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 1581 by gezelter, Mon Jun 13 22:13:12 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 2057 by gezelter, Tue Mar 3 15:22:26 2015 UTC

#	Line 35 | Line 35
35		*
36		* [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37		* [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38	<	* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
39	<	* [4]  Vardeman & Gezelter, in progress (2009).
38	>	* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008).
39	>	* [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40	>	* [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41		*/
42		#include "parallel/ForceMatrixDecomposition.hpp"
43		#include "math/SquareMatrix3.hpp"
#	Line 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	+	// Row and colum scans must visit all surrounding cells
54	+	cellOffsets_.clear();
55	+	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
56	+	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
57	+	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
58	+	cellOffsets_.push_back( Vector3i(-1, 0,-1) );
59	+	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
60	+	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
61	+	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
62	+	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
63	+	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
64	+	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
65	+	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
66	+	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
67	+	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
68	+	cellOffsets_.push_back( Vector3i( 0, 0, 0) );
69	+	cellOffsets_.push_back( Vector3i( 1, 0, 0) );
70	+	cellOffsets_.push_back( Vector3i(-1, 1, 0) );
71	+	cellOffsets_.push_back( Vector3i( 0, 1, 0) );
72	+	cellOffsets_.push_back( Vector3i( 1, 1, 0) );
73	+	cellOffsets_.push_back( Vector3i(-1,-1, 1) );
74	+	cellOffsets_.push_back( Vector3i( 0,-1, 1) );
75	+	cellOffsets_.push_back( Vector3i( 1,-1, 1) );
76	+	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
77	+	cellOffsets_.push_back( Vector3i( 0, 0, 1) );
78	+	cellOffsets_.push_back( Vector3i( 1, 0, 1) );
79	+	cellOffsets_.push_back( Vector3i(-1, 1, 1) );
80	+	cellOffsets_.push_back( Vector3i( 0, 1, 1) );
81	+	cellOffsets_.push_back( Vector3i( 1, 1, 1) );
82	+	}
83	+
84	+
85		/**
86		* distributeInitialData is essentially a copy of the older fortran
87		* SimulationSetup
88		*/
54	–
89		void ForceMatrixDecomposition::distributeInitialData() {
90		snap_ = sman_->getCurrentSnapshot();
91		storageLayout_ = sman_->getStorageLayout();
92		ff_ = info_->getForceField();
93		nLocal_ = snap_->getNumberOfAtoms();
94	<
94	>
95		nGroups_ = info_->getNLocalCutoffGroups();
62	–	cerr << "in dId, nGroups = " << nGroups_ << "\n";
96		// gather the information for atomtype IDs (atids):
97	<	identsLocal = info_->getIdentArray();
97	>	idents = info_->getIdentArray();
98	>	regions = info_->getRegions();
99		AtomLocalToGlobal = info_->getGlobalAtomIndices();
100		cgLocalToGlobal = info_->getGlobalGroupIndices();
101		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
102	+
103		massFactors = info_->getMassFactors();
69	–	PairList excludes = info_->getExcludedInteractions();
70	–	PairList oneTwo = info_->getOneTwoInteractions();
71	–	PairList oneThree = info_->getOneThreeInteractions();
72	–	PairList oneFour = info_->getOneFourInteractions();
104
105	+	PairList* excludes = info_->getExcludedInteractions();
106	+	PairList* oneTwo = info_->getOneTwoInteractions();
107	+	PairList* oneThree = info_->getOneThreeInteractions();
108	+	PairList* oneFour = info_->getOneFourInteractions();
109	+
110	+	if (needVelocities_)
111	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
112	+	DataStorage::dslVelocity);
113	+	else
114	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
115	+
116		#ifdef IS_MPI
117
118	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
119	<	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_);
118	>	MPI_Comm row = rowComm.getComm();
119	>	MPI_Comm col = colComm.getComm();
120
121	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
122	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
123	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
124	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
125	<	AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
121	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
122	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
123	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
124	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
125	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
126
127	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
128	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
129	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
130	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
127	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
128	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
129	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
130	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
131	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
132
133	<	nAtomsInRow_ = AtomCommIntRow->getSize();
134	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
135	<	nGroupsInRow_ = cgCommIntRow->getSize();
136	<	nGroupsInCol_ = cgCommIntColumn->getSize();
133	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
134	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
135	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
136	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
137
138	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
139	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
140	+	nGroupsInRow_ = cgPlanIntRow->getSize();
141	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
142	+
143		// Modify the data storage objects with the correct layouts and sizes:
144		atomRowData.resize(nAtomsInRow_);
145		atomRowData.setStorageLayout(storageLayout_);
#	Line 103 | Line 148 | namespace OpenMD {
148		cgRowData.resize(nGroupsInRow_);
149		cgRowData.setStorageLayout(DataStorage::dslPosition);
150		cgColData.resize(nGroupsInCol_);
151	<	cgColData.setStorageLayout(DataStorage::dslPosition);
152	<
151	>	if (needVelocities_)
152	>	// we only need column velocities if we need them.
153	>	cgColData.setStorageLayout(DataStorage::dslPosition |
154	>	DataStorage::dslVelocity);
155	>	else
156	>	cgColData.setStorageLayout(DataStorage::dslPosition);
157	>
158		identsRow.resize(nAtomsInRow_);
159		identsCol.resize(nAtomsInCol_);
160
161	<	AtomCommIntRow->gather(identsLocal, identsRow);
162	<	AtomCommIntColumn->gather(identsLocal, identsCol);
163	<
164	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
165	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
161	>	AtomPlanIntRow->gather(idents, identsRow);
162	>	AtomPlanIntColumn->gather(idents, identsCol);
163	>
164	>	regionsRow.resize(nAtomsInRow_);
165	>	regionsCol.resize(nAtomsInCol_);
166
167	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
168	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
167	>	AtomPlanIntRow->gather(regions, regionsRow);
168	>	AtomPlanIntColumn->gather(regions, regionsCol);
169	>
170	>	// allocate memory for the parallel objects
171	>	atypesRow.resize(nAtomsInRow_);
172	>	atypesCol.resize(nAtomsInCol_);
173
174	<	AtomCommRealRow->gather(massFactors, massFactorsRow);
175	<	AtomCommRealColumn->gather(massFactors, massFactorsCol);
174	>	for (int i = 0; i < nAtomsInRow_; i++)
175	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
176	>	for (int i = 0; i < nAtomsInCol_; i++)
177	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
178
179	+	pot_row.resize(nAtomsInRow_);
180	+	pot_col.resize(nAtomsInCol_);
181	+
182	+	expot_row.resize(nAtomsInRow_);
183	+	expot_col.resize(nAtomsInCol_);
184	+
185	+	AtomRowToGlobal.resize(nAtomsInRow_);
186	+	AtomColToGlobal.resize(nAtomsInCol_);
187	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
188	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
189	+
190	+	cgRowToGlobal.resize(nGroupsInRow_);
191	+	cgColToGlobal.resize(nGroupsInCol_);
192	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
193	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
194	+
195	+	massFactorsRow.resize(nAtomsInRow_);
196	+	massFactorsCol.resize(nAtomsInCol_);
197	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
198	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
199	+
200		groupListRow_.clear();
201		groupListRow_.resize(nGroupsInRow_);
202		for (int i = 0; i < nGroupsInRow_; i++) {
#	Line 142 | Line 219 | namespace OpenMD {
219		}
220		}
221
222	<	skipsForAtom.clear();
223	<	skipsForAtom.resize(nAtomsInRow_);
222	>	excludesForAtom.clear();
223	>	excludesForAtom.resize(nAtomsInRow_);
224		toposForAtom.clear();
225		toposForAtom.resize(nAtomsInRow_);
226		topoDist.clear();
#	Line 154 | Line 231 | namespace OpenMD {
231		for (int j = 0; j < nAtomsInCol_; j++) {
232		int jglob = AtomColToGlobal[j];
233
234	<	if (excludes.hasPair(iglob, jglob))
235	<	skipsForAtom[i].push_back(j);
234	>	if (excludes->hasPair(iglob, jglob))
235	>	excludesForAtom[i].push_back(j);
236
237	<	if (oneTwo.hasPair(iglob, jglob)) {
237	>	if (oneTwo->hasPair(iglob, jglob)) {
238		toposForAtom[i].push_back(j);
239		topoDist[i].push_back(1);
240		} else {
241	<	if (oneThree.hasPair(iglob, jglob)) {
241	>	if (oneThree->hasPair(iglob, jglob)) {
242		toposForAtom[i].push_back(j);
243		topoDist[i].push_back(2);
244		} else {
245	<	if (oneFour.hasPair(iglob, jglob)) {
245	>	if (oneFour->hasPair(iglob, jglob)) {
246		toposForAtom[i].push_back(j);
247		topoDist[i].push_back(3);
248		}
#	Line 174 | Line 251 | namespace OpenMD {
251		}
252		}
253
254	<	#endif
255	<
256	<	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_);
254	>	#else
255	>	excludesForAtom.clear();
256	>	excludesForAtom.resize(nLocal_);
257		toposForAtom.clear();
258		toposForAtom.resize(nLocal_);
259		topoDist.clear();
#	Line 201 | Line 265 | namespace OpenMD {
265		for (int j = 0; j < nLocal_; j++) {
266		int jglob = AtomLocalToGlobal[j];
267
268	<	if (excludes.hasPair(iglob, jglob))
269	<	skipsForAtom[i].push_back(j);
268	>	if (excludes->hasPair(iglob, jglob))
269	>	excludesForAtom[i].push_back(j);
270
271	<	if (oneTwo.hasPair(iglob, jglob)) {
271	>	if (oneTwo->hasPair(iglob, jglob)) {
272		toposForAtom[i].push_back(j);
273		topoDist[i].push_back(1);
274		} else {
275	<	if (oneThree.hasPair(iglob, jglob)) {
275	>	if (oneThree->hasPair(iglob, jglob)) {
276		toposForAtom[i].push_back(j);
277		topoDist[i].push_back(2);
278		} else {
279	<	if (oneFour.hasPair(iglob, jglob)) {
279	>	if (oneFour->hasPair(iglob, jglob)) {
280		toposForAtom[i].push_back(j);
281		topoDist[i].push_back(3);
282		}
#	Line 220 | Line 284 | namespace OpenMD {
284		}
285		}
286		}
287	<
224	<	createGtypeCutoffMap();
225	<	}
226	<
227	<	void ForceMatrixDecomposition::createGtypeCutoffMap() {
287	>	#endif
288
289	<	RealType tol = 1e-6;
290	<	RealType rc;
231	<	int atid;
232	<	set<AtomType*> atypes = info_->getSimulatedAtomTypes();
233	<	vector<RealType> atypeCutoff;
234	<	atypeCutoff.resize( atypes.size() );
289	>	// allocate memory for the parallel objects
290	>	atypesLocal.resize(nLocal_);
291
292	<	for (set<AtomType*>::iterator at = atypes.begin();
293	<	at != atypes.end(); ++at){
238	<	rc = interactionMan_->getSuggestedCutoffRadius(*at);
239	<	atid = (*at)->getIdent();
240	<	atypeCutoff[atid] = rc;
241	<	}
292	>	for (int i = 0; i < nLocal_; i++)
293	>	atypesLocal[i] = ff_->getAtomType(idents[i]);
294
295	<	vector<RealType> gTypeCutoffs;
296	<
297	<	// first we do a single loop over the cutoff groups to find the
298	<	// largest cutoff for any atypes present in this group.
