ViewVC Help

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

root/OpenMD/branches/development/src/parallel/ForceMatrixDecomposition.cpp

Comparing branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents):
Revision 1583 by gezelter, Thu Jun 16 22:00:08 2011 UTC vs.
Revision 1803 by gezelter, Wed Oct 3 14:20:07 2012 UTC

#	Line 36 | Line 36
36		* [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37		* [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38		* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
39	<	* [4]  Vardeman & Gezelter, in progress (2009).
39	>	* [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40	>	* [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41		*/
42		#include "parallel/ForceMatrixDecomposition.hpp"
43		#include "math/SquareMatrix3.hpp"
#	Line 47 | Line 48 | namespace OpenMD {
48		using namespace std;
49		namespace OpenMD {
50
51	+	ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52	+
53	+	// In a parallel computation, row and colum scans must visit all
54	+	// surrounding cells (not just the 14 upper triangular blocks that
55	+	// are used when the processor can see all pairs)
56	+	#ifdef IS_MPI
57	+	cellOffsets_.clear();
58	+	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59	+	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60	+	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
61	+	cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62	+	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63	+	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
64	+	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65	+	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
66	+	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67	+	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68	+	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69	+	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70	+	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
71	+	cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72	+	cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73	+	cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74	+	cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75	+	cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76	+	cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77	+	cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78	+	cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79	+	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80	+	cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81	+	cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82	+	cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83	+	cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84	+	cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85	+	#endif
86	+	}
87	+
88	+
89		/**
90		* distributeInitialData is essentially a copy of the older fortran
91		* SimulationSetup
92		*/
54	–
93		void ForceMatrixDecomposition::distributeInitialData() {
94		snap_ = sman_->getCurrentSnapshot();
95		storageLayout_ = sman_->getStorageLayout();
96		ff_ = info_->getForceField();
97		nLocal_ = snap_->getNumberOfAtoms();
98	<
98	>
99		nGroups_ = info_->getNLocalCutoffGroups();
62	–	cerr << "in dId, nGroups = " << nGroups_ << "\n";
100		// gather the information for atomtype IDs (atids):
101		idents = info_->getIdentArray();
102		AtomLocalToGlobal = info_->getGlobalAtomIndices();
103		cgLocalToGlobal = info_->getGlobalGroupIndices();
104		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
105	+
106		massFactors = info_->getMassFactors();
69	–	PairList excludes = info_->getExcludedInteractions();
70	–	PairList oneTwo = info_->getOneTwoInteractions();
71	–	PairList oneThree = info_->getOneThreeInteractions();
72	–	PairList oneFour = info_->getOneFourInteractions();
107
108	+	PairList* excludes = info_->getExcludedInteractions();
109	+	PairList* oneTwo = info_->getOneTwoInteractions();
110	+	PairList* oneThree = info_->getOneThreeInteractions();
111	+	PairList* oneFour = info_->getOneFourInteractions();
112	+
113	+	if (needVelocities_)
114	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
115	+	DataStorage::dslVelocity);
116	+	else
117	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
118	+
119		#ifdef IS_MPI
120
121	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
122	<	AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
78	<	AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
79	<	AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
80	<	AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
121	>	MPI::Intracomm row = rowComm.getComm();
122	>	MPI::Intracomm col = colComm.getComm();
123
124	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
125	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
126	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
127	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
128	<	AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
124	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
125	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
126	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
127	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
128	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
129
130	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
131	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
132	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
133	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
130	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
131	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
132	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
133	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
134	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
135
136	<	nAtomsInRow_ = AtomCommIntRow->getSize();
137	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
138	<	nGroupsInRow_ = cgCommIntRow->getSize();
139	<	nGroupsInCol_ = cgCommIntColumn->getSize();
136	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
137	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
138	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
139	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
140
141	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
142	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
143	+	nGroupsInRow_ = cgPlanIntRow->getSize();
144	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
