ViewVC Help

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

root/OpenMD/trunk/src/parallel/ForceMatrixDecomposition.cpp

Comparing:
branches/development/src/parallel/ForceDecomposition.cpp (file contents), Revision 1538 by chuckv, Tue Jan 11 18:58:12 2011 UTC vs.
branches/development/src/parallel/ForceMatrixDecomposition.cpp (file contents), Revision 1713 by gezelter, Sat May 19 14:21:02 2012 UTC

#	Line 1 | Line 1
1	<	/**
2	<	* @file ForceDecomposition.cpp
3	<	* @author Charles Vardeman <cvardema.at.nd.edu>
4	<	* @date 08/18/2010
5	<	* @time 11:56am
6	<	* @version 1.0
1	>	/*
2	>	* Copyright (c) 2005 The University of Notre Dame. All Rights Reserved.
3		*
8	–	* @section LICENSE
9	–	* Copyright (c) 2010 The University of Notre Dame. All Rights Reserved.
10	–	*
4		* The University of Notre Dame grants you ("Licensee") a
5		* non-exclusive, royalty free, license to use, modify and
6		* redistribute this software in source and binary code form, provided
#	Line 43 | Line 36
36		* [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37		* [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38		* [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
39	<	* [4]  Vardeman & Gezelter, in progress (2009).
39	>	* [4]  Kuang & Gezelter,  J. Chem. Phys. 133, 164101 (2010).
40	>	* [5]  Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011).
41		*/
42	+	#include "parallel/ForceMatrixDecomposition.hpp"
43	+	#include "math/SquareMatrix3.hpp"
44	+	#include "nonbonded/NonBondedInteraction.hpp"
45	+	#include "brains/SnapshotManager.hpp"
46	+	#include "brains/PairList.hpp"
47
48	+	using namespace std;
49	+	namespace OpenMD {
50
51	+	ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) {
52
53	<	/*  -*- c++ -*-  */
54	<	#include "config.h"
55	<	#include <stdlib.h>
53	>	// In a parallel computation, row and colum scans must visit all
54	>	// surrounding cells (not just the 14 upper triangular blocks that
55	>	// are used when the processor can see all pairs)
56		#ifdef IS_MPI
57	<	#include <mpi.h>
58	<	#endif
57	>	cellOffsets_.clear();
58	>	cellOffsets_.push_back( Vector3i(-1,-1,-1) );
59	>	cellOffsets_.push_back( Vector3i( 0,-1,-1) );
60	>	cellOffsets_.push_back( Vector3i( 1,-1,-1) );
61	>	cellOffsets_.push_back( Vector3i(-1, 0,-1) );
62	>	cellOffsets_.push_back( Vector3i( 0, 0,-1) );
63	>	cellOffsets_.push_back( Vector3i( 1, 0,-1) );
64	>	cellOffsets_.push_back( Vector3i(-1, 1,-1) );
65	>	cellOffsets_.push_back( Vector3i( 0, 1,-1) );
66	>	cellOffsets_.push_back( Vector3i( 1, 1,-1) );
67	>	cellOffsets_.push_back( Vector3i(-1,-1, 0) );
68	>	cellOffsets_.push_back( Vector3i( 0,-1, 0) );
69	>	cellOffsets_.push_back( Vector3i( 1,-1, 0) );
70	>	cellOffsets_.push_back( Vector3i(-1, 0, 0) );
71	>	cellOffsets_.push_back( Vector3i( 0, 0, 0) );
72	>	cellOffsets_.push_back( Vector3i( 1, 0, 0) );
73	>	cellOffsets_.push_back( Vector3i(-1, 1, 0) );
74	>	cellOffsets_.push_back( Vector3i( 0, 1, 0) );
75	>	cellOffsets_.push_back( Vector3i( 1, 1, 0) );
76	>	cellOffsets_.push_back( Vector3i(-1,-1, 1) );
77	>	cellOffsets_.push_back( Vector3i( 0,-1, 1) );
78	>	cellOffsets_.push_back( Vector3i( 1,-1, 1) );
79	>	cellOffsets_.push_back( Vector3i(-1, 0, 1) );
80	>	cellOffsets_.push_back( Vector3i( 0, 0, 1) );
81	>	cellOffsets_.push_back( Vector3i( 1, 0, 1) );
82	>	cellOffsets_.push_back( Vector3i(-1, 1, 1) );
83	>	cellOffsets_.push_back( Vector3i( 0, 1, 1) );
84	>	cellOffsets_.push_back( Vector3i( 1, 1, 1) );
85	>	#endif
86	>	}
87
58	–	#include <iostream>
59	–	#include <vector>
60	–	#include <algorithm>
61	–	#include <cmath>
62	–	#include "parallel/ForceDecomposition.hpp"
88
89	+	/**
90	+	* distributeInitialData is essentially a copy of the older fortran
91	+	* SimulationSetup
92	+	*/
93	+	void ForceMatrixDecomposition::distributeInitialData() {
94	+	snap_ = sman_->getCurrentSnapshot();
95	+	storageLayout_ = sman_->getStorageLayout();
96	+	ff_ = info_->getForceField();
97	+	nLocal_ = snap_->getNumberOfAtoms();
98	+
99	+	nGroups_ = info_->getNLocalCutoffGroups();
100	+	// gather the information for atomtype IDs (atids):
101	+	idents = info_->getIdentArray();
102	+	AtomLocalToGlobal = info_->getGlobalAtomIndices();
103	+	cgLocalToGlobal = info_->getGlobalGroupIndices();
104	+	vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
105
106	<	using namespace std;
66	<	using namespace OpenMD;
106	>	massFactors = info_->getMassFactors();
107
108	<	//__static
108	>	PairList* excludes = info_->getExcludedInteractions();
109	>	PairList* oneTwo = info_->getOneTwoInteractions();
110	>	PairList* oneThree = info_->getOneThreeInteractions();
111	>	PairList* oneFour = info_->getOneFourInteractions();
112	>
113		#ifdef IS_MPI
114	<	static vector<MPI:Comm> communictors;
115	<	#endif
114	>
115	>	MPI::Intracomm row = rowComm.getComm();
116	>	MPI::Intracomm col = colComm.getComm();
117
118	<	//____ MPITypeTraits
119	<	template<typename T>
120	<	struct MPITypeTraits;
118	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
119	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
120	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
121	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
122	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
123
124	<	#ifdef IS_MPI
125	<	template<>
126	<	struct MPITypeTraits<RealType> {
127	<	static const MPI::Datatype datatype;
128	<	};
82	<	const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL;
124	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
125	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
126	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
127	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
128	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
129
130	<	template<>
131	<	struct MPITypeTraits<int> {
132	<	static const MPI::Datatype datatype;
133	<	};
88	<	const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT;
89	<	#endif
130	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
131	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
132	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
133	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
134
135	<	/**
136	<	* Constructor for ForceDecomposition Parallel Decomposition Method
137	<	* Will try to construct a symmetric grid of processors. Ideally, the
138	<	* number of processors will be a square ex: 4, 9, 16, 25.
