ViewVC Help

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

root/OpenMD/branches/development/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 1821 by gezelter, Mon Jan 7 20:05:43 2013 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	>	if (needVelocities_)
114	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition |
115	>	DataStorage::dslVelocity);
116	>	else
117	>	snap_->cgData.setStorageLayout(DataStorage::dslPosition);
118	>
119		#ifdef IS_MPI
120	<	static vector<MPI:Comm> communictors;
121	<	#endif
120	>
121	>	MPI::Intracomm row = rowComm.getComm();
122	>	MPI::Intracomm col = colComm.getComm();
123
124	<	//____ MPITypeTraits
125	<	template<typename T>
126	<	struct MPITypeTraits;
124	>	AtomPlanIntRow = new Plan<int>(row, nLocal_);
125	>	AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
126	>	AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
127	>	AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
128	>	AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
129
130	<	#ifdef IS_MPI
131	<	template<>
132	<	struct MPITypeTraits<RealType> {
133	<	static const MPI::Datatype datatype;
134	<	};
82	<	const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL;
130	>	AtomPlanIntColumn = new Plan<int>(col, nLocal_);
131	>	AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
132	>	AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
133	>	AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
134	>	AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
135
136	<	template<>
137	<	struct MPITypeTraits<int> {
138	<	static const MPI::Datatype datatype;
139	<	};
88	<	const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT;
89	<	#endif
136	>	cgPlanIntRow = new Plan<int>(row, nGroups_);
137	>	cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
138	>	cgPlanIntColumn = new Plan<int>(col, nGroups_);
139	>	cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
140
141	<	/**
142	<	* Constructor for ForceDecomposition Parallel Decomposition Method
143	<	* Will try to construct a symmetric grid of processors. Ideally, the
144	<	* number of processors will be a square ex: 4, 9, 16, 25.
95	<	*
96	<	*/
141	>	nAtomsInRow_ = AtomPlanIntRow->getSize();
142	>	nAtomsInCol_ = AtomPlanIntColumn->getSize();
143	>	nGroupsInRow_ = cgPlanIntRow->getSize();
144	>	nGroupsInCol_ = cgPlanIntColumn->getSize();
145
146	<	ForceDecomposition::ForceDecomposition() {
146	>	// Modify the data storage objects with the correct layouts and sizes:
147	>	atomRowData.resize(nAtomsInRow_);
148	>	atomRowData.setStorageLayout(storageLayout_);
149	>	atomColData.resize(nAtomsInCol_);
150	>	atomColData.setStorageLayout(storageLayout_);
151	>	cgRowData.resize(nGroupsInRow_);
152	>	cgRowData.setStorageLayout(DataStorage::dslPosition);
153	>	cgColData.resize(nGroupsInCol_);
154	>	if (needVelocities_)
155	>	// we only need column velocities if we need them.
156	>	cgColData.setStorageLayout(DataStorage::dslPosition |
157	>	DataStorage::dslVelocity);
158	>	else
159	>	cgColData.setStorageLayout(DataStorage::dslPosition);
160	>
161	>	identsRow.resize(nAtomsInRow_);
162	>	identsCol.resize(nAtomsInCol_);
163	>
164	>	AtomPlanIntRow->gather(idents, identsRow);
165	>	AtomPlanIntColumn->gather(idents, identsCol);
166	>
167	>	// allocate memory for the parallel objects
168	>	atypesRow.resize(nAtomsInRow_);
169	>	atypesCol.resize(nAtomsInCol_);
170
171	<	#ifdef IS_MPI
172	<	int nProcs = MPI::COMM_WORLD.Get_size();
173	<	int worldRank = MPI::COMM_WORLD.Get_rank();
171	>	for (int i = 0; i < nAtomsInRow_; i++)
172	>	atypesRow[i] = ff_->getAtomType(identsRow[i]);
173	>	for (int i = 0; i < nAtomsInCol_; i++)
174	>	atypesCol[i] = ff_->getAtomType(identsCol[i]);
175	>
176	>	pot_row.resize(nAtomsInRow_);
177	>	pot_col.resize(nAtomsInCol_);
178	>
179	>	expot_row.resize(nAtomsInRow_);
180	>	expot_col.resize(nAtomsInCol_);
181	>
182	>	AtomRowToGlobal.resize(nAtomsInRow_);
183	>	AtomColToGlobal.resize(nAtomsInCol_);
184	>	AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
185	>	AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
186	>
187	>	cgRowToGlobal.resize(nGroupsInRow_);
188	>	cgColToGlobal.resize(nGroupsInCol_);
189	>	cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
190	>	cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
191	>
192	>	massFactorsRow.resize(nAtomsInRow_);
193	>	massFactorsCol.resize(nAtomsInCol_);
194	>	AtomPlanRealRow->gather(massFactors, massFactorsRow);
195	>	AtomPlanRealColumn->gather(massFactors, massFactorsCol);
196	>
197	>	groupListRow_.clear();
198	>	groupListRow_.resize(nGroupsInRow_);
199	>	for (int i = 0; i < nGroupsInRow_; i++) {
200	>	int gid = cgRowToGlobal[i];
201	>	for (int j = 0; j < nAtomsInRow_; j++) {
202	>	int aid = AtomRowToGlobal[j];
203	>	if (globalGroupMembership[aid] == gid)
204	>	groupListRow_[i].push_back(j);
205	>	}
206	>	}
207	>
208	>	groupListCol_.clear();
209	>	groupListCol_.resize(nGroupsInCol_);
210	>	for (int i = 0; i < nGroupsInCol_; i++) {
211	>	int gid = cgColToGlobal[i];
212	>	for (int j = 0; j < nAtomsInCol_; j++) {
213	>	int aid = AtomColToGlobal[j];
214	>	if (globalGroupMembership[aid] == gid)
215	>	groupListCol_[i].push_back(j);
216	>	}
217	>	}
218	>
219	>	excludesForAtom.clear();
220	>	excludesForAtom.resize(nAtomsInRow_);
221	>	toposForAtom.clear();
222	>	toposForAtom.resize(nAtomsInRow_);
223	>	topoDist.clear();
224	>	topoDist.resize(nAtomsInRow_);
225	>	for (int i = 0; i < nAtomsInRow_; i++) {
226	>	int iglob = AtomRowToGlobal[i];
227	>
228	>	for (int j = 0; j < nAtomsInCol_; j++) {
229	>	int jglob = AtomColToGlobal[j];
230	>
231	>	if (excludes->hasPair(iglob, jglob))
232	>	excludesForAtom[i].push_back(j);
233	>
234	>	if (oneTwo->hasPair(iglob, jglob)) {
235	>	toposForAtom[i].push_back(j);
236	>	topoDist[i].push_back(1);
237	>	} else {
238	>	if (oneThree->hasPair(iglob, jglob)) {
239	>	toposForAtom[i].push_back(j);
240	>	topoDist[i].push_back(2);
241	>	} else {
242	>	if (oneFour->hasPair(iglob, jglob)) {
243	>	toposForAtom[i].push_back(j);
244	>	topoDist[i].push_back(3);
245	>	}
246	>	}
247	>	}
248	>	}
249	>	}
250	>
251	>	#else
252	>	excludesForAtom.clear();
253	>	excludesForAtom.resize(nLocal_);
254	>	toposForAtom.clear();
255	>	toposForAtom.resize(nLocal_);
256	>	topoDist.clear();
257	>	topoDist.resize(nLocal_);
258	>
259	>	for (int i = 0; i < nLocal_; i++) {
260	>	int iglob = AtomLocalToGlobal[i];
261	>
262	>	for (int j = 0; j < nLocal_; j++) {
263	>	int jglob = AtomLocalToGlobal[j];
264	>
265	>	if (excludes->hasPair(iglob, jglob))
266	>	excludesForAtom[i].push_back(j);
267	>
268	>	if (oneTwo->hasPair(iglob, jglob)) {
269	>	toposForAtom[i].push_back(j);
270	>	topoDist[i].push_back(1);
271	>	} else {
272	>	if (oneThree->hasPair(iglob, jglob)) {
273	>	toposForAtom[i].push_back(j);
274	>	topoDist[i].push_back(2);
275	>	} else {
276	>	if (oneFour->hasPair(iglob, jglob)) {
277	>	toposForAtom[i].push_back(j);
278	>	topoDist[i].push_back(3);
279	>	}
280	>	}
281	>	}
282	>	}
283	>	}
284		#endif
285
286	<	// First time through, construct column stride.
