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 1586 by gezelter, Tue Jun 21 06:34:35 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1796 by gezelter, Mon Sep 10 18:38:44 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();
100		// gather the information for atomtype IDs (atids):
101		idents = info_->getIdentArray();
#	Line 67 | Line 105 | namespace OpenMD {
105
106		massFactors = info_->getMassFactors();
107
108	<	PairList excludes = info_->getExcludedInteractions();
109	<	PairList oneTwo = info_->getOneTwoInteractions();
110	<	PairList oneThree = info_->getOneThreeInteractions();
111	<	PairList oneFour = info_->getOneFourInteractions();
112	<
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_);
79	<	AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
80	<	AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
81	<	AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
121	>	MPI::Intracomm row = rowComm.getComm();
122	>	MPI::Intracomm col = colComm.getComm();
123
124	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
125	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
126	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
127	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
128	<	AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
124	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
125	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
126	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
127	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
128	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
129
130	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
131	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
132	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
133	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
130	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
131	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
132	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
133	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
134	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
135
136	<	nAtomsInRow_ = AtomCommIntRow->getSize();
137	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
138	<	nGroupsInRow_ = cgCommIntRow->getSize();
139	<	nGroupsInCol_ = cgCommIntColumn->getSize();
136	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
137	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
138	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
139	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
140
141	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
142	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
143	+	nGroupsInRow_ = cgPlanIntRow->getSize();
144	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
145	+
146		// Modify the data storage objects with the correct layouts and sizes:
147		atomRowData.resize(nAtomsInRow_);
148		atomRowData.setStorageLayout(storageLayout_);
#	Line 104 | Line 151 | namespace OpenMD {
151		cgRowData.resize(nGroupsInRow_);
152		cgRowData.setStorageLayout(DataStorage::dslPosition);
153		cgColData.resize(nGroupsInCol_);
154	<	cgColData.setStorageLayout(DataStorage::dslPosition);
155	<
154	>	if (needVelocities_)
155	>	// we only need column velocities if we need them.
156	>	cgColData.setStorageLayout(DataStorage::dslPosition |
157	>	DataStorage::dslVelocity);
158	>	else
159	>	cgColData.setStorageLayout(DataStorage::dslPosition);
160	>
161		identsRow.resize(nAtomsInRow_);
162		identsCol.resize(nAtomsInCol_);
163
164	<	AtomCommIntRow->gather(idents, identsRow);
165	<	AtomCommIntColumn->gather(idents, identsCol);
164	>	AtomPlanIntRow->gather(idents, identsRow);
165	>	AtomPlanIntColumn->gather(idents, identsCol);
166
167	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
168	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
169	<
118	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
119	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
167	>	// allocate memory for the parallel objects
168	>	atypesRow.resize(nAtomsInRow_);
169	>	atypesCol.resize(nAtomsInCol_);
170
171	<	AtomCommRealRow->gather(massFactors, massFactorsRow);
172	<	AtomCommRealColumn->gather(massFactors, massFactorsCol);
171	>	for (int i = 0; i < nAtomsInRow_; i++)
172	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
173	>	for (int i = 0; i < nAtomsInCol_; i++)
174	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
175
176	+	pot_row.resize(nAtomsInRow_);
177	+	pot_col.resize(nAtomsInCol_);
178	+
179	+	expot_row.resize(nAtomsInRow_);
180	+	expot_col.resize(nAtomsInCol_);
181	+
182	+	AtomRowToGlobal.resize(nAtomsInRow_);
183	+	AtomColToGlobal.resize(nAtomsInCol_);
184	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186	+
187	+	cgRowToGlobal.resize(nGroupsInRow_);
188	+	cgColToGlobal.resize(nGroupsInCol_);
189	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
191	+
192	+	massFactorsRow.resize(nAtomsInRow_);
193	+	massFactorsCol.resize(nAtomsInCol_);
194	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
195	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
196	+
197		groupListRow_.clear();
198		groupListRow_.resize(nGroupsInRow_);
199		for (int i = 0; i < nGroupsInRow_; i++) {
#	Line 143 | 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 155 | 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 175 | Line 248 | namespace OpenMD {
248		}
249		}
250
251	<	#endif
252	<
253	<	groupList_.clear();
181	<	groupList_.resize(nGroups_);
182	<	for (int i = 0; i < nGroups_; i++) {
183	<	int gid = cgLocalToGlobal[i];
184	<	for (int j = 0; j < nLocal_; j++) {
185	<	int aid = AtomLocalToGlobal[j];
186	<	if (globalGroupMembership[aid] == gid) {
187	<	groupList_[i].push_back(j);
188	<	}
189	<	}
190	<	}
191	<
192	<	skipsForAtom.clear();
193	<	skipsForAtom.resize(nLocal_);