95	<	*
96	<	*/
135	>	nAtomsInRow_ = AtomPlanIntRow->getSize();
136	>	nAtomsInCol_ = AtomPlanIntColumn->getSize();
137	>	nGroupsInRow_ = cgPlanIntRow->getSize();
138	>	nGroupsInCol_ = cgPlanIntColumn->getSize();
139
140	<	ForceDecomposition::ForceDecomposition() {
140	>	// Modify the data storage objects with the correct layouts and sizes:
141	>	atomRowData.resize(nAtomsInRow_);
142	>	atomRowData.setStorageLayout(storageLayout_);
143	>	atomColData.resize(nAtomsInCol_);
144	>	atomColData.setStorageLayout(storageLayout_);
145	>	cgRowData.resize(nGroupsInRow_);
146	>	cgRowData.setStorageLayout(DataStorage::dslPosition);
147	>	cgColData.resize(nGroupsInCol_);
148	>	cgColData.setStorageLayout(DataStorage::dslPosition);
149	>
150	>	identsRow.resize(nAtomsInRow_);
151	>	identsCol.resize(nAtomsInCol_);
152	>
153	>	AtomPlanIntRow->gather(idents, identsRow);
154	>	AtomPlanIntColumn->gather(idents, identsCol);
155	>
156	>	// allocate memory for the parallel objects
157	>	atypesRow.resize(nAtomsInRow_);
158	>	atypesCol.resize(nAtomsInCol_);
159
160	<	#ifdef IS_MPI
161	<	int nProcs = MPI::COMM_WORLD.Get_size();
162	<	int worldRank = MPI::COMM_WORLD.Get_rank();
160	>	for (int i = 0; i < nAtomsInRow_; i++)
161	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
162	>	for (int i = 0; i < nAtomsInCol_; i++)
163	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
164	>
165	>	pot_row.resize(nAtomsInRow_);
166	>	pot_col.resize(nAtomsInCol_);
167	>
168	>	AtomRowToGlobal.resize(nAtomsInRow_);
169	>	AtomColToGlobal.resize(nAtomsInCol_);
170	>	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
171	>	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
172	>
173	>	cgRowToGlobal.resize(nGroupsInRow_);
174	>	cgColToGlobal.resize(nGroupsInCol_);
175	>	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
176	>	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
177	>
178	>	massFactorsRow.resize(nAtomsInRow_);
179	>	massFactorsCol.resize(nAtomsInCol_);
180	>	AtomPlanRealRow->gather(massFactors, massFactorsRow);
181	>	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
182	>
183	>	groupListRow_.clear();
184	>	groupListRow_.resize(nGroupsInRow_);
185	>	for (int i = 0; i < nGroupsInRow_; i++) {
186	>	int gid = cgRowToGlobal[i];
187	>	for (int j = 0; j < nAtomsInRow_; j++) {
188	>	int aid = AtomRowToGlobal[j];
189	>	if (globalGroupMembership[aid] == gid)
190	>	groupListRow_[i].push_back(j);
191	>	}
192	>	}
193	>
194	>	groupListCol_.clear();
195	>	groupListCol_.resize(nGroupsInCol_);
196	>	for (int i = 0; i < nGroupsInCol_; i++) {
197	>	int gid = cgColToGlobal[i];
198	>	for (int j = 0; j < nAtomsInCol_; j++) {
199	>	int aid = AtomColToGlobal[j];
200	>	if (globalGroupMembership[aid] == gid)
201	>	groupListCol_[i].push_back(j);
202	>	}
203	>	}
204	>
205	>	excludesForAtom.clear();
206	>	excludesForAtom.resize(nAtomsInRow_);
207	>	toposForAtom.clear();
208	>	toposForAtom.resize(nAtomsInRow_);
209	>	topoDist.clear();
210	>	topoDist.resize(nAtomsInRow_);
211	>	for (int i = 0; i < nAtomsInRow_; i++) {
212	>	int iglob = AtomRowToGlobal[i];
213	>
214	>	for (int j = 0; j < nAtomsInCol_; j++) {
215	>	int jglob = AtomColToGlobal[j];
216	>
217	>	if (excludes->hasPair(iglob, jglob))
218	>	excludesForAtom[i].push_back(j);
219	>
220	>	if (oneTwo->hasPair(iglob, jglob)) {
221	>	toposForAtom[i].push_back(j);
222	>	topoDist[i].push_back(1);
223	>	} else {
224	>	if (oneThree->hasPair(iglob, jglob)) {
225	>	toposForAtom[i].push_back(j);
226	>	topoDist[i].push_back(2);
227	>	} else {
228	>	if (oneFour->hasPair(iglob, jglob)) {
229	>	toposForAtom[i].push_back(j);
230	>	topoDist[i].push_back(3);
231	>	}
232	>	}
233	>	}
234	>	}
235	>	}
236	>
237	>	#else
238	>	excludesForAtom.clear();
239	>	excludesForAtom.resize(nLocal_);
240	>	toposForAtom.clear();
241	>	toposForAtom.resize(nLocal_);
242	>	topoDist.clear();
243	>	topoDist.resize(nLocal_);
244	>
245	>	for (int i = 0; i < nLocal_; i++) {
246	>	int iglob = AtomLocalToGlobal[i];
247	>
248	>	for (int j = 0; j < nLocal_; j++) {
249	>	int jglob = AtomLocalToGlobal[j];
250	>
251	>	if (excludes->hasPair(iglob, jglob))
252	>	excludesForAtom[i].push_back(j);
253	>
254	>	if (oneTwo->hasPair(iglob, jglob)) {
255	>	toposForAtom[i].push_back(j);
256	>	topoDist[i].push_back(1);
257	>	} else {
258	>	if (oneThree->hasPair(iglob, jglob)) {
259	>	toposForAtom[i].push_back(j);
260	>	topoDist[i].push_back(2);
261	>	} else {
262	>	if (oneFour->hasPair(iglob, jglob)) {
263	>	toposForAtom[i].push_back(j);
264	>	topoDist[i].push_back(3);
265	>	}
266	>	}
267	>	}
268	>	}
269	>	}
270		#endif
271
272	<	// First time through, construct column stride.
