoff[atid];
339	<	}
340	<	}
341	<
342	<	bool gTypeFound = false;
343	<	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
344	<	if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
345	<	groupRowToGtype[cg1] = gt;
346	<	gTypeFound = true;
347	<	}
348	<	}
349	<	if (!gTypeFound) {
350	<	gTypeCutoffs.push_back( groupCutoffRow[cg1] );
351	<	groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
352	<	}
353	<
354	<	}
355	<	vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
356	<	groupColToGtype.resize(nGroupsInCol_);
357	<	for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
358	<	vector<int> atomListCol = getAtomsInGroupColumn(cg2);
359	<	for (vector<int>::iterator jb = atomListCol.begin();
360	<	jb != atomListCol.end(); ++jb) {
361	<	int atom2 = (*jb);
362	<	atid = identsCol[atom2];
363	<	if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
364	<	groupCutoffCol[cg2] = atypeCutoff[atid];
365	<	}
366	<	}
367	<	bool gTypeFound = false;
368	<	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
369	<	if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
370	<	groupColToGtype[cg2] = gt;
371	<	gTypeFound = true;
372	<	}
373	<	}
374	<	if (!gTypeFound) {
375	<	gTypeCutoffs.push_back( groupCutoffCol[cg2] );
376	<	groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
377	<	}
378	<	}
379	<	#else
380	<
381	<	vector<RealType> groupCutoff(nGroups_, 0.0);
382	<	groupToGtype.resize(nGroups_);
383	<	for (int cg1 = 0; cg1 < nGroups_; cg1++) {
384	<	groupCutoff[cg1] = 0.0;
385	<	vector<int> atomList = getAtomsInGroupRow(cg1);
386	<	for (vector<int>::iterator ia = atomList.begin();
387	<	ia != atomList.end(); ++ia) {
388	<	int atom1 = (*ia);
389	<	atid = idents[atom1];
390	<	if (atypeCutoff[atid] > groupCutoff[cg1])
391	<	groupCutoff[cg1] = atypeCutoff[atid];
392	<	}
393	<
394	<	bool gTypeFound = false;
395	<	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
396	<	if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
397	<	groupToGtype[cg1] = gt;
398	<	gTypeFound = true;
399	<	}
400	<	}
401	<	if (!gTypeFound) {
402	<	gTypeCutoffs.push_back( groupCutoff[cg1] );
403	<	groupToGtype[cg1] = gTypeCutoffs.size() - 1;
404	<	}
405	<	}
406	<	#endif
407	<
408	<	// Now we find the maximum group cutoff value present in the simulation
409	<
410	<	RealType groupMax = *max_element(gTypeCutoffs.begin(),
411	<	gTypeCutoffs.end());
412	<
413	<	#ifdef IS_MPI
414	<	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
415	<	MPI::MAX);
416	<	#endif
417	<
418	<	RealType tradRcut = groupMax;
419	<
420	<	for (int i = 0; i < gTypeCutoffs.size(); i++) {
421	<	for (int j = 0; j < gTypeCutoffs.size(); j++) {
422	<	RealType thisRcut;
423	<	switch(cutoffPolicy_) {
424	<	case TRADITIONAL:
425	<	thisRcut = tradRcut;
426	<	break;
427	<	case MIX:
428	<	thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
429	<	break;
430	<	case MAX:
431	<	thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
432	<	break;
433	<	default:
434	<	sprintf(painCave.errMsg,
435	<	"ForceMatrixDecomposition::createGtypeCutoffMap "
436	<	"hit an unknown cutoff policy!\n");
437	<	painCave.severity = OPENMD_ERROR;
438	<	painCave.isFatal = 1;
439	<	simError();
440	<	break;
441	<	}
442	<
443	<	pair<int,int> key = make_pair(i,j);
444	<	gTypeCutoffMap[key].first = thisRcut;
445	<	if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
446	<	gTypeCutoffMap[key].second = thisRcut*thisRcut;
447	<	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
448	<	// sanity check
449	<
450	<	if (userChoseCutoff_) {
451	<	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
452	<	sprintf(painCave.errMsg,
453	<	"ForceMatrixDecomposition::createGtypeCutoffMap "
454	<	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
455	<	painCave.severity = OPENMD_ERROR;
456	<	painCave.isFatal = 1;
457	<	simError();
458	<	}
459	<	}
460	<	}
461	<	}
462	<	}
463	<
464	<
465	<	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
466	<	int i, j;
467	<	#ifdef IS_MPI
468	<	i = groupRowToGtype[cg1];
469	<	j = groupColToGtype[cg2];
470	<	#else
471	<	i = groupToGtype[cg1];
472	<	j = groupToGtype[cg2];
473	<	#endif
474	<	return gTypeCutoffMap[make_pair(i,j)];
475	<	}
476	<
477	<	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
478	<	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
308	>	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
309	>	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
310		if (toposForAtom[atom1][j] == atom2)
311		return topoDist[atom1][j];
312	<	}
312	>	}
313		return 0;
314		}
315
316		void ForceMatrixDecomposition::zeroWorkArrays() {
317		pairwisePot = 0.0;
318		embeddingPot = 0.0;
319	+	excludedPot = 0.0;
320	+	excludedSelfPot = 0.0;
321
322		#ifdef IS_MPI
323		if (storageLayout_ & DataStorage::dslForce) {
#	Line 503 \| Line 336 \| namespace OpenMD {
336		fill(pot_col.begin(), pot_col.end(),
337		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
338
339	+	fill(expot_row.begin(), expot_row.end(),
340	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
341	+
342	+	fill(expot_col.begin(), expot_col.end(),
343	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
344	+
345		if (storageLayout_ & DataStorage::dslParticlePot) {
346		fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
347		0.0);
#	Line 550 \| Line 389 \| namespace OpenMD {
389		atomColData.electricField.end(), V3Zero);
390		}
391
392	<	if (storageLayout_ & DataStorage::dslFlucQForce) {
393	<	fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(),
394	<	0.0);
395	<	fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.end(),
396	<	0.0);
392	>	if (storageLayout_ & DataStorage::dslSitePotential) {