251	>	#else
252	>	excludesForAtom.clear();
253	>	excludesForAtom.resize(nLocal_);
254		toposForAtom.clear();
255		toposForAtom.resize(nLocal_);
256		topoDist.clear();
#	Line 202 | 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 221 | Line 281 | namespace OpenMD {
281		}
282		}
283		}
284	<
284	>	#endif
285	>
286	>	// allocate memory for the parallel objects
287	>	atypesLocal.resize(nLocal_);
288	>
289	>	for (int i = 0; i < nLocal_; i++)
290	>	atypesLocal[i] = ff_->getAtomType(idents[i]);
291	>
292	>	groupList_.clear();
293	>	groupList_.resize(nGroups_);
294	>	for (int i = 0; i < nGroups_; i++) {
295	>	int gid = cgLocalToGlobal[i];
296	>	for (int j = 0; j < nLocal_; j++) {
297	>	int aid = AtomLocalToGlobal[j];
298	>	if (globalGroupMembership[aid] == gid) {
299	>	groupList_[i].push_back(j);
300	>	}
301	>	}
302	>	}
303	>
304	>
305		createGtypeCutoffMap();
306	+
307		}
308
309		void ForceMatrixDecomposition::createGtypeCutoffMap() {
310
311		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	<
241	<	if (userChoseCutoff_)
321	>	if (userChoseCutoff_)
322		atypeCutoff[atid] = userCutoff_;
323		else
324		atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
325		}
326	<
326	>
327		vector<RealType> gTypeCutoffs;
248	–
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 303 | Line 382 | namespace OpenMD {
382
383		vector<RealType> groupCutoff(nGroups_, 0.0);
384		groupToGtype.resize(nGroups_);
306	–
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		}
#	Line 334 | Line 409 | namespace OpenMD {
409
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 367 | Line 444 | namespace OpenMD {
444
445		pair<int,int> key = make_pair(i,j);
446		gTypeCutoffMap[key].first = thisRcut;
370	–
447		if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
372	–
448		gTypeCutoffMap[key].second = thisRcut*thisRcut;
374	–
449		gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
376	–
450		// sanity check
451
452		if (userChoseCutoff_) {
#	Line 390 | Line 463 | namespace OpenMD {
463		}
464		}
465
393	–
466		groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
467		int i, j;
468		#ifdef IS_MPI
#	Line 404 | 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 414 | 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 432 | 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 443 | 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 539 | namespace OpenMD {
539		}
540
541		if (storageLayout_ & DataStorage::dslSkippedCharge) {
542	<	fill(atomRowData.skippedCharge.begin(), atomRowData.skippedCharge.end(), 0.0);
543	<	fill(atomColData.skippedCharge.begin(), atomColData.skippedCharge.end(), 0.0);
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	<	#else
549	<
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	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
563	>	fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(),
564	>	0.0);
565	>	fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.end(),
566	>	0.0);
567	>	}
568	>
569	>	#endif
570	>	// even in parallel, we need to zero out the local arrays:
571	>
572		if (storageLayout_ & DataStorage::dslParticlePot) {
573		fill(snap_->atomData.particlePot.begin(),
574		snap_->atomData.particlePot.end(), 0.0);
#	Line 470 | Line 578 | namespace OpenMD {
578		fill(snap_->atomData.density.begin(),
579		snap_->atomData.density.end(), 0.0);
580		}
581	+
582		if (storageLayout_ & DataStorage::dslFunctional) {
583		fill(snap_->atomData.functional.begin(),
584		snap_->atomData.functional.end(), 0.0);
585		}
586	+
587		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
588		fill(snap_->atomData.functionalDerivative.begin(),
589		snap_->atomData.functionalDerivative.end(), 0.0);
590		}
591	+
592		if (storageLayout_ & DataStorage::dslSkippedCharge) {
593		fill(snap_->atomData.skippedCharge.begin(),
594		snap_->atomData.skippedCharge.end(), 0.0);
595		}
596	<	#endif
597	<
596	>
597	>	if (storageLayout_ & DataStorage::dslElectricField) {
598	>	fill(snap_->atomData.electricField.begin(),
599	>	snap_->atomData.electricField.end(), V3Zero);
600	>	}
601		}
602
603
#	Line 493 | Line 607 | namespace OpenMD {
607		#ifdef IS_MPI
608
609		// gather up the atomic positions
610	<	AtomCommVectorRow->gather(snap_->atomData.position,
610	>	AtomPlanVectorRow->gather(snap_->atomData.position,
611		atomRowData.position);
612	<	AtomCommVectorColumn->gather(snap_->atomData.position,
612	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
613		atomColData.position);
614
615		// gather up the cutoff group positions
616	<	cgCommVectorRow->gather(snap_->cgData.position,
616	>
617	>	cgPlanVectorRow->gather(snap_->cgData.position,
618		cgRowData.position);
619	<	cgCommVectorColumn->gather(snap_->cgData.position,
619	>
620	>	cgPlanVectorColumn->gather(snap_->cgData.position,
621		cgColData.position);
622	+
623	+
624	+
625	+	if (needVelocities_) {
626	+	// gather up the atomic velocities
627	+	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
628	+	atomColData.velocity);
629	+
630	+	cgPlanVectorColumn->gather(snap_->cgData.velocity,
631	+	cgColData.velocity);
632	+	}
633	+
634
635		// if needed, gather the atomic rotation matrices
636		if (storageLayout_ & DataStorage::dslAmat) {
637	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
637	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
638		atomRowData.aMat);
639	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
639	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
640		atomColData.aMat);
641		}
642
643		// if needed, gather the atomic eletrostatic frames
644		if (storageLayout_ & DataStorage::dslElectroFrame) {
645	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
645	>	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
646		atomRowData.electroFrame);
647	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
