ViewVC Help

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

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

Comparing:
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1577 by gezelter, Wed Jun 8 20:26:56 2011 UTC vs.
branches/devel_omp/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1595 by chuckv, Tue Jul 19 18:50:04 2011 UTC

#	Line 47 | Line 47 | namespace OpenMD {
47		using namespace std;
48		namespace OpenMD {
49
50	+	ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
51	+
52	+	// In a parallel computation, row and colum scans must visit all
53	+	// surrounding cells (not just the 14 upper triangular blocks that
54	+	// are used when the processor can see all pairs)
55	+	#ifdef IS_MPI
56	+	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
57	+	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
58	+	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
59	+	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
60	+	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
61	+	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
62	+	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
63	+	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
64	+	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
65	+	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
66	+	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67	+	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
68	+	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
69	+	#endif
70	+	}
71	+
72	+
73		/**
74		* distributeInitialData is essentially a copy of the older fortran
75		* SimulationSetup
76		*/
54	–
77		void ForceMatrixDecomposition::distributeInitialData() {
78		snap_ = sman_->getCurrentSnapshot();
79		storageLayout_ = sman_->getStorageLayout();
80		ff_ = info_->getForceField();
81		nLocal_ = snap_->getNumberOfAtoms();
82	<
82	>
83		nGroups_ = info_->getNLocalCutoffGroups();
84		// gather the information for atomtype IDs (atids):
85	<	identsLocal = info_->getIdentArray();
85	>	idents = info_->getIdentArray();
86		AtomLocalToGlobal = info_->getGlobalAtomIndices();
87		cgLocalToGlobal = info_->getGlobalGroupIndices();
88		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();
89
90	+	massFactors = info_->getMassFactors();
91	+
92	+	PairList* excludes = info_->getExcludedInteractions();
93	+	PairList* oneTwo = info_->getOneTwoInteractions();
94	+	PairList* oneThree = info_->getOneThreeInteractions();
95	+	PairList* oneFour = info_->getOneFourInteractions();
96	+
97		#ifdef IS_MPI
98
99	<	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
100	<	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_);
99	>	MPI::Intracomm row = rowComm.getComm();
100	>	MPI::Intracomm col = colComm.getComm();
101
102	<	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
103	<	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
104	<	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
105	<	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
106	<	AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_);
102	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
103	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
104	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
105	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
106	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
107
108	<	cgCommIntRow = new Communicator<Row,int>(nGroups_);
109	<	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
110	<	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
111	<	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
108	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
109	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
110	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
111	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
112	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
113
114	<	nAtomsInRow_ = AtomCommIntRow->getSize();
115	<	nAtomsInCol_ = AtomCommIntColumn->getSize();
116	<	nGroupsInRow_ = cgCommIntRow->getSize();
117	<	nGroupsInCol_ = cgCommIntColumn->getSize();
114	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
115	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
116	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
117	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
118
119	+	nAtomsInRow_ = AtomPlanIntRow->getSize();
120	+	nAtomsInCol_ = AtomPlanIntColumn->getSize();
121	+	nGroupsInRow_ = cgPlanIntRow->getSize();
122	+	nGroupsInCol_ = cgPlanIntColumn->getSize();
123	+
124		// Modify the data storage objects with the correct layouts and sizes:
125		atomRowData.resize(nAtomsInRow_);
126		atomRowData.setStorageLayout(storageLayout_);
#	Line 107 | Line 134 | namespace OpenMD {
134		identsRow.resize(nAtomsInRow_);
135		identsCol.resize(nAtomsInCol_);
136
137	<	AtomCommIntRow->gather(identsLocal, identsRow);
138	<	AtomCommIntColumn->gather(identsLocal, identsCol);
137	>	AtomPlanIntRow->gather(idents, identsRow);
138	>	AtomPlanIntColumn->gather(idents, identsCol);
139
140	<	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
141	<	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
142	<
116	<	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
117	<	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
140	>	// allocate memory for the parallel objects
141	>	atypesRow.resize(nAtomsInRow_);
142	>	atypesCol.resize(nAtomsInCol_);
143
144	<	AtomCommRealRow->gather(massFactorsLocal, massFactorsRow);
145	<	AtomCommRealColumn->gather(massFactorsLocal, massFactorsCol);
144	>	for (int i = 0; i < nAtomsInRow_; i++)
145	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
146	>	for (int i = 0; i < nAtomsInCol_; i++)
147	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
148
149	+	pot_row.resize(nAtomsInRow_);
150	+	pot_col.resize(nAtomsInCol_);
151	+
152	+	AtomRowToGlobal.resize(nAtomsInRow_);
153	+	AtomColToGlobal.resize(nAtomsInCol_);
154	+	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
155	+	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
156	+
157	+	cerr << "Atoms in Local:\n";
158	+	for (int i = 0; i < AtomLocalToGlobal.size(); i++) {
159	+	cerr << "i =\t" << i << "\t localAt =\t" << AtomLocalToGlobal[i] << "\n";
160	+	}
161	+	cerr << "Atoms in Row:\n";