145	+
146		// Modify the data storage objects with the correct layouts and sizes:
147		atomRowData.resize(nAtomsInRow_);
148		atomRowData.setStorageLayout(storageLayout_);
#	Line 103 | Line 151 | namespace OpenMD {
151		cgRowData.resize(nGroupsInRow_);
152		cgRowData.setStorageLayout(DataStorage::dslPosition);
153		cgColData.resize(nGroupsInCol_);
154	<	cgColData.setStorageLayout(DataStorage::dslPosition);
155	<
154	>	if (needVelocities_)
155	>	// we only need column velocities if we need them.
156	>	cgColData.setStorageLayout(DataStorage::dslPosition |
157	>	DataStorage::dslVelocity);
158	>	else
159	>	cgColData.setStorageLayout(DataStorage::dslPosition);
160	>
161		identsRow.resize(nAtomsInRow_);
162		identsCol.resize(nAtomsInCol_);
163
164	<	AtomCommIntRow->gather(idents, identsRow);
165	<	AtomCommIntColumn->gather(idents, identsCol);
164	>	AtomPlanIntRow->gather(idents, identsRow);
165	>	AtomPlanIntColumn->gather(idents, identsCol);
166
167	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
168	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
169	<
117	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
118	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
167	>	// allocate memory for the parallel objects
168	>	atypesRow.resize(nAtomsInRow_);
169	>	atypesCol.resize(nAtomsInCol_);
170
171	<	AtomCommRealRow->gather(massFactors, massFactorsRow);
172	<	AtomCommRealColumn->gather(massFactors, massFactorsCol);
171	>	for (int i = 0; i < nAtomsInRow_; i++)
172	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
173	>	for (int i = 0; i < nAtomsInCol_; i++)
174	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
175
176	+	pot_row.resize(nAtomsInRow_);
177	+	pot_col.resize(nAtomsInCol_);
178	+
179	+	expot_row.resize(nAtomsInRow_);
180	+	expot_col.resize(nAtomsInCol_);
181	+
182	+	AtomRowToGlobal.resize(nAtomsInRow_);
183	+	AtomColToGlobal.resize(nAtomsInCol_);
184	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186	+
187	+	cgRowToGlobal.resize(nGroupsInRow_);
188	+	cgColToGlobal.resize(nGroupsInCol_);
189	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
191	+
192	+	massFactorsRow.resize(nAtomsInRow_);
193	+	massFactorsCol.resize(nAtomsInCol_);
194	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
195	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
196	+
197		groupListRow_.clear();
198		groupListRow_.resize(nGroupsInRow_);
199		for (int i = 0; i < nGroupsInRow_; i++) {
#	Line 142 | Line 216 | namespace OpenMD {
216		}
217		}
218
219	<	skipsForAtom.clear();
220	<	skipsForAtom.resize(nAtomsInRow_);
219	>	excludesForAtom.clear();
220	>	excludesForAtom.resize(nAtomsInRow_);
221		toposForAtom.clear();
222		toposForAtom.resize(nAtomsInRow_);
223		topoDist.clear();
#	Line 154 | Line 228 | namespace OpenMD {
228		for (int j = 0; j < nAtomsInCol_; j++) {
229		int jglob = AtomColToGlobal[j];
230
231	<	if (excludes.hasPair(iglob, jglob))
232	<	skipsForAtom[i].push_back(j);
231	>	if (excludes->hasPair(iglob, jglob))
232	>	excludesForAtom[i].push_back(j);
233
234	<	if (oneTwo.hasPair(iglob, jglob)) {
234	>	if (oneTwo->hasPair(iglob, jglob)) {
235		toposForAtom[i].push_back(j);
236		topoDist[i].push_back(1);
237		} else {
238	<	if (oneThree.hasPair(iglob, jglob)) {
238	>	if (oneThree->hasPair(iglob, jglob)) {
239		toposForAtom[i].push_back(j);
240		topoDist[i].push_back(2);
241		} else {
242	<	if (oneFour.hasPair(iglob, jglob)) {
242	>	if (oneFour->hasPair(iglob, jglob)) {
243		toposForAtom[i].push_back(j);
244		topoDist[i].push_back(3);
245		}
#	Line 174 | Line 248 | namespace OpenMD {
248		}
249		}
250
251	<	#endif
252	<
253	<	groupList_.clear();
180	<	groupList_.resize(nGroups_);
181	<	for (int i = 0; i < nGroups_; i++) {
182	<	int gid = cgLocalToGlobal[i];
183	<	for (int j = 0; j < nLocal_; j++) {
184	<	int aid = AtomLocalToGlobal[j];
185	<	if (globalGroupMembership[aid] == gid) {
186	<	groupList_[i].push_back(j);
187	<	}
188	<	}
189	<	}
190	<
191	<	skipsForAtom.clear();
192	<	skipsForAtom.resize(nLocal_);
251	>	#else
252	>	excludesForAtom.clear();
253	>	excludesForAtom.resize(nLocal_);
254		toposForAtom.clear();
255		toposForAtom.resize(nLocal_);
256		topoDist.clear();
#	Line 201 | Line 262 | namespace OpenMD {
262		for (int j = 0; j < nLocal_; j++) {
263		int jglob = AtomLocalToGlobal[j];
264
265	<	if (excludes.hasPair(iglob, jglob))
266	<	skipsForAtom[i].push_back(j);
265	>	if (excludes->hasPair(iglob, jglob))
266	>	excludesForAtom[i].push_back(j);
267
268	<	if (oneTwo.hasPair(iglob, jglob)) {
268	>	if (oneTwo->hasPair(iglob, jglob)) {
269		toposForAtom[i].push_back(j);
270		topoDist[i].push_back(1);
271		} else {
272	<	if (oneThree.hasPair(iglob, jglob)) {
272	>	if (oneThree->hasPair(iglob, jglob)) {
273		toposForAtom[i].push_back(j);
274		topoDist[i].push_back(2);
275		} else {
276	<	if (oneFour.hasPair(iglob, jglob)) {
276	>	if (oneFour->hasPair(iglob, jglob)) {
277		toposForAtom[i].push_back(j);
278		topoDist[i].push_back(3);
279		}
#	Line 220 | Line 281 | namespace OpenMD {
281		}
282		}
283		}
284	<
284	>	#endif
285	>
286	>	// allocate memory for the parallel objects
287	>	atypesLocal.resize(nLocal_);
288	>
289	>	for (int i = 0; i < nLocal_; i++)
290	>	atypesLocal[i] = ff_->getAtomType(idents[i]);
291	>
292	>	groupList_.clear();
293	>	groupList_.resize(nGroups_);
294	>	for (int i = 0; i < nGroups_; i++) {
295	>	int gid = cgLocalToGlobal[i];
296	>	for (int j = 0; j < nLocal_; j++) {
297	>	int aid = AtomLocalToGlobal[j];
298	>	if (globalGroupMembership[aid] == gid) {
299	>	groupList_[i].push_back(j);
300	>	}
301	>	}
302	>	}
303	>
304	>
305		createGtypeCutoffMap();
306	+
307		}
308
309		void ForceMatrixDecomposition::createGtypeCutoffMap() {
310	<
310	>
311		RealType tol = 1e-6;
312	<	RealType rc;
312	>	largestRcut_ = 0.0;
313		int atid;
314		set<AtomType*> atypes = info_->getSimulatedAtomTypes();
315	<	vector<RealType> atypeCutoff;
316	<	atypeCutoff.resize( atypes.size() );
315	>
316	>	map<int, RealType> atypeCutoff;
317
318		for (set<AtomType*>::iterator at = atypes.begin();
319		at != atypes.end(); ++at){
320		atid = (*at)->getIdent();
321	<
240	<	if (userChoseCutoff_)
321	>	if (userChoseCutoff_)
322		atypeCutoff[atid] = userCutoff_;
323		else
324		atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
325		}
326	<
326	>
327		vector<RealType> gTypeCutoffs;
247	–
328		// first we do a single loop over the cutoff groups to find the
329		// largest cutoff for any atypes present in this group.