647	>	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
648		atomColData.electroFrame);
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 532 | 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, field_tmp);
690	+	for (int i = 0; i < n; i++)
691	+	snap_->atomData.electricField[i] += field_tmp[i];
692	+	}
693		#endif
694		}
695
#	Line 553 | Line 702 | namespace OpenMD {
702		storageLayout_ = sman_->getStorageLayout();
703		#ifdef IS_MPI
704		if (storageLayout_ & DataStorage::dslFunctional) {
705	<	AtomCommRealRow->gather(snap_->atomData.functional,
705	>	AtomPlanRealRow->gather(snap_->atomData.functional,
706		atomRowData.functional);
707	<	AtomCommRealColumn->gather(snap_->atomData.functional,
707	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
708		atomColData.functional);
709		}
710
711		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
712	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
712	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
713		atomRowData.functionalDerivative);
714	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
714	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
715		atomColData.functionalDerivative);
716		}
717		#endif
#	Line 576 | Line 725 | namespace OpenMD {
725		int n = snap_->atomData.force.size();
726		vector<Vector3d> frc_tmp(n, V3Zero);
727
728	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
728	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
729		for (int i = 0; i < n; i++) {
730		snap_->atomData.force[i] += frc_tmp[i];
731		frc_tmp[i] = 0.0;
732		}
733
734	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
735	<	for (int i = 0; i < n; i++)
734	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
735	>	for (int i = 0; i < n; i++) {
736		snap_->atomData.force[i] += frc_tmp[i];
737	<
738	<
737	>	}
738	>
739		if (storageLayout_ & DataStorage::dslTorque) {
740
741	<	int nt = snap_->atomData.force.size();
741	>	int nt = snap_->atomData.torque.size();
742		vector<Vector3d> trq_tmp(nt, V3Zero);
743
744	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
745	<	for (int i = 0; i < n; i++) {
744	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
745	>	for (int i = 0; i < nt; i++) {
746		snap_->atomData.torque[i] += trq_tmp[i];
747		trq_tmp[i] = 0.0;
748		}
749
750	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
751	<	for (int i = 0; i < n; i++)
750	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
751	>	for (int i = 0; i < nt; i++)
752		snap_->atomData.torque[i] += trq_tmp[i];
753		}
754	+
755	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
756	+
757	+	int ns = snap_->atomData.skippedCharge.size();
758	+	vector<RealType> skch_tmp(ns, 0.0);
759	+
760	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
761	+	for (int i = 0; i < ns; i++) {
762	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
763	+	skch_tmp[i] = 0.0;
764	+	}
765	+
766	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
767	+	for (int i = 0; i < ns; i++)
768	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
769	+
770	+	}
771
772	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
773	+
774	+	int nq = snap_->atomData.flucQFrc.size();
775	+	vector<RealType> fqfrc_tmp(nq, 0.0);
776	+
777	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
778	+	for (int i = 0; i < nq; i++) {
779	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
780	+	fqfrc_tmp[i] = 0.0;
781	+	}
782	+
783	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
784	+	for (int i = 0; i < nq; i++)
785	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
786	+
787	+	}
788	+
789		nLocal_ = snap_->getNumberOfAtoms();
790
791		vector<potVec> pot_temp(nLocal_,
792		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
793	+	vector<potVec> expot_temp(nLocal_,
794	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
795
796		// scatter/gather pot_row into the members of my column
797
798	<	AtomCommPotRow->scatter(pot_row, pot_temp);
798	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
799	>	AtomPlanPotRow->scatter(expot_row, expot_temp);
800
801	<	for (int ii = 0;  ii < pot_temp.size(); ii++ )
801	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
802		pairwisePot += pot_temp[ii];
803	<
803	>
804	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
805	>	excludedPot += expot_temp[ii];
806	>
807	>	if (storageLayout_ & DataStorage::dslParticlePot) {
808	>	// This is the pairwise contribution to the particle pot.  The
809	>	// embedding contribution is added in each of the low level
810	>	// non-bonded routines.  In single processor, this is done in
811	>	// unpackInteractionData, not in collectData.
812	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
813	>	for (int i = 0; i < nLocal_; i++) {
814	>	// factor of two is because the total potential terms are divided
815	>	// by 2 in parallel due to row/ column scatter
816	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
817	>	}
818	>	}
819	>	}
820	>
821		fill(pot_temp.begin(), pot_temp.end(),
822		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
823	+	fill(expot_temp.begin(), expot_temp.end(),
824	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
825
826	<	AtomCommPotColumn->scatter(pot_col, pot_temp);
826	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
827	>	AtomPlanPotColumn->scatter(expot_col, expot_temp);
828
829		for (int ii = 0;  ii < pot_temp.size(); ii++ )
830		pairwisePot += pot_temp[ii];
831	+
832	+	for (int ii = 0;  ii < expot_temp.size(); ii++ )
833	+	excludedPot += expot_temp[ii];
834	+
835	+	if (storageLayout_ & DataStorage::dslParticlePot) {
836	+	// This is the pairwise contribution to the particle pot.  The
837	+	// embedding contribution is added in each of the low level
838	+	// non-bonded routines.  In single processor, this is done in
839	+	// unpackInteractionData, not in collectData.
