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 1571 by gezelter, Fri May 27 16:45:44 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1939 by gezelter, Thu Oct 31 18:18:57 2013 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	+	// 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	<	nGroups_ = snap_->getNumberOfCutoffGroups();
99	<
98	>
99	>	nGroups_ = info_->getNLocalCutoffGroups();
100		// gather the information for atomtype IDs (atids):
101	<	identsLocal = info_->getIdentArray();
101	>	idents = info_->getIdentArray();
102	>	regions = info_->getRegions();
103		AtomLocalToGlobal = info_->getGlobalAtomIndices();
104		cgLocalToGlobal = info_->getGlobalGroupIndices();
105		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
67	–	vector<RealType> massFactorsLocal = info_->getMassFactors();
68	–	PairList excludes = info_->getExcludedInteractions();
69	–	PairList oneTwo = info_->getOneTwoInteractions();
70	–	PairList oneThree = info_->getOneThreeInteractions();
71	–	PairList oneFour = info_->getOneFourInteractions();
72	–	vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
106
107	+	massFactors = info_->getMassFactors();
108	+
109	+	PairList* excludes = info_->getExcludedInteractions();
110	+	PairList* oneTwo = info_->getOneTwoInteractions();
111	+	PairList* oneThree = info_->getOneThreeInteractions();
112	+	PairList* oneFour = info_->getOneFourInteractions();
113	+
114	+	if (needVelocities_)
115	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
116	+	DataStorage::dslVelocity);
117	+	else
118	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
119	+
120		#ifdef IS_MPI
121
122	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
123	<	AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
78	<	AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
79	<	AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
122	>	MPI::Intracomm row = rowComm.getComm();
123	>	MPI::Intracomm col = colComm.getComm();
124
125	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
126	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
127	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
128	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
125	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
126	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
127	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
128	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
129	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
130
131	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
132	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
133	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
134	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
131	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
132	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
133	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
134	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
135	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
136
137	<	nAtomsInRow_ = AtomCommIntRow->getSize();
138	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
139	<	nGroupsInRow_ = cgCommIntRow->getSize();
140	<	nGroupsInCol_ = cgCommIntColumn->getSize();
137	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
138	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
139	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
140	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
141
142	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
143	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
144	+	nGroupsInRow_ = cgPlanIntRow->getSize();
145	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
146	+
147		// Modify the data storage objects with the correct layouts and sizes:
148		atomRowData.resize(nAtomsInRow_);
149		atomRowData.setStorageLayout(storageLayout_);
#	Line 101 | Line 152 | namespace OpenMD {
152		cgRowData.resize(nGroupsInRow_);
153		cgRowData.setStorageLayout(DataStorage::dslPosition);
154		cgColData.resize(nGroupsInCol_);
155	<	cgColData.setStorageLayout(DataStorage::dslPosition);
155	>	if (needVelocities_)
156	>	// we only need column velocities if we need them.
157	>	cgColData.setStorageLayout(DataStorage::dslPosition |
158	>	DataStorage::dslVelocity);
159	>	else
160	>	cgColData.setStorageLayout(DataStorage::dslPosition);
161	>
162	>	identsRow.resize(nAtomsInRow_);
163	>	identsCol.resize(nAtomsInCol_);
164
165	<	vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
166	<	vector<RealType> (nAtomsInRow_, 0.0));
167	<	vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
168	<	vector<RealType> (nAtomsInCol_, 0.0));
165	>	AtomPlanIntRow->gather(idents, identsRow);
166	>	AtomPlanIntColumn->gather(idents, identsCol);
167	>
168	>	regionsRow.resize(nAtomsInRow_);
169	>	regionsCol.resize(nAtomsInCol_);
170
171	<	identsRow.reserve(nAtomsInRow_);
172	<	identsCol.reserve(nAtomsInCol_);
171	>	AtomPlanIntRow->gather(regions, regionsRow);
172	>	AtomPlanIntColumn->gather(regions, regionsCol);
173
174	<	AtomCommIntRow->gather(identsLocal, identsRow);
175	<	AtomCommIntColumn->gather(identsLocal, identsCol);
176	<
117	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
118	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
119	<
120	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
121	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
174	>	// allocate memory for the parallel objects
175	>	atypesRow.resize(nAtomsInRow_);
176	>	atypesCol.resize(nAtomsInCol_);
177
178	<	AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
179	<	AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
178	>	for (int i = 0; i < nAtomsInRow_; i++)
179	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
180	>	for (int i = 0; i < nAtomsInCol_; i++)
181	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
182
183	+	pot_row.resize(nAtomsInRow_);
184	+	pot_col.resize(nAtomsInCol_);
185	+
186	+	expot_row.resize(nAtomsInRow_);
187	+	expot_col.resize(nAtomsInCol_);
188	+
189	+	AtomRowToGlobal.resize(nAtomsInRow_);
190	+	AtomColToGlobal.resize(nAtomsInCol_);
191	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
192	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
193	+
194	+	cgRowToGlobal.resize(nGroupsInRow_);
195	+	cgColToGlobal.resize(nGroupsInCol_);
196	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
197	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
198	+
199	+	massFactorsRow.resize(nAtomsInRow_);
200	+	massFactorsCol.resize(nAtomsInCol_);
201	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
202	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
203	+
204		groupListRow_.clear();
205	<	groupListRow_.reserve(nGroupsInRow_);
205	>	groupListRow_.resize(nGroupsInRow_);
206		for (int i = 0; i < nGroupsInRow_; i++) {
207		int gid = cgRowToGlobal[i];
208		for (int j = 0; j < nAtomsInRow_; j++) {
#	Line 135 | Line 213 | namespace OpenMD {
213		}
214
215		groupListCol_.clear();
216	<	groupListCol_.reserve(nGroupsInCol_);
216	>	groupListCol_.resize(nGroupsInCol_);
217		for (int i = 0; i < nGroupsInCol_; i++) {
218		int gid = cgColToGlobal[i];
219		for (int j = 0; j < nAtomsInCol_; j++) {
#	Line 145 | Line 223 | namespace OpenMD {
223		}
224		}
225
226	<	skipsForRowAtom.clear();
227	<	skipsForRowAtom.reserve(nAtomsInRow_);
226	>	excludesForAtom.clear();
227	>	excludesForAtom.resize(nAtomsInRow_);
228	>	toposForAtom.clear();
229	>	toposForAtom.resize(nAtomsInRow_);
230	>	topoDist.clear();
231	>	topoDist.resize(nAtomsInRow_);
232		for (int i = 0; i < nAtomsInRow_; i++) {
233		int iglob = AtomRowToGlobal[i];
234	+
235		for (int j = 0; j < nAtomsInCol_; j++) {
236	<	int jglob = AtomColToGlobal[j];
237	<	if (excludes.hasPair(iglob, jglob))
238	<	skipsForRowAtom[i].push_back(j);
236	>	int jglob = AtomColToGlobal[j];
237	>
238	>	if (excludes->hasPair(iglob, jglob))
239	>	excludesForAtom[i].push_back(j);
240	>
241	>	if (oneTwo->hasPair(iglob, jglob)) {
242	>	toposForAtom[i].push_back(j);
243	>	topoDist[i].push_back(1);
244	>	} else {
245	>	if (oneThree->hasPair(iglob, jglob)) {
246	>	toposForAtom[i].push_back(j);
247	>	topoDist[i].push_back(2);