287	<	if (communicators.size() == 0)
288	<	{
289	<	int nColumnsMax = (int) round(sqrt((float) nProcs));
290	<	for (int i = 0; i < nProcs; ++i)
291	<	{
292	<	if (nProcs%i==0) nColumns=i;
286	>	// allocate memory for the parallel objects
287	>	atypesLocal.resize(nLocal_);
288	>
289	>	for (int i = 0; i < nLocal_; i++)
290	>	atypesLocal[i] = ff_->getAtomType(idents[i]);
291	>
292	>	groupList_.clear();
293	>	groupList_.resize(nGroups_);
294	>	for (int i = 0; i < nGroups_; i++) {
295	>	int gid = cgLocalToGlobal[i];
296	>	for (int j = 0; j < nLocal_; j++) {
297	>	int aid = AtomLocalToGlobal[j];
298	>	if (globalGroupMembership[aid] == gid) {
299	>	groupList_[i].push_back(j);
300	>	}
301	>	}
302		}
303
304	<	int nRows = nProcs/nColumns;
305	<	myRank_ = (int) worldRank%nColumns;
304	>
305	>	createGtypeCutoffMap();
306	>
307		}
308	<	else
309	<	{
310	<	myRank_ = myRank/nColumns;
311	<	}
312	<	MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0);
313	<
314	<	isColumn_ = false;
315	<
316	<	}
308	>
309	>	void ForceMatrixDecomposition::createGtypeCutoffMap() {
310	>
311	>	RealType tol = 1e-6;
312	>	largestRcut_ = 0.0;
313	>	int atid;
314	>	set<AtomType*> atypes = info_->getSimulatedAtomTypes();
315	>
316	>	map<int, RealType> atypeCutoff;
317	>
318	>	for (set<AtomType*>::iterator at = atypes.begin();
319	>	at != atypes.end(); ++at){
320	>	atid = (*at)->getIdent();
321	>	if (userChoseCutoff_)
322	>	atypeCutoff[atid] = userCutoff_;
323	>	else
324	>	atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at);
325	>	}
326	>
327	>	vector<RealType> gTypeCutoffs;
328	>	// first we do a single loop over the cutoff groups to find the
329	>	// largest cutoff for any atypes present in this group.
330	>	#ifdef IS_MPI
331	>	vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0);
332	>	groupRowToGtype.resize(nGroupsInRow_);
333	>	for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) {
334	>	vector<int> atomListRow = getAtomsInGroupRow(cg1);
335	>	for (vector<int>::iterator ia = atomListRow.begin();
336	>	ia != atomListRow.end(); ++ia) {
337	>	int atom1 = (*ia);
338	>	atid = identsRow[atom1];
339	>	if (atypeCutoff[atid] > groupCutoffRow[cg1]) {
340	>	groupCutoffRow[cg1] = atypeCutoff[atid];
341	>	}
342	>	}
343
344	<	ForceDecomposition::gather(sendbuf, receivebuf){
345	<	communicators(myIndex_).Allgatherv();
346	<	}
344	>	bool gTypeFound = false;
345	>	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
346	>	if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) {
347	>	groupRowToGtype[cg1] = gt;
348	>	gTypeFound = true;
349	>	}
350	>	}
351	>	if (!gTypeFound) {
352	>	gTypeCutoffs.push_back( groupCutoffRow[cg1] );
353	>	groupRowToGtype[cg1] = gTypeCutoffs.size() - 1;
354	>	}
355	>
356	>	}
357	>	vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0);
358	>	groupColToGtype.resize(nGroupsInCol_);
359	>	for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) {
360	>	vector<int> atomListCol = getAtomsInGroupColumn(cg2);
361	>	for (vector<int>::iterator jb = atomListCol.begin();
362	>	jb != atomListCol.end(); ++jb) {
363	>	int atom2 = (*jb);
364	>	atid = identsCol[atom2];
365	>	if (atypeCutoff[atid] > groupCutoffCol[cg2]) {
366	>	groupCutoffCol[cg2] = atypeCutoff[atid];
367	>	}
368	>	}
369	>	bool gTypeFound = false;
370	>	for (int gt = 0; gt < gTypeCutoffs.size(); gt++) {
371	>	if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) {
372	>	groupColToGtype[cg2] = gt;
373	>	gTypeFound = true;
374	>	}
375	>	}
376	>	if (!gTypeFound) {
377	>	gTypeCutoffs.push_back( groupCutoffCol[cg2] );
378	>	groupColToGtype[cg2] = gTypeCutoffs.size() - 1;
379	>	}
380	>	}
381	>	#else
382	>
383	>	vector<RealType> groupCutoff(nGroups_, 0.0);
384	>	groupToGtype.resize(nGroups_);
385	>	for (int cg1 = 0; cg1 < nGroups_; cg1++) {
386	>	groupCutoff[cg1] = 0.0;
387	>	vector<int> atomList = getAtomsInGroupRow(cg1);
388	>	for (vector<int>::iterator ia = atomList.begin();
389	>	ia != atomList.end(); ++ia) {
390	>	int atom1 = (*ia);
391	>	atid = idents[atom1];
392	>	if (atypeCutoff[atid] > groupCutoff[cg1])
393	>	groupCutoff[cg1] = atypeCutoff[atid];
394	>	}
395	>
396	>	bool gTypeFound = false;
397	>	for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) {
398	>	if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) {
399	>	groupToGtype[cg1] = gt;
400	>	gTypeFound = true;
401	>	}
402	>	}
403	>	if (!gTypeFound) {
404	>	gTypeCutoffs.push_back( groupCutoff[cg1] );
405	>	groupToGtype[cg1] = gTypeCutoffs.size() - 1;
406	>	}
407	>	}
408	>	#endif
409
410	+	// Now we find the maximum group cutoff value present in the simulation
411
412	+	RealType groupMax = *max_element(gTypeCutoffs.begin(),
413	+	gTypeCutoffs.end());
414
415	<	ForceDecomposition::scatter(sbuffer, rbuffer){
416	<	communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM);
417	<	}
415	>	#ifdef IS_MPI
416	>	MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE,
417	>	MPI::MAX);
418	>	#endif
419	>
420	>	RealType tradRcut = groupMax;
421
422	+	for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) {
423	+	for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) {
424	+	RealType thisRcut;
425	+	switch(cutoffPolicy_) {
426	+	case TRADITIONAL:
427	+	thisRcut = tradRcut;
428	+	break;
429	+	case MIX:
430	+	thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]);
431	+	break;
432	+	case MAX:
433	+	thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]);
434	+	break;
435	+	default:
436	+	sprintf(painCave.errMsg,
437	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
438	+	"hit an unknown cutoff policy!\n");
439	+	painCave.severity = OPENMD_ERROR;
440	+	painCave.isFatal = 1;
441	+	simError();
442	+	break;
443	+	}
444
445	+	pair<int,int> key = make_pair(i,j);
446	+	gTypeCutoffMap[key].first = thisRcut;
447	+	if (thisRcut > largestRcut_) largestRcut_ = thisRcut;
448	+	gTypeCutoffMap[key].second = thisRcut*thisRcut;
449	+	gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2);
450	+	// sanity check
451	+
452	+	if (userChoseCutoff_) {