273	<	if (communicators.size() == 0)
274	<	{
275	<	int nColumnsMax = (int) round(sqrt((float) nProcs));
276	<	for (int i = 0; i < nProcs; ++i)
277	<	{
278	<	if (nProcs%i==0) nColumns=i;
272	>	// allocate memory for the parallel objects
273	>	atypesLocal.resize(nLocal_);
274	>
275	>	for (int i = 0; i < nLocal_; i++)
276	>	atypesLocal[i] = ff_->getAtomType(idents[i]);
277	>
278	>	groupList_.clear();
279	>	groupList_.resize(nGroups_);
280	>	for (int i = 0; i < nGroups_; i++) {
281	>	int gid = cgLocalToGlobal[i];
282	>	for (int j = 0; j < nLocal_; j++) {
283	>	int aid = AtomLocalToGlobal[j];
284	>	if (globalGroupMembership[aid] == gid) {
285	>	groupList_[i].push_back(j);
286	>	}
287	>	}
288		}
289
114	–	int nRows = nProcs/nColumns;
115	–	myRank_ = (int) worldRank%nColumns;
116	–	}
117	–	else
118	–	{
119	–	myRank_ = myRank/nColumns;
120	–	}
121	–	MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0);
122	–
123	–	isColumn_ = false;
124	–
125	–	}
290
291	<	ForceDecomposition::gather(sendbuf, receivebuf){
128	<	communicators(myIndex_).Allgatherv();
129	<	}
291	>	createGtypeCutoffMap();
292
293	+	}
294	+
295	+	void ForceMatrixDecomposition::createGtypeCutoffMap() {
296	+
297	+	RealType tol = 1e-6;
298	+	largestRcut_ = 0.0;
299	+	RealType rc;
300	+	int atid;
301	+	set<AtomType*> atypes = info_->getSimulatedAtomTypes();
302	+
303	+	map<int, RealType> atypeCutoff;
304	+
305	+	for (set<AtomType*>::iterator at = atypes.begin();
306	+	at != atypes.end(); ++at){
307	+	atid = (*at)->getIdent();
308	+	if (userChoseCutoff_)
309	+	atypeCutoff[atid] = userCutoff_;
310	+	else
311	+	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
312	+	}
313	+
314	+	vector<RealType> gTypeCutoffs;
315	+	// first we do a single loop over the cutoff groups to find the
316	+	// largest cutoff for any atypes present in this group.
317	+	#ifdef IS_MPI
318	+	vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
319	+	groupRowToGtype.resize(nGroupsInRow_);
320	+	for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
321	+	vector<int> atomListRow = getAtomsInGroupRow(cg1);
322	+	for (vector<int>::iterator ia = atomListRow.begin();
323	+	ia != atomListRow.end(); ++ia) {
324	+	int atom1 = (*ia);
325	+	atid = identsRow[atom1];
326	+	if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
327	+	groupCutoffRow[cg1] = atypeCutoff[atid];
328	+	}
329	+	}
330
331	+	bool gTypeFound = false;
332	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
333	+	if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
334	+	groupRowToGtype[cg1] = gt;
335	+	gTypeFound = true;
336	+	}
337	+	}
338	+	if (!gTypeFound) {
339	+	gTypeCutoffs.push_back( groupCutoffRow[cg1] );
340	+	groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
341	+	}
342	+
343	+	}
344	+	vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
345	+	groupColToGtype.resize(nGroupsInCol_);
346	+	for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
347	+	vector<int> atomListCol = getAtomsInGroupColumn(cg2);
348	+	for (vector<int>::iterator jb = atomListCol.begin();
349	+	jb != atomListCol.end(); ++jb) {
350	+	int atom2 = (*jb);
351	+	atid = identsCol[atom2];
352	+	if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
353	+	groupCutoffCol[cg2] = atypeCutoff[atid];
354	+	}
355	+	}
356	+	bool gTypeFound = false;
357	+	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
358	+	if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
359	+	groupColToGtype[cg2] = gt;
360	+	gTypeFound = true;
361	+	}
362	+	}
363	+	if (!gTypeFound) {
364	+	gTypeCutoffs.push_back( groupCutoffCol[cg2] );
365	+	groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
366	+	}
367	+	}
368	+	#else
369
370	<	ForceDecomposition::scatter(sbuffer, rbuffer){
371	<	communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM);
372	<	}
370	>	vector<RealType> groupCutoff(nGroups_, 0.0);
371	>	groupToGtype.resize(nGroups_);
372	>	for (int cg1 = 0; cg1 < nGroups_; cg1++) {
373	>	groupCutoff[cg1] = 0.0;
374	>	vector<int> atomList = getAtomsInGroupRow(cg1);
375	>	for (vector<int>::iterator ia = atomList.begin();
376	>	ia != atomList.end(); ++ia) {
377	>	int atom1 = (*ia);
378	>	atid = idents[atom1];
379	>	if (atypeCutoff[atid] > groupCutoff[cg1])
380	>	groupCutoff[cg1] = atypeCutoff[atid];
381	>	}
382	>
383	>	bool gTypeFound = false;
384	>	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
385	>	if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
386	>	groupToGtype[cg1] = gt;
387	>	gTypeFound = true;
388	>	}
389	>	}
390	>	if (!gTypeFound) {
391	>	gTypeCutoffs.push_back( groupCutoff[cg1] );
392	>	groupToGtype[cg1] = gTypeCutoffs.size() - 1;
393	>	}
394	>	}
395	>	#endif
396	>
397	>	// Now we find the maximum group cutoff value present in the simulation
398	>
399	>	RealType groupMax = *max_element(gTypeCutoffs.begin(),
400	>	gTypeCutoffs.end());
401	>
402	>	#ifdef IS_MPI
403	>	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
404	>	MPI::MAX);
405	>	#endif
406	>
407	>	RealType tradRcut = groupMax;
408	>
409	>	for (int i = 0; i < gTypeCutoffs.size();  i++) {
410	>	for (int j = 0; j < gTypeCutoffs.size();  j++) {
411	>	RealType thisRcut;
412	>	switch(cutoffPolicy_) {
413	>	case TRADITIONAL:
414	>	thisRcut = tradRcut;
415	>	break;
416	>	case MIX:
417	>	thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
418	>	break;
419	>	case MAX:
420	>	thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
421	>	break;
422	>	default:
423	>	sprintf(painCave.errMsg,
424	>	"ForceMatrixDecomposition::createGtypeCutoffMap "
425	>	"hit an unknown cutoff policy!\n");
426	>	painCave.severity = OPENMD_ERROR;
427	>	painCave.isFatal = 1;
428	>	simError();
429	>	break;
430	>	}
431	>
432	>	pair<int,int> key = make_pair(i,j);
433	>	gTypeCutoffMap[key].first = thisRcut;
434	>	if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
435	>	gTypeCutoffMap[key].second = thisRcut*thisRcut;
436	>	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
437	>	// sanity check
438	>
439	>	if (userChoseCutoff_) {
440	>	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
