ViewVC Help

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

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

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1688 by gezelter, Wed Mar 14 17:56:01 2012 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 2077 by gezelter, Mon Mar 9 17:10:26 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 95 | Line 91 | namespace OpenMD {
91		storageLayout_ = sman_->getStorageLayout();
92		ff_ = info_->getForceField();
93		nLocal_ = snap_->getNumberOfAtoms();
94	<
94	>
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 109 | Line 106 | namespace OpenMD {
106		PairList* oneTwo = info_->getOneTwoInteractions();
107		PairList* oneThree = info_->getOneThreeInteractions();
108		PairList* oneFour = info_->getOneFourInteractions();
109	<
109	>
110	>	if (needVelocities_)
111	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
112	>	DataStorage::dslVelocity);
113	>	else
114	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
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 145 | Line 148 | namespace OpenMD {
148		cgRowData.resize(nGroupsInRow_);
149		cgRowData.setStorageLayout(DataStorage::dslPosition);
150		cgColData.resize(nGroupsInCol_);
151	<	cgColData.setStorageLayout(DataStorage::dslPosition);
152	<
151	>	if (needVelocities_)
152	>	// we only need column velocities if we need them.
153	>	cgColData.setStorageLayout(DataStorage::dslPosition |
154	>	DataStorage::dslVelocity);
155	>	else
156	>	cgColData.setStorageLayout(DataStorage::dslPosition);
157	>
158		identsRow.resize(nAtomsInRow_);
159		identsCol.resize(nAtomsInCol_);
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 165 | 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 285 | Line 302 | namespace OpenMD {
302		groupList_[i].push_back(j);
303		}
304		}
305	<	}
289	<
290	<
291	<	createGtypeCutoffMap();
292	<
305	>	}
306		}
294	–
295	–	void ForceMatrixDecomposition::createGtypeCutoffMap() {
307
297	–	RealType tol = 1e-6;
298	–	largestRcut_ = 0.0;
299	–	RealType rc;
300	–	int atid;
301	–	set<AtomType*> atypes = info_->getSimulatedAtomTypes();
302	–
303	–	map<int, RealType> atypeCutoff;
304	–
305	–	for (set<AtomType*>::iterator at = atypes.begin();
306	–	at != atypes.end(); ++at){
307	–	atid = (*at)->getIdent();
308	–	if (userChoseCutoff_)
309	–	atypeCutoff[atid] = userCutoff_;
310	–	else
311	–	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
312	–	}
313	–
314	–	vector<RealType> gTypeCutoffs;
315	–	// first we do a single loop over the cutoff groups to find the
316	–	// largest cutoff for any atypes present in this group.
317	–	#ifdef IS_MPI
318	–	vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
319	–	groupRowToGtype.resize(nGroupsInRow_);
320	–	for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
321	–	vector<int> atomListRow = getAtomsInGroupRow(cg1);
322	–	for (vector<int>::iterator ia = atomListRow.begin();
323	–	ia != atomListRow.end(); ++ia) {
324	–	int atom1 = (*ia);
325	–	atid = identsRow[atom1];
326	–	if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
327	–	groupCutoffRow[cg1] = atypeCutoff[atid];
328	–	}
329	–	}
330	–
331	–	bool gTypeFound = false;
332	–	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
333	–	if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
334	–	groupRowToGtype[cg1] = gt;
335	–	gTypeFound = true;
336	–	}
337	–	}
338	–	if (!gTypeFound) {
339	–	gTypeCutoffs.push_back( groupCutoffRow[cg1] );
340	–	groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
341	–	}
342	–
343	–	}
344	–	vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
345	–	groupColToGtype.resize(nGroupsInCol_);
346	–	for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
347	–	vector<int> atomListCol = getAtomsInGroupColumn(cg2);
348	–	for (vector<int>::iterator jb = atomListCol.begin();
349	–	jb != atomListCol.end(); ++jb) {
350	–	int atom2 = (*jb);
351	–	atid = identsCol[atom2];
352	–	if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
353	–	groupCutoffCol[cg2] = atypeCutoff[atid];
354	–	}
355	–	}
356	–	bool gTypeFound = false;
357	–	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
358	–	if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
359	–	groupColToGtype[cg2] = gt;
360	–	gTypeFound = true;
361	–	}
362	–	}
363	–	if (!gTypeFound) {
364	–	gTypeCutoffs.push_back( groupCutoffCol[cg2] );
365	–	groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
366	–	}
367	–	}
368	–	#else
369	–
370	–	vector<RealType> groupCutoff(nGroups_, 0.0);
371	–	groupToGtype.resize(nGroups_);
372	–	for (int cg1 = 0; cg1 < nGroups_; cg1++) {
373	–	groupCutoff[cg1] = 0.0;
374	–	vector<int> atomList = getAtomsInGroupRow(cg1);
375	–	for (vector<int>::iterator ia = atomList.begin();
376	–	ia != atomList.end(); ++ia) {
377	–	int atom1 = (*ia);
378	–	atid = idents[atom1];
379	–	if (atypeCutoff[atid] > groupCutoff[cg1])
380	–	groupCutoff[cg1] = atypeCutoff[atid];
381	–	}
382	–
383	–	bool gTypeFound = false;
384	–	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
385	–	if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
386	–	groupToGtype[cg1] = gt;
387	–	gTypeFound = true;
388	–	}
389	–	}
390	–	if (!gTypeFound) {
391	–	gTypeCutoffs.push_back( groupCutoff[cg1] );
392	–	groupToGtype[cg1] = gTypeCutoffs.size() - 1;
393	–	}
394	–	}
395	–	#endif
396	–
397	–	// Now we find the maximum group cutoff value present in the simulation
398	–
399	–	RealType groupMax = *max_element(gTypeCutoffs.begin(),
400	–	gTypeCutoffs.end());
401	–
402	–	#ifdef IS_MPI
403	–	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
404	–	MPI::MAX);
405	–	#endif
406	–
407	–	RealType tradRcut = groupMax;
408	–
409	–	for (int i = 0; i < gTypeCutoffs.size();  i++) {
410	–	for (int j = 0; j < gTypeCutoffs.size();  j++) {
411	–	RealType thisRcut;
412	–	switch(cutoffPolicy_) {
413	–	case TRADITIONAL:
414	–	thisRcut = tradRcut;
415	–	break;
416	–	case MIX:
417	–	thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
418	–	break;
419	–	case MAX:
420	–	thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
421	–	break;
422	–	default:
423	–	sprintf(painCave.errMsg,
