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 1570 by gezelter, Thu May 26 21:56:04 2011 UTC vs.
Revision 1616 by gezelter, Tue Aug 30 15:45:35 2011 UTC

#	Line 47 | Line 47 | namespace OpenMD {
47		using namespace std;
48		namespace OpenMD {
49
50	+	ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
51	+
52	+	// In a parallel computation, row and colum scans must visit all
53	+	// surrounding cells (not just the 14 upper triangular blocks that
54	+	// are used when the processor can see all pairs)
55	+	#ifdef IS_MPI
56	+	cellOffsets_.clear();
57	+	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
58	+	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
59	+	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
60	+	cellOffsets_.push_back( Vector3i(-1, 0,-1) );
61	+	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
62	+	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
63	+	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
64	+	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
65	+	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
66	+	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
67	+	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
68	+	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
69	+	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
70	+	cellOffsets_.push_back( Vector3i( 0, 0, 0) );
71	+	cellOffsets_.push_back( Vector3i( 1, 0, 0) );
72	+	cellOffsets_.push_back( Vector3i(-1, 1, 0) );
73	+	cellOffsets_.push_back( Vector3i( 0, 1, 0) );
74	+	cellOffsets_.push_back( Vector3i( 1, 1, 0) );
75	+	cellOffsets_.push_back( Vector3i(-1,-1, 1) );
76	+	cellOffsets_.push_back( Vector3i( 0,-1, 1) );
77	+	cellOffsets_.push_back( Vector3i( 1,-1, 1) );
78	+	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
79	+	cellOffsets_.push_back( Vector3i( 0, 0, 1) );
80	+	cellOffsets_.push_back( Vector3i( 1, 0, 1) );
81	+	cellOffsets_.push_back( Vector3i(-1, 1, 1) );
82	+	cellOffsets_.push_back( Vector3i( 0, 1, 1) );
83	+	cellOffsets_.push_back( Vector3i( 1, 1, 1) );
84	+	#endif
85	+	}
86	+
87	+
88		/**
89		* distributeInitialData is essentially a copy of the older fortran
90		* SimulationSetup
91		*/
54	–
92		void ForceMatrixDecomposition::distributeInitialData() {
93		snap_ = sman_->getCurrentSnapshot();
94		storageLayout_ = sman_->getStorageLayout();
95	+	ff_ = info_->getForceField();
96		nLocal_ = snap_->getNumberOfAtoms();
97	<	nGroups_ = snap_->getNumberOfCutoffGroups();
98	<
97	>
98	>	nGroups_ = info_->getNLocalCutoffGroups();
99		// gather the information for atomtype IDs (atids):
100	<	vector<int> identsLocal = info_->getIdentArray();
100	>	idents = info_->getIdentArray();
101		AtomLocalToGlobal = info_->getGlobalAtomIndices();
102		cgLocalToGlobal = info_->getGlobalGroupIndices();
103		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
66	–	vector<RealType> massFactorsLocal = info_->getMassFactors();
67	–	PairList excludes = info_->getExcludedInteractions();
68	–	PairList oneTwo = info_->getOneTwoInteractions();
69	–	PairList oneThree = info_->getOneThreeInteractions();
70	–	PairList oneFour = info_->getOneFourInteractions();
71	–	vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
104
105	+	massFactors = info_->getMassFactors();
106	+
107	+	PairList* excludes = info_->getExcludedInteractions();
108	+	PairList* oneTwo = info_->getOneTwoInteractions();
109	+	PairList* oneThree = info_->getOneThreeInteractions();
110	+	PairList* oneFour = info_->getOneFourInteractions();
111	+
112		#ifdef IS_MPI
113
114	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
115	<	AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
77	<	AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
78	<	AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
114	>	MPI::Intracomm row = rowComm.getComm();
115	>	MPI::Intracomm col = colComm.getComm();
116
117	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
118	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
119	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
120	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
117	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
118	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
119	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
120	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
121	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
122
123	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
124	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
125	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
126	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
123	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
124	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
125	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
126	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
127	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
128
129	<	nAtomsInRow_ = AtomCommIntRow->getSize();
130	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
131	<	nGroupsInRow_ = cgCommIntRow->getSize();
132	<	nGroupsInCol_ = cgCommIntColumn->getSize();
129	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
130	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
131	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
132	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
133
134	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
135	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
136	+	nGroupsInRow_ = cgPlanIntRow->getSize();
137	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
138	+
139		// Modify the data storage objects with the correct layouts and sizes:
140		atomRowData.resize(nAtomsInRow_);
141		atomRowData.setStorageLayout(storageLayout_);
#	Line 101 | Line 145 | namespace OpenMD {
145		cgRowData.setStorageLayout(DataStorage::dslPosition);
146		cgColData.resize(nGroupsInCol_);
147		cgColData.setStorageLayout(DataStorage::dslPosition);
148	+
149	+	identsRow.resize(nAtomsInRow_);
150	+	identsCol.resize(nAtomsInCol_);
151
152	<	vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
153	<	vector<RealType> (nAtomsInRow_, 0.0));
107	<	vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
108	<	vector<RealType> (nAtomsInCol_, 0.0));
152	>	AtomPlanIntRow->gather(idents, identsRow);
153	>	AtomPlanIntColumn->gather(idents, identsCol);
154
155	<	identsRow.reserve(nAtomsInRow_);
156	<	identsCol.reserve(nAtomsInCol_);
157	<
113	<	AtomCommIntRow->gather(identsLocal, identsRow);
114	<	AtomCommIntColumn->gather(identsLocal, identsCol);
115	<
116	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
117	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
118	<
119	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
120	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
155	>	// allocate memory for the parallel objects