441	>	sprintf(painCave.errMsg,
442	>	"ForceMatrixDecomposition::createGtypeCutoffMap "
443	>	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
444	>	painCave.severity = OPENMD_ERROR;
445	>	painCave.isFatal = 1;
446	>	simError();
447	>	}
448	>	}
449	>	}
450	>	}
451	>	}
452	>
453	>
454	>	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
455	>	int i, j;
456	>	#ifdef IS_MPI
457	>	i = groupRowToGtype[cg1];
458	>	j = groupColToGtype[cg2];
459	>	#else
460	>	i = groupToGtype[cg1];
461	>	j = groupToGtype[cg2];
462	>	#endif
463	>	return gTypeCutoffMap[make_pair(i,j)];
464	>	}
465	>
466	>	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
467	>	for (int j = 0; j < toposForAtom[atom1].size(); j++) {
468	>	if (toposForAtom[atom1][j] == atom2)
469	>	return topoDist[atom1][j];
470	>	}
471	>	return 0;
472	>	}
473	>
474	>	void ForceMatrixDecomposition::zeroWorkArrays() {
475	>	pairwisePot = 0.0;
476	>	embeddingPot = 0.0;
477	>
478	>	#ifdef IS_MPI
479	>	if (storageLayout_ & DataStorage::dslForce) {
480	>	fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
481	>	fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
482	>	}
483	>
484	>	if (storageLayout_ & DataStorage::dslTorque) {
485	>	fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
486	>	fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
487	>	}
488	>
489	>	fill(pot_row.begin(), pot_row.end(),
490	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
491	>
492	>	fill(pot_col.begin(), pot_col.end(),
493	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
494	>
495	>	if (storageLayout_ & DataStorage::dslParticlePot) {
496	>	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
497	>	0.0);
498	>	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
499	>	0.0);
500	>	}
501	>
502	>	if (storageLayout_ & DataStorage::dslDensity) {
503	>	fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
504	>	fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
505	>	}
506	>
507	>	if (storageLayout_ & DataStorage::dslFunctional) {
508	>	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
509	>	0.0);
510	>	fill(atomColData.functional.begin(), atomColData.functional.end(),
511	>	0.0);
512	>	}
513	>
514	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
515	>	fill(atomRowData.functionalDerivative.begin(),
516	>	atomRowData.functionalDerivative.end(), 0.0);
517	>	fill(atomColData.functionalDerivative.begin(),
518	>	atomColData.functionalDerivative.end(), 0.0);
519	>	}
520	>
521	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
522	>	fill(atomRowData.skippedCharge.begin(),
523	>	atomRowData.skippedCharge.end(), 0.0);
524	>	fill(atomColData.skippedCharge.begin(),
525	>	atomColData.skippedCharge.end(), 0.0);
526	>	}
527	>
528	>	if (storageLayout_ & DataStorage::dslElectricField) {
529	>	fill(atomRowData.electricField.begin(),
530	>	atomRowData.electricField.end(), V3Zero);
531	>	fill(atomColData.electricField.begin(),
532	>	atomColData.electricField.end(), V3Zero);
533	>	}
534	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
535	>	fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(),
536	>	0.0);
537	>	fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.end(),
538	>	0.0);
539	>	}
540	>
541	>	#endif
542	>	// even in parallel, we need to zero out the local arrays:
543	>
544	>	if (storageLayout_ & DataStorage::dslParticlePot) {
545	>	fill(snap_->atomData.particlePot.begin(),
546	>	snap_->atomData.particlePot.end(), 0.0);
547	>	}
548	>
549	>	if (storageLayout_ & DataStorage::dslDensity) {
550	>	fill(snap_->atomData.density.begin(),
551	>	snap_->atomData.density.end(), 0.0);
552	>	}
553	>
554	>	if (storageLayout_ & DataStorage::dslFunctional) {
555	>	fill(snap_->atomData.functional.begin(),
556	>	snap_->atomData.functional.end(), 0.0);
557	>	}
558	>
559	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
560	>	fill(snap_->atomData.functionalDerivative.begin(),
561	>	snap_->atomData.functionalDerivative.end(), 0.0);
562	>	}
563	>
564	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
565	>	fill(snap_->atomData.skippedCharge.begin(),
566	>	snap_->atomData.skippedCharge.end(), 0.0);
567	>	}
568	>
569	>	if (storageLayout_ & DataStorage::dslElectricField) {
570	>	fill(snap_->atomData.electricField.begin(),
571	>	snap_->atomData.electricField.end(), V3Zero);
572	>	}
573	>	}
574	>
575	>
576	>	void ForceMatrixDecomposition::distributeData()  {
577	>	snap_ = sman_->getCurrentSnapshot();
578	>	storageLayout_ = sman_->getStorageLayout();
579	>	#ifdef IS_MPI
580	>
581	>	// gather up the atomic positions
582	>	AtomPlanVectorRow->gather(snap_->atomData.position,
583	>	atomRowData.position);
584	>	AtomPlanVectorColumn->gather(snap_->atomData.position,
585	>	atomColData.position);
586	>
587	>	// gather up the cutoff group positions
588	>
589	>	cgPlanVectorRow->gather(snap_->cgData.position,
590	>	cgRowData.position);
591	>
592	>	cgPlanVectorColumn->gather(snap_->cgData.position,
593	>	cgColData.position);
594	>
595	>
596	>	// if needed, gather the atomic rotation matrices
597	>	if (storageLayout_ & DataStorage::dslAmat) {
598	>	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
599	>	atomRowData.aMat);
600	>	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
601	>	atomColData.aMat);
602	>	}
603	>
604	>	// if needed, gather the atomic eletrostatic frames
605	>	if (storageLayout_ & DataStorage::dslElectroFrame) {
606	>	AtomPlanMatrixRow->gather(snap_->atomData.electroFrame,
607	>	atomRowData.electroFrame);
608	>	AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame,
609	>	atomColData.electroFrame);
610	>	}
611	>
612	>	// if needed, gather the atomic fluctuating charge values
613	>	if (storageLayout_ & DataStorage::dslFlucQPosition) {
614	>	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
615	>	atomRowData.flucQPos);
616	>	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
617	>	atomColData.flucQPos);
618	>	}
619	>
620	>	#endif
621	>	}
622	>
623	>	/* collects information obtained during the pre-pair loop onto local
624	>	* data structures.