330		#ifdef IS_MPI
#	Line 302 | Line 382 | namespace OpenMD {
382
383		vector<RealType> groupCutoff(nGroups_, 0.0);
384		groupToGtype.resize(nGroups_);
305	–
306	–	cerr << "nGroups = " << nGroups_ << "\n";
385		for (int cg1 = 0; cg1 < nGroups_; cg1++) {
308	–
386		groupCutoff[cg1] = 0.0;
387		vector<int> atomList = getAtomsInGroupRow(cg1);
311	–
388		for (vector<int>::iterator ia = atomList.begin();
389		ia != atomList.end(); ++ia) {
390		int atom1 = (*ia);
391		atid = idents[atom1];
392	<	if (atypeCutoff[atid] > groupCutoff[cg1]) {
392	>	if (atypeCutoff[atid] > groupCutoff[cg1])
393		groupCutoff[cg1] = atypeCutoff[atid];
318	–	}
394		}
395	<
395	>
396		bool gTypeFound = false;
397	<	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
397	>	for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
398		if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
399		groupToGtype[cg1] = gt;
400		gTypeFound = true;
401		}
402		}
403	<	if (!gTypeFound) {
403	>	if (!gTypeFound) {
404		gTypeCutoffs.push_back( groupCutoff[cg1] );
405		groupToGtype[cg1] = gTypeCutoffs.size() - 1;
406		}
407		}
408		#endif
409
335	–	cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
410		// Now we find the maximum group cutoff value present in the simulation
411
412	<	RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
412	>	RealType groupMax = *max_element(gTypeCutoffs.begin(),
413	>	gTypeCutoffs.end());
414
415		#ifdef IS_MPI
416	<	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
416	>	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
417	>	MPI::MAX);
418		#endif
419
420		RealType tradRcut = groupMax;
421
422	<	for (int i = 0; i < gTypeCutoffs.size();  i++) {
423	<	for (int j = 0; j < gTypeCutoffs.size();  j++) {
422	>	for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) {
423	>	for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) {
424		RealType thisRcut;
425		switch(cutoffPolicy_) {
426		case TRADITIONAL:
#	Line 368 | Line 444 | namespace OpenMD {
444
445		pair<int,int> key = make_pair(i,j);
446		gTypeCutoffMap[key].first = thisRcut;
371	–
447		if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
373	–
448		gTypeCutoffMap[key].second = thisRcut*thisRcut;
375	–
449		gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
377	–
450		// sanity check
451
452		if (userChoseCutoff_) {
#	Line 391 | Line 463 | namespace OpenMD {
463		}
464		}
465
394	–
466		groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
467		int i, j;
468		#ifdef IS_MPI
#	Line 405 | Line 476 | namespace OpenMD {
476		}
477
478		int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
479	<	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
479	>	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
480		if (toposForAtom[atom1][j] == atom2)
481		return topoDist[atom1][j];
482		}
#	Line 415 | Line 486 | namespace OpenMD {
486		void ForceMatrixDecomposition::zeroWorkArrays() {
487		pairwisePot = 0.0;
488		embeddingPot = 0.0;
489	+	excludedPot = 0.0;
490	+	excludedSelfPot = 0.0;
491
492		#ifdef IS_MPI
493		if (storageLayout_ & DataStorage::dslForce) {
#	Line 433 | Line 506 | namespace OpenMD {
506		fill(pot_col.begin(), pot_col.end(),
507		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
508
509	+	fill(expot_row.begin(), expot_row.end(),
510	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
511	+
512	+	fill(expot_col.begin(), expot_col.end(),
513	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
514	+
515		if (storageLayout_ & DataStorage::dslParticlePot) {
516	<	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
517	<	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
516	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
517	>	0.0);
518	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
519	>	0.0);
520		}
521
522		if (storageLayout_ & DataStorage::dslDensity) {
#	Line 444 | Line 525 | namespace OpenMD {
525		}
526
527		if (storageLayout_ & DataStorage::dslFunctional) {
528	<	fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
529	<	fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
528	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
529	>	0.0);
530	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
531	>	0.0);
532		}
533
534		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
#	Line 455 | Line 538 | namespace OpenMD {
538		atomColData.functionalDerivative.end(), 0.0);
539		}
540
541	<	#else
542	<
541	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
542	>	fill(atomRowData.skippedCharge.begin(),
543	>	atomRowData.skippedCharge.end(), 0.0);
544	>	fill(atomColData.skippedCharge.begin(),
545	>	atomColData.skippedCharge.end(), 0.0);
546	>	}
547	>
548	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
549	>	fill(atomRowData.flucQFrc.begin(),
550	>	atomRowData.flucQFrc.end(), 0.0);
551	>	fill(atomColData.flucQFrc.begin(),
552	>	atomColData.flucQFrc.end(), 0.0);
553	>	}
554	>
555	>	if (storageLayout_ & DataStorage::dslElectricField) {
556	>	fill(atomRowData.electricField.begin(),
557	>	atomRowData.electricField.end(), V3Zero);
558	>	fill(atomColData.electricField.begin(),
559	>	atomColData.electricField.end(), V3Zero);
560	>	}
561	>
562	>	#endif
563	>	// even in parallel, we need to zero out the local arrays:
564	>
565		if (storageLayout_ & DataStorage::dslParticlePot) {
566		fill(snap_->atomData.particlePot.begin(),
567		snap_->atomData.particlePot.end(), 0.0);
#	Line 466 | Line 571 | namespace OpenMD {
571		fill(snap_->atomData.density.begin(),
572		snap_->atomData.density.end(), 0.0);
573		}
574	+
575		if (storageLayout_ & DataStorage::dslFunctional) {
576		fill(snap_->atomData.functional.begin(),
577		snap_->atomData.functional.end(), 0.0);
578		}
579	+
580		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
581		fill(snap_->atomData.functionalDerivative.begin(),
582		snap_->atomData.functionalDerivative.end(), 0.0);
583		}
584	<	#endif
585	<
584	>
585	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
586	>	fill(snap_->atomData.skippedCharge.begin(),
587	>	snap_->atomData.skippedCharge.end(), 0.0);
588	>	}
589	>
590	>	if (storageLayout_ & DataStorage::dslElectricField) {
591	>	fill(snap_->atomData.electricField.begin(),
592	>	snap_->atomData.electricField.end(), V3Zero);
593	>	}
594		}
595
596
#	Line 485 | Line 600 | namespace OpenMD {
600		#ifdef IS_MPI
601
602		// gather up the atomic positions
603	<	AtomCommVectorRow->gather(snap_->atomData.position,
603	>	AtomPlanVectorRow->gather(snap_->atomData.position,
604		atomRowData.position);
605	<	AtomCommVectorColumn->gather(snap_->atomData.position,
605	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
606		atomColData.position);
607
608		// gather up the cutoff group positions
609	<	cgCommVectorRow->gather(snap_->cgData.position,
609	>
610	>	cgPlanVectorRow->gather(snap_->cgData.position,
611		cgRowData.position);
612	<	cgCommVectorColumn->gather(snap_->cgData.position,
612	>
613	>	cgPlanVectorColumn->gather(snap_->cgData.position,
614		cgColData.position);
615	+
616	+
617	+
618	+	if (needVelocities_) {
619	+	// gather up the atomic velocities
620	+	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
621	+	atomColData.velocity);
622	+
623	+	cgPlanVectorColumn->gather(snap_->cgData.velocity,
624	+	cgColData.velocity);
625	+	}
626	+
627
628		// if needed, gather the atomic rotation matrices
629		if (storageLayout_ & DataStorage::dslAmat) {
630	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
630	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
631		atomRowData.aMat);
632	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
632	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
633		atomColData.aMat);
634		}
635	<
636	<	// if needed, gather the atomic eletrostatic frames
637	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
638	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
639	<	atomRowData.electroFrame);
640	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
641	<	atomColData.electroFrame);
