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 1706 by gezelter, Fri Apr 27 20:44:16 2012 UTC

#	Line 36 | Line 36
36		* [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37		* [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38		* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
39	<	* [4]  Vardeman & Gezelter, in progress (2009).
39	>	* [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40	>	* [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41		*/
42		#include "parallel/ForceMatrixDecomposition.hpp"
43		#include "math/SquareMatrix3.hpp"
#	Line 47 | Line 48 | namespace OpenMD {
48		using namespace std;
49		namespace OpenMD {
50
51	+	ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52	+
53	+	// In a parallel computation, row and colum scans must visit all
54	+	// surrounding cells (not just the 14 upper triangular blocks that
55	+	// are used when the processor can see all pairs)
56	+	#ifdef IS_MPI
57	+	cellOffsets_.clear();
58	+	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59	+	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60	+	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
61	+	cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62	+	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63	+	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
64	+	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65	+	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
66	+	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67	+	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68	+	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69	+	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70	+	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
71	+	cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72	+	cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73	+	cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74	+	cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75	+	cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76	+	cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77	+	cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78	+	cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79	+	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80	+	cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81	+	cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82	+	cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83	+	cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84	+	cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85	+	#endif
86	+	}
87	+
88	+
89		/**
90		* distributeInitialData is essentially a copy of the older fortran
91		* SimulationSetup
92		*/
54	–
93		void ForceMatrixDecomposition::distributeInitialData() {
94		snap_ = sman_->getCurrentSnapshot();
95		storageLayout_ = sman_->getStorageLayout();
96	+	ff_ = info_->getForceField();
97		nLocal_ = snap_->getNumberOfAtoms();
98	<	nGroups_ = snap_->getNumberOfCutoffGroups();
99	<
98	>
99	>	nGroups_ = info_->getNLocalCutoffGroups();
100		// gather the information for atomtype IDs (atids):
101	<	vector<int> identsLocal = info_->getIdentArray();
101	>	idents = info_->getIdentArray();
102		AtomLocalToGlobal = info_->getGlobalAtomIndices();
103		cgLocalToGlobal = info_->getGlobalGroupIndices();
104		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);
105
106	+	massFactors = info_->getMassFactors();
107	+
108	+	PairList* excludes = info_->getExcludedInteractions();
109	+	PairList* oneTwo = info_->getOneTwoInteractions();
110	+	PairList* oneThree = info_->getOneThreeInteractions();
111	+	PairList* oneFour = info_->getOneFourInteractions();
112	+
113		#ifdef IS_MPI
114
115	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
116	<	AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
77	<	AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
78	<	AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
115	>	MPI::Intracomm row = rowComm.getComm();
116	>	MPI::Intracomm col = colComm.getComm();
117
118	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
119	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
120	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
121	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
118	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
119	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
120	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
121	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
122	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
123
124	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
125	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
126	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
127	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
124	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
125	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
126	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
127	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
128	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
129
130	<	nAtomsInRow_ = AtomCommIntRow->getSize();
131	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
132	<	nGroupsInRow_ = cgCommIntRow->getSize();
133	<	nGroupsInCol_ = cgCommIntColumn->getSize();
130	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
131	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
132	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
133	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
134
135	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
136	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
137	+	nGroupsInRow_ = cgPlanIntRow->getSize();
138	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
139	+
140		// Modify the data storage objects with the correct layouts and sizes:
141		atomRowData.resize(nAtomsInRow_);
142		atomRowData.setStorageLayout(storageLayout_);
#	Line 101 | Line 146 | namespace OpenMD {
146		cgRowData.setStorageLayout(DataStorage::dslPosition);
147		cgColData.resize(nGroupsInCol_);
148		cgColData.setStorageLayout(DataStorage::dslPosition);
149	+
150	+	identsRow.resize(nAtomsInRow_);
151	+	identsCol.resize(nAtomsInCol_);
152
153	<	vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
154	<	vector<RealType> (nAtomsInRow_, 0.0));
107	<	vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
108	<	vector<RealType> (nAtomsInCol_, 0.0));
153	>	AtomPlanIntRow->gather(idents, identsRow);
154	>	AtomPlanIntColumn->gather(idents, identsCol);
155
156	<	identsRow.reserve(nAtomsInRow_);