162	+	for (int i = 0; i < AtomRowToGlobal.size(); i++) {
163	+	cerr << "i =\t" << i << "\t rowAt =\t" << AtomRowToGlobal[i] << "\n";
164	+	}
165	+	cerr << "Atoms in Col:\n";
166	+	for (int i = 0; i < AtomColToGlobal.size(); i++) {
167	+	cerr << "i =\t" << i << "\t colAt =\t" << AtomColToGlobal[i] << "\n";
168	+	}
169	+
170	+	cgRowToGlobal.resize(nGroupsInRow_);
171	+	cgColToGlobal.resize(nGroupsInCol_);
172	+	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
173	+	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
174	+
175	+	cerr << "Gruops in Local:\n";
176	+	for (int i = 0; i < cgLocalToGlobal.size(); i++) {
177	+	cerr << "i =\t" << i << "\t localCG =\t" << cgLocalToGlobal[i] << "\n";
178	+	}
179	+	cerr << "Groups in Row:\n";
180	+	for (int i = 0; i < cgRowToGlobal.size(); i++) {
181	+	cerr << "i =\t" << i << "\t rowCG =\t" << cgRowToGlobal[i] << "\n";
182	+	}
183	+	cerr << "Groups in Col:\n";
184	+	for (int i = 0; i < cgColToGlobal.size(); i++) {
185	+	cerr << "i =\t" << i << "\t colCG =\t" << cgColToGlobal[i] << "\n";
186	+	}
187	+
188	+
189	+	massFactorsRow.resize(nAtomsInRow_);
190	+	massFactorsCol.resize(nAtomsInCol_);
191	+	AtomPlanRealRow->gather(massFactors, massFactorsRow);
192	+	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
193	+
194		groupListRow_.clear();
195		groupListRow_.resize(nGroupsInRow_);
196		for (int i = 0; i < nGroupsInRow_; i++) {
#	Line 141 | Line 213 | namespace OpenMD {
213		}
214		}
215
216	<	skipsForRowAtom.clear();
217	<	skipsForRowAtom.resize(nAtomsInRow_);
216	>	excludesForAtom.clear();
217	>	excludesForAtom.resize(nAtomsInRow_);
218	>	toposForAtom.clear();
219	>	toposForAtom.resize(nAtomsInRow_);
220	>	topoDist.clear();
221	>	topoDist.resize(nAtomsInRow_);
222		for (int i = 0; i < nAtomsInRow_; i++) {
223		int iglob = AtomRowToGlobal[i];
148	–	for (int j = 0; j < nAtomsInCol_; j++) {
149	–	int jglob = AtomColToGlobal[j];
150	–	if (excludes.hasPair(iglob, jglob))
151	–	skipsForRowAtom[i].push_back(j);
152	–	}
153	–	}
224
155	–	toposForRowAtom.clear();
156	–	toposForRowAtom.resize(nAtomsInRow_);
157	–	for (int i = 0; i < nAtomsInRow_; i++) {
158	–	int iglob = AtomRowToGlobal[i];
159	–	int nTopos = 0;
225		for (int j = 0; j < nAtomsInCol_; j++) {
226	<	int jglob = AtomColToGlobal[j];
227	<	if (oneTwo.hasPair(iglob, jglob)) {
228	<	toposForRowAtom[i].push_back(j);
229	<	topoDistRow[i][nTopos] = 1;
230	<	nTopos++;
226	>	int jglob = AtomColToGlobal[j];
227	>
228	>	if (excludes->hasPair(iglob, jglob))
229	>	excludesForAtom[i].push_back(j);
230	>
231	>	if (oneTwo->hasPair(iglob, jglob)) {
232	>	toposForAtom[i].push_back(j);
233	>	topoDist[i].push_back(1);
234	>	} else {
235	>	if (oneThree->hasPair(iglob, jglob)) {
236	>	toposForAtom[i].push_back(j);
237	>	topoDist[i].push_back(2);
238	>	} else {
239	>	if (oneFour->hasPair(iglob, jglob)) {
240	>	toposForAtom[i].push_back(j);
241	>	topoDist[i].push_back(3);
242	>	}
243	>	}
244		}
167	–	if (oneThree.hasPair(iglob, jglob)) {
168	–	toposForRowAtom[i].push_back(j);
169	–	topoDistRow[i][nTopos] = 2;
170	–	nTopos++;
171	–	}
172	–	if (oneFour.hasPair(iglob, jglob)) {
173	–	toposForRowAtom[i].push_back(j);
174	–	topoDistRow[i][nTopos] = 3;
175	–	nTopos++;
176	–	}
245		}
246		}
247
248		#endif
249	+
250	+	// allocate memory for the parallel objects
251	+	atypesLocal.resize(nLocal_);
252	+
253	+	for (int i = 0; i < nLocal_; i++)
254	+	atypesLocal[i] = ff_->getAtomType(idents[i]);
255	+
256		groupList_.clear();
257		groupList_.resize(nGroups_);
258		for (int i = 0; i < nGroups_; i++) {
#	Line 186 | Line 261 | namespace OpenMD {
261		int aid = AtomLocalToGlobal[j];
262		if (globalGroupMembership[aid] == gid) {
263		groupList_[i].push_back(j);
189	–
264		}
265		}
266		}
267
268	<	skipsForLocalAtom.clear();
269	<	skipsForLocalAtom.resize(nLocal_);
268	>	excludesForAtom.clear();
269	>	excludesForAtom.resize(nLocal_);
270	>	toposForAtom.clear();
271	>	toposForAtom.resize(nLocal_);
272	>	topoDist.clear();
273	>	topoDist.resize(nLocal_);
274
275		for (int i = 0; i < nLocal_; i++) {
276		int iglob = AtomLocalToGlobal[i];
277	+
278		for (int j = 0; j < nLocal_; j++) {
279	<	int jglob = AtomLocalToGlobal[j];
280	<	if (excludes.hasPair(iglob, jglob))
281	<	skipsForLocalAtom[i].push_back(j);
282	<	}
283	<	}
284	<	toposForLocalAtom.clear();
285	<	toposForLocalAtom.resize(nLocal_);
286	<	for (int i = 0; i < nLocal_; i++) {
287	<	int iglob = AtomLocalToGlobal[i];
288	<	int nTopos = 0;
289	<	for (int j = 0; j < nLocal_; j++) {
290	<	int jglob = AtomLocalToGlobal[j];
291	<	if (oneTwo.hasPair(iglob, jglob)) {
292	<	toposForLocalAtom[i].push_back(j);
293	<	topoDistLocal[i][nTopos] = 1;
294	<	nTopos++;
279	>	int jglob = AtomLocalToGlobal[j];
280	>
281	>	if (excludes->hasPair(iglob, jglob))
282	>	excludesForAtom[i].push_back(j);
283	>
284	>	if (oneTwo->hasPair(iglob, jglob)) {
285	>	toposForAtom[i].push_back(j);
286	>	topoDist[i].push_back(1);
287	>	} else {
288	>	if (oneThree->hasPair(iglob, jglob)) {
289	>	toposForAtom[i].push_back(j);
290	>	topoDist[i].push_back(2);
291	>	} else {
292	>	if (oneFour->hasPair(iglob, jglob)) {
293	>	toposForAtom[i].push_back(j);
294	>	topoDist[i].push_back(3);
295	>	}
296	>	}
297		}
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	–	}
298		}
299	<	}
299	>	}
300	>
301	>	createGtypeCutoffMap();
302
303		}
304
305		void ForceMatrixDecomposition::createGtypeCutoffMap() {
306	<
306	>
307		RealType tol = 1e-6;
308	+	largestRcut_ = 0.0;
309		RealType rc;
310		int atid;
311		set<AtomType*> atypes = info_->getSimulatedAtomTypes();
312	<	vector<RealType> atypeCutoff;
313	<	atypeCutoff.resize( atypes.size() );
314	<
315	<	for (set<AtomType*>::iterator at = atypes.begin(); at != atypes.end(); ++at){
316	<	rc = interactionMan_->getSuggestedCutoffRadius(*at);
312	>
313	>	map<int, RealType> atypeCutoff;
314	>
315	>	for (set<AtomType*>::iterator at = atypes.begin();
316	>	at != atypes.end(); ++at){
317		atid = (*at)->getIdent();
318	<	atypeCutoff[atid] = rc;
318	>	if (userChoseCutoff_)
319	>	atypeCutoff[atid] = userCutoff_;
320	>	else
321	>	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
322		}
323	<
323	>
324		vector<RealType> gTypeCutoffs;
248	–
325		// first we do a single loop over the cutoff groups to find the
326		// largest cutoff for any atypes present in this group.
