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 1575 by gezelter, Fri Jun 3 21:39:49 2011 UTC vs.
trunk/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 2061 by gezelter, Tue Mar 3 16:24:44 2015 UTC

#	Line 35 | Line 35
35		*
36		* [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37		* [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38	<	* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
39	<	* [4]  Vardeman & Gezelter, in progress (2009).
38	>	* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008).
39	>	* [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40	>	* [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41		*/
42		#include "parallel/ForceMatrixDecomposition.hpp"
43		#include "math/SquareMatrix3.hpp"
#	Line 47 | Line 48 | namespace OpenMD {
48		using namespace std;
49		namespace OpenMD {
50
51	+	ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52	+
53	+	// Row and colum scans must visit all surrounding cells
54	+	cellOffsets_.clear();
55	+	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
56	+	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
57	+	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
58	+	cellOffsets_.push_back( Vector3i(-1, 0,-1) );
59	+	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
60	+	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
61	+	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
62	+	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
63	+	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
64	+	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
65	+	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
66	+	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
67	+	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
68	+	cellOffsets_.push_back( Vector3i( 0, 0, 0) );
69	+	cellOffsets_.push_back( Vector3i( 1, 0, 0) );
70	+	cellOffsets_.push_back( Vector3i(-1, 1, 0) );
71	+	cellOffsets_.push_back( Vector3i( 0, 1, 0) );
72	+	cellOffsets_.push_back( Vector3i( 1, 1, 0) );
73	+	cellOffsets_.push_back( Vector3i(-1,-1, 1) );
74	+	cellOffsets_.push_back( Vector3i( 0,-1, 1) );
75	+	cellOffsets_.push_back( Vector3i( 1,-1, 1) );
76	+	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
77	+	cellOffsets_.push_back( Vector3i( 0, 0, 1) );
78	+	cellOffsets_.push_back( Vector3i( 1, 0, 1) );
79	+	cellOffsets_.push_back( Vector3i(-1, 1, 1) );
80	+	cellOffsets_.push_back( Vector3i( 0, 1, 1) );
81	+	cellOffsets_.push_back( Vector3i( 1, 1, 1) );
82	+	}
83	+
84	+
85		/**
86		* distributeInitialData is essentially a copy of the older fortran
87		* SimulationSetup
88		*/
54	–
89		void ForceMatrixDecomposition::distributeInitialData() {
90		snap_ = sman_->getCurrentSnapshot();
91		storageLayout_ = sman_->getStorageLayout();
92		ff_ = info_->getForceField();
93		nLocal_ = snap_->getNumberOfAtoms();
94	<	nGroups_ = snap_->getNumberOfCutoffGroups();
95	<
94	>
95	>	nGroups_ = info_->getNLocalCutoffGroups();
96		// gather the information for atomtype IDs (atids):
97	<	identsLocal = info_->getIdentArray();
97	>	idents = info_->getIdentArray();
98	>	regions = info_->getRegions();
99		AtomLocalToGlobal = info_->getGlobalAtomIndices();
100		cgLocalToGlobal = info_->getGlobalGroupIndices();
101		vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
67	–	vector<RealType> massFactorsLocal = info_->getMassFactors();
68	–	PairList excludes = info_->getExcludedInteractions();
69	–	PairList oneTwo = info_->getOneTwoInteractions();
70	–	PairList oneThree = info_->getOneThreeInteractions();
71	–	PairList oneFour = info_->getOneFourInteractions();
102
103	+	massFactors = info_->getMassFactors();
104	+
105	+	PairList* excludes = info_->getExcludedInteractions();
106	+	PairList* oneTwo = info_->getOneTwoInteractions();
107	+	PairList* oneThree = info_->getOneThreeInteractions();
108	+	PairList* oneFour = info_->getOneFourInteractions();
109	+
110	+	if (needVelocities_)
111	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
112	+	DataStorage::dslVelocity);
113	+	else
114	+	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
115	+
116		#ifdef IS_MPI
117
118	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
119	<	AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
77	<	AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
78	<	AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
79	<	AtomCommPotRow = new Communicator<Row,potVec>(nLocal_);
118	>	MPI_Comm row = rowComm.getComm();
119	>	MPI_Comm col = colComm.getComm();
120
121	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
122	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
123	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
124	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
125	<	AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
121	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
122	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
123	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
124	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
125	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
126
127	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
128	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
129	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
130	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
127	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
128	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
129	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
130	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
131	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
132
133	<	nAtomsInRow_ = AtomCommIntRow->getSize();
134	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
135	<	nGroupsInRow_ = cgCommIntRow->getSize();
136	<	nGroupsInCol_ = cgCommIntColumn->getSize();
133	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
134	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
135	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
136	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
137
138	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
139	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
140	+	nGroupsInRow_ = cgPlanIntRow->getSize();
141	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
142	+
143		// Modify the data storage objects with the correct layouts and sizes:
144		atomRowData.resize(nAtomsInRow_);
145		atomRowData.setStorageLayout(storageLayout_);
#	Line 102 | Line 148 | namespace OpenMD {
148		cgRowData.resize(nGroupsInRow_);
149		cgRowData.setStorageLayout(DataStorage::dslPosition);
150		cgColData.resize(nGroupsInCol_);
151	<	cgColData.setStorageLayout(DataStorage::dslPosition);
152	<
153	<	identsRow.reserve(nAtomsInRow_);
154	<	identsCol.reserve(nAtomsInCol_);
151	>	if (needVelocities_)
152	>	// we only need column velocities if we need them.
153	>	cgColData.setStorageLayout(DataStorage::dslPosition |
154	>	DataStorage::dslVelocity);
155	>	else
156	>	cgColData.setStorageLayout(DataStorage::dslPosition);
157	>
158	>	identsRow.resize(nAtomsInRow_);
159	>	identsCol.resize(nAtomsInCol_);
160
161	<	AtomCommIntRow->gather(identsLocal, identsRow);
162	<	AtomCommIntColumn->gather(identsLocal, identsCol);
161	>	AtomPlanIntRow->gather(idents, identsRow);
162	>	AtomPlanIntColumn->gather(idents, identsCol);
163	>
164	>	regionsRow.resize(nAtomsInRow_);
165	>	regionsCol.resize(nAtomsInCol_);
166
167	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
168	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
167	>	AtomPlanIntRow->gather(regions, regionsRow);
168	>	AtomPlanIntColumn->gather(regions, regionsCol);
169
170	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
171	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
170	>	// allocate memory for the parallel objects
171	>	atypesRow.resize(nAtomsInRow_);
172	>	atypesCol.resize(nAtomsInCol_);
173
174	<	AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
175	<	AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
174	>	for (int i = 0; i < nAtomsInRow_; i++)
175	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
176	>	for (int i = 0; i < nAtomsInCol_; i++)
177	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
178
179	+	pot_row.resize(nAtomsInRow_);
180	+	pot_col.resize(nAtomsInCol_);
181	+
182	+	expot_row.resize(nAtomsInRow_);
183	+	expot_col.resize(nAtomsInCol_);
184	+
185	+	AtomRowToGlobal.resize(nAtomsInRow_);
186	+	AtomColToGlobal.resize(nAtomsInCol_);
187	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
188	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
189	+
190	+	cgRowToGlobal.resize(nGroupsInRow_);
191	+	cgColToGlobal.resize(nGroupsInCol_);
192	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
193	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
194	+
195	+	massFactorsRow.resize(nAtomsInRow_);
196	+	massFactorsCol.resize(nAtomsInCol_);
197	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
198	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
199	+
200		groupListRow_.clear();
201	<	groupListRow_.reserve(nGroupsInRow_);
201	>	groupListRow_.resize(nGroupsInRow_);
202		for (int i = 0; i < nGroupsInRow_; i++) {
203		int gid = cgRowToGlobal[i];
204		for (int j = 0; j < nAtomsInRow_; j++) {