157	<	identsCol.reserve(nAtomsInCol_);
158	<
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);
156	>	// allocate memory for the parallel objects
157	>	atypesRow.resize(nAtomsInRow_);
158	>	atypesCol.resize(nAtomsInCol_);
159
160	<	AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
161	<	AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
160	>	for (int i = 0; i < nAtomsInRow_; i++)
161	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
162	>	for (int i = 0; i < nAtomsInCol_; i++)
163	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
164
165	+	pot_row.resize(nAtomsInRow_);
166	+	pot_col.resize(nAtomsInCol_);
167	+
168	+	AtomRowToGlobal.resize(nAtomsInRow_);
169	+	AtomColToGlobal.resize(nAtomsInCol_);
170	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
171	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
172	+
173	+	cgRowToGlobal.resize(nGroupsInRow_);
174	+	cgColToGlobal.resize(nGroupsInCol_);
175	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
176	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
177	+
178	+	massFactorsRow.resize(nAtomsInRow_);
179	+	massFactorsCol.resize(nAtomsInCol_);
180	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
181	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
182	+
183		groupListRow_.clear();
184	<	groupListRow_.reserve(nGroupsInRow_);
184	>	groupListRow_.resize(nGroupsInRow_);
185		for (int i = 0; i < nGroupsInRow_; i++) {
186		int gid = cgRowToGlobal[i];
187		for (int j = 0; j < nAtomsInRow_; j++) {
#	Line 134 | Line 192 | namespace OpenMD {
192		}
193
194		groupListCol_.clear();
195	<	groupListCol_.reserve(nGroupsInCol_);
195	>	groupListCol_.resize(nGroupsInCol_);
196		for (int i = 0; i < nGroupsInCol_; i++) {
197		int gid = cgColToGlobal[i];
198		for (int j = 0; j < nAtomsInCol_; j++) {
#	Line 144 | Line 202 | namespace OpenMD {
202		}
203		}
204
205	<	skipsForRowAtom.clear();
206	<	skipsForRowAtom.reserve(nAtomsInRow_);
207	<	for (int i = 0; i < nAtomsInRow_; i++) {
208	<	int iglob = AtomColToGlobal[i];
205	>	excludesForAtom.clear();
206	>	excludesForAtom.resize(nAtomsInRow_);
207	>	toposForAtom.clear();
208	>	toposForAtom.resize(nAtomsInRow_);
209	>	topoDist.clear();
210	>	topoDist.resize(nAtomsInRow_);
211	>	for (int i = 0; i < nAtomsInRow_; i++) {
212	>	int iglob = AtomRowToGlobal[i];
213	>
214		for (int j = 0; j < nAtomsInCol_; j++) {
215	<	int jglob = AtomRowToGlobal[j];
216	<	if (excludes.hasPair(iglob, jglob))
217	<	skipsForRowAtom[i].push_back(j);
215	>	int jglob = AtomColToGlobal[j];
216	>
217	>	if (excludes->hasPair(iglob, jglob))
218	>	excludesForAtom[i].push_back(j);
219	>
220	>	if (oneTwo->hasPair(iglob, jglob)) {
221	>	toposForAtom[i].push_back(j);
222	>	topoDist[i].push_back(1);
223	>	} else {
224	>	if (oneThree->hasPair(iglob, jglob)) {
225	>	toposForAtom[i].push_back(j);
226	>	topoDist[i].push_back(2);
227	>	} else {
228	>	if (oneFour->hasPair(iglob, jglob)) {
229	>	toposForAtom[i].push_back(j);
230	>	topoDist[i].push_back(3);
231	>	}
232	>	}
233	>	}
234		}
235		}
236
237	<	toposForRowAtom.clear();
238	<	toposForRowAtom.reserve(nAtomsInRow_);
239	<	for (int i = 0; i < nAtomsInRow_; i++) {
240	<	int iglob = AtomColToGlobal[i];
241	<	int nTopos = 0;
242	<	for (int j = 0; j < nAtomsInCol_; j++) {
243	<	int jglob = AtomRowToGlobal[j];
244	<	if (oneTwo.hasPair(iglob, jglob)) {
245	<	toposForRowAtom[i].push_back(j);
246	<	topoDistRow[i][nTopos] = 1;
247	<	nTopos++;
237	>	#else
238	>	excludesForAtom.clear();
239	>	excludesForAtom.resize(nLocal_);
240	>	toposForAtom.clear();
241	>	toposForAtom.resize(nLocal_);
242	>	topoDist.clear();
243	>	topoDist.resize(nLocal_);
244	>
245	>	for (int i = 0; i < nLocal_; i++) {
246	>	int iglob = AtomLocalToGlobal[i];
247	>
248	>	for (int j = 0; j < nLocal_; j++) {
249	>	int jglob = AtomLocalToGlobal[j];
250	>
251	>	if (excludes->hasPair(iglob, jglob))
252	>	excludesForAtom[i].push_back(j);
253	>
254	>	if (oneTwo->hasPair(iglob, jglob)) {
255	>	toposForAtom[i].push_back(j);
256	>	topoDist[i].push_back(1);
257	>	} else {
258	>	if (oneThree->hasPair(iglob, jglob)) {
259	>	toposForAtom[i].push_back(j);
260	>	topoDist[i].push_back(2);
261	>	} else {
262	>	if (oneFour->hasPair(iglob, jglob)) {
263	>	toposForAtom[i].push_back(j);
264	>	topoDist[i].push_back(3);
265	>	}
266	>	}
267		}
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	–	}
268		}
269		}
182	–
270		#endif
271
272	+	// allocate memory for the parallel objects
273	+	atypesLocal.resize(nLocal_);
274	+
275	+	for (int i = 0; i < nLocal_; i++)
276	+	atypesLocal[i] = ff_->getAtomType(idents[i]);
277	+
278		groupList_.clear();
279	<	groupList_.reserve(nGroups_);
279	>	groupList_.resize(nGroups_);
280		for (int i = 0; i < nGroups_; i++) {
281		int gid = cgLocalToGlobal[i];
282		for (int j = 0; j < nLocal_; j++) {
283		int aid = AtomLocalToGlobal[j];
284	<	if (globalGroupMembership[aid] == gid)
284	>	if (globalGroupMembership[aid] == gid) {
285		groupList_[i].push_back(j);
286	+	}
287		}
288		}
289
196	–	skipsForLocalAtom.clear();
197	–	skipsForLocalAtom.reserve(nLocal_);
290
291	<	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	<	}
291	>	createGtypeCutoffMap();
292
293	<	toposForLocalAtom.clear();
294	<	toposForLocalAtom.reserve(nLocal_);
295	<	for (int i = 0; i < nLocal_; i++) {
296	<	int iglob = AtomLocalToGlobal[i];
297	<	int nTopos = 0;
298	<	for (int j = 0; j < nLocal_; j++) {
299	<	int jglob = AtomLocalToGlobal[j];
300	<	if (oneTwo.hasPair(iglob, jglob)) {
301	<	toposForLocalAtom[i].push_back(j);
302	<	topoDistLocal[i][nTopos] = 1;
303	<	nTopos++;
304	<	}
305	<	if (oneThree.hasPair(iglob, jglob)) {
306	<	toposForLocalAtom[i].push_back(j);
307	<	topoDistLocal[i][nTopos] = 2;
308	<	nTopos++;
309	<	}
310	<	if (oneFour.hasPair(iglob, jglob)) {
311	<	toposForLocalAtom[i].push_back(j);
312	<	topoDistLocal[i][nTopos] = 3;
313	<	nTopos++;
293	>	}
294	>
295	>	void ForceMatrixDecomposition::createGtypeCutoffMap() {
296	>
297	>	RealType tol = 1e-6;
298	>	largestRcut_ = 0.0;
299	>	RealType rc;
300	>	int atid;
301	>	set<AtomType*> atypes = info_->getSimulatedAtomTypes();
302	>
303	>	map<int, RealType> atypeCutoff;
304	>
305	>	for (set<AtomType*>::iterator at = atypes.begin();
306	>	at != atypes.end(); ++at){
307	>	atid = (*at)->getIdent();
308	>	if (userChoseCutoff_)
309	>	atypeCutoff[atid] = userCutoff_;
310	>	else
311	>	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
312	>	}
313	>
314	>	vector<RealType> gTypeCutoffs;
315	>	// first we do a single loop over the cutoff groups to find the
316	>	// largest cutoff for any atypes present in this group.
