# | 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 42 | Line 35 | |
35 | * | |
36 | * [1] Meineke, et al., J. Comp. Chem. 26, 252-271 (2005). | |
37 | * [2] Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006). | |
38 | < | * [3] Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008). |
39 | < | * [4] Vardeman & Gezelter, in progress (2009). |
38 | > | * [3] Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008). |
39 | > | * [4] Kuang & Gezelter, J. Chem. Phys. 133, 164101 (2010). |
40 | > | * [5] Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011). |
41 | */ | |
42 | + | #include "parallel/ForceMatrixDecomposition.hpp" |
43 | + | #include "math/SquareMatrix3.hpp" |
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 | + | regions = info_->getRegions(); |
103 | + | AtomLocalToGlobal = info_->getGlobalAtomIndices(); |
104 | + | cgLocalToGlobal = info_->getGlobalGroupIndices(); |
105 | + | vector<int> globalGroupMembership = info_->getGlobalGroupMembership(); |
106 | ||
107 | < | using namespace std; |
66 | < | using namespace OpenMD; |
107 | > | massFactors = info_->getMassFactors(); |
108 | ||
109 | < | //__static |
109 | > | PairList* excludes = info_->getExcludedInteractions(); |
110 | > | PairList* oneTwo = info_->getOneTwoInteractions(); |
111 | > | PairList* oneThree = info_->getOneThreeInteractions(); |
112 | > | PairList* oneFour = info_->getOneFourInteractions(); |
113 | > | |
114 | > | if (needVelocities_) |
115 | > | snap_->cgData.setStorageLayout(DataStorage::dslPosition | |
116 | > | DataStorage::dslVelocity); |
117 | > | else |
118 | > | snap_->cgData.setStorageLayout(DataStorage::dslPosition); |
119 | > | |
120 | #ifdef IS_MPI | |
121 | < | static vector<MPI:Comm> communictors; |
122 | < | #endif |
121 | > | |
122 | > | MPI_Comm row = rowComm.getComm(); |
123 | > | MPI_Comm col = colComm.getComm(); |
124 | ||
125 | < | //____ MPITypeTraits |
126 | < | template<typename T> |
127 | < | struct MPITypeTraits; |
125 | > | AtomPlanIntRow = new Plan<int>(row, nLocal_); |
126 | > | AtomPlanRealRow = new Plan<RealType>(row, nLocal_); |
127 | > | AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_); |
128 | > | AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_); |
129 | > | AtomPlanPotRow = new Plan<potVec>(row, nLocal_); |
130 | ||
131 | < | #ifdef IS_MPI |
132 | < | template<> |
133 | < | struct MPITypeTraits<RealType> { |
134 | < | static const MPI::Datatype datatype; |
135 | < | }; |
82 | < | const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL; |
131 | > | AtomPlanIntColumn = new Plan<int>(col, nLocal_); |
132 | > | AtomPlanRealColumn = new Plan<RealType>(col, nLocal_); |
133 | > | AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_); |
134 | > | AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_); |
135 | > | AtomPlanPotColumn = new Plan<potVec>(col, nLocal_); |
136 | ||
137 | < | template<> |
138 | < | struct MPITypeTraits<int> { |
139 | < | static const MPI::Datatype datatype; |
140 | < | }; |
88 | < | const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT; |
89 | < | #endif |
137 | > | cgPlanIntRow = new Plan<int>(row, nGroups_); |
138 | > | cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_); |
139 | > | cgPlanIntColumn = new Plan<int>(col, nGroups_); |
140 | > | cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_); |
141 | ||
142 | < | /** |