156	>	atypesRow.resize(nAtomsInRow_);
157	>	atypesCol.resize(nAtomsInCol_);
158
159	<	AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
160	<	AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
159	>	for (int i = 0; i < nAtomsInRow_; i++)
160	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
161	>	for (int i = 0; i < nAtomsInCol_; i++)
162	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
163
164	+	pot_row.resize(nAtomsInRow_);
165	+	pot_col.resize(nAtomsInCol_);
166	+
167	+	AtomRowToGlobal.resize(nAtomsInRow_);
168	+	AtomColToGlobal.resize(nAtomsInCol_);
169	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
170	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
171	+
172	+	cgRowToGlobal.resize(nGroupsInRow_);
173	+	cgColToGlobal.resize(nGroupsInCol_);
174	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
175	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
176	+
177	+	massFactorsRow.resize(nAtomsInRow_);
178	+	massFactorsCol.resize(nAtomsInCol_);
179	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
180	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
181	+
182		groupListRow_.clear();
183	<	groupListRow_.reserve(nGroupsInRow_);
183	>	groupListRow_.resize(nGroupsInRow_);
184		for (int i = 0; i < nGroupsInRow_; i++) {
185		int gid = cgRowToGlobal[i];
186		for (int j = 0; j < nAtomsInRow_; j++) {
#	Line 134 | Line 191 | namespace OpenMD {
191		}
192
193		groupListCol_.clear();
194	<	groupListCol_.reserve(nGroupsInCol_);
194	>	groupListCol_.resize(nGroupsInCol_);
195		for (int i = 0; i < nGroupsInCol_; i++) {
196		int gid = cgColToGlobal[i];
197		for (int j = 0; j < nAtomsInCol_; j++) {
#	Line 144 | Line 201 | namespace OpenMD {
201		}
202		}
203
204	<	skipsForRowAtom.clear();
205	<	skipsForRowAtom.reserve(nAtomsInRow_);
206	<	for (int i = 0; i < nAtomsInRow_; i++) {
207	<	int iglob = AtomColToGlobal[i];
204	>	excludesForAtom.clear();
205	>	excludesForAtom.resize(nAtomsInRow_);
206	>	toposForAtom.clear();
207	>	toposForAtom.resize(nAtomsInRow_);
208	>	topoDist.clear();
209	>	topoDist.resize(nAtomsInRow_);
210	>	for (int i = 0; i < nAtomsInRow_; i++) {
211	>	int iglob = AtomRowToGlobal[i];
212	>
213		for (int j = 0; j < nAtomsInCol_; j++) {
214	<	int jglob = AtomRowToGlobal[j];
215	<	if (excludes.hasPair(iglob, jglob))
216	<	skipsForRowAtom[i].push_back(j);
214	>	int jglob = AtomColToGlobal[j];
215	>
216	>	if (excludes->hasPair(iglob, jglob))
217	>	excludesForAtom[i].push_back(j);
218	>
219	>	if (oneTwo->hasPair(iglob, jglob)) {
220	>	toposForAtom[i].push_back(j);
221	>	topoDist[i].push_back(1);
222	>	} else {
223	>	if (oneThree->hasPair(iglob, jglob)) {
224	>	toposForAtom[i].push_back(j);
225	>	topoDist[i].push_back(2);
226	>	} else {
227	>	if (oneFour->hasPair(iglob, jglob)) {
228	>	toposForAtom[i].push_back(j);
229	>	topoDist[i].push_back(3);
230	>	}
231	>	}
232	>	}
233		}
234		}
235
236	<	toposForRowAtom.clear();
237	<	toposForRowAtom.reserve(nAtomsInRow_);
238	<	for (int i = 0; i < nAtomsInRow_; i++) {
239	<	int iglob = AtomColToGlobal[i];
240	<	int nTopos = 0;
241	<	for (int j = 0; j < nAtomsInCol_; j++) {
242	<	int jglob = AtomRowToGlobal[j];
243	<	if (oneTwo.hasPair(iglob, jglob)) {
244	<	toposForRowAtom[i].push_back(j);
245	<	topoDistRow[i][nTopos] = 1;
246	<	nTopos++;
236	>	#else
237	>	excludesForAtom.clear();
238	>	excludesForAtom.resize(nLocal_);
239	>	toposForAtom.clear();
240	>	toposForAtom.resize(nLocal_);
241	>	topoDist.clear();
242	>	topoDist.resize(nLocal_);
243	>
244	>	for (int i = 0; i < nLocal_; i++) {
245	>	int iglob = AtomLocalToGlobal[i];
246	>
247	>	for (int j = 0; j < nLocal_; j++) {
248	>	int jglob = AtomLocalToGlobal[j];
249	>
250	>	if (excludes->hasPair(iglob, jglob))
251	>	excludesForAtom[i].push_back(j);
252	>
253	>	if (oneTwo->hasPair(iglob, jglob)) {
254	>	toposForAtom[i].push_back(j);
255	>	topoDist[i].push_back(1);
256	>	} else {
257	>	if (oneThree->hasPair(iglob, jglob)) {
258	>	toposForAtom[i].push_back(j);
259	>	topoDist[i].push_back(2);
260	>	} else {
261	>	if (oneFour->hasPair(iglob, jglob)) {
262	>	toposForAtom[i].push_back(j);
263	>	topoDist[i].push_back(3);
264	>	}
265	>	}
266		}
170	–	if (oneThree.hasPair(iglob, jglob)) {
171	–	toposForRowAtom[i].push_back(j);
172	–	topoDistRow[i][nTopos] = 2;
173	–	nTopos++;
174	–	}
175	–	if (oneFour.hasPair(iglob, jglob)) {
176	–	toposForRowAtom[i].push_back(j);
177	–	topoDistRow[i][nTopos] = 3;
178	–	nTopos++;
179	–	}
267		}
268		}
182	–
269		#endif
270
271	+	// allocate memory for the parallel objects
272	+	atypesLocal.resize(nLocal_);
273	+
274	+	for (int i = 0; i < nLocal_; i++)
275	+	atypesLocal[i] = ff_->getAtomType(idents[i]);
276	+
277		groupList_.clear();
278	<	groupList_.reserve(nGroups_);
278	>	groupList_.resize(nGroups_);
279		for (int i = 0; i < nGroups_; i++) {
280		int gid = cgLocalToGlobal[i];
281		for (int j = 0; j < nLocal_; j++) {
282		int aid = AtomLocalToGlobal[j];
283	<	if (globalGroupMembership[aid] == gid)
283	>	if (globalGroupMembership[aid] == gid) {
284		groupList_[i].push_back(j);
285	+	}
286		}
287		}
288
196	–	skipsForLocalAtom.clear();
197	–	skipsForLocalAtom.reserve(nLocal_);
289
290	<	for (int i = 0; i < nLocal_; i++) {
200	<	int iglob = AtomLocalToGlobal[i];
201	<	for (int j = 0; j < nLocal_; j++) {
202	<	int jglob = AtomLocalToGlobal[j];
203	<	if (excludes.hasPair(iglob, jglob))
204	<	skipsForLocalAtom[i].push_back(j);
205	<	}
206	<	}
290	>	createGtypeCutoffMap();
291
292	<	toposForLocalAtom.clear();
293	<	toposForLocalAtom.reserve(nLocal_);
294	<	for (int i = 0; i < nLocal_; i++) {
295	<	int iglob = AtomLocalToGlobal[i];
296	<	int nTopos = 0;
297	<	for (int j = 0; j < nLocal_; j++) {
298	<	int jglob = AtomLocalToGlobal[j];
299	<	if (oneTwo.hasPair(iglob, jglob)) {
300	<	toposForLocalAtom[i].push_back(j);
301	<	topoDistLocal[i][nTopos] = 1;
302	<	nTopos++;
303	<	}
304	<	if (oneThree.hasPair(iglob, jglob)) {
305	<	toposForLocalAtom[i].push_back(j);
306	<	topoDistLocal[i][nTopos] = 2;
307	<	nTopos++;
308	<	}
309	<	if (oneFour.hasPair(iglob, jglob)) {
310	<	toposForLocalAtom[i].push_back(j);
311	<	topoDistLocal[i][nTopos] = 3;
312	<	nTopos++;
292	>	}
293	>
294	>	void ForceMatrixDecomposition::createGtypeCutoffMap() {
295	>
296	>	RealType tol = 1e-6;
297	>	largestRcut_ = 0.0;
298	>	RealType rc;
299	>	int atid;
300	>	set<AtomType*> atypes = info_->getSimulatedAtomTypes();
301	>
302	>	map<int, RealType> atypeCutoff;
303	>
304	>	for (set<AtomType*>::iterator at = atypes.begin();
305	>	at != atypes.end(); ++at){
306	>	atid = (*at)->getIdent();
307	>	if (userChoseCutoff_)
308	>	atypeCutoff[atid] = userCutoff_;
309	>	else
310	>	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
311	>	}
312	>
313	>	vector<RealType> gTypeCutoffs;
314	>	// first we do a single loop over the cutoff groups to find the
315	>	// largest cutoff for any atypes present in this group.