317	>	#ifdef IS_MPI
318	>	vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
319	>	groupRowToGtype.resize(nGroupsInRow_);
320	>	for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
321	>	vector<int> atomListRow = getAtomsInGroupRow(cg1);
322	>	for (vector<int>::iterator ia = atomListRow.begin();
323	>	ia != atomListRow.end(); ++ia) {
324	>	int atom1 = (*ia);
325	>	atid = identsRow[atom1];
326	>	if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
327	>	groupCutoffRow[cg1] = atypeCutoff[atid];
328		}
329	+	}
330	+
331	+	bool gTypeFound = false;
332	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
333	+	if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
334	+	groupRowToGtype[cg1] = gt;
335	+	gTypeFound = true;
336	+	}
337	+	}
338	+	if (!gTypeFound) {
339	+	gTypeCutoffs.push_back( groupCutoffRow[cg1] );
340	+	groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
341	+	}
342	+
343	+	}
344	+	vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
345	+	groupColToGtype.resize(nGroupsInCol_);
346	+	for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
347	+	vector<int> atomListCol = getAtomsInGroupColumn(cg2);
348	+	for (vector<int>::iterator jb = atomListCol.begin();
349	+	jb != atomListCol.end(); ++jb) {
350	+	int atom2 = (*jb);
351	+	atid = identsCol[atom2];
352	+	if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
353	+	groupCutoffCol[cg2] = atypeCutoff[atid];
354	+	}
355	+	}
356	+	bool gTypeFound = false;
357	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
358	+	if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
359	+	groupColToGtype[cg2] = gt;
360	+	gTypeFound = true;
361	+	}
362	+	}
363	+	if (!gTypeFound) {
364	+	gTypeCutoffs.push_back( groupCutoffCol[cg2] );
365	+	groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
366	+	}
367	+	}
368	+	#else
369	+
370	+	vector<RealType> groupCutoff(nGroups_, 0.0);
371	+	groupToGtype.resize(nGroups_);
372	+	for (int cg1 = 0; cg1 < nGroups_; cg1++) {
373	+	groupCutoff[cg1] = 0.0;
374	+	vector<int> atomList = getAtomsInGroupRow(cg1);
375	+	for (vector<int>::iterator ia = atomList.begin();
376	+	ia != atomList.end(); ++ia) {
377	+	int atom1 = (*ia);
378	+	atid = idents[atom1];
379	+	if (atypeCutoff[atid] > groupCutoff[cg1])
380	+	groupCutoff[cg1] = atypeCutoff[atid];
381	+	}
382	+
383	+	bool gTypeFound = false;
384	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
385	+	if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
386	+	groupToGtype[cg1] = gt;
387	+	gTypeFound = true;
388	+	}
389	+	}
390	+	if (!gTypeFound) {
391	+	gTypeCutoffs.push_back( groupCutoff[cg1] );
392	+	groupToGtype[cg1] = gTypeCutoffs.size() - 1;
393		}
394		}
395	+	#endif
396	+
397	+	// Now we find the maximum group cutoff value present in the simulation
398	+
399	+	RealType groupMax = *max_element(gTypeCutoffs.begin(),
400	+	gTypeCutoffs.end());
401	+
402	+	#ifdef IS_MPI
403	+	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
404	+	MPI::MAX);
405	+	#endif
406	+
407	+	RealType tradRcut = groupMax;
408	+
409	+	for (int i = 0; i < gTypeCutoffs.size();  i++) {
410	+	for (int j = 0; j < gTypeCutoffs.size();  j++) {
411	+	RealType thisRcut;
412	+	switch(cutoffPolicy_) {
413	+	case TRADITIONAL:
414	+	thisRcut = tradRcut;
415	+	break;
416	+	case MIX:
417	+	thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
418	+	break;
419	+	case MAX:
420	+	thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
421	+	break;
422	+	default:
423	+	sprintf(painCave.errMsg,
424	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
425	+	"hit an unknown cutoff policy!\n");
426	+	painCave.severity = OPENMD_ERROR;
427	+	painCave.isFatal = 1;
428	+	simError();
429	+	break;
430	+	}
431	+
432	+	pair<int,int> key = make_pair(i,j);
433	+	gTypeCutoffMap[key].first = thisRcut;
434	+	if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
435	+	gTypeCutoffMap[key].second = thisRcut*thisRcut;
436	+	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
437	+	// sanity check
438	+
439	+	if (userChoseCutoff_) {
440	+	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
441	+	sprintf(painCave.errMsg,
442	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
443	+	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
444	+	painCave.severity = OPENMD_ERROR;
445	+	painCave.isFatal = 1;
446	+	simError();
447	+	}
448	+	}
449	+	}
450	+	}
451		}
452	<
452	>
453	>
454	>	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
455	>	int i, j;
456	>	#ifdef IS_MPI
457	>	i = groupRowToGtype[cg1];
458	>	j = groupColToGtype[cg2];
459	>	#else
460	>	i = groupToGtype[cg1];
461	>	j = groupToGtype[cg2];
462	>	#endif
463	>	return gTypeCutoffMap[make_pair(i,j)];
464	>	}
465	>
466	>	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