143 | < | * Constructor for ForceDecomposition Parallel Decomposition Method |
144 | < | * Will try to construct a symmetric grid of processors. Ideally, the |
145 | < | * number of processors will be a square ex: 4, 9, 16, 25. |
95 | < | * |
96 | < | */ |
142 | > | nAtomsInRow_ = AtomPlanIntRow->getSize(); |
143 | > | nAtomsInCol_ = AtomPlanIntColumn->getSize(); |
144 | > | nGroupsInRow_ = cgPlanIntRow->getSize(); |
145 | > | nGroupsInCol_ = cgPlanIntColumn->getSize(); |
146 | ||
147 | < | ForceDecomposition::ForceDecomposition() { |
147 | > | // Modify the data storage objects with the correct layouts and sizes: |
148 | > | atomRowData.resize(nAtomsInRow_); |
149 | > | atomRowData.setStorageLayout(storageLayout_); |
150 | > | atomColData.resize(nAtomsInCol_); |
151 | > | atomColData.setStorageLayout(storageLayout_); |
152 | > | cgRowData.resize(nGroupsInRow_); |
153 | > | cgRowData.setStorageLayout(DataStorage::dslPosition); |
154 | > | cgColData.resize(nGroupsInCol_); |
155 | > | if (needVelocities_) |
156 | > | // we only need column velocities if we need them. |
157 | > | cgColData.setStorageLayout(DataStorage::dslPosition | |
158 | > | DataStorage::dslVelocity); |
159 | > | else |
160 | > | cgColData.setStorageLayout(DataStorage::dslPosition); |
161 | > | |
162 | > | identsRow.resize(nAtomsInRow_); |
163 | > | identsCol.resize(nAtomsInCol_); |
164 | > | |
165 | > | AtomPlanIntRow->gather(idents, identsRow); |
166 | > | AtomPlanIntColumn->gather(idents, identsCol); |
167 | ||
168 | < | #ifdef IS_MPI |
169 | < | int nProcs = MPI::COMM_WORLD.Get_size(); |
170 | < | int worldRank = MPI::COMM_WORLD.Get_rank(); |
168 | > | regionsRow.resize(nAtomsInRow_); |
169 | > | regionsCol.resize(nAtomsInCol_); |
170 | > | |
171 | > | AtomPlanIntRow->gather(regions, regionsRow); |
172 | > | AtomPlanIntColumn->gather(regions, regionsCol); |
173 | > | |
174 | > | // allocate memory for the parallel objects |
175 | > | atypesRow.resize(nAtomsInRow_); |
176 | > | atypesCol.resize(nAtomsInCol_); |
177 | > | |
178 | > | for (int i = 0; i < nAtomsInRow_; i++) |
179 | > | atypesRow[i] = ff_->getAtomType(identsRow[i]); |
180 | > | for (int i = 0; i < nAtomsInCol_; i++) |
181 | > | atypesCol[i] = ff_->getAtomType(identsCol[i]); |
182 | > | |
183 | > | pot_row.resize(nAtomsInRow_); |
184 | > | pot_col.resize(nAtomsInCol_); |
185 | > | |
186 | > | expot_row.resize(nAtomsInRow_); |
187 | > | expot_col.resize(nAtomsInCol_); |
188 | > | |
189 | > | AtomRowToGlobal.resize(nAtomsInRow_); |
190 | > | AtomColToGlobal.resize(nAtomsInCol_); |
191 | > | AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal); |
192 | > | AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal); |
193 | > | |
194 | > | cgRowToGlobal.resize(nGroupsInRow_); |
195 | > | cgColToGlobal.resize(nGroupsInCol_); |
196 | > | cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal); |
197 | > | cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal); |
198 | > | |
199 | > | massFactorsRow.resize(nAtomsInRow_); |
200 | > | massFactorsCol.resize(nAtomsInCol_); |
201 | > | AtomPlanRealRow->gather(massFactors, massFactorsRow); |
202 | > | AtomPlanRealColumn->gather(massFactors, massFactorsCol); |
203 | > | |
204 | > | groupListRow_.clear(); |
205 | > | groupListRow_.resize(nGroupsInRow_); |
206 | > | for (int i = 0; i < nGroupsInRow_; i++) { |
207 | > | int gid = cgRowToGlobal[i]; |