316	>	#ifdef IS_MPI
317	>	vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
318	>	groupRowToGtype.resize(nGroupsInRow_);
319	>	for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
320	>	vector<int> atomListRow = getAtomsInGroupRow(cg1);
321	>	for (vector<int>::iterator ia = atomListRow.begin();
322	>	ia != atomListRow.end(); ++ia) {
323	>	int atom1 = (*ia);
324	>	atid = identsRow[atom1];
325	>	if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
326	>	groupCutoffRow[cg1] = atypeCutoff[atid];
327		}
328	+	}
329	+
330	+	bool gTypeFound = false;
331	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
332	+	if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
333	+	groupRowToGtype[cg1] = gt;
334	+	gTypeFound = true;
335	+	}
336	+	}
337	+	if (!gTypeFound) {
338	+	gTypeCutoffs.push_back( groupCutoffRow[cg1] );
339	+	groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
340	+	}
341	+
342	+	}
343	+	vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
344	+	groupColToGtype.resize(nGroupsInCol_);
345	+	for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
346	+	vector<int> atomListCol = getAtomsInGroupColumn(cg2);
347	+	for (vector<int>::iterator jb = atomListCol.begin();
348	+	jb != atomListCol.end(); ++jb) {
349	+	int atom2 = (*jb);
350	+	atid = identsCol[atom2];
351	+	if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
352	+	groupCutoffCol[cg2] = atypeCutoff[atid];
353	+	}
354	+	}
355	+	bool gTypeFound = false;
356	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
357	+	if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
358	+	groupColToGtype[cg2] = gt;
359	+	gTypeFound = true;
360	+	}
361	+	}
362	+	if (!gTypeFound) {
363	+	gTypeCutoffs.push_back( groupCutoffCol[cg2] );
364	+	groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
365	+	}
366	+	}
367	+	#else
368	+
369	+	vector<RealType> groupCutoff(nGroups_, 0.0);
370	+	groupToGtype.resize(nGroups_);
371	+	for (int cg1 = 0; cg1 < nGroups_; cg1++) {
372	+	groupCutoff[cg1] = 0.0;
373	+	vector<int> atomList = getAtomsInGroupRow(cg1);
374	+	for (vector<int>::iterator ia = atomList.begin();
375	+	ia != atomList.end(); ++ia) {
376	+	int atom1 = (*ia);
377	+	atid = idents[atom1];
378	+	if (atypeCutoff[atid] > groupCutoff[cg1])
379	+	groupCutoff[cg1] = atypeCutoff[atid];
380	+	}
381	+
382	+	bool gTypeFound = false;
383	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
384	+	if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
385	+	groupToGtype[cg1] = gt;
386	+	gTypeFound = true;
387	+	}
388	+	}
389	+	if (!gTypeFound) {
390	+	gTypeCutoffs.push_back( groupCutoff[cg1] );
391	+	groupToGtype[cg1] = gTypeCutoffs.size() - 1;
392		}
393		}
394	+	#endif
395	+
396	+	// Now we find the maximum group cutoff value present in the simulation
397	+
398	+	RealType groupMax = *max_element(gTypeCutoffs.begin(),
399	+	gTypeCutoffs.end());
400	+
401	+	#ifdef IS_MPI
402	+	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
403	+	MPI::MAX);
404	+	#endif
405	+
406	+	RealType tradRcut = groupMax;
407	+
408	+	for (int i = 0; i < gTypeCutoffs.size();  i++) {
409	+	for (int j = 0; j < gTypeCutoffs.size();  j++) {
410	+	RealType thisRcut;
411	+	switch(cutoffPolicy_) {
412	+	case TRADITIONAL:
413	+	thisRcut = tradRcut;
414	+	break;
415	+	case MIX:
416	+	thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
417	+	break;
418	+	case MAX:
419	+	thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
420	+	break;
421	+	default:
422	+	sprintf(painCave.errMsg,
423	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
424	+	"hit an unknown cutoff policy!\n");
425	+	painCave.severity = OPENMD_ERROR;
426	+	painCave.isFatal = 1;
427	+	simError();
428	+	break;
429	+	}
430	+
431	+	pair<int,int> key = make_pair(i,j);
432	+	gTypeCutoffMap[key].first = thisRcut;
433	+	if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
434	+	gTypeCutoffMap[key].second = thisRcut*thisRcut;
435	+	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
436	+	// sanity check
437	+
438	+	if (userChoseCutoff_) {
439	+	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
440	+	sprintf(painCave.errMsg,
441	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
442	+	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
443	+	painCave.severity = OPENMD_ERROR;
444	+	painCave.isFatal = 1;
445	+	simError();
446	+	}
447	+	}
448	+	}
449	+	}
450		}
451	<
451	>
452	>
453	>	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
454	>	int i, j;
455	>	#ifdef IS_MPI
456	>	i = groupRowToGtype[cg1];
457	>	j = groupColToGtype[cg2];
458	>	#else
459	>	i = groupToGtype[cg1];
460	>	j = groupToGtype[cg2];
461	>	#endif
462	>	return gTypeCutoffMap[make_pair(i,j)];
463	>	}
464	>
465	>	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
466	>	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
467	>	if (toposForAtom[atom1][j] == atom2)
468	>	return topoDist[atom1][j];
469	>	}
470	>	return 0;
471	>	}
472	>
473	>	void ForceMatrixDecomposition::zeroWorkArrays() {
474	>	pairwisePot = 0.0;
475	>	embeddingPot = 0.0;
476	>
477	>	#ifdef IS_MPI
478	>	if (storageLayout_ & DataStorage::dslForce) {
479	>	fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
480	>	fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
481	>	}
482	>
483	>	if (storageLayout_ & DataStorage::dslTorque) {
484	>	fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
485	>	fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
486	>	}
487	>
488	>	fill(pot_row.begin(), pot_row.end(),
489	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
490	>
491	>	fill(pot_col.begin(), pot_col.end(),
492	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
493	>
494	>	if (storageLayout_ & DataStorage::dslParticlePot) {
495	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
496	>	0.0);
497	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
498	>	0.0);
499	>	}
500	>
501	>	if (storageLayout_ & DataStorage::dslDensity) {
502	>	fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