840	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
841	+	for (int i = 0; i < nLocal_; i++) {
842	+	// factor of two is because the total potential terms are divided
843	+	// by 2 in parallel due to row/ column scatter
844	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
845	+	}
846	+	}
847	+	}
848	+
849	+	if (storageLayout_ & DataStorage::dslParticlePot) {
850	+	int npp = snap_->atomData.particlePot.size();
851	+	vector<RealType> ppot_temp(npp, 0.0);
852	+
853	+	// This is the direct or embedding contribution to the particle
854	+	// pot.
855	+
856	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
857	+	for (int i = 0; i < npp; i++) {
858	+	snap_->atomData.particlePot[i] += ppot_temp[i];
859	+	}
860	+
861	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
862	+
863	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
864	+	for (int i = 0; i < npp; i++) {
865	+	snap_->atomData.particlePot[i] += ppot_temp[i];
866	+	}
867	+	}
868	+
869	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
870	+	RealType ploc1 = pairwisePot[ii];
871	+	RealType ploc2 = 0.0;
872	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
873	+	pairwisePot[ii] = ploc2;
874	+	}
875	+
876	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
877	+	RealType ploc1 = excludedPot[ii];
878	+	RealType ploc2 = 0.0;
879	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
880	+	excludedPot[ii] = ploc2;
881	+	}
882	+
883	+	// Here be dragons.
884	+	MPI::Intracomm col = colComm.getComm();
885	+
886	+	col.Allreduce(MPI::IN_PLACE,
887	+	&snap_->frameData.conductiveHeatFlux[0], 3,
888	+	MPI::REALTYPE, MPI::SUM);
889	+
890	+
891		#endif
892
893	+	}
894	+
895	+	/**
896	+	* Collects information obtained during the post-pair (and embedding
897	+	* functional) loops onto local data structures.
898	+	*/
899	+	void ForceMatrixDecomposition::collectSelfData() {
900	+	snap_ = sman_->getCurrentSnapshot();
901	+	storageLayout_ = sman_->getStorageLayout();
902	+
903	+	#ifdef IS_MPI
904	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
905	+	RealType ploc1 = embeddingPot[ii];
906	+	RealType ploc2 = 0.0;
907	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
908	+	embeddingPot[ii] = ploc2;
909	+	}
910	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
911	+	RealType ploc1 = excludedSelfPot[ii];
912	+	RealType ploc2 = 0.0;
913	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
914	+	excludedSelfPot[ii] = ploc2;
915	+	}
916	+	#endif
917	+
918		}
919
920	+
921	+
922		int ForceMatrixDecomposition::getNAtomsInRow() {
923		#ifdef IS_MPI
924		return nAtomsInRow_;
#	Line 666 | Line 959 | namespace OpenMD {
959		return d;
960		}
961
962	+	Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
963	+	#ifdef IS_MPI
964	+	return cgColData.velocity[cg2];
965	+	#else
966	+	return snap_->cgData.velocity[cg2];
967	+	#endif
968	+	}
969
970	+	Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
971	+	#ifdef IS_MPI
972	+	return atomColData.velocity[atom2];
973	+	#else
974	+	return snap_->atomData.velocity[atom2];
975	+	#endif
976	+	}
977	+
978	+
979		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
980
981		Vector3d d;
#	Line 724 | Line 1033 | namespace OpenMD {
1033		return d;
1034		}
1035
1036	<	vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
1037	<	return skipsForAtom[atom1];
1036	>	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1037	>	return excludesForAtom[atom1];
1038		}
1039
1040		/**
1041	<	* There are a number of reasons to skip a pair or a
733	<	* particle. Mostly we do this to exclude atoms who are involved in
734	<	* short range interactions (bonds, bends, torsions), but we also
735	<	* need to exclude some overcounted interactions that result from
1041	>	* We need to exclude some overcounted interactions that result from
1042		* the parallel decomposition.
1043		*/
1044	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1044	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1045		int unique_id_1, unique_id_2;
1046	<
1046	>
1047		#ifdef IS_MPI
1048		// in MPI, we have to look up the unique IDs for each atom
1049		unique_id_1 = AtomRowToGlobal[atom1];
1050		unique_id_2 = AtomColToGlobal[atom2];
1051	+	// group1 = cgRowToGlobal[cg1];
1052	+	// group2 = cgColToGlobal[cg2];
1053	+	#else
1054	+	unique_id_1 = AtomLocalToGlobal[atom1];
1055	+	unique_id_2 = AtomLocalToGlobal[atom2];
1056	+	int group1 = cgLocalToGlobal[cg1];
1057	+	int group2 = cgLocalToGlobal[cg2];
1058	+	#endif
1059
746	–	// this situation should only arise in MPI simulations
1060		if (unique_id_1 == unique_id_2) return true;
1061	<
1061	>
1062	>	#ifdef IS_MPI
1063		// this prevents us from doing the pair on multiple processors
1064		if (unique_id_1 < unique_id_2) {
1065		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1066		} else {
1067	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1067	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1068		}
1069	<	#else
1070	<	// in the normal loop, the atom numbers are unique
1071	<	unique_id_1 = atom1;
1072	<	unique_id_2 = atom2;
1069	>	#endif
1070	>
1071	>	#ifndef IS_MPI
1072	>	if (group1 == group2) {
1073	>	if (unique_id_1 < unique_id_2) return true;
1074	>	}
1075		#endif
1076
1077	<	for (vector<int>::iterator i = skipsForAtom[atom1].begin();
1078	<	i != skipsForAtom[atom1].end(); ++i) {
1079	<	if ( (*i) == unique_id_2 ) return true;
1077	>	return false;
1078	>	}
1079	>
1080	>	/**
1081	>	* We need to handle the interactions for atoms who are involved in
1082	>	* the same rigid body as well as some short range interactions
1083	>	* (bonds, bends, torsions) differently from other interactions.