467	>	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
468	>	if (toposForAtom[atom1][j] == atom2)
469	>	return topoDist[atom1][j];
470	>	}
471	>	return 0;
472	>	}
473	>
474	>	void ForceMatrixDecomposition::zeroWorkArrays() {
475	>	pairwisePot = 0.0;
476	>	embeddingPot = 0.0;
477	>
478	>	#ifdef IS_MPI
479	>	if (storageLayout_ & DataStorage::dslForce) {
480	>	fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
481	>	fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
482	>	}
483	>
484	>	if (storageLayout_ & DataStorage::dslTorque) {
485	>	fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
486	>	fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
487	>	}
488	>
489	>	fill(pot_row.begin(), pot_row.end(),
490	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
491	>
492	>	fill(pot_col.begin(), pot_col.end(),
493	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
494	>
495	>	if (storageLayout_ & DataStorage::dslParticlePot) {
496	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
497	>	0.0);
498	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
499	>	0.0);
500	>	}
501	>
502	>	if (storageLayout_ & DataStorage::dslDensity) {
503	>	fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
504	>	fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
505	>	}
506	>
507	>	if (storageLayout_ & DataStorage::dslFunctional) {
508	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
509	>	0.0);
510	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
511	>	0.0);
512	>	}
513	>
514	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
515	>	fill(atomRowData.functionalDerivative.begin(),
516	>	atomRowData.functionalDerivative.end(), 0.0);
517	>	fill(atomColData.functionalDerivative.begin(),
518	>	atomColData.functionalDerivative.end(), 0.0);
519	>	}
520	>
521	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
522	>	fill(atomRowData.skippedCharge.begin(),
523	>	atomRowData.skippedCharge.end(), 0.0);
524	>	fill(atomColData.skippedCharge.begin(),
525	>	atomColData.skippedCharge.end(), 0.0);
526	>	}
527	>
528	>	#endif
529	>	// even in parallel, we need to zero out the local arrays:
530	>
531	>	if (storageLayout_ & DataStorage::dslParticlePot) {
532	>	fill(snap_->atomData.particlePot.begin(),
533	>	snap_->atomData.particlePot.end(), 0.0);
534	>	}
535	>
536	>	if (storageLayout_ & DataStorage::dslDensity) {
537	>	fill(snap_->atomData.density.begin(),
538	>	snap_->atomData.density.end(), 0.0);
539	>	}
540	>
541	>	if (storageLayout_ & DataStorage::dslFunctional) {
542	>	fill(snap_->atomData.functional.begin(),
543	>	snap_->atomData.functional.end(), 0.0);
544	>	}
545	>
546	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
547	>	fill(snap_->atomData.functionalDerivative.begin(),
548	>	snap_->atomData.functionalDerivative.end(), 0.0);
549	>	}
550	>
551	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
552	>	fill(snap_->atomData.skippedCharge.begin(),
553	>	snap_->atomData.skippedCharge.end(), 0.0);
554	>	}
555	>	}
556	>
557	>
558		void ForceMatrixDecomposition::distributeData()  {
559		snap_ = sman_->getCurrentSnapshot();
560		storageLayout_ = sman_->getStorageLayout();
561		#ifdef IS_MPI
562
563		// gather up the atomic positions
564	<	AtomCommVectorRow->gather(snap_->atomData.position,
564	>	AtomPlanVectorRow->gather(snap_->atomData.position,
565		atomRowData.position);
566	<	AtomCommVectorColumn->gather(snap_->atomData.position,
566	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
567		atomColData.position);
568
569		// gather up the cutoff group positions
570	<	cgCommVectorRow->gather(snap_->cgData.position,
570	>
571	>	cgPlanVectorRow->gather(snap_->cgData.position,
572		cgRowData.position);
573	<	cgCommVectorColumn->gather(snap_->cgData.position,
573	>
574	>	cgPlanVectorColumn->gather(snap_->cgData.position,
575		cgColData.position);
576	+
577
578		// if needed, gather the atomic rotation matrices
579		if (storageLayout_ & DataStorage::dslAmat) {
580	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
580	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
581		atomRowData.aMat);
582	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
582	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
583		atomColData.aMat);
584		}
585
586		// if needed, gather the atomic eletrostatic frames
587		if (storageLayout_ & DataStorage::dslElectroFrame) {
588	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
588	>	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
589		atomRowData.electroFrame);
590	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
590	>	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
591		atomColData.electroFrame);
592		}
593	+
594		#endif
595		}
596
597	+	/* collects information obtained during the pre-pair loop onto local
598	+	* data structures.