299	<	#ifdef IS_MPI
300	<	vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
301	<	groupRowToGtype.resize(nGroupsInRow_);
302	<	for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
251	<	vector<int> atomListRow = getAtomsInGroupRow(cg1);
252	<	for (vector<int>::iterator ia = atomListRow.begin();
253	<	ia != atomListRow.end(); ++ia) {
254	<	int atom1 = (*ia);
255	<	atid = identsRow[atom1];
256	<	if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
257	<	groupCutoffRow[cg1] = atypeCutoff[atid];
295	>	groupList_.clear();
296	>	groupList_.resize(nGroups_);
297	>	for (int i = 0; i < nGroups_; i++) {
298	>	int gid = cgLocalToGlobal[i];
299	>	for (int j = 0; j < nLocal_; j++) {
300	>	int aid = AtomLocalToGlobal[j];
301	>	if (globalGroupMembership[aid] == gid) {
302	>	groupList_[i].push_back(j);
303		}
259	–	}
260	–
261	–	bool gTypeFound = false;
262	–	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
263	–	if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
264	–	groupRowToGtype[cg1] = gt;
265	–	gTypeFound = true;
266	–	}
267	–	}
268	–	if (!gTypeFound) {
269	–	gTypeCutoffs.push_back( groupCutoffRow[cg1] );
270	–	groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
271	–	}
272	–
273	–	}
274	–	vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
275	–	groupColToGtype.resize(nGroupsInCol_);
276	–	for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
277	–	vector<int> atomListCol = getAtomsInGroupColumn(cg2);
278	–	for (vector<int>::iterator jb = atomListCol.begin();
279	–	jb != atomListCol.end(); ++jb) {
280	–	int atom2 = (*jb);
281	–	atid = identsCol[atom2];
282	–	if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
283	–	groupCutoffCol[cg2] = atypeCutoff[atid];
284	–	}
285	–	}
286	–	bool gTypeFound = false;
287	–	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
288	–	if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
289	–	groupColToGtype[cg2] = gt;
290	–	gTypeFound = true;
291	–	}
292	–	}
293	–	if (!gTypeFound) {
294	–	gTypeCutoffs.push_back( groupCutoffCol[cg2] );
295	–	groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
296	–	}
297	–	}
298	–	#else
299	–
300	–	vector<RealType> groupCutoff(nGroups_, 0.0);
301	–	groupToGtype.resize(nGroups_);
302	–
303	–	cerr << "nGroups = " << nGroups_ << "\n";
304	–	for (int cg1 = 0; cg1 < nGroups_; cg1++) {
305	–
306	–	groupCutoff[cg1] = 0.0;
307	–	vector<int> atomList = getAtomsInGroupRow(cg1);
308	–
309	–	for (vector<int>::iterator ia = atomList.begin();
310	–	ia != atomList.end(); ++ia) {
311	–	int atom1 = (*ia);
312	–	atid = identsLocal[atom1];
313	–	if (atypeCutoff[atid] > groupCutoff[cg1]) {
314	–	groupCutoff[cg1] = atypeCutoff[atid];
315	–	}
316	–	}
317	–
318	–	bool gTypeFound = false;
319	–	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
320	–	if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
321	–	groupToGtype[cg1] = gt;
322	–	gTypeFound = true;
323	–	}
324	–	}
325	–	if (!gTypeFound) {
326	–	gTypeCutoffs.push_back( groupCutoff[cg1] );
327	–	groupToGtype[cg1] = gTypeCutoffs.size() - 1;
304		}
305	<	}
330	<	#endif
331	<
332	<	cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n";
333	<	// Now we find the maximum group cutoff value present in the simulation
334	<
335	<	RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
336	<
337	<	#ifdef IS_MPI
338	<	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
339	<	#endif
340	<
341	<	RealType tradRcut = groupMax;
342	<
343	<	for (int i = 0; i < gTypeCutoffs.size();  i++) {
344	<	for (int j = 0; j < gTypeCutoffs.size();  j++) {
345	<	RealType thisRcut;
346	<	switch(cutoffPolicy_) {
347	<	case TRADITIONAL:
348	<	thisRcut = tradRcut;
349	<	break;
350	<	case MIX:
351	<	thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
352	<	break;
353	<	case MAX:
354	<	thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
355	<	break;
356	<	default:
357	<	sprintf(painCave.errMsg,
358	<	"ForceMatrixDecomposition::createGtypeCutoffMap "
359	<	"hit an unknown cutoff policy!\n");
360	<	painCave.severity = OPENMD_ERROR;
361	<	painCave.isFatal = 1;
362	<	simError();
363	<	break;
364	<	}
365	<
366	<	pair<int,int> key = make_pair(i,j);
367	<	gTypeCutoffMap[key].first = thisRcut;
368	<
369	<	if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
370	<
371	<	gTypeCutoffMap[key].second = thisRcut*thisRcut;
372	<
373	<	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
374	<
375	<	// sanity check
376	<
377	<	if (userChoseCutoff_) {
378	<	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
379	<	sprintf(painCave.errMsg,
380	<	"ForceMatrixDecomposition::createGtypeCutoffMap "
381	<	"user-specified rCut does not match computed group Cutoff\n");
382	<	painCave.severity = OPENMD_ERROR;
383	<	painCave.isFatal = 1;
384	<	simError();
385	<	}
386	<	}
387	<	}
388	<	}
305	>	}
306		}
307	<
391	<
392	<	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
393	<	int i, j;
394	<	#ifdef IS_MPI
395	<	i = groupRowToGtype[cg1];
396	<	j = groupColToGtype[cg2];
397	<	#else
398	<	i = groupToGtype[cg1];
399	<	j = groupToGtype[cg2];
400	<	#endif
401	<	return gTypeCutoffMap[make_pair(i,j)];
402	<	}
403	<
307	>
308		int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
309	<	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
309	>	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
310		if (toposForAtom[atom1][j] == atom2)
311		return topoDist[atom1][j];
312	<	}
312	>	}
313		return 0;
314		}
315
316		void ForceMatrixDecomposition::zeroWorkArrays() {
317	+	pairwisePot = 0.0;
318	+	embeddingPot = 0.0;
319	+	excludedPot = 0.0;
320	+	excludedSelfPot = 0.0;
321
414	–	for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
415	–	longRangePot_[j] = 0.0;
416	–	}
417	–
322		#ifdef IS_MPI
323		if (storageLayout_ & DataStorage::dslForce) {
324		fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
#	Line 430 | Line 334 | namespace OpenMD {
334		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
335
336		fill(pot_col.begin(), pot_col.end(),
337	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
338	+
339	+	fill(expot_row.begin(), expot_row.end(),
340		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
434	–
435	–	pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
341
342	+	fill(expot_col.begin(), expot_col.end(),
343	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
344	+
345		if (storageLayout_ & DataStorage::dslParticlePot) {
346	<	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
347	<	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
346	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
347	>	0.0);