327		#ifdef IS_MPI
328		vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
329	+	groupRowToGtype.resize(nGroupsInRow_);
330		for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
331		vector<int> atomListRow = getAtomsInGroupRow(cg1);
332		for (vector<int>::iterator ia = atomListRow.begin();
#	Line 275 | Line 352 | namespace OpenMD {
352
353		}
354		vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
355	+	groupColToGtype.resize(nGroupsInCol_);
356		for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
357		vector<int> atomListCol = getAtomsInGroupColumn(cg2);
358		for (vector<int>::iterator jb = atomListCol.begin();
#	Line 298 | Line 376 | namespace OpenMD {
376		}
377		}
378		#else
379	+
380		vector<RealType> groupCutoff(nGroups_, 0.0);
381	+	groupToGtype.resize(nGroups_);
382		for (int cg1 = 0; cg1 < nGroups_; cg1++) {
383		groupCutoff[cg1] = 0.0;
384		vector<int> atomList = getAtomsInGroupRow(cg1);
385		for (vector<int>::iterator ia = atomList.begin();
386		ia != atomList.end(); ++ia) {
387		int atom1 = (*ia);
388	<	atid = identsLocal[atom1];
389	<	if (atypeCutoff[atid] > groupCutoff[cg1]) {
388	>	atid = idents[atom1];
389	>	if (atypeCutoff[atid] > groupCutoff[cg1])
390		groupCutoff[cg1] = atypeCutoff[atid];
311	–	}
391		}
392	<
392	>
393		bool gTypeFound = false;
394		for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
395		if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
#	Line 318 | Line 397 | namespace OpenMD {
397		gTypeFound = true;
398		}
399		}
400	<	if (!gTypeFound) {
400	>	if (!gTypeFound) {
401		gTypeCutoffs.push_back( groupCutoff[cg1] );
402		groupToGtype[cg1] = gTypeCutoffs.size() - 1;
403		}
#	Line 327 | Line 406 | namespace OpenMD {
406
407		// Now we find the maximum group cutoff value present in the simulation
408
409	<	vector<RealType>::iterator groupMaxLoc = max_element(gTypeCutoffs.begin(), gTypeCutoffs.end());
410	<	RealType groupMax = *groupMaxLoc;
409	>	RealType groupMax = *max_element(gTypeCutoffs.begin(),
410	>	gTypeCutoffs.end());
411
412		#ifdef IS_MPI
413	<	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX);
413	>	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
414	>	MPI::MAX);
415		#endif
416
417		RealType tradRcut = groupMax;
418
419		for (int i = 0; i < gTypeCutoffs.size();  i++) {
420	<	for (int j = 0; j < gTypeCutoffs.size();  j++) {
341	<
420	>	for (int j = 0; j < gTypeCutoffs.size();  j++) {
421		RealType thisRcut;
422		switch(cutoffPolicy_) {
423		case TRADITIONAL:
424		thisRcut = tradRcut;
425	+	break;
426		case MIX:
427		thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
428	+	break;
429		case MAX:
430		thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
431	+	break;
432		default:
433		sprintf(painCave.errMsg,
434		"ForceMatrixDecomposition::createGtypeCutoffMap "
435		"hit an unknown cutoff policy!\n");
436		painCave.severity = OPENMD_ERROR;
437		painCave.isFatal = 1;
438	<	simError();
438	>	simError();
439	>	break;
440		}
441
442		pair<int,int> key = make_pair(i,j);
443		gTypeCutoffMap[key].first = thisRcut;
361	–
444		if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
363	–
445		gTypeCutoffMap[key].second = thisRcut*thisRcut;
365	–
446		gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
367	–
447		// sanity check
448
449		if (userChoseCutoff_) {
450		if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
451		sprintf(painCave.errMsg,
452		"ForceMatrixDecomposition::createGtypeCutoffMap "
453	<	"user-specified rCut does not match computed group Cutoff\n");
453	>	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
454		painCave.severity = OPENMD_ERROR;
455		painCave.isFatal = 1;
456		simError();
#	Line 383 | Line 462 | namespace OpenMD {
462
463
464		groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
465	<	int i, j;
387	<
465	>	int i, j;
466		#ifdef IS_MPI
467		i = groupRowToGtype[cg1];
468		j = groupColToGtype[cg2];
469		#else
470		i = groupToGtype[cg1];
471		j = groupToGtype[cg2];
472	<	#endif
395	<
472	>	#endif
473		return gTypeCutoffMap[make_pair(i,j)];
474		}
475
476	+	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
477	+	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
478	+	if (toposForAtom[atom1][j] == atom2)
479	+	return topoDist[atom1][j];
480	+	}
481	+	return 0;
482	+	}
483
484		void ForceMatrixDecomposition::zeroWorkArrays() {
485	+	pairwisePot = 0.0;
486	+	embeddingPot = 0.0;
487
402	–	for (int j = 0; j < N_INTERACTION_FAMILIES; j++) {
403	–	longRangePot_[j] = 0.0;
404	–	}
405	–
488		#ifdef IS_MPI
489		if (storageLayout_ & DataStorage::dslForce) {
490		fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
#	Line 418 | Line 500 | namespace OpenMD {
500		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
501
502		fill(pot_col.begin(), pot_col.end(),
503	<	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
422	<
423	<	pot_local = Vector<RealType, N_INTERACTION_FAMILIES>(0.0);
503	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
504
505		if (storageLayout_ & DataStorage::dslParticlePot) {
506	<	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0);
507	<	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0);
506	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
507	>	0.0);
508	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
509	>	0.0);
510		}
511
512		if (storageLayout_ & DataStorage::dslDensity) {
#	Line 433 | Line 515 | namespace OpenMD {
515		}
516
517		if (storageLayout_ & DataStorage::dslFunctional) {
518	<	fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0);
519	<	fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0);
518	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
519	>	0.0);
520	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
521	>	0.0);
522		}
523
524		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
#	Line 444 | Line 528 | namespace OpenMD {
528		atomColData.functionalDerivative.end(), 0.0);
529		}
530
531	<	#else
532	<
531	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
532	>	fill(atomRowData.skippedCharge.begin(),
533	>	atomRowData.skippedCharge.end(), 0.0);
534	>	fill(atomColData.skippedCharge.begin(),
535	>	atomColData.skippedCharge.end(), 0.0);
536	>	}
537	>
538	>	#endif
539	>	// even in parallel, we need to zero out the local arrays:
540	>
541		if (storageLayout_ & DataStorage::dslParticlePot) {
542		fill(snap_->atomData.particlePot.begin(),
543		snap_->atomData.particlePot.end(), 0.0);
#	Line 463 | Line 555 | namespace OpenMD {
555		fill(snap_->atomData.functionalDerivative.begin(),
556		snap_->atomData.functionalDerivative.end(), 0.0);
557		}
558	<	#endif
558	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
559	>	fill(snap_->atomData.skippedCharge.begin(),
560	>	snap_->atomData.skippedCharge.end(), 0.0);
561	>	}
562
563		}
564
#	Line 474 | Line 569 | namespace OpenMD {
569		#ifdef IS_MPI
570
571		// gather up the atomic positions
572	<	AtomCommVectorRow->gather(snap_->atomData.position,
572	>	AtomPlanVectorRow->gather(snap_->atomData.position,
573		atomRowData.position);
574	<	AtomCommVectorColumn->gather(snap_->atomData.position,
574	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
575		atomColData.position);
576
577		// gather up the cutoff group positions
578	<	cgCommVectorRow->gather(snap_->cgData.position,
578	>
579	>	cerr  << "before gather\n";
580	>	for (int i = 0; i < snap_->cgData.position.size(); i++) {