625	>	*/
626	>	void ForceMatrixDecomposition::collectIntermediateData() {
627	>	snap_ = sman_->getCurrentSnapshot();
628	>	storageLayout_ = sman_->getStorageLayout();
629	>	#ifdef IS_MPI
630	>
631	>	if (storageLayout_ & DataStorage::dslDensity) {
632	>
633	>	AtomPlanRealRow->scatter(atomRowData.density,
634	>	snap_->atomData.density);
635	>
636	>	int n = snap_->atomData.density.size();
637	>	vector<RealType> rho_tmp(n, 0.0);
638	>	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
639	>	for (int i = 0; i < n; i++)
640	>	snap_->atomData.density[i] += rho_tmp[i];
641	>	}
642	>
643	>	if (storageLayout_ & DataStorage::dslElectricField) {
644	>
645	>	AtomPlanVectorRow->scatter(atomRowData.electricField,
646	>	snap_->atomData.electricField);
647	>
648	>	int n = snap_->atomData.electricField.size();
649	>	vector<Vector3d> field_tmp(n, V3Zero);
650	>	AtomPlanVectorColumn->scatter(atomColData.electricField, field_tmp);
651	>	for (int i = 0; i < n; i++)
652	>	snap_->atomData.electricField[i] += field_tmp[i];
653	>	}
654	>	#endif
655	>	}
656	>
657	>	/*
658	>	* redistributes information obtained during the pre-pair loop out to
659	>	* row and column-indexed data structures
660	>	*/
661	>	void ForceMatrixDecomposition::distributeIntermediateData() {
662	>	snap_ = sman_->getCurrentSnapshot();
663	>	storageLayout_ = sman_->getStorageLayout();
664	>	#ifdef IS_MPI
665	>	if (storageLayout_ & DataStorage::dslFunctional) {
666	>	AtomPlanRealRow->gather(snap_->atomData.functional,
667	>	atomRowData.functional);
668	>	AtomPlanRealColumn->gather(snap_->atomData.functional,
669	>	atomColData.functional);
670	>	}
671	>
672	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
673	>	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
674	>	atomRowData.functionalDerivative);
675	>	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
676	>	atomColData.functionalDerivative);
677	>	}
678	>	#endif
679	>	}
680	>
681	>
682	>	void ForceMatrixDecomposition::collectData() {
683	>	snap_ = sman_->getCurrentSnapshot();
684	>	storageLayout_ = sman_->getStorageLayout();
685	>	#ifdef IS_MPI
686	>	int n = snap_->atomData.force.size();
687	>	vector<Vector3d> frc_tmp(n, V3Zero);
688	>
689	>	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
690	>	for (int i = 0; i < n; i++) {
691	>	snap_->atomData.force[i] += frc_tmp[i];
692	>	frc_tmp[i] = 0.0;
693	>	}
694	>
695	>	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
696	>	for (int i = 0; i < n; i++) {
697	>	snap_->atomData.force[i] += frc_tmp[i];
698	>	}
699	>
700	>	if (storageLayout_ & DataStorage::dslTorque) {
701	>
702	>	int nt = snap_->atomData.torque.size();
703	>	vector<Vector3d> trq_tmp(nt, V3Zero);
704	>
705	>	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
706	>	for (int i = 0; i < nt; i++) {
707	>	snap_->atomData.torque[i] += trq_tmp[i];
708	>	trq_tmp[i] = 0.0;
709	>	}
710	>
711	>	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
712	>	for (int i = 0; i < nt; i++)
713	>	snap_->atomData.torque[i] += trq_tmp[i];
714	>	}
715	>
716	>	if (storageLayout_ & DataStorage::dslSkippedCharge) {
717	>
718	>	int ns = snap_->atomData.skippedCharge.size();
719	>	vector<RealType> skch_tmp(ns, 0.0);
720	>
721	>	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
722	>	for (int i = 0; i < ns; i++) {
723	>	snap_->atomData.skippedCharge[i] += skch_tmp[i];
724	>	skch_tmp[i] = 0.0;
725	>	}
726	>
727	>	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
728	>	for (int i = 0; i < ns; i++)
729	>	snap_->atomData.skippedCharge[i] += skch_tmp[i];
730	>
731	>	}
732	>
733	>	if (storageLayout_ & DataStorage::dslFlucQForce) {
734	>
735	>	int nq = snap_->atomData.flucQFrc.size();
736	>	vector<RealType> fqfrc_tmp(nq, 0.0);
737	>
738	>	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
739	>	for (int i = 0; i < nq; i++) {
740	>	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
741	>	fqfrc_tmp[i] = 0.0;
742	>	}
743	>
744	>	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
745	>	for (int i = 0; i < nq; i++)
746	>	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
747	>
748	>	}
749	>
750	>	nLocal_ = snap_->getNumberOfAtoms();
751	>
752	>	vector<potVec> pot_temp(nLocal_,
753	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
754	>
755	>	// scatter/gather pot_row into the members of my column
756	>
757	>	AtomPlanPotRow->scatter(pot_row, pot_temp);
758	>
759	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
760	>	pairwisePot += pot_temp[ii];
761	>
762	>	fill(pot_temp.begin(), pot_temp.end(),
763	>	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
764	>
765	>	AtomPlanPotColumn->scatter(pot_col, pot_temp);
766	>
767	>	for (int ii = 0;  ii < pot_temp.size(); ii++ )
768	>	pairwisePot += pot_temp[ii];
769	>
770	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
771	>	RealType ploc1 = pairwisePot[ii];
772	>	RealType ploc2 = 0.0;
773	>	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
774	>	pairwisePot[ii] = ploc2;
775	>	}
776	>
777	>	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
778	>	RealType ploc1 = embeddingPot[ii];
779	>	RealType ploc2 = 0.0;
780	>	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
781	>	embeddingPot[ii] = ploc2;
782	>	}
783	>
784	>	#endif
785	>
786	>	}
787	>
788	>	int ForceMatrixDecomposition::getNAtomsInRow() {
789	>	#ifdef IS_MPI
790	>	return nAtomsInRow_;
791	>	#else
792	>	return nLocal_;
793	>	#endif
794	>	}
795	>
796	>	/**
797	>	* returns the list of atoms belonging to this group.