348	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
349	>	0.0);
350		}
351
352		if (storageLayout_ & DataStorage::dslDensity) {
#	Line 445 | Line 355 | namespace OpenMD {
355		}
356
357		if (storageLayout_ & DataStorage::dslFunctional) {
358	<	fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
359	<	fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
358	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
359	>	0.0);
360	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
361	>	0.0);
362		}
363
364		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
#	Line 456 | Line 368 | namespace OpenMD {
368		atomColData.functionalDerivative.end(), 0.0);
369		}
370
371	<	#else
372	<
371	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
372	>	fill(atomRowData.skippedCharge.begin(),
373	>	atomRowData.skippedCharge.end(), 0.0);
374	>	fill(atomColData.skippedCharge.begin(),
375	>	atomColData.skippedCharge.end(), 0.0);
376	>	}
377	>
378	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
379	>	fill(atomRowData.flucQFrc.begin(),
380	>	atomRowData.flucQFrc.end(), 0.0);
381	>	fill(atomColData.flucQFrc.begin(),
382	>	atomColData.flucQFrc.end(), 0.0);
383	>	}
384	>
385	>	if (storageLayout_ & DataStorage::dslElectricField) {
386	>	fill(atomRowData.electricField.begin(),
387	>	atomRowData.electricField.end(), V3Zero);
388	>	fill(atomColData.electricField.begin(),
389	>	atomColData.electricField.end(), V3Zero);
390	>	}
391	>
392	>	if (storageLayout_ & DataStorage::dslSitePotential) {
393	>	fill(atomRowData.sitePotential.begin(),
394	>	atomRowData.sitePotential.end(), 0.0);
395	>	fill(atomColData.sitePotential.begin(),
396	>	atomColData.sitePotential.end(), 0.0);
397	>	}
398	>
399	>	#endif
400	>	// even in parallel, we need to zero out the local arrays:
401	>
402		if (storageLayout_ & DataStorage::dslParticlePot) {
403		fill(snap_->atomData.particlePot.begin(),
404		snap_->atomData.particlePot.end(), 0.0);
#	Line 467 | Line 408 | namespace OpenMD {
408		fill(snap_->atomData.density.begin(),
409		snap_->atomData.density.end(), 0.0);
410		}
411	+
412		if (storageLayout_ & DataStorage::dslFunctional) {
413		fill(snap_->atomData.functional.begin(),
414		snap_->atomData.functional.end(), 0.0);
415		}
416	+
417		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
418		fill(snap_->atomData.functionalDerivative.begin(),
419		snap_->atomData.functionalDerivative.end(), 0.0);
420		}
421	<	#endif
422	<
421	>
422	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
423	>	fill(snap_->atomData.skippedCharge.begin(),
424	>	snap_->atomData.skippedCharge.end(), 0.0);
425	>	}
426	>
427	>	if (storageLayout_ & DataStorage::dslElectricField) {
428	>	fill(snap_->atomData.electricField.begin(),
429	>	snap_->atomData.electricField.end(), V3Zero);
430	>	}
431	>	if (storageLayout_ & DataStorage::dslSitePotential) {
432	>	fill(snap_->atomData.sitePotential.begin(),
433	>	snap_->atomData.sitePotential.end(), 0.0);
434	>	}
435		}
436
437
438		void ForceMatrixDecomposition::distributeData()  {
439		snap_ = sman_->getCurrentSnapshot();
440		storageLayout_ = sman_->getStorageLayout();
441	+
442	+	bool needsCG = true;
443	+	if(info_->getNCutoffGroups() != info_->getNAtoms())
444	+	needsCG = false;
445	+
446		#ifdef IS_MPI
447
448		// gather up the atomic positions
449	<	AtomCommVectorRow->gather(snap_->atomData.position,
449	>	AtomPlanVectorRow->gather(snap_->atomData.position,
450		atomRowData.position);
451	<	AtomCommVectorColumn->gather(snap_->atomData.position,
451	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
452		atomColData.position);
453
454		// gather up the cutoff group positions
455	<	cgCommVectorRow->gather(snap_->cgData.position,
456	<	cgRowData.position);
457	<	cgCommVectorColumn->gather(snap_->cgData.position,
458	<	cgColData.position);
455	>
456	>	if (needsCG) {
457	>	cgPlanVectorRow->gather(snap_->cgData.position,
458	>	cgRowData.position);
459	>
460	>	cgPlanVectorColumn->gather(snap_->cgData.position,
461	>	cgColData.position);
462	>	}
463	>
464	>
465	>	if (needVelocities_) {
466	>	// gather up the atomic velocities
467	>	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
468	>	atomColData.velocity);
469	>
470	>	if (needsCG) {
471	>	cgPlanVectorColumn->gather(snap_->cgData.velocity,
472	>	cgColData.velocity);
473	>	}
474	>	}
475	>
476
477		// if needed, gather the atomic rotation matrices
478		if (storageLayout_ & DataStorage::dslAmat) {
479	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
479	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
480		atomRowData.aMat);
481	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
481	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
482		atomColData.aMat);
483		}
484	<
485	<	// if needed, gather the atomic eletrostatic frames
486	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
487	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
488	<	atomRowData.electroFrame);
489	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
490	<	atomColData.electroFrame);
484	>
485	>	// if needed, gather the atomic eletrostatic information
486	>	if (storageLayout_ & DataStorage::dslDipole) {
487	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
488	>	atomRowData.dipole);
489	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
490	>	atomColData.dipole);
491		}
492	+
493	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
494	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
495	+	atomRowData.quadrupole);
496	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
497	+	atomColData.quadrupole);
498	+	}
499	+
500	+	// if needed, gather the atomic fluctuating charge values
501	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
502	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
503	+	atomRowData.flucQPos);
504	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
505	+	atomColData.flucQPos);
506	+	}
507	+
508		#endif
509		}
510
#	Line 525 | Line 518 | namespace OpenMD {
518
519		if (storageLayout_ & DataStorage::dslDensity) {
520
521	<	AtomCommRealRow->scatter(atomRowData.density,
521	>	AtomPlanRealRow->scatter(atomRowData.density,
522		snap_->atomData.density);
523
524		int n = snap_->atomData.density.size();
525		vector<RealType> rho_tmp(n, 0.0);
526	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
526	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
527		for (int i = 0; i < n; i++)
528		snap_->atomData.density[i] += rho_tmp[i];
529		}
530	+
531	+	// this isn't necessary if we don't have polarizable atoms, but
532	+	// we'll leave it here for now.
