ViewVC Help

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

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

Comparing:
branches/development/src/parallel/ForceDecomposition.cpp (file contents), Revision 1544 by gezelter, Fri Mar 18 19:31:52 2011 UTC vs.
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1803 by gezelter, Wed Oct 3 14:20:07 2012 UTC

#	Line 36 | Line 36
36		* [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37		* [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38		* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
39	<	* [4]  Vardeman & Gezelter, in progress (2009).
39	>	* [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40	>	* [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41		*/
42	<	#include "parallel/ForceDecomposition.hpp"
42	>	#include "parallel/ForceMatrixDecomposition.hpp"
43		#include "math/SquareMatrix3.hpp"
44		#include "nonbonded/NonBondedInteraction.hpp"
45		#include "brains/SnapshotManager.hpp"
46	+	#include "brains/PairList.hpp"
47
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		*/
93	<
94	<	void ForceDecomposition::distributeInitialData() {
95	<	#ifdef IS_MPI
96	<	Snapshot* snap = sman_->getCurrentSnapshot();
97	<	int nLocal = snap->getNumberOfAtoms();
98	<	int nGroups = snap->getNumberOfCutoffGroups();
93	>	void ForceMatrixDecomposition::distributeInitialData() {
94	>	snap_ = sman_->getCurrentSnapshot();
95	>	storageLayout_ = sman_->getStorageLayout();
96	>	ff_ = info_->getForceField();
97	>	nLocal_ = snap_->getNumberOfAtoms();
98	>
99	>	nGroups_ = info_->getNLocalCutoffGroups();
100	>	// gather the information for atomtype IDs (atids):
101	>	idents = info_->getIdentArray();
102	>	AtomLocalToGlobal = info_->getGlobalAtomIndices();
103	>	cgLocalToGlobal = info_->getGlobalGroupIndices();
104	>	vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
105
106	<	AtomCommIntI = new Communicator<Row,int>(nLocal);
61	<	AtomCommRealI = new Communicator<Row,RealType>(nLocal);
62	<	AtomCommVectorI = new Communicator<Row,Vector3d>(nLocal);
63	<	AtomCommMatrixI = new Communicator<Row,Mat3x3d>(nLocal);
106	>	massFactors = info_->getMassFactors();
107
108	<	AtomCommIntJ = new Communicator<Column,int>(nLocal);
109	<	AtomCommRealJ = new Communicator<Column,RealType>(nLocal);
110	<	AtomCommVectorJ = new Communicator<Column,Vector3d>(nLocal);
111	<	AtomCommMatrixJ = new Communicator<Column,Mat3x3d>(nLocal);
108	>	PairList* excludes = info_->getExcludedInteractions();
109	>	PairList* oneTwo = info_->getOneTwoInteractions();
110	>	PairList* oneThree = info_->getOneThreeInteractions();
111	>	PairList* oneFour = info_->getOneFourInteractions();
112	>
113	>	if (needVelocities_)
114	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
115	>	DataStorage::dslVelocity);
116	>	else
117	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
118	>
119	>	#ifdef IS_MPI
120	>
121	>	MPI::Intracomm row = rowComm.getComm();
122	>	MPI::Intracomm col = colComm.getComm();
123
124	<	cgCommIntI = new Communicator<Row,int>(nGroups);
125	<	cgCommVectorI = new Communicator<Row,Vector3d>(nGroups);
126	<	cgCommIntJ = new Communicator<Column,int>(nGroups);
127	<	cgCommVectorJ = new Communicator<Column,Vector3d>(nGroups);
124	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
125	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
126	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
127	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
128	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
129
130	<	int nAtomsInRow = AtomCommIntI->getSize();
131	<	int nAtomsInCol = AtomCommIntJ->getSize();
132	<	int nGroupsInRow = cgCommIntI->getSize();
133	<	int nGroupsInCol = cgCommIntJ->getSize();
130	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
131	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
132	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
133	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
134	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
135
136	<	vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
137	<	vector<RealType> (nAtomsInRow, 0.0));
138	<	vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
139	<	vector<RealType> (nAtomsInCol, 0.0));
136	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
137	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
138	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
139	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
140	>
141	>	nAtomsInRow_ = AtomPlanIntRow->getSize();
142	>	nAtomsInCol_ = AtomPlanIntColumn->getSize();
143	>	nGroupsInRow_ = cgPlanIntRow->getSize();
144	>	nGroupsInCol_ = cgPlanIntColumn->getSize();
145	>
146	>	// Modify the data storage objects with the correct layouts and sizes:
147	>	atomRowData.resize(nAtomsInRow_);
148	>	atomRowData.setStorageLayout(storageLayout_);
149	>	atomColData.resize(nAtomsInCol_);
150	>	atomColData.setStorageLayout(storageLayout_);
151	>	cgRowData.resize(nGroupsInRow_);
152	>	cgRowData.setStorageLayout(DataStorage::dslPosition);
153	>	cgColData.resize(nGroupsInCol_);
154	>	if (needVelocities_)
155	>	// we only need column velocities if we need them.
156	>	cgColData.setStorageLayout(DataStorage::dslPosition |
157	>	DataStorage::dslVelocity);
158	>	else
159	>	cgColData.setStorageLayout(DataStorage::dslPosition);
160	>
161	>	identsRow.resize(nAtomsInRow_);
162	>	identsCol.resize(nAtomsInCol_);
163
164	<	vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