#	Line 131 | Line 209 | namespace OpenMD {
209		}
210
211		groupListCol_.clear();
212	<	groupListCol_.reserve(nGroupsInCol_);
212	>	groupListCol_.resize(nGroupsInCol_);
213		for (int i = 0; i < nGroupsInCol_; i++) {
214		int gid = cgColToGlobal[i];
215		for (int j = 0; j < nAtomsInCol_; j++) {
#	Line 141 | Line 219 | namespace OpenMD {
219		}
220		}
221
222	<	skipsForRowAtom.clear();
223	<	skipsForRowAtom.reserve(nAtomsInRow_);
222	>	excludesForAtom.clear();
223	>	excludesForAtom.resize(nAtomsInRow_);
224	>	toposForAtom.clear();
225	>	toposForAtom.resize(nAtomsInRow_);
226	>	topoDist.clear();
227	>	topoDist.resize(nAtomsInRow_);
228		for (int i = 0; i < nAtomsInRow_; i++) {
229		int iglob = AtomRowToGlobal[i];
230	+
231		for (int j = 0; j < nAtomsInCol_; j++) {
232	<	int jglob = AtomColToGlobal[j];
233	<	if (excludes.hasPair(iglob, jglob))
234	<	skipsForRowAtom[i].push_back(j);
232	>	int jglob = AtomColToGlobal[j];
233	>
234	>	if (excludes->hasPair(iglob, jglob))
235	>	excludesForAtom[i].push_back(j);
236	>
237	>	if (oneTwo->hasPair(iglob, jglob)) {
238	>	toposForAtom[i].push_back(j);
239	>	topoDist[i].push_back(1);
240	>	} else {
241	>	if (oneThree->hasPair(iglob, jglob)) {
242	>	toposForAtom[i].push_back(j);
243	>	topoDist[i].push_back(2);
244	>	} else {
245	>	if (oneFour->hasPair(iglob, jglob)) {
246	>	toposForAtom[i].push_back(j);
247	>	topoDist[i].push_back(3);
248	>	}
249	>	}
250	>	}
251		}
252		}
253
254	<	toposForRowAtom.clear();
255	<	toposForRowAtom.reserve(nAtomsInRow_);
256	<	for (int i = 0; i < nAtomsInRow_; i++) {
257	<	int iglob = AtomRowToGlobal[i];
258	<	int nTopos = 0;
259	<	for (int j = 0; j < nAtomsInCol_; j++) {
260	<	int jglob = AtomColToGlobal[j];
261	<	if (oneTwo.hasPair(iglob, jglob)) {
262	<	toposForRowAtom[i].push_back(j);
263	<	topoDistRow[i][nTopos] = 1;
264	<	nTopos++;
265	<	}
266	<	if (oneThree.hasPair(iglob, jglob)) {
267	<	toposForRowAtom[i].push_back(j);
268	<	topoDistRow[i][nTopos] = 2;
269	<	nTopos++;
270	<	}
271	<	if (oneFour.hasPair(iglob, jglob)) {
272	<	toposForRowAtom[i].push_back(j);
273	<	topoDistRow[i][nTopos] = 3;
274	<	nTopos++;
254	>	#else
255	>	excludesForAtom.clear();
256	>	excludesForAtom.resize(nLocal_);
257	>	toposForAtom.clear();
258	>	toposForAtom.resize(nLocal_);
259	>	topoDist.clear();
260	>	topoDist.resize(nLocal_);
261	>
262	>	for (int i = 0; i < nLocal_; i++) {
263	>	int iglob = AtomLocalToGlobal[i];
264	>
265	>	for (int j = 0; j < nLocal_; j++) {
266	>	int jglob = AtomLocalToGlobal[j];
267	>
268	>	if (excludes->hasPair(iglob, jglob))
269	>	excludesForAtom[i].push_back(j);
270	>
271	>	if (oneTwo->hasPair(iglob, jglob)) {
272	>	toposForAtom[i].push_back(j);
273	>	topoDist[i].push_back(1);
274	>	} else {
275	>	if (oneThree->hasPair(iglob, jglob)) {
276	>	toposForAtom[i].push_back(j);
277	>	topoDist[i].push_back(2);
278	>	} else {
279	>	if (oneFour->hasPair(iglob, jglob)) {
280	>	toposForAtom[i].push_back(j);
281	>	topoDist[i].push_back(3);
282	>	}
283	>	}
284		}
285		}
286		}
179	–
287		#endif
288
289	+	// allocate memory for the parallel objects
290	+	atypesLocal.resize(nLocal_);
291	+
292	+	for (int i = 0; i < nLocal_; i++)
293	+	atypesLocal[i] = ff_->getAtomType(idents[i]);
294	+
295		groupList_.clear();
296	<	groupList_.reserve(nGroups_);
296	>	groupList_.resize(nGroups_);
297		for (int i = 0; i < nGroups_; i++) {
298		int gid = cgLocalToGlobal[i];
299		for (int j = 0; j < nLocal_; j++) {
300		int aid = AtomLocalToGlobal[j];
301	<	if (globalGroupMembership[aid] == gid)
301	>	if (globalGroupMembership[aid] == gid) {
302		groupList_[i].push_back(j);
190	–	}
191	–	}
192	–
193	–	skipsForLocalAtom.clear();
194	–	skipsForLocalAtom.reserve(nLocal_);
195	–
196	–	for (int i = 0; i < nLocal_; i++) {
197	–	int iglob = AtomLocalToGlobal[i];
198	–	for (int j = 0; j < nLocal_; j++) {
199	–	int jglob = AtomLocalToGlobal[j];
200	–	if (excludes.hasPair(iglob, jglob))
201	–	skipsForLocalAtom[i].push_back(j);
202	–	}
203	–	}
204	–
205	–	toposForLocalAtom.clear();
206	–	toposForLocalAtom.reserve(nLocal_);
207	–	for (int i = 0; i < nLocal_; i++) {
208	–	int iglob = AtomLocalToGlobal[i];
209	–	int nTopos = 0;
210	–	for (int j = 0; j < nLocal_; j++) {
211	–	int jglob = AtomLocalToGlobal[j];
212	–	if (oneTwo.hasPair(iglob, jglob)) {
213	–	toposForLocalAtom[i].push_back(j);
214	–	topoDistLocal[i][nTopos] = 1;
215	–	nTopos++;
303		}
217	–	if (oneThree.hasPair(iglob, jglob)) {
218	–	toposForLocalAtom[i].push_back(j);
219	–	topoDistLocal[i][nTopos] = 2;
220	–	nTopos++;
221	–	}
222	–	if (oneFour.hasPair(iglob, jglob)) {
223	–	toposForLocalAtom[i].push_back(j);
224	–	topoDistLocal[i][nTopos] = 3;
225	–	nTopos++;
226	–	}
304		}
305	<	}
305	>	}
306		}
307	<
307	>
308	>	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
309	>	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
310	>	if (toposForAtom[atom1][j] == atom2)
311	>	return topoDist[atom1][j];
312	>	}
313	>	return 0;
314	>	}
315	>
316		void ForceMatrixDecomposition::zeroWorkArrays() {
317	+	pairwisePot = 0.0;
318	+	embeddingPot = 0.0;
319	+	excludedPot = 0.0;
320	+	excludedSelfPot = 0.0;
321
233	–	for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
234	–	longRangePot_[j] = 0.0;
235	–	}
236	–
322		#ifdef IS_MPI
323		if (storageLayout_ & DataStorage::dslForce) {
324		fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
#	Line 249 | Line 334 | namespace OpenMD {
334		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
335
336		fill(pot_col.begin(), pot_col.end(),
337	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
338	+
339	+	fill(expot_row.begin(), expot_row.end(),
340		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
253	–
254	–	pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
341
342	+	fill(expot_col.begin(), expot_col.end(),
343	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
344	+
345		if (storageLayout_ & DataStorage::dslParticlePot) {
346	<	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
347	<	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
346	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
347	>	0.0);
348	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
349	>	0.0);
350		}
351
352		if (storageLayout_ & DataStorage::dslDensity) {
#	Line 264 | Line 355 | namespace OpenMD {
355		}
356
357		if (storageLayout_ & DataStorage::dslFunctional) {
358	<	fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
359	<	fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
358	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
359	>	0.0);
360	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
361	>	0.0);
362		}
363
364		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
#	Line 275 | Line 368 | namespace OpenMD {
368		atomColData.functionalDerivative.end(), 0.0);
369		}
370
371	<	#else
372	<
371	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
372	>	fill(atomRowData.skippedCharge.begin(),
373	>	atomRowData.skippedCharge.end(), 0.0);
374	>	fill(atomColData.skippedCharge.begin(),
375	>	atomColData.skippedCharge.end(), 0.0);
376	>	}
377	>
378	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
379	>	fill(atomRowData.flucQFrc.begin(),
380	>	atomRowData.flucQFrc.end(), 0.0);
381	>	fill(atomColData.flucQFrc.begin(),
382	>	atomColData.flucQFrc.end(), 0.0);
383	>	}
384	>
385	>	if (storageLayout_ & DataStorage::dslElectricField) {
386	>	fill(atomRowData.electricField.begin(),
387	>	atomRowData.electricField.end(), V3Zero);
388	>	fill(atomColData.electricField.begin(),
389	>	atomColData.electricField.end(), V3Zero);
390	>	}
391	>
392	>	if (storageLayout_ & DataStorage::dslSitePotential) {
393	>	fill(atomRowData.sitePotential.begin(),
394	>	atomRowData.sitePotential.end(), 0.0);
395	>	fill(atomColData.sitePotential.begin(),
396	>	atomColData.sitePotential.end(), 0.0);
397	>	}
398	>
399	>	#endif
400	>	// even in parallel, we need to zero out the local arrays:
401	>
402		if (storageLayout_ & DataStorage::dslParticlePot) {
403		fill(snap_->atomData.particlePot.begin(),
404		snap_->atomData.particlePot.end(), 0.0);
#	Line 286 | Line 408 | namespace OpenMD {
408		fill(snap_->atomData.density.begin(),
409		snap_->atomData.density.end(), 0.0);
410		}
411	+
412		if (storageLayout_ & DataStorage::dslFunctional) {
413		fill(snap_->atomData.functional.begin(),
414		snap_->atomData.functional.end(), 0.0);
415		}
416	+
417		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
418		fill(snap_->atomData.functionalDerivative.begin(),
419		snap_->atomData.functionalDerivative.end(), 0.0);
420		}
421	<	#endif
422	<
421	>
422	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
423	>	fill(snap_->atomData.skippedCharge.begin(),
424	>	snap_->atomData.skippedCharge.end(), 0.0);
425	>	}
426	>
427	>	if (storageLayout_ & DataStorage::dslElectricField) {
428	>	fill(snap_->atomData.electricField.begin(),
429	>	snap_->atomData.electricField.end(), V3Zero);
430	>	}
431	>	if (storageLayout_ & DataStorage::dslSitePotential) {
432	>	fill(snap_->atomData.sitePotential.begin(),
433	>	snap_->atomData.sitePotential.end(), 0.0);
434	>	}
435		}
436
437
438		void ForceMatrixDecomposition::distributeData()  {
439	+
440	+	#ifdef IS_MPI
441	+
442		snap_ = sman_->getCurrentSnapshot();
443		storageLayout_ = sman_->getStorageLayout();
305	–	#ifdef IS_MPI
444
445	+	bool needsCG = true;