581	>	cerr << "cgpos = " << snap_->cgData.position[i] << "\n";
582	>	}
583	>
584	>	cgPlanVectorRow->gather(snap_->cgData.position,
585		cgRowData.position);
586	<	cgCommVectorColumn->gather(snap_->cgData.position,
586	>
587	>	cerr  << "after gather\n";
588	>	for (int i = 0; i < cgRowData.position.size(); i++) {
589	>	cerr << "cgRpos = " << cgRowData.position[i] << "\n";
590	>	}
591	>
592	>	cgPlanVectorColumn->gather(snap_->cgData.position,
593		cgColData.position);
594	+	for (int i = 0; i < cgColData.position.size(); i++) {
595	+	cerr << "cgCpos = " << cgColData.position[i] << "\n";
596	+	}
597	+
598
599		// if needed, gather the atomic rotation matrices
600		if (storageLayout_ & DataStorage::dslAmat) {
601	<	AtomCommMatrixRow->gather(snap_->atomData.aMat,
601	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
602		atomRowData.aMat);
603	<	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
603	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
604		atomColData.aMat);
605		}
606
607		// if needed, gather the atomic eletrostatic frames
608		if (storageLayout_ & DataStorage::dslElectroFrame) {
609	<	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
609	>	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
610		atomRowData.electroFrame);
611	<	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
611	>	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
612		atomColData.electroFrame);
613		}
614	+
615		#endif
616		}
617
#	Line 513 | Line 625 | namespace OpenMD {
625
626		if (storageLayout_ & DataStorage::dslDensity) {
627
628	<	AtomCommRealRow->scatter(atomRowData.density,
628	>	AtomPlanRealRow->scatter(atomRowData.density,
629		snap_->atomData.density);
630
631		int n = snap_->atomData.density.size();
632		vector<RealType> rho_tmp(n, 0.0);
633	<	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
633	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
634		for (int i = 0; i < n; i++)
635		snap_->atomData.density[i] += rho_tmp[i];
636		}
#	Line 534 | Line 646 | namespace OpenMD {
646		storageLayout_ = sman_->getStorageLayout();
647		#ifdef IS_MPI
648		if (storageLayout_ & DataStorage::dslFunctional) {
649	<	AtomCommRealRow->gather(snap_->atomData.functional,
649	>	AtomPlanRealRow->gather(snap_->atomData.functional,
650		atomRowData.functional);
651	<	AtomCommRealColumn->gather(snap_->atomData.functional,
651	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
652		atomColData.functional);
653		}
654
655		if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
656	<	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
656	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
657		atomRowData.functionalDerivative);
658	<	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
658	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
659		atomColData.functionalDerivative);
660		}
661		#endif
#	Line 557 | Line 669 | namespace OpenMD {
669		int n = snap_->atomData.force.size();
670		vector<Vector3d> frc_tmp(n, V3Zero);
671
672	<	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
672	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
673		for (int i = 0; i < n; i++) {
674		snap_->atomData.force[i] += frc_tmp[i];
675		frc_tmp[i] = 0.0;
676		}
677
678	<	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
679	<	for (int i = 0; i < n; i++)
678	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
679	>	for (int i = 0; i < n; i++) {
680		snap_->atomData.force[i] += frc_tmp[i];
681	<
682	<
681	>	}
682	>
683		if (storageLayout_ & DataStorage::dslTorque) {
684
685	<	int nt = snap_->atomData.force.size();
685	>	int nt = snap_->atomData.torque.size();
686		vector<Vector3d> trq_tmp(nt, V3Zero);
687
688	<	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
689	<	for (int i = 0; i < n; i++) {
688	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
689	>	for (int i = 0; i < nt; i++) {
690		snap_->atomData.torque[i] += trq_tmp[i];
691		trq_tmp[i] = 0.0;
692		}
693
694	<	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
695	<	for (int i = 0; i < n; i++)
694	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
695	>	for (int i = 0; i < nt; i++)
696		snap_->atomData.torque[i] += trq_tmp[i];
697		}
698	+
699	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
700	+
701	+	int ns = snap_->atomData.skippedCharge.size();
702	+	vector<RealType> skch_tmp(ns, 0.0);
703	+
704	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
705	+	for (int i = 0; i < ns; i++) {
706	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
707	+	skch_tmp[i] = 0.0;
708	+	}
709	+
710	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
711	+	for (int i = 0; i < ns; i++)
712	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
713	+	}
714
715		nLocal_ = snap_->getNumberOfAtoms();
716
#	Line 591 | Line 719 | namespace OpenMD {
719
720		// scatter/gather pot_row into the members of my column
721
722	<	AtomCommPotRow->scatter(pot_row, pot_temp);
722	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
723
724		for (int ii = 0;  ii < pot_temp.size(); ii++ )
725	<	pot_local += pot_temp[ii];
725	>	pairwisePot += pot_temp[ii];
726
727		fill(pot_temp.begin(), pot_temp.end(),
728		Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
729
730	<	AtomCommPotColumn->scatter(pot_col, pot_temp);
730	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
731
732		for (int ii = 0;  ii < pot_temp.size(); ii++ )
733	<	pot_local += pot_temp[ii];
606	<
733	>	pairwisePot += pot_temp[ii];
734		#endif
735	+
736	+	cerr << "pairwisePot = " <<  pairwisePot << "\n";
737		}
738
739		int ForceMatrixDecomposition::getNAtomsInRow() {
#	Line 639 | Line 768 | namespace OpenMD {
768
769		#ifdef IS_MPI
770		d = cgColData.position[cg2] - cgRowData.position[cg1];
771	+	cerr << "cg1 = " << cg1 << "\tcg1p = " << cgRowData.position[cg1] << "\n";
772	+	cerr << "cg2 = " << cg2 << "\tcg2p = " << cgColData.position[cg2] << "\n";
773		#else
774		d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
775	+	cerr << "cg1 = " << cg1 << "\tcg1p = " << snap_->cgData.position[cg1] << "\n";
776	+	cerr << "cg2 = " << cg2 << "\tcg2p = " << snap_->cgData.position[cg2] << "\n";
777		#endif
778
779		snap_->wrapVector(d);
#	Line 679 | Line 812 | namespace OpenMD {
812		#ifdef IS_MPI
813		return massFactorsRow[atom1];
814		#else
815	<	return massFactorsLocal[atom1];
815	>	return massFactors[atom1];
816		#endif
817		}
818
#	Line 687 | Line 820 | namespace OpenMD {
820		#ifdef IS_MPI
821		return massFactorsCol[atom2];
822		#else
823	<	return massFactorsLocal[atom2];
823	>	return massFactors[atom2];
824		#endif
825
826		}
#	Line 705 | Line 838 | namespace OpenMD {
838		return d;
839		}
840
841	<	vector<int> ForceMatrixDecomposition::getSkipsForRowAtom(int atom1) {
842	<	#ifdef IS_MPI
710	<	return skipsForRowAtom[atom1];
711	<	#else
712	<	return skipsForLocalAtom[atom1];
713	<	#endif
841	>	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
842	>	return excludesForAtom[atom1];
843		}
844
845		/**
846	<	* There are a number of reasons to skip a pair or a
718	<	* particle. Mostly we do this to exclude atoms who are involved in
719	<	* short range interactions (bonds, bends, torsions), but we also
720	<	* need to exclude some overcounted interactions that result from
846	>	* We need to exclude some overcounted interactions that result from
847		* the parallel decomposition.