164	>	AtomPlanIntRow->gather(idents, identsRow);
165	>	AtomPlanIntColumn->gather(idents, identsCol);
166	>
167	>	// allocate memory for the parallel objects
168	>	atypesRow.resize(nAtomsInRow_);
169	>	atypesCol.resize(nAtomsInCol_);
170
171	<	// gather the information for atomtype IDs (atids):
172	<	AtomCommIntI->gather(info_->getIdentArray(), identsRow);
173	<	AtomCommIntJ->gather(info_->getIdentArray(), identsCol);
171	>	for (int i = 0; i < nAtomsInRow_; i++)
172	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
173	>	for (int i = 0; i < nAtomsInCol_; i++)
174	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
175
176	<	AtomLocalToGlobal = info_->getLocalToGlobalAtomIndex();
177	<	AtomCommIntI->gather(AtomLocalToGlobal, AtomRowToGlobal);
93	<	AtomCommIntJ->gather(AtomLocalToGlobal, AtomColToGlobal);
176	>	pot_row.resize(nAtomsInRow_);
177	>	pot_col.resize(nAtomsInCol_);
178
179	<	cgLocalToGlobal = info_->getLocalToGlobalCutoffGroupIndex();
180	<	cgCommIntI->gather(cgLocalToGlobal, cgRowToGlobal);
97	<	cgCommIntJ->gather(cgLocalToGlobal, cgColToGlobal);
179	>	expot_row.resize(nAtomsInRow_);
180	>	expot_col.resize(nAtomsInCol_);
181
182	+	AtomRowToGlobal.resize(nAtomsInRow_);
183	+	AtomColToGlobal.resize(nAtomsInCol_);
184	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186	+
187	+	cgRowToGlobal.resize(nGroupsInRow_);
188	+	cgColToGlobal.resize(nGroupsInCol_);
189	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
191	+
192	+	massFactorsRow.resize(nAtomsInRow_);
193	+	massFactorsCol.resize(nAtomsInCol_);
194	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
195	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
196	+
197	+	groupListRow_.clear();
198	+	groupListRow_.resize(nGroupsInRow_);
199	+	for (int i = 0; i < nGroupsInRow_; i++) {
200	+	int gid = cgRowToGlobal[i];
201	+	for (int j = 0; j < nAtomsInRow_; j++) {
202	+	int aid = AtomRowToGlobal[j];
203	+	if (globalGroupMembership[aid] == gid)
204	+	groupListRow_[i].push_back(j);
205	+	}
206	+	}
207	+
208	+	groupListCol_.clear();
209	+	groupListCol_.resize(nGroupsInCol_);
210	+	for (int i = 0; i < nGroupsInCol_; i++) {
211	+	int gid = cgColToGlobal[i];
212	+	for (int j = 0; j < nAtomsInCol_; j++) {
213	+	int aid = AtomColToGlobal[j];
214	+	if (globalGroupMembership[aid] == gid)
215	+	groupListCol_[i].push_back(j);
216	+	}
217	+	}
218	+
219	+	excludesForAtom.clear();
220	+	excludesForAtom.resize(nAtomsInRow_);
221	+	toposForAtom.clear();
222	+	toposForAtom.resize(nAtomsInRow_);
223	+	topoDist.clear();
224	+	topoDist.resize(nAtomsInRow_);
225	+	for (int i = 0; i < nAtomsInRow_; i++) {
226	+	int iglob = AtomRowToGlobal[i];
227	+
228	+	for (int j = 0; j < nAtomsInCol_; j++) {
229	+	int jglob = AtomColToGlobal[j];
230	+
231	+	if (excludes->hasPair(iglob, jglob))
232	+	excludesForAtom[i].push_back(j);
233	+
234	+	if (oneTwo->hasPair(iglob, jglob)) {
235	+	toposForAtom[i].push_back(j);
236	+	topoDist[i].push_back(1);
237	+	} else {
238	+	if (oneThree->hasPair(iglob, jglob)) {
239	+	toposForAtom[i].push_back(j);
240	+	topoDist[i].push_back(2);
241	+	} else {
242	+	if (oneFour->hasPair(iglob, jglob)) {
243	+	toposForAtom[i].push_back(j);
244	+	topoDist[i].push_back(3);
245	+	}
246	+	}
247	+	}
248	+	}
249	+	}
250	+
251	+	#else
252	+	excludesForAtom.clear();
253	+	excludesForAtom.resize(nLocal_);
254	+	toposForAtom.clear();
255	+	toposForAtom.resize(nLocal_);
256	+	topoDist.clear();
257	+	topoDist.resize(nLocal_);
258	+
259	+	for (int i = 0; i < nLocal_; i++) {
260	+	int iglob = AtomLocalToGlobal[i];
261	+
262	+	for (int j = 0; j < nLocal_; j++) {
263	+	int jglob = AtomLocalToGlobal[j];
264	+
265	+	if (excludes->hasPair(iglob, jglob))
266	+	excludesForAtom[i].push_back(j);
267	+
268	+	if (oneTwo->hasPair(iglob, jglob)) {
269	+	toposForAtom[i].push_back(j);
270	+	topoDist[i].push_back(1);
271	+	} else {
272	+	if (oneThree->hasPair(iglob, jglob)) {
273	+	toposForAtom[i].push_back(j);
274	+	topoDist[i].push_back(2);
275	+	} else {
276	+	if (oneFour->hasPair(iglob, jglob)) {
277	+	toposForAtom[i].push_back(j);
278	+	topoDist[i].push_back(3);
279	+	}
280	+	}
281	+	}
282	+	}
283	+	}
284	+	#endif
285	+
286	+	// allocate memory for the parallel objects
287	+	atypesLocal.resize(nLocal_);
288	+
289	+	for (int i = 0; i < nLocal_; i++)
290	+	atypesLocal[i] = ff_->getAtomType(idents[i]);
291	+
292	+	groupList_.clear();
293	+	groupList_.resize(nGroups_);
294	+	for (int i = 0; i < nGroups_; i++) {
295	+	int gid = cgLocalToGlobal[i];
296	+	for (int j = 0; j < nLocal_; j++) {
297	+	int aid = AtomLocalToGlobal[j];
298	+	if (globalGroupMembership[aid] == gid) {
299	+	groupList_[i].push_back(j);
300	+	}
301	+	}
302	+	}
303	+
304	+
305	+	createGtypeCutoffMap();
306	+
307	+	}
308	+
309	+	void ForceMatrixDecomposition::createGtypeCutoffMap() {
310	+
311	+	RealType tol = 1e-6;
312	+	largestRcut_ = 0.0;
313	+	int atid;
314	+	set<AtomType*> atypes = info_->getSimulatedAtomTypes();
315	+
316	+	map<int, RealType> atypeCutoff;
317
318	+	for (set<AtomType*>::iterator at = atypes.begin();
319	+	at != atypes.end(); ++at){
320	+	atid = (*at)->getIdent();
321	+	if (userChoseCutoff_)
322	+	atypeCutoff[atid] = userCutoff_;
323	+	else
324	+	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
325	+	}
326	+
327	+	vector<RealType> gTypeCutoffs;
328	+	// first we do a single loop over the cutoff groups to find the
329	+	// largest cutoff for any atypes present in this group.