453	+	if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) {
454	+	sprintf(painCave.errMsg,
455	+	"ForceMatrixDecomposition::createGtypeCutoffMap "
456	+	"user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_);
457	+	painCave.severity = OPENMD_ERROR;
458	+	painCave.isFatal = 1;
459	+	simError();
460	+	}
461	+	}
462	+	}
463	+	}
464	+	}
465	+
466	+	groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) {
467	+	int i, j;
468	+	#ifdef IS_MPI
469	+	i = groupRowToGtype[cg1];
470	+	j = groupColToGtype[cg2];
471	+	#else
472	+	i = groupToGtype[cg1];
473	+	j = groupToGtype[cg2];
474	+	#endif
475	+	return gTypeCutoffMap[make_pair(i,j)];
476	+	}
477	+
478	+	int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
479	+	for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
480	+	if (toposForAtom[atom1][j] == atom2)
481	+	return topoDist[atom1][j];
482	+	}
483	+	return 0;
484	+	}
485	+
486	+	void ForceMatrixDecomposition::zeroWorkArrays() {
487	+	pairwisePot = 0.0;
488	+	embeddingPot = 0.0;
489	+	excludedPot = 0.0;
490	+	excludedSelfPot = 0.0;
491	+
492	+	#ifdef IS_MPI
493	+	if (storageLayout_ & DataStorage::dslForce) {
494	+	fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
495	+	fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
496	+	}
497	+
498	+	if (storageLayout_ & DataStorage::dslTorque) {
499	+	fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
500	+	fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
501	+	}
502	+
503	+	fill(pot_row.begin(), pot_row.end(),
504	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
505	+
506	+	fill(pot_col.begin(), pot_col.end(),
507	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
508	+
509	+	fill(expot_row.begin(), expot_row.end(),
510	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
511	+
512	+	fill(expot_col.begin(), expot_col.end(),
513	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
514	+
515	+	if (storageLayout_ & DataStorage::dslParticlePot) {
516	+	fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
517	+	0.0);
518	+	fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
519	+	0.0);
520	+	}
521	+
522	+	if (storageLayout_ & DataStorage::dslDensity) {
523	+	fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
524	+	fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
525	+	}
526	+
527	+	if (storageLayout_ & DataStorage::dslFunctional) {
528	+	fill(atomRowData.functional.begin(), atomRowData.functional.end(),
529	+	0.0);
530	+	fill(atomColData.functional.begin(), atomColData.functional.end(),
531	+	0.0);
532	+	}
533	+
534	+	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
535	+	fill(atomRowData.functionalDerivative.begin(),
536	+	atomRowData.functionalDerivative.end(), 0.0);
537	+	fill(atomColData.functionalDerivative.begin(),
538	+	atomColData.functionalDerivative.end(), 0.0);
539	+	}
540	+
541	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
542	+	fill(atomRowData.skippedCharge.begin(),
543	+	atomRowData.skippedCharge.end(), 0.0);
544	+	fill(atomColData.skippedCharge.begin(),
545	+	atomColData.skippedCharge.end(), 0.0);
546	+	}
547	+
548	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
549	+	fill(atomRowData.flucQFrc.begin(),
550	+	atomRowData.flucQFrc.end(), 0.0);
551	+	fill(atomColData.flucQFrc.begin(),
552	+	atomColData.flucQFrc.end(), 0.0);
553	+	}
554	+
555	+	if (storageLayout_ & DataStorage::dslElectricField) {
556	+	fill(atomRowData.electricField.begin(),
557	+	atomRowData.electricField.end(), V3Zero);
558	+	fill(atomColData.electricField.begin(),
559	+	atomColData.electricField.end(), V3Zero);
560	+	}
561	+
562	+	#endif
563	+	// even in parallel, we need to zero out the local arrays:
564	+
565	+	if (storageLayout_ & DataStorage::dslParticlePot) {
566	+	fill(snap_->atomData.particlePot.begin(),
567	+	snap_->atomData.particlePot.end(), 0.0);
568	+	}
569	+
570	+	if (storageLayout_ & DataStorage::dslDensity) {
571	+	fill(snap_->atomData.density.begin(),
572	+	snap_->atomData.density.end(), 0.0);
573	+	}
574	+
575	+	if (storageLayout_ & DataStorage::dslFunctional) {
576	+	fill(snap_->atomData.functional.begin(),
577	+	snap_->atomData.functional.end(), 0.0);
578	+	}
579	+
580	+	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
581	+	fill(snap_->atomData.functionalDerivative.begin(),
582	+	snap_->atomData.functionalDerivative.end(), 0.0);
583	+	}
584	+
585	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
586	+	fill(snap_->atomData.skippedCharge.begin(),
587	+	snap_->atomData.skippedCharge.end(), 0.0);
588	+	}
589	+
590	+	if (storageLayout_ & DataStorage::dslElectricField) {
591	+	fill(snap_->atomData.electricField.begin(),
592	+	snap_->atomData.electricField.end(), V3Zero);
593	+	}
594	+	}
595	+
596	+
597	+	void ForceMatrixDecomposition::distributeData()  {
598	+	snap_ = sman_->getCurrentSnapshot();
599	+	storageLayout_ = sman_->getStorageLayout();
600	+	#ifdef IS_MPI
601	+
602	+	// gather up the atomic positions
603	+	AtomPlanVectorRow->gather(snap_->atomData.position,
604	+	atomRowData.position);
605	+	AtomPlanVectorColumn->gather(snap_->atomData.position,
606	+	atomColData.position);
607	+
608	+	// gather up the cutoff group positions
609	+
610	+	cgPlanVectorRow->gather(snap_->cgData.position,
611	+	cgRowData.position);
612	+
613	+	cgPlanVectorColumn->gather(snap_->cgData.position,
614	+	cgColData.position);
615	+
616	+
617	+
618	+	if (needVelocities_) {
619	+	// gather up the atomic velocities
620	+	AtomPlanVectorColumn->gather(snap_->atomData.velocity,
621	+	atomColData.velocity);
622	+
623	+	cgPlanVectorColumn->gather(snap_->cgData.velocity,
624	+	cgColData.velocity);
625	+	}
626	+
627	+
628	+	// if needed, gather the atomic rotation matrices
629	+	if (storageLayout_ & DataStorage::dslAmat) {
630	+	AtomPlanMatrixRow->gather(snap_->atomData.aMat,
631	+	atomRowData.aMat);
632	+	AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
633	+	atomColData.aMat);
634	+	}
635	+
636	+	// if needed, gather the atomic eletrostatic information