599	+	*/
600		void ForceMatrixDecomposition::collectIntermediateData() {
601		snap_ = sman_->getCurrentSnapshot();
602		storageLayout_ = sman_->getStorageLayout();
#	Line 273 | Line 604 | namespace OpenMD {
604
605		if (storageLayout_ & DataStorage::dslDensity) {
606
607	<	AtomCommRealRow->scatter(atomRowData.density,
607	>	AtomPlanRealRow->scatter(atomRowData.density,
608		snap_->atomData.density);
609
610		int n = snap_->atomData.density.size();
611	<	std::vector<RealType> rho_tmp(n, 0.0);
612	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
611	>	vector<RealType> rho_tmp(n, 0.0);
612	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
613		for (int i = 0; i < n; i++)
614		snap_->atomData.density[i] += rho_tmp[i];
615		}
616		#endif
617		}
618	<
618	>
619	>	/*
620	>	* redistributes information obtained during the pre-pair loop out to
621	>	* row and column-indexed data structures
622	>	*/
623		void ForceMatrixDecomposition::distributeIntermediateData() {
624		snap_ = sman_->getCurrentSnapshot();
625		storageLayout_ = sman_->getStorageLayout();
626		#ifdef IS_MPI
627		if (storageLayout_ & DataStorage::dslFunctional) {
628	<	AtomCommRealRow->gather(snap_->atomData.functional,
628	>	AtomPlanRealRow->gather(snap_->atomData.functional,
629		atomRowData.functional);
630	<	AtomCommRealColumn->gather(snap_->atomData.functional,
630	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
631		atomColData.functional);
632		}
633
634		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
635	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
635	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
636		atomRowData.functionalDerivative);
637	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
637	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
638		atomColData.functionalDerivative);
639		}
640		#endif
#	Line 313 | Line 648 | namespace OpenMD {
648		int n = snap_->atomData.force.size();
649		vector<Vector3d> frc_tmp(n, V3Zero);
650
651	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
651	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
652		for (int i = 0; i < n; i++) {
653		snap_->atomData.force[i] += frc_tmp[i];
654		frc_tmp[i] = 0.0;
655		}
656
657	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
658	<	for (int i = 0; i < n; i++)
657	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
658	>	for (int i = 0; i < n; i++) {
659		snap_->atomData.force[i] += frc_tmp[i];
660	<
661	<
660	>	}
661	>
662		if (storageLayout_ & DataStorage::dslTorque) {
663
664	<	int nt = snap_->atomData.force.size();
664	>	int nt = snap_->atomData.torque.size();
665		vector<Vector3d> trq_tmp(nt, V3Zero);
666
667	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
668	<	for (int i = 0; i < n; i++) {
667	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
668	>	for (int i = 0; i < nt; i++) {
669		snap_->atomData.torque[i] += trq_tmp[i];
670		trq_tmp[i] = 0.0;
671		}
672
673	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
674	<	for (int i = 0; i < n; i++)
673	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
674	>	for (int i = 0; i < nt; i++)
675		snap_->atomData.torque[i] += trq_tmp[i];
676		}
677	+
678	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
679	+
680	+	int ns = snap_->atomData.skippedCharge.size();
681	+	vector<RealType> skch_tmp(ns, 0.0);
682	+
683	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
684	+	for (int i = 0; i < ns; i++) {
685	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
686	+	skch_tmp[i] = 0.0;
687	+	}
688	+
689	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
690	+	for (int i = 0; i < ns; i++)
691	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
692	+
693	+	}
694
695		nLocal_ = snap_->getNumberOfAtoms();
696
697	<	vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
698	<	vector<RealType> (nLocal_, 0.0));
697	>	vector<potVec> pot_temp(nLocal_,
698	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
699	>
700	>	// scatter/gather pot_row into the members of my column
701	>
702	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
703	>
704	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
705	>	pairwisePot += pot_temp[ii];
706
707	<	for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
708	<	AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
709	<	for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
710	<	pot_local[i] += pot_temp[i][ii];
711	<	}
707	>	fill(pot_temp.begin(), pot_temp.end(),
708	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
709	>
710	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
711	>
712	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
713	>	pairwisePot += pot_temp[ii];
714	>
715	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
716	>	RealType ploc1 = pairwisePot[ii];
717	>	RealType ploc2 = 0.0;
718	>	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
719	>	pairwisePot[ii] = ploc2;
720		}
721	+
722	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
723	+	RealType ploc1 = embeddingPot[ii];
724	+	RealType ploc2 = 0.0;
725	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
726	+	embeddingPot[ii] = ploc2;
727	+	}
728	+
729		#endif
730	+
731		}
732
733		int ForceMatrixDecomposition::getNAtomsInRow() {
#	Line 426 | Line 802 | namespace OpenMD {
802		#ifdef IS_MPI
803		return massFactorsRow[atom1];
804		#else
805	<	return massFactorsLocal[atom1];
805	>	return massFactors[atom1];
806		#endif
807		}
808
#	Line 434 | Line 810 | namespace OpenMD {
810		#ifdef IS_MPI
811		return massFactorsCol[atom2];
812		#else
813	<	return massFactorsLocal[atom2];
813	>	return massFactors[atom2];
814		#endif
815
816		}
#	Line 452 | Line 828 | namespace OpenMD {
828		return d;
829		}
830
831	<	vector<int> ForceMatrixDecomposition::getSkipsForRowAtom(int atom1) {
832	<	#ifdef IS_MPI
457	<	return skipsForRowAtom[atom1];
458	<	#else
459	<	return skipsForLocalAtom[atom1];
460	<	#endif
831	>	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
832	>	return excludesForAtom[atom1];
833		}
834
835		/**
836	<	* there are a number of reasons to skip a pair or a particle mostly
837	<	* 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.
836	>	* We need to exclude some overcounted interactions that result from
837	>	* the parallel decomposition.
838		*/
839		bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
840		int unique_id_1, unique_id_2;
841	<
841	>
842		#ifdef IS_MPI
843		// in MPI, we have to look up the unique IDs for each atom
844		unique_id_1 = AtomRowToGlobal[atom1];
845		unique_id_2 = AtomColToGlobal[atom2];
846	+	#else
847	+	unique_id_1 = AtomLocalToGlobal[atom1];
848	+	unique_id_2 = AtomLocalToGlobal[atom2];
849	+	#endif
850
478	–	// this situation should only arise in MPI simulations
851		if (unique_id_1 == unique_id_2) return true;
852	<
852	>
853	>	#ifdef IS_MPI
854		// this prevents us from doing the pair on multiple processors
855		if (unique_id_1 < unique_id_2) {
856		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
857		} else {
858	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
858	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
859		}
487	–	#else
488	–	// in the normal loop, the atom numbers are unique
489	–	unique_id_1 = atom1;
490	–	unique_id_2 = atom2;
860		#endif
861
862	<	#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
862	>	return false;
863		}
864
865	<	int ForceMatrixDecomposition::getTopoDistance(int atom1, int atom2) {
865	>	/**
866	>	* We need to handle the interactions for atoms who are involved in
867	>	* the same rigid body as well as some short range interactions
868	>	* (bonds, bends, torsions) differently from other interactions.
869	>	* We'll still visit the pairwise routines, but with a flag that
870	>	* tells those routines to exclude the pair from direct long range
871	>	* interactions.  Some indirect interactions (notably reaction
872	>	* field) must still be handled for these pairs.