330	+	#ifdef IS_MPI
331	+	vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
332	+	groupRowToGtype.resize(nGroupsInRow_);
333	+	for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
334	+	vector<int> atomListRow = getAtomsInGroupRow(cg1);
335	+	for (vector<int>::iterator ia = atomListRow.begin();
336	+	ia != atomListRow.end(); ++ia) {
337	+	int atom1 = (*ia);
338	+	atid = identsRow[atom1];
339	+	if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
340	+	groupCutoffRow[cg1] = atypeCutoff[atid];
341	+	}
342	+	}
343	+
344	+	bool gTypeFound = false;
345	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
346	+	if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
347	+	groupRowToGtype[cg1] = gt;
348	+	gTypeFound = true;
349	+	}
350	+	}
351	+	if (!gTypeFound) {
352	+	gTypeCutoffs.push_back( groupCutoffRow[cg1] );
353	+	groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
354	+	}
355
356	+	}
357	+	vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
358	+	groupColToGtype.resize(nGroupsInCol_);
359	+	for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
360	+	vector<int> atomListCol = getAtomsInGroupColumn(cg2);
361	+	for (vector<int>::iterator jb = atomListCol.begin();
362	+	jb != atomListCol.end(); ++jb) {
363	+	int atom2 = (*jb);
364	+	atid = identsCol[atom2];
365	+	if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
366	+	groupCutoffCol[cg2] = atypeCutoff[atid];
367	+	}
368	+	}
369	+	bool gTypeFound = false;
370	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
371	+	if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
372	+	groupColToGtype[cg2] = gt;
373	+	gTypeFound = true;
374	+	}
375	+	}
376	+	if (!gTypeFound) {
377	+	gTypeCutoffs.push_back( groupCutoffCol[cg2] );
378	+	groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
379	+	}
380	+	}
381	+	#else
382
383	+	vector<RealType> groupCutoff(nGroups_, 0.0);
384	+	groupToGtype.resize(nGroups_);
385	+	for (int cg1 = 0; cg1 < nGroups_; cg1++) {
386	+	groupCutoff[cg1] = 0.0;
387	+	vector<int> atomList = getAtomsInGroupRow(cg1);
388	+	for (vector<int>::iterator ia = atomList.begin();
389	+	ia != atomList.end(); ++ia) {
390	+	int atom1 = (*ia);
391	+	atid = idents[atom1];
392	+	if (atypeCutoff[atid] > groupCutoff[cg1])
393	+	groupCutoff[cg1] = atypeCutoff[atid];
394	+	}
395	+
396	+	bool gTypeFound = false;
397	+	for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
398	+	if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
399	+	groupToGtype[cg1] = gt;
400	+	gTypeFound = true;
401	+	}
402	+	}
403	+	if (!gTypeFound) {
404	+	gTypeCutoffs.push_back( groupCutoff[cg1] );
405	+	groupToGtype[cg1] = gTypeCutoffs.size() - 1;
406	+	}
407	+	}
408	+	#endif
409
410	+	// Now we find the maximum group cutoff value present in the simulation
411
412	<	// still need:
413	<	// topoDist
414	<	// exclude
412	>	RealType groupMax = *max_element(gTypeCutoffs.begin(),
413	>	gTypeCutoffs.end());
414	>
415	>	#ifdef IS_MPI
416	>	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
417	>	MPI::MAX);
418		#endif
419	+
420	+	RealType tradRcut = groupMax;
421	+
422	+	for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) {
423	+	for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) {
424	+	RealType thisRcut;
425	+	switch(cutoffPolicy_) {
426	+	case TRADITIONAL:
427	+	thisRcut = tradRcut;
428	+	break;
429	+	case MIX:
430	+	thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
431	+	break;
432	+	case MAX:
433	+	thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
434	+	break;
435	+	default:
436	+	sprintf(painCave.errMsg,
437	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
438	+	"hit an unknown cutoff policy!\n");
439	+	painCave.severity = OPENMD_ERROR;
440	+	painCave.isFatal = 1;
441	+	simError();
442	+	break;
443	+	}
444	+
445	+	pair<int,int> key = make_pair(i,j);
446	+	gTypeCutoffMap[key].first = thisRcut;
447	+	if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
448	+	gTypeCutoffMap[key].second = thisRcut*thisRcut;
449	+	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
450	+	// sanity check
451	+
452	+	if (userChoseCutoff_) {
453	+	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
454	+	sprintf(painCave.errMsg,
455	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
456	+	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
457	+	painCave.severity = OPENMD_ERROR;
458	+	painCave.isFatal = 1;
459	+	simError();
460	+	}
461	+	}
462	+	}
463	+	}
464		}
465	+
466	+	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
467	+	int i, j;
468	+	#ifdef IS_MPI
469	+	i = groupRowToGtype[cg1];
470	+	j = groupColToGtype[cg2];
471	+	#else
472	+	i = groupToGtype[cg1];
473	+	j = groupToGtype[cg2];
474	+	#endif
475	+	return gTypeCutoffMap[make_pair(i,j)];
476	+	}
477	+
478	+	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
479	+	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
480	+	if (toposForAtom[atom1][j] == atom2)
481	+	return topoDist[atom1][j];
482	+	}
483	+	return 0;
484	+	}
485	+
486	+	void ForceMatrixDecomposition::zeroWorkArrays() {
487	+	pairwisePot = 0.0;
488	+	embeddingPot = 0.0;
489	+	excludedPot = 0.0;
490	+	excludedSelfPot = 0.0;
491	+
492	+	#ifdef IS_MPI
493	+	if (storageLayout_ & DataStorage::dslForce) {
494	+	fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
495	+	fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
496	+	}
497	+
498	+	if (storageLayout_ & DataStorage::dslTorque) {
499	+	fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
500	+	fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
501	+	}
502
503	+	fill(pot_row.begin(), pot_row.end(),
504	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
505
506	+	fill(pot_col.begin(), pot_col.end(),
507	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
508
509	<	void ForceDecomposition::distributeData()  {
509	>	fill(expot_row.begin(), expot_row.end(),
510	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
511	>
512	>	fill(expot_col.begin(), expot_col.end(),
513	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
514	>
515	>	if (storageLayout_ & DataStorage::dslParticlePot) {
516	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
517	>	0.0);
518	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
519	>	0.0);
520	>	}
521	>
522	>	if (storageLayout_ & DataStorage::dslDensity) {
523	>	fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
524	>	fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
525	>	}
526	>
527	>	if (storageLayout_ & DataStorage::dslFunctional) {
528	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
529	>	0.0);
530	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
531	>	0.0);
532	>	}
533	>
534	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
535	>	fill(atomRowData.functionalDerivative.begin(),
536	>	atomRowData.functionalDerivative.end(), 0.0);
537	>	fill(atomColData.functionalDerivative.begin(),
538	>	atomColData.functionalDerivative.end(), 0.0);
539	>	}
540	>
541	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
542	>	fill(atomRowData.skippedCharge.begin(),
543	>	atomRowData.skippedCharge.end(), 0.0);
544	>	fill(atomColData.skippedCharge.begin(),
545	>	atomColData.skippedCharge.end(), 0.0);
546	>	}
547	>
548	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
549	>	fill(atomRowData.flucQFrc.begin(),
550	>	atomRowData.flucQFrc.end(), 0.0);
551	>	fill(atomColData.flucQFrc.begin(),
552	>	atomColData.flucQFrc.end(), 0.0);
553	>	}
554	>
555	>	if (storageLayout_ & DataStorage::dslElectricField) {
556	>	fill(atomRowData.electricField.begin(),
557	>	atomRowData.electricField.end(), V3Zero);
558	>	fill(atomColData.electricField.begin(),
559	>	atomColData.electricField.end(), V3Zero);
560	>	}
561	>
562	>	#endif
563	>	// even in parallel, we need to zero out the local arrays:
564	>
565	>	if (storageLayout_ & DataStorage::dslParticlePot) {
566	>	fill(snap_->atomData.particlePot.begin(),
567	>	snap_->atomData.particlePot.end(), 0.0);
568	>	}
569	>
570	>	if (storageLayout_ & DataStorage::dslDensity) {
571	>	fill(snap_->atomData.density.begin(),
572	>	snap_->atomData.density.end(), 0.0);
573	>	}
574	>
575	>	if (storageLayout_ & DataStorage::dslFunctional) {
576	>	fill(snap_->atomData.functional.begin(),
577	>	snap_->atomData.functional.end(), 0.0);
578	>	}
579	>
580	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
581	>	fill(snap_->atomData.functionalDerivative.begin(),
582	>	snap_->atomData.functionalDerivative.end(), 0.0);
583	>	}
584	>
585	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
586	>	fill(snap_->atomData.skippedCharge.begin(),
587	>	snap_->atomData.skippedCharge.end(), 0.0);
588	>	}
589	>
590	>	if (storageLayout_ & DataStorage::dslElectricField) {
591	>	fill(snap_->atomData.electricField.begin(),
592	>	snap_->atomData.electricField.end(), V3Zero);
593	>	}
594	>	}
595	>
596	>
597	>	void ForceMatrixDecomposition::distributeData()  {
598	>	snap_ = sman_->getCurrentSnapshot();
599	>	storageLayout_ = sman_->getStorageLayout();
600		#ifdef IS_MPI
114	–	Snapshot* snap = sman_->getCurrentSnapshot();
601
602		// gather up the atomic positions
603	<	AtomCommVectorI->gather(snap->atomData.position,
604	<	snap->atomIData.position);
605	<	AtomCommVectorJ->gather(snap->atomData.position,
606	<	snap->atomJData.position);
603	>	AtomPlanVectorRow->gather(snap_->atomData.position,
604	>	atomRowData.position);
605	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
606	>	atomColData.position);
607
608		// gather up the cutoff group positions
609	<	cgCommVectorI->gather(snap->cgData.position,
610	<	snap->cgIData.position);
611	<	cgCommVectorJ->gather(snap->cgData.position,
612	<	snap->cgJData.position);
609	>
610	>	cgPlanVectorRow->gather(snap_->cgData.position,
611	>	cgRowData.position);
612	>
613	>	cgPlanVectorColumn->gather(snap_->cgData.position,
614	>	cgColData.position);
615	>
616	>
617	>
618	>	if (needVelocities_) {
619	>	// gather up the atomic velocities
620	>	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
621	>	atomColData.velocity);
622	>
623	>	cgPlanVectorColumn->gather(snap_->cgData.velocity,
624	>	cgColData.velocity);
625	>	}
626	>
627
628		// if needed, gather the atomic rotation matrices
629	<	if (snap->atomData.getStorageLayout() & DataStorage::dslAmat) {
630	<	AtomCommMatrixI->gather(snap->atomData.aMat,
631	<	snap->atomIData.aMat);
632	<	AtomCommMatrixJ->gather(snap->atomData.aMat,
633	<	snap->atomJData.aMat);
629	>	if (storageLayout_ & DataStorage::dslAmat) {
630	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
631	>	atomRowData.aMat);
632	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
633	>	atomColData.aMat);
634		}
635	<
636	<	// if needed, gather the atomic eletrostatic frames
637	<	if (snap->atomData.getStorageLayout() & DataStorage::dslElectroFrame) {
638	<	AtomCommMatrixI->gather(snap->atomData.electroFrame,
639	<	snap->atomIData.electroFrame);
640	<	AtomCommMatrixJ->gather(snap->atomData.electroFrame,
641	<	snap->atomJData.electroFrame);
635	>
636	>	// if needed, gather the atomic eletrostatic information
637	>	if (storageLayout_ & DataStorage::dslDipole) {
638	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
639	>	atomRowData.dipole);
640	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
641	>	atomColData.dipole);
642		}
643	+
644	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
645	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
646	+	atomRowData.quadrupole);
647	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
648	+	atomColData.quadrupole);
649	+	}
650	+
651	+	// if needed, gather the atomic fluctuating charge values
652	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
653	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
654	+	atomRowData.flucQPos);
655	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
656	+	atomColData.flucQPos);
657	+	}
658	+
659		#endif
660		}
661
662	<	void ForceDecomposition::collectIntermediateData() {
662	>	/* collects information obtained during the pre-pair loop onto local
663	>	* data structures.
664	>	*/
665	>	void ForceMatrixDecomposition::collectIntermediateData() {
666	>	snap_ = sman_->getCurrentSnapshot();
667	>	storageLayout_ = sman_->getStorageLayout();
668		#ifdef IS_MPI
148	–	Snapshot* snap = sman_->getCurrentSnapshot();
669
670	<	if (snap->atomData.getStorageLayout() & DataStorage::dslDensity) {
670	>	if (storageLayout_ & DataStorage::dslDensity) {
671	>
672	>	AtomPlanRealRow->scatter(atomRowData.density,
673	>	snap_->atomData.density);
674	>
675	>	int n = snap_->atomData.density.size();
676	>	vector<RealType> rho_tmp(n, 0.0);
677	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
678	>	for (int i = 0; i < n; i++)
679	>	snap_->atomData.density[i] += rho_tmp[i];
680	>	}
681
682	<	AtomCommRealI->scatter(snap->atomIData.density,
683	<	snap->atomData.density);
684	<
685	<	int n = snap->atomData.density.size();
686	<	std::vector<RealType> rho_tmp(n, 0.0);
687	<	AtomCommRealJ->scatter(snap->atomJData.density, rho_tmp);
682	>	if (storageLayout_ & DataStorage::dslElectricField) {
683	>
684	>	AtomPlanVectorRow->scatter(atomRowData.electricField,
685	>	snap_->atomData.electricField);
686	>
687	>	int n = snap_->atomData.electricField.size();
688	>	vector<Vector3d> field_tmp(n, V3Zero);
689	>	AtomPlanVectorColumn->scatter(atomColData.electricField,
690	>	field_tmp);
691		for (int i = 0; i < n; i++)
692	<	snap->atomData.density[i] += rho_tmp[i];
692	>	snap_->atomData.electricField[i] += field_tmp[i];
693		}
694		#endif
695		}
696	<
697	<	void ForceDecomposition::distributeIntermediateData() {
696	>
697	>	/*
698	>	* redistributes information obtained during the pre-pair loop out to
699	>	* row and column-indexed data structures
700	>	*/
701	>	void ForceMatrixDecomposition::distributeIntermediateData() {
702	>	snap_ = sman_->getCurrentSnapshot();
703	>	storageLayout_ = sman_->getStorageLayout();
704		#ifdef IS_MPI
705	<	Snapshot* snap = sman_->getCurrentSnapshot();
706	<	if (snap->atomData.getStorageLayout() & DataStorage::dslFunctional) {
707	<	AtomCommRealI->gather(snap->atomData.functional,
708	<	snap->atomIData.functional);
709	<	AtomCommRealJ->gather(snap->atomData.functional,
171	<	snap->atomJData.functional);
705	>	if (storageLayout_ & DataStorage::dslFunctional) {
706	>	AtomPlanRealRow->gather(snap_->atomData.functional,
707	>	atomRowData.functional);
708	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
709	>	atomColData.functional);
710		}
711
712	<	if (snap->atomData.getStorageLayout() & DataStorage::dslFunctionalDerivative) {
713	<	AtomCommRealI->gather(snap->atomData.functionalDerivative,
714	<	snap->atomIData.functionalDerivative);
715	<	AtomCommRealJ->gather(snap->atomData.functionalDerivative,
716	<	snap->atomJData.functionalDerivative);
712	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
713	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
714	>	atomRowData.functionalDerivative);
715	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
716	>	atomColData.functionalDerivative);
717		}
718		#endif
719		}
720
721
722	<	void ForceDecomposition::collectData() {
723	<	#ifdef IS_MPI
724	<	Snapshot* snap = sman_->getCurrentSnapshot();
725	<
726	<	int n = snap->atomData.force.size();
722	>	void ForceMatrixDecomposition::collectData() {
723	>	snap_ = sman_->getCurrentSnapshot();
724	>	storageLayout_ = sman_->getStorageLayout();
725	>	#ifdef IS_MPI
726	>	int n = snap_->atomData.force.size();
727		vector<Vector3d> frc_tmp(n, V3Zero);
728