635	>
636	>	// if needed, gather the atomic eletrostatic information
637	>	if (storageLayout_ & DataStorage::dslDipole) {
638	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
639	>	atomRowData.dipole);
640	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
641	>	atomColData.dipole);
642		}
643	+
644	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
645	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
646	+	atomRowData.quadrupole);
647	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
648	+	atomColData.quadrupole);
649	+	}
650	+
651	+	// if needed, gather the atomic fluctuating charge values
652	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
653	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
654	+	atomRowData.flucQPos);
655	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
656	+	atomColData.flucQPos);
657	+	}
658	+
659		#endif
660		}
661
#	Line 524 | Line 669 | namespace OpenMD {
669
670		if (storageLayout_ & DataStorage::dslDensity) {
671
672	<	AtomCommRealRow->scatter(atomRowData.density,
672	>	AtomPlanRealRow->scatter(atomRowData.density,
673		snap_->atomData.density);
674
675		int n = snap_->atomData.density.size();
676		vector<RealType> rho_tmp(n, 0.0);
677	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
677	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
678		for (int i = 0; i < n; i++)
679		snap_->atomData.density[i] += rho_tmp[i];
680		}
681	+
682	+	if (storageLayout_ & DataStorage::dslElectricField) {
683	+
684	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
685	+	snap_->atomData.electricField);
686	+
687	+	int n = snap_->atomData.electricField.size();
688	+	vector<Vector3d> field_tmp(n, V3Zero);
689	+	AtomPlanVectorColumn->scatter(atomColData.electricField,
690	+	field_tmp);
691	+	for (int i = 0; i < n; i++)
692	+	snap_->atomData.electricField[i] += field_tmp[i];
693	+	}
694		#endif
695		}
696
#	Line 545 | Line 703 | namespace OpenMD {
703		storageLayout_ = sman_->getStorageLayout();
704		#ifdef IS_MPI
705		if (storageLayout_ & DataStorage::dslFunctional) {
706	<	AtomCommRealRow->gather(snap_->atomData.functional,
706	>	AtomPlanRealRow->gather(snap_->atomData.functional,
707		atomRowData.functional);
708	<	AtomCommRealColumn->gather(snap_->atomData.functional,
708	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
709		atomColData.functional);
710		}
711
712		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
713	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
713	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
714		atomRowData.functionalDerivative);
715	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
715	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
716		atomColData.functionalDerivative);
717		}
718		#endif
#	Line 568 | Line 726 | namespace OpenMD {
726		int n = snap_->atomData.force.size();
727		vector<Vector3d> frc_tmp(n, V3Zero);
728
729	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
729	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
730		for (int i = 0; i < n; i++) {
731		snap_->atomData.force[i] += frc_tmp[i];
732		frc_tmp[i] = 0.0;
733		}
734
735	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
736	<	for (int i = 0; i < n; i++)
735	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
736	>	for (int i = 0; i < n; i++) {
737		snap_->atomData.force[i] += frc_tmp[i];
738	<
739	<
738	>	}
739	>
740		if (storageLayout_ & DataStorage::dslTorque) {
741
742	<	int nt = snap_->atomData.force.size();
742	>	int nt = snap_->atomData.torque.size();
743		vector<Vector3d> trq_tmp(nt, V3Zero);
744
745	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
746	<	for (int i = 0; i < n; i++) {
745	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
746	>	for (int i = 0; i < nt; i++) {
747		snap_->atomData.torque[i] += trq_tmp[i];
748		trq_tmp[i] = 0.0;
749		}
750
751	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
752	<	for (int i = 0; i < n; i++)
751	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
752	>	for (int i = 0; i < nt; i++)
753		snap_->atomData.torque[i] += trq_tmp[i];
754	+	}
755	+
756	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
757	+
758	+	int ns = snap_->atomData.skippedCharge.size();
759	+	vector<RealType> skch_tmp(ns, 0.0);
760	+
761	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
762	+	for (int i = 0; i < ns; i++) {
763	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
764	+	skch_tmp[i] = 0.0;
765	+	}
766	+
767	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
768	+	for (int i = 0; i < ns; i++)
769	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
770	+
771		}
772
773	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
774	+
775	+	int nq = snap_->atomData.flucQFrc.size();
776	+	vector<RealType> fqfrc_tmp(nq, 0.0);
777	+
778	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
779	+	for (int i = 0; i < nq; i++) {
780	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
781	+	fqfrc_tmp[i] = 0.0;
782	+	}
783	+
784	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
785	+	for (int i = 0; i < nq; i++)
786	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
787	+
788	+	}
789	+
790		nLocal_ = snap_->getNumberOfAtoms();
791
792		vector<potVec> pot_temp(nLocal_,
793		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
794	+	vector<potVec> expot_temp(nLocal_,
795	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
796
797		// scatter/gather pot_row into the members of my column
798
799	<	AtomCommPotRow->scatter(pot_row, pot_temp);
799	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
800	>	AtomPlanPotRow->scatter(expot_row, expot_temp);
801
802	<	for (int ii = 0;  ii < pot_temp.size(); ii++ )
802	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
803		pairwisePot += pot_temp[ii];
804	<
804	>
805	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
806	>	excludedPot += expot_temp[ii];
807	>
808	>	if (storageLayout_ & DataStorage::dslParticlePot) {
809	>	// This is the pairwise contribution to the particle pot.  The
810	>	// embedding contribution is added in each of the low level
811	>	// non-bonded routines.  In single processor, this is done in
812	>	// unpackInteractionData, not in collectData.
813	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
814	>	for (int i = 0; i < nLocal_; i++) {
815	>	// factor of two is because the total potential terms are divided
816	>	// by 2 in parallel due to row/ column scatter
817	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
818	>	}
819	>	}
820	>	}
821	>
822		fill(pot_temp.begin(), pot_temp.end(),
823		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
824	+	fill(expot_temp.begin(), expot_temp.end(),
825	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
826
827	<	AtomCommPotColumn->scatter(pot_col, pot_temp);
827	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
828	>	AtomPlanPotColumn->scatter(expot_col, expot_temp);
829
830		for (int ii = 0;  ii < pot_temp.size(); ii++ )
831		pairwisePot += pot_temp[ii];
832	+
833	+	for (int ii = 0;  ii < expot_temp.size(); ii++ )
834	+	excludedPot += expot_temp[ii];
835	+
836	+	if (storageLayout_ & DataStorage::dslParticlePot) {
837	+	// This is the pairwise contribution to the particle pot.  The
838	+	// embedding contribution is added in each of the low level
839	+	// non-bonded routines.  In single processor, this is done in
840	+	// unpackInteractionData, not in collectData.
841	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
842	+	for (int i = 0; i < nLocal_; i++) {
843	+	// factor of two is because the total potential terms are divided
844	+	// by 2 in parallel due to row/ column scatter
845	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
846	+	}
847	+	}
848	+	}
849	+
850	+	if (storageLayout_ & DataStorage::dslParticlePot) {
851	+	int npp = snap_->atomData.particlePot.size();
852	+	vector<RealType> ppot_temp(npp, 0.0);
853	+
854	+	// This is the direct or embedding contribution to the particle
855	+	// pot.