424	–	"ForceMatrixDecomposition::createGtypeCutoffMap "
425	–	"hit an unknown cutoff policy!\n");
426	–	painCave.severity = OPENMD_ERROR;
427	–	painCave.isFatal = 1;
428	–	simError();
429	–	break;
430	–	}
431	–
432	–	pair<int,int> key = make_pair(i,j);
433	–	gTypeCutoffMap[key].first = thisRcut;
434	–	if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
435	–	gTypeCutoffMap[key].second = thisRcut*thisRcut;
436	–	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
437	–	// sanity check
438	–
439	–	if (userChoseCutoff_) {
440	–	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
441	–	sprintf(painCave.errMsg,
442	–	"ForceMatrixDecomposition::createGtypeCutoffMap "
443	–	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
444	–	painCave.severity = OPENMD_ERROR;
445	–	painCave.isFatal = 1;
446	–	simError();
447	–	}
448	–	}
449	–	}
450	–	}
451	–	}
452	–
453	–
454	–	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
455	–	int i, j;
456	–	#ifdef IS_MPI
457	–	i = groupRowToGtype[cg1];
458	–	j = groupColToGtype[cg2];
459	–	#else
460	–	i = groupToGtype[cg1];
461	–	j = groupToGtype[cg2];
462	–	#endif
463	–	return gTypeCutoffMap[make_pair(i,j)];
464	–	}
465	–
308		int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
309	<	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
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 492 | 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 525 | Line 375 | namespace OpenMD {
375		atomColData.skippedCharge.end(), 0.0);
376		}
377
378	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
379	+	fill(atomRowData.flucQFrc.begin(),
380	+	atomRowData.flucQFrc.end(), 0.0);
381	+	fill(atomColData.flucQFrc.begin(),
382	+	atomColData.flucQFrc.end(), 0.0);
383	+	}
384	+
385	+	if (storageLayout_ & DataStorage::dslElectricField) {
386	+	fill(atomRowData.electricField.begin(),
387	+	atomRowData.electricField.end(), V3Zero);
388	+	fill(atomColData.electricField.begin(),
389	+	atomColData.electricField.end(), V3Zero);
390	+	}
391	+
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
400		// even in parallel, we need to zero out the local arrays:
401
#	Line 537 | Line 408 | namespace OpenMD {
408		fill(snap_->atomData.density.begin(),
409		snap_->atomData.density.end(), 0.0);
410		}
411	+
412		if (storageLayout_ & DataStorage::dslFunctional) {
413		fill(snap_->atomData.functional.begin(),
414		snap_->atomData.functional.end(), 0.0);
415		}
416	+
417		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
418		fill(snap_->atomData.functionalDerivative.begin(),
419		snap_->atomData.functionalDerivative.end(), 0.0);
420		}
421	+
422		if (storageLayout_ & DataStorage::dslSkippedCharge) {
423		fill(snap_->atomData.skippedCharge.begin(),
424		snap_->atomData.skippedCharge.end(), 0.0);
425		}
426	<
426	>
427	>	if (storageLayout_ & DataStorage::dslElectricField) {
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	+
440	+	#ifdef IS_MPI
441	+
442		snap_ = sman_->getCurrentSnapshot();
443		storageLayout_ = sman_->getStorageLayout();
559	–	#ifdef IS_MPI
444
445	+	bool needsCG = true;
446	+	if(info_->getNCutoffGroups() != info_->getNAtoms())
447	+	needsCG = false;
448	+
449		// gather up the atomic positions
450		AtomPlanVectorRow->gather(snap_->atomData.position,
451		atomRowData.position);
#	Line 566 | Line 454 | namespace OpenMD {
454
455		// gather up the cutoff group positions
456
457	<	cgPlanVectorRow->gather(snap_->cgData.position,
458	<	cgRowData.position);
457	>	if (needsCG) {
458	>	cgPlanVectorRow->gather(snap_->cgData.position,
459	>	cgRowData.position);
460	>
461	>	cgPlanVectorColumn->gather(snap_->cgData.position,
462	>	cgColData.position);
463	>	}
464
572	–	cgPlanVectorColumn->gather(snap_->cgData.position,
573	–	cgColData.position);
465
466	+	if (needVelocities_) {
467	+	// gather up the atomic velocities
468	+	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
469	+	atomColData.velocity);
470	+
471	+	if (needsCG) {
472	+	cgPlanVectorColumn->gather(snap_->cgData.velocity,
473	+	cgColData.velocity);
474	+	}
475	+	}
476	+
477
478		// if needed, gather the atomic rotation matrices
479		if (storageLayout_ & DataStorage::dslAmat) {
#	Line 580 | Line 482 | namespace OpenMD {
482		AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
483		atomColData.aMat);
484		}
485	<
486	<	// if needed, gather the atomic eletrostatic frames
487	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
488	<	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
489	<	atomRowData.electroFrame);
490	<	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
491	<	atomColData.electroFrame);
485	>
486	>	// if needed, gather the atomic eletrostatic information
487	>	if (storageLayout_ & DataStorage::dslDipole) {
488	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
489	>	atomRowData.dipole);
490	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
491	>	atomColData.dipole);
492		}
493
494	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
495	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
496	+	atomRowData.quadrupole);
497	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
498	+	atomColData.quadrupole);
499	+	}
500	+
501	+	// if needed, gather the atomic fluctuating charge values
502	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
503	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
504	+	atomRowData.flucQPos);
505	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
506	+	atomColData.flucQPos);
507	+	}
508	+
509		#endif
510		}
511
#	Line 596 | Line 513 | namespace OpenMD {
513		* data structures.
514		*/
515		void ForceMatrixDecomposition::collectIntermediateData() {
516	+	#ifdef IS_MPI
517	+
518		snap_ = sman_->getCurrentSnapshot();
519		storageLayout_ = sman_->getStorageLayout();
520	<	#ifdef IS_MPI
602	<
520	>
521		if (storageLayout_ & DataStorage::dslDensity) {
522
523		AtomPlanRealRow->scatter(atomRowData.density,
#	Line 611 | Line 529 | namespace OpenMD {
529		for (int i = 0; i < n; i++)
530		snap_->atomData.density[i] += rho_tmp[i];
531		}
532	+
533	+	// this isn't necessary if we don't have polarizable atoms, but
534	+	// we'll leave it here for now.