533	+	if (storageLayout_ & DataStorage::dslElectricField) {
534	+
535	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
536	+	snap_->atomData.electricField);
537	+
538	+	int n = snap_->atomData.electricField.size();
539	+	vector<Vector3d> field_tmp(n, V3Zero);
540	+	AtomPlanVectorColumn->scatter(atomColData.electricField,
541	+	field_tmp);
542	+	for (int i = 0; i < n; i++)
543	+	snap_->atomData.electricField[i] += field_tmp[i];
544	+	}
545		#endif
546		}
547
#	Line 546 | Line 554 | namespace OpenMD {
554		storageLayout_ = sman_->getStorageLayout();
555		#ifdef IS_MPI
556		if (storageLayout_ & DataStorage::dslFunctional) {
557	<	AtomCommRealRow->gather(snap_->atomData.functional,
557	>	AtomPlanRealRow->gather(snap_->atomData.functional,
558		atomRowData.functional);
559	<	AtomCommRealColumn->gather(snap_->atomData.functional,
559	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
560		atomColData.functional);
561		}
562
563		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
564	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
564	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
565		atomRowData.functionalDerivative);
566	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
566	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
567		atomColData.functionalDerivative);
568		}
569		#endif
#	Line 569 | Line 577 | namespace OpenMD {
577		int n = snap_->atomData.force.size();
578		vector<Vector3d> frc_tmp(n, V3Zero);
579
580	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
580	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
581		for (int i = 0; i < n; i++) {
582		snap_->atomData.force[i] += frc_tmp[i];
583		frc_tmp[i] = 0.0;
584		}
585
586	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
587	<	for (int i = 0; i < n; i++)
586	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
587	>	for (int i = 0; i < n; i++) {
588		snap_->atomData.force[i] += frc_tmp[i];
589	<
590	<
589	>	}
590	>
591		if (storageLayout_ & DataStorage::dslTorque) {
592
593	<	int nt = snap_->atomData.force.size();
593	>	int nt = snap_->atomData.torque.size();
594		vector<Vector3d> trq_tmp(nt, V3Zero);
595
596	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
597	<	for (int i = 0; i < n; i++) {
596	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
597	>	for (int i = 0; i < nt; i++) {
598		snap_->atomData.torque[i] += trq_tmp[i];
599		trq_tmp[i] = 0.0;
600		}
601
602	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
603	<	for (int i = 0; i < n; i++)
602	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
603	>	for (int i = 0; i < nt; i++)
604		snap_->atomData.torque[i] += trq_tmp[i];
605		}
606	+
607	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
608	+
609	+	int ns = snap_->atomData.skippedCharge.size();
610	+	vector<RealType> skch_tmp(ns, 0.0);
611	+
612	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
613	+	for (int i = 0; i < ns; i++) {
614	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
615	+	skch_tmp[i] = 0.0;
616	+	}
617	+
618	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
619	+	for (int i = 0; i < ns; i++)
620	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
621	+
622	+	}
623
624	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
625	+
626	+	int nq = snap_->atomData.flucQFrc.size();
627	+	vector<RealType> fqfrc_tmp(nq, 0.0);
628	+
629	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
630	+	for (int i = 0; i < nq; i++) {
631	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
632	+	fqfrc_tmp[i] = 0.0;
633	+	}
634	+
635	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
636	+	for (int i = 0; i < nq; i++)
637	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
638	+
639	+	}
640	+
641	+	if (storageLayout_ & DataStorage::dslElectricField) {
642	+
643	+	int nef = snap_->atomData.electricField.size();
644	+	vector<Vector3d> efield_tmp(nef, V3Zero);
645	+
646	+	AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
647	+	for (int i = 0; i < nef; i++) {
648	+	snap_->atomData.electricField[i] += efield_tmp[i];
649	+	efield_tmp[i] = 0.0;
650	+	}
651	+
652	+	AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
653	+	for (int i = 0; i < nef; i++)
654	+	snap_->atomData.electricField[i] += efield_tmp[i];
655	+	}
656	+
657	+	if (storageLayout_ & DataStorage::dslSitePotential) {
658	+
659	+	int nsp = snap_->atomData.sitePotential.size();
660	+	vector<RealType> sp_tmp(nsp, 0.0);
661	+
662	+	AtomPlanRealRow->scatter(atomRowData.sitePotential, sp_tmp);
663	+	for (int i = 0; i < nsp; i++) {
664	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
665	+	sp_tmp[i] = 0.0;
666	+	}
667	+
668	+	AtomPlanRealColumn->scatter(atomColData.sitePotential, sp_tmp);
669	+	for (int i = 0; i < nsp; i++)
670	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
671	+	}
672	+
673		nLocal_ = snap_->getNumberOfAtoms();
674
675		vector<potVec> pot_temp(nLocal_,
676		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
677	+	vector<potVec> expot_temp(nLocal_,
678	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
679
680		// scatter/gather pot_row into the members of my column
681
682	<	AtomCommPotRow->scatter(pot_row, pot_temp);
682	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
683	>	AtomPlanPotRow->scatter(expot_row, expot_temp);
684
685	<	for (int ii = 0;  ii < pot_temp.size(); ii++ )
686	<	pot_local += pot_temp[ii];
687	<
685	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
686	>	pairwisePot += pot_temp[ii];
687	>
688	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
689	>	excludedPot += expot_temp[ii];
690	>
691	>	if (storageLayout_ & DataStorage::dslParticlePot) {
692	>	// This is the pairwise contribution to the particle pot.  The
693	>	// embedding contribution is added in each of the low level
694	>	// non-bonded routines.  In single processor, this is done in
695	>	// unpackInteractionData, not in collectData.
696	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
697	>	for (int i = 0; i < nLocal_; i++) {
698	>	// factor of two is because the total potential terms are divided
699	>	// by 2 in parallel due to row/ column scatter
700	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
701	>	}
702	>	}
703	>	}
704	>
705		fill(pot_temp.begin(), pot_temp.end(),
706		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
707	+	fill(expot_temp.begin(), expot_temp.end(),
708	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
709
710	<	AtomCommPotColumn->scatter(pot_col, pot_temp);
710	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
711	>	AtomPlanPotColumn->scatter(expot_col, expot_temp);
712
713		for (int ii = 0;  ii < pot_temp.size(); ii++ )
714	<	pot_local += pot_temp[ii];
714	>	pairwisePot += pot_temp[ii];
715	>
716	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
717	>	excludedPot += expot_temp[ii];
718	>
719	>	if (storageLayout_ & DataStorage::dslParticlePot) {
720	>	// This is the pairwise contribution to the particle pot.  The
721	>	// embedding contribution is added in each of the low level
722	>	// non-bonded routines.  In single processor, this is done in
723	>	// unpackInteractionData, not in collectData.
724	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
725	>	for (int i = 0; i < nLocal_; i++) {
726	>	// factor of two is because the total potential terms are divided
727	>	// by 2 in parallel due to row/ column scatter
728	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
729	>	}
730	>	}
731	>	}
732
733	+	if (storageLayout_ & DataStorage::dslParticlePot) {
734	+	int npp = snap_->atomData.particlePot.size();
735	+	vector<RealType> ppot_temp(npp, 0.0);
736	+
737	+	// This is the direct or embedding contribution to the particle
738	+	// pot.
739	+
740	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
741	+	for (int i = 0; i < npp; i++) {
742	+	snap_->atomData.particlePot[i] += ppot_temp[i];
743	+	}
744	+
745	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
746	+
747	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
748	+	for (int i = 0; i < npp; i++) {
749	+	snap_->atomData.particlePot[i] += ppot_temp[i];
750	+	}
751	+	}
752	+
753	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
754	+	RealType ploc1 = pairwisePot[ii];
755	+	RealType ploc2 = 0.0;
756	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
757	+	pairwisePot[ii] = ploc2;
758	+	}
759	+
760	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
761	+	RealType ploc1 = excludedPot[ii];
762	+	RealType ploc2 = 0.0;
763	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
764	+	excludedPot[ii] = ploc2;
765	+	}
766	+
767	+	// Here be dragons.
768	+	MPI_Comm col = colComm.getComm();
769	+
770	+	MPI_Allreduce(MPI_IN_PLACE,
771	+	&snap_->frameData.conductiveHeatFlux[0], 3,
772	+	MPI_REALTYPE, MPI_SUM, col);
773	+
774	+
775		#endif
776	+
777		}
778
779	<	int ForceMatrixDecomposition::getNAtomsInRow() {
779	>	/**
780	>	* Collects information obtained during the post-pair (and embedding
781	>	* functional) loops onto local data structures.
782	>	*/
783	>	void ForceMatrixDecomposition::collectSelfData() {
784	>	snap_ = sman_->getCurrentSnapshot();
785	>	storageLayout_ = sman_->getStorageLayout();
786	>
787		#ifdef IS_MPI
788	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
789	+	RealType ploc1 = embeddingPot[ii];
790	+	RealType ploc2 = 0.0;
791	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
792	+	embeddingPot[ii] = ploc2;
793	+	}
794	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
795	+	RealType ploc1 = excludedSelfPot[ii];
796	+	RealType ploc2 = 0.0;
797	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
798	+	excludedSelfPot[ii] = ploc2;
799	+	}
800	+	#endif
801	+
802	+	}
803	+
804	+
805	+
806	+	int& ForceMatrixDecomposition::getNAtomsInRow() {
807	+	#ifdef IS_MPI
808		return nAtomsInRow_;
809		#else
810		return nLocal_;
#	Line 630 | Line 814 | namespace OpenMD {
814		/**
815		* returns the list of atoms belonging to this group.