873	>	*/
874	>	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
875	>
876	>	// excludesForAtom was constructed to use row/column indices in the MPI
877	>	// version, and to use local IDs in the non-MPI version:
878
879	<	#ifdef IS_MPI
880	<	for (int i = 0; i < toposForRowAtom[atom1].size(); i++) {
881	<	if ( toposForRowAtom[atom1][i] == atom2 ) return topoDistRow[atom1][i];
879	>	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
880	>	i != excludesForAtom[atom1].end(); ++i) {
881	>	if ( (*i) == atom2 ) return true;
882		}
512	–	#else
513	–	for (int i = 0; i < toposForLocalAtom[atom1].size(); i++) {
514	–	if ( toposForLocalAtom[atom1][i] == atom2 ) return topoDistLocal[atom1][i];
515	–	}
516	–	#endif
883
884	<	// zero is default for unconnected (i.e. normal) pair interactions
519	<	return 0;
884	>	return false;
885		}
886
887	+
888		void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
889		#ifdef IS_MPI
890		atomRowData.force[atom1] += fg;
#	Line 536 | Line 902 | namespace OpenMD {
902		}
903
904		// filling interaction blocks with pointers
905	<	InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {
906	<	InteractionData idat;
905	>	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
906	>	int atom1, int atom2) {
907
908	+	idat.excluded = excludeAtomPair(atom1, atom2);
909	+
910		#ifdef IS_MPI
911	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
912	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
913	+	//                         ff_->getAtomType(identsCol[atom2]) );
914	+
915		if (storageLayout_ & DataStorage::dslAmat) {
916		idat.A1 = &(atomRowData.aMat[atom1]);
917		idat.A2 = &(atomColData.aMat[atom2]);
#	Line 560 | Line 932 | namespace OpenMD {
932		idat.rho2 = &(atomColData.density[atom2]);
933		}
934
935	+	if (storageLayout_ & DataStorage::dslFunctional) {
936	+	idat.frho1 = &(atomRowData.functional[atom1]);
937	+	idat.frho2 = &(atomColData.functional[atom2]);
938	+	}
939	+
940		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
941		idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
942		idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
943		}
944
945	+	if (storageLayout_ & DataStorage::dslParticlePot) {
946	+	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
947	+	idat.particlePot2 = &(atomColData.particlePot[atom2]);
948	+	}
949	+
950	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
951	+	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
952	+	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
953	+	}
954	+
955		#else
956	+
957	+
958	+	// cerr << "atoms = " << atom1 << " " << atom2 << "\n";
959	+	// cerr << "pos1 = " << snap_->atomData.position[atom1] << "\n";
960	+	// cerr << "pos2 = " << snap_->atomData.position[atom2] << "\n";
961	+
962	+	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
963	+	//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
964	+	//                         ff_->getAtomType(idents[atom2]) );
965	+
966		if (storageLayout_ & DataStorage::dslAmat) {
967		idat.A1 = &(snap_->atomData.aMat[atom1]);
968		idat.A2 = &(snap_->atomData.aMat[atom2]);
#	Line 581 | Line 978 | namespace OpenMD {
978		idat.t2 = &(snap_->atomData.torque[atom2]);
979		}
980
981	<	if (storageLayout_ & DataStorage::dslDensity) {
981	>	if (storageLayout_ & DataStorage::dslDensity) {
982		idat.rho1 = &(snap_->atomData.density[atom1]);
983		idat.rho2 = &(snap_->atomData.density[atom2]);
984		}
985
986	+	if (storageLayout_ & DataStorage::dslFunctional) {
987	+	idat.frho1 = &(snap_->atomData.functional[atom1]);
988	+	idat.frho2 = &(snap_->atomData.functional[atom2]);
989	+	}
990	+
991		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
992		idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
993		idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
994		}
995	+
996	+	if (storageLayout_ & DataStorage::dslParticlePot) {
997	+	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
998	+	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
999	+	}
1000	+
1001	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1002	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1003	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1004	+	}
1005		#endif
594	–	return idat;
1006		}
1007
1008	<	InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
1009	<
599	<	InteractionData idat;
1008	>
1009	>	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1010		#ifdef IS_MPI
1011	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1012	<	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1013	<	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1014	<	}
1015	<	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	<	}
1011	>	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1012	>	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1013	>
1014	>	atomRowData.force[atom1] += *(idat.f1);
1015	>	atomColData.force[atom2] -= *(idat.f1);
1016		#else
1017	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1018	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1019	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1020	<	}
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	<	}
1017	>	pairwisePot += *(idat.pot);
1018	>
1019	>	snap_->atomData.force[atom1] += *(idat.f1);
1020	>	snap_->atomData.force[atom2] -= *(idat.f1);
1021		#endif
1022
1023		}
1024
630	–
631	–
632	–
1025		/*
1026		* buildNeighborList
1027		*
#	Line 639 | Line 1031 | namespace OpenMD {
1031		vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1032
1033		vector<pair<int, int> > neighborList;
1034	+	groupCutoffs cuts;
1035	+	bool doAllPairs = false;
1036	+
1037		#ifdef IS_MPI
1038		cellListRow_.clear();
1039		cellListCol_.clear();
#	Line 646 | Line 1041 | namespace OpenMD {
1041		cellList_.clear();
1042		#endif
1043
1044	<	// dangerous to not do error checking.