637	+	if (storageLayout_ & DataStorage::dslDipole) {
638	+	AtomPlanVectorRow->gather(snap_->atomData.dipole,
639	+	atomRowData.dipole);
640	+	AtomPlanVectorColumn->gather(snap_->atomData.dipole,
641	+	atomColData.dipole);
642	+	}
643	+
644	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
645	+	AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
646	+	atomRowData.quadrupole);
647	+	AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
648	+	atomColData.quadrupole);
649	+	}
650	+
651	+	// if needed, gather the atomic fluctuating charge values
652	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
653	+	AtomPlanRealRow->gather(snap_->atomData.flucQPos,
654	+	atomRowData.flucQPos);
655	+	AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
656	+	atomColData.flucQPos);
657	+	}
658	+
659	+	#endif
660	+	}
661	+
662	+	/* collects information obtained during the pre-pair loop onto local
663	+	* data structures.
664	+	*/
665	+	void ForceMatrixDecomposition::collectIntermediateData() {
666	+	snap_ = sman_->getCurrentSnapshot();
667	+	storageLayout_ = sman_->getStorageLayout();
668	+	#ifdef IS_MPI
669	+
670	+	if (storageLayout_ & DataStorage::dslDensity) {
671	+
672	+	AtomPlanRealRow->scatter(atomRowData.density,
673	+	snap_->atomData.density);
674	+
675	+	int n = snap_->atomData.density.size();
676	+	vector<RealType> rho_tmp(n, 0.0);
677	+	AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
678	+	for (int i = 0; i < n; i++)
679	+	snap_->atomData.density[i] += rho_tmp[i];
680	+	}
681	+
682	+	if (storageLayout_ & DataStorage::dslElectricField) {
683	+
684	+	AtomPlanVectorRow->scatter(atomRowData.electricField,
685	+	snap_->atomData.electricField);
686	+
687	+	int n = snap_->atomData.electricField.size();
688	+	vector<Vector3d> field_tmp(n, V3Zero);
689	+	AtomPlanVectorColumn->scatter(atomColData.electricField,
690	+	field_tmp);
691	+	for (int i = 0; i < n; i++)
692	+	snap_->atomData.electricField[i] += field_tmp[i];
693	+	}
694	+	#endif
695	+	}
696	+
697	+	/*
698	+	* redistributes information obtained during the pre-pair loop out to
699	+	* row and column-indexed data structures
700	+	*/
701	+	void ForceMatrixDecomposition::distributeIntermediateData() {
702	+	snap_ = sman_->getCurrentSnapshot();
703	+	storageLayout_ = sman_->getStorageLayout();
704	+	#ifdef IS_MPI
705	+	if (storageLayout_ & DataStorage::dslFunctional) {
706	+	AtomPlanRealRow->gather(snap_->atomData.functional,
707	+	atomRowData.functional);
708	+	AtomPlanRealColumn->gather(snap_->atomData.functional,
709	+	atomColData.functional);
710	+	}
711	+
712	+	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
713	+	AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
714	+	atomRowData.functionalDerivative);
715	+	AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
716	+	atomColData.functionalDerivative);
717	+	}
718	+	#endif
719	+	}
720	+
721	+
722	+	void ForceMatrixDecomposition::collectData() {
723	+	snap_ = sman_->getCurrentSnapshot();
724	+	storageLayout_ = sman_->getStorageLayout();
725	+	#ifdef IS_MPI
726	+	int n = snap_->atomData.force.size();
727	+	vector<Vector3d> frc_tmp(n, V3Zero);
728	+
729	+	AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
730	+	for (int i = 0; i < n; i++) {
731	+	snap_->atomData.force[i] += frc_tmp[i];
732	+	frc_tmp[i] = 0.0;
733	+	}
734	+
735	+	AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
736	+	for (int i = 0; i < n; i++) {
737	+	snap_->atomData.force[i] += frc_tmp[i];
738	+	}
739	+
740	+	if (storageLayout_ & DataStorage::dslTorque) {
741	+
742	+	int nt = snap_->atomData.torque.size();
743	+	vector<Vector3d> trq_tmp(nt, V3Zero);
744	+
745	+	AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
746	+	for (int i = 0; i < nt; i++) {
747	+	snap_->atomData.torque[i] += trq_tmp[i];
748	+	trq_tmp[i] = 0.0;
749	+	}
750	+
751	+	AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
752	+	for (int i = 0; i < nt; i++)
753	+	snap_->atomData.torque[i] += trq_tmp[i];
754	+	}
755	+
756	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
757	+
758	+	int ns = snap_->atomData.skippedCharge.size();
759	+	vector<RealType> skch_tmp(ns, 0.0);
760	+
761	+	AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
762	+	for (int i = 0; i < ns; i++) {
763	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
764	+	skch_tmp[i] = 0.0;
765	+	}
766	+
767	+	AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
768	+	for (int i = 0; i < ns; i++)
769	+	snap_->atomData.skippedCharge[i] += skch_tmp[i];
770	+
771	+	}
772	+
773	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
774	+
775	+	int nq = snap_->atomData.flucQFrc.size();
776	+	vector<RealType> fqfrc_tmp(nq, 0.0);
777	+
778	+	AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
779	+	for (int i = 0; i < nq; i++) {
780	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
781	+	fqfrc_tmp[i] = 0.0;
782	+	}
783	+
784	+	AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
785	+	for (int i = 0; i < nq; i++)
786	+	snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
787	+
788	+	}
789	+
790	+	nLocal_ = snap_->getNumberOfAtoms();
791	+
792	+	vector<potVec> pot_temp(nLocal_,
793	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
794	+	vector<potVec> expot_temp(nLocal_,
795	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
796	+
797	+	// scatter/gather pot_row into the members of my column
798	+
799	+	AtomPlanPotRow->scatter(pot_row, pot_temp);
800	+	AtomPlanPotRow->scatter(expot_row, expot_temp);
801	+
802	+	for (int ii = 0;  ii < pot_temp.size(); ii++ )
803	+	pairwisePot += pot_temp[ii];
804	+
805	+	for (int ii = 0;  ii < expot_temp.size(); ii++ )
806	+	excludedPot += expot_temp[ii];
807	+
808	+	if (storageLayout_ & DataStorage::dslParticlePot) {
809	+	// This is the pairwise contribution to the particle pot.  The
810	+	// embedding contribution is added in each of the low level
811	+	// non-bonded routines.  In single processor, this is done in
812	+	// unpackInteractionData, not in collectData.