798	>	*/
799	>	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
800	>	#ifdef IS_MPI
801	>	return groupListRow_[cg1];
802	>	#else
803	>	return groupList_[cg1];
804	>	#endif
805	>	}
806	>
807	>	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
808	>	#ifdef IS_MPI
809	>	return groupListCol_[cg2];
810	>	#else
811	>	return groupList_[cg2];
812	>	#endif
813	>	}
814	>
815	>	Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
816	>	Vector3d d;
817	>
818	>	#ifdef IS_MPI
819	>	d = cgColData.position[cg2] - cgRowData.position[cg1];
820	>	#else
821	>	d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
822	>	#endif
823	>
824	>	snap_->wrapVector(d);
825	>	return d;
826	>	}
827	>
828	>
829	>	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
830	>
831	>	Vector3d d;
832	>
833	>	#ifdef IS_MPI
834	>	d = cgRowData.position[cg1] - atomRowData.position[atom1];
835	>	#else
836	>	d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
837	>	#endif
838	>
839	>	snap_->wrapVector(d);
840	>	return d;
841	>	}
842	>
843	>	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
844	>	Vector3d d;
845	>
846	>	#ifdef IS_MPI
847	>	d = cgColData.position[cg2] - atomColData.position[atom2];
848	>	#else
849	>	d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
850	>	#endif
851	>
852	>	snap_->wrapVector(d);
853	>	return d;
854	>	}
855	>
856	>	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
857	>	#ifdef IS_MPI
858	>	return massFactorsRow[atom1];
859	>	#else
860	>	return massFactors[atom1];
861	>	#endif
862	>	}
863	>
864	>	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
865	>	#ifdef IS_MPI
866	>	return massFactorsCol[atom2];
867	>	#else
868	>	return massFactors[atom2];
869	>	#endif
870	>
871	>	}
872	>
873	>	Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
874	>	Vector3d d;
875	>
876	>	#ifdef IS_MPI
877	>	d = atomColData.position[atom2] - atomRowData.position[atom1];
878	>	#else
879	>	d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
880	>	#endif
881	>
882	>	snap_->wrapVector(d);
883	>	return d;
884	>	}
885	>
886	>	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
887	>	return excludesForAtom[atom1];
888	>	}
889	>
890	>	/**
891	>	* We need to exclude some overcounted interactions that result from
892	>	* the parallel decomposition.
893	>	*/
894	>	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) {
895	>	int unique_id_1, unique_id_2;
896	>
897	>	#ifdef IS_MPI
898	>	// in MPI, we have to look up the unique IDs for each atom
899	>	unique_id_1 = AtomRowToGlobal[atom1];
900	>	unique_id_2 = AtomColToGlobal[atom2];
901	>	#else
902	>	unique_id_1 = AtomLocalToGlobal[atom1];
903	>	unique_id_2 = AtomLocalToGlobal[atom2];
904	>	#endif
905	>
906	>	if (unique_id_1 == unique_id_2) return true;
907	>
908	>	#ifdef IS_MPI
909	>	// this prevents us from doing the pair on multiple processors
910	>	if (unique_id_1 < unique_id_2) {
911	>	if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
912	>	} else {
913	>	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
914	>	}
915	>	#endif
916	>
917	>	return false;
918	>	}
919	>
920	>	/**
921	>	* We need to handle the interactions for atoms who are involved in
922	>	* the same rigid body as well as some short range interactions
923	>	* (bonds, bends, torsions) differently from other interactions.
924	>	* We'll still visit the pairwise routines, but with a flag that
925	>	* tells those routines to exclude the pair from direct long range
926	>	* interactions.  Some indirect interactions (notably reaction
927	>	* field) must still be handled for these pairs.
928	>	*/
929	>	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
930	>
931	>	// excludesForAtom was constructed to use row/column indices in the MPI
932	>	// version, and to use local IDs in the non-MPI version:
933	>
934	>	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
935	>	i != excludesForAtom[atom1].end(); ++i) {
936	>	if ( (*i) == atom2 ) return true;
937	>	}
938	>
939	>	return false;
940	>	}
941	>
942	>
943	>	void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
944	>	#ifdef IS_MPI
945	>	atomRowData.force[atom1] += fg;
946	>	#else
947	>	snap_->atomData.force[atom1] += fg;
948	>	#endif
949	>	}
950	>
951	>	void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
952	>	#ifdef IS_MPI
953	>	atomColData.force[atom2] += fg;
954	>	#else
955	>	snap_->atomData.force[atom2] += fg;
956	>	#endif
957	>	}
958	>
959	>	// filling interaction blocks with pointers
960	>	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
961	>	int atom1, int atom2) {
962	>
963	>	idat.excluded = excludeAtomPair(atom1, atom2);
964	>
965	>	#ifdef IS_MPI
966	>	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
967	>	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
968	>	//                         ff_->getAtomType(identsCol[atom2]) );
969	>
970	>	if (storageLayout_ & DataStorage::dslAmat) {
971	>	idat.A1 = &(atomRowData.aMat[atom1]);
972	>	idat.A2 = &(atomColData.aMat[atom2]);
973	>	}
974	>
975	>	if (storageLayout_ & DataStorage::dslElectroFrame) {
976	>	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
977	>	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
978	>	}
979	>
980	>	if (storageLayout_ & DataStorage::dslTorque) {
981	>	idat.t1 = &(atomRowData.torque[atom1]);
982	>	idat.t2 = &(atomColData.torque[atom2]);
983	>	}
984	>
985	>	if (storageLayout_ & DataStorage::dslDensity) {
986	>	idat.rho1 = &(atomRowData.density[atom1]);
987	>	idat.rho2 = &(atomColData.density[atom2]);
988	>	}
989	>
990	>	if (storageLayout_ & DataStorage::dslFunctional) {
991	>	idat.frho1 = &(atomRowData.functional[atom1]);
992	>	idat.frho2 = &(atomColData.functional[atom2]);
993	>	}
994	>
995	>	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
996	>	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
997	>	idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
998	>	}
999	>
1000	>	if (storageLayout_ & DataStorage::dslParticlePot) {