856	+
857	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
858	+	for (int i = 0; i < npp; i++) {
859	+	snap_->atomData.particlePot[i] += ppot_temp[i];
860	+	}
861	+
862	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
863	+
864	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
865	+	for (int i = 0; i < npp; i++) {
866	+	snap_->atomData.particlePot[i] += ppot_temp[i];
867	+	}
868	+	}
869	+
870	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
871	+	RealType ploc1 = pairwisePot[ii];
872	+	RealType ploc2 = 0.0;
873	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
874	+	pairwisePot[ii] = ploc2;
875	+	}
876	+
877	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
878	+	RealType ploc1 = excludedPot[ii];
879	+	RealType ploc2 = 0.0;
880	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
881	+	excludedPot[ii] = ploc2;
882	+	}
883	+
884	+	// Here be dragons.
885	+	MPI::Intracomm col = colComm.getComm();
886	+
887	+	col.Allreduce(MPI::IN_PLACE,
888	+	&snap_->frameData.conductiveHeatFlux[0], 3,
889	+	MPI::REALTYPE, MPI::SUM);
890	+
891	+
892		#endif
893
894		}
895
896	+	/**
897	+	* Collects information obtained during the post-pair (and embedding
898	+	* functional) loops onto local data structures.
899	+	*/
900	+	void ForceMatrixDecomposition::collectSelfData() {
901	+	snap_ = sman_->getCurrentSnapshot();
902	+	storageLayout_ = sman_->getStorageLayout();
903	+
904	+	#ifdef IS_MPI
905	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
906	+	RealType ploc1 = embeddingPot[ii];
907	+	RealType ploc2 = 0.0;
908	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
909	+	embeddingPot[ii] = ploc2;
910	+	}
911	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
912	+	RealType ploc1 = excludedSelfPot[ii];
913	+	RealType ploc2 = 0.0;
914	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
915	+	excludedSelfPot[ii] = ploc2;
916	+	}
917	+	#endif
918	+
919	+	}
920	+
921	+
922	+
923		int ForceMatrixDecomposition::getNAtomsInRow() {
924		#ifdef IS_MPI
925		return nAtomsInRow_;
#	Line 658 | Line 960 | namespace OpenMD {
960		return d;
961		}
962
963	+	Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
964	+	#ifdef IS_MPI
965	+	return cgColData.velocity[cg2];
966	+	#else
967	+	return snap_->cgData.velocity[cg2];
968	+	#endif
969	+	}
970
971	+	Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
972	+	#ifdef IS_MPI
973	+	return atomColData.velocity[atom2];
974	+	#else
975	+	return snap_->atomData.velocity[atom2];
976	+	#endif
977	+	}
978	+
979	+
980		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
981
982		Vector3d d;
#	Line 716 | Line 1034 | namespace OpenMD {
1034		return d;
1035		}
1036
1037	<	vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
1038	<	return skipsForAtom[atom1];
1037	>	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1038	>	return excludesForAtom[atom1];
1039		}
1040
1041		/**
1042	<	* There are a number of reasons to skip a pair or a
725	<	* particle. Mostly we do this to exclude atoms who are involved in
726	<	* short range interactions (bonds, bends, torsions), but we also
727	<	* need to exclude some overcounted interactions that result from
1042	>	* We need to exclude some overcounted interactions that result from
1043		* the parallel decomposition.
1044		*/
1045	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1045	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1046		int unique_id_1, unique_id_2;
1047	<
1047	>
1048		#ifdef IS_MPI
1049		// in MPI, we have to look up the unique IDs for each atom
1050		unique_id_1 = AtomRowToGlobal[atom1];
1051		unique_id_2 = AtomColToGlobal[atom2];
1052	+	// group1 = cgRowToGlobal[cg1];
1053	+	// group2 = cgColToGlobal[cg2];
1054	+	#else
1055	+	unique_id_1 = AtomLocalToGlobal[atom1];
1056	+	unique_id_2 = AtomLocalToGlobal[atom2];
1057	+	int group1 = cgLocalToGlobal[cg1];
1058	+	int group2 = cgLocalToGlobal[cg2];
1059	+	#endif
1060
738	–	// this situation should only arise in MPI simulations
1061		if (unique_id_1 == unique_id_2) return true;
1062	<
1062	>
1063	>	#ifdef IS_MPI
1064		// this prevents us from doing the pair on multiple processors
1065		if (unique_id_1 < unique_id_2) {
1066		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1067		} else {
1068	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1068	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1069		}
1070	<	#else
1071	<	// in the normal loop, the atom numbers are unique
1072	<	unique_id_1 = atom1;
1073	<	unique_id_2 = atom2;
1070	>	#endif
1071	>
1072	>	#ifndef IS_MPI
1073	>	if (group1 == group2) {
1074	>	if (unique_id_1 < unique_id_2) return true;
1075	>	}
1076		#endif
1077
1078	<	for (vector<int>::iterator i = skipsForAtom[atom1].begin();
1079	<	i != skipsForAtom[atom1].end(); ++i) {
1080	<	if ( (*i) == unique_id_2 ) return true;
1078	>	return false;
1079	>	}
1080	>
1081	>	/**
1082	>	* We need to handle the interactions for atoms who are involved in
1083	>	* the same rigid body as well as some short range interactions
1084	>	* (bonds, bends, torsions) differently from other interactions.
1085	>	* We'll still visit the pairwise routines, but with a flag that
1086	>	* tells those routines to exclude the pair from direct long range
1087	>	* interactions.  Some indirect interactions (notably reaction
1088	>	* field) must still be handled for these pairs.