248	>	} else {
249	>	if (oneFour->hasPair(iglob, jglob)) {
250	>	toposForAtom[i].push_back(j);
251	>	topoDist[i].push_back(3);
252	>	}
253	>	}
254	>	}
255		}
256		}
257
258	<	toposForRowAtom.clear();
259	<	toposForRowAtom.reserve(nAtomsInRow_);
260	<	for (int i = 0; i < nAtomsInRow_; i++) {
261	<	int iglob = AtomRowToGlobal[i];
262	<	int nTopos = 0;
263	<	for (int j = 0; j < nAtomsInCol_; j++) {
264	<	int jglob = AtomColToGlobal[j];
265	<	if (oneTwo.hasPair(iglob, jglob)) {
266	<	toposForRowAtom[i].push_back(j);
267	<	topoDistRow[i][nTopos] = 1;
268	<	nTopos++;
258	>	#else
259	>	excludesForAtom.clear();
260	>	excludesForAtom.resize(nLocal_);
261	>	toposForAtom.clear();
262	>	toposForAtom.resize(nLocal_);
263	>	topoDist.clear();
264	>	topoDist.resize(nLocal_);
265	>
266	>	for (int i = 0; i < nLocal_; i++) {
267	>	int iglob = AtomLocalToGlobal[i];
268	>
269	>	for (int j = 0; j < nLocal_; j++) {
270	>	int jglob = AtomLocalToGlobal[j];
271	>
272	>	if (excludes->hasPair(iglob, jglob))
273	>	excludesForAtom[i].push_back(j);
274	>
275	>	if (oneTwo->hasPair(iglob, jglob)) {
276	>	toposForAtom[i].push_back(j);
277	>	topoDist[i].push_back(1);
278	>	} else {
279	>	if (oneThree->hasPair(iglob, jglob)) {
280	>	toposForAtom[i].push_back(j);
281	>	topoDist[i].push_back(2);
282	>	} else {
283	>	if (oneFour->hasPair(iglob, jglob)) {
284	>	toposForAtom[i].push_back(j);
285	>	topoDist[i].push_back(3);
286	>	}
287	>	}
288		}
171	–	if (oneThree.hasPair(iglob, jglob)) {
172	–	toposForRowAtom[i].push_back(j);
173	–	topoDistRow[i][nTopos] = 2;
174	–	nTopos++;
175	–	}
176	–	if (oneFour.hasPair(iglob, jglob)) {
177	–	toposForRowAtom[i].push_back(j);
178	–	topoDistRow[i][nTopos] = 3;
179	–	nTopos++;
180	–	}
289		}
290		}
183	–
291		#endif
292
293	+	// allocate memory for the parallel objects
294	+	atypesLocal.resize(nLocal_);
295	+
296	+	for (int i = 0; i < nLocal_; i++)
297	+	atypesLocal[i] = ff_->getAtomType(idents[i]);
298	+
299		groupList_.clear();
300	<	groupList_.reserve(nGroups_);
300	>	groupList_.resize(nGroups_);
301		for (int i = 0; i < nGroups_; i++) {
302		int gid = cgLocalToGlobal[i];
303		for (int j = 0; j < nLocal_; j++) {
304		int aid = AtomLocalToGlobal[j];
305	<	if (globalGroupMembership[aid] == gid)
305	>	if (globalGroupMembership[aid] == gid) {
306		groupList_[i].push_back(j);
307	+	}
308		}
309		}
310
197	–	skipsForLocalAtom.clear();
198	–	skipsForLocalAtom.reserve(nLocal_);
311
312	<	for (int i = 0; i < nLocal_; i++) {
313	<	int iglob = AtomLocalToGlobal[i];
314	<	for (int j = 0; j < nLocal_; j++) {
315	<	int jglob = AtomLocalToGlobal[j];
316	<	if (excludes.hasPair(iglob, jglob))
317	<	skipsForLocalAtom[i].push_back(j);
318	<	}
312	>	createGtypeCutoffMap();
313	>
314	>	}
315	>
316	>	void ForceMatrixDecomposition::createGtypeCutoffMap() {
317	>
318	>	GrCut.clear();
319	>	GrCutSq.clear();
320	>	GrlistSq.clear();
321	>
322	>	RealType tol = 1e-6;
323	>	largestRcut_ = 0.0;
324	>	int atid;
325	>	set<AtomType*> atypes = info_->getSimulatedAtomTypes();
326	>
327	>	map<int, RealType> atypeCutoff;
328	>
329	>	for (set<AtomType*>::iterator at = atypes.begin();
330	>	at != atypes.end(); ++at){
331	>	atid = (*at)->getIdent();
332	>	if (userChoseCutoff_)
333	>	atypeCutoff[atid] = userCutoff_;
334	>	else
335	>	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
336		}
337	+
338	+	vector<RealType> gTypeCutoffs;
339	+	// first we do a single loop over the cutoff groups to find the
340	+	// largest cutoff for any atypes present in this group.
341	+	#ifdef IS_MPI
342	+	vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
343	+	groupRowToGtype.resize(nGroupsInRow_);
344	+	for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
345	+	vector<int> atomListRow = getAtomsInGroupRow(cg1);
346	+	for (vector<int>::iterator ia = atomListRow.begin();
347	+	ia != atomListRow.end(); ++ia) {
348	+	int atom1 = (*ia);
349	+	atid = identsRow[atom1];
350	+	if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
351	+	groupCutoffRow[cg1] = atypeCutoff[atid];
352	+	}
353	+	}
354
355	<	toposForLocalAtom.clear();
356	<	toposForLocalAtom.reserve(nLocal_);
357	<	for (int i = 0; i < nLocal_; i++) {
358	<	int iglob = AtomLocalToGlobal[i];
359	<	int nTopos = 0;
360	<	for (int j = 0; j < nLocal_; j++) {
361	<	int jglob = AtomLocalToGlobal[j];
362	<	if (oneTwo.hasPair(iglob, jglob)) {
363	<	toposForLocalAtom[i].push_back(j);
364	<	topoDistLocal[i][nTopos] = 1;
365	<	nTopos++;
355	>	bool gTypeFound = false;
356	>	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
357	>	if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
358	>	groupRowToGtype[cg1] = gt;
359	>	gTypeFound = true;
360	>	}
361	>	}
362	>	if (!gTypeFound) {
363	>	gTypeCutoffs.push_back( groupCutoffRow[cg1] );
364	>	groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
365	>	}
366	>
367	>	}
368	>	vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
369	>	groupColToGtype.resize(nGroupsInCol_);
370	>	for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
371	>	vector<int> atomListCol = getAtomsInGroupColumn(cg2);
372	>	for (vector<int>::iterator jb = atomListCol.begin();
373	>	jb != atomListCol.end(); ++jb) {
374	>	int atom2 = (*jb);
375	>	atid = identsCol[atom2];
376	>	if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
377	>	groupCutoffCol[cg2] = atypeCutoff[atid];
378		}
379	<	if (oneThree.hasPair(iglob, jglob)) {
380	<	toposForLocalAtom[i].push_back(j);
381	<	topoDistLocal[i][nTopos] = 2;
382	<	nTopos++;
383	<	}
384	<	if (oneFour.hasPair(iglob, jglob)) {
385	<	toposForLocalAtom[i].push_back(j);
386	<	topoDistLocal[i][nTopos] = 3;
387	<	nTopos++;
388	<	}
379	>	}
380	>	bool gTypeFound = false;
381	>	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
382	>	if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
383	>	groupColToGtype[cg2] = gt;
384	>	gTypeFound = true;
385	>	}
386	>	}
387	>	if (!gTypeFound) {
388	>	gTypeCutoffs.push_back( groupCutoffCol[cg2] );
389	>	groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
390	>	}
391	>	}
392	>	#else
393	>
394	>	vector<RealType> groupCutoff(nGroups_, 0.0);
395	>	groupToGtype.resize(nGroups_);
396	>	for (int cg1 = 0; cg1 < nGroups_; cg1++) {
397	>	groupCutoff[cg1] = 0.0;
398	>	vector<int> atomList = getAtomsInGroupRow(cg1);
399	>	for (vector<int>::iterator ia = atomList.begin();
400	>	ia != atomList.end(); ++ia) {
401	>	int atom1 = (*ia);
402	>	atid = idents[atom1];
403	>	if (atypeCutoff[atid] > groupCutoff[cg1])
404	>	groupCutoff[cg1] = atypeCutoff[atid];
405	>	}
406	>
407	>	bool gTypeFound = false;
408	>	for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
409	>	if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
410	>	groupToGtype[cg1] = gt;
411	>	gTypeFound = true;
412	>	}
413	>	}
414	>	if (!gTypeFound) {
415	>	gTypeCutoffs.push_back( groupCutoff[cg1] );
416	>	groupToGtype[cg1] = gTypeCutoffs.size() - 1;
417		}
418	+	}
419	+	#endif
420	+
421	+	// Now we find the maximum group cutoff value present in the simulation
422	+
423	+	RealType groupMax = *max_element(gTypeCutoffs.begin(),
424	+	gTypeCutoffs.end());
425	+
426	+	#ifdef IS_MPI
427	+	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
428	+	MPI::MAX);
429	+	#endif
430	+
431	+	RealType tradRcut = groupMax;
432	+
433	+	GrCut.resize( gTypeCutoffs.size() );
434	+	GrCutSq.resize( gTypeCutoffs.size() );
435	+	GrlistSq.resize( gTypeCutoffs.size() );
436	+
437	+
438	+	for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) {
439	+	GrCut[i].resize( gTypeCutoffs.size() , 0.0);
440	+	GrCutSq[i].resize( gTypeCutoffs.size(), 0.0 );
441	+	GrlistSq[i].resize( gTypeCutoffs.size(), 0.0 );
442	+
443	+	for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) {
444	+	RealType thisRcut;
445	+	switch(cutoffPolicy_) {
446	+	case TRADITIONAL:
447	+	thisRcut = tradRcut;
448	+	break;
449	+	case MIX:
450	+	thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
451	+	break;
452	+	case MAX:
453	+	thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
454	+	break;
455	+	default:
456	+	sprintf(painCave.errMsg,
457	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
458	+	"hit an unknown cutoff policy!\n");
459	+	painCave.severity = OPENMD_ERROR;