535	+	if (storageLayout_ & DataStorage::dslElectricField) {
536	+
537	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
538	+	snap_->atomData.electricField);
539	+
540	+	int n = snap_->atomData.electricField.size();
541	+	vector<Vector3d> field_tmp(n, V3Zero);
542	+	AtomPlanVectorColumn->scatter(atomColData.electricField,
543	+	field_tmp);
544	+	for (int i = 0; i < n; i++)
545	+	snap_->atomData.electricField[i] += field_tmp[i];
546	+	}
547		#endif
548		}
549
#	Line 619 | Line 552 | namespace OpenMD {
552		* row and column-indexed data structures
553		*/
554		void ForceMatrixDecomposition::distributeIntermediateData() {
555	+	#ifdef IS_MPI
556		snap_ = sman_->getCurrentSnapshot();
557		storageLayout_ = sman_->getStorageLayout();
558	<	#ifdef IS_MPI
558	>
559		if (storageLayout_ & DataStorage::dslFunctional) {
560		AtomPlanRealRow->gather(snap_->atomData.functional,
561		atomRowData.functional);
#	Line 640 | Line 574 | namespace OpenMD {
574
575
576		void ForceMatrixDecomposition::collectData() {
577	+	#ifdef IS_MPI
578		snap_ = sman_->getCurrentSnapshot();
579		storageLayout_ = sman_->getStorageLayout();
580	<	#ifdef IS_MPI
580	>
581		int n = snap_->atomData.force.size();
582		vector<Vector3d> frc_tmp(n, V3Zero);
583
#	Line 690 | Line 625 | namespace OpenMD {
625
626		}
627
628	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
629	+
630	+	int nq = snap_->atomData.flucQFrc.size();
631	+	vector<RealType> fqfrc_tmp(nq, 0.0);
632	+
633	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
634	+	for (int i = 0; i < nq; i++) {
635	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
636	+	fqfrc_tmp[i] = 0.0;
637	+	}
638	+
639	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
640	+	for (int i = 0; i < nq; i++)
641	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
642	+
643	+	}
644	+
645	+	if (storageLayout_ & DataStorage::dslElectricField) {
646	+
647	+	int nef = snap_->atomData.electricField.size();
648	+	vector<Vector3d> efield_tmp(nef, V3Zero);
649	+
650	+	AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
651	+	for (int i = 0; i < nef; i++) {
652	+	snap_->atomData.electricField[i] += efield_tmp[i];
653	+	efield_tmp[i] = 0.0;
654	+	}
655	+
656	+	AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
657	+	for (int i = 0; i < nef; i++)
658	+	snap_->atomData.electricField[i] += efield_tmp[i];
659	+	}
660	+
661	+	if (storageLayout_ & DataStorage::dslSitePotential) {
662	+
663	+	int nsp = snap_->atomData.sitePotential.size();
664	+	vector<RealType> sp_tmp(nsp, 0.0);
665	+
666	+	AtomPlanRealRow->scatter(atomRowData.sitePotential, sp_tmp);
667	+	for (int i = 0; i < nsp; i++) {
668	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
669	+	sp_tmp[i] = 0.0;
670	+	}
671	+
672	+	AtomPlanRealColumn->scatter(atomColData.sitePotential, sp_tmp);
673	+	for (int i = 0; i < nsp; i++)
674	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
675	+	}
676	+
677		nLocal_ = snap_->getNumberOfAtoms();
678
679		vector<potVec> pot_temp(nLocal_,
680		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
681	+	vector<potVec> expot_temp(nLocal_,
682	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
683
684		// scatter/gather pot_row into the members of my column
685
686		AtomPlanPotRow->scatter(pot_row, pot_temp);
687	+	AtomPlanPotRow->scatter(expot_row, expot_temp);
688
689	<	for (int ii = 0;  ii < pot_temp.size(); ii++ )
689	>	for (std::size_t ii = 0;  ii < pot_temp.size(); ii++ )
690		pairwisePot += pot_temp[ii];
691	+
692	+	for (std::size_t ii = 0;  ii < expot_temp.size(); ii++ )
693	+	excludedPot += expot_temp[ii];
694
695	+	if (storageLayout_ & DataStorage::dslParticlePot) {
696	+	// This is the pairwise contribution to the particle pot.  The
697	+	// embedding contribution is added in each of the low level
698	+	// non-bonded routines.  In single processor, this is done in
699	+	// unpackInteractionData, not in collectData.
700	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
701	+	for (int i = 0; i < nLocal_; i++) {
702	+	// factor of two is because the total potential terms are divided
703	+	// by 2 in parallel due to row/ column scatter
704	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
705	+	}
706	+	}
707	+	}
708	+
709		fill(pot_temp.begin(), pot_temp.end(),
710		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
711	+	fill(expot_temp.begin(), expot_temp.end(),
712	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
713
714		AtomPlanPotColumn->scatter(pot_col, pot_temp);
715	+	AtomPlanPotColumn->scatter(expot_col, expot_temp);
716
717	<	for (int ii = 0;  ii < pot_temp.size(); ii++ )
717	>	for (std::size_t ii = 0;  ii < pot_temp.size(); ii++ )
718		pairwisePot += pot_temp[ii];
719	+
720	+	for (std::size_t ii = 0;  ii < expot_temp.size(); ii++ )
721	+	excludedPot += expot_temp[ii];
722
723	+	if (storageLayout_ & DataStorage::dslParticlePot) {
724	+	// This is the pairwise contribution to the particle pot.  The
725	+	// embedding contribution is added in each of the low level
726	+	// non-bonded routines.  In single processor, this is done in
727	+	// unpackInteractionData, not in collectData.
728	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
729	+	for (int i = 0; i < nLocal_; i++) {
730	+	// factor of two is because the total potential terms are divided
731	+	// by 2 in parallel due to row/ column scatter
732	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
733	+	}
734	+	}
735	+	}
736	+
737	+	if (storageLayout_ & DataStorage::dslParticlePot) {
738	+	int npp = snap_->atomData.particlePot.size();
739	+	vector<RealType> ppot_temp(npp, 0.0);
740	+
741	+	// This is the direct or embedding contribution to the particle
742	+	// pot.
743	+
744	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
745	+	for (int i = 0; i < npp; i++) {
746	+	snap_->atomData.particlePot[i] += ppot_temp[i];
747	+	}
748	+
749	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
750	+
751	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
752	+	for (int i = 0; i < npp; i++) {
753	+	snap_->atomData.particlePot[i] += ppot_temp[i];
754	+	}
755	+	}
756	+
757		for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
758		RealType ploc1 = pairwisePot[ii];
759		RealType ploc2 = 0.0;
760	<	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
760	>	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
761		pairwisePot[ii] = ploc2;
762		}
763
764		for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
765	<	RealType ploc1 = embeddingPot[ii];
765	>	RealType ploc1 = excludedPot[ii];
766		RealType ploc2 = 0.0;
767	<	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
768	<	embeddingPot[ii] = ploc2;
767	>	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
768	>	excludedPot[ii] = ploc2;
769		}
770
771	+	// Here be dragons.