1084	>	* We'll still visit the pairwise routines, but with a flag that
1085	>	* tells those routines to exclude the pair from direct long range
1086	>	* interactions.  Some indirect interactions (notably reaction
1087	>	* field) must still be handled for these pairs.
1088	>	*/
1089	>	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1090	>
1091	>	// excludesForAtom was constructed to use row/column indices in the MPI
1092	>	// version, and to use local IDs in the non-MPI version:
1093	>
1094	>	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1095	>	i != excludesForAtom[atom1].end(); ++i) {
1096	>	if ( (*i) == atom2 ) return true;
1097		}
1098
1099		return false;
#	Line 785 | Line 1118 | namespace OpenMD {
1118
1119		// filling interaction blocks with pointers
1120		void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1121	<	int atom1, int atom2) {
1121	>	int atom1, int atom2) {
1122	>
1123	>	idat.excluded = excludeAtomPair(atom1, atom2);
1124	>
1125		#ifdef IS_MPI
1126	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1127	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1128	+	//                         ff_->getAtomType(identsCol[atom2]) );
1129
791	–	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
792	–	ff_->getAtomType(identsCol[atom2]) );
793	–
1130		if (storageLayout_ & DataStorage::dslAmat) {
1131		idat.A1 = &(atomRowData.aMat[atom1]);
1132		idat.A2 = &(atomColData.aMat[atom2]);
#	Line 826 | Line 1162 | namespace OpenMD {
1162		idat.particlePot2 = &(atomColData.particlePot[atom2]);
1163		}
1164
1165	<	#else
1165	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1166	>	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1167	>	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1168	>	}
1169
1170	<	idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1171	<	ff_->getAtomType(idents[atom2]) );
1170	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1171	>	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1172	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1173	>	}
1174
1175	+	#else
1176	+
1177	+	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1178	+
1179		if (storageLayout_ & DataStorage::dslAmat) {
1180		idat.A1 = &(snap_->atomData.aMat[atom1]);
1181		idat.A2 = &(snap_->atomData.aMat[atom2]);
#	Line 866 | Line 1211 | namespace OpenMD {
1211		idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1212		}
1213
1214	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1215	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1216	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1217	+	}
1218	+
1219	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1220	+	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1221	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1222	+	}
1223	+
1224		#endif
1225		}
1226
1227
1228		void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1229		#ifdef IS_MPI
1230	<	pot_row[atom1] += 0.5 *  *(idat.pot);
1231	<	pot_col[atom2] += 0.5 *  *(idat.pot);
1230	>	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1231	>	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1232	>	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1233	>	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1234
1235		atomRowData.force[atom1] += *(idat.f1);
1236		atomColData.force[atom2] -= *(idat.f1);
1237	+
1238	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1239	+	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1240	+	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1241	+	}
1242	+
1243	+	if (storageLayout_ & DataStorage::dslElectricField) {
1244	+	atomRowData.electricField[atom1] += *(idat.eField1);
1245	+	atomColData.electricField[atom2] += *(idat.eField2);
1246	+	}
1247	+
1248		#else
1249		pairwisePot += *(idat.pot);
1250	+	excludedPot += *(idat.excludedPot);
1251
1252		snap_->atomData.force[atom1] += *(idat.f1);
1253		snap_->atomData.force[atom2] -= *(idat.f1);
885	–	#endif
886	–
887	–	}
1254
1255	<
1256	<	void ForceMatrixDecomposition::fillSkipData(InteractionData &idat,
1257	<	int atom1, int atom2) {
1258	<	#ifdef IS_MPI
1259	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1260	<	ff_->getAtomType(identsCol[atom2]) );
1261	<
896	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
897	<	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
898	<	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1255	>	if (idat.doParticlePot) {
1256	>	// This is the pairwise contribution to the particle pot.  The
1257	>	// embedding contribution is added in each of the low level
1258	>	// non-bonded routines.  In parallel, this calculation is done
1259	>	// in collectData, not in unpackInteractionData.