503	>	fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
504	>	}
505	>
506	>	if (storageLayout_ & DataStorage::dslFunctional) {
507	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
508	>	0.0);
509	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
510	>	0.0);
511	>	}
512	>
513	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
514	>	fill(atomRowData.functionalDerivative.begin(),
515	>	atomRowData.functionalDerivative.end(), 0.0);
516	>	fill(atomColData.functionalDerivative.begin(),
517	>	atomColData.functionalDerivative.end(), 0.0);
518	>	}
519	>
520	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
521	>	fill(atomRowData.skippedCharge.begin(),
522	>	atomRowData.skippedCharge.end(), 0.0);
523	>	fill(atomColData.skippedCharge.begin(),
524	>	atomColData.skippedCharge.end(), 0.0);
525	>	}
526	>
527	>	#endif
528	>	// even in parallel, we need to zero out the local arrays:
529	>
530	>	if (storageLayout_ & DataStorage::dslParticlePot) {
531	>	fill(snap_->atomData.particlePot.begin(),
532	>	snap_->atomData.particlePot.end(), 0.0);
533	>	}
534	>
535	>	if (storageLayout_ & DataStorage::dslDensity) {
536	>	fill(snap_->atomData.density.begin(),
537	>	snap_->atomData.density.end(), 0.0);
538	>	}
539	>	if (storageLayout_ & DataStorage::dslFunctional) {
540	>	fill(snap_->atomData.functional.begin(),
541	>	snap_->atomData.functional.end(), 0.0);
542	>	}
543	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
544	>	fill(snap_->atomData.functionalDerivative.begin(),
545	>	snap_->atomData.functionalDerivative.end(), 0.0);
546	>	}
547	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
548	>	fill(snap_->atomData.skippedCharge.begin(),
549	>	snap_->atomData.skippedCharge.end(), 0.0);
550	>	}
551	>
552	>	}
553	>
554	>
555		void ForceMatrixDecomposition::distributeData()  {
556		snap_ = sman_->getCurrentSnapshot();
557		storageLayout_ = sman_->getStorageLayout();
558		#ifdef IS_MPI
559
560		// gather up the atomic positions
561	<	AtomCommVectorRow->gather(snap_->atomData.position,
561	>	AtomPlanVectorRow->gather(snap_->atomData.position,
562		atomRowData.position);
563	<	AtomCommVectorColumn->gather(snap_->atomData.position,
563	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
564		atomColData.position);
565
566		// gather up the cutoff group positions
567	<	cgCommVectorRow->gather(snap_->cgData.position,
567	>
568	>	cgPlanVectorRow->gather(snap_->cgData.position,
569		cgRowData.position);
570	<	cgCommVectorColumn->gather(snap_->cgData.position,
570	>
571	>	cgPlanVectorColumn->gather(snap_->cgData.position,
572		cgColData.position);
573	+
574
575		// if needed, gather the atomic rotation matrices
576		if (storageLayout_ & DataStorage::dslAmat) {
577	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
577	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
578		atomRowData.aMat);
579	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
579	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
580		atomColData.aMat);
581		}
582
583		// if needed, gather the atomic eletrostatic frames
584		if (storageLayout_ & DataStorage::dslElectroFrame) {
585	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
585	>	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
586		atomRowData.electroFrame);
587	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
587	>	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
588		atomColData.electroFrame);
589		}
590	+
591		#endif
592		}
593
594	+	/* collects information obtained during the pre-pair loop onto local
595	+	* data structures.
596	+	*/
597		void ForceMatrixDecomposition::collectIntermediateData() {
598		snap_ = sman_->getCurrentSnapshot();
599		storageLayout_ = sman_->getStorageLayout();
#	Line 273 | Line 601 | namespace OpenMD {
601
602		if (storageLayout_ & DataStorage::dslDensity) {
603
604	<	AtomCommRealRow->scatter(atomRowData.density,
604	>	AtomPlanRealRow->scatter(atomRowData.density,
605		snap_->atomData.density);
606
607		int n = snap_->atomData.density.size();
608	<	std::vector<RealType> rho_tmp(n, 0.0);
609	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
608	>	vector<RealType> rho_tmp(n, 0.0);
609	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
610		for (int i = 0; i < n; i++)
611		snap_->atomData.density[i] += rho_tmp[i];
612		}
613		#endif
614		}
615	<
615	>
616	>	/*
617	>	* redistributes information obtained during the pre-pair loop out to
618	>	* row and column-indexed data structures
619	>	*/
620		void ForceMatrixDecomposition::distributeIntermediateData() {
621		snap_ = sman_->getCurrentSnapshot();
622		storageLayout_ = sman_->getStorageLayout();
623		#ifdef IS_MPI
624		if (storageLayout_ & DataStorage::dslFunctional) {
625	<	AtomCommRealRow->gather(snap_->atomData.functional,
625	>	AtomPlanRealRow->gather(snap_->atomData.functional,
626		atomRowData.functional);
627	<	AtomCommRealColumn->gather(snap_->atomData.functional,
627	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
628		atomColData.functional);
629		}
630
631		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
632	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
632	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
633		atomRowData.functionalDerivative);
634	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
634	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
635		atomColData.functionalDerivative);
636		}
637		#endif
#	Line 313 | Line 645 | namespace OpenMD {
645		int n = snap_->atomData.force.size();
646		vector<Vector3d> frc_tmp(n, V3Zero);
647
648	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
648	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
649		for (int i = 0; i < n; i++) {
650		snap_->atomData.force[i] += frc_tmp[i];
651		frc_tmp[i] = 0.0;
652		}
653
654	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
655	<	for (int i = 0; i < n; i++)
654	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
655	>	for (int i = 0; i < n; i++) {
656		snap_->atomData.force[i] += frc_tmp[i];
657	<