460	+	painCave.isFatal = 1;
461	+	simError();
462	+	break;
463	+	}
464	+
465	+	GrCut[i][j] = thisRcut;
466	+	if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
467	+	GrCutSq[i][j] = thisRcut * thisRcut;
468	+	GrlistSq[i][j] = pow(thisRcut + skinThickness_, 2);
469	+
470	+	// pair<int,int> key = make_pair(i,j);
471	+	// gTypeCutoffMap[key].first = thisRcut;
472	+	// gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
473	+	// sanity check
474	+
475	+	if (userChoseCutoff_) {
476	+	if (abs(GrCut[i][j] - userCutoff_) > 0.0001) {
477	+	sprintf(painCave.errMsg,
478	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
479	+	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
480	+	painCave.severity = OPENMD_ERROR;
481	+	painCave.isFatal = 1;
482	+	simError();
483	+	}
484	+	}
485	+	}
486		}
487		}
488	<
488	>
489	>	void ForceMatrixDecomposition::getGroupCutoffs(int &cg1, int &cg2, RealType &rcut, RealType &rcutsq, RealType &rlistsq) {
490	>	int i, j;
491	>	#ifdef IS_MPI
492	>	i = groupRowToGtype[cg1];
493	>	j = groupColToGtype[cg2];
494	>	#else
495	>	i = groupToGtype[cg1];
496	>	j = groupToGtype[cg2];
497	>	#endif
498	>	rcut = GrCut[i][j];
499	>	rcutsq = GrCutSq[i][j];
500	>	rlistsq = GrlistSq[i][j];
501	>	return;
502	>	//return gTypeCutoffMap[make_pair(i,j)];
503	>	}
504	>
505	>	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
506	>	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
507	>	if (toposForAtom[atom1][j] == atom2)
508	>	return topoDist[atom1][j];
509	>	}
510	>	return 0;
511	>	}
512	>
513	>	void ForceMatrixDecomposition::zeroWorkArrays() {
514	>	pairwisePot = 0.0;
515	>	embeddingPot = 0.0;
516	>	excludedPot = 0.0;
517	>	excludedSelfPot = 0.0;
518	>
519	>	#ifdef IS_MPI
520	>	if (storageLayout_ & DataStorage::dslForce) {
521	>	fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
522	>	fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
523	>	}
524	>
525	>	if (storageLayout_ & DataStorage::dslTorque) {
526	>	fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
527	>	fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
528	>	}
529	>
530	>	fill(pot_row.begin(), pot_row.end(),
531	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
532	>
533	>	fill(pot_col.begin(), pot_col.end(),
534	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
535	>
536	>	fill(expot_row.begin(), expot_row.end(),
537	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
538	>
539	>	fill(expot_col.begin(), expot_col.end(),
540	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
541	>
542	>	if (storageLayout_ & DataStorage::dslParticlePot) {
543	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
544	>	0.0);
545	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
546	>	0.0);
547	>	}
548	>
549	>	if (storageLayout_ & DataStorage::dslDensity) {
550	>	fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
551	>	fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
552	>	}
553	>
554	>	if (storageLayout_ & DataStorage::dslFunctional) {
555	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
556	>	0.0);
557	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
558	>	0.0);
559	>	}
560	>
561	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
562	>	fill(atomRowData.functionalDerivative.begin(),
563	>	atomRowData.functionalDerivative.end(), 0.0);
564	>	fill(atomColData.functionalDerivative.begin(),
565	>	atomColData.functionalDerivative.end(), 0.0);
566	>	}
567	>
568	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
569	>	fill(atomRowData.skippedCharge.begin(),
570	>	atomRowData.skippedCharge.end(), 0.0);
571	>	fill(atomColData.skippedCharge.begin(),
572	>	atomColData.skippedCharge.end(), 0.0);
573	>	}
574	>
575	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
576	>	fill(atomRowData.flucQFrc.begin(),
577	>	atomRowData.flucQFrc.end(), 0.0);
578	>	fill(atomColData.flucQFrc.begin(),
579	>	atomColData.flucQFrc.end(), 0.0);
580	>	}
581	>
582	>	if (storageLayout_ & DataStorage::dslElectricField) {
583	>	fill(atomRowData.electricField.begin(),
584	>	atomRowData.electricField.end(), V3Zero);
585	>	fill(atomColData.electricField.begin(),
586	>	atomColData.electricField.end(), V3Zero);
587	>	}
588	>
589	>	#endif
590	>	// even in parallel, we need to zero out the local arrays:
591	>
592	>	if (storageLayout_ & DataStorage::dslParticlePot) {
593	>	fill(snap_->atomData.particlePot.begin(),
594	>	snap_->atomData.particlePot.end(), 0.0);
595	>	}
596	>
597	>	if (storageLayout_ & DataStorage::dslDensity) {
598	>	fill(snap_->atomData.density.begin(),
599	>	snap_->atomData.density.end(), 0.0);
600	>	}
601	>
602	>	if (storageLayout_ & DataStorage::dslFunctional) {
603	>	fill(snap_->atomData.functional.begin(),
604	>	snap_->atomData.functional.end(), 0.0);
605	>	}
606	>
607	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
608	>	fill(snap_->atomData.functionalDerivative.begin(),
609	>	snap_->atomData.functionalDerivative.end(), 0.0);
610	>	}
611	>
612	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
613	>	fill(snap_->atomData.skippedCharge.begin(),
614	>	snap_->atomData.skippedCharge.end(), 0.0);
615	>	}
616	>
617	>	if (storageLayout_ & DataStorage::dslElectricField) {
618	>	fill(snap_->atomData.electricField.begin(),
619	>	snap_->atomData.electricField.end(), V3Zero);
620	>	}
621	>	}
622	>
623	>
624		void ForceMatrixDecomposition::distributeData()  {
625		snap_ = sman_->getCurrentSnapshot();
626		storageLayout_ = sman_->getStorageLayout();
627		#ifdef IS_MPI
628
629		// gather up the atomic positions
630	<	AtomCommVectorRow->gather(snap_->atomData.position,
630	>	AtomPlanVectorRow->gather(snap_->atomData.position,
631		atomRowData.position);
632	<	AtomCommVectorColumn->gather(snap_->atomData.position,
632	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
633		atomColData.position);
634
635		// gather up the cutoff group positions
636	<	cgCommVectorRow->gather(snap_->cgData.position,
636	>
637	>	cgPlanVectorRow->gather(snap_->cgData.position,
638		cgRowData.position);
639	<	cgCommVectorColumn->gather(snap_->cgData.position,
639	>
640	>	cgPlanVectorColumn->gather(snap_->cgData.position,
641		cgColData.position);
642	+
643	+
644	+
645	+	if (needVelocities_) {
646	+	// gather up the atomic velocities
647	+	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
648	+	atomColData.velocity);
649	+
650	+	cgPlanVectorColumn->gather(snap_->cgData.velocity,
651	+	cgColData.velocity);
652	+	}
653	+
654
655		// if needed, gather the atomic rotation matrices
656		if (storageLayout_ & DataStorage::dslAmat) {
657	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
657	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
658		atomRowData.aMat);
659	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
659	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
660		atomColData.aMat);
661		}
662	<
663	<	// if needed, gather the atomic eletrostatic frames
664	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
665	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
666	<	atomRowData.electroFrame);
667	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
668	<	atomColData.electroFrame);
662	>
663	>	// if needed, gather the atomic eletrostatic information
664	>	if (storageLayout_ & DataStorage::dslDipole) {
665	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
666	>	atomRowData.dipole);
667	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
668	>	atomColData.dipole);
669		}
670	+
671	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
672	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
673	+	atomRowData.quadrupole);
674	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
675	+	atomColData.quadrupole);
676	+	}
677	+
678	+	// if needed, gather the atomic fluctuating charge values
679	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
680	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
681	+	atomRowData.flucQPos);
682	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
683	+	atomColData.flucQPos);
684	+	}
685	+
686		#endif
687		}
688
689	+	/* collects information obtained during the pre-pair loop onto local
690	+	* data structures.