772	+	MPI_Comm col = colComm.getComm();
773	+
774	+	MPI_Allreduce(MPI_IN_PLACE,
775	+	&snap_->frameData.conductiveHeatFlux[0], 3,
776	+	MPI_REALTYPE, MPI_SUM, col);
777	+
778	+
779		#endif
780
781		}
782
783	<	int ForceMatrixDecomposition::getNAtomsInRow() {
783	>	/**
784	>	* Collects information obtained during the post-pair (and embedding
785	>	* functional) loops onto local data structures.
786	>	*/
787	>	void ForceMatrixDecomposition::collectSelfData() {
788	>
789		#ifdef IS_MPI
790	+	snap_ = sman_->getCurrentSnapshot();
791	+	storageLayout_ = sman_->getStorageLayout();
792	+
793	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
794	+	RealType ploc1 = embeddingPot[ii];
795	+	RealType ploc2 = 0.0;
796	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
797	+	embeddingPot[ii] = ploc2;
798	+	}
799	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
800	+	RealType ploc1 = excludedSelfPot[ii];
801	+	RealType ploc2 = 0.0;
802	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
803	+	excludedSelfPot[ii] = ploc2;
804	+	}
805	+	#endif
806	+
807	+	}
808	+
809	+	int& ForceMatrixDecomposition::getNAtomsInRow() {
810	+	#ifdef IS_MPI
811		return nAtomsInRow_;
812		#else
813		return nLocal_;
#	Line 739 | Line 817 | namespace OpenMD {
817		/**
818		* returns the list of atoms belonging to this group.
819		*/
820	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
820	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
821		#ifdef IS_MPI
822		return groupListRow_[cg1];
823		#else
#	Line 747 | Line 825 | namespace OpenMD {
825		#endif
826		}
827
828	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
828	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
829		#ifdef IS_MPI
830		return groupListCol_[cg2];
831		#else
#	Line 755 | Line 833 | namespace OpenMD {
833		#endif
834		}
835
836	<	Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
836	>	Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1,
837	>	int cg2){
838	>
839		Vector3d d;
760	–
840		#ifdef IS_MPI
841		d = cgColData.position[cg2] - cgRowData.position[cg1];
842		#else
843		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
844		#endif
845
846	<	snap_->wrapVector(d);
846	>	if (usePeriodicBoundaryConditions_) {
847	>	snap_->wrapVector(d);
848	>	}
849		return d;
850		}
851
852	+	Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
853	+	#ifdef IS_MPI
854	+	return cgColData.velocity[cg2];
855	+	#else
856	+	return snap_->cgData.velocity[cg2];
857	+	#endif
858	+	}
859
860	<	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
860	>	Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
861	>	#ifdef IS_MPI
862	>	return atomColData.velocity[atom2];
863	>	#else
864	>	return snap_->atomData.velocity[atom2];
865	>	#endif
866	>	}
867
868	+
869	+	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1,
870	+	int cg1) {
871		Vector3d d;
872
873		#ifdef IS_MPI
#	Line 778 | Line 875 | namespace OpenMD {
875		#else
876		d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
877		#endif
878	<
879	<	snap_->wrapVector(d);
878	>	if (usePeriodicBoundaryConditions_) {
879	>	snap_->wrapVector(d);
880	>	}
881		return d;
882		}
883
884	<	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
884	>	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2,
885	>	int cg2) {
886		Vector3d d;
887
888		#ifdef IS_MPI
#	Line 791 | Line 890 | namespace OpenMD {
890		#else
891		d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
892		#endif
893	<
894	<	snap_->wrapVector(d);
893	>	if (usePeriodicBoundaryConditions_) {
894	>	snap_->wrapVector(d);
895	>	}
896		return d;
897		}
898
899	<	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
899	>	RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
900		#ifdef IS_MPI
901		return massFactorsRow[atom1];
902		#else
#	Line 804 | Line 904 | namespace OpenMD {
904		#endif
905		}
906
907	<	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
907	>	RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
908		#ifdef IS_MPI
909		return massFactorsCol[atom2];
910		#else
#	Line 813 | Line 913 | namespace OpenMD {
913
914		}
915
916	<	Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
916	>	Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1,
917	>	int atom2){
918		Vector3d d;
919
920		#ifdef IS_MPI
#	Line 821 | Line 922 | namespace OpenMD {
922		#else
923		d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
924		#endif
925	<
926	<	snap_->wrapVector(d);
925	>	if (usePeriodicBoundaryConditions_) {
926	>	snap_->wrapVector(d);
927	>	}
928		return d;
929		}
930
931	<	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
931	>	vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
932		return excludesForAtom[atom1];
933		}
934
#	Line 834 | Line 936 | namespace OpenMD {
936		* We need to exclude some overcounted interactions that result from
937		* the parallel decomposition.
938		*/
939	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
939	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2,
940	>	int cg1, int cg2) {
941		int unique_id_1, unique_id_2;
942
943		#ifdef IS_MPI
944		// in MPI, we have to look up the unique IDs for each atom
945		unique_id_1 = AtomRowToGlobal[atom1];
946		unique_id_2 = AtomColToGlobal[atom2];
947	+	// group1 = cgRowToGlobal[cg1];
948	+	// group2 = cgColToGlobal[cg2];
949		#else
950		unique_id_1 = AtomLocalToGlobal[atom1];
951		unique_id_2 = AtomLocalToGlobal[atom2];
952	+	int group1 = cgLocalToGlobal[cg1];
953	+	int group2 = cgLocalToGlobal[cg2];
954		#endif
955
956		if (unique_id_1 == unique_id_2) return true;
#	Line 855 | Line 962 | namespace OpenMD {
962		} else {
963		if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
964		}
965	+	#endif
966	+
967	+	#ifndef IS_MPI
968	+	if (group1 == group2) {
969	+	if (unique_id_1 < unique_id_2) return true;
970	+	}
971		#endif
972
973		return false;
#	Line 901 | Line 1014 | namespace OpenMD {
1014
1015		// filling interaction blocks with pointers
1016		void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1017	<	int atom1, int atom2) {
1017	>	int atom1, int atom2,
1018	>	bool newAtom1) {
1019
1020		idat.excluded = excludeAtomPair(atom1, atom2);
1021	<
1021	>
1022	>	if (newAtom1) {
1023	>
1024		#ifdef IS_MPI
1025	<	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1026	<	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1027	<	//                         ff_->getAtomType(identsCol[atom2]) );