816		*/
817	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
817	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
818		#ifdef IS_MPI
819		return groupListRow_[cg1];
820		#else
#	Line 638 | Line 822 | namespace OpenMD {
822		#endif
823		}
824
825	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
825	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
826		#ifdef IS_MPI
827		return groupListCol_[cg2];
828		#else
#	Line 655 | Line 839 | namespace OpenMD {
839		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
840		#endif
841
842	<	snap_->wrapVector(d);
842	>	if (usePeriodicBoundaryConditions_) {
843	>	snap_->wrapVector(d);
844	>	}
845		return d;
846		}
847
848	+	Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
849	+	#ifdef IS_MPI
850	+	return cgColData.velocity[cg2];
851	+	#else
852	+	return snap_->cgData.velocity[cg2];
853	+	#endif
854	+	}
855
856	+	Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
857	+	#ifdef IS_MPI
858	+	return atomColData.velocity[atom2];
859	+	#else
860	+	return snap_->atomData.velocity[atom2];
861	+	#endif
862	+	}
863	+
864	+
865		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
866
867		Vector3d d;
#	Line 669 | Line 871 | namespace OpenMD {
871		#else
872		d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
873		#endif
874	<
875	<	snap_->wrapVector(d);
874	>	if (usePeriodicBoundaryConditions_) {
875	>	snap_->wrapVector(d);
876	>	}
877		return d;
878		}
879
#	Line 682 | Line 885 | namespace OpenMD {
885		#else
886		d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
887		#endif
888	<
889	<	snap_->wrapVector(d);
888	>	if (usePeriodicBoundaryConditions_) {
889	>	snap_->wrapVector(d);
890	>	}
891		return d;
892		}
893
894	<	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
894	>	RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
895		#ifdef IS_MPI
896		return massFactorsRow[atom1];
897		#else
#	Line 695 | Line 899 | namespace OpenMD {
899		#endif
900		}
901
902	<	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
902	>	RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
903		#ifdef IS_MPI
904		return massFactorsCol[atom2];
905		#else
#	Line 712 | Line 916 | namespace OpenMD {
916		#else
917		d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
918		#endif
919	<
920	<	snap_->wrapVector(d);
919	>	if (usePeriodicBoundaryConditions_) {
920	>	snap_->wrapVector(d);
921	>	}
922		return d;
923		}
924
925	<	vector<int> ForceMatrixDecomposition::getSkipsForAtom(int atom1) {
926	<	return skipsForAtom[atom1];
925	>	vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
926	>	return excludesForAtom[atom1];
927		}
928
929		/**
930	<	* There are a number of reasons to skip a pair or a
726	<	* particle. Mostly we do this to exclude atoms who are involved in
727	<	* short range interactions (bonds, bends, torsions), but we also
728	<	* need to exclude some overcounted interactions that result from
930	>	* We need to exclude some overcounted interactions that result from
931		* the parallel decomposition.
932		*/
933	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
933	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
934		int unique_id_1, unique_id_2;
935	<
935	>
936		#ifdef IS_MPI
937		// in MPI, we have to look up the unique IDs for each atom
938		unique_id_1 = AtomRowToGlobal[atom1];
939		unique_id_2 = AtomColToGlobal[atom2];
940	<
941	<	// this situation should only arise in MPI simulations
940	>	// group1 = cgRowToGlobal[cg1];
941	>	// group2 = cgColToGlobal[cg2];
942	>	#else
943	>	unique_id_1 = AtomLocalToGlobal[atom1];
944	>	unique_id_2 = AtomLocalToGlobal[atom2];
945	>	int group1 = cgLocalToGlobal[cg1];
946	>	int group2 = cgLocalToGlobal[cg2];
947	>	#endif
948	>
949		if (unique_id_1 == unique_id_2) return true;
950	<
950	>
951	>	#ifdef IS_MPI
952		// this prevents us from doing the pair on multiple processors
953		if (unique_id_1 < unique_id_2) {
954		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
955		} else {
956	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
956	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
957		}
958	<	#else
959	<	// in the normal loop, the atom numbers are unique
960	<	unique_id_1 = atom1;
961	<	unique_id_2 = atom2;
958	>	#endif
959	>
960	>	#ifndef IS_MPI
961	>	if (group1 == group2) {
962	>	if (unique_id_1 < unique_id_2) return true;
963	>	}
964		#endif
965
966	<	for (vector<int>::iterator i = skipsForAtom[atom1].begin();
967	<	i != skipsForAtom[atom1].end(); ++i) {
756	<	if ( (*i) == unique_id_2 ) return true;
757	<	}
966	>	return false;
967	>	}
968
969	+	/**
970	+	* We need to handle the interactions for atoms who are involved in
971	+	* the same rigid body as well as some short range interactions
972	+	* (bonds, bends, torsions) differently from other interactions.
973	+	* We'll still visit the pairwise routines, but with a flag that
974	+	* tells those routines to exclude the pair from direct long range
975	+	* interactions.  Some indirect interactions (notably reaction
976	+	* field) must still be handled for these pairs.
977	+	*/
978	+	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
979	+
980	+	// excludesForAtom was constructed to use row/column indices in the MPI
981	+	// version, and to use local IDs in the non-MPI version:
982	+
983	+	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
984	+	i != excludesForAtom[atom1].end(); ++i) {
985	+	if ( (*i) == atom2 ) return true;
986	+	}
987	+
988	+	return false;
989		}
990
991
#	Line 776 | Line 1006 | namespace OpenMD {
1006		}
1007
1008		// filling interaction blocks with pointers
1009	<	void ForceMatrixDecomposition::fillInteractionData(InteractionData idat,
1010	<	int atom1, int atom2) {
1009	>	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1010	>	int atom1, int atom2,
1011	>	bool newAtom1) {
1012	>
1013	>	idat.excluded = excludeAtomPair(atom1, atom2);
1014	>
1015	>	if (newAtom1) {
1016	>
1017		#ifdef IS_MPI
1018	<
1019	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1020	<	ff_->getAtomType(identsCol[atom2]) );
1021	<
1018	>	idat.atid1 = identsRow[atom1];
1019	>	idat.atid2 = identsCol[atom2];
1020	>
1021	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1022	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1023	>	} else {
1024	>	idat.sameRegion = false;
1025	>	}
1026	>
1027	>	if (storageLayout_ & DataStorage::dslAmat) {
1028	>	idat.A1 = &(atomRowData.aMat[atom1]);
1029	>	idat.A2 = &(atomColData.aMat[atom2]);
1030	>	}
1031	>
1032	>	if (storageLayout_ & DataStorage::dslTorque) {
1033	>	idat.t1 = &(atomRowData.torque[atom1]);
1034	>	idat.t2 = &(atomColData.torque[atom2]);
1035	>	}
1036	>
1037	>	if (storageLayout_ & DataStorage::dslDipole) {
1038	>	idat.dipole1 = &(atomRowData.dipole[atom1]);
1039	>	idat.dipole2 = &(atomColData.dipole[atom2]);
1040	>	}
1041	>
1042	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1043	>	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1044	>	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1045	>	}
1046	>
1047	>	if (storageLayout_ & DataStorage::dslDensity) {
1048	>	idat.rho1 = &(atomRowData.density[atom1]);
1049	>	idat.rho2 = &(atomColData.density[atom2]);
1050	>	}
1051	>
1052	>	if (storageLayout_ & DataStorage::dslFunctional) {
1053	>	idat.frho1 = &(atomRowData.functional[atom1]);
1054	>	idat.frho2 = &(atomColData.functional[atom2]);
1055	>	}
1056	>
1057	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1058	>	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1059	>	idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1060	>	}
1061	>
1062	>	if (storageLayout_ & DataStorage::dslParticlePot) {
1063	>	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1064	>	idat.particlePot2 = &(atomColData.particlePot[atom2]);
1065	>	}
1066	>
1067	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1068	>	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1069	>	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1070	>	}
1071	>
1072	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1073	>	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1074	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1075	>	}
1076	>
1077	>	#else
1078	>
1079	>	idat.atid1 = idents[atom1];
1080	>	idat.atid2 = idents[atom2];
1081	>
1082	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1083	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1084	>	} else {
1085	>	idat.sameRegion = false;
1086	>	}
1087	>
1088	>	if (storageLayout_ & DataStorage::dslAmat) {
1089	>	idat.A1 = &(snap_->atomData.aMat[atom1]);
1090	>	idat.A2 = &(snap_->atomData.aMat[atom2]);
1091	>	}
1092	>
1093	>	if (storageLayout_ & DataStorage::dslTorque) {
1094	>	idat.t1 = &(snap_->atomData.torque[atom1]);
1095	>	idat.t2 = &(snap_->atomData.torque[atom2]);
1096	>	}
1097	>
1098	>	if (storageLayout_ & DataStorage::dslDipole) {
1099	>	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1100	>	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1101	>	}
1102	>
1103	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1104	>	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1105	>	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1106	>	}
1107	>
1108	>	if (storageLayout_ & DataStorage::dslDensity) {
1109	>	idat.rho1 = &(snap_->atomData.density[atom1]);
1110	>	idat.rho2 = &(snap_->atomData.density[atom2]);
1111	>	}
1112	>
1113	>	if (storageLayout_ & DataStorage::dslFunctional) {
1114	>	idat.frho1 = &(snap_->atomData.functional[atom1]);
1115	>	idat.frho2 = &(snap_->atomData.functional[atom2]);