446	+	if(info_->getNCutoffGroups() != info_->getNAtoms())
447	+	needsCG = false;
448	+
449		// gather up the atomic positions
450	<	AtomCommVectorRow->gather(snap_->atomData.position,
450	>	AtomPlanVectorRow->gather(snap_->atomData.position,
451		atomRowData.position);
452	<	AtomCommVectorColumn->gather(snap_->atomData.position,
452	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
453		atomColData.position);
454
455		// gather up the cutoff group positions
456	<	cgCommVectorRow->gather(snap_->cgData.position,
457	<	cgRowData.position);
458	<	cgCommVectorColumn->gather(snap_->cgData.position,
459	<	cgColData.position);
456	>
457	>	if (needsCG) {
458	>	cgPlanVectorRow->gather(snap_->cgData.position,
459	>	cgRowData.position);
460	>
461	>	cgPlanVectorColumn->gather(snap_->cgData.position,
462	>	cgColData.position);
463	>	}
464	>
465	>
466	>	if (needVelocities_) {
467	>	// gather up the atomic velocities
468	>	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
469	>	atomColData.velocity);
470	>
471	>	if (needsCG) {
472	>	cgPlanVectorColumn->gather(snap_->cgData.velocity,
473	>	cgColData.velocity);
474	>	}
475	>	}
476	>
477
478		// if needed, gather the atomic rotation matrices
479		if (storageLayout_ & DataStorage::dslAmat) {
480	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
480	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
481		atomRowData.aMat);
482	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
482	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
483		atomColData.aMat);
484		}
485	<
486	<	// if needed, gather the atomic eletrostatic frames
487	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
488	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
489	<	atomRowData.electroFrame);
490	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
491	<	atomColData.electroFrame);
485	>
486	>	// if needed, gather the atomic eletrostatic information
487	>	if (storageLayout_ & DataStorage::dslDipole) {
488	>	AtomPlanVectorRow->gather(snap_->atomData.dipole,
489	>	atomRowData.dipole);
490	>	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
491	>	atomColData.dipole);
492		}
493	+
494	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
495	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
496	+	atomRowData.quadrupole);
497	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
498	+	atomColData.quadrupole);
499	+	}
500	+
501	+	// if needed, gather the atomic fluctuating charge values
502	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
503	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
504	+	atomRowData.flucQPos);
505	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
506	+	atomColData.flucQPos);
507	+	}
508	+
509		#endif
510		}
511
#	Line 344 | Line 519 | namespace OpenMD {
519
520		if (storageLayout_ & DataStorage::dslDensity) {
521
522	<	AtomCommRealRow->scatter(atomRowData.density,
522	>	AtomPlanRealRow->scatter(atomRowData.density,
523		snap_->atomData.density);
524
525		int n = snap_->atomData.density.size();
526		vector<RealType> rho_tmp(n, 0.0);
527	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
527	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
528		for (int i = 0; i < n; i++)
529		snap_->atomData.density[i] += rho_tmp[i];
530		}
531	+
532	+	// this isn't necessary if we don't have polarizable atoms, but
533	+	// we'll leave it here for now.
534	+	if (storageLayout_ & DataStorage::dslElectricField) {
535	+
536	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
537	+	snap_->atomData.electricField);
538	+
539	+	int n = snap_->atomData.electricField.size();
540	+	vector<Vector3d> field_tmp(n, V3Zero);
541	+	AtomPlanVectorColumn->scatter(atomColData.electricField,
542	+	field_tmp);
543	+	for (int i = 0; i < n; i++)
544	+	snap_->atomData.electricField[i] += field_tmp[i];
545	+	}
546		#endif
547		}
548
#	Line 365 | Line 555 | namespace OpenMD {
555		storageLayout_ = sman_->getStorageLayout();
556		#ifdef IS_MPI
557		if (storageLayout_ & DataStorage::dslFunctional) {
558	<	AtomCommRealRow->gather(snap_->atomData.functional,
558	>	AtomPlanRealRow->gather(snap_->atomData.functional,
559		atomRowData.functional);
560	<	AtomCommRealColumn->gather(snap_->atomData.functional,
560	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
561		atomColData.functional);
562		}
563
564		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
565	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
565	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
566		atomRowData.functionalDerivative);
567	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
567	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
568		atomColData.functionalDerivative);
569		}
570		#endif
#	Line 388 | Line 578 | namespace OpenMD {
578		int n = snap_->atomData.force.size();
579		vector<Vector3d> frc_tmp(n, V3Zero);
580
581	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
581	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
582		for (int i = 0; i < n; i++) {
583		snap_->atomData.force[i] += frc_tmp[i];
584		frc_tmp[i] = 0.0;
585		}
586
587	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
588	<	for (int i = 0; i < n; i++)
587	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
588	>	for (int i = 0; i < n; i++) {
589		snap_->atomData.force[i] += frc_tmp[i];
590	<
591	<
590	>	}
591	>
592		if (storageLayout_ & DataStorage::dslTorque) {
593
594	<	int nt = snap_->atomData.force.size();
594	>	int nt = snap_->atomData.torque.size();
595		vector<Vector3d> trq_tmp(nt, V3Zero);
596
597	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
598	<	for (int i = 0; i < n; i++) {
597	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
598	>	for (int i = 0; i < nt; i++) {
599		snap_->atomData.torque[i] += trq_tmp[i];
600		trq_tmp[i] = 0.0;
601		}
602
603	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
604	<	for (int i = 0; i < n; i++)
603	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
604	>	for (int i = 0; i < nt; i++)
605		snap_->atomData.torque[i] += trq_tmp[i];
606		}
607	+
608	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
609	+
610	+	int ns = snap_->atomData.skippedCharge.size();
611	+	vector<RealType> skch_tmp(ns, 0.0);
612	+
613	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
614	+	for (int i = 0; i < ns; i++) {
615	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
616	+	skch_tmp[i] = 0.0;
617	+	}
618	+
619	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
620	+	for (int i = 0; i < ns; i++)
621	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
622	+
623	+	}
624
625	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
626	+
627	+	int nq = snap_->atomData.flucQFrc.size();
628	+	vector<RealType> fqfrc_tmp(nq, 0.0);
629	+
630	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
631	+	for (int i = 0; i < nq; i++) {
632	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
633	+	fqfrc_tmp[i] = 0.0;
634	+	}
635	+
636	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
637	+	for (int i = 0; i < nq; i++)
638	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
639	+
640	+	}
641	+
642	+	if (storageLayout_ & DataStorage::dslElectricField) {
643	+
644	+	int nef = snap_->atomData.electricField.size();
645	+	vector<Vector3d> efield_tmp(nef, V3Zero);
646	+
647	+	AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
648	+	for (int i = 0; i < nef; i++) {
649	+	snap_->atomData.electricField[i] += efield_tmp[i];
650	+	efield_tmp[i] = 0.0;
651	+	}
652	+
653	+	AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
654	+	for (int i = 0; i < nef; i++)
655	+	snap_->atomData.electricField[i] += efield_tmp[i];
656	+	}
657	+
658	+	if (storageLayout_ & DataStorage::dslSitePotential) {
659	+
660	+	int nsp = snap_->atomData.sitePotential.size();
661	+	vector<RealType> sp_tmp(nsp, 0.0);
662	+
663	+	AtomPlanRealRow->scatter(atomRowData.sitePotential, sp_tmp);
664	+	for (int i = 0; i < nsp; i++) {
665	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
666	+	sp_tmp[i] = 0.0;
667	+	}
668	+
669	+	AtomPlanRealColumn->scatter(atomColData.sitePotential, sp_tmp);
670	+	for (int i = 0; i < nsp; i++)
671	+	snap_->atomData.sitePotential[i] += sp_tmp[i];
672	+	}
673	+
674		nLocal_ = snap_->getNumberOfAtoms();
675
676		vector<potVec> pot_temp(nLocal_,
677		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
678	+	vector<potVec> expot_temp(nLocal_,
679	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
680
681		// scatter/gather pot_row into the members of my column
682
683	<	AtomCommPotRow->scatter(pot_row, pot_temp);
683	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
684	>	AtomPlanPotRow->scatter(expot_row, expot_temp);
685
686	<	for (int ii = 0;  ii < pot_temp.size(); ii++ )
687	<	pot_local += pot_temp[ii];
688	<
686	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
687	>	pairwisePot += pot_temp[ii];
688	>
689	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
690	>	excludedPot += expot_temp[ii];
691	>
692	>	if (storageLayout_ & DataStorage::dslParticlePot) {
693	>	// This is the pairwise contribution to the particle pot.  The
694	>	// embedding contribution is added in each of the low level
695	>	// non-bonded routines.  In single processor, this is done in
696	>	// unpackInteractionData, not in collectData.