393	>	fill(atomRowData.sitePotential.begin(),
394	>	atomRowData.sitePotential.end(), 0.0);
395	>	fill(atomColData.sitePotential.begin(),
396	>	atomColData.sitePotential.end(), 0.0);
397		}
398
399		#endif
#	Line 589 \| Line 428 \| namespace OpenMD {
428		fill(snap_->atomData.electricField.begin(),
429		snap_->atomData.electricField.end(), V3Zero);
430		}
431	+	if (storageLayout_ & DataStorage::dslSitePotential) {
432	+	fill(snap_->atomData.sitePotential.begin(),
433	+	snap_->atomData.sitePotential.end(), 0.0);
434	+	}
435		}
436
437
438		void ForceMatrixDecomposition::distributeData() {
439		snap_ = sman_->getCurrentSnapshot();
440		storageLayout_ = sman_->getStorageLayout();
441	+
442	+	bool needsCG = true;
443	+	if(info_->getNCutoffGroups() != info_->getNAtoms())
444	+	needsCG = false;
445	+
446		#ifdef IS_MPI
447
448		// gather up the atomic positions
#	Line 605 \| Line 453 \| namespace OpenMD {
453
454		// gather up the cutoff group positions
455
456	<	cgPlanVectorRow->gather(snap_->cgData.position,
457	<	cgRowData.position);
456	>	if (needsCG) {
457	>	cgPlanVectorRow->gather(snap_->cgData.position,
458	>	cgRowData.position);
459	>
460	>	cgPlanVectorColumn->gather(snap_->cgData.position,
461	>	cgColData.position);
462	>	}
463
611	–	cgPlanVectorColumn->gather(snap_->cgData.position,
612	–	cgColData.position);
464
614	–
615	–
465		if (needVelocities_) {
466		// gather up the atomic velocities
467		AtomPlanVectorColumn->gather(snap_->atomData.velocity,
468		atomColData.velocity);
469	<
470	<	cgPlanVectorColumn->gather(snap_->cgData.velocity,
471	<	cgColData.velocity);
469	>
470	>	if (needsCG) {
471	>	cgPlanVectorColumn->gather(snap_->cgData.velocity,
472	>	cgColData.velocity);
473	>	}
474		}
475
476
#	Line 630 \| Line 481 \| namespace OpenMD {
481		AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
482		atomColData.aMat);
483		}
484	<
485	<	// if needed, gather the atomic eletrostatic frames
486	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
487	<	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
488	<	atomRowData.electroFrame);
489	<	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
490	<	atomColData.electroFrame);
484	>
485	>	// if needed, gather the atomic eletrostatic information
486	>	if (storageLayout_ & DataStorage::dslDipole) {
487	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
488	>	atomRowData.dipole);
489	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
490	>	atomColData.dipole);
491		}
492
493	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
494	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
495	+	atomRowData.quadrupole);
496	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
497	+	atomColData.quadrupole);
498	+	}
499	+
500		// if needed, gather the atomic fluctuating charge values
501		if (storageLayout_ & DataStorage::dslFlucQPosition) {
502		AtomPlanRealRow->gather(snap_->atomData.flucQPos,
#	Line 670 \| Line 528 \| namespace OpenMD {
528		snap_->atomData.density[i] += rho_tmp[i];
529		}
530
531	+	// this isn't necessary if we don't have polarizable atoms, but
532	+	// we'll leave it here for now.
533		if (storageLayout_ & DataStorage::dslElectricField) {
534
535		AtomPlanVectorRow->scatter(atomRowData.electricField,
#	Line 677 \| Line 537 \| namespace OpenMD {
537
538		int n = snap_->atomData.electricField.size();
539		vector<Vector3d> field_tmp(n, V3Zero);
540	<	AtomPlanVectorColumn->scatter(atomColData.electricField, field_tmp);
540	>	AtomPlanVectorColumn->scatter(atomColData.electricField,
541	>	field_tmp);
542		for (int i = 0; i < n; i++)
543		snap_->atomData.electricField[i] += field_tmp[i];
544		}
#	Line 777 \| Line 638 \| namespace OpenMD {
638
639		}
640
641	+	if (storageLayout_ & DataStorage::dslElectricField) {
642	+
643	+	int nef = snap_->atomData.electricField.size();
644	+	vector<Vector3d> efield_tmp(nef, V3Zero);
645	+
646	+	AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
647	+	for (int i = 0; i < nef; i++) {
648	+	snap_->atomData.electricField[i] += efield_tmp[i];
649	+	efield_tmp[i] = 0.0;
650	+	}
651	+
652	+	AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
653	+	for (int i = 0; i < nef; i++)
654	+	snap_->atomData.electricField[i] += efield_tmp[i];
655	+	}
656	+
657	+	if (storageLayout_ & DataStorage::dslSitePotential) {
658	+
659	+	int nsp = snap_->atomData.sitePotential.size();
660	+	vector<RealType> sp_tmp(nsp, 0.0);
661	+
662	+	AtomPlanRealRow->scatter(atomRowData.sitePotential, sp_tmp);
663	+	for (int i = 0; i < nsp; i++) {
664	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
665	+	sp_tmp[i] = 0.0;
666	+	}
667	+
668	+	AtomPlanRealColumn->scatter(atomColData.sitePotential, sp_tmp);
669	+	for (int i = 0; i < nsp; i++)
670	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
671	+	}
672	+
673		nLocal_ = snap_->getNumberOfAtoms();
674
675		vector<potVec> pot_temp(nLocal_,
676		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
677	+	vector<potVec> expot_temp(nLocal_,
678	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
679
680		// scatter/gather pot_row into the members of my column
681
682		AtomPlanPotRow->scatter(pot_row, pot_temp);
683	+	AtomPlanPotRow->scatter(expot_row, expot_temp);
684
685	<	for (int ii = 0; ii < pot_temp.size(); ii++ )
685	>	for (int ii = 0; ii < pot_temp.size(); ii++ )
686		pairwisePot += pot_temp[ii];
687	+
688	+	for (int ii = 0; ii < expot_temp.size(); ii++ )
689	+	excludedPot += expot_temp[ii];
690
691		if (storageLayout_ & DataStorage::dslParticlePot) {
692		// This is the pairwise contribution to the particle pot. The