1028	<
1025	>	idat.atid1 = identsRow[atom1];
1026	>	idat.atid2 = identsCol[atom2];
1027	>
1028	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1029	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1030	>	} else {
1031	>	idat.sameRegion = false;
1032	>	}
1033	>
1034	>	if (storageLayout_ & DataStorage::dslAmat) {
1035	>	idat.A1 = &(atomRowData.aMat[atom1]);
1036	>	idat.A2 = &(atomColData.aMat[atom2]);
1037	>	}
1038	>
1039	>	if (storageLayout_ & DataStorage::dslTorque) {
1040	>	idat.t1 = &(atomRowData.torque[atom1]);
1041	>	idat.t2 = &(atomColData.torque[atom2]);
1042	>	}
1043	>
1044	>	if (storageLayout_ & DataStorage::dslDipole) {
1045	>	idat.dipole1 = &(atomRowData.dipole[atom1]);
1046	>	idat.dipole2 = &(atomColData.dipole[atom2]);
1047	>	}
1048	>
1049	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1050	>	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1051	>	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1052	>	}
1053	>
1054	>	if (storageLayout_ & DataStorage::dslDensity) {
1055	>	idat.rho1 = &(atomRowData.density[atom1]);
1056	>	idat.rho2 = &(atomColData.density[atom2]);
1057	>	}
1058	>
1059	>	if (storageLayout_ & DataStorage::dslFunctional) {
1060	>	idat.frho1 = &(atomRowData.functional[atom1]);
1061	>	idat.frho2 = &(atomColData.functional[atom2]);
1062	>	}
1063	>
1064	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1065	>	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1066	>	idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1067	>	}
1068	>
1069	>	if (storageLayout_ & DataStorage::dslParticlePot) {
1070	>	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1071	>	idat.particlePot2 = &(atomColData.particlePot[atom2]);
1072	>	}
1073	>
1074	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1075	>	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1076	>	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1077	>	}
1078	>
1079	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1080	>	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1081	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1082	>	}
1083	>
1084	>	#else
1085	>
1086	>	idat.atid1 = idents[atom1];
1087	>	idat.atid2 = idents[atom2];
1088	>
1089	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1090	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1091	>	} else {
1092	>	idat.sameRegion = false;
1093	>	}
1094	>
1095	>	if (storageLayout_ & DataStorage::dslAmat) {
1096	>	idat.A1 = &(snap_->atomData.aMat[atom1]);
1097	>	idat.A2 = &(snap_->atomData.aMat[atom2]);
1098	>	}
1099	>
1100	>	if (storageLayout_ & DataStorage::dslTorque) {
1101	>	idat.t1 = &(snap_->atomData.torque[atom1]);
1102	>	idat.t2 = &(snap_->atomData.torque[atom2]);
1103	>	}
1104	>
1105	>	if (storageLayout_ & DataStorage::dslDipole) {
1106	>	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1107	>	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1108	>	}
1109	>
1110	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1111	>	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1112	>	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1113	>	}
1114	>
1115	>	if (storageLayout_ & DataStorage::dslDensity) {
1116	>	idat.rho1 = &(snap_->atomData.density[atom1]);
1117	>	idat.rho2 = &(snap_->atomData.density[atom2]);
1118	>	}
1119	>
1120	>	if (storageLayout_ & DataStorage::dslFunctional) {
1121	>	idat.frho1 = &(snap_->atomData.functional[atom1]);
1122	>	idat.frho2 = &(snap_->atomData.functional[atom2]);
1123	>	}
1124	>
1125	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1126	>	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1127	>	idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1128	>	}
1129	>
1130	>	if (storageLayout_ & DataStorage::dslParticlePot) {
1131	>	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1132	>	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1133	>	}
1134	>
1135	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1136	>	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1137	>	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1138	>	}
1139	>
1140	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1141	>	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1142	>	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1143	>	}
1144	>	#endif
1145	>
1146	>	} else {
1147	>	// atom1 is not new, so don't bother updating properties of that atom:
1148	>	#ifdef IS_MPI
1149	>	idat.atid2 = identsCol[atom2];
1150	>
1151	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1152	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1153	>	} else {
1154	>	idat.sameRegion = false;
1155	>	}
1156	>
1157		if (storageLayout_ & DataStorage::dslAmat) {
914	–	idat.A1 = &(atomRowData.aMat[atom1]);
1158		idat.A2 = &(atomColData.aMat[atom2]);
1159		}
1160
918	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
919	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
920	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
921	–	}
922	–
1161		if (storageLayout_ & DataStorage::dslTorque) {
924	–	idat.t1 = &(atomRowData.torque[atom1]);
1162		idat.t2 = &(atomColData.torque[atom2]);
1163		}
1164
1165	+	if (storageLayout_ & DataStorage::dslDipole) {
1166	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1167	+	}
1168	+
1169	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1170	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1171	+	}
1172	+
1173		if (storageLayout_ & DataStorage::dslDensity) {
929	–	idat.rho1 = &(atomRowData.density[atom1]);
1174		idat.rho2 = &(atomColData.density[atom2]);
1175		}
1176
1177		if (storageLayout_ & DataStorage::dslFunctional) {
934	–	idat.frho1 = &(atomRowData.functional[atom1]);
1178		idat.frho2 = &(atomColData.functional[atom2]);
1179		}
1180
1181		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
939	–	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1182		idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1183		}
1184
1185		if (storageLayout_ & DataStorage::dslParticlePot) {
944	–	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1186		idat.particlePot2 = &(atomColData.particlePot[atom2]);
1187		}
1188
1189		if (storageLayout_ & DataStorage::dslSkippedCharge) {
949	–	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1190		idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1191		}
1192
1193	<	#else
1194	<