848		*/
849		bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
850		int unique_id_1, unique_id_2;
851	+
852
853	+	cerr << "sap with atom1, atom2 =\t" << atom1 << "\t" << atom2 << "\n";
854		#ifdef IS_MPI
855		// in MPI, we have to look up the unique IDs for each atom
856		unique_id_1 = AtomRowToGlobal[atom1];
857		unique_id_2 = AtomColToGlobal[atom2];
858
859	+	cerr << "sap with uid1, uid2 =\t" << unique_id_1 << "\t" << unique_id_2 << "\n";
860		// this situation should only arise in MPI simulations
861		if (unique_id_1 == unique_id_2) return true;
862
#	Line 737 | Line 866 | namespace OpenMD {
866		} else {
867		if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
868		}
869	+	#endif
870	+	return false;
871	+	}
872	+
873	+	/**
874	+	* We need to handle the interactions for atoms who are involved in
875	+	* the same rigid body as well as some short range interactions
876	+	* (bonds, bends, torsions) differently from other interactions.
877	+	* We'll still visit the pairwise routines, but with a flag that
878	+	* tells those routines to exclude the pair from direct long range
879	+	* interactions.  Some indirect interactions (notably reaction
880	+	* field) must still be handled for these pairs.
881	+	*/
882	+	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
883	+	int unique_id_2;
884	+	#ifdef IS_MPI
885	+	// in MPI, we have to look up the unique IDs for the row atom.
886	+	unique_id_2 = AtomColToGlobal[atom2];
887		#else
888		// in the normal loop, the atom numbers are unique
742	–	unique_id_1 = atom1;
889		unique_id_2 = atom2;
890		#endif
891
892	<	#ifdef IS_MPI
893	<	for (vector<int>::iterator i = skipsForRowAtom[atom1].begin();
748	<	i != skipsForRowAtom[atom1].end(); ++i) {
892	>	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
893	>	i != excludesForAtom[atom1].end(); ++i) {
894		if ( (*i) == unique_id_2 ) return true;
750	–	}
751	–	#else
752	–	for (vector<int>::iterator i = skipsForLocalAtom[atom1].begin();
753	–	i != skipsForLocalAtom[atom1].end(); ++i) {
754	–	if ( (*i) == unique_id_2 ) return true;
755	–	}
756	–	#endif
757	–	}
758	–
759	–	int ForceMatrixDecomposition::getTopoDistance(int atom1, int atom2) {
760	–
761	–	#ifdef IS_MPI
762	–	for (int i = 0; i < toposForRowAtom[atom1].size(); i++) {
763	–	if ( toposForRowAtom[atom1][i] == atom2 ) return topoDistRow[atom1][i];
895		}
896	<	#else
897	<	for (int i = 0; i < toposForLocalAtom[atom1].size(); i++) {
767	<	if ( toposForLocalAtom[atom1][i] == atom2 ) return topoDistLocal[atom1][i];
768	<	}
769	<	#endif
770	<
771	<	// zero is default for unconnected (i.e. normal) pair interactions
772	<	return 0;
896	>
897	>	return false;
898		}
899
900	+
901		void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
902		#ifdef IS_MPI
903		atomRowData.force[atom1] += fg;
#	Line 789 | Line 915 | namespace OpenMD {
915		}
916
917		// filling interaction blocks with pointers
918	<	InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {
919	<	InteractionData idat;
918	>	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
919	>	int atom1, int atom2) {
920
921	+	idat.excluded = excludeAtomPair(atom1, atom2);
922	+
923		#ifdef IS_MPI
924	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
925	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
926	+	//                         ff_->getAtomType(identsCol[atom2]) );
927
797	–	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
798	–	ff_->getAtomType(identsCol[atom2]) );
799	–
800	–
928		if (storageLayout_ & DataStorage::dslAmat) {
929		idat.A1 = &(atomRowData.aMat[atom1]);
930		idat.A2 = &(atomColData.aMat[atom2]);
#	Line 833 | Line 960 | namespace OpenMD {
960		idat.particlePot2 = &(atomColData.particlePot[atom2]);
961		}
962
963	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
964	+	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
965	+	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
966	+	}
967	+
968		#else
969
970	<	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
971	<	ff_->getAtomType(identsLocal[atom2]) );
970	>	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
971	>	//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
972	>	//                         ff_->getAtomType(idents[atom2]) );
973
974		if (storageLayout_ & DataStorage::dslAmat) {
975		idat.A1 = &(snap_->atomData.aMat[atom1]);
#	Line 853 | Line 986 | namespace OpenMD {
986		idat.t2 = &(snap_->atomData.torque[atom2]);
987		}
988
989	<	if (storageLayout_ & DataStorage::dslDensity) {
989	>	if (storageLayout_ & DataStorage::dslDensity) {
990		idat.rho1 = &(snap_->atomData.density[atom1]);
991		idat.rho2 = &(snap_->atomData.density[atom2]);
992		}
#	Line 873 | Line 1006 | namespace OpenMD {
1006		idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1007		}
1008
1009	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1010	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1011	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1012	+	}
1013		#endif
877	–	return idat;
1014		}
1015
1016
1017	<	void ForceMatrixDecomposition::unpackInteractionData(InteractionData idat, int atom1, int atom2) {
1017	>	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1018		#ifdef IS_MPI
1019		pot_row[atom1] += 0.5 *  *(idat.pot);
1020		pot_col[atom2] += 0.5 *  *(idat.pot);
#	Line 886 | Line 1022 | namespace OpenMD {
1022		atomRowData.force[atom1] += *(idat.f1);
1023		atomColData.force[atom2] -= *(idat.f1);
1024		#else
1025	<	longRangePot_ += *(idat.pot);
1026	<
1025	>	pairwisePot += *(idat.pot);
1026	>
1027		snap_->atomData.force[atom1] += *(idat.f1);
1028		snap_->atomData.force[atom2] -= *(idat.f1);
1029		#endif
1030	<
1030	>
1031		}
1032
1033	+	vector<vector<int> > ForceMatrixDecomposition::buildLayerBasedNeighborList() {
1034	+	printf("buildLayerBasedNeighborList; nGroups:%d\n", nGroups_);
1035	+	// Na = nGroups_
1036	+	/* cell occupancy counter */
1037	+	vector<int> k_c;
1038	+	/* c_i - has cell containing atom i (size Na) */
1039	+	vector<int> c;
1040	+	/* l_i - layer containing atom i (size Na) */
1041	+	vector<int> l;
1042
1043	<	InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
1043	>	//      cellList_.clear();
1044
1045	<	InteractionData idat;
1046	<	#ifdef IS_MPI
1047	<	idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1048	<	ff_->getAtomType(identsCol[atom2]) );
1045	>	RealType rList_ = (largestRcut_ + skinThickness_);
1046	>	Snapshot* snap_ = sman_->getCurrentSnapshot();