1089	>	*/
1090	>	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1091	>
1092	>	// excludesForAtom was constructed to use row/column indices in the MPI
1093	>	// version, and to use local IDs in the non-MPI version:
1094	>
1095	>	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1096	>	i != excludesForAtom[atom1].end(); ++i) {
1097	>	if ( (*i) == atom2 ) return true;
1098		}
1099
1100		return false;
#	Line 777 | Line 1119 | namespace OpenMD {
1119
1120		// filling interaction blocks with pointers
1121		void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1122	<	int atom1, int atom2) {
1122	>	int atom1, int atom2) {
1123	>
1124	>	idat.excluded = excludeAtomPair(atom1, atom2);
1125	>
1126		#ifdef IS_MPI
1127	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1128	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1129	+	//                         ff_->getAtomType(identsCol[atom2]) );
1130
783	–	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
784	–	ff_->getAtomType(identsCol[atom2]) );
785	–
1131		if (storageLayout_ & DataStorage::dslAmat) {
1132		idat.A1 = &(atomRowData.aMat[atom1]);
1133		idat.A2 = &(atomColData.aMat[atom2]);
1134		}
1135
791	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
792	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
793	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
794	–	}
795	–
1136		if (storageLayout_ & DataStorage::dslTorque) {
1137		idat.t1 = &(atomRowData.torque[atom1]);
1138		idat.t2 = &(atomColData.torque[atom2]);
1139		}
1140
1141	+	if (storageLayout_ & DataStorage::dslDipole) {
1142	+	idat.dipole1 = &(atomRowData.dipole[atom1]);
1143	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1144	+	}
1145	+
1146	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1147	+	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1148	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1149	+	}
1150	+
1151		if (storageLayout_ & DataStorage::dslDensity) {
1152		idat.rho1 = &(atomRowData.density[atom1]);
1153		idat.rho2 = &(atomColData.density[atom2]);
#	Line 818 | Line 1168 | namespace OpenMD {
1168		idat.particlePot2 = &(atomColData.particlePot[atom2]);
1169		}
1170
1171	<	#else
1171	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1172	>	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1173	>	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1174	>	}
1175
1176	<	idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1177	<	ff_->getAtomType(idents[atom2]) );
1176	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1177	>	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1178	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1179	>	}
1180
1181	+	#else
1182	+
1183	+	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1184	+
1185		if (storageLayout_ & DataStorage::dslAmat) {
1186		idat.A1 = &(snap_->atomData.aMat[atom1]);
1187		idat.A2 = &(snap_->atomData.aMat[atom2]);
1188		}
1189
831	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
832	–	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
833	–	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
834	–	}
835	–
1190		if (storageLayout_ & DataStorage::dslTorque) {
1191		idat.t1 = &(snap_->atomData.torque[atom1]);
1192		idat.t2 = &(snap_->atomData.torque[atom2]);
1193		}
1194
1195	+	if (storageLayout_ & DataStorage::dslDipole) {
1196	+	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1197	+	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1198	+	}
1199	+
1200	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1201	+	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1202	+	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1203	+	}
1204	+
1205		if (storageLayout_ & DataStorage::dslDensity) {
1206		idat.rho1 = &(snap_->atomData.density[atom1]);
1207		idat.rho2 = &(snap_->atomData.density[atom2]);
#	Line 858 | Line 1222 | namespace OpenMD {
1222		idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1223		}
1224
1225	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1226	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1227	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1228	+	}
1229	+
1230	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1231	+	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1232	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1233	+	}
1234	+
1235		#endif
1236		}
1237
1238
1239		void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1240		#ifdef IS_MPI
1241	<	pot_row[atom1] += 0.5 *  *(idat.pot);
1242	<	pot_col[atom2] += 0.5 *  *(idat.pot);
1241	>	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1242	>	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1243	>	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1244	>	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1245
1246		atomRowData.force[atom1] += *(idat.f1);
1247		atomColData.force[atom2] -= *(idat.f1);
1248	+
1249	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1250	+	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1251	+	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1252	+	}
1253	+
1254	+	if (storageLayout_ & DataStorage::dslElectricField) {
1255	+	atomRowData.electricField[atom1] += *(idat.eField1);
1256	+	atomColData.electricField[atom2] += *(idat.eField2);
1257	+	}
1258	+
1259		#else
1260		pairwisePot += *(idat.pot);
1261	+	excludedPot += *(idat.excludedPot);
1262
1263		snap_->atomData.force[atom1] += *(idat.f1);
1264		snap_->atomData.force[atom2] -= *(idat.f1);
877	–	#endif
1265
1266	<	}
1267	<
1268	<
1269	<	void ForceMatrixDecomposition::fillSkipData(InteractionData &idat,
1270	<	int atom1, int atom2) {
1271	<	// Still Missing:: skippedCharge fill must be added to DataStorage
1272	<	#ifdef IS_MPI
886	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
887	<	ff_->getAtomType(identsCol[atom2]) );
888	<
889	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
890	<	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
891	<	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1266	>	if (idat.doParticlePot) {
1267	>	// This is the pairwise contribution to the particle pot.  The
1268	>	// embedding contribution is added in each of the low level
1269	>	// non-bonded routines.  In parallel, this calculation is done
1270	>	// in collectData, not in unpackInteractionData.
1271	>	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1272	>	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1273		}
1274	<	if (storageLayout_ & DataStorage::dslTorque) {
1275	<	idat.t1 = &(atomRowData.torque[atom1]);
1276	<	idat.t2 = &(atomColData.torque[atom2]);
1274	>
1275	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1276	>	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1277	>	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1278		}
897	–	#else
898	–	idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
899	–	ff_->getAtomType(idents[atom2]) );
1279
1280	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1281	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1282	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1280	>	if (storageLayout_ & DataStorage::dslElectricField) {
1281	>	snap_->atomData.electricField[atom1] += *(idat.eField1);
1282	>	snap_->atomData.electricField[atom2] += *(idat.eField2);
1283		}
905	–	if (storageLayout_ & DataStorage::dslTorque) {
906	–	idat.t1 = &(snap_->atomData.torque[atom1]);
907	–	idat.t2 = &(snap_->atomData.torque[atom2]);
908	–	}
909	–	#endif
910	–	}
1284
912	–
913	–	void ForceMatrixDecomposition::unpackSkipData(InteractionData &idat, int atom1, int atom2) {
914	–	#ifdef IS_MPI
915	–	pot_row[atom1] += 0.5 *  *(idat.pot);
916	–	pot_col[atom2] += 0.5 *  *(idat.pot);
917	–	#else
918	–	pairwisePot += *(idat.pot);
1285		#endif
1286	<
1286	>
1287		}
1288
923	–
1289		/*
1290		* buildNeighborList
1291		*
#	Line 931 | Line 1296 | namespace OpenMD {
1296
1297		vector<pair<int, int> > neighborList;
1298		groupCutoffs cuts;
1299	+	bool doAllPairs = false;
1300	+
1301		#ifdef IS_MPI
1302		cellListRow_.clear();
1303		cellListCol_.clear();
#	Line 950 | Line 1317 | namespace OpenMD {
1317		nCells_.y() = (int) ( Hy.length() )/ rList_;
1318		nCells_.z() = (int) ( Hz.length() )/ rList_;
1319
1320	+	// handle small boxes where the cell offsets can end up repeating cells
1321	+
1322	+	if (nCells_.x() < 3) doAllPairs = true;
1323	+	if (nCells_.y() < 3) doAllPairs = true;
1324	+	if (nCells_.z() < 3) doAllPairs = true;
1325	+
1326		Mat3x3d invHmat = snap_->getInvHmat();
1327		Vector3d rs, scaled, dr;
1328		Vector3i whichCell;
#	Line 963 | Line 1336 | namespace OpenMD {
1336		cellList_.resize(nCtot);
1337		#endif
1338
1339	+	if (!doAllPairs) {
1340		#ifdef IS_MPI
967	–	for (int i = 0; i < nGroupsInRow_; i++) {
968	–	rs = cgRowData.position[i];
1341
1342	<	// scaled positions relative to the box vectors
1343	<	scaled = invHmat * rs;
1344	<
1345	<	// wrap the vector back into the unit box by subtracting integer box
1346	<	// numbers
1347	<	for (int j = 0; j < 3; j++) {
1348	<	scaled[j] -= roundMe(scaled[j]);
1349	<	scaled[j] += 0.5;
1342	>	for (int i = 0; i < nGroupsInRow_; i++) {
1343	>	rs = cgRowData.position[i];
1344	>
1345	>	// scaled positions relative to the box vectors
1346	>	scaled = invHmat * rs;
1347	>
1348	>	// wrap the vector back into the unit box by subtracting integer box
1349	>	// numbers
1350	>	for (int j = 0; j < 3; j++) {
1351	>	scaled[j] -= roundMe(scaled[j]);
1352	>	scaled[j] += 0.5;
1353	>	// Handle the special case when an object is exactly on the
1354	>	// boundary (a scaled coordinate of 1.0 is the same as
1355	>	// scaled coordinate of 0.0)
1356	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1357	>	}
1358	>
1359	>	// find xyz-indices of cell that cutoffGroup is in.