650	<	RealType rCut_;
651	<
652	<	RealType rList_ = (rCut_ + skinThickness_);
1044	>	RealType rList_ = (largestRcut_ + skinThickness_);
1045		RealType rl2 = rList_ * rList_;
1046		Snapshot* snap_ = sman_->getCurrentSnapshot();
1047		Mat3x3d Hmat = snap_->getHmat();
#	Line 661 | Line 1053 | namespace OpenMD {
1053		nCells_.y() = (int) ( Hy.length() )/ rList_;
1054		nCells_.z() = (int) ( Hz.length() )/ rList_;
1055
1056	+	// handle small boxes where the cell offsets can end up repeating cells
1057	+
1058	+	if (nCells_.x() < 3) doAllPairs = true;
1059	+	if (nCells_.y() < 3) doAllPairs = true;
1060	+	if (nCells_.z() < 3) doAllPairs = true;
1061	+
1062		Mat3x3d invHmat = snap_->getInvHmat();
1063		Vector3d rs, scaled, dr;
1064		Vector3i whichCell;
1065		int cellIndex;
1066	+	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1067
1068		#ifdef IS_MPI
1069	<	for (int i = 0; i < nGroupsInRow_; i++) {
1070	<	rs = cgRowData.position[i];
1071	<	// scaled positions relative to the box vectors
1072	<	scaled = invHmat * rs;
1073	<	// 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();
683	<
684	<	// find single index of this cell:
685	<	cellIndex = Vlinear(whichCell, nCells_);
686	<	// add this cutoff group to the list of groups in this cell;
687	<	cellListRow_[cellIndex].push_back(i);
688	<	}
1069	>	cellListRow_.resize(nCtot);
1070	>	cellListCol_.resize(nCtot);
1071	>	#else
1072	>	cellList_.resize(nCtot);
1073	>	#endif
1074
1075	<	for (int i = 0; i < nGroupsInCol_; i++) {
1076	<	rs = cgColData.position[i];
692	<	// scaled positions relative to the box vectors
693	<	scaled = invHmat * rs;
694	<	// wrap the vector back into the unit box by subtracting integer box
695	<	// numbers
696	<	for (int j = 0; j < 3; j++)
697	<	scaled[j] -= roundMe(scaled[j]);
1075	>	if (!doAllPairs) {
1076	>	#ifdef IS_MPI
1077
1078	<	// find xyz-indices of cell that cutoffGroup is in.
1079	<	whichCell.x() = nCells_.x() * scaled.x();
1080	<	whichCell.y() = nCells_.y() * scaled.y();
1081	<	whichCell.z() = nCells_.z() * scaled.z();
1082	<
1083	<	// find single index of this cell:
1084	<	cellIndex = Vlinear(whichCell, nCells_);
1085	<	// add this cutoff group to the list of groups in this cell;
1086	<	cellListCol_[cellIndex].push_back(i);
1087	<	}
1078	>	for (int i = 0; i < nGroupsInRow_; i++) {
1079	>	rs = cgRowData.position[i];
1080	>
1081	>	// scaled positions relative to the box vectors
1082	>	scaled = invHmat * rs;
1083	>
1084	>	// wrap the vector back into the unit box by subtracting integer box
1085	>	// numbers
1086	>	for (int j = 0; j < 3; j++) {
1087	>	scaled[j] -= roundMe(scaled[j]);
1088	>	scaled[j] += 0.5;
1089	>	}
1090	>
1091	>	// find xyz-indices of cell that cutoffGroup is in.
1092	>	whichCell.x() = nCells_.x() * scaled.x();
1093	>	whichCell.y() = nCells_.y() * scaled.y();
1094	>	whichCell.z() = nCells_.z() * scaled.z();
1095	>
1096	>	// find single index of this cell:
1097	>	cellIndex = Vlinear(whichCell, nCells_);
1098	>
1099	>	// add this cutoff group to the list of groups in this cell;
1100	>	cellListRow_[cellIndex].push_back(i);
1101	>	}
1102	>	for (int i = 0; i < nGroupsInCol_; i++) {
1103	>	rs = cgColData.position[i];
1104	>
1105	>	// scaled positions relative to the box vectors
1106	>	scaled = invHmat * rs;
1107	>
1108	>	// wrap the vector back into the unit box by subtracting integer box
1109	>	// numbers
1110	>	for (int j = 0; j < 3; j++) {
1111	>	scaled[j] -= roundMe(scaled[j]);
1112	>	scaled[j] += 0.5;
1113	>	}
1114	>
1115	>	// find xyz-indices of cell that cutoffGroup is in.
1116	>	whichCell.x() = nCells_.x() * scaled.x();
1117	>	whichCell.y() = nCells_.y() * scaled.y();
1118	>	whichCell.z() = nCells_.z() * scaled.z();
1119	>
1120	>	// find single index of this cell:
1121	>	cellIndex = Vlinear(whichCell, nCells_);
1122	>
1123	>	// add this cutoff group to the list of groups in this cell;
1124	>	cellListCol_[cellIndex].push_back(i);
1125	>	}
1126	>
1127		#else
1128	<	for (int i = 0; i < nGroups_; i++) {
1129	<	rs = snap_->cgData.position[i];
1130	<	// scaled positions relative to the box vectors
1131	<	scaled = invHmat * rs;
1132	<	// wrap the vector back into the unit box by subtracting integer box
1133	<	// numbers
1134	<	for (int j = 0; j < 3; j++)
1135	<	scaled[j] -= roundMe(scaled[j]);
1128	>	for (int i = 0; i < nGroups_; i++) {
1129	>	rs = snap_->cgData.position[i];
1130	>
1131	>	// scaled positions relative to the box vectors
1132	>	scaled = invHmat * rs;
1133	>
1134	>	// wrap the vector back into the unit box by subtracting integer box
1135	>	// numbers
1136	>	for (int j = 0; j < 3; j++) {
1137	>	scaled[j] -= roundMe(scaled[j]);
1138	>	scaled[j] += 0.5;
1139	>	}
1140	>
1141	>	// find xyz-indices of cell that cutoffGroup is in.