658	<
657	>	}
658	>
659		if (storageLayout_ & DataStorage::dslTorque) {
660
661	<	int nt = snap_->atomData.force.size();
661	>	int nt = snap_->atomData.torque.size();
662		vector<Vector3d> trq_tmp(nt, V3Zero);
663
664	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
665	<	for (int i = 0; i < n; i++) {
664	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
665	>	for (int i = 0; i < nt; i++) {
666		snap_->atomData.torque[i] += trq_tmp[i];
667		trq_tmp[i] = 0.0;
668		}
669
670	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
671	<	for (int i = 0; i < n; i++)
670	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
671	>	for (int i = 0; i < nt; i++)
672		snap_->atomData.torque[i] += trq_tmp[i];
673		}
674	+
675	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
676	+
677	+	int ns = snap_->atomData.skippedCharge.size();
678	+	vector<RealType> skch_tmp(ns, 0.0);
679	+
680	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
681	+	for (int i = 0; i < ns; i++) {
682	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
683	+	skch_tmp[i] = 0.0;
684	+	}
685	+
686	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
687	+	for (int i = 0; i < ns; i++)
688	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
689	+
690	+	}
691
692		nLocal_ = snap_->getNumberOfAtoms();
693
694	<	vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
695	<	vector<RealType> (nLocal_, 0.0));
694	>	vector<potVec> pot_temp(nLocal_,
695	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
696	>
697	>	// scatter/gather pot_row into the members of my column
698	>
699	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
700	>
701	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
702	>	pairwisePot += pot_temp[ii];
703
704	<	for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
705	<	AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
706	<	for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
707	<	pot_local[i] += pot_temp[i][ii];
708	<	}
704	>	fill(pot_temp.begin(), pot_temp.end(),
705	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
706	>
707	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
708	>
709	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
710	>	pairwisePot += pot_temp[ii];
711	>
712	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
713	>	RealType ploc1 = pairwisePot[ii];
714	>	RealType ploc2 = 0.0;
715	>	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
716	>	pairwisePot[ii] = ploc2;
717		}
718	+
719	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
720	+	RealType ploc1 = embeddingPot[ii];
721	+	RealType ploc2 = 0.0;
722	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
723	+	embeddingPot[ii] = ploc2;
724	+	}
725	+
726		#endif
727	+
728		}
729
730		int ForceMatrixDecomposition::getNAtomsInRow() {
#	Line 426 | Line 799 | namespace OpenMD {
799		#ifdef IS_MPI
800		return massFactorsRow[atom1];
801		#else
802	<	return massFactorsLocal[atom1];
802	>	return massFactors[atom1];
803		#endif
804		}
805
#	Line 434 | Line 807 | namespace OpenMD {
807		#ifdef IS_MPI
808		return massFactorsCol[atom2];
809		#else
810	<	return massFactorsLocal[atom2];
810	>	return massFactors[atom2];
811		#endif
812
813		}
#	Line 452 | Line 825 | namespace OpenMD {
825		return d;
826		}
827
828	<	vector<int> ForceMatrixDecomposition::getSkipsForRowAtom(int atom1) {
829	<	#ifdef IS_MPI
457	<	return skipsForRowAtom[atom1];
458	<	#else
459	<	return skipsForLocalAtom[atom1];
460	<	#endif
828	>	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
829	>	return excludesForAtom[atom1];
830		}
831
832		/**
833	<	* there are a number of reasons to skip a pair or a particle mostly
834	<	* we do this to exclude atoms who are involved in short range
466	<	* interactions (bonds, bends, torsions), but we also need to
467	<	* exclude some overcounted interactions that result from the
468	<	* parallel decomposition.
833	>	* We need to exclude some overcounted interactions that result from
834	>	* the parallel decomposition.
835		*/
836		bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
837		int unique_id_1, unique_id_2;
838	<
838	>
839		#ifdef IS_MPI
840		// in MPI, we have to look up the unique IDs for each atom
841		unique_id_1 = AtomRowToGlobal[atom1];
842		unique_id_2 = AtomColToGlobal[atom2];
843	+	#else
844	+	unique_id_1 = AtomLocalToGlobal[atom1];
845	+	unique_id_2 = AtomLocalToGlobal[atom2];
846	+	#endif
847
478	–	// this situation should only arise in MPI simulations
848		if (unique_id_1 == unique_id_2) return true;
849	<
849	>
850	>	#ifdef IS_MPI
851		// this prevents us from doing the pair on multiple processors
852		if (unique_id_1 < unique_id_2) {
853		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
854		} else {
855	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
855	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
856		}
487	–	#else
488	–	// in the normal loop, the atom numbers are unique
489	–	unique_id_1 = atom1;
490	–	unique_id_2 = atom2;
857		#endif
858
859	<	#ifdef IS_MPI
494	<	for (vector<int>::iterator i = skipsForRowAtom[atom1].begin();
495	<	i != skipsForRowAtom[atom1].end(); ++i) {
496	<	if ( (*i) == unique_id_2 ) return true;
497	<	}
498	<	#else
499	<	for (vector<int>::iterator i = skipsForLocalAtom[atom1].begin();
500	<	i != skipsForLocalAtom[atom1].end(); ++i) {
501	<	if ( (*i) == unique_id_2 ) return true;
502	<	}
503	<	#endif
859	>	return false;
860		}
861
862	<	int ForceMatrixDecomposition::getTopoDistance(int atom1, int atom2) {
862	>	/**
863	>	* We need to handle the interactions for atoms who are involved in
864	>	* the same rigid body as well as some short range interactions
865	>	* (bonds, bends, torsions) differently from other interactions.
866	>	* We'll still visit the pairwise routines, but with a flag that
867	>	* tells those routines to exclude the pair from direct long range
868	>	* interactions.  Some indirect interactions (notably reaction
869	>	* field) must still be handled for these pairs.