1360	>	whichCell.x() = nCells_.x() * scaled.x();
1361	>	whichCell.y() = nCells_.y() * scaled.y();
1362	>	whichCell.z() = nCells_.z() * scaled.z();
1363	>
1364	>	// find single index of this cell:
1365	>	cellIndex = Vlinear(whichCell, nCells_);
1366	>
1367	>	// add this cutoff group to the list of groups in this cell;
1368	>	cellListRow_[cellIndex].push_back(i);
1369		}
1370	<
1371	<	// find xyz-indices of cell that cutoffGroup is in.
1372	<	whichCell.x() = nCells_.x() * scaled.x();
1373	<	whichCell.y() = nCells_.y() * scaled.y();
1374	<	whichCell.z() = nCells_.z() * scaled.z();
1375	<
1376	<	// find single index of this cell:
1377	<	cellIndex = Vlinear(whichCell, nCells_);
1378	<
1379	<	// add this cutoff group to the list of groups in this cell;
1380	<	cellListRow_[cellIndex].push_back(i);
1381	<	}
1382	<
1383	<	for (int i = 0; i < nGroupsInCol_; i++) {
1384	<	rs = cgColData.position[i];
1385	<
1386	<	// scaled positions relative to the box vectors
1387	<	scaled = invHmat * rs;
1388	<
1389	<	// wrap the vector back into the unit box by subtracting integer box
1390	<	// numbers
1391	<	for (int j = 0; j < 3; j++) {
1392	<	scaled[j] -= roundMe(scaled[j]);
1393	<	scaled[j] += 0.5;
1370	>	for (int i = 0; i < nGroupsInCol_; i++) {
1371	>	rs = cgColData.position[i];
1372	>
1373	>	// scaled positions relative to the box vectors
1374	>	scaled = invHmat * rs;
1375	>
1376	>	// wrap the vector back into the unit box by subtracting integer box
1377	>	// numbers
1378	>	for (int j = 0; j < 3; j++) {
1379	>	scaled[j] -= roundMe(scaled[j]);
1380	>	scaled[j] += 0.5;
1381	>	// Handle the special case when an object is exactly on the
1382	>	// boundary (a scaled coordinate of 1.0 is the same as
1383	>	// scaled coordinate of 0.0)
1384	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1385	>	}
1386	>
1387	>	// find xyz-indices of cell that cutoffGroup is in.
1388	>	whichCell.x() = nCells_.x() * scaled.x();
1389	>	whichCell.y() = nCells_.y() * scaled.y();
1390	>	whichCell.z() = nCells_.z() * scaled.z();
1391	>
1392	>	// find single index of this cell:
1393	>	cellIndex = Vlinear(whichCell, nCells_);
1394	>
1395	>	// add this cutoff group to the list of groups in this cell;
1396	>	cellListCol_[cellIndex].push_back(i);
1397		}
1398	<
1005	<	// find xyz-indices of cell that cutoffGroup is in.
1006	<	whichCell.x() = nCells_.x() * scaled.x();
1007	<	whichCell.y() = nCells_.y() * scaled.y();
1008	<	whichCell.z() = nCells_.z() * scaled.z();
1009	<
1010	<	// find single index of this cell:
1011	<	cellIndex = Vlinear(whichCell, nCells_);
1012	<
1013	<	// add this cutoff group to the list of groups in this cell;
1014	<	cellListCol_[cellIndex].push_back(i);
1015	<	}
1398	>
1399		#else
1400	<	for (int i = 0; i < nGroups_; i++) {
1401	<	rs = snap_->cgData.position[i];
1402	<
1403	<	// scaled positions relative to the box vectors
1404	<	scaled = invHmat * rs;
1405	<
1406	<	// wrap the vector back into the unit box by subtracting integer box
1407	<	// numbers
1408	<	for (int j = 0; j < 3; j++) {
1409	<	scaled[j] -= roundMe(scaled[j]);
1410	<	scaled[j] += 0.5;
1400	>	for (int i = 0; i < nGroups_; i++) {
1401	>	rs = snap_->cgData.position[i];
1402	>
1403	>	// scaled positions relative to the box vectors
1404	>	scaled = invHmat * rs;
1405	>
1406	>	// wrap the vector back into the unit box by subtracting integer box
1407	>	// numbers
1408	>	for (int j = 0; j < 3; j++) {
1409	>	scaled[j] -= roundMe(scaled[j]);
1410	>	scaled[j] += 0.5;
1411	>	// Handle the special case when an object is exactly on the
1412	>	// boundary (a scaled coordinate of 1.0 is the same as
1413	>	// scaled coordinate of 0.0)
1414	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1415	>	}
1416	>
1417	>	// find xyz-indices of cell that cutoffGroup is in.
1418	>	whichCell.x() = nCells_.x() * scaled.x();
1419	>	whichCell.y() = nCells_.y() * scaled.y();
1420	>	whichCell.z() = nCells_.z() * scaled.z();
1421	>
1422	>	// find single index of this cell:
1423	>	cellIndex = Vlinear(whichCell, nCells_);
1424	>
1425	>	// add this cutoff group to the list of groups in this cell;
1426	>	cellList_[cellIndex].push_back(i);
1427		}
1428
1030	–	// find xyz-indices of cell that cutoffGroup is in.