729	<	AtomCommVectorI->scatter(snap->atomIData.force, frc_tmp);
729	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
730		for (int i = 0; i < n; i++) {
731	<	snap->atomData.force[i] += frc_tmp[i];
731	>	snap_->atomData.force[i] += frc_tmp[i];
732		frc_tmp[i] = 0.0;
733		}
734
735	<	AtomCommVectorJ->scatter(snap->atomJData.force, frc_tmp);
736	<	for (int i = 0; i < n; i++)
737	<	snap->atomData.force[i] += frc_tmp[i];
738	<
739	<
740	<	if (snap->atomData.getStorageLayout() & DataStorage::dslTorque) {
735	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
736	>	for (int i = 0; i < n; i++) {
737	>	snap_->atomData.force[i] += frc_tmp[i];
738	>	}
739	>
740	>	if (storageLayout_ & DataStorage::dslTorque) {
741
742	<	int nt = snap->atomData.force.size();
742	>	int nt = snap_->atomData.torque.size();
743		vector<Vector3d> trq_tmp(nt, V3Zero);
744
745	<	AtomCommVectorI->scatter(snap->atomIData.torque, trq_tmp);
746	<	for (int i = 0; i < n; i++) {
747	<	snap->atomData.torque[i] += trq_tmp[i];
745	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
746	>	for (int i = 0; i < nt; i++) {
747	>	snap_->atomData.torque[i] += trq_tmp[i];
748		trq_tmp[i] = 0.0;
749		}
750
751	<	AtomCommVectorJ->scatter(snap->atomJData.torque, trq_tmp);
752	<	for (int i = 0; i < n; i++)
753	<	snap->atomData.torque[i] += trq_tmp[i];
751	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
752	>	for (int i = 0; i < nt; i++)
753	>	snap_->atomData.torque[i] += trq_tmp[i];
754		}
755	+
756	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
757	+
758	+	int ns = snap_->atomData.skippedCharge.size();
759	+	vector<RealType> skch_tmp(ns, 0.0);
760	+
761	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
762	+	for (int i = 0; i < ns; i++) {
763	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
764	+	skch_tmp[i] = 0.0;
765	+	}
766	+
767	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
768	+	for (int i = 0; i < ns; i++)
769	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
770	+
771	+	}
772
773	<	int nLocal = snap->getNumberOfAtoms();
773	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
774
775	<	vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
776	<	vector<RealType> (nLocal, 0.0));
775	>	int nq = snap_->atomData.flucQFrc.size();
776	>	vector<RealType> fqfrc_tmp(nq, 0.0);
777	>
778	>	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
779	>	for (int i = 0; i < nq; i++) {
780	>	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
781	>	fqfrc_tmp[i] = 0.0;
782	>	}
783	>
784	>	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
785	>	for (int i = 0; i < nq; i++)
786	>	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
787	>
788	>	}
789	>
790	>	nLocal_ = snap_->getNumberOfAtoms();
791	>
792	>	vector<potVec> pot_temp(nLocal_,
793	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
794	>	vector<potVec> expot_temp(nLocal_,
795	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
796	>
797	>	// scatter/gather pot_row into the members of my column
798	>
799	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
800	>	AtomPlanPotRow->scatter(expot_row, expot_temp);
801	>
802	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
803	>	pairwisePot += pot_temp[ii];
804	>
805	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
806	>	excludedPot += expot_temp[ii];
807	>
808	>	if (storageLayout_ & DataStorage::dslParticlePot) {
809	>	// This is the pairwise contribution to the particle pot.  The
810	>	// embedding contribution is added in each of the low level
811	>	// non-bonded routines.  In single processor, this is done in
812	>	// unpackInteractionData, not in collectData.
813	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
814	>	for (int i = 0; i < nLocal_; i++) {
815	>	// factor of two is because the total potential terms are divided
816	>	// by 2 in parallel due to row/ column scatter
817	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
818	>	}
819	>	}
820	>	}
821	>
822	>	fill(pot_temp.begin(), pot_temp.end(),
823	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
824	>	fill(expot_temp.begin(), expot_temp.end(),
825	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
826	>
827	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
828	>	AtomPlanPotColumn->scatter(expot_col, expot_temp);
829
830	<	for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
831	<	AtomCommRealI->scatter(pot_row[i], pot_temp[i]);
832	<	for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
833	<	pot_local[i] += pot_temp[i][ii];
830	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
831	>	pairwisePot += pot_temp[ii];
832	>
833	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
834	>	excludedPot += expot_temp[ii];
835	>
836	>	if (storageLayout_ & DataStorage::dslParticlePot) {
837	>	// This is the pairwise contribution to the particle pot.  The
838	>	// embedding contribution is added in each of the low level
839	>	// non-bonded routines.  In single processor, this is done in
840	>	// unpackInteractionData, not in collectData.
841	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
842	>	for (int i = 0; i < nLocal_; i++) {
843	>	// factor of two is because the total potential terms are divided
844	>	// by 2 in parallel due to row/ column scatter
845	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
846	>	}
847		}
848		}
849	+
850	+	if (storageLayout_ & DataStorage::dslParticlePot) {
851	+	int npp = snap_->atomData.particlePot.size();
852	+	vector<RealType> ppot_temp(npp, 0.0);
853	+
854	+	// This is the direct or embedding contribution to the particle
855	+	// pot.
856	+
857	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
858	+	for (int i = 0; i < npp; i++) {
859	+	snap_->atomData.particlePot[i] += ppot_temp[i];
860	+	}
861	+
862	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
863	+
864	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
865	+	for (int i = 0; i < npp; i++) {
866	+	snap_->atomData.particlePot[i] += ppot_temp[i];
867	+	}
868	+	}
869	+
870	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
871	+	RealType ploc1 = pairwisePot[ii];
872	+	RealType ploc2 = 0.0;
873	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
874	+	pairwisePot[ii] = ploc2;
875	+	}
876	+
877	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
878	+	RealType ploc1 = excludedPot[ii];
879	+	RealType ploc2 = 0.0;
880	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
881	+	excludedPot[ii] = ploc2;
882	+	}
883	+
884	+	// Here be dragons.
885	+	MPI::Intracomm col = colComm.getComm();
886	+
887	+	col.Allreduce(MPI::IN_PLACE,
888	+	&snap_->frameData.conductiveHeatFlux[0], 3,
889	+	MPI::REALTYPE, MPI::SUM);
890	+
891	+
892		#endif
893	+
894		}
895	+
896	+	/**
897	+	* Collects information obtained during the post-pair (and embedding
898	+	* functional) loops onto local data structures.
899	+	*/
900	+	void ForceMatrixDecomposition::collectSelfData() {
901	+	snap_ = sman_->getCurrentSnapshot();
902	+	storageLayout_ = sman_->getStorageLayout();
903	+
904	+	#ifdef IS_MPI
905	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
906	+	RealType ploc1 = embeddingPot[ii];
907	+	RealType ploc2 = 0.0;
908	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
909	+	embeddingPot[ii] = ploc2;
910	+	}
911	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
912	+	RealType ploc1 = excludedSelfPot[ii];
913	+	RealType ploc2 = 0.0;
914	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
915	+	excludedSelfPot[ii] = ploc2;
916	+	}
917	+	#endif
918	+
919	+	}
920	+
921	+
922	+
923	+	int ForceMatrixDecomposition::getNAtomsInRow() {
924	+	#ifdef IS_MPI
925	+	return nAtomsInRow_;
926	+	#else
927	+	return nLocal_;
928	+	#endif
929	+	}
930	+
931	+	/**
932	+	* returns the list of atoms belonging to this group.