691	+	*/
692		void ForceMatrixDecomposition::collectIntermediateData() {
693		snap_ = sman_->getCurrentSnapshot();
694		storageLayout_ = sman_->getStorageLayout();
#	Line 274 | Line 696 | namespace OpenMD {
696
697		if (storageLayout_ & DataStorage::dslDensity) {
698
699	<	AtomCommRealRow->scatter(atomRowData.density,
699	>	AtomPlanRealRow->scatter(atomRowData.density,
700		snap_->atomData.density);
701
702		int n = snap_->atomData.density.size();
703	<	std::vector<RealType> rho_tmp(n, 0.0);
704	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
703	>	vector<RealType> rho_tmp(n, 0.0);
704	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
705		for (int i = 0; i < n; i++)
706		snap_->atomData.density[i] += rho_tmp[i];
707		}
708	+
709	+	// this isn't necessary if we don't have polarizable atoms, but
710	+	// we'll leave it here for now.
711	+	if (storageLayout_ & DataStorage::dslElectricField) {
712	+
713	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
714	+	snap_->atomData.electricField);
715	+
716	+	int n = snap_->atomData.electricField.size();
717	+	vector<Vector3d> field_tmp(n, V3Zero);
718	+	AtomPlanVectorColumn->scatter(atomColData.electricField,
719	+	field_tmp);
720	+	for (int i = 0; i < n; i++)
721	+	snap_->atomData.electricField[i] += field_tmp[i];
722	+	}
723		#endif
724		}
725	<
725	>
726	>	/*
727	>	* redistributes information obtained during the pre-pair loop out to
728	>	* row and column-indexed data structures
729	>	*/
730		void ForceMatrixDecomposition::distributeIntermediateData() {
731		snap_ = sman_->getCurrentSnapshot();
732		storageLayout_ = sman_->getStorageLayout();
733		#ifdef IS_MPI
734		if (storageLayout_ & DataStorage::dslFunctional) {
735	<	AtomCommRealRow->gather(snap_->atomData.functional,
735	>	AtomPlanRealRow->gather(snap_->atomData.functional,
736		atomRowData.functional);
737	<	AtomCommRealColumn->gather(snap_->atomData.functional,
737	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
738		atomColData.functional);
739		}
740
741		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
742	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
742	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
743		atomRowData.functionalDerivative);
744	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
744	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
745		atomColData.functionalDerivative);
746		}
747		#endif
#	Line 314 | Line 755 | namespace OpenMD {
755		int n = snap_->atomData.force.size();
756		vector<Vector3d> frc_tmp(n, V3Zero);
757
758	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
758	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
759		for (int i = 0; i < n; i++) {
760		snap_->atomData.force[i] += frc_tmp[i];
761		frc_tmp[i] = 0.0;
762		}
763
764	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
765	<	for (int i = 0; i < n; i++)
764	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
765	>	for (int i = 0; i < n; i++) {
766		snap_->atomData.force[i] += frc_tmp[i];
767	<
768	<
767	>	}
768	>
769		if (storageLayout_ & DataStorage::dslTorque) {
770
771	<	int nt = snap_->atomData.force.size();
771	>	int nt = snap_->atomData.torque.size();
772		vector<Vector3d> trq_tmp(nt, V3Zero);
773
774	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
775	<	for (int i = 0; i < n; i++) {
774	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
775	>	for (int i = 0; i < nt; i++) {
776		snap_->atomData.torque[i] += trq_tmp[i];
777		trq_tmp[i] = 0.0;
778		}
779
780	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
781	<	for (int i = 0; i < n; i++)
780	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
781	>	for (int i = 0; i < nt; i++)
782		snap_->atomData.torque[i] += trq_tmp[i];
783		}
784	+
785	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
786	+
787	+	int ns = snap_->atomData.skippedCharge.size();
788	+	vector<RealType> skch_tmp(ns, 0.0);
789	+
790	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
791	+	for (int i = 0; i < ns; i++) {
792	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
793	+	skch_tmp[i] = 0.0;
794	+	}
795	+
796	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
797	+	for (int i = 0; i < ns; i++)
798	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
799	+
800	+	}
801
802	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
803	+
804	+	int nq = snap_->atomData.flucQFrc.size();
805	+	vector<RealType> fqfrc_tmp(nq, 0.0);
806	+
807	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
808	+	for (int i = 0; i < nq; i++) {
809	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
810	+	fqfrc_tmp[i] = 0.0;
811	+	}
812	+
813	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
814	+	for (int i = 0; i < nq; i++)
815	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
816	+
817	+	}
818	+
819	+	if (storageLayout_ & DataStorage::dslElectricField) {
820	+
821	+	int nef = snap_->atomData.electricField.size();
822	+	vector<Vector3d> efield_tmp(nef, V3Zero);
823	+
824	+	AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
825	+	for (int i = 0; i < nef; i++) {
826	+	snap_->atomData.electricField[i] += efield_tmp[i];
827	+	efield_tmp[i] = 0.0;
828	+	}
829	+
830	+	AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
831	+	for (int i = 0; i < nef; i++)
832	+	snap_->atomData.electricField[i] += efield_tmp[i];
833	+	}
834	+
835	+
836		nLocal_ = snap_->getNumberOfAtoms();
837
838	<	vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
839	<	vector<RealType> (nLocal_, 0.0));
838	>	vector<potVec> pot_temp(nLocal_,
839	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
840	>	vector<potVec> expot_temp(nLocal_,
841	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
842	>
843	>	// scatter/gather pot_row into the members of my column
844	>
845	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
846	>	AtomPlanPotRow->scatter(expot_row, expot_temp);
847	>
848	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
849	>	pairwisePot += pot_temp[ii];
850	>
851	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
852	>	excludedPot += expot_temp[ii];
853	>
854	>	if (storageLayout_ & DataStorage::dslParticlePot) {
855	>	// This is the pairwise contribution to the particle pot.  The
856	>	// embedding contribution is added in each of the low level
857	>	// non-bonded routines.  In single processor, this is done in
858	>	// unpackInteractionData, not in collectData.
859	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
860	>	for (int i = 0; i < nLocal_; i++) {
861	>	// factor of two is because the total potential terms are divided
862	>	// by 2 in parallel due to row/ column scatter
863	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
864	>	}
865	>	}
866	>	}
867	>
868	>	fill(pot_temp.begin(), pot_temp.end(),
869	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
870	>	fill(expot_temp.begin(), expot_temp.end(),
871	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
872	>
873	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
874	>	AtomPlanPotColumn->scatter(expot_col, expot_temp);
875
876	<	for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
877	<	AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
878	<	for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
879	<	pot_local[i] += pot_temp[i][ii];
876	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
877	>	pairwisePot += pot_temp[ii];
878	>
879	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
880	>	excludedPot += expot_temp[ii];
881	>
882	>	if (storageLayout_ & DataStorage::dslParticlePot) {
883	>	// This is the pairwise contribution to the particle pot.  The
884	>	// embedding contribution is added in each of the low level
885	>	// non-bonded routines.  In single processor, this is done in
886	>	// unpackInteractionData, not in collectData.
887	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
888	>	for (int i = 0; i < nLocal_; i++) {
889	>	// factor of two is because the total potential terms are divided
890	>	// by 2 in parallel due to row/ column scatter
891	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
892	>	}
893		}
894		}
895	+
896	+	if (storageLayout_ & DataStorage::dslParticlePot) {
897	+	int npp = snap_->atomData.particlePot.size();
898	+	vector<RealType> ppot_temp(npp, 0.0);
899	+
900	+	// This is the direct or embedding contribution to the particle
901	+	// pot.