1193	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1194	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1195	>	}
1196
1197	<	// cerr << "atoms = " << atom1 << " " << atom2 << "\n";
1198	<	// cerr << "pos1 = " << snap_->atomData.position[atom1] << "\n";
958	<	// cerr << "pos2 = " << snap_->atomData.position[atom2] << "\n";
1197	>	#else
1198	>	idat.atid2 = idents[atom2];
1199
1200	<	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1201	<	//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1202	<	//                         ff_->getAtomType(idents[atom2]) );
1200	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1201	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1202	>	} else {
1203	>	idat.sameRegion = false;
1204	>	}
1205
1206		if (storageLayout_ & DataStorage::dslAmat) {
965	–	idat.A1 = &(snap_->atomData.aMat[atom1]);
1207		idat.A2 = &(snap_->atomData.aMat[atom2]);
1208		}
1209
969	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
970	–	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
971	–	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
972	–	}
973	–
1210		if (storageLayout_ & DataStorage::dslTorque) {
975	–	idat.t1 = &(snap_->atomData.torque[atom1]);
1211		idat.t2 = &(snap_->atomData.torque[atom2]);
1212		}
1213
1214	+	if (storageLayout_ & DataStorage::dslDipole) {
1215	+	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1216	+	}
1217	+
1218	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1219	+	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1220	+	}
1221	+
1222		if (storageLayout_ & DataStorage::dslDensity) {
980	–	idat.rho1 = &(snap_->atomData.density[atom1]);
1223		idat.rho2 = &(snap_->atomData.density[atom2]);
1224		}
1225
1226		if (storageLayout_ & DataStorage::dslFunctional) {
985	–	idat.frho1 = &(snap_->atomData.functional[atom1]);
1227		idat.frho2 = &(snap_->atomData.functional[atom2]);
1228		}
1229
1230		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
990	–	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1231		idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1232		}
1233
1234		if (storageLayout_ & DataStorage::dslParticlePot) {
995	–	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1235		idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1236		}
1237
1238		if (storageLayout_ & DataStorage::dslSkippedCharge) {
1000	–	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1239		idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1240		}
1241	+
1242	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1243	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1244	+	}
1245	+
1246		#endif
1247	+	}
1248		}
1005	–
1249
1250	<	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1250	>	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat,
1251	>	int atom1, int atom2) {
1252		#ifdef IS_MPI
1253		pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1254		pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1255	+	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1256	+	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1257
1258		atomRowData.force[atom1] += *(idat.f1);
1259		atomColData.force[atom2] -= *(idat.f1);
1260	+
1261	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1262	+	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1263	+	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1264	+	}
1265	+
1266	+	if (storageLayout_ & DataStorage::dslElectricField) {
1267	+	atomRowData.electricField[atom1] += *(idat.eField1);
1268	+	atomColData.electricField[atom2] += *(idat.eField2);
1269	+	}
1270	+
1271	+	if (storageLayout_ & DataStorage::dslSitePotential) {
1272	+	atomRowData.sitePotential[atom1] += *(idat.sPot1);
1273	+	atomColData.sitePotential[atom2] += *(idat.sPot2);
1274	+	}
1275	+
1276		#else
1277		pairwisePot += *(idat.pot);
1278	+	excludedPot += *(idat.excludedPot);
1279
1280		snap_->atomData.force[atom1] += *(idat.f1);
1281		snap_->atomData.force[atom2] -= *(idat.f1);
1282	+
1283	+	if (idat.doParticlePot) {
1284	+	// This is the pairwise contribution to the particle pot.  The
1285	+	// embedding contribution is added in each of the low level
1286	+	// non-bonded routines.  In parallel, this calculation is done
1287	+	// in collectData, not in unpackInteractionData.
1288	+	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1289	+	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1290	+	}
1291	+
1292	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1293	+	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1294	+	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1295	+	}
1296	+
1297	+	if (storageLayout_ & DataStorage::dslElectricField) {
1298	+	snap_->atomData.electricField[atom1] += *(idat.eField1);
1299	+	snap_->atomData.electricField[atom2] += *(idat.eField2);
1300	+	}
1301	+
1302	+	if (storageLayout_ & DataStorage::dslSitePotential) {
1303	+	snap_->atomData.sitePotential[atom1] += *(idat.sPot1);
1304	+	snap_->atomData.sitePotential[atom2] += *(idat.sPot2);
1305	+	}
1306	+
1307		#endif
1308
1309		}
#	Line 1023 | Line 1311 | namespace OpenMD {
1311		/*
1312		* buildNeighborList
1313		*
1314	<	* first element of pair is row-indexed CutoffGroup
1315	<	* second element of pair is column-indexed CutoffGroup
1314	>	* Constructs the Verlet neighbor list for a force-matrix
1315	>	* decomposition.  In this case, each processor is responsible for
1316	>	* row-site interactions with column-sites.
1317	>	*
1318	>	* neighborList is returned as a packed array of neighboring
1319	>	* column-ordered CutoffGroups.  The starting position in
1320	>	* neighborList for each row-ordered CutoffGroup is given by the
1321	>	* returned vector point.
1322		*/
1323	<	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1324	<
1325	<	vector<pair<int, int> > neighborList;
1326	<	groupCutoffs cuts;
1323	>	void ForceMatrixDecomposition::buildNeighborList(vector<int>& neighborList,
1324	>	vector<int>& point) {
1325	>	neighborList.clear();
1326	>	point.clear();
1327	>	int len = 0;
1328	>
1329		bool doAllPairs = false;
1330
1331	+	Snapshot* snap_ = sman_->getCurrentSnapshot();
1332	+	Mat3x3d box;
1333	+	Mat3x3d invBox;
1334	+
1335	+	Vector3d rs, scaled, dr;
1336	+	Vector3i whichCell;
1337	+	int cellIndex;
1338	+
1339		#ifdef IS_MPI
1340		cellListRow_.clear();
1341		cellListCol_.clear();
1342	+	point.resize(nGroupsInRow_+1);
1343		#else
1344		cellList_.clear();