#	Line 805 \| Line 704 \| namespace OpenMD {
704
705		fill(pot_temp.begin(), pot_temp.end(),
706		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
707	+	fill(expot_temp.begin(), expot_temp.end(),
708	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
709
710		AtomPlanPotColumn->scatter(pot_col, pot_temp);
711	+	AtomPlanPotColumn->scatter(expot_col, expot_temp);
712
713		for (int ii = 0; ii < pot_temp.size(); ii++ )
714		pairwisePot += pot_temp[ii];
715
716	+	for (int ii = 0; ii < expot_temp.size(); ii++ )
717	+	excludedPot += expot_temp[ii];
718	+
719		if (storageLayout_ & DataStorage::dslParticlePot) {
720		// This is the pairwise contribution to the particle pot. The
721		// embedding contribution is added in each of the low level
#	Line 848 \| Line 753 \| namespace OpenMD {
753		for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
754		RealType ploc1 = pairwisePot[ii];
755		RealType ploc2 = 0.0;
756	<	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
756	>	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
757		pairwisePot[ii] = ploc2;
758		}
759
760		for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
761	<	RealType ploc1 = embeddingPot[ii];
761	>	RealType ploc1 = excludedPot[ii];
762		RealType ploc2 = 0.0;
763	<	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
764	<	embeddingPot[ii] = ploc2;
763	>	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
764	>	excludedPot[ii] = ploc2;
765		}
766	<
766	>
767		// Here be dragons.
768	<	MPI::Intracomm col = colComm.getComm();
768	>	MPI_Comm col = colComm.getComm();
769
770	<	col.Allreduce(MPI::IN_PLACE,
770	>	MPI_Allreduce(MPI_IN_PLACE,
771		&snap_->frameData.conductiveHeatFlux[0], 3,
772	<	MPI::REALTYPE, MPI::SUM);
772	>	MPI_REALTYPE, MPI_SUM, col);
773
774
775		#endif
776
777		}
778
779	<	int ForceMatrixDecomposition::getNAtomsInRow() {
779	>	/**
780	>	* Collects information obtained during the post-pair (and embedding
781	>	* functional) loops onto local data structures.
782	>	*/
783	>	void ForceMatrixDecomposition::collectSelfData() {
784	>	snap_ = sman_->getCurrentSnapshot();
785	>	storageLayout_ = sman_->getStorageLayout();
786	>
787		#ifdef IS_MPI
788	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
789	+	RealType ploc1 = embeddingPot[ii];
790	+	RealType ploc2 = 0.0;
791	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
792	+	embeddingPot[ii] = ploc2;
793	+	}
794	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
795	+	RealType ploc1 = excludedSelfPot[ii];
796	+	RealType ploc2 = 0.0;
797	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
798	+	excludedSelfPot[ii] = ploc2;
799	+	}
800	+	#endif
801	+
802	+	}
803	+
804	+
805	+
806	+	int& ForceMatrixDecomposition::getNAtomsInRow() {
807	+	#ifdef IS_MPI
808		return nAtomsInRow_;
809		#else
810		return nLocal_;
#	Line 882 \| Line 814 \| namespace OpenMD {
814		/**
815		* returns the list of atoms belonging to this group.
816		*/
817	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
817	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
818		#ifdef IS_MPI
819		return groupListRow_[cg1];
820		#else
#	Line 890 \| Line 822 \| namespace OpenMD {
822		#endif
823		}
824
825	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
825	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
826		#ifdef IS_MPI
827		return groupListCol_[cg2];
828		#else
#	Line 907 \| Line 839 \| namespace OpenMD {
839		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
840		#endif
841
842	<	snap_->wrapVector(d);
842	>	if (usePeriodicBoundaryConditions_) {
843	>	snap_->wrapVector(d);
844	>	}
845		return d;
846		}
847
848	<	Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
848	>	Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
849		#ifdef IS_MPI
850		return cgColData.velocity[cg2];
851		#else
#	Line 919 \| Line 853 \| namespace OpenMD {
853		#endif
854		}
855
856	<	Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
856	>	Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
857		#ifdef IS_MPI
858		return atomColData.velocity[atom2];
859		#else
#	Line 937 \| Line 871 \| namespace OpenMD {
871		#else
872		d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
873		#endif
874	<
875	<	snap_->wrapVector(d);
874	>	if (usePeriodicBoundaryConditions_) {
875	>	snap_->wrapVector(d);
876	>	}
877		return d;
878		}
879
#	Line 950 \| Line 885 \| namespace OpenMD {
885		#else
886		d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
887		#endif
888	<
889	<	snap_->wrapVector(d);
888	>	if (usePeriodicBoundaryConditions_) {
889	>	snap_->wrapVector(d);
890	>	}
891		return d;
892		}
893
894	<	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
894	>	RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
895		#ifdef IS_MPI
896		return massFactorsRow[atom1];
897		#else
#	Line 963 \| Line 899 \| namespace OpenMD {
899		#endif
900		}
901
902	<	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
902	>	RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
903		#ifdef IS_MPI
904		return massFactorsCol[atom2];
905		#else
#	Line 980 \| Line 916 \| namespace OpenMD {
916		#else
917		d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
918		#endif
919	<
920	<	snap_->wrapVector(d);
919	>	if (usePeriodicBoundaryConditions_) {
920	>	snap_->wrapVector(d);
921	>	}
922		return d;
923		}
924
925	<	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
925	>	vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
926		return excludesForAtom[atom1];
927		}
928
#	Line 993 \| Line 930 \| namespace OpenMD {
930		* We need to exclude some overcounted interactions that result from
931		* the parallel decomposition.