902	+
903	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
904	+	for (int i = 0; i < npp; i++) {
905	+	snap_->atomData.particlePot[i] += ppot_temp[i];
906	+	}
907	+
908	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
909	+
910	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
911	+	for (int i = 0; i < npp; i++) {
912	+	snap_->atomData.particlePot[i] += ppot_temp[i];
913	+	}
914	+	}
915	+
916	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
917	+	RealType ploc1 = pairwisePot[ii];
918	+	RealType ploc2 = 0.0;
919	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
920	+	pairwisePot[ii] = ploc2;
921	+	}
922	+
923	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
924	+	RealType ploc1 = excludedPot[ii];
925	+	RealType ploc2 = 0.0;
926	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
927	+	excludedPot[ii] = ploc2;
928	+	}
929	+
930	+	// Here be dragons.
931	+	MPI::Intracomm col = colComm.getComm();
932	+
933	+	col.Allreduce(MPI::IN_PLACE,
934	+	&snap_->frameData.conductiveHeatFlux[0], 3,
935	+	MPI::REALTYPE, MPI::SUM);
936	+
937	+
938		#endif
939	+
940		}
941
942	<	int ForceMatrixDecomposition::getNAtomsInRow() {
942	>	/**
943	>	* Collects information obtained during the post-pair (and embedding
944	>	* functional) loops onto local data structures.
945	>	*/
946	>	void ForceMatrixDecomposition::collectSelfData() {
947	>	snap_ = sman_->getCurrentSnapshot();
948	>	storageLayout_ = sman_->getStorageLayout();
949	>
950		#ifdef IS_MPI
951	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
952	+	RealType ploc1 = embeddingPot[ii];
953	+	RealType ploc2 = 0.0;
954	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
955	+	embeddingPot[ii] = ploc2;
956	+	}
957	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
958	+	RealType ploc1 = excludedSelfPot[ii];
959	+	RealType ploc2 = 0.0;
960	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
961	+	excludedSelfPot[ii] = ploc2;
962	+	}
963	+	#endif
964	+
965	+	}
966	+
967	+
968	+
969	+	int& ForceMatrixDecomposition::getNAtomsInRow() {
970	+	#ifdef IS_MPI
971		return nAtomsInRow_;
972		#else
973		return nLocal_;
#	Line 366 | Line 977 | namespace OpenMD {
977		/**
978		* returns the list of atoms belonging to this group.
979		*/
980	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
980	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
981		#ifdef IS_MPI
982		return groupListRow_[cg1];
983		#else
#	Line 374 | Line 985 | namespace OpenMD {
985		#endif
986		}
987
988	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
988	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
989		#ifdef IS_MPI
990		return groupListCol_[cg2];
991		#else
#	Line 391 | Line 1002 | namespace OpenMD {
1002		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
1003		#endif
1004
1005	<	snap_->wrapVector(d);
1005	>	if (usePeriodicBoundaryConditions_) {
1006	>	snap_->wrapVector(d);
1007	>	}
1008		return d;
1009		}
1010
1011	+	Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
1012	+	#ifdef IS_MPI
1013	+	return cgColData.velocity[cg2];
1014	+	#else
1015	+	return snap_->cgData.velocity[cg2];
1016	+	#endif
1017	+	}
1018
1019	+	Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
1020	+	#ifdef IS_MPI
1021	+	return atomColData.velocity[atom2];
1022	+	#else
1023	+	return snap_->atomData.velocity[atom2];
1024	+	#endif
1025	+	}
1026	+
1027	+
1028		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
1029
1030		Vector3d d;
#	Line 405 | Line 1034 | namespace OpenMD {
1034		#else
1035		d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
1036		#endif
1037	<
1038	<	snap_->wrapVector(d);
1037	>	if (usePeriodicBoundaryConditions_) {
1038	>	snap_->wrapVector(d);
1039	>	}
1040		return d;
1041		}
1042
#	Line 418 | Line 1048 | namespace OpenMD {
1048		#else
1049		d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1050		#endif
1051	<
1052	<	snap_->wrapVector(d);
1051	>	if (usePeriodicBoundaryConditions_) {
1052	>	snap_->wrapVector(d);
1053	>	}
1054		return d;
1055		}
1056
1057	<	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1057	>	RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1058		#ifdef IS_MPI
1059		return massFactorsRow[atom1];
1060		#else
1061	<	return massFactorsLocal[atom1];
1061	>	return massFactors[atom1];
1062		#endif
1063		}
1064
1065	<	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1065	>	RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1066		#ifdef IS_MPI
1067		return massFactorsCol[atom2];
1068		#else
1069	<	return massFactorsLocal[atom2];
1069	>	return massFactors[atom2];
1070		#endif
1071
1072		}
#	Line 448 | Line 1079 | namespace OpenMD {
1079		#else
1080		d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1081		#endif
1082	<
1083	<	snap_->wrapVector(d);
1082	>	if (usePeriodicBoundaryConditions_) {
1083	>	snap_->wrapVector(d);
1084	>	}
1085		return d;
1086		}
1087
1088	<	vector<int> ForceMatrixDecomposition::getSkipsForRowAtom(int atom1) {
1089	<	#ifdef IS_MPI
458	<	return skipsForRowAtom[atom1];
459	<	#else
460	<	return skipsForLocalAtom[atom1];
461	<	#endif
1088	>	vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1089	>	return excludesForAtom[atom1];
1090		}
1091
1092		/**
1093	<	* there are a number of reasons to skip a pair or a particle mostly
1094	<	* we do this to exclude atoms who are involved in short range
467	<	* interactions (bonds, bends, torsions), but we also need to
468	<	* exclude some overcounted interactions that result from the
469	<	* parallel decomposition.
1093	>	* We need to exclude some overcounted interactions that result from
1094	>	* the parallel decomposition.
1095		*/
1096	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
1096	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1097		int unique_id_1, unique_id_2;
1098	<
1098	>
1099		#ifdef IS_MPI
1100		// in MPI, we have to look up the unique IDs for each atom
1101		unique_id_1 = AtomRowToGlobal[atom1];
1102		unique_id_2 = AtomColToGlobal[atom2];
1103	+	// group1 = cgRowToGlobal[cg1];
1104	+	// group2 = cgColToGlobal[cg2];
1105	+	#else
1106	+	unique_id_1 = AtomLocalToGlobal[atom1];
1107	+	unique_id_2 = AtomLocalToGlobal[atom2];
1108	+	int group1 = cgLocalToGlobal[cg1];
1109	+	int group2 = cgLocalToGlobal[cg2];
1110	+	#endif
1111
479	–	// this situation should only arise in MPI simulations
1112		if (unique_id_1 == unique_id_2) return true;
1113	<
1113	>
1114	>	#ifdef IS_MPI
1115		// this prevents us from doing the pair on multiple processors
1116		if (unique_id_1 < unique_id_2) {
1117		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1118		} else {
1119	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1119	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1120		}
1121	<	#else
1122	<	// in the normal loop, the atom numbers are unique
1123	<	unique_id_1 = atom1;
1124	<	unique_id_2 = atom2;
1121	>	#endif
1122	>
1123	>	#ifndef IS_MPI
1124	>	if (group1 == group2) {
1125	>	if (unique_id_1 < unique_id_2) return true;
1126	>	}
1127		#endif
1128
1129	<	#ifdef IS_MPI
495	<	for (vector<int>::iterator i = skipsForRowAtom[atom1].begin();
496	<	i != skipsForRowAtom[atom1].end(); ++i) {
497	<	if ( (*i) == unique_id_2 ) return true;
498	<	}
499	<	#else
500	<	for (vector<int>::iterator i = skipsForLocalAtom[atom1].begin();
501	<	i != skipsForLocalAtom[atom1].end(); ++i) {
502	<	if ( (*i) == unique_id_2 ) return true;
503	<	}
504	<	#endif
1129	>	return false;
1130		}
1131
1132	<	int ForceMatrixDecomposition::getTopoDistance(int atom1, int atom2) {
1132	>	/**
1133	>	* We need to handle the interactions for atoms who are involved in
1134	>	* the same rigid body as well as some short range interactions
1135	>	* (bonds, bends, torsions) differently from other interactions.
1136	>	* We'll still visit the pairwise routines, but with a flag that
1137	>	* tells those routines to exclude the pair from direct long range
1138	>	* interactions.  Some indirect interactions (notably reaction
1139	>	* field) must still be handled for these pairs.