1031	–	whichCell.x() = nCells_.x() * scaled.x();
1032	–	whichCell.y() = nCells_.y() * scaled.y();
1033	–	whichCell.z() = nCells_.z() * scaled.z();
1034	–
1035	–	// find single index of this cell:
1036	–	cellIndex = Vlinear(whichCell, nCells_);
1037	–
1038	–	// add this cutoff group to the list of groups in this cell;
1039	–	cellList_[cellIndex].push_back(i);
1040	–	}
1429		#endif
1430
1431	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1432	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1433	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1434	<	Vector3i m1v(m1x, m1y, m1z);
1435	<	int m1 = Vlinear(m1v, nCells_);
1048	<
1049	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1050	<	os != cellOffsets_.end(); ++os) {
1431	>	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1432	>	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1433	>	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1434	>	Vector3i m1v(m1x, m1y, m1z);
1435	>	int m1 = Vlinear(m1v, nCells_);
1436
1437	<	Vector3i m2v = m1v + (*os);
1438	<
1439	<	if (m2v.x() >= nCells_.x()) {
1440	<	m2v.x() = 0;
1441	<	} else if (m2v.x() < 0) {
1057	<	m2v.x() = nCells_.x() - 1;
1058	<	}
1059	<
1060	<	if (m2v.y() >= nCells_.y()) {
1061	<	m2v.y() = 0;
1062	<	} else if (m2v.y() < 0) {
1063	<	m2v.y() = nCells_.y() - 1;
1064	<	}
1065	<
1066	<	if (m2v.z() >= nCells_.z()) {
1067	<	m2v.z() = 0;
1068	<	} else if (m2v.z() < 0) {
1069	<	m2v.z() = nCells_.z() - 1;
1070	<	}
1071	<
1072	<	int m2 = Vlinear (m2v, nCells_);
1437	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1438	>	os != cellOffsets_.end(); ++os) {
1439	>
1440	>	Vector3i m2v = m1v + (*os);
1441	>
1442
1443	<	#ifdef IS_MPI
1444	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1445	<	j1 != cellListRow_[m1].end(); ++j1) {
1446	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1447	<	j2 != cellListCol_[m2].end(); ++j2) {
1448	<
1449	<	// Always do this if we're in different cells or if
1450	<	// we're in the same cell and the global index of the
1451	<	// j2 cutoff group is less than the j1 cutoff group
1443	>	if (m2v.x() >= nCells_.x()) {
1444	>	m2v.x() = 0;
1445	>	} else if (m2v.x() < 0) {
1446	>	m2v.x() = nCells_.x() - 1;
1447	>	}
1448	>
1449	>	if (m2v.y() >= nCells_.y()) {
1450	>	m2v.y() = 0;
1451	>	} else if (m2v.y() < 0) {
1452	>	m2v.y() = nCells_.y() - 1;
1453	>	}
1454	>
1455	>	if (m2v.z() >= nCells_.z()) {
1456	>	m2v.z() = 0;
1457	>	} else if (m2v.z() < 0) {
1458	>	m2v.z() = nCells_.z() - 1;
1459	>	}
1460
1461	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1461	>	int m2 = Vlinear (m2v, nCells_);
1462	>
1463	>	#ifdef IS_MPI
1464	>	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1465	>	j1 != cellListRow_[m1].end(); ++j1) {
1466	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1467	>	j2 != cellListCol_[m2].end(); ++j2) {
1468	>
1469	>	// In parallel, we need to visit *all* pairs of row
1470	>	// & column indicies and will divide labor in the
1471	>	// force evaluation later.
1472		dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1473		snap_->wrapVector(dr);
1474		cuts = getGroupCutoffs( (*j1), (*j2) );
1475		if (dr.lengthSquare() < cuts.third) {
1476		neighborList.push_back(make_pair((*j1), (*j2)));
1477	<	}
1477	>	}
1478		}
1479		}
1093	–	}
1480		#else
1481	+	for (vector<int>::iterator j1 = cellList_[m1].begin();
1482	+	j1 != cellList_[m1].end(); ++j1) {
1483	+	for (vector<int>::iterator j2 = cellList_[m2].begin();
1484	+	j2 != cellList_[m2].end(); ++j2) {
1485	+
1486	+	// Always do this if we're in different cells or if
1487	+	// we're in the same cell and the global index of
1488	+	// the j2 cutoff group is greater than or equal to
1489	+	// the j1 cutoff group.  Note that Rappaport's code
1490	+	// has a "less than" conditional here, but that
1491	+	// deals with atom-by-atom computation.  OpenMD
1492	+	// allows atoms within a single cutoff group to
1493	+	// interact with each other.
1494
1096	–	for (vector<int>::iterator j1 = cellList_[m1].begin();
1097	–	j1 != cellList_[m1].end(); ++j1) {
1098	–	for (vector<int>::iterator j2 = cellList_[m2].begin();
1099	–	j2 != cellList_[m2].end(); ++j2) {
1495
1101	–	// Always do this if we're in different cells or if
1102	–	// we're in the same cell and the global index of the
1103	–	// j2 cutoff group is less than the j1 cutoff group
1496
1497	<	if (m2 != m1 || (*j2) < (*j1)) {
1498	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1499	<	snap_->wrapVector(dr);
1500	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1501	<	if (dr.lengthSquare() < cuts.third) {
1502	<	neighborList.push_back(make_pair((*j1), (*j2)));
1497	>	if (m2 != m1 || (*j2) >= (*j1) ) {
1498	>
1499	>	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1500	>	snap_->wrapVector(dr);
1501	>	cuts = getGroupCutoffs( (*j1), (*j2) );
1502	>	if (dr.lengthSquare() < cuts.third) {
1503	>	neighborList.push_back(make_pair((*j1), (*j2)));
1504	>	}
1505		}
1506		}
1507		}
1114	–	}
1508		#endif
1509	+	}
1510		}
1511		}
1512		}
1513	+	} else {
1514	+	// branch to do all cutoff group pairs
1515	+	#ifdef IS_MPI
1516	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1517	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1518	+	dr = cgColData.position[j2] - cgRowData.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	+	}
1526	+	#else
1527	+	// include all groups here.
1528	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1529	+	// include self group interactions j2 == j1
1530	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1531	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1532	+	snap_->wrapVector(dr);
1533	+	cuts = getGroupCutoffs( j1, j2 );
1534	+	if (dr.lengthSquare() < cuts.third) {
1535	+	neighborList.push_back(make_pair(j1, j2));
1536	+	}
1537	+	}
1538	+	}
1539	+	#endif
1540		}
1541	<
1541	>
1542		// save the local cutoff group positions for the check that is
1543		// done on each loop:
1544		saved_CG_positions_.clear();
1545		for (int i = 0; i < nGroups_; i++)
1546		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1547	<
1547	>
1548		return neighborList;
1549		}
1550		} //end namespace OpenMD

Diff Legend

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