697	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
698	>	for (int i = 0; i < nLocal_; i++) {
699	>	// factor of two is because the total potential terms are divided
700	>	// by 2 in parallel due to row/ column scatter
701	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
702	>	}
703	>	}
704	>	}
705	>
706		fill(pot_temp.begin(), pot_temp.end(),
707		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
708	+	fill(expot_temp.begin(), expot_temp.end(),
709	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
710
711	<	AtomCommPotColumn->scatter(pot_col, pot_temp);
711	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
712	>	AtomPlanPotColumn->scatter(expot_col, expot_temp);
713
714		for (int ii = 0;  ii < pot_temp.size(); ii++ )
715	<	pot_local += pot_temp[ii];
715	>	pairwisePot += pot_temp[ii];
716	>
717	>	for (int ii = 0;  ii < expot_temp.size(); ii++ )
718	>	excludedPot += expot_temp[ii];
719	>
720	>	if (storageLayout_ & DataStorage::dslParticlePot) {
721	>	// This is the pairwise contribution to the particle pot.  The
722	>	// embedding contribution is added in each of the low level
723	>	// non-bonded routines.  In single processor, this is done in
724	>	// unpackInteractionData, not in collectData.
725	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
726	>	for (int i = 0; i < nLocal_; i++) {
727	>	// factor of two is because the total potential terms are divided
728	>	// by 2 in parallel due to row/ column scatter
729	>	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
730	>	}
731	>	}
732	>	}
733
734	+	if (storageLayout_ & DataStorage::dslParticlePot) {
735	+	int npp = snap_->atomData.particlePot.size();
736	+	vector<RealType> ppot_temp(npp, 0.0);
737	+
738	+	// This is the direct or embedding contribution to the particle
739	+	// pot.
740	+
741	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
742	+	for (int i = 0; i < npp; i++) {
743	+	snap_->atomData.particlePot[i] += ppot_temp[i];
744	+	}
745	+
746	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
747	+
748	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
749	+	for (int i = 0; i < npp; i++) {
750	+	snap_->atomData.particlePot[i] += ppot_temp[i];
751	+	}
752	+	}
753	+
754	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
755	+	RealType ploc1 = pairwisePot[ii];
756	+	RealType ploc2 = 0.0;
757	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
758	+	pairwisePot[ii] = ploc2;
759	+	}
760	+
761	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
762	+	RealType ploc1 = excludedPot[ii];
763	+	RealType ploc2 = 0.0;
764	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
765	+	excludedPot[ii] = ploc2;
766	+	}
767	+
768	+	// Here be dragons.
769	+	MPI_Comm col = colComm.getComm();
770	+
771	+	MPI_Allreduce(MPI_IN_PLACE,
772	+	&snap_->frameData.conductiveHeatFlux[0], 3,
773	+	MPI_REALTYPE, MPI_SUM, col);
774	+
775	+
776		#endif
777	+
778		}
779
780	<	int ForceMatrixDecomposition::getNAtomsInRow() {
780	>	/**
781	>	* Collects information obtained during the post-pair (and embedding
782	>	* functional) loops onto local data structures.
783	>	*/
784	>	void ForceMatrixDecomposition::collectSelfData() {
785	>	snap_ = sman_->getCurrentSnapshot();
786	>	storageLayout_ = sman_->getStorageLayout();
787	>
788		#ifdef IS_MPI
789	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
790	+	RealType ploc1 = embeddingPot[ii];
791	+	RealType ploc2 = 0.0;
792	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
793	+	embeddingPot[ii] = ploc2;
794	+	}
795	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
796	+	RealType ploc1 = excludedSelfPot[ii];
797	+	RealType ploc2 = 0.0;
798	+	MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
799	+	excludedSelfPot[ii] = ploc2;
800	+	}
801	+	#endif
802	+
803	+	}
804	+
805	+
806	+
807	+	int& ForceMatrixDecomposition::getNAtomsInRow() {
808	+	#ifdef IS_MPI
809		return nAtomsInRow_;
810		#else
811		return nLocal_;
#	Line 449 | Line 815 | namespace OpenMD {
815		/**
816		* returns the list of atoms belonging to this group.
817		*/
818	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
818	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
819		#ifdef IS_MPI
820		return groupListRow_[cg1];
821		#else
#	Line 457 | Line 823 | namespace OpenMD {
823		#endif
824		}
825
826	<	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
826	>	vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
827		#ifdef IS_MPI
828		return groupListCol_[cg2];
829		#else
#	Line 474 | Line 840 | namespace OpenMD {
840		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
841		#endif
842
843	<	snap_->wrapVector(d);
843	>	if (usePeriodicBoundaryConditions_) {
844	>	snap_->wrapVector(d);
845	>	}
846		return d;
847		}
848
849	+	Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
850	+	#ifdef IS_MPI
851	+	return cgColData.velocity[cg2];
852	+	#else
853	+	return snap_->cgData.velocity[cg2];
854	+	#endif
855	+	}
856
857	+	Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
858	+	#ifdef IS_MPI
859	+	return atomColData.velocity[atom2];
860	+	#else
861	+	return snap_->atomData.velocity[atom2];
862	+	#endif
863	+	}
864	+
865	+
866		Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
867
868		Vector3d d;
#	Line 488 | Line 872 | namespace OpenMD {
872		#else
873		d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
874		#endif
875	<
876	<	snap_->wrapVector(d);
875	>	if (usePeriodicBoundaryConditions_) {
876	>	snap_->wrapVector(d);
877	>	}
878		return d;
879		}
880
#	Line 501 | Line 886 | namespace OpenMD {
886		#else
887		d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
888		#endif
889	<
890	<	snap_->wrapVector(d);
889	>	if (usePeriodicBoundaryConditions_) {
890	>	snap_->wrapVector(d);
891	>	}
892		return d;
893		}
894
895	<	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
895	>	RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
896		#ifdef IS_MPI
897		return massFactorsRow[atom1];
898		#else
899	<	return massFactorsLocal[atom1];
899	>	return massFactors[atom1];
900		#endif
901		}
902
903	<	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
903	>	RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
904		#ifdef IS_MPI
905		return massFactorsCol[atom2];
906		#else
907	<	return massFactorsLocal[atom2];
907	>	return massFactors[atom2];
908		#endif
909
910		}
#	Line 531 | Line 917 | namespace OpenMD {
917		#else
918		d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
919		#endif
920	<
921	<	snap_->wrapVector(d);
920	>	if (usePeriodicBoundaryConditions_) {
921	>	snap_->wrapVector(d);
922	>	}
923		return d;
924		}
925
926	<	vector<int> ForceMatrixDecomposition::getSkipsForRowAtom(int atom1) {
927	<	#ifdef IS_MPI
541	<	return skipsForRowAtom[atom1];
542	<	#else
543	<	return skipsForLocalAtom[atom1];
544	<	#endif
926	>	vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
927	>	return excludesForAtom[atom1];
928		}
929
930		/**
931	<	* There are a number of reasons to skip a pair or a
549	<	* particle. Mostly we do this to exclude atoms who are involved in
550	<	* short range interactions (bonds, bends, torsions), but we also
551	<	* need to exclude some overcounted interactions that result from
931	>	* We need to exclude some overcounted interactions that result from
932		* the parallel decomposition.