813	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
814	+	for (int i = 0; i < nLocal_; i++) {
815	+	// factor of two is because the total potential terms are divided
816	+	// by 2 in parallel due to row/ column scatter
817	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
818	+	}
819	+	}
820	+	}
821	+
822	+	fill(pot_temp.begin(), pot_temp.end(),
823	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
824	+	fill(expot_temp.begin(), expot_temp.end(),
825	+	Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
826	+
827	+	AtomPlanPotColumn->scatter(pot_col, pot_temp);
828	+	AtomPlanPotColumn->scatter(expot_col, expot_temp);
829	+
830	+	for (int ii = 0;  ii < pot_temp.size(); ii++ )
831	+	pairwisePot += pot_temp[ii];
832	+
833	+	for (int ii = 0;  ii < expot_temp.size(); ii++ )
834	+	excludedPot += expot_temp[ii];
835	+
836	+	if (storageLayout_ & DataStorage::dslParticlePot) {
837	+	// This is the pairwise contribution to the particle pot.  The
838	+	// embedding contribution is added in each of the low level
839	+	// non-bonded routines.  In single processor, this is done in
840	+	// unpackInteractionData, not in collectData.
841	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
842	+	for (int i = 0; i < nLocal_; i++) {
843	+	// factor of two is because the total potential terms are divided
844	+	// by 2 in parallel due to row/ column scatter
845	+	snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
846	+	}
847	+	}
848	+	}
849	+
850	+	if (storageLayout_ & DataStorage::dslParticlePot) {
851	+	int npp = snap_->atomData.particlePot.size();
852	+	vector<RealType> ppot_temp(npp, 0.0);
853	+
854	+	// This is the direct or embedding contribution to the particle
855	+	// pot.
856	+
857	+	AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
858	+	for (int i = 0; i < npp; i++) {
859	+	snap_->atomData.particlePot[i] += ppot_temp[i];
860	+	}
861	+
862	+	fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
863	+
864	+	AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
865	+	for (int i = 0; i < npp; i++) {
866	+	snap_->atomData.particlePot[i] += ppot_temp[i];
867	+	}
868	+	}
869	+
870	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
871	+	RealType ploc1 = pairwisePot[ii];
872	+	RealType ploc2 = 0.0;
873	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
874	+	pairwisePot[ii] = ploc2;
875	+	}
876	+
877	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
878	+	RealType ploc1 = excludedPot[ii];
879	+	RealType ploc2 = 0.0;
880	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
881	+	excludedPot[ii] = ploc2;
882	+	}
883	+
884	+	// Here be dragons.
885	+	MPI::Intracomm col = colComm.getComm();
886	+
887	+	col.Allreduce(MPI::IN_PLACE,
888	+	&snap_->frameData.conductiveHeatFlux[0], 3,
889	+	MPI::REALTYPE, MPI::SUM);
890	+
891	+
892	+	#endif
893	+
894	+	}
895	+
896	+	/**
897	+	* Collects information obtained during the post-pair (and embedding
898	+	* functional) loops onto local data structures.
899	+	*/
900	+	void ForceMatrixDecomposition::collectSelfData() {
901	+	snap_ = sman_->getCurrentSnapshot();
902	+	storageLayout_ = sman_->getStorageLayout();
903	+
904	+	#ifdef IS_MPI
905	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
906	+	RealType ploc1 = embeddingPot[ii];
907	+	RealType ploc2 = 0.0;
908	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
909	+	embeddingPot[ii] = ploc2;
910	+	}
911	+	for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
912	+	RealType ploc1 = excludedSelfPot[ii];
913	+	RealType ploc2 = 0.0;
914	+	MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM);
915	+	excludedSelfPot[ii] = ploc2;
916	+	}
917	+	#endif
918	+
919	+	}
920	+
921	+
922	+
923	+	int ForceMatrixDecomposition::getNAtomsInRow() {
924	+	#ifdef IS_MPI
925	+	return nAtomsInRow_;
926	+	#else
927	+	return nLocal_;
928	+	#endif
929	+	}
930	+
931	+	/**
932	+	* returns the list of atoms belonging to this group.
933	+	*/
934	+	vector<int> ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
935	+	#ifdef IS_MPI
936	+	return groupListRow_[cg1];
937	+	#else
938	+	return groupList_[cg1];
939	+	#endif
940	+	}
941	+
942	+	vector<int> ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
943	+	#ifdef IS_MPI
944	+	return groupListCol_[cg2];
945	+	#else
946	+	return groupList_[cg2];
947	+	#endif
948	+	}
949	+
950	+	Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
951	+	Vector3d d;
952	+
953	+	#ifdef IS_MPI
954	+	d = cgColData.position[cg2] - cgRowData.position[cg1];
955	+	#else
956	+	d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
957	+	#endif
958	+
959	+	snap_->wrapVector(d);
960	+	return d;
961	+	}
962	+
963	+	Vector3d ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
964	+	#ifdef IS_MPI
965	+	return cgColData.velocity[cg2];
966	+	#else
967	+	return snap_->cgData.velocity[cg2];
968	+	#endif
969	+	}
970	+
971	+	Vector3d ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
972	+	#ifdef IS_MPI
973	+	return atomColData.velocity[atom2];
974	+	#else
975	+	return snap_->atomData.velocity[atom2];
976	+	#endif
977	+	}
978	+
979	+
980	+	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
981	+
982	+	Vector3d d;
983	+
984	+	#ifdef IS_MPI
985	+	d = cgRowData.position[cg1] - atomRowData.position[atom1];
986	+	#else
987	+	d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
988	+	#endif
989	+
990	+	snap_->wrapVector(d);
991	+	return d;
992	+	}
993	+
994	+	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
995	+	Vector3d d;
996	+
997	+	#ifdef IS_MPI
998	+	d = cgColData.position[cg2] - atomColData.position[atom2];
999	+	#else
1000	+	d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
1001	+	#endif
1002	+
1003	+	snap_->wrapVector(d);
1004	+	return d;
1005	+	}
1006	+
1007	+	RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) {
1008	+	#ifdef IS_MPI
1009	+	return massFactorsRow[atom1];
1010	+	#else
1011	+	return massFactors[atom1];
1012	+	#endif
1013	+	}
1014	+
1015	+	RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
1016	+	#ifdef IS_MPI
1017	+	return massFactorsCol[atom2];
1018	+	#else
1019	+	return massFactors[atom2];
1020	+	#endif
1021	+
1022	+	}
1023	+
1024	+	Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
1025	+	Vector3d d;
1026	+
1027	+	#ifdef IS_MPI
1028	+	d = atomColData.position[atom2] - atomRowData.position[atom1];
1029	+	#else
1030	+	d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
1031	+	#endif
1032	+
1033	+	snap_->wrapVector(d);
1034	+	return d;
1035	+	}
1036	+
1037	+	vector<int> ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
1038	+	return excludesForAtom[atom1];
1039	+	}
1040	+
1041	+	/**
1042	+	* We need to exclude some overcounted interactions that result from
1043	+	* the parallel decomposition.