870	>	*/
871	>	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
872	>
873	>	// excludesForAtom was constructed to use row/column indices in the MPI
874	>	// version, and to use local IDs in the non-MPI version:
875
876	<	#ifdef IS_MPI
877	<	for (int i = 0; i < toposForRowAtom[atom1].size(); i++) {
878	<	if ( toposForRowAtom[atom1][i] == atom2 ) return topoDistRow[atom1][i];
511	<	}
512	<	#else
513	<	for (int i = 0; i < toposForLocalAtom[atom1].size(); i++) {
514	<	if ( toposForLocalAtom[atom1][i] == atom2 ) return topoDistLocal[atom1][i];
876	>	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
877	>	i != excludesForAtom[atom1].end(); ++i) {
878	>	if ( (*i) == atom2 ) return true;
879		}
516	–	#endif
880
881	<	// zero is default for unconnected (i.e. normal) pair interactions
519	<	return 0;
881	>	return false;
882		}
883
884	+
885		void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
886		#ifdef IS_MPI
887		atomRowData.force[atom1] += fg;
#	Line 536 | Line 899 | namespace OpenMD {
899		}
900
901		// filling interaction blocks with pointers
902	<	InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {
903	<	InteractionData idat;
902	>	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
903	>	int atom1, int atom2) {
904
905	+	idat.excluded = excludeAtomPair(atom1, atom2);
906	+
907		#ifdef IS_MPI
908	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
909	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
910	+	//                         ff_->getAtomType(identsCol[atom2]) );
911	+
912		if (storageLayout_ & DataStorage::dslAmat) {
913		idat.A1 = &(atomRowData.aMat[atom1]);
914		idat.A2 = &(atomColData.aMat[atom2]);
#	Line 560 | Line 929 | namespace OpenMD {
929		idat.rho2 = &(atomColData.density[atom2]);
930		}
931
932	+	if (storageLayout_ & DataStorage::dslFunctional) {
933	+	idat.frho1 = &(atomRowData.functional[atom1]);
934	+	idat.frho2 = &(atomColData.functional[atom2]);
935	+	}
936	+
937		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
938		idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
939		idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
940		}
941
942	+	if (storageLayout_ & DataStorage::dslParticlePot) {
943	+	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
944	+	idat.particlePot2 = &(atomColData.particlePot[atom2]);
945	+	}
946	+
947	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
948	+	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
949	+	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
950	+	}
951	+
952		#else
953	+
954	+	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
955	+	//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
956	+	//                         ff_->getAtomType(idents[atom2]) );
957	+
958		if (storageLayout_ & DataStorage::dslAmat) {
959		idat.A1 = &(snap_->atomData.aMat[atom1]);
960		idat.A2 = &(snap_->atomData.aMat[atom2]);
#	Line 581 | Line 970 | namespace OpenMD {
970		idat.t2 = &(snap_->atomData.torque[atom2]);
971		}
972
973	<	if (storageLayout_ & DataStorage::dslDensity) {
973	>	if (storageLayout_ & DataStorage::dslDensity) {
974		idat.rho1 = &(snap_->atomData.density[atom1]);
975		idat.rho2 = &(snap_->atomData.density[atom2]);
976		}
977
978	+	if (storageLayout_ & DataStorage::dslFunctional) {
979	+	idat.frho1 = &(snap_->atomData.functional[atom1]);
980	+	idat.frho2 = &(snap_->atomData.functional[atom2]);
981	+	}
982	+
983		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
984		idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
985		idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
986		}
987	+
988	+	if (storageLayout_ & DataStorage::dslParticlePot) {
989	+	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
990	+	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
991	+	}
992	+
993	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
994	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
995	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
996	+	}
997		#endif
594	–	return idat;
998		}
999
1000	<	InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
1001	<
599	<	InteractionData idat;
1000	>
1001	>	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1002		#ifdef IS_MPI
1003	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1004	<	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1005	<	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1006	<	}
1007	<	if (storageLayout_ & DataStorage::dslTorque) {
606	<	idat.t1 = &(atomRowData.torque[atom1]);
607	<	idat.t2 = &(atomColData.torque[atom2]);
608	<	}
609	<	if (storageLayout_ & DataStorage::dslForce) {
610	<	idat.t1 = &(atomRowData.force[atom1]);
611	<	idat.t2 = &(atomColData.force[atom2]);
612	<	}
1003	>	pot_row[atom1] += 0.5 *  *(idat.pot);
1004	>	pot_col[atom2] += 0.5 *  *(idat.pot);
1005	>
1006	>	atomRowData.force[atom1] += *(idat.f1);
1007	>	atomColData.force[atom2] -= *(idat.f1);
1008		#else
1009	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1010	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1011	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1012	<	}
618	<	if (storageLayout_ & DataStorage::dslTorque) {
619	<	idat.t1 = &(snap_->atomData.torque[atom1]);
620	<	idat.t2 = &(snap_->atomData.torque[atom2]);
621	<	}
622	<	if (storageLayout_ & DataStorage::dslForce) {
623	<	idat.t1 = &(snap_->atomData.force[atom1]);
624	<	idat.t2 = &(snap_->atomData.force[atom2]);
625	<	}
1009	>	pairwisePot += *(idat.pot);
1010	>
1011	>	snap_->atomData.force[atom1] += *(idat.f1);
1012	>	snap_->atomData.force[atom2] -= *(idat.f1);
1013		#endif
1014
1015		}
1016
630	–
631	–
632	–
1017		/*
1018		* buildNeighborList
1019		*
#	Line 639 | Line 1023 | namespace OpenMD {
1023		vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1024
1025		vector<pair<int, int> > neighborList;
1026	+	groupCutoffs cuts;
1027	+	bool doAllPairs = false;
1028	+
1029		#ifdef IS_MPI
1030		cellListRow_.clear();
1031		cellListCol_.clear();
#	Line 646 | Line 1033 | namespace OpenMD {
1033		cellList_.clear();
1034		#endif
1035
1036	<	// dangerous to not do error checking.
650	<	RealType rCut_;
651	<
652	<	RealType rList_ = (rCut_ + skinThickness_);
1036	>	RealType rList_ = (largestRcut_ + skinThickness_);
1037		RealType rl2 = rList_ * rList_;
1038		Snapshot* snap_ = sman_->getCurrentSnapshot();
1039		Mat3x3d Hmat = snap_->getHmat();
#	Line 661 | Line 1045 | namespace OpenMD {
1045		nCells_.y() = (int) ( Hy.length() )/ rList_;
1046		nCells_.z() = (int) ( Hz.length() )/ rList_;
1047
1048	+	// handle small boxes where the cell offsets can end up repeating cells
1049	+
1050	+	if (nCells_.x() < 3) doAllPairs = true;
1051	+	if (nCells_.y() < 3) doAllPairs = true;
1052	+	if (nCells_.z() < 3) doAllPairs = true;
1053	+
1054		Mat3x3d invHmat = snap_->getInvHmat();
1055		Vector3d rs, scaled, dr;
1056		Vector3i whichCell;
1057		int cellIndex;
1058	+	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1059
1060		#ifdef IS_MPI
1061	<	for (int i = 0; i < nGroupsInRow_; i++) {
1062	<	rs = cgRowData.position[i];
1063	<	// scaled positions relative to the box vectors
1064	<	scaled = invHmat * rs;
1065	<	// wrap the vector back into the unit box by subtracting integer box
675	<	// numbers
676	<	for (int j = 0; j < 3; j++)
677	<	scaled[j] -= roundMe(scaled[j]);
678	<
679	<	// find xyz-indices of cell that cutoffGroup is in.