933		*/
934	<	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
934	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
935		int unique_id_1, unique_id_2;
936	<
936	>
937		#ifdef IS_MPI
938		// in MPI, we have to look up the unique IDs for each atom
939		unique_id_1 = AtomRowToGlobal[atom1];
940		unique_id_2 = AtomColToGlobal[atom2];
941	+	// group1 = cgRowToGlobal[cg1];
942	+	// group2 = cgColToGlobal[cg2];
943	+	#else
944	+	unique_id_1 = AtomLocalToGlobal[atom1];
945	+	unique_id_2 = AtomLocalToGlobal[atom2];
946	+	int group1 = cgLocalToGlobal[cg1];
947	+	int group2 = cgLocalToGlobal[cg2];
948	+	#endif
949
562	–	// this situation should only arise in MPI simulations
950		if (unique_id_1 == unique_id_2) return true;
951	<
951	>
952	>	#ifdef IS_MPI
953		// this prevents us from doing the pair on multiple processors
954		if (unique_id_1 < unique_id_2) {
955		if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
956		} else {
957	<	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
957	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
958		}
959	<	#else
960	<	// in the normal loop, the atom numbers are unique
961	<	unique_id_1 = atom1;
962	<	unique_id_2 = atom2;
959	>	#endif
960	>
961	>	#ifndef IS_MPI
962	>	if (group1 == group2) {
963	>	if (unique_id_1 < unique_id_2) return true;
964	>	}
965		#endif
966
967	<	#ifdef IS_MPI
578	<	for (vector<int>::iterator i = skipsForRowAtom[atom1].begin();
579	<	i != skipsForRowAtom[atom1].end(); ++i) {
580	<	if ( (*i) == unique_id_2 ) return true;
581	<	}
582	<	#else
583	<	for (vector<int>::iterator i = skipsForLocalAtom[atom1].begin();
584	<	i != skipsForLocalAtom[atom1].end(); ++i) {
585	<	if ( (*i) == unique_id_2 ) return true;
586	<	}
587	<	#endif
967	>	return false;
968		}
969
970	<	int ForceMatrixDecomposition::getTopoDistance(int atom1, int atom2) {
971	<
972	<	#ifdef IS_MPI
973	<	for (int i = 0; i < toposForRowAtom[atom1].size(); i++) {
974	<	if ( toposForRowAtom[atom1][i] == atom2 ) return topoDistRow[atom1][i];
975	<	}
976	<	#else
977	<	for (int i = 0; i < toposForLocalAtom[atom1].size(); i++) {
978	<	if ( toposForLocalAtom[atom1][i] == atom2 ) return topoDistLocal[atom1][i];
970	>	/**
971	>	* We need to handle the interactions for atoms who are involved in
972	>	* the same rigid body as well as some short range interactions
973	>	* (bonds, bends, torsions) differently from other interactions.
974	>	* We'll still visit the pairwise routines, but with a flag that
975	>	* tells those routines to exclude the pair from direct long range
976	>	* interactions.  Some indirect interactions (notably reaction
977	>	* field) must still be handled for these pairs.
978	>	*/
979	>	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
980	>
981	>	// excludesForAtom was constructed to use row/column indices in the MPI
982	>	// version, and to use local IDs in the non-MPI version:
983	>
984	>	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
985	>	i != excludesForAtom[atom1].end(); ++i) {
986	>	if ( (*i) == atom2 ) return true;
987		}
600	–	#endif
988
989	<	// zero is default for unconnected (i.e. normal) pair interactions
603	<	return 0;
989	>	return false;
990		}
991
992	+
993		void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
994		#ifdef IS_MPI
995		atomRowData.force[atom1] += fg;
#	Line 620 | Line 1007 | namespace OpenMD {
1007		}
1008
1009		// filling interaction blocks with pointers
1010	<	InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {
1011	<	InteractionData idat;
1010	>	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1011	>	int atom1, int atom2,
1012	>	bool newAtom1) {
1013
1014	+	idat.excluded = excludeAtomPair(atom1, atom2);
1015	+
1016	+	if (newAtom1) {
1017	+
1018		#ifdef IS_MPI
1019	<
1020	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1021	<	ff_->getAtomType(identsCol[atom2]) );
1019	>	idat.atid1 = identsRow[atom1];
1020	>	idat.atid2 = identsCol[atom2];
1021	>
1022	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1023	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1024	>	} else {
1025	>	idat.sameRegion = false;
1026	>	}
1027	>
1028	>	if (storageLayout_ & DataStorage::dslAmat) {
1029	>	idat.A1 = &(atomRowData.aMat[atom1]);
1030	>	idat.A2 = &(atomColData.aMat[atom2]);
1031	>	}
1032	>
1033	>	if (storageLayout_ & DataStorage::dslTorque) {
1034	>	idat.t1 = &(atomRowData.torque[atom1]);
1035	>	idat.t2 = &(atomColData.torque[atom2]);
1036	>	}
1037	>
1038	>	if (storageLayout_ & DataStorage::dslDipole) {
1039	>	idat.dipole1 = &(atomRowData.dipole[atom1]);
1040	>	idat.dipole2 = &(atomColData.dipole[atom2]);
1041	>	}
1042	>
1043	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1044	>	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1045	>	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1046	>	}
1047	>
1048	>	if (storageLayout_ & DataStorage::dslDensity) {
1049	>	idat.rho1 = &(atomRowData.density[atom1]);
1050	>	idat.rho2 = &(atomColData.density[atom2]);
1051	>	}
1052	>
1053	>	if (storageLayout_ & DataStorage::dslFunctional) {
1054	>	idat.frho1 = &(atomRowData.functional[atom1]);
1055	>	idat.frho2 = &(atomColData.functional[atom2]);
1056	>	}
1057	>
1058	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1059	>	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1060	>	idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1061	>	}
1062	>
1063	>	if (storageLayout_ & DataStorage::dslParticlePot) {
1064	>	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1065	>	idat.particlePot2 = &(atomColData.particlePot[atom2]);
1066	>	}
1067	>
1068	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1069	>	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1070	>	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1071	>	}
1072	>
1073	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1074	>	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1075	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1076	>	}
1077	>
1078	>	#else
1079	>
1080	>	idat.atid1 = idents[atom1];
1081	>	idat.atid2 = idents[atom2];
1082	>
1083	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1084	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1085	>	} else {
1086	>	idat.sameRegion = false;
1087	>	}
1088	>
1089	>	if (storageLayout_ & DataStorage::dslAmat) {
1090	>	idat.A1 = &(snap_->atomData.aMat[atom1]);
1091	>	idat.A2 = &(snap_->atomData.aMat[atom2]);
1092	>	}
1093	>
1094	>	if (storageLayout_ & DataStorage::dslTorque) {
1095	>	idat.t1 = &(snap_->atomData.torque[atom1]);
1096	>	idat.t2 = &(snap_->atomData.torque[atom2]);
1097	>	}
1098	>
1099	>	if (storageLayout_ & DataStorage::dslDipole) {
1100	>	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1101	>	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1102	>	}
1103	>
1104	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1105	>	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1106	>	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1107	>	}
1108	>
1109	>	if (storageLayout_ & DataStorage::dslDensity) {
1110	>	idat.rho1 = &(snap_->atomData.density[atom1]);
1111	>	idat.rho2 = &(snap_->atomData.density[atom2]);
1112	>	}
1113	>
1114	>	if (storageLayout_ & DataStorage::dslFunctional) {
1115	>	idat.frho1 = &(snap_->atomData.functional[atom1]);
1116	>	idat.frho2 = &(snap_->atomData.functional[atom2]);
1117	>	}
1118	>
1119	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1120	>	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1121	>	idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1122	>	}
1123	>
1124	>	if (storageLayout_ & DataStorage::dslParticlePot) {
1125	>	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1126	>	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1127	>	}
1128	>
1129	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1130	>	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1131	>	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1132	>	}
1133	>
1134	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1135	>	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1136	>	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1137	>	}
1138	>	#endif
1139	>
1140	>	} else {
1141	>	// atom1 is not new, so don't bother updating properties of that atom:
1142	>	#ifdef IS_MPI
1143	>	idat.atid2 = identsCol[atom2];
1144
1145	<
1145	>	if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1146	>	idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1147	>	} else {
1148	>	idat.sameRegion = false;
1149	>	}
1150	>
1151		if (storageLayout_ & DataStorage::dslAmat) {
633	–	idat.A1 = &(atomRowData.aMat[atom1]);
1152		idat.A2 = &(atomColData.aMat[atom2]);
1153		}
1154
637	–	if (storageLayout_ & DataStorage::dslElectroFrame) {
638	–	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
639	–	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
640	–	}
641	–
1155		if (storageLayout_ & DataStorage::dslTorque) {
643	–	idat.t1 = &(atomRowData.torque[atom1]);
1156		idat.t2 = &(atomColData.torque[atom2]);
1157		}
1158