1116	>	}
1117	>
1118	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1119	>	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1120	>	idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1121	>	}
1122	>
1123	>	if (storageLayout_ & DataStorage::dslParticlePot) {
1124	>	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1125	>	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1126	>	}
1127	>
1128	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1129	>	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1130	>	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1131	>	}
1132	>
1133	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1134	>	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1135	>	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1136	>	}
1137	>	#endif
1138	>
1139	>	} else {
1140	>	// atom1 is not new, so don't bother updating properties of that atom:
1141	>	#ifdef IS_MPI
1142	>	idat.atid2 = identsCol[atom2];
1143	>
1144	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1145	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1146	>	} else {
1147	>	idat.sameRegion = false;
1148	>	}
1149	>
1150		if (storageLayout_ & DataStorage::dslAmat) {
787	–	idat.A1 = &(atomRowData.aMat[atom1]);
1151		idat.A2 = &(atomColData.aMat[atom2]);
1152		}
1153
791	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
792	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
793	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
794	–	}
795	–
1154		if (storageLayout_ & DataStorage::dslTorque) {
797	–	idat.t1 = &(atomRowData.torque[atom1]);
1155		idat.t2 = &(atomColData.torque[atom2]);
1156		}
1157
1158	+	if (storageLayout_ & DataStorage::dslDipole) {
1159	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1160	+	}
1161	+
1162	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1163	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1164	+	}
1165	+
1166		if (storageLayout_ & DataStorage::dslDensity) {
802	–	idat.rho1 = &(atomRowData.density[atom1]);
1167		idat.rho2 = &(atomColData.density[atom2]);
1168		}
1169
1170		if (storageLayout_ & DataStorage::dslFunctional) {
807	–	idat.frho1 = &(atomRowData.functional[atom1]);
1171		idat.frho2 = &(atomColData.functional[atom2]);
1172		}
1173
1174		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
812	–	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1175		idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1176		}
1177
1178		if (storageLayout_ & DataStorage::dslParticlePot) {
817	–	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1179		idat.particlePot2 = &(atomColData.particlePot[atom2]);
1180		}
1181
1182	<	#else
1183	<
1184	<	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
824	<	ff_->getAtomType(identsLocal[atom2]) );
1182	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1183	>	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1184	>	}
1185
1186	<	if (storageLayout_ & DataStorage::dslAmat) {
1187	<	idat.A1 = &(snap_->atomData.aMat[atom1]);
828	<	idat.A2 = &(snap_->atomData.aMat[atom2]);
1186	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1187	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1188		}
1189
1190	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1191	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1192	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1190	>	#else
1191	>	idat.atid2 = idents[atom2];
1192	>
1193	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1194	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1195	>	} else {
1196	>	idat.sameRegion = false;
1197		}
1198
1199	+	if (storageLayout_ & DataStorage::dslAmat) {
1200	+	idat.A2 = &(snap_->atomData.aMat[atom2]);
1201	+	}
1202	+
1203		if (storageLayout_ & DataStorage::dslTorque) {
837	–	idat.t1 = &(snap_->atomData.torque[atom1]);
1204		idat.t2 = &(snap_->atomData.torque[atom2]);
1205		}
1206
1207	<	if (storageLayout_ & DataStorage::dslDensity) {
1208	<	idat.rho1 = &(snap_->atomData.density[atom1]);
1207	>	if (storageLayout_ & DataStorage::dslDipole) {
1208	>	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1209	>	}
1210	>
1211	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1212	>	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1213	>	}
1214	>
1215	>	if (storageLayout_ & DataStorage::dslDensity) {
1216		idat.rho2 = &(snap_->atomData.density[atom2]);
1217		}
1218
1219		if (storageLayout_ & DataStorage::dslFunctional) {
847	–	idat.frho1 = &(snap_->atomData.functional[atom1]);
1220		idat.frho2 = &(snap_->atomData.functional[atom2]);
1221		}
1222
1223		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
852	–	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1224		idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1225		}
1226
1227		if (storageLayout_ & DataStorage::dslParticlePot) {
857	–	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1228		idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1229		}
1230
1231	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1232	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1233	+	}
1234	+
1235	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1236	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1237	+	}
1238	+
1239		#endif
1240	+	}
1241		}
863	–
1242
1243	<	void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {
1243	>	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat,
1244	>	int atom1, int atom2) {
1245		#ifdef IS_MPI
1246	<	pot_row[atom1] += 0.5 *  *(idat.pot);
1247	<	pot_col[atom2] += 0.5 *  *(idat.pot);
1246	>	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1247	>	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1248	>	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1249	>	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1250
1251		atomRowData.force[atom1] += *(idat.f1);
1252		atomColData.force[atom2] -= *(idat.f1);
872	–	#else
873	–	longRangePot_ += *(idat.pot);
874	–
875	–	snap_->atomData.force[atom1] += *(idat.f1);
876	–	snap_->atomData.force[atom2] -= *(idat.f1);
877	–	#endif
1253
1254	<	}
1254	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1255	>	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1256	>	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1257	>	}
1258
1259	+	if (storageLayout_ & DataStorage::dslElectricField) {
1260	+	atomRowData.electricField[atom1] += *(idat.eField1);
1261	+	atomColData.electricField[atom2] += *(idat.eField2);
1262	+	}
1263
1264	<	void ForceMatrixDecomposition::fillSkipData(InteractionData idat,
1265	<	int atom1, int atom2) {
1266	<	#ifdef IS_MPI
1267	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
886	<	ff_->getAtomType(identsCol[atom2]) );
1264	>	if (storageLayout_ & DataStorage::dslSitePotential) {
1265	>	atomRowData.sitePotential[atom1] += *(idat.sPot1);
1266	>	atomColData.sitePotential[atom2] += *(idat.sPot2);
1267	>	}
1268
1269	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1270	<	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1271	<	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1269	>	#else
1270	>	pairwisePot += *(idat.pot);
1271	>	excludedPot += *(idat.excludedPot);
1272	>
1273	>	snap_->atomData.force[atom1] += *(idat.f1);
1274	>	snap_->atomData.force[atom2] -= *(idat.f1);
1275	>
1276	>	if (idat.doParticlePot) {
1277	>	// This is the pairwise contribution to the particle pot.  The
1278	>	// embedding contribution is added in each of the low level
1279	>	// non-bonded routines.  In parallel, this calculation is done
1280	>	// in collectData, not in unpackInteractionData.
1281	>	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1282	>	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1283		}
1284	<	if (storageLayout_ & DataStorage::dslTorque) {
1285	<	idat.t1 = &(atomRowData.torque[atom1]);
1286	<	idat.t2 = &(atomColData.torque[atom2]);
1284	>
1285	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1286	>	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1287	>	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1288		}
896	–	#else
897	–	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
898	–	ff_->getAtomType(identsLocal[atom2]) );
1289
1290	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1291	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1292	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1290	>	if (storageLayout_ & DataStorage::dslElectricField) {
1291	>	snap_->atomData.electricField[atom1] += *(idat.eField1);
1292	>	snap_->atomData.electricField[atom2] += *(idat.eField2);
1293		}
1294	<	if (storageLayout_ & DataStorage::dslTorque) {
1295	<	idat.t1 = &(snap_->atomData.torque[atom1]);
1296	<	idat.t2 = &(snap_->atomData.torque[atom2]);
1294	>
1295	>	if (storageLayout_ & DataStorage::dslSitePotential) {
1296	>	snap_->atomData.sitePotential[atom1] += *(idat.sPot1);
1297	>	snap_->atomData.sitePotential[atom2] += *(idat.sPot2);
1298		}
1299	<	#endif
1299	>
1300	>	#endif
1301	>
1302		}
1303
1304		/*
1305		* buildNeighborList
1306		*
1307	<	* first element of pair is row-indexed CutoffGroup
1308	<	* second element of pair is column-indexed CutoffGroup
1307	>	* Constructs the Verlet neighbor list for a force-matrix
1308	>	* decomposition.  In this case, each processor is responsible for
1309	>	* row-site interactions with column-sites.
1310	>	*
1311	>	* neighborList is returned as a packed array of neighboring
1312	>	* column-ordered CutoffGroups.  The starting position in
1313	>	* neighborList for each row-ordered CutoffGroup is given by the
1314	>	* returned vector point.