1001	>	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1002	>	idat.particlePot2 = &(atomColData.particlePot[atom2]);
1003	>	}
1004
1005	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1006	+	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1007	+	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1008	+	}
1009
1010	+	#else
1011	+
1012	+
1013	+	// cerr << "atoms = " << atom1 << " " << atom2 << "\n";
1014	+	// cerr << "pos1 = " << snap_->atomData.position[atom1] << "\n";
1015	+	// cerr << "pos2 = " << snap_->atomData.position[atom2] << "\n";
1016	+
1017	+	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1018	+	//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]),
1019	+	//                         ff_->getAtomType(idents[atom2]) );
1020	+
1021	+	if (storageLayout_ & DataStorage::dslAmat) {
1022	+	idat.A1 = &(snap_->atomData.aMat[atom1]);
1023	+	idat.A2 = &(snap_->atomData.aMat[atom2]);
1024	+	}
1025	+
1026	+	if (storageLayout_ & DataStorage::dslElectroFrame) {
1027	+	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
1028	+	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
1029	+	}
1030	+
1031	+	if (storageLayout_ & DataStorage::dslTorque) {
1032	+	idat.t1 = &(snap_->atomData.torque[atom1]);
1033	+	idat.t2 = &(snap_->atomData.torque[atom2]);
1034	+	}
1035	+
1036	+	if (storageLayout_ & DataStorage::dslDensity) {
1037	+	idat.rho1 = &(snap_->atomData.density[atom1]);
1038	+	idat.rho2 = &(snap_->atomData.density[atom2]);
1039	+	}
1040	+
1041	+	if (storageLayout_ & DataStorage::dslFunctional) {
1042	+	idat.frho1 = &(snap_->atomData.functional[atom1]);
1043	+	idat.frho2 = &(snap_->atomData.functional[atom2]);
1044	+	}
1045	+
1046	+	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1047	+	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1048	+	idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1049	+	}
1050	+
1051	+	if (storageLayout_ & DataStorage::dslParticlePot) {
1052	+	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1053	+	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1054	+	}
1055	+
1056	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1057	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1058	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1059	+	}
1060	+	#endif
1061	+	}
1062	+
1063	+
1064	+	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1065	+	#ifdef IS_MPI
1066	+	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1067	+	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1068	+
1069	+	atomRowData.force[atom1] += *(idat.f1);
1070	+	atomColData.force[atom2] -= *(idat.f1);
1071	+
1072	+	// should particle pot be done here also?
1073	+	#else
1074	+	pairwisePot += *(idat.pot);
1075	+
1076	+	snap_->atomData.force[atom1] += *(idat.f1);
1077	+	snap_->atomData.force[atom2] -= *(idat.f1);
1078	+
1079	+	if (idat.doParticlePot) {
1080	+	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1081	+	snap_->atomData.particlePot[atom2] -= *(idat.vpair) * *(idat.sw);
1082	+	}
1083	+
1084	+	#endif
1085	+
1086	+	}
1087	+
1088	+	/*
1089	+	* buildNeighborList
1090	+	*
1091	+	* first element of pair is row-indexed CutoffGroup
1092	+	* second element of pair is column-indexed CutoffGroup
1093	+	*/
1094	+	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1095	+
1096	+	vector<pair<int, int> > neighborList;
1097	+	groupCutoffs cuts;
1098	+	bool doAllPairs = false;
1099	+
1100	+	#ifdef IS_MPI
1101	+	cellListRow_.clear();
1102	+	cellListCol_.clear();
1103	+	#else
1104	+	cellList_.clear();
1105	+	#endif
1106	+
1107	+	RealType rList_ = (largestRcut_ + skinThickness_);
1108	+	RealType rl2 = rList_ * rList_;
1109	+	Snapshot* snap_ = sman_->getCurrentSnapshot();
1110	+	Mat3x3d Hmat = snap_->getHmat();
1111	+	Vector3d Hx = Hmat.getColumn(0);
1112	+	Vector3d Hy = Hmat.getColumn(1);
1113	+	Vector3d Hz = Hmat.getColumn(2);
1114	+
1115	+	nCells_.x() = (int) ( Hx.length() )/ rList_;
1116	+	nCells_.y() = (int) ( Hy.length() )/ rList_;
1117	+	nCells_.z() = (int) ( Hz.length() )/ rList_;
1118	+
1119	+	// handle small boxes where the cell offsets can end up repeating cells
1120	+
1121	+	if (nCells_.x() < 3) doAllPairs = true;
1122	+	if (nCells_.y() < 3) doAllPairs = true;
1123	+	if (nCells_.z() < 3) doAllPairs = true;
1124	+
1125	+	Mat3x3d invHmat = snap_->getInvHmat();
1126	+	Vector3d rs, scaled, dr;
1127	+	Vector3i whichCell;
1128	+	int cellIndex;
1129	+	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1130	+
1131	+	#ifdef IS_MPI
1132	+	cellListRow_.resize(nCtot);
1133	+	cellListCol_.resize(nCtot);
1134	+	#else
1135	+	cellList_.resize(nCtot);
1136	+	#endif
1137	+
1138	+	if (!doAllPairs) {
1139	+	#ifdef IS_MPI
1140	+
1141	+	for (int i = 0; i < nGroupsInRow_; i++) {
1142	+	rs = cgRowData.position[i];
1143	+
1144	+	// scaled positions relative to the box vectors
1145	+	scaled = invHmat * rs;
1146	+
1147	+	// wrap the vector back into the unit box by subtracting integer box
1148	+	// numbers
1149	+	for (int j = 0; j < 3; j++) {
1150	+	scaled[j] -= roundMe(scaled[j]);
1151	+	scaled[j] += 0.5;
1152	+	}
1153	+
1154	+	// find xyz-indices of cell that cutoffGroup is in.
1155	+	whichCell.x() = nCells_.x() * scaled.x();
1156	+	whichCell.y() = nCells_.y() * scaled.y();
1157	+	whichCell.z() = nCells_.z() * scaled.z();
1158	+
1159	+	// find single index of this cell:
1160	+	cellIndex = Vlinear(whichCell, nCells_);
1161	+
1162	+	// add this cutoff group to the list of groups in this cell;
1163	+	cellListRow_[cellIndex].push_back(i);
1164	+	}
1165	+	for (int i = 0; i < nGroupsInCol_; i++) {
1166	+	rs = cgColData.position[i];
1167	+
1168	+	// scaled positions relative to the box vectors
1169	+	scaled = invHmat * rs;
1170	+
1171	+	// wrap the vector back into the unit box by subtracting integer box
1172	+	// numbers
1173	+	for (int j = 0; j < 3; j++) {
1174	+	scaled[j] -= roundMe(scaled[j]);
1175	+	scaled[j] += 0.5;
1176	+	}
1177	+
1178	+	// find xyz-indices of cell that cutoffGroup is in.