1159	+	if (storageLayout_ & DataStorage::dslDipole) {
1160	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1161	+	}
1162	+
1163	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1164	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1165	+	}
1166	+
1167		if (storageLayout_ & DataStorage::dslDensity) {
648	–	idat.rho1 = &(atomRowData.density[atom1]);
1168		idat.rho2 = &(atomColData.density[atom2]);
1169		}
1170
1171		if (storageLayout_ & DataStorage::dslFunctional) {
653	–	idat.frho1 = &(atomRowData.functional[atom1]);
1172		idat.frho2 = &(atomColData.functional[atom2]);
1173		}
1174
1175		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
658	–	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1176		idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1177		}
1178
1179		if (storageLayout_ & DataStorage::dslParticlePot) {
663	–	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1180		idat.particlePot2 = &(atomColData.particlePot[atom2]);
1181		}
1182
1183	<	#else
1183	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1184	>	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1185	>	}
1186
1187	<	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
1188	<	ff_->getAtomType(identsLocal[atom2]) );
1187	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1188	>	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1189	>	}
1190
1191	<	if (storageLayout_ & DataStorage::dslAmat) {
1192	<	idat.A1 = &(snap_->atomData.aMat[atom1]);
1193	<	idat.A2 = &(snap_->atomData.aMat[atom2]);
1191	>	#else
1192	>	idat.atid2 = idents[atom2];
1193	>
1194	>	if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1195	>	idat.sameRegion = (regions[atom1] == regions[atom2]);
1196	>	} else {
1197	>	idat.sameRegion = false;
1198		}
1199
1200	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1201	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
679	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1200	>	if (storageLayout_ & DataStorage::dslAmat) {
1201	>	idat.A2 = &(snap_->atomData.aMat[atom2]);
1202		}
1203
1204		if (storageLayout_ & DataStorage::dslTorque) {
683	–	idat.t1 = &(snap_->atomData.torque[atom1]);
1205		idat.t2 = &(snap_->atomData.torque[atom2]);
1206		}
1207
1208	<	if (storageLayout_ & DataStorage::dslDensity) {
1209	<	idat.rho1 = &(snap_->atomData.density[atom1]);
1208	>	if (storageLayout_ & DataStorage::dslDipole) {
1209	>	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1210	>	}
1211	>
1212	>	if (storageLayout_ & DataStorage::dslQuadrupole) {
1213	>	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1214	>	}
1215	>
1216	>	if (storageLayout_ & DataStorage::dslDensity) {
1217		idat.rho2 = &(snap_->atomData.density[atom2]);
1218		}
1219
1220		if (storageLayout_ & DataStorage::dslFunctional) {
693	–	idat.frho1 = &(snap_->atomData.functional[atom1]);
1221		idat.frho2 = &(snap_->atomData.functional[atom2]);
1222		}
1223
1224		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
698	–	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1225		idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1226		}
1227
1228		if (storageLayout_ & DataStorage::dslParticlePot) {
703	–	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1229		idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1230		}
1231
1232	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1233	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1234	+	}
1235	+
1236	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1237	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1238	+	}
1239	+
1240		#endif
1241	<	return idat;
1241	>	}
1242		}
710	–
1243
1244	<	void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {
1244	>	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat,
1245	>	int atom1, int atom2) {
1246		#ifdef IS_MPI
1247	<	pot_row[atom1] += 0.5 *  *(idat.pot);
1248	<	pot_col[atom2] += 0.5 *  *(idat.pot);
1247	>	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1248	>	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1249	>	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1250	>	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1251
1252		atomRowData.force[atom1] += *(idat.f1);
1253		atomColData.force[atom2] -= *(idat.f1);
719	–	#else
720	–	longRangePot_ += *(idat.pot);
721	–
722	–	snap_->atomData.force[atom1] += *(idat.f1);
723	–	snap_->atomData.force[atom2] -= *(idat.f1);
724	–	#endif
1254
1255	<	}
1255	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1256	>	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1257	>	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1258	>	}
1259
1260	+	if (storageLayout_ & DataStorage::dslElectricField) {
1261	+	atomRowData.electricField[atom1] += *(idat.eField1);
1262	+	atomColData.electricField[atom2] += *(idat.eField2);
1263	+	}
1264
1265	<	InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
1265	>	if (storageLayout_ & DataStorage::dslSitePotential) {
1266	>	atomRowData.sitePotential[atom1] += *(idat.sPot1);
1267	>	atomColData.sitePotential[atom2] += *(idat.sPot2);
1268	>	}
1269
1270	<	InteractionData idat;
1271	<	#ifdef IS_MPI
1272	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
734	<	ff_->getAtomType(identsCol[atom2]) );
1270	>	#else
1271	>	pairwisePot += *(idat.pot);
1272	>	excludedPot += *(idat.excludedPot);
1273
1274	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1275	<	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1276	<	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
1274	>	snap_->atomData.force[atom1] += *(idat.f1);
1275	>	snap_->atomData.force[atom2] -= *(idat.f1);
1276	>
1277	>	if (idat.doParticlePot) {
1278	>	// This is the pairwise contribution to the particle pot.  The
1279	>	// embedding contribution is added in each of the low level
1280	>	// non-bonded routines.  In parallel, this calculation is done
1281	>	// in collectData, not in unpackInteractionData.
1282	>	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1283	>	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1284		}
1285	<	if (storageLayout_ & DataStorage::dslTorque) {
1286	<	idat.t1 = &(atomRowData.torque[atom1]);
1287	<	idat.t2 = &(atomColData.torque[atom2]);
1285	>
1286	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
1287	>	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1288	>	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1289		}
744	–	#else
745	–	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
746	–	ff_->getAtomType(identsLocal[atom2]) );
1290
1291	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1292	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1293	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1291	>	if (storageLayout_ & DataStorage::dslElectricField) {
1292	>	snap_->atomData.electricField[atom1] += *(idat.eField1);
1293	>	snap_->atomData.electricField[atom2] += *(idat.eField2);
1294		}
1295	<	if (storageLayout_ & DataStorage::dslTorque) {
1296	<	idat.t1 = &(snap_->atomData.torque[atom1]);
1297	<	idat.t2 = &(snap_->atomData.torque[atom2]);
1295	>
1296	>	if (storageLayout_ & DataStorage::dslSitePotential) {
1297	>	snap_->atomData.sitePotential[atom1] += *(idat.sPot1);
1298	>	snap_->atomData.sitePotential[atom2] += *(idat.sPot2);
1299		}
1300	<	#endif
1300	>
1301	>	#endif
1302	>
1303		}
1304
1305		/*
1306		* buildNeighborList
1307		*
1308	<	* first element of pair is row-indexed CutoffGroup
1309	<	* second element of pair is column-indexed CutoffGroup
1308	>	* Constructs the Verlet neighbor list for a force-matrix
1309	>	* decomposition.  In this case, each processor is responsible for
1310	>	* row-site interactions with column-sites.
1311	>	*
1312	>	* neighborList is returned as a packed array of neighboring
1313	>	* column-ordered CutoffGroups.  The starting position in
1314	>	* neighborList for each row-ordered CutoffGroup is given by the
1315	>	* returned vector point.
1316		*/
1317	<	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1318	<
1319	<	vector<pair<int, int> > neighborList;
1317	>	void ForceMatrixDecomposition::buildNeighborList(vector<int>& neighborList,
1318	>	vector<int>& point) {
1319	>	neighborList.clear();
1320	>	point.clear();
1321	>	int len = 0;
1322	>
1323	>	bool doAllPairs = false;
1324	>
1325	>	Snapshot* snap_ = sman_->getCurrentSnapshot();
1326	>	Mat3x3d box;
1327	>	Mat3x3d invBox;
1328	>
1329	>	Vector3d rs, scaled, dr;
1330	>	Vector3i whichCell;
1331	>	int cellIndex;
1332	>
1333		#ifdef IS_MPI
1334		cellListRow_.clear();
1335		cellListCol_.clear();
1336	+	point.resize(nGroupsInRow_+1);
1337		#else
1338		cellList_.clear();
1339	+	point.resize(nGroups_+1);
1340		#endif
1341	+
1342	+	if (!usePeriodicBoundaryConditions_) {
1343	+	box = snap_->getBoundingBox();
1344	+	invBox = snap_->getInvBoundingBox();
1345	+	} else {
1346	+	box = snap_->getHmat();
1347	+	invBox = snap_->getInvHmat();
1348	+	}
1349	+
1350	+	Vector3d A = box.getColumn(0);
1351	+	Vector3d B = box.getColumn(1);
1352	+	Vector3d C = box.getColumn(2);
1353
1354	<	// dangerous to not do error checking.