1140	>	*/
1141	>	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1142	>
1143	>	// excludesForAtom was constructed to use row/column indices in the MPI
1144	>	// version, and to use local IDs in the non-MPI version:
1145
1146	<	#ifdef IS_MPI
1147	<	for (int i = 0; i < toposForRowAtom[atom1].size(); i++) {
1148	<	if ( toposForRowAtom[atom1][i] == atom2 ) return topoDistRow[atom1][i];
512	<	}
513	<	#else
514	<	for (int i = 0; i < toposForLocalAtom[atom1].size(); i++) {
515	<	if ( toposForLocalAtom[atom1][i] == atom2 ) return topoDistLocal[atom1][i];
1146	>	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1147	>	i != excludesForAtom[atom1].end(); ++i) {
1148	>	if ( (*i) == atom2 ) return true;
1149		}
517	–	#endif
1150
1151	<	// zero is default for unconnected (i.e. normal) pair interactions
520	<	return 0;
1151	>	return false;
1152		}
1153
1154	+
1155		void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
1156		#ifdef IS_MPI
1157		atomRowData.force[atom1] += fg;
#	Line 537 | Line 1169 | namespace OpenMD {
1169		}
1170
1171		// filling interaction blocks with pointers
1172	<	InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {
1173	<	InteractionData idat;
1172	>	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1173	>	int atom1, int atom2) {
1174
1175	+	idat.excluded = excludeAtomPair(atom1, atom2);
1176	+
1177		#ifdef IS_MPI
1178	<
1179	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1180	<	ff_->getAtomType(identsCol[atom2]) );
1178	>	//idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1179	>	idat.atid1 = identsRow[atom1];
1180	>	idat.atid2 = identsCol[atom2];
1181
1182	+	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1183	+	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1184	+	} else {
1185	+	idat.sameRegion = false;
1186	+	}
1187	+
1188		if (storageLayout_ & DataStorage::dslAmat) {
1189		idat.A1 = &(atomRowData.aMat[atom1]);
1190		idat.A2 = &(atomColData.aMat[atom2]);
1191		}
1192
553	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
554	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
555	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
556	–	}
557	–
1193		if (storageLayout_ & DataStorage::dslTorque) {
1194		idat.t1 = &(atomRowData.torque[atom1]);
1195		idat.t2 = &(atomColData.torque[atom2]);
1196		}
1197
1198	+	if (storageLayout_ & DataStorage::dslDipole) {
1199	+	idat.dipole1 = &(atomRowData.dipole[atom1]);
1200	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1201	+	}
1202	+
1203	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1204	+	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1205	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1206	+	}
1207	+
1208		if (storageLayout_ & DataStorage::dslDensity) {
1209		idat.rho1 = &(atomRowData.density[atom1]);
1210		idat.rho2 = &(atomColData.density[atom2]);
1211		}
1212
1213	+	if (storageLayout_ & DataStorage::dslFunctional) {
1214	+	idat.frho1 = &(atomRowData.functional[atom1]);
1215	+	idat.frho2 = &(atomColData.functional[atom2]);
1216	+	}
1217	+
1218		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1219		idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1220		idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1221		}
1222
1223	+	if (storageLayout_ & DataStorage::dslParticlePot) {
1224	+	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1225	+	idat.particlePot2 = &(atomColData.particlePot[atom2]);
1226	+	}
1227	+
1228	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1229	+	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1230	+	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1231	+	}
1232	+
1233	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1234	+	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1235	+	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1236	+	}
1237	+
1238		#else
1239	+
1240	+	//idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1241	+	idat.atid1 = idents[atom1];
1242	+	idat.atid2 = idents[atom2];
1243
1244	<	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
1245	<	ff_->getAtomType(identsLocal[atom2]) );
1244	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1245	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1246	>	} else {
1247	>	idat.sameRegion = false;
1248	>	}
1249
1250		if (storageLayout_ & DataStorage::dslAmat) {
1251		idat.A1 = &(snap_->atomData.aMat[atom1]);
1252		idat.A2 = &(snap_->atomData.aMat[atom2]);
1253		}
1254
583	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
584	–	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
585	–	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
586	–	}
587	–
1255		if (storageLayout_ & DataStorage::dslTorque) {
1256		idat.t1 = &(snap_->atomData.torque[atom1]);
1257		idat.t2 = &(snap_->atomData.torque[atom2]);
1258		}
1259
1260	<	if (storageLayout_ & DataStorage::dslDensity) {
1260	>	if (storageLayout_ & DataStorage::dslDipole) {
1261	>	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1262	>	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1263	>	}
1264	>
1265	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1266	>	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1267	>	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1268	>	}
1269	>
1270	>	if (storageLayout_ & DataStorage::dslDensity) {
1271		idat.rho1 = &(snap_->atomData.density[atom1]);
1272		idat.rho2 = &(snap_->atomData.density[atom2]);
1273		}
1274
1275	+	if (storageLayout_ & DataStorage::dslFunctional) {
1276	+	idat.frho1 = &(snap_->atomData.functional[atom1]);
1277	+	idat.frho2 = &(snap_->atomData.functional[atom2]);
1278	+	}
1279	+
1280		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1281		idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1282		idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1283		}
1284	+
1285	+	if (storageLayout_ & DataStorage::dslParticlePot) {
1286	+	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1287	+	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1288	+	}
1289	+
1290	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1291	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1292	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1293	+	}
1294	+
1295	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1296	+	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1297	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1298	+	}
1299	+
1300		#endif
603	–	return idat;
1301		}
1302
1303	<	InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
1304	<
608	<	InteractionData idat;
1303	>
1304	>	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1305		#ifdef IS_MPI
1306	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1307	<	ff_->getAtomType(identsCol[atom2]) );
1306	>	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1307	>	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1308	>	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1309	>	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1310
1311	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1312	<	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1313	<	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1311	>	atomRowData.force[atom1] += *(idat.f1);
1312	>	atomColData.force[atom2] -= *(idat.f1);
1313	>
1314	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1315	>	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1316	>	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1317		}
1318	<	if (storageLayout_ & DataStorage::dslTorque) {
1319	<	idat.t1 = &(atomRowData.torque[atom1]);
1320	<	idat.t2 = &(atomColData.torque[atom2]);
1318	>
1319	>	if (storageLayout_ & DataStorage::dslElectricField) {
1320	>	atomRowData.electricField[atom1] += *(idat.eField1);
1321	>	atomColData.electricField[atom2] += *(idat.eField2);
1322		}
1323	<	if (storageLayout_ & DataStorage::dslForce) {
622	<	idat.t1 = &(atomRowData.force[atom1]);
623	<	idat.t2 = &(atomColData.force[atom2]);
624	<	}
1323	>
1324		#else
1325	<	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
1326	<	ff_->getAtomType(identsLocal[atom2]) );
1325	>	pairwisePot += *(idat.pot);
1326	>	excludedPot += *(idat.excludedPot);
1327
1328	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1329	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1330	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1328	>	snap_->atomData.force[atom1] += *(idat.f1);
1329	>	snap_->atomData.force[atom2] -= *(idat.f1);
1330	>
1331	>	if (idat.doParticlePot) {
1332	>	// This is the pairwise contribution to the particle pot.  The
1333	>	// embedding contribution is added in each of the low level
1334	>	// non-bonded routines.  In parallel, this calculation is done
1335	>	// in collectData, not in unpackInteractionData.
1336	>	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1337	>	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1338		}
1339	<	if (storageLayout_ & DataStorage::dslTorque) {
1340	<	idat.t1 = &(snap_->atomData.torque[atom1]);
1341	<	idat.t2 = &(snap_->atomData.torque[atom2]);
1339	>
1340	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1341	>	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1342	>	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1343		}
1344	<	if (storageLayout_ & DataStorage::dslForce) {
1345	<	idat.t1 = &(snap_->atomData.force[atom1]);
1346	<	idat.t2 = &(snap_->atomData.force[atom2]);
1344	>
1345	>	if (storageLayout_ & DataStorage::dslElectricField) {
1346	>	snap_->atomData.electricField[atom1] += *(idat.eField1);
1347	>	snap_->atomData.electricField[atom2] += *(idat.eField2);
1348		}
1349	<	#endif
1349	>
1350	>	#endif
1351	>
1352		}
1353
1354		/*
#	Line 647 | Line 1357 | namespace OpenMD {
1357		* first element of pair is row-indexed CutoffGroup
1358		* second element of pair is column-indexed CutoffGroup
1359		*/
1360	<	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1361	<
1362	<	vector<pair<int, int> > neighborList;
1363	<	#ifdef IS_MPI
1364	<	cellListRow_.clear();
655	<	cellListCol_.clear();
656	<	#else
657	<	cellList_.clear();
658	<	#endif
1360	>	void ForceMatrixDecomposition::buildNeighborList(vector<pair<int,int> >& neighborList) {
1361	>
1362	>	neighborList.clear();
1363	>	groupCutoffs cuts;
1364	>	bool doAllPairs = false;
1365
1366	<	// dangerous to not do error checking.