933	+	*/
934	+	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
935	+	#ifdef IS_MPI
936	+	return groupListRow_[cg1];
937	+	#else
938	+	return groupList_[cg1];
939	+	#endif
940	+	}
941	+
942	+	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
943	+	#ifdef IS_MPI
944	+	return groupListCol_[cg2];
945	+	#else
946	+	return groupList_[cg2];
947	+	#endif
948	+	}
949
950	+	Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
951	+	Vector3d d;
952	+
953	+	#ifdef IS_MPI
954	+	d = cgColData.position[cg2] - cgRowData.position[cg1];
955	+	#else
956	+	d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
957	+	#endif
958	+
959	+	snap_->wrapVector(d);
960	+	return d;
961	+	}
962	+
963	+	Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
964	+	#ifdef IS_MPI
965	+	return cgColData.velocity[cg2];
966	+	#else
967	+	return snap_->cgData.velocity[cg2];
968	+	#endif
969	+	}
970	+
971	+	Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
972	+	#ifdef IS_MPI
973	+	return atomColData.velocity[atom2];
974	+	#else
975	+	return snap_->atomData.velocity[atom2];
976	+	#endif
977	+	}
978	+
979	+
980	+	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
981	+
982	+	Vector3d d;
983	+
984	+	#ifdef IS_MPI
985	+	d = cgRowData.position[cg1] - atomRowData.position[atom1];
986	+	#else
987	+	d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
988	+	#endif
989	+
990	+	snap_->wrapVector(d);
991	+	return d;
992	+	}
993	+
994	+	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
995	+	Vector3d d;
996	+
997	+	#ifdef IS_MPI
998	+	d = cgColData.position[cg2] - atomColData.position[atom2];
999	+	#else
1000	+	d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1001	+	#endif
1002	+
1003	+	snap_->wrapVector(d);
1004	+	return d;
1005	+	}
1006	+
1007	+	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1008	+	#ifdef IS_MPI
1009	+	return massFactorsRow[atom1];
1010	+	#else
1011	+	return massFactors[atom1];
1012	+	#endif
1013	+	}
1014	+
1015	+	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1016	+	#ifdef IS_MPI
1017	+	return massFactorsCol[atom2];
1018	+	#else
1019	+	return massFactors[atom2];
1020	+	#endif
1021	+
1022	+	}
1023	+
1024	+	Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
1025	+	Vector3d d;
1026	+
1027	+	#ifdef IS_MPI
1028	+	d = atomColData.position[atom2] - atomRowData.position[atom1];
1029	+	#else
1030	+	d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1031	+	#endif
1032	+
1033	+	snap_->wrapVector(d);
1034	+	return d;
1035	+	}
1036	+
1037	+	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1038	+	return excludesForAtom[atom1];
1039	+	}
1040	+
1041	+	/**
1042	+	* We need to exclude some overcounted interactions that result from
1043	+	* the parallel decomposition.
1044	+	*/
1045	+	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1046	+	int unique_id_1, unique_id_2;
1047	+
1048	+	#ifdef IS_MPI
1049	+	// in MPI, we have to look up the unique IDs for each atom
1050	+	unique_id_1 = AtomRowToGlobal[atom1];
1051	+	unique_id_2 = AtomColToGlobal[atom2];
1052	+	// group1 = cgRowToGlobal[cg1];
1053	+	// group2 = cgColToGlobal[cg2];
1054	+	#else
1055	+	unique_id_1 = AtomLocalToGlobal[atom1];
1056	+	unique_id_2 = AtomLocalToGlobal[atom2];
1057	+	int group1 = cgLocalToGlobal[cg1];
1058	+	int group2 = cgLocalToGlobal[cg2];
1059	+	#endif
1060	+
1061	+	if (unique_id_1 == unique_id_2) return true;
1062	+
1063	+	#ifdef IS_MPI
1064	+	// this prevents us from doing the pair on multiple processors
1065	+	if (unique_id_1 < unique_id_2) {
1066	+	if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1067	+	} else {
1068	+	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1069	+	}
1070	+	#endif
1071	+
1072	+	#ifndef IS_MPI
1073	+	if (group1 == group2) {
1074	+	if (unique_id_1 < unique_id_2) return true;
1075	+	}
1076	+	#endif
1077	+
1078	+	return false;
1079	+	}
1080	+
1081	+	/**
1082	+	* We need to handle the interactions for atoms who are involved in
1083	+	* the same rigid body as well as some short range interactions
1084	+	* (bonds, bends, torsions) differently from other interactions.
1085	+	* We'll still visit the pairwise routines, but with a flag that
1086	+	* tells those routines to exclude the pair from direct long range
1087	+	* interactions.  Some indirect interactions (notably reaction
1088	+	* field) must still be handled for these pairs.
1089	+	*/
1090	+	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1091	+
1092	+	// excludesForAtom was constructed to use row/column indices in the MPI
1093	+	// version, and to use local IDs in the non-MPI version:
1094	+
1095	+	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1096	+	i != excludesForAtom[atom1].end(); ++i) {
1097	+	if ( (*i) == atom2 ) return true;
1098	+	}
1099	+
1100	+	return false;
1101	+	}
1102	+
1103	+
1104	+	void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
1105	+	#ifdef IS_MPI
1106	+	atomRowData.force[atom1] += fg;
1107	+	#else
1108	+	snap_->atomData.force[atom1] += fg;
1109	+	#endif
1110	+	}
1111	+
1112	+	void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
1113	+	#ifdef IS_MPI
1114	+	atomColData.force[atom2] += fg;
1115	+	#else
1116	+	snap_->atomData.force[atom2] += fg;
1117	+	#endif
1118	+	}
1119	+
1120	+	// filling interaction blocks with pointers
1121	+	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1122	+	int atom1, int atom2) {
1123	+
1124	+	idat.excluded = excludeAtomPair(atom1, atom2);
1125	+
1126	+	#ifdef IS_MPI
1127	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1128	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1129	+	//                         ff_->getAtomType(identsCol[atom2]) );
1130	+
1131	+	if (storageLayout_ & DataStorage::dslAmat) {
1132	+	idat.A1 = &(atomRowData.aMat[atom1]);
1133	+	idat.A2 = &(atomColData.aMat[atom2]);
1134	+	}
1135	+
1136	+	if (storageLayout_ & DataStorage::dslTorque) {
1137	+	idat.t1 = &(atomRowData.torque[atom1]);
1138	+	idat.t2 = &(atomColData.torque[atom2]);
1139	+	}
1140	+
1141	+	if (storageLayout_ & DataStorage::dslDipole) {
1142	+	idat.dipole1 = &(atomRowData.dipole[atom1]);
1143	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1144	+	}
1145	+
1146	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1147	+	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1148	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1149	+	}
1150	+
1151	+	if (storageLayout_ & DataStorage::dslDensity) {
1152	+	idat.rho1 = &(atomRowData.density[atom1]);
1153	+	idat.rho2 = &(atomColData.density[atom2]);
1154	+	}
1155	+
1156	+	if (storageLayout_ & DataStorage::dslFunctional) {
1157	+	idat.frho1 = &(atomRowData.functional[atom1]);
1158	+	idat.frho2 = &(atomColData.functional[atom2]);
1159	+	}
1160	+
1161	+	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1162	+	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1163	+	idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1164	+	}