1047	>	Mat3x3d Hmat = snap_->getHmat();
1048	>	Vector3d Hx = Hmat.getColumn(0);
1049	>	Vector3d Hy = Hmat.getColumn(1);
1050	>	Vector3d Hz = Hmat.getColumn(2);
1051
1052	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1053	<	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
1054	<	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
908	<	}
909	<	if (storageLayout_ & DataStorage::dslTorque) {
910	<	idat.t1 = &(atomRowData.torque[atom1]);
911	<	idat.t2 = &(atomColData.torque[atom2]);
912	<	}
913	<	#else
914	<	idat.atypes = make_pair( ff_->getAtomType(identsLocal[atom1]),
915	<	ff_->getAtomType(identsLocal[atom2]) );
1052	>	nCells_.x() = (int) (Hx.length()) / rList_;
1053	>	nCells_.y() = (int) (Hy.length()) / rList_;
1054	>	nCells_.z() = (int) (Hz.length()) / rList_;
1055
1056	<	if (storageLayout_ & DataStorage::dslElectroFrame) {
1057	<	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1058	<	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1059	<	}
1060	<	if (storageLayout_ & DataStorage::dslTorque) {
922	<	idat.t1 = &(snap_->atomData.torque[atom1]);
923	<	idat.t2 = &(snap_->atomData.torque[atom2]);
924	<	}
925	<	#endif
926	<	}
1056	>	Mat3x3d invHmat = snap_->getInvHmat();
1057	>	Vector3d rs, scaled, dr;
1058	>	Vector3i whichCell;
1059	>	int cellIndex;
1060	>	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1061
1062	+	//      cellList_.resize(nCtot);
1063	+	k_c = vector<int>(nCtot, 0);
1064	+
1065	+	for (int i = 0; i < nGroups_; i++)
1066	+	{
1067	+	rs = snap_->cgData.position[i];
1068	+
1069	+	// scaled positions relative to the box vectors
1070	+	scaled = invHmat * rs;
1071	+
1072	+	// wrap the vector back into the unit box by subtracting integer box
1073	+	// numbers
1074	+	for (int j = 0; j < 3; j++)
1075	+	{
1076	+	scaled[j] -= roundMe(scaled[j]);
1077	+	scaled[j] += 0.5;
1078	+	}
1079	+
1080	+	// find xyz-indices of cell that cutoffGroup is in.
1081	+	whichCell.x() = nCells_.x() * scaled.x();
1082	+	whichCell.y() = nCells_.y() * scaled.y();
1083	+	whichCell.z() = nCells_.z() * scaled.z();
1084	+
1085	+	// find single index of this cell:
1086	+	cellIndex = Vlinear(whichCell, nCells_);
1087	+
1088	+	c.push_back(cellIndex);
1089	+
1090	+	//              // add this cutoff group to the list of groups in this cell;
1091	+	//              cellList_[cellIndex].push_back(i);
1092	+	}
1093	+
1094	+	int k_c_curr;
1095	+	int k_c_max = 0;
1096	+	/* the cell-layer occupancy matrix */
1097	+	vector<vector<int> > H_c_l = vector<vector<int> >(nCtot);
1098	+
1099	+	for(int i = 0; i < nGroups_; ++i)
1100	+	{
1101	+	k_c_curr = ++k_c[c[i]];
1102	+	l.push_back(k_c_curr);
1103	+
1104	+	/* determines the number of layers in use */
1105	+	if(k_c_max < k_c_curr)
1106	+	{
1107	+	k_c_max = k_c_curr;
1108	+	}
1109	+
1110	+	H_c_l[c[i]].push_back(/*l[*/i/*]*/);
1111	+	}
1112	+
1113	+	int m;
1114	+	/* the neighbor matrix */
1115	+	vector<vector<int> >neighborMatW = vector<vector<int> >(nGroups_);
1116	+
1117	+	//      vector<pair<int, int> > neighborList;
1118	+	groupCutoffs cuts;
1119	+
1120	+	/* loops over objects(atoms, rigidBodies, cutoffGroups, etc.) */
1121	+	for(int i = 0; i < nGroups_; ++i)
1122	+	{
1123	+	m = 0;
1124	+	/* c' */
1125	+	int c1 = c[i];
1126	+	Vector3i c1v = idxToV(c1, nCells_);
1127	+
1128	+	/* loops over the neighboring cells c'' */
1129	+	for (vector<Vector3i>::iterator os = cellOffsets_.begin(); os != cellOffsets_.end(); ++os)
1130	+	{
1131	+	Vector3i c2v = c1v + (*os);
1132	+
1133	+	if (c2v.x() >= nCells_.x())
1134	+	{
1135	+	c2v.x() = 0;
1136	+	} else if (c2v.x() < 0)
1137	+	{
1138	+	c2v.x() = nCells_.x() - 1;
1139	+	}
1140	+
1141	+	if (c2v.y() >= nCells_.y())
1142	+	{
1143	+	c2v.y() = 0;
1144	+	} else if (c2v.y() < 0)
1145	+	{
1146	+	c2v.y() = nCells_.y() - 1;
1147	+	}
1148	+
1149	+	if (c2v.z() >= nCells_.z())
1150	+	{
1151	+	c2v.z() = 0;
1152	+	} else if (c2v.z() < 0)
1153	+	{
1154	+	c2v.z() = nCells_.z() - 1;
1155	+	}
1156	+
1157	+	int c2 = Vlinear(c2v, nCells_);
1158	+	/* loops over layers l to access the neighbor atoms */
1159	+	for (vector<int>::iterator j = H_c_l[c2].begin(); j != H_c_l[c2].end(); ++j)
1160	+	{
1161	+	//                              if i'' = 0 then break // doesn't apply to vector implementation of matrix
1162	+	//                              if(i != *j)
1163	+	if (c2 != c1 || (*j) < (i))
1164	+	{
1165	+	dr = snap_->cgData.position[(*j)] - snap_->cgData.position[(i)];
1166	+	snap_->wrapVector(dr);
1167	+	cuts = getGroupCutoffs((i), (*j));
1168	+	if (dr.lengthSquare() < cuts.third)
1169	+	{
1170	+	++m;
1171	+	/* transposed version of Rapaport W mat, to occupy successive memory locations on CPU */
1172	+	neighborMatW[i].push_back(*j);
1173	+	//                                              neighborList.push_back(make_pair((i), (*j)));
1174	+	}
1175	+	}
1176	+	}
1177	+	}
1178	+	}
1179	+
1180	+	// save the local cutoff group positions for the check that is
1181	+	// done on each loop:
1182	+	saved_CG_positions_.clear();
1183	+	for (int i = 0; i < nGroups_; i++)
1184	+	saved_CG_positions_.push_back(snap_->cgData.position[i]);
1185	+
1186	+	return neighborMatW;
1187	+	}
1188	+
1189		/*
1190		* buildNeighborList
1191		*
#	Line 935 | Line 1196 | namespace OpenMD {
1196
1197		vector<pair<int, int> > neighborList;
1198		groupCutoffs cuts;
1199	+	bool doAllPairs = false;
1200	+
1201		#ifdef IS_MPI
1202		cellListRow_.clear();
1203		cellListCol_.clear();
#	Line 954 | Line 1217 | namespace OpenMD {
1217		nCells_.y() = (int) ( Hy.length() )/ rList_;
1218		nCells_.z() = (int) ( Hz.length() )/ rList_;
1219
1220	+	// handle small boxes where the cell offsets can end up repeating cells
1221	+
1222	+	if (nCells_.x() < 3) doAllPairs = true;
1223	+	if (nCells_.y() < 3) doAllPairs = true;
1224	+	if (nCells_.z() < 3) doAllPairs = true;
1225	+
1226		Mat3x3d invHmat = snap_->getInvHmat();
1227		Vector3d rs, scaled, dr;
1228		Vector3i whichCell;
1229		int cellIndex;
1230	+	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1231
1232		#ifdef IS_MPI
1233	<	for (int i = 0; i < nGroupsInRow_; i++) {
1234	<	rs = cgRowData.position[i];
1235	<	// scaled positions relative to the box vectors
1236	<	scaled = invHmat * rs;
1237	<	// wrap the vector back into the unit box by subtracting integer box
968	<	// numbers
969	<	for (int j = 0; j < 3; j++)
970	<	scaled[j] -= roundMe(scaled[j]);
971	<
972	<	// find xyz-indices of cell that cutoffGroup is in.