1367	<	RealType rCut_;
662	<
663	<	RealType rList_ = (rCut_ + skinThickness_);
664	<	RealType rl2 = rList_ * rList_;
1366	>	RealType rList_ = (largestRcut_ + skinThickness_);
1367	>	RealType rcut, rcutsq, rlistsq;
1368		Snapshot* snap_ = sman_->getCurrentSnapshot();
1369	<	Mat3x3d Hmat = snap_->getHmat();
1370	<	Vector3d Hx = Hmat.getColumn(0);
668	<	Vector3d Hy = Hmat.getColumn(1);
669	<	Vector3d Hz = Hmat.getColumn(2);
1369	>	Mat3x3d box;
1370	>	Mat3x3d invBox;
1371
671	–	nCells_.x() = (int) ( Hx.length() )/ rList_;
672	–	nCells_.y() = (int) ( Hy.length() )/ rList_;
673	–	nCells_.z() = (int) ( Hz.length() )/ rList_;
674	–
675	–	Mat3x3d invHmat = snap_->getInvHmat();
1372		Vector3d rs, scaled, dr;
1373		Vector3i whichCell;
1374		int cellIndex;
1375
1376		#ifdef IS_MPI
1377	<	for (int i = 0; i < nGroupsInRow_; i++) {
1378	<	rs = cgRowData.position[i];
1379	<	// scaled positions relative to the box vectors
1380	<	scaled = invHmat * rs;
1381	<	// wrap the vector back into the unit box by subtracting integer box
1382	<	// numbers
1383	<	for (int j = 0; j < 3; j++)
1384	<	scaled[j] -= roundMe(scaled[j]);
1385	<
1386	<	// find xyz-indices of cell that cutoffGroup is in.
1387	<	whichCell.x() = nCells_.x() * scaled.x();
1388	<	whichCell.y() = nCells_.y() * scaled.y();
693	<	whichCell.z() = nCells_.z() * scaled.z();
694	<
695	<	// find single index of this cell:
696	<	cellIndex = Vlinear(whichCell, nCells_);
697	<	// add this cutoff group to the list of groups in this cell;
698	<	cellListRow_[cellIndex].push_back(i);
1377	>	cellListRow_.clear();
1378	>	cellListCol_.clear();
1379	>	#else
1380	>	cellList_.clear();
1381	>	#endif
1382	>
1383	>	if (!usePeriodicBoundaryConditions_) {
1384	>	box = snap_->getBoundingBox();
1385	>	invBox = snap_->getInvBoundingBox();
1386	>	} else {
1387	>	box = snap_->getHmat();
1388	>	invBox = snap_->getInvHmat();
1389		}
1390	<
1391	<	for (int i = 0; i < nGroupsInCol_; i++) {
1392	<	rs = cgColData.position[i];
1393	<	// scaled positions relative to the box vectors
1394	<	scaled = invHmat * rs;
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	<
1400	<	// find xyz-indices of cell that cutoffGroup is in.
1401	<	whichCell.x() = nCells_.x() * scaled.x();
1402	<	whichCell.y() = nCells_.y() * scaled.y();
1403	<	whichCell.z() = nCells_.z() * scaled.z();
1404	<
1405	<	// find single index of this cell:
1406	<	cellIndex = Vlinear(whichCell, nCells_);
1407	<	// add this cutoff group to the list of groups in this cell;
1408	<	cellListCol_[cellIndex].push_back(i);
1409	<	}
1390	>
1391	>	Vector3d boxX = box.getColumn(0);
1392	>	Vector3d boxY = box.getColumn(1);
1393	>	Vector3d boxZ = box.getColumn(2);
1394	>
1395	>	nCells_.x() = int( boxX.length() / rList_ );
1396	>	nCells_.y() = int( boxY.length() / rList_ );
1397	>	nCells_.z() = int( boxZ.length() / rList_ );
1398	>
1399	>	// handle small boxes where the cell offsets can end up repeating cells
1400	>
1401	>	if (nCells_.x() < 3) doAllPairs = true;
1402	>	if (nCells_.y() < 3) doAllPairs = true;
1403	>	if (nCells_.z() < 3) doAllPairs = true;
1404	>
1405	>	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1406	>
1407	>	#ifdef IS_MPI
1408	>	cellListRow_.resize(nCtot);
1409	>	cellListCol_.resize(nCtot);
1410		#else
1411	<	for (int i = 0; i < nGroups_; i++) {
1412	<	rs = snap_->cgData.position[i];
1413	<	// scaled positions relative to the box vectors
1414	<	scaled = invHmat * rs;
1415	<	// wrap the vector back into the unit box by subtracting integer box
1416	<	// numbers
1417	<	for (int j = 0; j < 3; j++)
1418	<	scaled[j] -= roundMe(scaled[j]);
1411	>	cellList_.resize(nCtot);
1412	>	#endif
1413	>
1414	>	if (!doAllPairs) {
1415	>	#ifdef IS_MPI
1416	>
1417	>	for (int i = 0; i < nGroupsInRow_; i++) {
1418	>	rs = cgRowData.position[i];
1419	>
1420	>	// scaled positions relative to the box vectors
1421	>	scaled = invBox * rs;
1422	>
1423	>	// wrap the vector back into the unit box by subtracting integer box
1424	>	// numbers
1425	>	for (int j = 0; j < 3; j++) {
1426	>	scaled[j] -= roundMe(scaled[j]);
1427	>	scaled[j] += 0.5;
1428	>	// Handle the special case when an object is exactly on the
1429	>	// boundary (a scaled coordinate of 1.0 is the same as
1430	>	// scaled coordinate of 0.0)
1431	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1432	>	}
1433	>
1434	>	// find xyz-indices of cell that cutoffGroup is in.
1435	>	whichCell.x() = nCells_.x() * scaled.x();
1436	>	whichCell.y() = nCells_.y() * scaled.y();
1437	>	whichCell.z() = nCells_.z() * scaled.z();
1438	>
1439	>	// find single index of this cell:
1440	>	cellIndex = Vlinear(whichCell, nCells_);
1441	>
1442	>	// add this cutoff group to the list of groups in this cell;
1443	>	cellListRow_[cellIndex].push_back(i);
1444	>	}
1445	>	for (int i = 0; i < nGroupsInCol_; i++) {
1446	>	rs = cgColData.position[i];
1447	>
1448	>	// scaled positions relative to the box vectors
1449	>	scaled = invBox * rs;
1450	>
1451	>	// wrap the vector back into the unit box by subtracting integer box
1452	>	// numbers
1453	>	for (int j = 0; j < 3; j++) {
1454	>	scaled[j] -= roundMe(scaled[j]);
1455	>	scaled[j] += 0.5;
1456	>	// Handle the special case when an object is exactly on the
1457	>	// boundary (a scaled coordinate of 1.0 is the same as
1458	>	// scaled coordinate of 0.0)
1459	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1460	>	}
1461	>
1462	>	// find xyz-indices of cell that cutoffGroup is in.
1463	>	whichCell.x() = nCells_.x() * scaled.x();
1464	>	whichCell.y() = nCells_.y() * scaled.y();
1465	>	whichCell.z() = nCells_.z() * scaled.z();
1466	>
1467	>	// find single index of this cell:
1468	>	cellIndex = Vlinear(whichCell, nCells_);
1469	>
1470	>	// add this cutoff group to the list of groups in this cell;
1471	>	cellListCol_[cellIndex].push_back(i);
1472	>	}
1473	>
1474	>	#else
1475	>	for (int i = 0; i < nGroups_; i++) {
1476	>	rs = snap_->cgData.position[i];
1477	>
1478	>	// scaled positions relative to the box vectors
1479	>	scaled = invBox * rs;
1480	>
1481	>	// wrap the vector back into the unit box by subtracting integer box
1482	>	// numbers
1483	>	for (int j = 0; j < 3; j++) {
1484	>	scaled[j] -= roundMe(scaled[j]);
1485	>	scaled[j] += 0.5;
1486	>	// Handle the special case when an object is exactly on the
1487	>	// boundary (a scaled coordinate of 1.0 is the same as
1488	>	// scaled coordinate of 0.0)
1489	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1490	>	}
1491	>
1492	>	// find xyz-indices of cell that cutoffGroup is in.