1260	>	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1261	>	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1262		}
1263	<
1264	<	if (storageLayout_ & DataStorage::dslTorque) {
1265	<	idat.t1 = &(atomRowData.torque[atom1]);
1266	<	idat.t2 = &(atomColData.torque[atom2]);
1263	>
1264	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1265	>	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1266	>	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1267		}
1268
1269	<	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1270	<	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1271	<	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1269	>	if (storageLayout_ & DataStorage::dslElectricField) {
1270	>	snap_->atomData.electricField[atom1] += *(idat.eField1);
1271	>	snap_->atomData.electricField[atom2] += *(idat.eField2);
1272		}
910	–	#else
911	–	idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
912	–	ff_->getAtomType(idents[atom2]) );
1273
914	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
915	–	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
916	–	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
917	–	}
918	–
919	–	if (storageLayout_ & DataStorage::dslTorque) {
920	–	idat.t1 = &(snap_->atomData.torque[atom1]);
921	–	idat.t2 = &(snap_->atomData.torque[atom2]);
922	–	}
923	–
924	–	if (storageLayout_ & DataStorage::dslSkippedCharge) {
925	–	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
926	–	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
927	–	}
928	–	#endif
929	–	}
930	–
931	–
932	–	void ForceMatrixDecomposition::unpackSkipData(InteractionData &idat, int atom1, int atom2) {
933	–	#ifdef IS_MPI
934	–	pot_row[atom1] += 0.5 *  *(idat.pot);
935	–	pot_col[atom2] += 0.5 *  *(idat.pot);
936	–	#else
937	–	pairwisePot += *(idat.pot);
1274		#endif
1275	<
1275	>
1276		}
1277
942	–
1278		/*
1279		* buildNeighborList
1280		*
#	Line 950 | Line 1285 | namespace OpenMD {
1285
1286		vector<pair<int, int> > neighborList;
1287		groupCutoffs cuts;
1288	+	bool doAllPairs = false;
1289	+
1290		#ifdef IS_MPI
1291		cellListRow_.clear();
1292		cellListCol_.clear();
#	Line 958 | Line 1295 | namespace OpenMD {
1295		#endif
1296
1297		RealType rList_ = (largestRcut_ + skinThickness_);
961	–	RealType rl2 = rList_ * rList_;
1298		Snapshot* snap_ = sman_->getCurrentSnapshot();
1299		Mat3x3d Hmat = snap_->getHmat();
1300		Vector3d Hx = Hmat.getColumn(0);
#	Line 969 | Line 1305 | namespace OpenMD {
1305		nCells_.y() = (int) ( Hy.length() )/ rList_;
1306		nCells_.z() = (int) ( Hz.length() )/ rList_;
1307
1308	+	// handle small boxes where the cell offsets can end up repeating cells
1309	+
1310	+	if (nCells_.x() < 3) doAllPairs = true;
1311	+	if (nCells_.y() < 3) doAllPairs = true;
1312	+	if (nCells_.z() < 3) doAllPairs = true;
1313	+
1314		Mat3x3d invHmat = snap_->getInvHmat();
1315		Vector3d rs, scaled, dr;
1316		Vector3i whichCell;
#	Line 982 | Line 1324 | namespace OpenMD {
1324		cellList_.resize(nCtot);
1325		#endif
1326
1327	+	if (!doAllPairs) {
1328		#ifdef IS_MPI
986	–	for (int i = 0; i < nGroupsInRow_; i++) {
987	–	rs = cgRowData.position[i];
1329
1330	<	// scaled positions relative to the box vectors
1331	<	scaled = invHmat * rs;
1332	<
1333	<	// wrap the vector back into the unit box by subtracting integer box
1334	<	// numbers
1335	<	for (int j = 0; j < 3; j++) {
1336	<	scaled[j] -= roundMe(scaled[j]);
1337	<	scaled[j] += 0.5;
1330	>	for (int i = 0; i < nGroupsInRow_; i++) {
1331	>	rs = cgRowData.position[i];
1332	>
1333	>	// scaled positions relative to the box vectors
1334	>	scaled = invHmat * rs;
1335	>
1336	>	// wrap the vector back into the unit box by subtracting integer box
1337	>	// numbers
1338	>	for (int j = 0; j < 3; j++) {
1339	>	scaled[j] -= roundMe(scaled[j]);
1340	>	scaled[j] += 0.5;
1341	>	// Handle the special case when an object is exactly on the
1342	>	// boundary (a scaled coordinate of 1.0 is the same as
1343	>	// scaled coordinate of 0.0)
1344	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1345	>	}
1346	>
1347	>	// find xyz-indices of cell that cutoffGroup is in.
1348	>	whichCell.x() = nCells_.x() * scaled.x();
1349	>	whichCell.y() = nCells_.y() * scaled.y();
1350	>	whichCell.z() = nCells_.z() * scaled.z();
1351	>
1352	>	// find single index of this cell:
1353	>	cellIndex = Vlinear(whichCell, nCells_);
1354	>
1355	>	// add this cutoff group to the list of groups in this cell;
1356	>	cellListRow_[cellIndex].push_back(i);
1357	>	}
1358	>	for (int i = 0; i < nGroupsInCol_; i++) {
1359	>	rs = cgColData.position[i];
1360	>
1361	>	// scaled positions relative to the box vectors
1362	>	scaled = invHmat * rs;
1363	>
1364	>	// wrap the vector back into the unit box by subtracting integer box
1365	>	// numbers
1366	>	for (int j = 0; j < 3; j++) {
1367	>	scaled[j] -= roundMe(scaled[j]);
1368	>	scaled[j] += 0.5;
1369	>	// Handle the special case when an object is exactly on the
1370	>	// boundary (a scaled coordinate of 1.0 is the same as
1371	>	// scaled coordinate of 0.0)
1372	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1373	>	}
1374	>
1375	>	// find xyz-indices of cell that cutoffGroup is in.
1376	>	whichCell.x() = nCells_.x() * scaled.x();
1377	>	whichCell.y() = nCells_.y() * scaled.y();
1378	>	whichCell.z() = nCells_.z() * scaled.z();
1379	>
1380	>	// find single index of this cell:
1381	>	cellIndex = Vlinear(whichCell, nCells_);
1382	>
1383	>	// add this cutoff group to the list of groups in this cell;
1384	>	cellListCol_[cellIndex].push_back(i);
1385		}
1386
999	–	// find xyz-indices of cell that cutoffGroup is in.