1044	+	*/
1045	+	bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
1046	+	int unique_id_1, unique_id_2;
1047	+
1048	+	#ifdef IS_MPI
1049	+	// in MPI, we have to look up the unique IDs for each atom
1050	+	unique_id_1 = AtomRowToGlobal[atom1];
1051	+	unique_id_2 = AtomColToGlobal[atom2];
1052	+	// group1 = cgRowToGlobal[cg1];
1053	+	// group2 = cgColToGlobal[cg2];
1054	+	#else
1055	+	unique_id_1 = AtomLocalToGlobal[atom1];
1056	+	unique_id_2 = AtomLocalToGlobal[atom2];
1057	+	int group1 = cgLocalToGlobal[cg1];
1058	+	int group2 = cgLocalToGlobal[cg2];
1059	+	#endif
1060	+
1061	+	if (unique_id_1 == unique_id_2) return true;
1062	+
1063	+	#ifdef IS_MPI
1064	+	// this prevents us from doing the pair on multiple processors
1065	+	if (unique_id_1 < unique_id_2) {
1066	+	if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
1067	+	} else {
1068	+	if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
1069	+	}
1070	+	#endif
1071	+
1072	+	#ifndef IS_MPI
1073	+	if (group1 == group2) {
1074	+	if (unique_id_1 < unique_id_2) return true;
1075	+	}
1076	+	#endif
1077	+
1078	+	return false;
1079	+	}
1080	+
1081	+	/**
1082	+	* We need to handle the interactions for atoms who are involved in
1083	+	* the same rigid body as well as some short range interactions
1084	+	* (bonds, bends, torsions) differently from other interactions.
1085	+	* We'll still visit the pairwise routines, but with a flag that
1086	+	* tells those routines to exclude the pair from direct long range
1087	+	* interactions.  Some indirect interactions (notably reaction
1088	+	* field) must still be handled for these pairs.
1089	+	*/
1090	+	bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
1091	+
1092	+	// excludesForAtom was constructed to use row/column indices in the MPI
1093	+	// version, and to use local IDs in the non-MPI version:
1094	+
1095	+	for (vector<int>::iterator i = excludesForAtom[atom1].begin();
1096	+	i != excludesForAtom[atom1].end(); ++i) {
1097	+	if ( (*i) == atom2 ) return true;
1098	+	}
1099	+
1100	+	return false;
1101	+	}
1102	+
1103	+
1104	+	void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
1105	+	#ifdef IS_MPI
1106	+	atomRowData.force[atom1] += fg;
1107	+	#else
1108	+	snap_->atomData.force[atom1] += fg;
1109	+	#endif
1110	+	}
1111	+
1112	+	void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
1113	+	#ifdef IS_MPI
1114	+	atomColData.force[atom2] += fg;
1115	+	#else
1116	+	snap_->atomData.force[atom2] += fg;
1117	+	#endif
1118	+	}
1119	+
1120	+	// filling interaction blocks with pointers
1121	+	void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1122	+	int atom1, int atom2) {
1123	+
1124	+	idat.excluded = excludeAtomPair(atom1, atom2);
1125	+
1126	+	#ifdef IS_MPI
1127	+	idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]);
1128	+	//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]),
1129	+	//                         ff_->getAtomType(identsCol[atom2]) );
1130	+
1131	+	if (storageLayout_ & DataStorage::dslAmat) {
1132	+	idat.A1 = &(atomRowData.aMat[atom1]);
1133	+	idat.A2 = &(atomColData.aMat[atom2]);
1134	+	}
1135	+
1136	+	if (storageLayout_ & DataStorage::dslTorque) {
1137	+	idat.t1 = &(atomRowData.torque[atom1]);
1138	+	idat.t2 = &(atomColData.torque[atom2]);
1139	+	}
1140	+
1141	+	if (storageLayout_ & DataStorage::dslDipole) {
1142	+	idat.dipole1 = &(atomRowData.dipole[atom1]);
1143	+	idat.dipole2 = &(atomColData.dipole[atom2]);
1144	+	}
1145	+
1146	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1147	+	idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1148	+	idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1149	+	}
1150	+
1151	+	if (storageLayout_ & DataStorage::dslDensity) {
1152	+	idat.rho1 = &(atomRowData.density[atom1]);
1153	+	idat.rho2 = &(atomColData.density[atom2]);
1154	+	}
1155	+
1156	+	if (storageLayout_ & DataStorage::dslFunctional) {
1157	+	idat.frho1 = &(atomRowData.functional[atom1]);
1158	+	idat.frho2 = &(atomColData.functional[atom2]);
1159	+	}
1160	+
1161	+	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1162	+	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1163	+	idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1164	+	}
1165	+
1166	+	if (storageLayout_ & DataStorage::dslParticlePot) {
1167	+	idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1168	+	idat.particlePot2 = &(atomColData.particlePot[atom2]);
1169	+	}
1170	+
1171	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1172	+	idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1173	+	idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1174	+	}
1175	+
1176	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1177	+	idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1178	+	idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1179	+	}
1180	+
1181	+	#else
1182	+
1183	+	idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]);
1184	+
1185	+	if (storageLayout_ & DataStorage::dslAmat) {
1186	+	idat.A1 = &(snap_->atomData.aMat[atom1]);
1187	+	idat.A2 = &(snap_->atomData.aMat[atom2]);
1188	+	}
1189	+
1190	+	RealType ct = dot(idat.A1->getColumn(2), idat.A2->getColumn(2));
1191	+
1192	+	if (storageLayout_ & DataStorage::dslTorque) {
1193	+	idat.t1 = &(snap_->atomData.torque[atom1]);
1194	+	idat.t2 = &(snap_->atomData.torque[atom2]);
1195	+	}
1196	+
1197	+	if (storageLayout_ & DataStorage::dslDipole) {
1198	+	idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1199	+	idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1200	+	}
1201	+
1202	+	if (storageLayout_ & DataStorage::dslQuadrupole) {
1203	+	idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1204	+	idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1205	+	}
1206	+
1207	+	if (storageLayout_ & DataStorage::dslDensity) {
1208	+	idat.rho1 = &(snap_->atomData.density[atom1]);
1209	+	idat.rho2 = &(snap_->atomData.density[atom2]);
1210	+	}
1211	+
1212	+	if (storageLayout_ & DataStorage::dslFunctional) {
1213	+	idat.frho1 = &(snap_->atomData.functional[atom1]);
1214	+	idat.frho2 = &(snap_->atomData.functional[atom2]);
1215	+	}
1216	+