932		*/
933	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
933	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
934		int unique_id_1, unique_id_2;
935
936		#ifdef IS_MPI
937		// in MPI, we have to look up the unique IDs for each atom
938		unique_id_1 = AtomRowToGlobal[atom1];
939		unique_id_2 = AtomColToGlobal[atom2];
940	+	// group1 = cgRowToGlobal[cg1];
941	+	// group2 = cgColToGlobal[cg2];
942		#else
943		unique_id_1 = AtomLocalToGlobal[atom1];
944		unique_id_2 = AtomLocalToGlobal[atom2];
945	+	int group1 = cgLocalToGlobal[cg1];
946	+	int group2 = cgLocalToGlobal[cg2];
947		#endif
948
949		if (unique_id_1 == unique_id_2) return true;
#	Line 1014 \| Line 955 \| namespace OpenMD {
955		} else {
956		if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
957		}
958	+	#endif
959	+
960	+	#ifndef IS_MPI
961	+	if (group1 == group2) {
962	+	if (unique_id_1 < unique_id_2) return true;
963	+	}
964		#endif
965
966		return false;
#	Line 1060 \| Line 1007 \| namespace OpenMD {
1007
1008		// filling interaction blocks with pointers
1009		void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1010	<	int atom1, int atom2) {
1010	>	int atom1, int atom2,
1011	>	bool newAtom1) {
1012
1013		idat.excluded = excludeAtomPair(atom1, atom2);
1014	<
1014	>
1015	>	if (newAtom1) {
1016	>
1017		#ifdef IS_MPI
1018	<	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1019	<	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1020	<	// ff_->getAtomType(identsCol[atom2]) );
1021	<
1018	>	idat.atid1 = identsRow[atom1];
1019	>	idat.atid2 = identsCol[atom2];
1020	>
1021	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1022	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1023	>	} else {
1024	>	idat.sameRegion = false;
1025	>	}
1026	>
1027	>	if (storageLayout_ & DataStorage::dslAmat) {
1028	>	idat.A1 = &(atomRowData.aMat[atom1]);
1029	>	idat.A2 = &(atomColData.aMat[atom2]);
1030	>	}
1031	>
1032	>	if (storageLayout_ & DataStorage::dslTorque) {
1033	>	idat.t1 = &(atomRowData.torque[atom1]);
1034	>	idat.t2 = &(atomColData.torque[atom2]);
1035	>	}
1036	>
1037	>	if (storageLayout_ & DataStorage::dslDipole) {
1038	>	idat.dipole1 = &(atomRowData.dipole[atom1]);
1039	>	idat.dipole2 = &(atomColData.dipole[atom2]);
1040	>	}
1041	>
1042	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1043	>	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1044	>	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1045	>	}
1046	>
1047	>	if (storageLayout_ & DataStorage::dslDensity) {
1048	>	idat.rho1 = &(atomRowData.density[atom1]);
1049	>	idat.rho2 = &(atomColData.density[atom2]);
1050	>	}
1051	>
1052	>	if (storageLayout_ & DataStorage::dslFunctional) {
1053	>	idat.frho1 = &(atomRowData.functional[atom1]);
1054	>	idat.frho2 = &(atomColData.functional[atom2]);
1055	>	}
1056	>
1057	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1058	>	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1059	>	idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1060	>	}
1061	>
1062	>	if (storageLayout_ & DataStorage::dslParticlePot) {
1063	>	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1064	>	idat.particlePot2 = &(atomColData.particlePot[atom2]);
1065	>	}
1066	>
1067	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1068	>	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1069	>	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1070	>	}
1071	>
1072	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1073	>	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1074	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1075	>	}
1076	>
1077	>	#else
1078	>
1079	>	idat.atid1 = idents[atom1];
1080	>	idat.atid2 = idents[atom2];
1081	>
1082	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1083	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1084	>	} else {
1085	>	idat.sameRegion = false;
1086	>	}
1087	>
1088	>	if (storageLayout_ & DataStorage::dslAmat) {
1089	>	idat.A1 = &(snap_->atomData.aMat[atom1]);
1090	>	idat.A2 = &(snap_->atomData.aMat[atom2]);
1091	>	}
1092	>
1093	>	if (storageLayout_ & DataStorage::dslTorque) {
1094	>	idat.t1 = &(snap_->atomData.torque[atom1]);
1095	>	idat.t2 = &(snap_->atomData.torque[atom2]);
1096	>	}
1097	>
1098	>	if (storageLayout_ & DataStorage::dslDipole) {
1099	>	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1100	>	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1101	>	}
1102	>
1103	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1104	>	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1105	>	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1106	>	}
1107	>
1108	>	if (storageLayout_ & DataStorage::dslDensity) {
1109	>	idat.rho1 = &(snap_->atomData.density[atom1]);
1110	>	idat.rho2 = &(snap_->atomData.density[atom2]);
1111	>	}
1112	>
1113	>	if (storageLayout_ & DataStorage::dslFunctional) {
1114	>	idat.frho1 = &(snap_->atomData.functional[atom1]);
1115	>	idat.frho2 = &(snap_->atomData.functional[atom2]);
1116	>	}
1117	>
1118	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1119	>	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1120	>	idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1121	>	}
1122	>
1123	>	if (storageLayout_ & DataStorage::dslParticlePot) {
1124	>	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1125	>	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1126	>	}
1127	>
1128	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1129	>	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1130	>	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1131	>	}
1132	>
1133	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1134	>	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1135	>	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1136	>	}
1137	>	#endif
1138	>
1139	>	} else {
1140	>	// atom1 is not new, so don't bother updating properties of that atom:
1141	>	#ifdef IS_MPI
1142	>	idat.atid2 = identsCol[atom2];
1143	>
1144	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1145	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1146	>	} else {
1147	>	idat.sameRegion = false;
1148	>	}
1149	>
1150		if (storageLayout_ & DataStorage::dslAmat) {
1073	–	idat.A1 = &(atomRowData.aMat[atom1]);
1151		idat.A2 = &(atomColData.aMat[atom2]);
1152		}
1153
1077	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
1078	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1079	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1080	–	}
1081	–
1154		if (storageLayout_ & DataStorage::dslTorque) {
1083	–	idat.t1 = &(atomRowData.torque[atom1]);