1165	+
1166	+	if (storageLayout_ & DataStorage::dslParticlePot) {
1167	+	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1168	+	idat.particlePot2 = &(atomColData.particlePot[atom2]);
1169	+	}
1170	+
1171	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1172	+	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1173	+	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1174	+	}
1175	+
1176	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1177	+	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1178	+	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1179	+	}
1180	+
1181	+	#else
1182	+
1183	+	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1184	+
1185	+	if (storageLayout_ & DataStorage::dslAmat) {
1186	+	idat.A1 = &(snap_->atomData.aMat[atom1]);
1187	+	idat.A2 = &(snap_->atomData.aMat[atom2]);
1188	+	}
1189	+
1190	+	if (storageLayout_ & DataStorage::dslTorque) {
1191	+	idat.t1 = &(snap_->atomData.torque[atom1]);
1192	+	idat.t2 = &(snap_->atomData.torque[atom2]);
1193	+	}
1194	+
1195	+	if (storageLayout_ & DataStorage::dslDipole) {
1196	+	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1197	+	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1198	+	}
1199	+
1200	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1201	+	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1202	+	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1203	+	}
1204	+
1205	+	if (storageLayout_ & DataStorage::dslDensity) {
1206	+	idat.rho1 = &(snap_->atomData.density[atom1]);
1207	+	idat.rho2 = &(snap_->atomData.density[atom2]);
1208	+	}
1209	+
1210	+	if (storageLayout_ & DataStorage::dslFunctional) {
1211	+	idat.frho1 = &(snap_->atomData.functional[atom1]);
1212	+	idat.frho2 = &(snap_->atomData.functional[atom2]);
1213	+	}
1214	+
1215	+	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1216	+	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1217	+	idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1218	+	}
1219	+
1220	+	if (storageLayout_ & DataStorage::dslParticlePot) {
1221	+	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1222	+	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1223	+	}
1224	+
1225	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1226	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1227	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1228	+	}
1229	+
1230	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1231	+	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1232	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1233	+	}
1234	+
1235	+	#endif
1236	+	}
1237	+
1238	+
1239	+	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1240	+	#ifdef IS_MPI
1241	+	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1242	+	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1243	+	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1244	+	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1245	+
1246	+	atomRowData.force[atom1] += *(idat.f1);
1247	+	atomColData.force[atom2] -= *(idat.f1);
1248	+
1249	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1250	+	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1251	+	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1252	+	}
1253	+
1254	+	if (storageLayout_ & DataStorage::dslElectricField) {
1255	+	atomRowData.electricField[atom1] += *(idat.eField1);
1256	+	atomColData.electricField[atom2] += *(idat.eField2);
1257	+	}
1258	+
1259	+	#else
1260	+	pairwisePot += *(idat.pot);
1261	+	excludedPot += *(idat.excludedPot);
1262	+
1263	+	snap_->atomData.force[atom1] += *(idat.f1);
1264	+	snap_->atomData.force[atom2] -= *(idat.f1);
1265	+
1266	+	if (idat.doParticlePot) {
1267	+	// This is the pairwise contribution to the particle pot.  The
1268	+	// embedding contribution is added in each of the low level
1269	+	// non-bonded routines.  In parallel, this calculation is done
1270	+	// in collectData, not in unpackInteractionData.
1271	+	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1272	+	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1273	+	}
1274	+
1275	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1276	+	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1277	+	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1278	+	}
1279	+
1280	+	if (storageLayout_ & DataStorage::dslElectricField) {
1281	+	snap_->atomData.electricField[atom1] += *(idat.eField1);
1282	+	snap_->atomData.electricField[atom2] += *(idat.eField2);
1283	+	}
1284	+
1285	+	#endif
1286	+
1287	+	}
1288	+
1289	+	/*
1290	+	* buildNeighborList
1291	+	*
1292	+	* first element of pair is row-indexed CutoffGroup
1293	+	* second element of pair is column-indexed CutoffGroup
1294	+	*/
1295	+	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1296	+
1297	+	vector<pair<int, int> > neighborList;
1298	+	groupCutoffs cuts;
1299	+	bool doAllPairs = false;
1300	+
1301	+	#ifdef IS_MPI
1302	+	cellListRow_.clear();
1303	+	cellListCol_.clear();
1304	+	#else
1305	+	cellList_.clear();
1306	+	#endif
1307	+
1308	+	RealType rList_ = (largestRcut_ + skinThickness_);
1309	+	RealType rl2 = rList_ * rList_;
1310	+	Snapshot* snap_ = sman_->getCurrentSnapshot();
1311	+	Mat3x3d Hmat = snap_->getHmat();
1312	+	Vector3d Hx = Hmat.getColumn(0);
1313	+	Vector3d Hy = Hmat.getColumn(1);
1314	+	Vector3d Hz = Hmat.getColumn(2);
1315	+
1316	+	nCells_.x() = (int) ( Hx.length() )/ rList_;
1317	+	nCells_.y() = (int) ( Hy.length() )/ rList_;
1318	+	nCells_.z() = (int) ( Hz.length() )/ rList_;
1319	+
1320	+	// handle small boxes where the cell offsets can end up repeating cells
1321	+
1322	+	if (nCells_.x() < 3) doAllPairs = true;
1323	+	if (nCells_.y() < 3) doAllPairs = true;
1324	+	if (nCells_.z() < 3) doAllPairs = true;
1325	+
1326	+	Mat3x3d invHmat = snap_->getInvHmat();
1327	+	Vector3d rs, scaled, dr;
1328	+	Vector3i whichCell;
1329	+	int cellIndex;
1330	+	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1331	+
1332	+	#ifdef IS_MPI
1333	+	cellListRow_.resize(nCtot);
1334	+	cellListCol_.resize(nCtot);
1335	+	#else
1336	+	cellList_.resize(nCtot);
1337	+	#endif
1338	+
1339	+	if (!doAllPairs) {
1340	+	#ifdef IS_MPI
1341	+
1342	+	for (int i = 0; i < nGroupsInRow_; i++) {
1343	+	rs = cgRowData.position[i];
1344	+
1345	+	// scaled positions relative to the box vectors
1346	+	scaled = invHmat * rs;
1347	+
1348	+	// wrap the vector back into the unit box by subtracting integer box
1349	+	// numbers
1350	+	for (int j = 0; j < 3; j++) {
1351	+	scaled[j] -= roundMe(scaled[j]);
1352	+	scaled[j] += 0.5;
1353	+	// Handle the special case when an object is exactly on the
1354	+	// boundary (a scaled coordinate of 1.0 is the same as
1355	+	// scaled coordinate of 0.0)
1356	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1357	+	}
1358	+
1359	+	// find xyz-indices of cell that cutoffGroup is in.