1217	+	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1218	+	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1219	+	idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1220	+	}
1221	+
1222	+	if (storageLayout_ & DataStorage::dslParticlePot) {
1223	+	idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1224	+	idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1225	+	}
1226	+
1227	+	if (storageLayout_ & DataStorage::dslSkippedCharge) {
1228	+	idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1229	+	idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1230	+	}
1231	+
1232	+	if (storageLayout_ & DataStorage::dslFlucQPosition) {
1233	+	idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1234	+	idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1235	+	}
1236	+
1237	+	#endif
1238	+	}
1239	+
1240	+
1241	+	void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) {
1242	+	#ifdef IS_MPI
1243	+	pot_row[atom1] += RealType(0.5) *  *(idat.pot);
1244	+	pot_col[atom2] += RealType(0.5) *  *(idat.pot);
1245	+	expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot);
1246	+	expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot);
1247	+
1248	+	atomRowData.force[atom1] += *(idat.f1);
1249	+	atomColData.force[atom2] -= *(idat.f1);
1250	+
1251	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1252	+	atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1253	+	atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1254	+	}
1255	+
1256	+	if (storageLayout_ & DataStorage::dslElectricField) {
1257	+	atomRowData.electricField[atom1] += *(idat.eField1);
1258	+	atomColData.electricField[atom2] += *(idat.eField2);
1259	+	}
1260	+
1261	+	#else
1262	+	pairwisePot += *(idat.pot);
1263	+	excludedPot += *(idat.excludedPot);
1264	+
1265	+	snap_->atomData.force[atom1] += *(idat.f1);
1266	+	snap_->atomData.force[atom2] -= *(idat.f1);
1267	+
1268	+	if (idat.doParticlePot) {
1269	+	// This is the pairwise contribution to the particle pot.  The
1270	+	// embedding contribution is added in each of the low level
1271	+	// non-bonded routines.  In parallel, this calculation is done
1272	+	// in collectData, not in unpackInteractionData.
1273	+	snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1274	+	snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1275	+	}
1276	+
1277	+	if (storageLayout_ & DataStorage::dslFlucQForce) {
1278	+	snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1279	+	snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1280	+	}
1281	+
1282	+	if (storageLayout_ & DataStorage::dslElectricField) {
1283	+	snap_->atomData.electricField[atom1] += *(idat.eField1);
1284	+	snap_->atomData.electricField[atom2] += *(idat.eField2);
1285	+	}
1286	+
1287	+	#endif
1288	+
1289	+	}
1290	+
1291	+	/*
1292	+	* buildNeighborList
1293	+	*
1294	+	* first element of pair is row-indexed CutoffGroup
1295	+	* second element of pair is column-indexed CutoffGroup
1296	+	*/
1297	+	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
1298	+
1299	+	vector<pair<int, int> > neighborList;
1300	+	groupCutoffs cuts;
1301	+	bool doAllPairs = false;
1302	+
1303	+	#ifdef IS_MPI
1304	+	cellListRow_.clear();
1305	+	cellListCol_.clear();
1306	+	#else
1307	+	cellList_.clear();
1308	+	#endif
1309	+
1310	+	RealType rList_ = (largestRcut_ + skinThickness_);
1311	+	RealType rl2 = rList_ * rList_;
1312	+	Snapshot* snap_ = sman_->getCurrentSnapshot();
1313	+	Mat3x3d Hmat = snap_->getHmat();
1314	+	Vector3d Hx = Hmat.getColumn(0);
1315	+	Vector3d Hy = Hmat.getColumn(1);
1316	+	Vector3d Hz = Hmat.getColumn(2);
1317	+
1318	+	nCells_.x() = (int) ( Hx.length() )/ rList_;
1319	+	nCells_.y() = (int) ( Hy.length() )/ rList_;
1320	+	nCells_.z() = (int) ( Hz.length() )/ rList_;
1321	+
1322	+	// handle small boxes where the cell offsets can end up repeating cells
1323	+
1324	+	if (nCells_.x() < 3) doAllPairs = true;
1325	+	if (nCells_.y() < 3) doAllPairs = true;
1326	+	if (nCells_.z() < 3) doAllPairs = true;
1327	+
1328	+	Mat3x3d invHmat = snap_->getInvHmat();
1329	+	Vector3d rs, scaled, dr;
1330	+	Vector3i whichCell;
1331	+	int cellIndex;
1332	+	int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1333	+
1334	+	#ifdef IS_MPI
1335	+	cellListRow_.resize(nCtot);
1336	+	cellListCol_.resize(nCtot);
1337	+	#else
1338	+	cellList_.resize(nCtot);
1339	+	#endif
1340	+
1341	+	if (!doAllPairs) {
1342	+	#ifdef IS_MPI
1343	+
1344	+	for (int i = 0; i < nGroupsInRow_; i++) {
1345	+	rs = cgRowData.position[i];
1346	+
1347	+	// scaled positions relative to the box vectors
1348	+	scaled = invHmat * rs;
1349	+
1350	+	// wrap the vector back into the unit box by subtracting integer box
1351	+	// numbers
1352	+	for (int j = 0; j < 3; j++) {
1353	+	scaled[j] -= roundMe(scaled[j]);
1354	+	scaled[j] += 0.5;
1355	+	// Handle the special case when an object is exactly on the
1356	+	// boundary (a scaled coordinate of 1.0 is the same as
1357	+	// scaled coordinate of 0.0)
1358	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1359	+	}
1360	+
1361	+	// find xyz-indices of cell that cutoffGroup is in.
1362	+	whichCell.x() = nCells_.x() * scaled.x();
1363	+	whichCell.y() = nCells_.y() * scaled.y();
1364	+	whichCell.z() = nCells_.z() * scaled.z();
1365	+
1366	+	// find single index of this cell:
1367	+	cellIndex = Vlinear(whichCell, nCells_);
1368	+
1369	+	// add this cutoff group to the list of groups in this cell;
1370	+	cellListRow_[cellIndex].push_back(i);
1371	+	}
1372	+	for (int i = 0; i < nGroupsInCol_; i++) {
1373	+	rs = cgColData.position[i];
1374	+
1375	+	// scaled positions relative to the box vectors
1376	+	scaled = invHmat * rs;
1377	+
1378	+	// wrap the vector back into the unit box by subtracting integer box
1379	+	// numbers
1380	+	for (int j = 0; j < 3; j++) {
1381	+	scaled[j] -= roundMe(scaled[j]);
1382	+	scaled[j] += 0.5;
1383	+	// Handle the special case when an object is exactly on the
1384	+	// boundary (a scaled coordinate of 1.0 is the same as
1385	+	// scaled coordinate of 0.0)
1386	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1387	+	}
1388	+
1389	+	// find xyz-indices of cell that cutoffGroup is in.