1355	<	RealType rCut_;
1356	<
1357	<	RealType rList_ = (rCut_ + skinThickness_);
779	<	RealType rl2 = rList_ * rList_;
780	<	Snapshot* snap_ = sman_->getCurrentSnapshot();
781	<	Mat3x3d Hmat = snap_->getHmat();
782	<	Vector3d Hx = Hmat.getColumn(0);
783	<	Vector3d Hy = Hmat.getColumn(1);
784	<	Vector3d Hz = Hmat.getColumn(2);
1354	>	// Required for triclinic cells
1355	>	Vector3d AxB = cross(A, B);
1356	>	Vector3d BxC = cross(B, C);
1357	>	Vector3d CxA = cross(C, A);
1358
1359	<	nCells_.x() = (int) ( Hx.length() )/ rList_;
1360	<	nCells_.y() = (int) ( Hy.length() )/ rList_;
1361	<	nCells_.z() = (int) ( Hz.length() )/ rList_;
1359	>	// unit vectors perpendicular to the faces of the triclinic cell:
1360	>	AxB.normalize();
1361	>	BxC.normalize();
1362	>	CxA.normalize();
1363
1364	<	Mat3x3d invHmat = snap_->getInvHmat();
1365	<	Vector3d rs, scaled, dr;
1366	<	Vector3i whichCell;
1367	<	int cellIndex;
1368	<
1364	>	// A set of perpendicular lengths in triclinic cells:
1365	>	RealType Wa = abs(dot(A, BxC));
1366	>	RealType Wb = abs(dot(B, CxA));
1367	>	RealType Wc = abs(dot(C, AxB));
1368	>
1369	>	nCells_.x() = int( Wa / rList_ );
1370	>	nCells_.y() = int( Wb / rList_ );
1371	>	nCells_.z() = int( Wc / rList_ );
1372	>
1373	>	// handle small boxes where the cell offsets can end up repeating cells
1374	>	if (nCells_.x() < 3) doAllPairs = true;
1375	>	if (nCells_.y() < 3) doAllPairs = true;
1376	>	if (nCells_.z() < 3) doAllPairs = true;
1377	>
1378	>	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1379	>
1380		#ifdef IS_MPI
1381	<	for (int i = 0; i < nGroupsInRow_; i++) {
1382	<	rs = cgRowData.position[i];
1383	<	// scaled positions relative to the box vectors
1384	<	scaled = invHmat * rs;
1385	<	// wrap the vector back into the unit box by subtracting integer box
1386	<	// numbers
1387	<	for (int j = 0; j < 3; j++)
1388	<	scaled[j] -= roundMe(scaled[j]);
1389	<
1390	<	// find xyz-indices of cell that cutoffGroup is in.
1391	<	whichCell.x() = nCells_.x() * scaled.x();
1392	<	whichCell.y() = nCells_.y() * scaled.y();
1393	<	whichCell.z() = nCells_.z() * scaled.z();
1381	>	cellListRow_.resize(nCtot);
1382	>	cellListCol_.resize(nCtot);
1383	>	#else
1384	>	cellList_.resize(nCtot);
1385	>	#endif
1386	>
1387	>	if (!doAllPairs) {
1388	>
1389	>	#ifdef IS_MPI
1390	>
1391	>	for (int i = 0; i < nGroupsInRow_; i++) {
1392	>	rs = cgRowData.position[i];
1393	>
1394	>	// scaled positions relative to the box vectors
1395	>	scaled = invBox * rs;
1396	>
1397	>	// wrap the vector back into the unit box by subtracting integer box
1398	>	// numbers
1399	>	for (int j = 0; j < 3; j++) {
1400	>	scaled[j] -= roundMe(scaled[j]);
1401	>	scaled[j] += 0.5;
1402	>	// Handle the special case when an object is exactly on the
1403	>	// boundary (a scaled coordinate of 1.0 is the same as
1404	>	// scaled coordinate of 0.0)
1405	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1406	>	}
1407	>
1408	>	// find xyz-indices of cell that cutoffGroup is in.
1409	>	whichCell.x() = nCells_.x() * scaled.x();
1410	>	whichCell.y() = nCells_.y() * scaled.y();
1411	>	whichCell.z() = nCells_.z() * scaled.z();
1412	>
1413	>	// find single index of this cell:
1414	>	cellIndex = Vlinear(whichCell, nCells_);
1415	>
1416	>	// add this cutoff group to the list of groups in this cell;
1417	>	cellListRow_[cellIndex].push_back(i);
1418	>	}
1419	>	for (int i = 0; i < nGroupsInCol_; i++) {
1420	>	rs = cgColData.position[i];
1421	>
1422	>	// scaled positions relative to the box vectors
1423	>	scaled = invBox * rs;
1424	>
1425	>	// wrap the vector back into the unit box by subtracting integer box
1426	>	// numbers
1427	>	for (int j = 0; j < 3; j++) {
1428	>	scaled[j] -= roundMe(scaled[j]);
1429	>	scaled[j] += 0.5;
1430	>	// Handle the special case when an object is exactly on the
1431	>	// boundary (a scaled coordinate of 1.0 is the same as
1432	>	// scaled coordinate of 0.0)
1433	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1434	>	}
1435	>
1436	>	// find xyz-indices of cell that cutoffGroup is in.
1437	>	whichCell.x() = nCells_.x() * scaled.x();
1438	>	whichCell.y() = nCells_.y() * scaled.y();
1439	>	whichCell.z() = nCells_.z() * scaled.z();
1440	>
1441	>	// find single index of this cell:
1442	>	cellIndex = Vlinear(whichCell, nCells_);
1443	>
1444	>	// add this cutoff group to the list of groups in this cell;
1445	>	cellListCol_[cellIndex].push_back(i);
1446	>	}
1447	>
1448	>	#else
1449	>	for (int i = 0; i < nGroups_; i++) {
1450	>	rs = snap_->cgData.position[i];
1451	>
1452	>	// scaled positions relative to the box vectors
1453	>	scaled = invBox * rs;
1454	>
1455	>	// wrap the vector back into the unit box by subtracting integer box
1456	>	// numbers
1457	>	for (int j = 0; j < 3; j++) {
1458	>	scaled[j] -= roundMe(scaled[j]);
1459	>	scaled[j] += 0.5;
1460	>	// Handle the special case when an object is exactly on the
1461	>	// boundary (a scaled coordinate of 1.0 is the same as
1462	>	// scaled coordinate of 0.0)
1463	>	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1464	>	}
1465	>
1466	>	// find xyz-indices of cell that cutoffGroup is in.
1467	>	whichCell.x() = int(nCells_.x() * scaled.x());
1468	>	whichCell.y() = int(nCells_.y() * scaled.y());
1469	>	whichCell.z() = int(nCells_.z() * scaled.z());
1470	>
1471	>	// find single index of this cell:
1472	>	cellIndex = Vlinear(whichCell, nCells_);
1473	>
1474	>	// add this cutoff group to the list of groups in this cell;
1475	>	cellList_[cellIndex].push_back(i);
1476	>	}
1477
1478	<	// find single index of this cell:
811	<	cellIndex = Vlinear(whichCell, nCells_);
812	<	// add this cutoff group to the list of groups in this cell;
813	<	cellListRow_[cellIndex].push_back(i);
814	<	}
1478	>	#endif
1479
1480	<	for (int i = 0; i < nGroupsInCol_; i++) {
1481	<	rs = cgColData.position[i];
1482	<	// scaled positions relative to the box vectors
819	<	scaled = invHmat * rs;
820	<	// wrap the vector back into the unit box by subtracting integer box
821	<	// numbers
822	<	for (int j = 0; j < 3; j++)
823	<	scaled[j] -= roundMe(scaled[j]);
824	<
825	<	// find xyz-indices of cell that cutoffGroup is in.
826	<	whichCell.x() = nCells_.x() * scaled.x();
827	<	whichCell.y() = nCells_.y() * scaled.y();
828	<	whichCell.z() = nCells_.z() * scaled.z();
829	<
830	<	// find single index of this cell:
831	<	cellIndex = Vlinear(whichCell, nCells_);
832	<	// add this cutoff group to the list of groups in this cell;
833	<	cellListCol_[cellIndex].push_back(i);
834	<	}
1480	>	#ifdef IS_MPI
1481	>	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1482	>	rs = cgRowData.position[j1];
1483		#else
836	–	for (int i = 0; i < nGroups_; i++) {
837	–	rs = snap_->cgData.position[i];
838	–	// scaled positions relative to the box vectors
839	–	scaled = invHmat * rs;
840	–	// wrap the vector back into the unit box by subtracting integer box
841	–	// numbers
842	–	for (int j = 0; j < 3; j++)
843	–	scaled[j] -= roundMe(scaled[j]);
1484
1485	<	// find xyz-indices of cell that cutoffGroup is in.