1000	–	whichCell.x() = nCells_.x() * scaled.x();
1001	–	whichCell.y() = nCells_.y() * scaled.y();
1002	–	whichCell.z() = nCells_.z() * scaled.z();
1003	–
1004	–	// find single index of this cell:
1005	–	cellIndex = Vlinear(whichCell, nCells_);
1006	–
1007	–	// add this cutoff group to the list of groups in this cell;
1008	–	cellListRow_[cellIndex].push_back(i);
1009	–	}
1010	–
1011	–	for (int i = 0; i < nGroupsInCol_; i++) {
1012	–	rs = cgColData.position[i];
1013	–
1014	–	// scaled positions relative to the box vectors
1015	–	scaled = invHmat * rs;
1016	–
1017	–	// wrap the vector back into the unit box by subtracting integer box
1018	–	// numbers
1019	–	for (int j = 0; j < 3; j++) {
1020	–	scaled[j] -= roundMe(scaled[j]);
1021	–	scaled[j] += 0.5;
1022	–	}
1023	–
1024	–	// find xyz-indices of cell that cutoffGroup is in.
1025	–	whichCell.x() = nCells_.x() * scaled.x();
1026	–	whichCell.y() = nCells_.y() * scaled.y();
1027	–	whichCell.z() = nCells_.z() * scaled.z();
1028	–
1029	–	// find single index of this cell:
1030	–	cellIndex = Vlinear(whichCell, nCells_);
1031	–
1032	–	// add this cutoff group to the list of groups in this cell;
1033	–	cellListCol_[cellIndex].push_back(i);
1034	–	}
1387		#else
1388	<	for (int i = 0; i < nGroups_; i++) {
1389	<	rs = snap_->cgData.position[i];
1390	<
1391	<	// scaled positions relative to the box vectors
1392	<	scaled = invHmat * rs;
1393	<
1394	<	// wrap the vector back into the unit box by subtracting integer box
1395	<	// numbers
1396	<	for (int j = 0; j < 3; j++) {
1397	<	scaled[j] -= roundMe(scaled[j]);
1398	<	scaled[j] += 0.5;
1388	>	for (int i = 0; i < nGroups_; i++) {
1389	>	rs = snap_->cgData.position[i];
1390	>
1391	>	// scaled positions relative to the box vectors
1392	>	scaled = invHmat * rs;
1393	>
1394	>	// wrap the vector back into the unit box by subtracting integer box
1395	>	// numbers
1396	>	for (int j = 0; j < 3; j++) {
1397	>	scaled[j] -= roundMe(scaled[j]);
1398	>	scaled[j] += 0.5;
1399	>	// Handle the special case when an object is exactly on the
1400	>	// boundary (a scaled coordinate of 1.0 is the same as
1401	>	// scaled coordinate of 0.0)
1402	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1403	>	}
1404	>
1405	>	// find xyz-indices of cell that cutoffGroup is in.
1406	>	whichCell.x() = nCells_.x() * scaled.x();
1407	>	whichCell.y() = nCells_.y() * scaled.y();
1408	>	whichCell.z() = nCells_.z() * scaled.z();
1409	>
1410	>	// find single index of this cell:
1411	>	cellIndex = Vlinear(whichCell, nCells_);
1412	>
1413	>	// add this cutoff group to the list of groups in this cell;
1414	>	cellList_[cellIndex].push_back(i);
1415		}
1416
1049	–	// find xyz-indices of cell that cutoffGroup is in.
1050	–	whichCell.x() = nCells_.x() * scaled.x();
1051	–	whichCell.y() = nCells_.y() * scaled.y();
1052	–	whichCell.z() = nCells_.z() * scaled.z();
1053	–
1054	–	// find single index of this cell:
1055	–	cellIndex = Vlinear(whichCell, nCells_);
1056	–
1057	–	// add this cutoff group to the list of groups in this cell;
1058	–	cellList_[cellIndex].push_back(i);
1059	–	}
1417		#endif
1418
1419	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1420	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1421	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1422	<	Vector3i m1v(m1x, m1y, m1z);
1423	<	int m1 = Vlinear(m1v, nCells_);
1067	<
1068	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1069	<	os != cellOffsets_.end(); ++os) {
1419	>	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1420	>	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1421	>	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1422	>	Vector3i m1v(m1x, m1y, m1z);
1423	>	int m1 = Vlinear(m1v, nCells_);
1424
1425	<	Vector3i m2v = m1v + (*os);
1426	<
1427	<	if (m2v.x() >= nCells_.x()) {
1428	<	m2v.x() = 0;
1429	<	} else if (m2v.x() < 0) {
1076	<	m2v.x() = nCells_.x() - 1;
1077	<	}
1078	<
1079	<	if (m2v.y() >= nCells_.y()) {
1080	<	m2v.y() = 0;
1081	<	} else if (m2v.y() < 0) {
1082	<	m2v.y() = nCells_.y() - 1;
1083	<	}
1084	<
1085	<	if (m2v.z() >= nCells_.z()) {
1086	<	m2v.z() = 0;
1087	<	} else if (m2v.z() < 0) {
1088	<	m2v.z() = nCells_.z() - 1;
1089	<	}
1090	<
1091	<	int m2 = Vlinear (m2v, nCells_);
1425	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1426	>	os != cellOffsets_.end(); ++os) {
1427	>
1428	>	Vector3i m2v = m1v + (*os);
1429	>
1430
1431	<	#ifdef IS_MPI
1432	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1433	<	j1 != cellListRow_[m1].end(); ++j1) {