1390	+	whichCell.x() = nCells_.x() * scaled.x();
1391	+	whichCell.y() = nCells_.y() * scaled.y();
1392	+	whichCell.z() = nCells_.z() * scaled.z();
1393	+
1394	+	// find single index of this cell:
1395	+	cellIndex = Vlinear(whichCell, nCells_);
1396	+
1397	+	// add this cutoff group to the list of groups in this cell;
1398	+	cellListCol_[cellIndex].push_back(i);
1399	+	}
1400	+
1401	+	#else
1402	+	for (int i = 0; i < nGroups_; i++) {
1403	+	rs = snap_->cgData.position[i];
1404	+
1405	+	// scaled positions relative to the box vectors
1406	+	scaled = invHmat * rs;
1407	+
1408	+	// wrap the vector back into the unit box by subtracting integer box
1409	+	// numbers
1410	+	for (int j = 0; j < 3; j++) {
1411	+	scaled[j] -= roundMe(scaled[j]);
1412	+	scaled[j] += 0.5;
1413	+	// Handle the special case when an object is exactly on the
1414	+	// boundary (a scaled coordinate of 1.0 is the same as
1415	+	// scaled coordinate of 0.0)
1416	+	if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1417	+	}
1418	+
1419	+	// find xyz-indices of cell that cutoffGroup is in.
1420	+	whichCell.x() = nCells_.x() * scaled.x();
1421	+	whichCell.y() = nCells_.y() * scaled.y();
1422	+	whichCell.z() = nCells_.z() * scaled.z();
1423	+
1424	+	// find single index of this cell:
1425	+	cellIndex = Vlinear(whichCell, nCells_);
1426	+
1427	+	// add this cutoff group to the list of groups in this cell;
1428	+	cellList_[cellIndex].push_back(i);
1429	+	}
1430	+
1431	+	#endif
1432	+
1433	+	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
1434	+	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
1435	+	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
1436	+	Vector3i m1v(m1x, m1y, m1z);
1437	+	int m1 = Vlinear(m1v, nCells_);
1438	+
1439	+	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1440	+	os != cellOffsets_.end(); ++os) {
1441	+
1442	+	Vector3i m2v = m1v + (*os);
1443	+
1444	+
1445	+	if (m2v.x() >= nCells_.x()) {
1446	+	m2v.x() = 0;
1447	+	} else if (m2v.x() < 0) {
1448	+	m2v.x() = nCells_.x() - 1;
1449	+	}
1450	+
1451	+	if (m2v.y() >= nCells_.y()) {
1452	+	m2v.y() = 0;
1453	+	} else if (m2v.y() < 0) {
1454	+	m2v.y() = nCells_.y() - 1;
1455	+	}
1456	+
1457	+	if (m2v.z() >= nCells_.z()) {
1458	+	m2v.z() = 0;
1459	+	} else if (m2v.z() < 0) {
1460	+	m2v.z() = nCells_.z() - 1;
1461	+	}
1462	+
1463	+	int m2 = Vlinear (m2v, nCells_);
1464	+
1465	+	#ifdef IS_MPI
1466	+	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
1467	+	j1 != cellListRow_[m1].end(); ++j1) {
1468	+	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1469	+	j2 != cellListCol_[m2].end(); ++j2) {
1470	+
1471	+	// In parallel, we need to visit *all* pairs of row
1472	+	// & column indicies and will divide labor in the
1473	+	// force evaluation later.
1474	+	dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)];
1475	+	snap_->wrapVector(dr);
1476	+	cuts = getGroupCutoffs( (*j1), (*j2) );
1477	+	if (dr.lengthSquare() < cuts.third) {
1478	+	neighborList.push_back(make_pair((*j1), (*j2)));
1479	+	}
1480	+	}
1481	+	}
1482	+	#else
1483	+	for (vector<int>::iterator j1 = cellList_[m1].begin();
1484	+	j1 != cellList_[m1].end(); ++j1) {
1485	+	for (vector<int>::iterator j2 = cellList_[m2].begin();
1486	+	j2 != cellList_[m2].end(); ++j2) {
1487	+
1488	+	// Always do this if we're in different cells or if
1489	+	// we're in the same cell and the global index of
1490	+	// the j2 cutoff group is greater than or equal to
1491	+	// the j1 cutoff group.  Note that Rappaport's code
1492	+	// has a "less than" conditional here, but that
1493	+	// deals with atom-by-atom computation.  OpenMD
1494	+	// allows atoms within a single cutoff group to
1495	+	// interact with each other.
1496	+
1497	+
1498	+
1499	+	if (m2 != m1 || (*j2) >= (*j1) ) {
1500	+
1501	+	dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)];
1502	+	snap_->wrapVector(dr);
1503	+	cuts = getGroupCutoffs( (*j1), (*j2) );
1504	+	if (dr.lengthSquare() < cuts.third) {
1505	+	neighborList.push_back(make_pair((*j1), (*j2)));
1506	+	}
1507	+	}
1508	+	}
1509	+	}
1510	+	#endif
1511	+	}
1512	+	}
1513	+	}
1514	+	}
1515	+	} else {
1516	+	// branch to do all cutoff group pairs
1517	+	#ifdef IS_MPI
1518	+	for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1519	+	for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1520	+	dr = cgColData.position[j2] - cgRowData.position[j1];
1521	+	snap_->wrapVector(dr);
1522	+	cuts = getGroupCutoffs( j1, j2 );
1523	+	if (dr.lengthSquare() < cuts.third) {
1524	+	neighborList.push_back(make_pair(j1, j2));
1525	+	}
1526	+	}
1527	+	}
1528	+	#else
1529	+	// include all groups here.
1530	+	for (int j1 = 0; j1 < nGroups_; j1++) {
1531	+	// include self group interactions j2 == j1
1532	+	for (int j2 = j1; j2 < nGroups_; j2++) {
1533	+	dr = snap_->cgData.position[j2] - snap_->cgData.position[j1];
1534	+	snap_->wrapVector(dr);
1535	+	cuts = getGroupCutoffs( j1, j2 );
1536	+	if (dr.lengthSquare() < cuts.third) {
1537	+	neighborList.push_back(make_pair(j1, j2));
1538	+	}
1539	+	}
1540	+	}
1541	+	#endif
1542	+	}
1543	+
1544	+	// save the local cutoff group positions for the check that is
1545	+	// done on each loop:
1546	+	saved_CG_positions_.clear();
1547	+	for (int i = 0; i < nGroups_; i++)
1548	+	saved_CG_positions_.push_back(snap_->cgData.position[i]);
1549	+
1550	+	return neighborList;
1551	+	}
1552	+	} //end namespace OpenMD

Diff Legend

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