1155		idat.t2 = &(atomColData.torque[atom2]);
1156		}
1157
1158	+	if (storageLayout_ & DataStorage::dslDipole) {
1159	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1160	+	}
1161	+
1162	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1163	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1164	+	}
1165	+
1166		if (storageLayout_ & DataStorage::dslDensity) {
1088	–	idat.rho1 = &(atomRowData.density[atom1]);
1167		idat.rho2 = &(atomColData.density[atom2]);
1168		}
1169
1170		if (storageLayout_ & DataStorage::dslFunctional) {
1093	–	idat.frho1 = &(atomRowData.functional[atom1]);
1171		idat.frho2 = &(atomColData.functional[atom2]);
1172		}
1173
1174		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1098	–	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1175		idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1176		}
1177
1178		if (storageLayout_ & DataStorage::dslParticlePot) {
1103	–	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1179		idat.particlePot2 = &(atomColData.particlePot[atom2]);
1180		}
1181
1182		if (storageLayout_ & DataStorage::dslSkippedCharge) {
1108	–	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1183		idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1184		}
1185
1186	<	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1113	<	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1186	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1187		idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1188		}
1189
1190	<	#else
1191	<
1190	>	#else
1191	>	idat.atid2 = idents[atom2];
1192
1193	<	// cerr << "atoms = " << atom1 << " " << atom2 << "\n";
1194	<	// cerr << "pos1 = " << snap_->atomData.position[atom1] << "\n";
1195	<	// cerr << "pos2 = " << snap_->atomData.position[atom2] << "\n";
1193	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1194	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1195	>	} else {
1196	>	idat.sameRegion = false;
1197	>	}
1198
1124	–	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1125	–	//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1126	–	// ff_->getAtomType(idents[atom2]) );
1127	–
1199		if (storageLayout_ & DataStorage::dslAmat) {
1129	–	idat.A1 = &(snap_->atomData.aMat[atom1]);
1200		idat.A2 = &(snap_->atomData.aMat[atom2]);
1201		}
1202
1133	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
1134	–	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1135	–	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1136	–	}
1137	–
1203		if (storageLayout_ & DataStorage::dslTorque) {
1139	–	idat.t1 = &(snap_->atomData.torque[atom1]);
1204		idat.t2 = &(snap_->atomData.torque[atom2]);
1205		}
1206
1207	+	if (storageLayout_ & DataStorage::dslDipole) {
1208	+	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1209	+	}
1210	+
1211	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1212	+	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1213	+	}
1214	+
1215		if (storageLayout_ & DataStorage::dslDensity) {
1144	–	idat.rho1 = &(snap_->atomData.density[atom1]);
1216		idat.rho2 = &(snap_->atomData.density[atom2]);
1217		}
1218
1219		if (storageLayout_ & DataStorage::dslFunctional) {
1149	–	idat.frho1 = &(snap_->atomData.functional[atom1]);
1220		idat.frho2 = &(snap_->atomData.functional[atom2]);
1221		}
1222
1223		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1154	–	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1224		idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1225		}
1226
1227		if (storageLayout_ & DataStorage::dslParticlePot) {
1159	–	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1228		idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1229		}
1230
1231		if (storageLayout_ & DataStorage::dslSkippedCharge) {
1164	–	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1232		idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1233		}
1234
1235		if (storageLayout_ & DataStorage::dslFlucQPosition) {
1169	–	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1236		idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1237		}
1238
1239		#endif
1240	+	}
1241		}
1175	–
1242
1243	<	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1243	>	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat,
1244	>	int atom1, int atom2) {
1245		#ifdef IS_MPI
1246		pot_row[atom1] += RealType(0.5) * *(idat.pot);
1247		pot_col[atom2] += RealType(0.5) * *(idat.pot);
1248	+	expot_row[atom1] += RealType(0.5) * *(idat.excludedPot);
1249	+	expot_col[atom2] += RealType(0.5) * *(idat.excludedPot);
1250
1251		atomRowData.force[atom1] += *(idat.f1);
1252		atomColData.force[atom2] -= *(idat.f1);
1253
1254		if (storageLayout_ & DataStorage::dslFlucQForce) {
1255	<	atomRowData.flucQFrc[atom1] += *(idat.dVdFQ1);
1256	<	atomColData.flucQFrc[atom2] += *(idat.dVdFQ2);
1255	>	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1256	>	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1257		}
1258
1259		if (storageLayout_ & DataStorage::dslElectricField) {
#	Line 1192 \| Line 1261 \| namespace OpenMD {
1261		atomColData.electricField[atom2] += *(idat.eField2);
1262		}
1263
1264	+	if (storageLayout_ & DataStorage::dslSitePotential) {
1265	+	atomRowData.sitePotential[atom1] += *(idat.sPot1);
1266	+	atomColData.sitePotential[atom2] += *(idat.sPot2);
1267	+	}
1268	+
1269		#else
1270		pairwisePot += *(idat.pot);
1271	+	excludedPot += *(idat.excludedPot);
1272
1273		snap_->atomData.force[atom1] += *(idat.f1);
1274		snap_->atomData.force[atom2] -= *(idat.f1);
#	Line 1208 \| Line 1283 \| namespace OpenMD {
1283		}
1284
1285		if (storageLayout_ & DataStorage::dslFlucQForce) {
1286	<	snap_->atomData.flucQFrc[atom1] += *(idat.dVdFQ1);
1286	>	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1287		snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1288		}
1289
#	Line 1217 \| Line 1292 \| namespace OpenMD {
1292		snap_->atomData.electricField[atom2] += *(idat.eField2);
1293		}
1294
1295	+	if (storageLayout_ & DataStorage::dslSitePotential) {
1296	+	snap_->atomData.sitePotential[atom1] += *(idat.sPot1);
1297	+	snap_->atomData.sitePotential[atom2] += *(idat.sPot2);
1298	+	}
1299	+
1300		#endif
1301
1302		}
#	Line 1224 \| Line 1304 \| namespace OpenMD {
1304		/*
1305		* buildNeighborList
1306		*
1307	<	* first element of pair is row-indexed CutoffGroup
1308	<	* second element of pair is column-indexed CutoffGroup
1307	>	* Constructs the Verlet neighbor list for a force-matrix
1308	>	* decomposition. In this case, each processor is responsible for
1309	>	* row-site interactions with column-sites.