973	<	whichCell.x() = nCells_.x() * scaled.x();
974	<	whichCell.y() = nCells_.y() * scaled.y();
975	<	whichCell.z() = nCells_.z() * scaled.z();
1233	>	cellListRow_.resize(nCtot);
1234	>	cellListCol_.resize(nCtot);
1235	>	#else
1236	>	cellList_.resize(nCtot);
1237	>	#endif
1238
1239	<	// find single index of this cell:
1240	<	cellIndex = Vlinear(whichCell, nCells_);
979	<	// add this cutoff group to the list of groups in this cell;
980	<	cellListRow_[cellIndex].push_back(i);
981	<	}
1239	>	if (!doAllPairs) {
1240	>	#ifdef IS_MPI
1241
1242	<	for (int i = 0; i < nGroupsInCol_; i++) {
1243	<	rs = cgColData.position[i];
1244	<	// scaled positions relative to the box vectors
1245	<	scaled = invHmat * rs;
1246	<	// wrap the vector back into the unit box by subtracting integer box
1247	<	// numbers
1248	<	for (int j = 0; j < 3; j++)
1249	<	scaled[j] -= roundMe(scaled[j]);
1250	<
1251	<	// find xyz-indices of cell that cutoffGroup is in.
1252	<	whichCell.x() = nCells_.x() * scaled.x();
1253	<	whichCell.y() = nCells_.y() * scaled.y();
1254	<	whichCell.z() = nCells_.z() * scaled.z();
1255	<
1256	<	// find single index of this cell:
1257	<	cellIndex = Vlinear(whichCell, nCells_);
1258	<	// add this cutoff group to the list of groups in this cell;
1259	<	cellListCol_[cellIndex].push_back(i);
1260	<	}
1242	>	for (int i = 0; i < nGroupsInRow_; i++) {
1243	>	rs = cgRowData.position[i];
1244	>
1245	>	// scaled positions relative to the box vectors
1246	>	scaled = invHmat * rs;
1247	>
1248	>	// wrap the vector back into the unit box by subtracting integer box
1249	>	// numbers
1250	>	for (int j = 0; j < 3; j++) {
1251	>	scaled[j] -= roundMe(scaled[j]);
1252	>	scaled[j] += 0.5;
1253	>	}
1254	>
1255	>	// find xyz-indices of cell that cutoffGroup is in.
1256	>	whichCell.x() = nCells_.x() * scaled.x();
1257	>	whichCell.y() = nCells_.y() * scaled.y();
1258	>	whichCell.z() = nCells_.z() * scaled.z();
1259	>
1260	>	// find single index of this cell:
1261	>	cellIndex = Vlinear(whichCell, nCells_);
1262	>
1263	>	// add this cutoff group to the list of groups in this cell;
1264	>	cellListRow_[cellIndex].push_back(i);
1265	>	}
1266	>	for (int i = 0; i < nGroupsInCol_; i++) {
1267	>	rs = cgColData.position[i];
1268	>
1269	>	// scaled positions relative to the box vectors
1270	>	scaled = invHmat * rs;
1271	>
1272	>	// wrap the vector back into the unit box by subtracting integer box
1273	>	// numbers
1274	>	for (int j = 0; j < 3; j++) {
1275	>	scaled[j] -= roundMe(scaled[j]);
1276	>	scaled[j] += 0.5;
1277	>	}
1278	>
1279	>	// find xyz-indices of cell that cutoffGroup is in.
1280	>	whichCell.x() = nCells_.x() * scaled.x();
1281	>	whichCell.y() = nCells_.y() * scaled.y();
1282	>	whichCell.z() = nCells_.z() * scaled.z();
1283	>
1284	>	// find single index of this cell:
1285	>	cellIndex = Vlinear(whichCell, nCells_);
1286	>
1287	>	// add this cutoff group to the list of groups in this cell;
1288	>	cellListCol_[cellIndex].push_back(i);
1289	>	}
1290		#else
1291	<	for (int i = 0; i < nGroups_; i++) {
1292	<	rs = snap_->cgData.position[i];
1293	<	// scaled positions relative to the box vectors
1294	<	scaled = invHmat * rs;
1295	<	// wrap the vector back into the unit box by subtracting integer box
1296	<	// numbers
1297	<	for (int j = 0; j < 3; j++)
1298	<	scaled[j] -= roundMe(scaled[j]);
1299	<
1300	<	// find xyz-indices of cell that cutoffGroup is in.
1301	<	whichCell.x() = nCells_.x() * scaled.x();
1302	<	whichCell.y() = nCells_.y() * scaled.y();
1303	<	whichCell.z() = nCells_.z() * scaled.z();
1304	<
1305	<	// find single index of this cell:
1306	<	cellIndex = Vlinear(whichCell, nCells_);
1307	<	// add this cutoff group to the list of groups in this cell;
1308	<	cellList_[cellIndex].push_back(i);
1309	<	}
1291	>	for (int i = 0; i < nGroups_; i++) {
1292	>	rs = snap_->cgData.position[i];
1293	>
1294	>	// scaled positions relative to the box vectors
1295	>	scaled = invHmat * rs;
1296	>
1297	>	// wrap the vector back into the unit box by subtracting integer box
1298	>	// numbers
1299	>	for (int j = 0; j < 3; j++) {
1300	>	scaled[j] -= roundMe(scaled[j]);
1301	>	scaled[j] += 0.5;
1302	>	}
1303	>
1304	>	// find xyz-indices of cell that cutoffGroup is in.