208 | > | for (int j = 0; j < nAtomsInRow_; j++) { |
209 | > | int aid = AtomRowToGlobal[j]; |
210 | > | if (globalGroupMembership[aid] == gid) |
211 | > | groupListRow_[i].push_back(j); |
212 | > | } |
213 | > | } |
214 | > | |
215 | > | groupListCol_.clear(); |
216 | > | groupListCol_.resize(nGroupsInCol_); |
217 | > | for (int i = 0; i < nGroupsInCol_; i++) { |
218 | > | int gid = cgColToGlobal[i]; |
219 | > | for (int j = 0; j < nAtomsInCol_; j++) { |
220 | > | int aid = AtomColToGlobal[j]; |
221 | > | if (globalGroupMembership[aid] == gid) |
222 | > | groupListCol_[i].push_back(j); |
223 | > | } |
224 | > | } |
225 | > | |
226 | > | excludesForAtom.clear(); |
227 | > | excludesForAtom.resize(nAtomsInRow_); |
228 | > | toposForAtom.clear(); |
229 | > | toposForAtom.resize(nAtomsInRow_); |
230 | > | topoDist.clear(); |
231 | > | topoDist.resize(nAtomsInRow_); |
232 | > | for (int i = 0; i < nAtomsInRow_; i++) { |
233 | > | int iglob = AtomRowToGlobal[i]; |
234 | > | |
235 | > | for (int j = 0; j < nAtomsInCol_; j++) { |
236 | > | int jglob = AtomColToGlobal[j]; |
237 | > | |
238 | > | if (excludes->hasPair(iglob, jglob)) |
239 | > | excludesForAtom[i].push_back(j); |
240 | > | |
241 | > | if (oneTwo->hasPair(iglob, jglob)) { |
242 | > | toposForAtom[i].push_back(j); |
243 | > | topoDist[i].push_back(1); |
244 | > | } else { |
245 | > | if (oneThree->hasPair(iglob, jglob)) { |
246 | > | toposForAtom[i].push_back(j); |
247 | > | topoDist[i].push_back(2); |
248 | > | } else { |
249 | > | if (oneFour->hasPair(iglob, jglob)) { |
250 | > | toposForAtom[i].push_back(j); |
251 | > | topoDist[i].push_back(3); |
252 | > | } |
253 | > | } |
254 | > | } |
255 | > | } |
256 | > | } |
257 | > | |
258 | > | #else |
259 | > | excludesForAtom.clear(); |
260 | > | excludesForAtom.resize(nLocal_); |
261 | > | toposForAtom.clear(); |
262 | > | toposForAtom.resize(nLocal_); |
263 | > | topoDist.clear(); |
264 | > | topoDist.resize(nLocal_); |
265 | > | |
266 | > | for (int i = 0; i < nLocal_; i++) { |
267 | > | int iglob = AtomLocalToGlobal[i]; |
268 | > | |
269 | > | for (int j = 0; j < nLocal_; j++) { |
270 | > | int jglob = AtomLocalToGlobal[j]; |
271 | > | |
272 | > | if (excludes->hasPair(iglob, jglob)) |
273 | > | excludesForAtom[i].push_back(j); |
274 | > | |
275 | > | if (oneTwo->hasPair(iglob, jglob)) { |
276 | > | toposForAtom[i].push_back(j); |
277 | > | topoDist[i].push_back(1); |
278 | > | } else { |
279 | > | if (oneThree->hasPair(iglob, jglob)) { |
280 | > | toposForAtom[i].push_back(j); |
281 | > | topoDist[i].push_back(2); |
282 | > | } else { |
283 | > | if (oneFour->hasPair(iglob, jglob)) { |
284 | > | toposForAtom[i].push_back(j); |
285 | > | topoDist[i].push_back(3); |
286 | > | } |
287 | > | } |
288 | > | } |
289 | > | } |
290 | > | } |
291 | #endif | |
292 | ||
293 | < | // First time through, construct column stride. |
294 | < | if (communicators.size() == 0) |
295 | < | { |
296 | < | int nColumnsMax = (int) round(sqrt((float) nProcs)); |
297 | < | for (int i = 0; i < nProcs; ++i) |
298 | < | { |
299 | < | if (nProcs%i==0) nColumns=i; |
293 | > | // allocate memory for the parallel objects |
294 | > | atypesLocal.resize(nLocal_); |
295 | > | |
296 | > | for (int i = 0; i < nLocal_; i++) |
297 | > | atypesLocal[i] = ff_->getAtomType(idents[i]); |
298 | > | |
299 | > | groupList_.clear(); |
300 | > | groupList_.resize(nGroups_); |