1310	>	*
1311	>	* neighborList is returned as a packed array of neighboring
1312	>	* column-ordered CutoffGroups. The starting position in
1313	>	* neighborList for each row-ordered CutoffGroup is given by the
1314	>	* returned vector point.
1315		*/
1316	<	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1317	<
1318	<	vector<pair<int, int> > neighborList;
1319	<	groupCutoffs cuts;
1316	>	void ForceMatrixDecomposition::buildNeighborList(vector<int>& neighborList,
1317	>	vector<int>& point) {
1318	>	neighborList.clear();
1319	>	point.clear();
1320	>	int len = 0;
1321	>
1322		bool doAllPairs = false;
1323
1324	+	Snapshot* snap_ = sman_->getCurrentSnapshot();
1325	+	Mat3x3d box;
1326	+	Mat3x3d invBox;
1327	+
1328	+	Vector3d rs, scaled, dr;
1329	+	Vector3i whichCell;
1330	+	int cellIndex;
1331	+
1332		#ifdef IS_MPI
1333		cellListRow_.clear();
1334		cellListCol_.clear();
1335	+	point.resize(nGroupsInRow_+1);
1336		#else
1337		cellList_.clear();
1338	+	point.resize(nGroups_+1);
1339		#endif
1340	+
1341	+	if (!usePeriodicBoundaryConditions_) {
1342	+	box = snap_->getBoundingBox();
1343	+	invBox = snap_->getInvBoundingBox();
1344	+	} else {
1345	+	box = snap_->getHmat();
1346	+	invBox = snap_->getInvHmat();
1347	+	}
1348	+
1349	+	Vector3d A = box.getColumn(0);
1350	+	Vector3d B = box.getColumn(1);
1351	+	Vector3d C = box.getColumn(2);
1352
1353	<	RealType rList_ = (largestRcut_ + skinThickness_);
1354	<	RealType rl2 = rList_ * rList_;
1355	<	Snapshot* snap_ = sman_->getCurrentSnapshot();
1356	<	Mat3x3d Hmat = snap_->getHmat();
1247	<	Vector3d Hx = Hmat.getColumn(0);
1248	<	Vector3d Hy = Hmat.getColumn(1);
1249	<	Vector3d Hz = Hmat.getColumn(2);
1353	>	// Required for triclinic cells
1354	>	Vector3d AxB = cross(A, B);
1355	>	Vector3d BxC = cross(B, C);
1356	>	Vector3d CxA = cross(C, A);
1357
1358	<	nCells_.x() = (int) ( Hx.length() )/ rList_;
1359	<	nCells_.y() = (int) ( Hy.length() )/ rList_;
1360	<	nCells_.z() = (int) ( Hz.length() )/ rList_;
1358	>	// unit vectors perpendicular to the faces of the triclinic cell:
1359	>	AxB.normalize();
1360	>	BxC.normalize();
1361	>	CxA.normalize();
1362
1363	<	// handle small boxes where the cell offsets can end up repeating cells
1363	>	// A set of perpendicular lengths in triclinic cells:
1364	>	RealType Wa = abs(dot(A, BxC));
1365	>	RealType Wb = abs(dot(B, CxA));
1366	>	RealType Wc = abs(dot(C, AxB));
1367
1368	+	nCells_.x() = int( Wa / rList_ );
1369	+	nCells_.y() = int( Wb / rList_ );
1370	+	nCells_.z() = int( Wc / rList_ );
1371	+
1372	+	// handle small boxes where the cell offsets can end up repeating cells
1373		if (nCells_.x() < 3) doAllPairs = true;
1374		if (nCells_.y() < 3) doAllPairs = true;
1375		if (nCells_.z() < 3) doAllPairs = true;
1376	<
1261	<	Mat3x3d invHmat = snap_->getInvHmat();
1262	<	Vector3d rs, scaled, dr;
1263	<	Vector3i whichCell;
1264	<	int cellIndex;
1376	>
1377		int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1378	<
1378	>
1379		#ifdef IS_MPI
1380		cellListRow_.resize(nCtot);
1381		cellListCol_.resize(nCtot);
1382		#else
1383		cellList_.resize(nCtot);
1384		#endif
1385	<
1385	>
1386		if (!doAllPairs) {
1387	+
1388		#ifdef IS_MPI
1389	<
1389	>
1390		for (int i = 0; i < nGroupsInRow_; i++) {
1391		rs = cgRowData.position[i];
1392
1393		// scaled positions relative to the box vectors
1394	<	scaled = invHmat * rs;
1394	>	scaled = invBox * rs;
1395
1396		// wrap the vector back into the unit box by subtracting integer box
1397		// numbers
1398		for (int j = 0; j < 3; j++) {
1399		scaled[j] -= roundMe(scaled[j]);
1400		scaled[j] += 0.5;
1401	+	// Handle the special case when an object is exactly on the
1402	+	// boundary (a scaled coordinate of 1.0 is the same as
1403	+	// scaled coordinate of 0.0)
1404	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1405		}
1406
1407		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1302 \| Line 1419 \| namespace OpenMD {
1419		rs = cgColData.position[i];
1420
1421		// scaled positions relative to the box vectors
1422	<	scaled = invHmat * rs;
1422	>	scaled = invBox * rs;
1423
1424		// wrap the vector back into the unit box by subtracting integer box
1425		// numbers
1426		for (int j = 0; j < 3; j++) {
1427		scaled[j] -= roundMe(scaled[j]);
1428		scaled[j] += 0.5;
1429	+	// Handle the special case when an object is exactly on the
1430	+	// boundary (a scaled coordinate of 1.0 is the same as
1431	+	// scaled coordinate of 0.0)
1432	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1433		}
1434
1435		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1322 \| Line 1443 \| namespace OpenMD {
1443		// add this cutoff group to the list of groups in this cell;
1444		cellListCol_[cellIndex].push_back(i);
1445		}
1446	<
1446	>
1447		#else
1448		for (int i = 0; i < nGroups_; i++) {
1449		rs = snap_->cgData.position[i];
1450
1451		// scaled positions relative to the box vectors
1452	<	scaled = invHmat * rs;
1452	>	scaled = invBox * rs;
1453
1454		// wrap the vector back into the unit box by subtracting integer box
1455		// numbers
1456		for (int j = 0; j < 3; j++) {
1457		scaled[j] -= roundMe(scaled[j]);
1458		scaled[j] += 0.5;
1459	+	// Handle the special case when an object is exactly on the
1460	+	// boundary (a scaled coordinate of 1.0 is the same as
1461	+	// scaled coordinate of 0.0)
1462	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1463		}
1464
1465		// find xyz-indices of cell that cutoffGroup is in.