1434	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1435	<	j2 != cellListCol_[m2].end(); ++j2) {
1436	<
1437	<	// Always do this if we're in different cells or if
1438	<	// we're in the same cell and the global index of the
1439	<	// j2 cutoff group is less than the j1 cutoff group
1431	>	if (m2v.x() >= nCells_.x()) {
1432	>	m2v.x() = 0;
1433	>	} else if (m2v.x() < 0) {
1434	>	m2v.x() = nCells_.x() - 1;
1435	>	}
1436	>
1437	>	if (m2v.y() >= nCells_.y()) {
1438	>	m2v.y() = 0;
1439	>	} else if (m2v.y() < 0) {
1440	>	m2v.y() = nCells_.y() - 1;
1441	>	}
1442	>
1443	>	if (m2v.z() >= nCells_.z()) {
1444	>	m2v.z() = 0;
1445	>	} else if (m2v.z() < 0) {
1446	>	m2v.z() = nCells_.z() - 1;
1447	>	}
1448
1449	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1449	>	int m2 = Vlinear (m2v, nCells_);
1450	>
1451	>	#ifdef IS_MPI
1452	>	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1453	>	j1 != cellListRow_[m1].end(); ++j1) {
1454	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1455	>	j2 != cellListCol_[m2].end(); ++j2) {
1456	>
1457	>	// In parallel, we need to visit *all* pairs of row
1458	>	// & column indicies and will divide labor in the
1459	>	// force evaluation later.
1460		dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1461		snap_->wrapVector(dr);
1462		cuts = getGroupCutoffs( (*j1), (*j2) );
1463		if (dr.lengthSquare() < cuts.third) {
1464		neighborList.push_back(make_pair((*j1), (*j2)));
1465	<	}
1465	>	}
1466		}
1467		}
1112	–	}
1468		#else
1469	+	for (vector<int>::iterator j1 = cellList_[m1].begin();
1470	+	j1 != cellList_[m1].end(); ++j1) {
1471	+	for (vector<int>::iterator j2 = cellList_[m2].begin();
1472	+	j2 != cellList_[m2].end(); ++j2) {
1473	+
1474	+	// Always do this if we're in different cells or if
1475	+	// we're in the same cell and the global index of
1476	+	// the j2 cutoff group is greater than or equal to
1477	+	// the j1 cutoff group.  Note that Rappaport's code
1478	+	// has a "less than" conditional here, but that
1479	+	// deals with atom-by-atom computation.  OpenMD
1480	+	// allows atoms within a single cutoff group to
1481	+	// interact with each other.
1482
1115	–	for (vector<int>::iterator j1 = cellList_[m1].begin();
1116	–	j1 != cellList_[m1].end(); ++j1) {
1117	–	for (vector<int>::iterator j2 = cellList_[m2].begin();
1118	–	j2 != cellList_[m2].end(); ++j2) {
1483
1120	–	// Always do this if we're in different cells or if
1121	–	// we're in the same cell and the global index of the
1122	–	// j2 cutoff group is less than the j1 cutoff group
1484
1485	<	if (m2 != m1 || (*j2) < (*j1)) {
1486	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1487	<	snap_->wrapVector(dr);
1488	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1489	<	if (dr.lengthSquare() < cuts.third) {
1490	<	neighborList.push_back(make_pair((*j1), (*j2)));
1485	>	if (m2 != m1 || (*j2) >= (*j1) ) {
1486	>
1487	>	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1488	>	snap_->wrapVector(dr);
1489	>	cuts = getGroupCutoffs( (*j1), (*j2) );
1490	>	if (dr.lengthSquare() < cuts.third) {
1491	>	neighborList.push_back(make_pair((*j1), (*j2)));
1492	>	}
1493		}
1494		}
1495		}
1133	–	}
1496		#endif
1497	+	}
1498		}
1499		}
1500		}
1501	+	} else {
1502	+	// branch to do all cutoff group pairs
1503	+	#ifdef IS_MPI
1504	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1505	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1506	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1507	+	snap_->wrapVector(dr);
1508	+	cuts = getGroupCutoffs( j1, j2 );
1509	+	if (dr.lengthSquare() < cuts.third) {
1510	+	neighborList.push_back(make_pair(j1, j2));
1511	+	}
1512	+	}
1513	+	}
1514	+	#else
1515	+	// include all groups here.
1516	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1517	+	// include self group interactions j2 == j1
1518	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1519	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1520	+	snap_->wrapVector(dr);
1521	+	cuts = getGroupCutoffs( j1, j2 );
1522	+	if (dr.lengthSquare() < cuts.third) {
1523	+	neighborList.push_back(make_pair(j1, j2));
1524	+	}
1525	+	}
1526	+	}
1527	+	#endif
1528		}
1529	<
1529	>
1530		// save the local cutoff group positions for the check that is
1531		// done on each loop:
1532		saved_CG_positions_.clear();
1533		for (int i = 0; i < nGroups_; i++)
1534		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1535	<
1535	>
1536		return neighborList;
1537		}
1538		} //end namespace OpenMD

Diff Legend

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