301 | > | for (int i = 0; i < nGroups_; i++) { |
302 | > | int gid = cgLocalToGlobal[i]; |
303 | > | for (int j = 0; j < nLocal_; j++) { |
304 | > | int aid = AtomLocalToGlobal[j]; |
305 | > | if (globalGroupMembership[aid] == gid) { |
306 | > | groupList_[i].push_back(j); |
307 | > | } |
308 | > | } |
309 | } | |
310 | ||
311 | < | int nRows = nProcs/nColumns; |
312 | < | myRank_ = (int) worldRank%nColumns; |
311 | > | |
312 | > | createGtypeCutoffMap(); |
313 | > | |
314 | } | |
315 | < | else |
316 | < | { |
317 | < | myRank_ = myRank/nColumns; |
318 | < | } |
319 | < | MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0); |
320 | < | |
123 | < | isColumn_ = false; |
124 | < | |
125 | < | } |
315 | > | |
316 | > | void ForceMatrixDecomposition::createGtypeCutoffMap() { |
317 | > | |
318 | > | GrCut.clear(); |
319 | > | GrCutSq.clear(); |
320 | > | GrlistSq.clear(); |
321 | ||
322 | < | ForceDecomposition::gather(sendbuf, receivebuf){ |
323 | < | communicators(myIndex_).Allgatherv(); |
324 | < | } |
322 | > | RealType tol = 1e-6; |
323 | > | largestRcut_ = 0.0; |
324 | > | int atid; |
325 | > | set<AtomType*> atypes = info_->getSimulatedAtomTypes(); |
326 | > | |
327 | > | map<int, RealType> atypeCutoff; |
328 | > | |
329 | > | for (set<AtomType*>::iterator at = atypes.begin(); |
330 | > | at != atypes.end(); ++at){ |
331 | > | atid = (*at)->getIdent(); |
332 | > | if (userChoseCutoff_) |
333 | > | atypeCutoff[atid] = userCutoff_; |
334 | > | else |
335 | > | atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at); |
336 | > | } |
337 | > | |
338 | > | vector<RealType> gTypeCutoffs; |
339 | > | // first we do a single loop over the cutoff groups to find the |
340 | > | // largest cutoff for any atypes present in this group. |
341 | > | #ifdef IS_MPI |
342 | > | vector<RealType> groupCutoffRow(nGroupsInRow_, 0.0); |
343 | > | groupRowToGtype.resize(nGroupsInRow_); |
344 | > | for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) { |
345 | > | vector<int> atomListRow = getAtomsInGroupRow(cg1); |
346 | > | for (vector<int>::iterator ia = atomListRow.begin(); |
347 | > | ia != atomListRow.end(); ++ia) { |
348 | > | int atom1 = (*ia); |
349 | > | atid = identsRow[atom1]; |
350 | > | if (atypeCutoff[atid] > groupCutoffRow[cg1]) { |
351 | > | groupCutoffRow[cg1] = atypeCutoff[atid]; |
352 | > | } |
353 | > | } |
354 | > | |
355 | > | bool gTypeFound = false; |
356 | > | for (int gt = 0; gt < gTypeCutoffs.size(); gt++) { |
357 | > | if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) { |
358 | > | groupRowToGtype[cg1] = gt; |
359 | > | gTypeFound = true; |
360 | > | } |
361 | > | } |
362 | > | if (!gTypeFound) { |
363 | > | gTypeCutoffs.push_back( groupCutoffRow[cg1] ); |
364 | > | groupRowToGtype[cg1] = gTypeCutoffs.size() - 1; |
365 | > | } |
366 | > | |
367 | > | } |
368 | > | vector<RealType> groupCutoffCol(nGroupsInCol_, 0.0); |
369 | > | groupColToGtype.resize(nGroupsInCol_); |
370 | > | for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) { |
371 | > | vector<int> atomListCol = getAtomsInGroupColumn(cg2); |
372 | > | for (vector<int>::iterator jb = atomListCol.begin(); |
373 | > | jb != atomListCol.end(); ++jb) { |
374 | > | int atom2 = (*jb); |
375 | > | atid = identsCol[atom2]; |
376 | > | if (atypeCutoff[atid] > groupCutoffCol[cg2]) { |
377 | > | groupCutoffCol[cg2] = atypeCutoff[atid]; |
378 | > | } |
379 | > | } |