1466	<	whichCell.x() = nCells_.x() * scaled.x();
1467	<	whichCell.y() = nCells_.y() * scaled.y();
1468	<	whichCell.z() = nCells_.z() * scaled.z();
1466	>	whichCell.x() = int(nCells_.x() * scaled.x());
1467	>	whichCell.y() = int(nCells_.y() * scaled.y());
1468	>	whichCell.z() = int(nCells_.z() * scaled.z());
1469
1470		// find single index of this cell:
1471		cellIndex = Vlinear(whichCell, nCells_);
#	Line 1351 \| Line 1476 \| namespace OpenMD {
1476
1477		#endif
1478
1354	–	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1355	–	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1356	–	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1357	–	Vector3i m1v(m1x, m1y, m1z);
1358	–	int m1 = Vlinear(m1v, nCells_);
1359	–
1360	–	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1361	–	os != cellOffsets_.end(); ++os) {
1362	–
1363	–	Vector3i m2v = m1v + (*os);
1364	–
1365	–
1366	–	if (m2v.x() >= nCells_.x()) {
1367	–	m2v.x() = 0;
1368	–	} else if (m2v.x() < 0) {
1369	–	m2v.x() = nCells_.x() - 1;
1370	–	}
1371	–
1372	–	if (m2v.y() >= nCells_.y()) {
1373	–	m2v.y() = 0;
1374	–	} else if (m2v.y() < 0) {
1375	–	m2v.y() = nCells_.y() - 1;
1376	–	}
1377	–
1378	–	if (m2v.z() >= nCells_.z()) {
1379	–	m2v.z() = 0;
1380	–	} else if (m2v.z() < 0) {
1381	–	m2v.z() = nCells_.z() - 1;
1382	–	}
1383	–
1384	–	int m2 = Vlinear (m2v, nCells_);
1385	–
1479		#ifdef IS_MPI
1480	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1481	<	j1 != cellListRow_[m1].end(); ++j1) {
1389	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1390	<	j2 != cellListCol_[m2].end(); ++j2) {
1391	<
1392	<	// In parallel, we need to visit all pairs of row
1393	<	// & column indicies and will divide labor in the
1394	<	// force evaluation later.
1395	<	dr = cgColData.position[(j2)] - cgRowData.position[(j1)];
1396	<	snap_->wrapVector(dr);
1397	<	cuts = getGroupCutoffs( (j1), (j2) );
1398	<	if (dr.lengthSquare() < cuts.third) {
1399	<	neighborList.push_back(make_pair((j1), (j2)));
1400	<	}
1401	<	}
1402	<	}
1480	>	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1481	>	rs = cgRowData.position[j1];
1482		#else
1404	–	for (vector<int>::iterator j1 = cellList_[m1].begin();
1405	–	j1 != cellList_[m1].end(); ++j1) {
1406	–	for (vector<int>::iterator j2 = cellList_[m2].begin();
1407	–	j2 != cellList_[m2].end(); ++j2) {
1408	–
1409	–	// Always do this if we're in different cells or if
1410	–	// we're in the same cell and the global index of
1411	–	// the j2 cutoff group is greater than or equal to
1412	–	// the j1 cutoff group. Note that Rappaport's code
1413	–	// has a "less than" conditional here, but that
1414	–	// deals with atom-by-atom computation. OpenMD
1415	–	// allows atoms within a single cutoff group to
1416	–	// interact with each other.
1483
1484	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1485	+	rs = snap_->cgData.position[j1];
1486	+	#endif
1487	+	point[j1] = len;
1488	+
1489	+	// scaled positions relative to the box vectors
1490	+	scaled = invBox * rs;
1491	+
1492	+	// wrap the vector back into the unit box by subtracting integer box
1493	+	// numbers
1494	+	for (int j = 0; j < 3; j++) {
1495	+	scaled[j] -= roundMe(scaled[j]);
1496	+	scaled[j] += 0.5;
1497	+	// Handle the special case when an object is exactly on the
1498	+	// boundary (a scaled coordinate of 1.0 is the same as
1499	+	// scaled coordinate of 0.0)
1500	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1501	+	}
1502	+
1503	+	// find xyz-indices of cell that cutoffGroup is in.