1360	+	whichCell.x() = nCells_.x() * scaled.x();
1361	+	whichCell.y() = nCells_.y() * scaled.y();
1362	+	whichCell.z() = nCells_.z() * scaled.z();
1363	+
1364	+	// find single index of this cell:
1365	+	cellIndex = Vlinear(whichCell, nCells_);
1366	+
1367	+	// add this cutoff group to the list of groups in this cell;
1368	+	cellListRow_[cellIndex].push_back(i);
1369	+	}
1370	+	for (int i = 0; i < nGroupsInCol_; i++) {
1371	+	rs = cgColData.position[i];
1372	+
1373	+	// scaled positions relative to the box vectors
1374	+	scaled = invHmat * rs;
1375	+
1376	+	// wrap the vector back into the unit box by subtracting integer box
1377	+	// numbers
1378	+	for (int j = 0; j < 3; j++) {
1379	+	scaled[j] -= roundMe(scaled[j]);
1380	+	scaled[j] += 0.5;
1381	+	// Handle the special case when an object is exactly on the
1382	+	// boundary (a scaled coordinate of 1.0 is the same as
1383	+	// scaled coordinate of 0.0)
1384	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1385	+	}
1386	+
1387	+	// find xyz-indices of cell that cutoffGroup is in.
1388	+	whichCell.x() = nCells_.x() * scaled.x();
1389	+	whichCell.y() = nCells_.y() * scaled.y();
1390	+	whichCell.z() = nCells_.z() * scaled.z();
1391	+
1392	+	// find single index of this cell:
1393	+	cellIndex = Vlinear(whichCell, nCells_);
1394	+
1395	+	// add this cutoff group to the list of groups in this cell;
1396	+	cellListCol_[cellIndex].push_back(i);
1397	+	}
1398	+
1399	+	#else
1400	+	for (int i = 0; i < nGroups_; i++) {
1401	+	rs = snap_->cgData.position[i];
1402	+
1403	+	// scaled positions relative to the box vectors
1404	+	scaled = invHmat * rs;
1405	+
1406	+	// wrap the vector back into the unit box by subtracting integer box
1407	+	// numbers
1408	+	for (int j = 0; j < 3; j++) {
1409	+	scaled[j] -= roundMe(scaled[j]);
1410	+	scaled[j] += 0.5;
1411	+	// Handle the special case when an object is exactly on the
1412	+	// boundary (a scaled coordinate of 1.0 is the same as
1413	+	// scaled coordinate of 0.0)
1414	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1415	+	}
1416	+
1417	+	// find xyz-indices of cell that cutoffGroup is in.
1418	+	whichCell.x() = nCells_.x() * scaled.x();
1419	+	whichCell.y() = nCells_.y() * scaled.y();
1420	+	whichCell.z() = nCells_.z() * scaled.z();
1421	+
1422	+	// find single index of this cell:
1423	+	cellIndex = Vlinear(whichCell, nCells_);
1424	+
1425	+	// add this cutoff group to the list of groups in this cell;
1426	+	cellList_[cellIndex].push_back(i);
1427	+	}
1428	+
1429	+	#endif
1430	+
1431	+	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1432	+	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1433	+	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1434	+	Vector3i m1v(m1x, m1y, m1z);
1435	+	int m1 = Vlinear(m1v, nCells_);
1436	+
1437	+	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1438	+	os != cellOffsets_.end(); ++os) {
1439	+
1440	+	Vector3i m2v = m1v + (*os);
1441	+
1442	+
1443	+	if (m2v.x() >= nCells_.x()) {
1444	+	m2v.x() = 0;
1445	+	} else if (m2v.x() < 0) {
1446	+	m2v.x() = nCells_.x() - 1;
1447	+	}
1448	+
1449	+	if (m2v.y() >= nCells_.y()) {
1450	+	m2v.y() = 0;
1451	+	} else if (m2v.y() < 0) {
1452	+	m2v.y() = nCells_.y() - 1;
1453	+	}
1454	+
1455	+	if (m2v.z() >= nCells_.z()) {
1456	+	m2v.z() = 0;
1457	+	} else if (m2v.z() < 0) {
1458	+	m2v.z() = nCells_.z() - 1;
1459	+	}
1460	+
1461	+	int m2 = Vlinear (m2v, nCells_);
1462	+
1463	+	#ifdef IS_MPI
1464	+	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1465	+	j1 != cellListRow_[m1].end(); ++j1) {
1466	+	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1467	+	j2 != cellListCol_[m2].end(); ++j2) {
1468	+
1469	+	// In parallel, we need to visit *all* pairs of row
1470	+	// & column indicies and will divide labor in the
1471	+	// force evaluation later.
1472	+	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1473	+	snap_->wrapVector(dr);
1474	+	cuts = getGroupCutoffs( (*j1), (*j2) );
1475	+	if (dr.lengthSquare() < cuts.third) {
1476	+	neighborList.push_back(make_pair((*j1), (*j2)));
1477	+	}
1478	+	}
1479	+	}
1480	+	#else
1481	+	for (vector<int>::iterator j1 = cellList_[m1].begin();
1482	+	j1 != cellList_[m1].end(); ++j1) {
1483	+	for (vector<int>::iterator j2 = cellList_[m2].begin();
1484	+	j2 != cellList_[m2].end(); ++j2) {
1485	+
1486	+	// Always do this if we're in different cells or if
1487	+	// we're in the same cell and the global index of
1488	+	// the j2 cutoff group is greater than or equal to
1489	+	// the j1 cutoff group.  Note that Rappaport's code
1490	+	// has a "less than" conditional here, but that
1491	+	// deals with atom-by-atom computation.  OpenMD
1492	+	// allows atoms within a single cutoff group to
1493	+	// interact with each other.
1494	+
1495	+
1496	+
1497	+	if (m2 != m1 || (*j2) >= (*j1) ) {
1498	+
1499	+	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1500	+	snap_->wrapVector(dr);
1501	+	cuts = getGroupCutoffs( (*j1), (*j2) );
1502	+	if (dr.lengthSquare() < cuts.third) {
1503	+	neighborList.push_back(make_pair((*j1), (*j2)));
1504	+	}
1505	+	}
1506	+	}
1507	+	}
1508	+	#endif
1509	+	}
1510	+	}
1511	+	}
1512	+	}
1513	+	} else {
1514	+	// branch to do all cutoff group pairs
1515	+	#ifdef IS_MPI
1516	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1517	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1518	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1519	+	snap_->wrapVector(dr);
1520	+	cuts = getGroupCutoffs( j1, j2 );
1521	+	if (dr.lengthSquare() < cuts.third) {
1522	+	neighborList.push_back(make_pair(j1, j2));
1523	+	}
1524	+	}
1525	+	}
1526	+	#else
1527	+	// include all groups here.
1528	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1529	+	// include self group interactions j2 == j1
1530	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1531	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1532	+	snap_->wrapVector(dr);
1533	+	cuts = getGroupCutoffs( j1, j2 );
1534	+	if (dr.lengthSquare() < cuts.third) {
1535	+	neighborList.push_back(make_pair(j1, j2));
1536	+	}
1537	+	}
1538	+	}
1539	+	#endif
1540	+	}
1541	+
1542	+	// save the local cutoff group positions for the check that is
1543	+	// done on each loop:
1544	+	saved_CG_positions_.clear();
1545	+	for (int i = 0; i < nGroups_; i++)
1546	+	saved_CG_positions_.push_back(snap_->cgData.position[i]);
1547	+
1548	+	return neighborList;
1549	+	}
1550		} //end namespace OpenMD

Diff Legend

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