1179	+	whichCell.x() = nCells_.x() * scaled.x();
1180	+	whichCell.y() = nCells_.y() * scaled.y();
1181	+	whichCell.z() = nCells_.z() * scaled.z();
1182	+
1183	+	// find single index of this cell:
1184	+	cellIndex = Vlinear(whichCell, nCells_);
1185	+
1186	+	// add this cutoff group to the list of groups in this cell;
1187	+	cellListCol_[cellIndex].push_back(i);
1188	+	}
1189	+
1190	+	#else
1191	+	for (int i = 0; i < nGroups_; i++) {
1192	+	rs = snap_->cgData.position[i];
1193	+
1194	+	// scaled positions relative to the box vectors
1195	+	scaled = invHmat * rs;
1196	+
1197	+	// wrap the vector back into the unit box by subtracting integer box
1198	+	// numbers
1199	+	for (int j = 0; j < 3; j++) {
1200	+	scaled[j] -= roundMe(scaled[j]);
1201	+	scaled[j] += 0.5;
1202	+	}
1203	+
1204	+	// find xyz-indices of cell that cutoffGroup is in.
1205	+	whichCell.x() = nCells_.x() * scaled.x();
1206	+	whichCell.y() = nCells_.y() * scaled.y();
1207	+	whichCell.z() = nCells_.z() * scaled.z();
1208	+
1209	+	// find single index of this cell:
1210	+	cellIndex = Vlinear(whichCell, nCells_);
1211	+
1212	+	// add this cutoff group to the list of groups in this cell;
1213	+	cellList_[cellIndex].push_back(i);
1214	+	}
1215	+
1216	+	#endif
1217	+
1218	+	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1219	+	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1220	+	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1221	+	Vector3i m1v(m1x, m1y, m1z);
1222	+	int m1 = Vlinear(m1v, nCells_);
1223	+
1224	+	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1225	+	os != cellOffsets_.end(); ++os) {
1226	+
1227	+	Vector3i m2v = m1v + (*os);
1228	+
1229	+
1230	+	if (m2v.x() >= nCells_.x()) {
1231	+	m2v.x() = 0;
1232	+	} else if (m2v.x() < 0) {
1233	+	m2v.x() = nCells_.x() - 1;
1234	+	}
1235	+
1236	+	if (m2v.y() >= nCells_.y()) {
1237	+	m2v.y() = 0;
1238	+	} else if (m2v.y() < 0) {
1239	+	m2v.y() = nCells_.y() - 1;
1240	+	}
1241	+
1242	+	if (m2v.z() >= nCells_.z()) {
1243	+	m2v.z() = 0;
1244	+	} else if (m2v.z() < 0) {
1245	+	m2v.z() = nCells_.z() - 1;
1246	+	}
1247	+
1248	+	int m2 = Vlinear (m2v, nCells_);
1249	+
1250	+	#ifdef IS_MPI
1251	+	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1252	+	j1 != cellListRow_[m1].end(); ++j1) {
1253	+	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1254	+	j2 != cellListCol_[m2].end(); ++j2) {
1255	+
1256	+	// In parallel, we need to visit *all* pairs of row
1257	+	// & column indicies and will divide labor in the
1258	+	// force evaluation later.
1259	+	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1260	+	snap_->wrapVector(dr);
1261	+	cuts = getGroupCutoffs( (*j1), (*j2) );
1262	+	if (dr.lengthSquare() < cuts.third) {
1263	+	neighborList.push_back(make_pair((*j1), (*j2)));
1264	+	}
1265	+	}
1266	+	}
1267	+	#else
1268	+	for (vector<int>::iterator j1 = cellList_[m1].begin();
1269	+	j1 != cellList_[m1].end(); ++j1) {
1270	+	for (vector<int>::iterator j2 = cellList_[m2].begin();
1271	+	j2 != cellList_[m2].end(); ++j2) {
1272	+
1273	+	// Always do this if we're in different cells or if
1274	+	// we're in the same cell and the global index of
1275	+	// the j2 cutoff group is greater than or equal to
1276	+	// the j1 cutoff group.  Note that Rappaport's code
1277	+	// has a "less than" conditional here, but that
1278	+	// deals with atom-by-atom computation.  OpenMD
1279	+	// allows atoms within a single cutoff group to
1280	+	// interact with each other.
1281	+
1282	+
1283	+
1284	+	if (m2 != m1 || (*j2) >= (*j1) ) {
1285	+
1286	+	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1287	+	snap_->wrapVector(dr);
1288	+	cuts = getGroupCutoffs( (*j1), (*j2) );
1289	+	if (dr.lengthSquare() < cuts.third) {
1290	+	neighborList.push_back(make_pair((*j1), (*j2)));
1291	+	}
1292	+	}
1293	+	}
1294	+	}
1295	+	#endif
1296	+	}
1297	+	}
1298	+	}
1299	+	}
1300	+	} else {
1301	+	// branch to do all cutoff group pairs
1302	+	#ifdef IS_MPI
1303	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1304	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1305	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1306	+	snap_->wrapVector(dr);
1307	+	cuts = getGroupCutoffs( j1, j2 );
1308	+	if (dr.lengthSquare() < cuts.third) {
1309	+	neighborList.push_back(make_pair(j1, j2));
1310	+	}
1311	+	}
1312	+	}
1313	+	#else
1314	+	// include all groups here.
1315	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1316	+	// include self group interactions j2 == j1
1317	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1318	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1319	+	snap_->wrapVector(dr);
1320	+	cuts = getGroupCutoffs( j1, j2 );
1321	+	if (dr.lengthSquare() < cuts.third) {
1322	+	neighborList.push_back(make_pair(j1, j2));
1323	+	}
1324	+	}
1325	+	}
1326	+	#endif
1327	+	}
1328	+
1329	+	// save the local cutoff group positions for the check that is
1330	+	// done on each loop:
1331	+	saved_CG_positions_.clear();
1332	+	for (int i = 0; i < nGroups_; i++)
1333	+	saved_CG_positions_.push_back(snap_->cgData.position[i]);
1334	+
1335	+	return neighborList;
1336	+	}
1337	+	} //end namespace OpenMD

Diff Legend

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