1504	+	whichCell.x() = nCells_.x() * scaled.x();
1505	+	whichCell.y() = nCells_.y() * scaled.y();
1506	+	whichCell.z() = nCells_.z() * scaled.z();
1507	+
1508	+	// find single index of this cell:
1509	+	int m1 = Vlinear(whichCell, nCells_);
1510
1511	+	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1512	+	os != cellOffsets_.end(); ++os) {
1513	+
1514	+	Vector3i m2v = whichCell + (*os);
1515
1516	<	if (m2 != m1 \|\| (j2) >= (j1) ) {
1517	<
1518	<	dr = snap_->cgData.position[(j2)] - snap_->cgData.position[(j1)];
1519	<	snap_->wrapVector(dr);
1520	<	cuts = getGroupCutoffs( (j1), (j2) );
1521	<	if (dr.lengthSquare() < cuts.third) {
1522	<	neighborList.push_back(make_pair((j1), (j2)));
1523	<	}
1524	<	}
1525	<	}
1516	>	if (m2v.x() >= nCells_.x()) {
1517	>	m2v.x() = 0;
1518	>	} else if (m2v.x() < 0) {
1519	>	m2v.x() = nCells_.x() - 1;
1520	>	}
1521	>
1522	>	if (m2v.y() >= nCells_.y()) {
1523	>	m2v.y() = 0;
1524	>	} else if (m2v.y() < 0) {
1525	>	m2v.y() = nCells_.y() - 1;
1526	>	}
1527	>
1528	>	if (m2v.z() >= nCells_.z()) {
1529	>	m2v.z() = 0;
1530	>	} else if (m2v.z() < 0) {
1531	>	m2v.z() = nCells_.z() - 1;
1532	>	}
1533	>	int m2 = Vlinear (m2v, nCells_);
1534	>	#ifdef IS_MPI
1535	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1536	>	j2 != cellListCol_[m2].end(); ++j2) {
1537	>
1538	>	// In parallel, we need to visit all pairs of row
1539	>	// & column indicies and will divide labor in the
1540	>	// force evaluation later.
1541	>	dr = cgColData.position[(*j2)] - rs;
1542	>	if (usePeriodicBoundaryConditions_) {
1543	>	snap_->wrapVector(dr);
1544	>	}
1545	>	if (dr.lengthSquare() < rListSq_) {
1546	>	neighborList.push_back( (*j2) );
1547	>	++len;
1548	>	}
1549	>	}
1550	>	#else
1551	>	for (vector<int>::iterator j2 = cellList_[m2].begin();
1552	>	j2 != cellList_[m2].end(); ++j2) {
1553	>
1554	>	// Always do this if we're in different cells or if
1555	>	// we're in the same cell and the global index of
1556	>	// the j2 cutoff group is greater than or equal to
1557	>	// the j1 cutoff group. Note that Rappaport's code
1558	>	// has a "less than" conditional here, but that
1559	>	// deals with atom-by-atom computation. OpenMD
1560	>	// allows atoms within a single cutoff group to
1561	>	// interact with each other.
1562	>
1563	>	if ( (*j2) >= j1 ) {
1564	>
1565	>	dr = snap_->cgData.position[(*j2)] - rs;
1566	>	if (usePeriodicBoundaryConditions_) {
1567	>	snap_->wrapVector(dr);
1568		}
1569	<	#endif
1569	>	if ( dr.lengthSquare() < rListSq_) {
1570	>	neighborList.push_back( (*j2) );
1571	>	++len;
1572	>	}
1573		}
1574	<	}
1574	>	}
1575	>	#endif
1576		}
1577	<	}
1577	>	}
1578		} else {
1579		// branch to do all cutoff group pairs
1580		#ifdef IS_MPI
1581		for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1582	+	point[j1] = len;
1583	+	rs = cgRowData.position[j1];
1584		for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1585	<	dr = cgColData.position[j2] - cgRowData.position[j1];
1586	<	snap_->wrapVector(dr);
1587	<	cuts = getGroupCutoffs( j1, j2 );
1444	<	if (dr.lengthSquare() < cuts.third) {
1445	<	neighborList.push_back(make_pair(j1, j2));
1585	>	dr = cgColData.position[j2] - rs;
1586	>	if (usePeriodicBoundaryConditions_) {
1587	>	snap_->wrapVector(dr);
1588		}
1589	+	if (dr.lengthSquare() < rListSq_) {
1590	+	neighborList.push_back( j2 );
1591	+	++len;
1592	+	}
1593		}
1594		}
1595		#else
1596		// include all groups here.
1597		for (int j1 = 0; j1 < nGroups_; j1++) {
1598	+	point[j1] = len;
1599	+	rs = snap_->cgData.position[j1];
1600		// include self group interactions j2 == j1
1601		for (int j2 = j1; j2 < nGroups_; j2++) {
1602	<	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1603	<	snap_->wrapVector(dr);
1604	<	cuts = getGroupCutoffs( j1, j2 );
1457	<	if (dr.lengthSquare() < cuts.third) {
1458	<	neighborList.push_back(make_pair(j1, j2));
1602	>	dr = snap_->cgData.position[j2] - rs;
1603	>	if (usePeriodicBoundaryConditions_) {
1604	>	snap_->wrapVector(dr);
1605		}
1606	+	if (dr.lengthSquare() < rListSq_) {
1607	+	neighborList.push_back( j2 );
1608	+	++len;
1609	+	}
1610		}
1611		}
1612		#endif
1613		}
1614	<
1614	>
1615	>	#ifdef IS_MPI
1616	>	point[nGroupsInRow_] = len;
1617	>	#else
1618	>	point[nGroups_] = len;
1619	>	#endif
1620	>
1621		// save the local cutoff group positions for the check that is
1622		// done on each loop:
1623		saved_CG_positions_.clear();
1624	+	saved_CG_positions_.reserve(nGroups_);
1625		for (int i = 0; i < nGroups_; i++)
1626		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1470	–
1471	–	return neighborList;
1627		}
1628		} //end namespace OpenMD

Diff Legend

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

Comparing: branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1723 by gezelter, Thu May 24 20:59:54 2012 UTC vs. trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 2057 by gezelter, Tue Mar 3 15:22:26 2015 UTC

Diff Legend

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1723 by gezelter, Thu May 24 20:59:54 2012 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 2057 by gezelter, Tue Mar 3 15:22:26 2015 UTC