1315		*/
1316	<	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1317	<
1318	<	vector<pair<int, int> > neighborList;
1319	<	groupCutoffs cuts;
1316	>	void ForceMatrixDecomposition::buildNeighborList(vector<int>& neighborList,
1317	>	vector<int>& point) {
1318	>	neighborList.clear();
1319	>	point.clear();
1320	>	int len = 0;
1321	>
1322	>	bool doAllPairs = false;
1323	>
1324	>	Snapshot* snap_ = sman_->getCurrentSnapshot();
1325	>	Mat3x3d box;
1326	>	Mat3x3d invBox;
1327	>
1328	>	Vector3d rs, scaled, dr;
1329	>	Vector3i whichCell;
1330	>	int cellIndex;
1331	>
1332		#ifdef IS_MPI
1333		cellListRow_.clear();
1334		cellListCol_.clear();
1335	+	point.resize(nGroupsInRow_+1);
1336		#else
1337		cellList_.clear();
1338	+	point.resize(nGroups_+1);
1339		#endif
1340	+
1341	+	if (!usePeriodicBoundaryConditions_) {
1342	+	box = snap_->getBoundingBox();
1343	+	invBox = snap_->getInvBoundingBox();
1344	+	} else {
1345	+	box = snap_->getHmat();
1346	+	invBox = snap_->getInvHmat();
1347	+	}
1348	+
1349	+	Vector3d A = box.getColumn(0);
1350	+	Vector3d B = box.getColumn(1);
1351	+	Vector3d C = box.getColumn(2);
1352
1353	<	RealType rList_ = (largestRcut_ + skinThickness_);
1354	<	RealType rl2 = rList_ * rList_;
1355	<	Snapshot* snap_ = sman_->getCurrentSnapshot();
1356	<	Mat3x3d Hmat = snap_->getHmat();
932	<	Vector3d Hx = Hmat.getColumn(0);
933	<	Vector3d Hy = Hmat.getColumn(1);
934	<	Vector3d Hz = Hmat.getColumn(2);
1353	>	// Required for triclinic cells
1354	>	Vector3d AxB = cross(A, B);
1355	>	Vector3d BxC = cross(B, C);
1356	>	Vector3d CxA = cross(C, A);
1357
1358	<	nCells_.x() = (int) ( Hx.length() )/ rList_;
1359	<	nCells_.y() = (int) ( Hy.length() )/ rList_;
1360	<	nCells_.z() = (int) ( Hz.length() )/ rList_;
1361	<
940	<	Mat3x3d invHmat = snap_->getInvHmat();
941	<	Vector3d rs, scaled, dr;
942	<	Vector3i whichCell;
943	<	int cellIndex;
944	<	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1358	>	// unit vectors perpendicular to the faces of the triclinic cell:
1359	>	AxB.normalize();
1360	>	BxC.normalize();
1361	>	CxA.normalize();
1362
1363	<	cerr << "flag1\n";
1363	>	// A set of perpendicular lengths in triclinic cells:
1364	>	RealType Wa = abs(dot(A, BxC));
1365	>	RealType Wb = abs(dot(B, CxA));
1366	>	RealType Wc = abs(dot(C, AxB));
1367	>
1368	>	nCells_.x() = int( Wa / rList_ );
1369	>	nCells_.y() = int( Wb / rList_ );
1370	>	nCells_.z() = int( Wc / rList_ );
1371	>
1372	>	// handle small boxes where the cell offsets can end up repeating cells
1373	>	if (nCells_.x() < 3) doAllPairs = true;
1374	>	if (nCells_.y() < 3) doAllPairs = true;
1375	>	if (nCells_.z() < 3) doAllPairs = true;
1376	>
1377	>	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1378	>
1379		#ifdef IS_MPI
1380		cellListRow_.resize(nCtot);
1381		cellListCol_.resize(nCtot);
1382		#else
1383		cellList_.resize(nCtot);
1384		#endif
1385	<	cerr << "flag2\n";
1385	>
1386	>	if (!doAllPairs) {
1387	>
1388		#ifdef IS_MPI
1389	<	for (int i = 0; i < nGroupsInRow_; i++) {
1390	<	rs = cgRowData.position[i];
1391	<
1392	<	// scaled positions relative to the box vectors
1393	<	scaled = invHmat * rs;
1394	<
1395	<	// wrap the vector back into the unit box by subtracting integer box
1396	<	// numbers
1397	<	for (int j = 0; j < 3; j++) {
1398	<	scaled[j] -= roundMe(scaled[j]);
1399	<	scaled[j] += 0.5;
1389	>
1390	>	for (int i = 0; i < nGroupsInRow_; i++) {
1391	>	rs = cgRowData.position[i];
1392	>
1393	>	// scaled positions relative to the box vectors
1394	>	scaled = invBox * rs;
1395	>
1396	>	// wrap the vector back into the unit box by subtracting integer box
1397	>	// numbers
1398	>	for (int j = 0; j < 3; j++) {
1399	>	scaled[j] -= roundMe(scaled[j]);
1400	>	scaled[j] += 0.5;
1401	>	// Handle the special case when an object is exactly on the
1402	>	// boundary (a scaled coordinate of 1.0 is the same as
1403	>	// scaled coordinate of 0.0)
1404	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1405	>	}
1406	>
1407	>	// find xyz-indices of cell that cutoffGroup is in.
1408	>	whichCell.x() = nCells_.x() * scaled.x();
1409	>	whichCell.y() = nCells_.y() * scaled.y();
1410	>	whichCell.z() = nCells_.z() * scaled.z();
1411	>
1412	>	// find single index of this cell:
1413	>	cellIndex = Vlinear(whichCell, nCells_);
1414	>
1415	>	// add this cutoff group to the list of groups in this cell;
1416	>	cellListRow_[cellIndex].push_back(i);
1417		}
1418	<
1419	<	// find xyz-indices of cell that cutoffGroup is in.
1420	<	whichCell.x() = nCells_.x() * scaled.x();
1421	<	whichCell.y() = nCells_.y() * scaled.y();
1422	<	whichCell.z() = nCells_.z() * scaled.z();
1423	<
1424	<	// find single index of this cell:
1425	<	cellIndex = Vlinear(whichCell, nCells_);
1426	<
1427	<	// add this cutoff group to the list of groups in this cell;
1428	<	cellListRow_[cellIndex].push_back(i);
1429	<	}
1430	<
1431	<	for (int i = 0; i < nGroupsInCol_; i++) {
1432	<	rs = cgColData.position[i];
1433	<
1434	<	// scaled positions relative to the box vectors
1435	<	scaled = invHmat * rs;
1436	<
1437	<	// wrap the vector back into the unit box by subtracting integer box
1438	<	// numbers
1439	<	for (int j = 0; j < 3; j++) {
1440	<	scaled[j] -= roundMe(scaled[j]);
1441	<	scaled[j] += 0.5;
1418	>	for (int i = 0; i < nGroupsInCol_; i++) {
1419	>	rs = cgColData.position[i];
1420	>
1421	>	// scaled positions relative to the box vectors
1422	>	scaled = invBox * rs;
1423	>
1424	>	// wrap the vector back into the unit box by subtracting integer box
1425	>	// numbers
1426	>	for (int j = 0; j < 3; j++) {
1427	>	scaled[j] -= roundMe(scaled[j]);
1428	>	scaled[j] += 0.5;
1429	>	// Handle the special case when an object is exactly on the
1430	>	// boundary (a scaled coordinate of 1.0 is the same as
1431	>	// scaled coordinate of 0.0)
1432	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1433	>	}
1434	>
1435	>	// find xyz-indices of cell that cutoffGroup is in.
1436	>	whichCell.x() = nCells_.x() * scaled.x();
1437	>	whichCell.y() = nCells_.y() * scaled.y();
1438	>	whichCell.z() = nCells_.z() * scaled.z();
1439	>
1440	>	// find single index of this cell:
1441	>	cellIndex = Vlinear(whichCell, nCells_);
1442	>
1443	>	// add this cutoff group to the list of groups in this cell;
1444	>	cellListCol_[cellIndex].push_back(i);
1445		}
1446	<
993	<	// find xyz-indices of cell that cutoffGroup is in.
994	<	whichCell.x() = nCells_.x() * scaled.x();
995	<	whichCell.y() = nCells_.y() * scaled.y();
996	<	whichCell.z() = nCells_.z() * scaled.z();
997	<
998	<	// find single index of this cell:
999	<	cellIndex = Vlinear(whichCell, nCells_);
1000	<
1001	<	// add this cutoff group to the list of groups in this cell;
1002	<	cellListCol_[cellIndex].push_back(i);
1003	<	}
1446	>
1447		#else
1448	<	for (int i = 0; i < nGroups_; i++) {
1449	<	rs = snap_->cgData.position[i];
1450	<
1451	<	// scaled positions relative to the box vectors
1452	<	scaled = invHmat * rs;
1453	<
1454	<	// wrap the vector back into the unit box by subtracting integer box
1455	<	// numbers
1456	<	for (int j = 0; j < 3; j++) {
1457	<	scaled[j] -= roundMe(scaled[j]);
1458	<	scaled[j] += 0.5;
1448	>	for (int i = 0; i < nGroups_; i++) {
1449	>	rs = snap_->cgData.position[i];
1450	>
1451	>	// scaled positions relative to the box vectors
1452	>	scaled = invBox * rs;
1453	>
1454	>	// wrap the vector back into the unit box by subtracting integer box
1455	>	// numbers
1456	>	for (int j = 0; j < 3; j++) {
1457	>	scaled[j] -= roundMe(scaled[j]);
1458	>	scaled[j] += 0.5;
1459	>	// Handle the special case when an object is exactly on the
1460	>	// boundary (a scaled coordinate of 1.0 is the same as
1461	>	// scaled coordinate of 0.0)
1462	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1463	>	}
1464	>
1465	>	// find xyz-indices of cell that cutoffGroup is in.
1466	>	whichCell.x() = int(nCells_.x() * scaled.x());
1467	>	whichCell.y() = int(nCells_.y() * scaled.y());
1468	>	whichCell.z() = int(nCells_.z() * scaled.z());
1469	>
1470	>	// find single index of this cell:
1471	>	cellIndex = Vlinear(whichCell, nCells_);
1472	>
1473	>	// add this cutoff group to the list of groups in this cell;
1474	>	cellList_[cellIndex].push_back(i);
1475		}
1476
1477	<	// find xyz-indices of cell that cutoffGroup is in.