680	<	whichCell.x() = nCells_.x() * scaled.x();
681	<	whichCell.y() = nCells_.y() * scaled.y();
682	<	whichCell.z() = nCells_.z() * scaled.z();
1061	>	cellListRow_.resize(nCtot);
1062	>	cellListCol_.resize(nCtot);
1063	>	#else
1064	>	cellList_.resize(nCtot);
1065	>	#endif
1066
1067	<	// find single index of this cell:
1068	<	cellIndex = Vlinear(whichCell, nCells_);
686	<	// add this cutoff group to the list of groups in this cell;
687	<	cellListRow_[cellIndex].push_back(i);
688	<	}
1067	>	if (!doAllPairs) {
1068	>	#ifdef IS_MPI
1069
1070	<	for (int i = 0; i < nGroupsInCol_; i++) {
1071	<	rs = cgColData.position[i];
1072	<	// scaled positions relative to the box vectors
1073	<	scaled = invHmat * rs;
1074	<	// wrap the vector back into the unit box by subtracting integer box
1075	<	// numbers
1076	<	for (int j = 0; j < 3; j++)
1077	<	scaled[j] -= roundMe(scaled[j]);
1078	<
1079	<	// find xyz-indices of cell that cutoffGroup is in.
1080	<	whichCell.x() = nCells_.x() * scaled.x();
1081	<	whichCell.y() = nCells_.y() * scaled.y();
1082	<	whichCell.z() = nCells_.z() * scaled.z();
1083	<
1084	<	// find single index of this cell:
1085	<	cellIndex = Vlinear(whichCell, nCells_);
1086	<	// add this cutoff group to the list of groups in this cell;
1087	<	cellListCol_[cellIndex].push_back(i);
1088	<	}
1070	>	for (int i = 0; i < nGroupsInRow_; i++) {
1071	>	rs = cgRowData.position[i];
1072	>
1073	>	// scaled positions relative to the box vectors
1074	>	scaled = invHmat * rs;
1075	>
1076	>	// wrap the vector back into the unit box by subtracting integer box
1077	>	// numbers
1078	>	for (int j = 0; j < 3; j++) {
1079	>	scaled[j] -= roundMe(scaled[j]);
1080	>	scaled[j] += 0.5;
1081	>	}
1082	>
1083	>	// find xyz-indices of cell that cutoffGroup is in.
1084	>	whichCell.x() = nCells_.x() * scaled.x();
1085	>	whichCell.y() = nCells_.y() * scaled.y();
1086	>	whichCell.z() = nCells_.z() * scaled.z();
1087	>
1088	>	// find single index of this cell:
1089	>	cellIndex = Vlinear(whichCell, nCells_);
1090	>
1091	>	// add this cutoff group to the list of groups in this cell;
1092	>	cellListRow_[cellIndex].push_back(i);
1093	>	}
1094	>	for (int i = 0; i < nGroupsInCol_; i++) {
1095	>	rs = cgColData.position[i];
1096	>
1097	>	// scaled positions relative to the box vectors
1098	>	scaled = invHmat * rs;
1099	>
1100	>	// wrap the vector back into the unit box by subtracting integer box
1101	>	// numbers
1102	>	for (int j = 0; j < 3; j++) {
1103	>	scaled[j] -= roundMe(scaled[j]);
1104	>	scaled[j] += 0.5;
1105	>	}
1106	>
1107	>	// find xyz-indices of cell that cutoffGroup is in.
1108	>	whichCell.x() = nCells_.x() * scaled.x();
1109	>	whichCell.y() = nCells_.y() * scaled.y();
1110	>	whichCell.z() = nCells_.z() * scaled.z();
1111	>
1112	>	// find single index of this cell:
1113	>	cellIndex = Vlinear(whichCell, nCells_);
1114	>
1115	>	// add this cutoff group to the list of groups in this cell;
1116	>	cellListCol_[cellIndex].push_back(i);
1117	>	}
1118	>
1119		#else
1120	<	for (int i = 0; i < nGroups_; i++) {
1121	<	rs = snap_->cgData.position[i];
1122	<	// scaled positions relative to the box vectors
1123	<	scaled = invHmat * rs;
1124	<	// wrap the vector back into the unit box by subtracting integer box
1125	<	// numbers
1126	<	for (int j = 0; j < 3; j++)
1127	<	scaled[j] -= roundMe(scaled[j]);
1120	>	for (int i = 0; i < nGroups_; i++) {
1121	>	rs = snap_->cgData.position[i];
1122	>
1123	>	// scaled positions relative to the box vectors
1124	>	scaled = invHmat * rs;
1125	>
1126	>	// wrap the vector back into the unit box by subtracting integer box
1127	>	// numbers
1128	>	for (int j = 0; j < 3; j++) {
1129	>	scaled[j] -= roundMe(scaled[j]);
1130	>	scaled[j] += 0.5;
1131	>	}
1132	>
1133	>	// find xyz-indices of cell that cutoffGroup is in.
1134	>	whichCell.x() = nCells_.x() * scaled.x();
1135	>	whichCell.y() = nCells_.y() * scaled.y();
1136	>	whichCell.z() = nCells_.z() * scaled.z();
1137	>
1138	>	// find single index of this cell:
1139	>	cellIndex = Vlinear(whichCell, nCells_);
1140	>
1141	>	// add this cutoff group to the list of groups in this cell;
1142	>	cellList_[cellIndex].push_back(i);
1143	>	}
1144
719	–	// find xyz-indices of cell that cutoffGroup is in.