1345	+	point.resize(nGroups_+1);
1346		#endif
1347	<
1348	<	RealType rList_ = (largestRcut_ + skinThickness_);
1349	<	RealType rl2 = rList_ * rList_;
1350	<	Snapshot* snap_ = sman_->getCurrentSnapshot();
1351	<	Mat3x3d Hmat = snap_->getHmat();
1352	<	Vector3d Hx = Hmat.getColumn(0);
1353	<	Vector3d Hy = Hmat.getColumn(1);
1354	<	Vector3d Hz = Hmat.getColumn(2);
1355	<
1356	<	nCells_.x() = (int) ( Hx.length() )/ rList_;
1357	<	nCells_.y() = (int) ( Hy.length() )/ rList_;
1358	<	nCells_.z() = (int) ( Hz.length() )/ rList_;
1347	>
1348	>	if (!usePeriodicBoundaryConditions_) {
1349	>	box = snap_->getBoundingBox();
1350	>	invBox = snap_->getInvBoundingBox();
1351	>	} else {
1352	>	box = snap_->getHmat();
1353	>	invBox = snap_->getInvHmat();
1354	>	}
1355	>
1356	>	Vector3d A = box.getColumn(0);
1357	>	Vector3d B = box.getColumn(1);
1358	>	Vector3d C = box.getColumn(2);
1359
1360	<	// handle small boxes where the cell offsets can end up repeating cells
1360	>	// Required for triclinic cells
1361	>	Vector3d AxB = cross(A, B);
1362	>	Vector3d BxC = cross(B, C);
1363	>	Vector3d CxA = cross(C, A);
1364	>
1365	>	// unit vectors perpendicular to the faces of the triclinic cell:
1366	>	AxB.normalize();
1367	>	BxC.normalize();
1368	>	CxA.normalize();
1369	>
1370	>	// A set of perpendicular lengths in triclinic cells:
1371	>	RealType Wa = abs(dot(A, BxC));
1372	>	RealType Wb = abs(dot(B, CxA));
1373	>	RealType Wc = abs(dot(C, AxB));
1374
1375	+	nCells_.x() = int( Wa / rList_ );
1376	+	nCells_.y() = int( Wb / rList_ );
1377	+	nCells_.z() = int( Wc / rList_ );
1378	+
1379	+	// handle small boxes where the cell offsets can end up repeating cells
1380		if (nCells_.x() < 3) doAllPairs = true;
1381		if (nCells_.y() < 3) doAllPairs = true;
1382		if (nCells_.z() < 3) doAllPairs = true;
1383	<
1060	<	Mat3x3d invHmat = snap_->getInvHmat();
1061	<	Vector3d rs, scaled, dr;
1062	<	Vector3i whichCell;
1063	<	int cellIndex;
1383	>
1384		int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1385	<
1385	>
1386		#ifdef IS_MPI
1387		cellListRow_.resize(nCtot);
1388		cellListCol_.resize(nCtot);
1389		#else
1390		cellList_.resize(nCtot);
1391		#endif
1392	<
1392	>
1393		if (!doAllPairs) {
1394	+
1395		#ifdef IS_MPI
1396	<
1396	>
1397		for (int i = 0; i < nGroupsInRow_; i++) {
1398		rs = cgRowData.position[i];
1399
1400		// scaled positions relative to the box vectors
1401	<	scaled = invHmat * rs;
1401	>	scaled = invBox * rs;
1402
1403		// wrap the vector back into the unit box by subtracting integer box
1404		// numbers
1405		for (int j = 0; j < 3; j++) {
1406		scaled[j] -= roundMe(scaled[j]);
1407		scaled[j] += 0.5;
1408	+	// Handle the special case when an object is exactly on the
1409	+	// boundary (a scaled coordinate of 1.0 is the same as
1410	+	// scaled coordinate of 0.0)
1411	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1412		}
1413
1414		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1101 | Line 1426 | namespace OpenMD {
1426		rs = cgColData.position[i];
1427
1428		// scaled positions relative to the box vectors
1429	<	scaled = invHmat * rs;
1429	>	scaled = invBox * rs;
1430
1431		// wrap the vector back into the unit box by subtracting integer box
1432		// numbers
1433		for (int j = 0; j < 3; j++) {
1434		scaled[j] -= roundMe(scaled[j]);
1435		scaled[j] += 0.5;
1436	+	// Handle the special case when an object is exactly on the
1437	+	// boundary (a scaled coordinate of 1.0 is the same as
1438	+	// scaled coordinate of 0.0)
1439	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1440		}
1441
1442		// find xyz-indices of cell that cutoffGroup is in.
#	Line 1121 | Line 1450 | namespace OpenMD {
1450		// add this cutoff group to the list of groups in this cell;
1451		cellListCol_[cellIndex].push_back(i);
1452		}
1453	<
1453	>
1454		#else
1455		for (int i = 0; i < nGroups_; i++) {
1456		rs = snap_->cgData.position[i];
1457
1458		// scaled positions relative to the box vectors
1459	<	scaled = invHmat * rs;
1459	>	scaled = invBox * rs;
1460
1461		// wrap the vector back into the unit box by subtracting integer box
1462		// numbers
1463		for (int j = 0; j < 3; j++) {
1464		scaled[j] -= roundMe(scaled[j]);
1465		scaled[j] += 0.5;
1466	+	// Handle the special case when an object is exactly on the
1467	+	// boundary (a scaled coordinate of 1.0 is the same as
1468	+	// scaled coordinate of 0.0)
1469	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1470		}
1471
1472		// find xyz-indices of cell that cutoffGroup is in.
1473	<	whichCell.x() = nCells_.x() * scaled.x();
1474	<	whichCell.y() = nCells_.y() * scaled.y();
1475	<	whichCell.z() = nCells_.z() * scaled.z();
1473	>	whichCell.x() = int(nCells_.x() * scaled.x());
1474	>	whichCell.y() = int(nCells_.y() * scaled.y());
1475	>	whichCell.z() = int(nCells_.z() * scaled.z());
1476
1477		// find single index of this cell:
1478		cellIndex = Vlinear(whichCell, nCells_);
#	Line 1150 | Line 1483 | namespace OpenMD {
1483
1484		#endif
1485
1486	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1487	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1488	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1489	<	Vector3i m1v(m1x, m1y, m1z);
1157	<	int m1 = Vlinear(m1v, nCells_);
1158	<
1159	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1160	<	os != cellOffsets_.end(); ++os) {
1161	<
1162	<	Vector3i m2v = m1v + (*os);
1163	<
1486	>	#ifdef IS_MPI
1487	>	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1488	>	rs = cgRowData.position[j1];
1489	>	#else
1490
1491	<	if (m2v.x() >= nCells_.x()) {
1492	<	m2v.x() = 0;
1493	<	} else if (m2v.x() < 0) {
1494	<	m2v.x() = nCells_.x() - 1;
1495	<	}
1491	>	for (int j1 = 0; j1 < nGroups_; j1++) {
1492	>	rs = snap_->cgData.position[j1];
1493	>	#endif
1494	>	point[j1] = len;
1495	>
1496	>	// scaled positions relative to the box vectors
1497	>	scaled = invBox * rs;
1498	>
1499	>	// wrap the vector back into the unit box by subtracting integer box
1500	>	// numbers
1501	>	for (int j = 0; j < 3; j++) {
1502	>	scaled[j] -= roundMe(scaled[j]);
1503	>	scaled[j] += 0.5;
1504	>	// Handle the special case when an object is exactly on the
1505	>	// boundary (a scaled coordinate of 1.0 is the same as
1506	>	// scaled coordinate of 0.0)
1507	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1508	>	}
1509	>
1510	>	// find xyz-indices of cell that cutoffGroup is in.