1019	<	whichCell.x() = nCells_.x() * scaled.x();
1020	<	whichCell.y() = nCells_.y() * scaled.y();
1021	<	whichCell.z() = nCells_.z() * scaled.z();
1477	>	#endif
1478
1479	<	// find single index of this cell:
1480	<	cellIndex = Vlinear(whichCell, nCells_);
1479	>	#ifdef IS_MPI
1480	>	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1481	>	rs = cgRowData.position[j1];
1482	>	#else
1483
1484	<	// add this cutoff group to the list of groups in this cell;
1485	<	cellList_[cellIndex].push_back(i);
1028	<	}
1484	>	for (int j1 = 0; j1 < nGroups_; j1++) {
1485	>	rs = snap_->cgData.position[j1];
1486		#endif
1487	+	point[j1] = len;
1488	+
1489	+	// scaled positions relative to the box vectors
1490	+	scaled = invBox * rs;
1491	+
1492	+	// wrap the vector back into the unit box by subtracting integer box
1493	+	// numbers
1494	+	for (int j = 0; j < 3; j++) {
1495	+	scaled[j] -= roundMe(scaled[j]);
1496	+	scaled[j] += 0.5;
1497	+	// Handle the special case when an object is exactly on the
1498	+	// boundary (a scaled coordinate of 1.0 is the same as
1499	+	// scaled coordinate of 0.0)
1500	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1501	+	}
1502	+
1503	+	// find xyz-indices of cell that cutoffGroup is in.
1504	+	whichCell.x() = nCells_.x() * scaled.x();
1505	+	whichCell.y() = nCells_.y() * scaled.y();
1506	+	whichCell.z() = nCells_.z() * scaled.z();
1507	+
1508	+	// find single index of this cell:
1509	+	int m1 = Vlinear(whichCell, nCells_);
1510
1511	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1512	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1513	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1514	<	Vector3i m1v(m1x, m1y, m1z);
1035	<	int m1 = Vlinear(m1v, nCells_);
1511	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1512	>	os != cellOffsets_.end(); ++os) {
1513	>
1514	>	Vector3i m2v = whichCell + (*os);
1515
1516	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1517	<	os != cellOffsets_.end(); ++os) {
1516	>	if (m2v.x() >= nCells_.x()) {
1517	>	m2v.x() = 0;
1518	>	} else if (m2v.x() < 0) {
1519	>	m2v.x() = nCells_.x() - 1;
1520	>	}
1521	>
1522	>	if (m2v.y() >= nCells_.y()) {
1523	>	m2v.y() = 0;
1524	>	} else if (m2v.y() < 0) {
1525	>	m2v.y() = nCells_.y() - 1;
1526	>	}
1527	>
1528	>	if (m2v.z() >= nCells_.z()) {
1529	>	m2v.z() = 0;
1530	>	} else if (m2v.z() < 0) {
1531	>	m2v.z() = nCells_.z() - 1;
1532	>	}
1533	>	int m2 = Vlinear (m2v, nCells_);
1534	>	#ifdef IS_MPI
1535	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1536	>	j2 != cellListCol_[m2].end(); ++j2) {
1537
1538	<	Vector3i m2v = m1v + (*os);
1539	<
1540	<	if (m2v.x() >= nCells_.x()) {
1541	<	m2v.x() = 0;
1542	<	} else if (m2v.x() < 0) {
1543	<	m2v.x() = nCells_.x() - 1;
1538	>	// In parallel, we need to visit *all* pairs of row
1539	>	// & column indicies and will divide labor in the
1540	>	// force evaluation later.
1541	>	dr = cgColData.position[(*j2)] - rs;
1542	>	if (usePeriodicBoundaryConditions_) {
1543	>	snap_->wrapVector(dr);
1544		}
1545	+	if (dr.lengthSquare() < rListSq_) {
1546	+	neighborList.push_back( (*j2) );
1547	+	++len;
1548	+	}
1549	+	}
1550	+	#else
1551	+	for (vector<int>::iterator j2 = cellList_[m2].begin();
1552	+	j2 != cellList_[m2].end(); ++j2) {
1553	+
1554	+	// Always do this if we're in different cells or if
1555	+	// we're in the same cell and the global index of
1556	+	// the j2 cutoff group is greater than or equal to
1557	+	// the j1 cutoff group.  Note that Rappaport's code
1558	+	// has a "less than" conditional here, but that
1559	+	// deals with atom-by-atom computation.  OpenMD
1560	+	// allows atoms within a single cutoff group to
1561	+	// interact with each other.
1562
1563	<	if (m2v.y() >= nCells_.y()) {
1564	<	m2v.y() = 0;
1565	<	} else if (m2v.y() < 0) {
1566	<	m2v.y() = nCells_.y() - 1;
1567	<	}
1053	<
1054	<	if (m2v.z() >= nCells_.z()) {
1055	<	m2v.z() = 0;
1056	<	} else if (m2v.z() < 0) {
1057	<	m2v.z() = nCells_.z() - 1;
1058	<	}
1059	<
1060	<	int m2 = Vlinear (m2v, nCells_);
1061	<
1062	<	#ifdef IS_MPI
1063	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1064	<	j1 != cellListRow_[m1].end(); ++j1) {
1065	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1066	<	j2 != cellListCol_[m2].end(); ++j2) {
1067	<
1068	<	// Always do this if we're in different cells or if
1069	<	// we're in the same cell and the global index of the
1070	<	// j2 cutoff group is less than the j1 cutoff group
1071	<
1072	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1073	<	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1074	<	snap_->wrapVector(dr);
1075	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1076	<	if (dr.lengthSquare() < cuts.third) {
1077	<	neighborList.push_back(make_pair((*j1), (*j2)));
1078	<	}
1079	<	}
1563	>	if ( (*j2) >= j1 ) {
1564	>
1565	>	dr = snap_->cgData.position[(*j2)] - rs;
1566	>	if (usePeriodicBoundaryConditions_) {
1567	>	snap_->wrapVector(dr);
1568		}
1569	<	}
1570	<	#else
1571	<
1084	<	for (vector<int>::iterator j1 = cellList_[m1].begin();
1085	<	j1 != cellList_[m1].end(); ++j1) {
1086	<	for (vector<int>::iterator j2 = cellList_[m2].begin();
1087	<	j2 != cellList_[m2].end(); ++j2) {
1088	<
1089	<	// Always do this if we're in different cells or if
1090	<	// we're in the same cell and the global index of the
1091	<	// j2 cutoff group is less than the j1 cutoff group
1092	<
1093	<	if (m2 != m1 || (*j2) < (*j1)) {
1094	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1095	<	snap_->wrapVector(dr);
1096	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1097	<	if (dr.lengthSquare() < cuts.third) {
1098	<	neighborList.push_back(make_pair((*j1), (*j2)));
1099	<	}
1100	<	}
1569	>	if ( dr.lengthSquare() < rListSq_) {
1570	>	neighborList.push_back( (*j2) );
1571	>	++len;
1572		}
1573		}
1574	+	}
1575		#endif
1576	+	}
1577	+	}
1578	+	} else {
1579	+	// branch to do all cutoff group pairs
1580	+	#ifdef IS_MPI
1581	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1582	+	point[j1] = len;
1583	+	rs = cgRowData.position[j1];
1584	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1585	+	dr = cgColData.position[j2] - rs;
1586	+	if (usePeriodicBoundaryConditions_) {
1587	+	snap_->wrapVector(dr);
1588		}
1589	+	if (dr.lengthSquare() < rListSq_) {
1590	+	neighborList.push_back( j2 );
1591	+	++len;
1592	+	}
1593		}
1594	+	}
1595	+	#else
1596	+	// include all groups here.
1597	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1598	+	point[j1] = len;
1599	+	rs = snap_->cgData.position[j1];
1600	+	// include self group interactions j2 == j1
1601	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1602	+	dr = snap_->cgData.position[j2] - rs;
1603	+	if (usePeriodicBoundaryConditions_) {
1604	+	snap_->wrapVector(dr);
1605	+	}
1606	+	if (dr.lengthSquare() < rListSq_) {
1607	+	neighborList.push_back( j2 );
1608	+	++len;
1609	+	}
1610	+	}
1611		}
1612	+	#endif
1613		}
1614	<
1614	>
1615	>	#ifdef IS_MPI
1616	>	point[nGroupsInRow_] = len;
1617	>	#else
1618	>	point[nGroups_] = len;
1619	>	#endif
1620	>
1621		// save the local cutoff group positions for the check that is
1622		// done on each loop:
1623		saved_CG_positions_.clear();
1624	+	saved_CG_positions_.reserve(nGroups_);
1625		for (int i = 0; i < nGroups_; i++)
1626		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1114	–
1115	–	return neighborList;
1627		}
1628		} //end namespace OpenMD

Diff Legend

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