720	–	whichCell.x() = nCells_.x() * scaled.x();
721	–	whichCell.y() = nCells_.y() * scaled.y();
722	–	whichCell.z() = nCells_.z() * scaled.z();
723	–
724	–	// find single index of this cell:
725	–	cellIndex = Vlinear(whichCell, nCells_);
726	–	// add this cutoff group to the list of groups in this cell;
727	–	cellList_[cellIndex].push_back(i);
728	–	}
1145		#endif
1146
1147	<
1148	<
1149	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1150	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1151	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
736	<	Vector3i m1v(m1x, m1y, m1z);
737	<	int m1 = Vlinear(m1v, nCells_);
738	<
739	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
740	<	os != cellOffsets_.end(); ++os) {
1147	>	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1148	>	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1149	>	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1150	>	Vector3i m1v(m1x, m1y, m1z);
1151	>	int m1 = Vlinear(m1v, nCells_);
1152
1153	<	Vector3i m2v = m1v + (*os);
1154	<
1155	<	if (m2v.x() >= nCells_.x()) {
1156	<	m2v.x() = 0;
1157	<	} else if (m2v.x() < 0) {
747	<	m2v.x() = nCells_.x() - 1;
748	<	}
749	<
750	<	if (m2v.y() >= nCells_.y()) {
751	<	m2v.y() = 0;
752	<	} else if (m2v.y() < 0) {
753	<	m2v.y() = nCells_.y() - 1;
754	<	}
755	<
756	<	if (m2v.z() >= nCells_.z()) {
757	<	m2v.z() = 0;
758	<	} else if (m2v.z() < 0) {
759	<	m2v.z() = nCells_.z() - 1;
760	<	}
761	<
762	<	int m2 = Vlinear (m2v, nCells_);
1153	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1154	>	os != cellOffsets_.end(); ++os) {
1155	>
1156	>	Vector3i m2v = m1v + (*os);
1157	>
1158
1159	<	#ifdef IS_MPI
1160	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1161	<	j1 != cellListRow_[m1].end(); ++j1) {
1162	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1163	<	j2 != cellListCol_[m2].end(); ++j2) {
1164	<
1165	<	// Always do this if we're in different cells or if
1166	<	// we're in the same cell and the global index of the
1167	<	// j2 cutoff group is less than the j1 cutoff group
1159	>	if (m2v.x() >= nCells_.x()) {
1160	>	m2v.x() = 0;
1161	>	} else if (m2v.x() < 0) {
1162	>	m2v.x() = nCells_.x() - 1;
1163	>	}
1164	>
1165	>	if (m2v.y() >= nCells_.y()) {
1166	>	m2v.y() = 0;
1167	>	} else if (m2v.y() < 0) {
1168	>	m2v.y() = nCells_.y() - 1;
1169	>	}
1170	>
1171	>	if (m2v.z() >= nCells_.z()) {
1172	>	m2v.z() = 0;
1173	>	} else if (m2v.z() < 0) {
1174	>	m2v.z() = nCells_.z() - 1;
1175	>	}
1176
1177	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1177	>	int m2 = Vlinear (m2v, nCells_);
1178	>
1179	>	#ifdef IS_MPI
1180	>	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1181	>	j1 != cellListRow_[m1].end(); ++j1) {
1182	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1183	>	j2 != cellListCol_[m2].end(); ++j2) {
1184	>
1185	>	// In parallel, we need to visit *all* pairs of row
1186	>	// & column indicies and will divide labor in the
1187	>	// force evaluation later.
1188		dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1189		snap_->wrapVector(dr);
1190	<	if (dr.lengthSquare() < rl2) {
1190	>	cuts = getGroupCutoffs( (*j1), (*j2) );
1191	>	if (dr.lengthSquare() < cuts.third) {
1192		neighborList.push_back(make_pair((*j1), (*j2)));
1193	<	}
1193	>	}
1194		}
1195		}
782	–	}
1196		#else
1197	<	for (vector<int>::iterator j1 = cellList_[m1].begin();
1198	<	j1 != cellList_[m1].end(); ++j1) {
1199	<	for (vector<int>::iterator j2 = cellList_[m2].begin();
1200	<	j2 != cellList_[m2].end(); ++j2) {
1201	<
1202	<	// Always do this if we're in different cells or if
1203	<	// we're in the same cell and the global index of the
1204	<	// j2 cutoff group is less than the j1 cutoff group
1197	>	for (vector<int>::iterator j1 = cellList_[m1].begin();
1198	>	j1 != cellList_[m1].end(); ++j1) {
1199	>	for (vector<int>::iterator j2 = cellList_[m2].begin();
1200	>	j2 != cellList_[m2].end(); ++j2) {
1201	>
1202	>	// Always do this if we're in different cells or if
1203	>	// we're in the same cell and the global index of
1204	>	// the j2 cutoff group is greater than or equal to
1205	>	// the j1 cutoff group.  Note that Rappaport's code
1206	>	// has a "less than" conditional here, but that
1207	>	// deals with atom-by-atom computation.  OpenMD
1208	>	// allows atoms within a single cutoff group to
1209	>	// interact with each other.
1210
1211	<	if (m2 != m1 || (*j2) < (*j1)) {
1212	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1213	<	snap_->wrapVector(dr);
1214	<	if (dr.lengthSquare() < rl2) {
1215	<	neighborList.push_back(make_pair((*j1), (*j2)));
1211	>
1212	>
1213	>	if (m2 != m1 || (*j2) >= (*j1) ) {
1214	>
1215	>	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1216	>	snap_->wrapVector(dr);
1217	>	cuts = getGroupCutoffs( (*j1), (*j2) );
1218	>	if (dr.lengthSquare() < cuts.third) {
1219	>	neighborList.push_back(make_pair((*j1), (*j2)));
1220	>	}
1221		}
1222		}
1223		}
801	–	}
1224		#endif
1225	+	}
1226		}
1227		}
1228		}
1229	+	} else {
1230	+	// branch to do all cutoff group pairs
1231	+	#ifdef IS_MPI
1232	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1233	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1234	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1235	+	snap_->wrapVector(dr);
1236	+	cuts = getGroupCutoffs( j1, j2 );
1237	+	if (dr.lengthSquare() < cuts.third) {
1238	+	neighborList.push_back(make_pair(j1, j2));
1239	+	}
1240	+	}
1241	+	}
1242	+	#else
1243	+	// include all groups here.
1244	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1245	+	// include self group interactions j2 == j1
1246	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1247	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1248	+	snap_->wrapVector(dr);
1249	+	cuts = getGroupCutoffs( j1, j2 );
1250	+	if (dr.lengthSquare() < cuts.third) {
1251	+	neighborList.push_back(make_pair(j1, j2));
1252	+	}
1253	+	}
1254	+	}
1255	+	#endif
1256		}
1257	<
1257	>
1258		// save the local cutoff group positions for the check that is
1259		// done on each loop:
1260		saved_CG_positions_.clear();
1261		for (int i = 0; i < nGroups_; i++)
1262		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1263	<
1263	>
1264		return neighborList;
1265		}
1266		} //end namespace OpenMD

Diff Legend

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