1142	>	whichCell.x() = nCells_.x() * scaled.x();
1143	>	whichCell.y() = nCells_.y() * scaled.y();
1144	>	whichCell.z() = nCells_.z() * scaled.z();
1145	>
1146	>	// find single index of this cell:
1147	>	cellIndex = Vlinear(whichCell, nCells_);
1148	>
1149	>	// add this cutoff group to the list of groups in this cell;
1150	>	cellList_[cellIndex].push_back(i);
1151	>	}
1152
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	–	}
1153		#endif
1154
1155	<
1156	<
1157	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1158	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1159	<	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) {
1155	>	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1156	>	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1157	>	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1158	>	Vector3i m1v(m1x, m1y, m1z);
1159	>	int m1 = Vlinear(m1v, nCells_);
1160
1161	<	Vector3i m2v = m1v + (*os);
1162	<
1163	<	if (m2v.x() >= nCells_.x()) {
1164	<	m2v.x() = 0;
1165	<	} 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_);
1161	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1162	>	os != cellOffsets_.end(); ++os) {
1163	>
1164	>	Vector3i m2v = m1v + (*os);
1165	>
1166
1167	<	#ifdef IS_MPI
1168	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1169	<	j1 != cellListRow_[m1].end(); ++j1) {
1170	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1171	<	j2 != cellListCol_[m2].end(); ++j2) {
1172	<
1173	<	// Always do this if we're in different cells or if
1174	<	// we're in the same cell and the global index of the
1175	<	// j2 cutoff group is less than the j1 cutoff group
1167	>	if (m2v.x() >= nCells_.x()) {
1168	>	m2v.x() = 0;
1169	>	} else if (m2v.x() < 0) {
1170	>	m2v.x() = nCells_.x() - 1;
1171	>	}
1172	>
1173	>	if (m2v.y() >= nCells_.y()) {
1174	>	m2v.y() = 0;
1175	>	} else if (m2v.y() < 0) {
1176	>	m2v.y() = nCells_.y() - 1;
1177	>	}
1178	>
1179	>	if (m2v.z() >= nCells_.z()) {
1180	>	m2v.z() = 0;
1181	>	} else if (m2v.z() < 0) {
1182	>	m2v.z() = nCells_.z() - 1;
1183	>	}
1184
1185	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1185	>	int m2 = Vlinear (m2v, nCells_);
1186	>
1187	>	#ifdef IS_MPI
1188	>	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1189	>	j1 != cellListRow_[m1].end(); ++j1) {
1190	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1191	>	j2 != cellListCol_[m2].end(); ++j2) {
1192	>
1193	>	// In parallel, we need to visit *all* pairs of row
1194	>	// & column indicies and will divide labor in the
1195	>	// force evaluation later.
1196		dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1197		snap_->wrapVector(dr);
1198	<	if (dr.lengthSquare() < rl2) {
1198	>	cuts = getGroupCutoffs( (*j1), (*j2) );
1199	>	if (dr.lengthSquare() < cuts.third) {
1200		neighborList.push_back(make_pair((*j1), (*j2)));
1201	<	}
1201	>	}
1202		}
1203		}
782	–	}
1204		#else
1205	<	for (vector<int>::iterator j1 = cellList_[m1].begin();
1206	<	j1 != cellList_[m1].end(); ++j1) {
1207	<	for (vector<int>::iterator j2 = cellList_[m2].begin();
1208	<	j2 != cellList_[m2].end(); ++j2) {
1209	<
1210	<	// Always do this if we're in different cells or if
1211	<	// we're in the same cell and the global index of the
1212	<	// j2 cutoff group is less than the j1 cutoff group
1205	>	for (vector<int>::iterator j1 = cellList_[m1].begin();
1206	>	j1 != cellList_[m1].end(); ++j1) {
1207	>	for (vector<int>::iterator j2 = cellList_[m2].begin();
1208	>	j2 != cellList_[m2].end(); ++j2) {
1209	>
1210	>	// Always do this if we're in different cells or if
1211	>	// we're in the same cell and the global index of
1212	>	// the j2 cutoff group is greater than or equal to
1213	>	// the j1 cutoff group.  Note that Rappaport's code
1214	>	// has a "less than" conditional here, but that
1215	>	// deals with atom-by-atom computation.  OpenMD
1216	>	// allows atoms within a single cutoff group to
1217	>	// interact with each other.
1218
1219	<	if (m2 != m1 || (*j2) < (*j1)) {
1220	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1221	<	snap_->wrapVector(dr);
1222	<	if (dr.lengthSquare() < rl2) {
1223	<	neighborList.push_back(make_pair((*j1), (*j2)));
1219	>
1220	>
1221	>	if (m2 != m1 || (*j2) >= (*j1) ) {
1222	>
1223	>	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1224	>	snap_->wrapVector(dr);
1225	>	cuts = getGroupCutoffs( (*j1), (*j2) );
1226	>	if (dr.lengthSquare() < cuts.third) {
1227	>	neighborList.push_back(make_pair((*j1), (*j2)));
1228	>	}
1229		}
1230		}
1231		}
801	–	}
1232		#endif
1233	+	}
1234		}
1235		}
1236		}
1237	+	} else {
1238	+	// branch to do all cutoff group pairs
1239	+	#ifdef IS_MPI
1240	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1241	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1242	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1243	+	snap_->wrapVector(dr);
1244	+	cuts = getGroupCutoffs( j1, j2 );
1245	+	if (dr.lengthSquare() < cuts.third) {
1246	+	neighborList.push_back(make_pair(j1, j2));
1247	+	}
1248	+	}
1249	+	}
1250	+	#else
1251	+	// include all groups here.
1252	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1253	+	// include self group interactions j2 == j1
1254	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1255	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1256	+	snap_->wrapVector(dr);
1257	+	cuts = getGroupCutoffs( j1, j2 );
1258	+	if (dr.lengthSquare() < cuts.third) {
1259	+	neighborList.push_back(make_pair(j1, j2));
1260	+	}
1261	+	}
1262	+	}
1263	+	#endif
1264		}
1265	<
1265	>
1266		// save the local cutoff group positions for the check that is
1267		// done on each loop:
1268		saved_CG_positions_.clear();
1269		for (int i = 0; i < nGroups_; i++)
1270		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1271	<
1271	>
1272		return neighborList;
1273		}
1274		} //end namespace OpenMD

Diff Legend

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