1493	>	whichCell.x() = int(nCells_.x() * scaled.x());
1494	>	whichCell.y() = int(nCells_.y() * scaled.y());
1495	>	whichCell.z() = int(nCells_.z() * scaled.z());
1496	>
1497	>	// find single index of this cell:
1498	>	cellIndex = Vlinear(whichCell, nCells_);
1499	>
1500	>	// add this cutoff group to the list of groups in this cell;
1501	>	cellList_[cellIndex].push_back(i);
1502	>	}
1503
730	–	// find xyz-indices of cell that cutoffGroup is in.
731	–	whichCell.x() = nCells_.x() * scaled.x();
732	–	whichCell.y() = nCells_.y() * scaled.y();
733	–	whichCell.z() = nCells_.z() * scaled.z();
734	–
735	–	// find single index of this cell:
736	–	cellIndex = Vlinear(whichCell, nCells_);
737	–	// add this cutoff group to the list of groups in this cell;
738	–	cellList_[cellIndex].push_back(i);
739	–	}
1504		#endif
1505
1506	<
1507	<
1508	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1509	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1510	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
747	<	Vector3i m1v(m1x, m1y, m1z);
748	<	int m1 = Vlinear(m1v, nCells_);
749	<
750	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
751	<	os != cellOffsets_.end(); ++os) {
1506	>	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1507	>	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1508	>	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1509	>	Vector3i m1v(m1x, m1y, m1z);
1510	>	int m1 = Vlinear(m1v, nCells_);
1511
1512	<	Vector3i m2v = m1v + (*os);
1513	<
1514	<	if (m2v.x() >= nCells_.x()) {
1515	<	m2v.x() = 0;
1516	<	} else if (m2v.x() < 0) {
758	<	m2v.x() = nCells_.x() - 1;
759	<	}
760	<
761	<	if (m2v.y() >= nCells_.y()) {
762	<	m2v.y() = 0;
763	<	} else if (m2v.y() < 0) {
764	<	m2v.y() = nCells_.y() - 1;
765	<	}
766	<
767	<	if (m2v.z() >= nCells_.z()) {
768	<	m2v.z() = 0;
769	<	} else if (m2v.z() < 0) {
770	<	m2v.z() = nCells_.z() - 1;
771	<	}
772	<
773	<	int m2 = Vlinear (m2v, nCells_);
1512	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1513	>	os != cellOffsets_.end(); ++os) {
1514	>
1515	>	Vector3i m2v = m1v + (*os);
1516	>
1517
1518	<	#ifdef IS_MPI
1519	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1520	<	j1 != cellListRow_[m1].end(); ++j1) {
1521	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1522	<	j2 != cellListCol_[m2].end(); ++j2) {
1523	<
1524	<	// Always do this if we're in different cells or if
1525	<	// we're in the same cell and the global index of the
1526	<	// j2 cutoff group is less than the j1 cutoff group
1518	>	if (m2v.x() >= nCells_.x()) {
1519	>	m2v.x() = 0;
1520	>	} else if (m2v.x() < 0) {
1521	>	m2v.x() = nCells_.x() - 1;
1522	>	}
1523	>
1524	>	if (m2v.y() >= nCells_.y()) {
1525	>	m2v.y() = 0;
1526	>	} else if (m2v.y() < 0) {
1527	>	m2v.y() = nCells_.y() - 1;
1528	>	}
1529	>
1530	>	if (m2v.z() >= nCells_.z()) {
1531	>	m2v.z() = 0;
1532	>	} else if (m2v.z() < 0) {
1533	>	m2v.z() = nCells_.z() - 1;
1534	>	}
1535
1536	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1536	>	int m2 = Vlinear (m2v, nCells_);
1537	>
1538	>	#ifdef IS_MPI
1539	>	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1540	>	j1 != cellListRow_[m1].end(); ++j1) {
1541	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1542	>	j2 != cellListCol_[m2].end(); ++j2) {
1543	>
1544	>	// In parallel, we need to visit *all* pairs of row
1545	>	// & column indicies and will divide labor in the
1546	>	// force evaluation later.
1547		dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1548	<	snap_->wrapVector(dr);
1549	<	if (dr.lengthSquare() < rl2) {
789	<	neighborList.push_back(make_pair((*j1), (*j2)));
1548	>	if (usePeriodicBoundaryConditions_) {
1549	>	snap_->wrapVector(dr);
1550		}
1551	+	getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq );
1552	+	if (dr.lengthSquare() < rlistsq) {
1553	+	neighborList.push_back(make_pair((*j1), (*j2)));
1554	+	}
1555		}
1556		}
793	–	}
1557		#else
1558	<	for (vector<int>::iterator j1 = cellList_[m1].begin();
1559	<	j1 != cellList_[m1].end(); ++j1) {
1560	<	for (vector<int>::iterator j2 = cellList_[m2].begin();
1561	<	j2 != cellList_[m2].end(); ++j2) {
1562	<
1563	<	// Always do this if we're in different cells or if
1564	<	// we're in the same cell and the global index of the
1565	<	// j2 cutoff group is less than the j1 cutoff group
1558	>	for (vector<int>::iterator j1 = cellList_[m1].begin();
1559	>	j1 != cellList_[m1].end(); ++j1) {
1560	>	for (vector<int>::iterator j2 = cellList_[m2].begin();
1561	>	j2 != cellList_[m2].end(); ++j2) {
1562	>
1563	>	// Always do this if we're in different cells or if
1564	>	// we're in the same cell and the global index of
1565	>	// the j2 cutoff group is greater than or equal to
1566	>	// the j1 cutoff group.  Note that Rappaport's code
1567	>	// has a "less than" conditional here, but that
1568	>	// deals with atom-by-atom computation.  OpenMD
1569	>	// allows atoms within a single cutoff group to
1570	>	// interact with each other.
1571
1572	<	if (m2 != m1 || (*j2) < (*j1)) {
1573	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1574	<	snap_->wrapVector(dr);
1575	<	if (dr.lengthSquare() < rl2) {
1576	<	neighborList.push_back(make_pair((*j1), (*j2)));
1572	>	if (m2 != m1 || (*j2) >= (*j1) ) {
1573	>
1574	>	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1575	>	if (usePeriodicBoundaryConditions_) {
1576	>	snap_->wrapVector(dr);
1577	>	}
1578	>	getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq );
1579	>	if (dr.lengthSquare() < rlistsq) {
1580	>	neighborList.push_back(make_pair((*j1), (*j2)));
1581	>	}
1582		}
1583		}
1584		}
812	–	}
1585		#endif
1586	+	}
1587		}
1588		}
1589		}
1590	+	} else {
1591	+	// branch to do all cutoff group pairs
1592	+	#ifdef IS_MPI
1593	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1594	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1595	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1596	+	if (usePeriodicBoundaryConditions_) {
1597	+	snap_->wrapVector(dr);
1598	+	}
1599	+	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq);
1600	+	if (dr.lengthSquare() < rlistsq) {
1601	+	neighborList.push_back(make_pair(j1, j2));
1602	+	}
1603	+	}
1604	+	}
1605	+	#else
1606	+	// include all groups here.
1607	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1608	+	// include self group interactions j2 == j1
1609	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1610	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1611	+	if (usePeriodicBoundaryConditions_) {
1612	+	snap_->wrapVector(dr);
1613	+	}
1614	+	getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq );
1615	+	if (dr.lengthSquare() < rlistsq) {
1616	+	neighborList.push_back(make_pair(j1, j2));
1617	+	}
1618	+	}
1619	+	}
1620	+	#endif
1621		}
1622	<
1622	>
1623		// save the local cutoff group positions for the check that is
1624		// done on each loop:
1625		saved_CG_positions_.clear();
1626		for (int i = 0; i < nGroups_; i++)
1627		saved_CG_positions_.push_back(snap_->cgData.position[i]);
824	–
825	–	return neighborList;
1628		}
1629		} //end namespace OpenMD

Diff Legend

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