1511	>	whichCell.x() = int(nCells_.x() * scaled.x());
1512	>	whichCell.y() = int(nCells_.y() * scaled.y());
1513	>	whichCell.z() = int(nCells_.z() * scaled.z());
1514	>
1515	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1516	>	os != cellOffsets_.end(); ++os) {
1517
1518	<	if (m2v.y() >= nCells_.y()) {
1172	<	m2v.y() = 0;
1173	<	} else if (m2v.y() < 0) {
1174	<	m2v.y() = nCells_.y() - 1;
1175	<	}
1176	<
1177	<	if (m2v.z() >= nCells_.z()) {
1178	<	m2v.z() = 0;
1179	<	} else if (m2v.z() < 0) {
1180	<	m2v.z() = nCells_.z() - 1;
1181	<	}
1518	>	Vector3i m2v = whichCell + (*os);
1519
1520	<	int m2 = Vlinear (m2v, nCells_);
1521	<
1520	>	if (m2v.x() >= nCells_.x()) {
1521	>	m2v.x() = 0;
1522	>	} else if (m2v.x() < 0) {
1523	>	m2v.x() = nCells_.x() - 1;
1524	>	}
1525	>
1526	>	if (m2v.y() >= nCells_.y()) {
1527	>	m2v.y() = 0;
1528	>	} else if (m2v.y() < 0) {
1529	>	m2v.y() = nCells_.y() - 1;
1530	>	}
1531	>
1532	>	if (m2v.z() >= nCells_.z()) {
1533	>	m2v.z() = 0;
1534	>	} else if (m2v.z() < 0) {
1535	>	m2v.z() = nCells_.z() - 1;
1536	>	}
1537	>	int m2 = Vlinear (m2v, nCells_);
1538		#ifdef IS_MPI
1539	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1540	<	j1 != cellListRow_[m1].end(); ++j1) {
1541	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1542	<	j2 != cellListCol_[m2].end(); ++j2) {
1543	<
1544	<	// In parallel, we need to visit *all* pairs of row
1545	<	// & column indicies and will divide labor in the
1546	<	// force evaluation later.
1547	<	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1548	<	snap_->wrapVector(dr);
1549	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1550	<	if (dr.lengthSquare() < cuts.third) {
1551	<	neighborList.push_back(make_pair((*j1), (*j2)));
1552	<	}
1553	<	}
1201	<	}
1539	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1540	>	j2 != cellListCol_[m2].end(); ++j2) {
1541	>
1542	>	// In parallel, we need to visit *all* pairs of row
1543	>	// & column indicies and will divide labor in the
1544	>	// force evaluation later.
1545	>	dr = cgColData.position[(*j2)] - rs;
1546	>	if (usePeriodicBoundaryConditions_) {
1547	>	snap_->wrapVector(dr);
1548	>	}
1549	>	if (dr.lengthSquare() < rListSq_) {
1550	>	neighborList.push_back( (*j2) );
1551	>	++len;
1552	>	}
1553	>	}
1554		#else
1555	<	for (vector<int>::iterator j1 = cellList_[m1].begin();
1556	<	j1 != cellList_[m1].end(); ++j1) {
1557	<	for (vector<int>::iterator j2 = cellList_[m2].begin();
1558	<	j2 != cellList_[m2].end(); ++j2) {
1559	<
1560	<	// Always do this if we're in different cells or if
1561	<	// we're in the same cell and the global index of
1562	<	// the j2 cutoff group is greater than or equal to
1563	<	// the j1 cutoff group.  Note that Rappaport's code
1564	<	// has a "less than" conditional here, but that
1565	<	// deals with atom-by-atom computation.  OpenMD
1566	<	// allows atoms within a single cutoff group to
1567	<	// interact with each other.
1568	<
1569	<
1570	<
1571	<	if (m2 != m1 || (*j2) >= (*j1) ) {
1220	<
1221	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1222	<	snap_->wrapVector(dr);
1223	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1224	<	if (dr.lengthSquare() < cuts.third) {
1225	<	neighborList.push_back(make_pair((*j1), (*j2)));
1226	<	}
1227	<	}
1228	<	}
1555	>	for (vector<int>::iterator j2 = cellList_[m2].begin();
1556	>	j2 != cellList_[m2].end(); ++j2) {
1557	>
1558	>	// Always do this if we're in different cells or if
1559	>	// we're in the same cell and the global index of
1560	>	// the j2 cutoff group is greater than or equal to
1561	>	// the j1 cutoff group.  Note that Rappaport's code
1562	>	// has a "less than" conditional here, but that
1563	>	// deals with atom-by-atom computation.  OpenMD
1564	>	// allows atoms within a single cutoff group to
1565	>	// interact with each other.
1566	>
1567	>	if ( (*j2) >= j1 ) {
1568	>
1569	>	dr = snap_->cgData.position[(*j2)] - rs;
1570	>	if (usePeriodicBoundaryConditions_) {
1571	>	snap_->wrapVector(dr);
1572		}
1573	<	#endif
1573	>	if ( dr.lengthSquare() < rListSq_) {
1574	>	neighborList.push_back( (*j2) );
1575	>	++len;
1576	>	}
1577		}
1578	<	}
1578	>	}
1579	>	#endif
1580		}
1581	<	}
1581	>	}
1582		} else {
1583		// branch to do all cutoff group pairs
1584		#ifdef IS_MPI
1585		for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1586	+	point[j1] = len;
1587	+	rs = cgRowData.position[j1];
1588		for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1589	<	dr = cgColData.position[j2] - cgRowData.position[j1];
1590	<	snap_->wrapVector(dr);
1591	<	cuts = getGroupCutoffs( j1, j2 );
1243	<	if (dr.lengthSquare() < cuts.third) {
1244	<	neighborList.push_back(make_pair(j1, j2));
1589	>	dr = cgColData.position[j2] - rs;
1590	>	if (usePeriodicBoundaryConditions_) {
1591	>	snap_->wrapVector(dr);
1592		}
1593	+	if (dr.lengthSquare() < rListSq_) {
1594	+	neighborList.push_back( j2 );
1595	+	++len;
1596	+	}
1597		}
1598		}
1599		#else
1600		// include all groups here.
1601		for (int j1 = 0; j1 < nGroups_; j1++) {
1602	+	point[j1] = len;
1603	+	rs = snap_->cgData.position[j1];
1604		// include self group interactions j2 == j1
1605		for (int j2 = j1; j2 < nGroups_; j2++) {
1606	<	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1607	<	snap_->wrapVector(dr);
1608	<	cuts = getGroupCutoffs( j1, j2 );
1256	<	if (dr.lengthSquare() < cuts.third) {
1257	<	neighborList.push_back(make_pair(j1, j2));
1606	>	dr = snap_->cgData.position[j2] - rs;
1607	>	if (usePeriodicBoundaryConditions_) {
1608	>	snap_->wrapVector(dr);
1609		}
1610	+	if (dr.lengthSquare() < rListSq_) {
1611	+	neighborList.push_back( j2 );
1612	+	++len;
1613	+	}
1614		}
1615		}
1616		#endif
1617		}
1618	<
1618	>
1619	>	#ifdef IS_MPI
1620	>	point[nGroupsInRow_] = len;
1621	>	#else
1622	>	point[nGroups_] = len;
1623	>	#endif
1624	>
1625		// save the local cutoff group positions for the check that is
1626		// done on each loop:
1627		saved_CG_positions_.clear();
1628	+	saved_CG_positions_.reserve(nGroups_);
1629		for (int i = 0; i < nGroups_; i++)
1630		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1269	–
1270	–	return neighborList;
1631		}
1632		} //end namespace OpenMD

Diff Legend

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