1486	<	whichCell.x() = nCells_.x() * scaled.x();
847	<	whichCell.y() = nCells_.y() * scaled.y();
848	<	whichCell.z() = nCells_.z() * scaled.z();
849	<
850	<	// find single index of this cell:
851	<	cellIndex = Vlinear(whichCell, nCells_);
852	<	// add this cutoff group to the list of groups in this cell;
853	<	cellList_[cellIndex].push_back(i);
854	<	}
1485	>	for (int j1 = 0; j1 < nGroups_; j1++) {
1486	>	rs = snap_->cgData.position[j1];
1487		#endif
1488	+	point[j1] = len;
1489	+
1490	+	// scaled positions relative to the box vectors
1491	+	scaled = invBox * rs;
1492	+
1493	+	// wrap the vector back into the unit box by subtracting integer box
1494	+	// numbers
1495	+	for (int j = 0; j < 3; j++) {
1496	+	scaled[j] -= roundMe(scaled[j]);
1497	+	scaled[j] += 0.5;
1498	+	// Handle the special case when an object is exactly on the
1499	+	// boundary (a scaled coordinate of 1.0 is the same as
1500	+	// scaled coordinate of 0.0)
1501	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1502	+	}
1503	+
1504	+	// find xyz-indices of cell that cutoffGroup is in.
1505	+	whichCell.x() = nCells_.x() * scaled.x();
1506	+	whichCell.y() = nCells_.y() * scaled.y();
1507	+	whichCell.z() = nCells_.z() * scaled.z();
1508	+
1509	+	// find single index of this cell:
1510	+	int m1 = Vlinear(whichCell, nCells_);
1511
1512	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1513	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1514	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1515	<	Vector3i m1v(m1x, m1y, m1z);
861	<	int m1 = Vlinear(m1v, nCells_);
1512	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1513	>	os != cellOffsets_.end(); ++os) {
1514	>
1515	>	Vector3i m2v = whichCell + (*os);
1516
1517	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1518	<	os != cellOffsets_.end(); ++os) {
1517	>	if (m2v.x() >= nCells_.x()) {
1518	>	m2v.x() = 0;
1519	>	} else if (m2v.x() < 0) {
1520	>	m2v.x() = nCells_.x() - 1;
1521	>	}
1522	>
1523	>	if (m2v.y() >= nCells_.y()) {
1524	>	m2v.y() = 0;
1525	>	} else if (m2v.y() < 0) {
1526	>	m2v.y() = nCells_.y() - 1;
1527	>	}
1528	>
1529	>	if (m2v.z() >= nCells_.z()) {
1530	>	m2v.z() = 0;
1531	>	} else if (m2v.z() < 0) {
1532	>	m2v.z() = nCells_.z() - 1;
1533	>	}
1534	>	int m2 = Vlinear (m2v, nCells_);
1535	>	#ifdef IS_MPI
1536	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1537	>	j2 != cellListCol_[m2].end(); ++j2) {
1538
1539	<	Vector3i m2v = m1v + (*os);
1540	<
1541	<	if (m2v.x() >= nCells_.x()) {
1542	<	m2v.x() = 0;
1543	<	} else if (m2v.x() < 0) {
1544	<	m2v.x() = nCells_.x() - 1;
1539	>	// In parallel, we need to visit *all* pairs of row
1540	>	// & column indicies and will divide labor in the
1541	>	// force evaluation later.
1542	>	dr = cgColData.position[(*j2)] - rs;
1543	>	if (usePeriodicBoundaryConditions_) {
1544	>	snap_->wrapVector(dr);
1545		}
1546	+	if (dr.lengthSquare() < rListSq_) {
1547	+	neighborList.push_back( (*j2) );
1548	+	++len;
1549	+	}
1550	+	}
1551	+	#else
1552	+	for (vector<int>::iterator j2 = cellList_[m2].begin();
1553	+	j2 != cellList_[m2].end(); ++j2) {
1554	+
1555	+	// Always do this if we're in different cells or if
1556	+	// we're in the same cell and the global index of
1557	+	// the j2 cutoff group is greater than or equal to
1558	+	// the j1 cutoff group.  Note that Rappaport's code
1559	+	// has a "less than" conditional here, but that
1560	+	// deals with atom-by-atom computation.  OpenMD
1561	+	// allows atoms within a single cutoff group to
1562	+	// interact with each other.
1563
1564	<	if (m2v.y() >= nCells_.y()) {
1565	<	m2v.y() = 0;
1566	<	} else if (m2v.y() < 0) {
1567	<	m2v.y() = nCells_.y() - 1;
1568	<	}
879	<
880	<	if (m2v.z() >= nCells_.z()) {
881	<	m2v.z() = 0;
882	<	} else if (m2v.z() < 0) {
883	<	m2v.z() = nCells_.z() - 1;
884	<	}
885	<
886	<	int m2 = Vlinear (m2v, nCells_);
887	<
888	<	#ifdef IS_MPI
889	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
890	<	j1 != cellListRow_[m1].end(); ++j1) {
891	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
892	<	j2 != cellListCol_[m2].end(); ++j2) {
893	<
894	<	// Always do this if we're in different cells or if
895	<	// we're in the same cell and the global index of the
896	<	// j2 cutoff group is less than the j1 cutoff group
897	<
898	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
899	<	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
900	<	snap_->wrapVector(dr);
901	<	if (dr.lengthSquare() < rl2) {
902	<	neighborList.push_back(make_pair((*j1), (*j2)));
903	<	}
904	<	}
1564	>	if ( (*j2) >= j1 ) {
1565	>
1566	>	dr = snap_->cgData.position[(*j2)] - rs;
1567	>	if (usePeriodicBoundaryConditions_) {
1568	>	snap_->wrapVector(dr);
1569		}
1570	<	}
1571	<	#else
1572	<	for (vector<int>::iterator j1 = cellList_[m1].begin();
909	<	j1 != cellList_[m1].end(); ++j1) {
910	<	for (vector<int>::iterator j2 = cellList_[m2].begin();
911	<	j2 != cellList_[m2].end(); ++j2) {
912	<
913	<	// Always do this if we're in different cells or if
914	<	// we're in the same cell and the global index of the
915	<	// j2 cutoff group is less than the j1 cutoff group
916	<
917	<	if (m2 != m1 || (*j2) < (*j1)) {
918	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
919	<	snap_->wrapVector(dr);
920	<	if (dr.lengthSquare() < rl2) {
921	<	neighborList.push_back(make_pair((*j1), (*j2)));
922	<	}
923	<	}
1570	>	if ( dr.lengthSquare() < rListSq_) {
1571	>	neighborList.push_back( (*j2) );
1572	>	++len;
1573		}
1574		}
1575	+	}
1576		#endif
1577	+	}
1578	+	}
1579	+	} else {
1580	+	// branch to do all cutoff group pairs
1581	+	#ifdef IS_MPI
1582	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1583	+	point[j1] = len;
1584	+	rs = cgRowData.position[j1];
1585	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1586	+	dr = cgColData.position[j2] - rs;
1587	+	if (usePeriodicBoundaryConditions_) {
1588	+	snap_->wrapVector(dr);
1589		}
1590	+	if (dr.lengthSquare() < rListSq_) {
1591	+	neighborList.push_back( j2 );
1592	+	++len;
1593	+	}
1594		}
1595	+	}
1596	+	#else
1597	+	// include all groups here.
1598	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1599	+	point[j1] = len;
1600	+	rs = snap_->cgData.position[j1];
1601	+	// include self group interactions j2 == j1
1602	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1603	+	dr = snap_->cgData.position[j2] - rs;
1604	+	if (usePeriodicBoundaryConditions_) {
1605	+	snap_->wrapVector(dr);
1606	+	}
1607	+	if (dr.lengthSquare() < rListSq_) {
1608	+	neighborList.push_back( j2 );
1609	+	++len;
1610	+	}
1611	+	}
1612		}
1613	+	#endif
1614		}
1615
1616	+	#ifdef IS_MPI
1617	+	point[nGroupsInRow_] = len;
1618	+	#else
1619	+	point[nGroups_] = len;
1620	+	#endif
1621	+
1622		// save the local cutoff group positions for the check that is
1623		// done on each loop:
1624		saved_CG_positions_.clear();
1625	+	saved_CG_positions_.reserve(nGroups_);
1626		for (int i = 0; i < nGroups_; i++)
1627		saved_CG_positions_.push_back(snap_->cgData.position[i]);
937	–
938	–	return neighborList;
1628		}
1629		} //end namespace OpenMD

Diff Legend

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