1305	>	whichCell.x() = nCells_.x() * scaled.x();
1306	>	whichCell.y() = nCells_.y() * scaled.y();
1307	>	whichCell.z() = nCells_.z() * scaled.z();
1308	>
1309	>	// find single index of this cell:
1310	>	cellIndex = Vlinear(whichCell, nCells_);
1311	>
1312	>	// add this cutoff group to the list of groups in this cell;
1313	>	cellList_[cellIndex].push_back(i);
1314	>	}
1315		#endif
1316
1317	<	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1318	<	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1319	<	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1320	<	Vector3i m1v(m1x, m1y, m1z);
1321	<	int m1 = Vlinear(m1v, nCells_);
1029	<
1030	<	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1031	<	os != cellOffsets_.end(); ++os) {
1317	>	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1318	>	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1319	>	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1320	>	Vector3i m1v(m1x, m1y, m1z);
1321	>	int m1 = Vlinear(m1v, nCells_);
1322
1323	<	Vector3i m2v = m1v + (*os);
1324	<
1325	<	if (m2v.x() >= nCells_.x()) {
1326	<	m2v.x() = 0;
1327	<	} else if (m2v.x() < 0) {
1328	<	m2v.x() = nCells_.x() - 1;
1329	<	}
1330	<
1331	<	if (m2v.y() >= nCells_.y()) {
1332	<	m2v.y() = 0;
1333	<	} else if (m2v.y() < 0) {
1334	<	m2v.y() = nCells_.y() - 1;
1335	<	}
1336	<
1337	<	if (m2v.z() >= nCells_.z()) {
1338	<	m2v.z() = 0;
1339	<	} else if (m2v.z() < 0) {
1340	<	m2v.z() = nCells_.z() - 1;
1341	<	}
1342	<
1343	<	int m2 = Vlinear (m2v, nCells_);
1344	<
1323	>	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1324	>	os != cellOffsets_.end(); ++os) {
1325	>
1326	>	Vector3i m2v = m1v + (*os);
1327	>
1328	>	if (m2v.x() >= nCells_.x()) {
1329	>	m2v.x() = 0;
1330	>	} else if (m2v.x() < 0) {
1331	>	m2v.x() = nCells_.x() - 1;
1332	>	}
1333	>
1334	>	if (m2v.y() >= nCells_.y()) {
1335	>	m2v.y() = 0;
1336	>	} else if (m2v.y() < 0) {
1337	>	m2v.y() = nCells_.y() - 1;
1338	>	}
1339	>
1340	>	if (m2v.z() >= nCells_.z()) {
1341	>	m2v.z() = 0;
1342	>	} else if (m2v.z() < 0) {
1343	>	m2v.z() = nCells_.z() - 1;
1344	>	}
1345	>
1346	>	int m2 = Vlinear (m2v, nCells_);
1347	>
1348		#ifdef IS_MPI
1349	<	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1350	<	j1 != cellListRow_[m1].end(); ++j1) {
1351	<	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1352	<	j2 != cellListCol_[m2].end(); ++j2) {
1353	<
1354	<	// Always do this if we're in different cells or if
1355	<	// we're in the same cell and the global index of the
1063	<	// j2 cutoff group is less than the j1 cutoff group
1064	<
1065	<	if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) {
1349	>	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1350	>	j1 != cellListRow_[m1].end(); ++j1) {
1351	>	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1352	>	j2 != cellListCol_[m2].end(); ++j2) {
1353	>
1354	>	// In parallel, we need to visit *all* pairs of row &
1355	>	// column indicies and will truncate later on.
1356		dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1357		snap_->wrapVector(dr);
1358		cuts = getGroupCutoffs( (*j1), (*j2) );
1359		if (dr.lengthSquare() < cuts.third) {
1360		neighborList.push_back(make_pair((*j1), (*j2)));
1361	<	}
1361	>	}
1362		}
1363		}
1074	–	}
1364		#else
1365	<	for (vector<int>::iterator j1 = cellList_[m1].begin();
1366	<	j1 != cellList_[m1].end(); ++j1) {
1367	<	for (vector<int>::iterator j2 = cellList_[m2].begin();
1368	<	j2 != cellList_[m2].end(); ++j2) {
1369	<
1370	<	// Always do this if we're in different cells or if
1371	<	// we're in the same cell and the global index of the
1372	<	// j2 cutoff group is less than the j1 cutoff group
1373	<
1374	<	if (m2 != m1 || (*j2) < (*j1)) {
1375	<	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1376	<	snap_->wrapVector(dr);
1377	<	cuts = getGroupCutoffs( (*j1), (*j2) );
1378	<	if (dr.lengthSquare() < cuts.third) {
1379	<	neighborList.push_back(make_pair((*j1), (*j2)));
1365	>
1366	>	for (vector<int>::iterator j1 = cellList_[m1].begin();
1367	>	j1 != cellList_[m1].end(); ++j1) {
1368	>	for (vector<int>::iterator j2 = cellList_[m2].begin();
1369	>	j2 != cellList_[m2].end(); ++j2) {
1370	>
1371	>	// Always do this if we're in different cells or if
1372	>	// we're in the same cell and the global index of the
1373	>	// j2 cutoff group is less than the j1 cutoff group
1374	>
1375	>	if (m2 != m1 || (*j2) < (*j1)) {
1376	>	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1377	>	snap_->wrapVector(dr);
1378	>	cuts = getGroupCutoffs( (*j1), (*j2) );
1379	>	if (dr.lengthSquare() < cuts.third) {
1380	>	neighborList.push_back(make_pair((*j1), (*j2)));
1381	>	}
1382		}
1383		}
1384		}
1094	–	}
1385		#endif
1386	+	}
1387		}
1388		}
1389		}
1390	+	} else {
1391	+	// branch to do all cutoff group pairs
1392	+	#ifdef IS_MPI
1393	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1394	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1395	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1396	+	snap_->wrapVector(dr);
1397	+	cuts = getGroupCutoffs( j1, j2 );
1398	+	if (dr.lengthSquare() < cuts.third) {
1399	+	neighborList.push_back(make_pair(j1, j2));
1400	+	}
1401	+	}
1402	+	}
1403	+	#else
1404	+	for (int j1 = 0; j1 < nGroups_ - 1; j1++) {
1405	+	for (int j2 = j1 + 1; j2 < nGroups_; j2++) {
1406	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1407	+	snap_->wrapVector(dr);
1408	+	cuts = getGroupCutoffs( j1, j2 );
1409	+	if (dr.lengthSquare() < cuts.third) {
1410	+	neighborList.push_back(make_pair(j1, j2));
1411	+	}
1412	+	}
1413	+	}
1414	+	#endif
1415		}
1416	<
1416	>
1417		// save the local cutoff group positions for the check that is
1418		// done on each loop:
1419		saved_CG_positions_.clear();
1420		for (int i = 0; i < nGroups_; i++)
1421		saved_CG_positions_.push_back(snap_->cgData.position[i]);
1422	<
1422	>
1423		return neighborList;
1424		}
1425		} //end namespace OpenMD

Diff Legend

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