380 | > | bool gTypeFound = false; |
381 | > | for (int gt = 0; gt < gTypeCutoffs.size(); gt++) { |
382 | > | if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) { |
383 | > | groupColToGtype[cg2] = gt; |
384 | > | gTypeFound = true; |
385 | > | } |
386 | > | } |
387 | > | if (!gTypeFound) { |
388 | > | gTypeCutoffs.push_back( groupCutoffCol[cg2] ); |
389 | > | groupColToGtype[cg2] = gTypeCutoffs.size() - 1; |
390 | > | } |
391 | > | } |
392 | > | #else |
393 | > | |
394 | > | vector<RealType> groupCutoff(nGroups_, 0.0); |
395 | > | groupToGtype.resize(nGroups_); |
396 | > | for (int cg1 = 0; cg1 < nGroups_; cg1++) { |
397 | > | groupCutoff[cg1] = 0.0; |
398 | > | vector<int> atomList = getAtomsInGroupRow(cg1); |
399 | > | for (vector<int>::iterator ia = atomList.begin(); |
400 | > | ia != atomList.end(); ++ia) { |
401 | > | int atom1 = (*ia); |
402 | > | atid = idents[atom1]; |
403 | > | if (atypeCutoff[atid] > groupCutoff[cg1]) |
404 | > | groupCutoff[cg1] = atypeCutoff[atid]; |
405 | > | } |
406 | > | |
407 | > | bool gTypeFound = false; |
408 | > | for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) { |
409 | > | if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) { |
410 | > | groupToGtype[cg1] = gt; |
411 | > | gTypeFound = true; |
412 | > | } |
413 | > | } |
414 | > | if (!gTypeFound) { |
415 | > | gTypeCutoffs.push_back( groupCutoff[cg1] ); |
416 | > | groupToGtype[cg1] = gTypeCutoffs.size() - 1; |
417 | > | } |
418 | > | } |
419 | > | #endif |
420 | ||
421 | + | // Now we find the maximum group cutoff value present in the simulation |
422 | ||
423 | + | RealType groupMax = *max_element(gTypeCutoffs.begin(), |
424 | + | gTypeCutoffs.end()); |
425 | ||
426 | < | ForceDecomposition::scatter(sbuffer, rbuffer){ |
427 | < | communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM); |
428 | < | } |
426 | > | #ifdef IS_MPI |
427 | > | MPI_Allreduce(MPI_IN_PLACE, &groupMax, 1, MPI_REALTYPE, |
428 | > | MPI_MAX, MPI_COMM_WORLD); |
429 | > | #endif |
430 | > | |
431 | > | RealType tradRcut = groupMax; |
432 | > | |
433 | > | GrCut.resize( gTypeCutoffs.size() ); |
434 | > | GrCutSq.resize( gTypeCutoffs.size() ); |
435 | > | GrlistSq.resize( gTypeCutoffs.size() ); |
436 | > | |
437 | > | |
438 | > | for (unsigned int i = 0; i < gTypeCutoffs.size(); i++) { |
439 | > | GrCut[i].resize( gTypeCutoffs.size() , 0.0); |
440 | > | GrCutSq[i].resize( gTypeCutoffs.size(), 0.0 ); |
441 | > | GrlistSq[i].resize( gTypeCutoffs.size(), 0.0 ); |
442 | > | |
443 | > | for (unsigned int j = 0; j < gTypeCutoffs.size(); j++) { |
444 | > | RealType thisRcut; |
445 | > | switch(cutoffPolicy_) { |
446 | > | case TRADITIONAL: |
447 | > | thisRcut = tradRcut; |
448 | > | break; |
449 | > | case MIX: |
450 | > | thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]); |
451 | > | break; |
452 | > | case MAX: |
453 | > | thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]); |
454 | > | break; |
455 | > | default: |
456 | > | sprintf(painCave.errMsg, |
457 | > | "ForceMatrixDecomposition::createGtypeCutoffMap " |
458 | > | "hit an unknown cutoff policy!\n"); |
459 | > | painCave.severity = OPENMD_ERROR; |
460 | > | painCave.isFatal = 1; |
461 | > | simError(); |
462 | > | break; |
463 | > | } |
464 | > | |
465 | > | GrCut[i][j] = thisRcut; |
466 | > | if (thisRcut > largestRcut_) largestRcut_ = thisRcut; |
467 | > | GrCutSq[i][j] = thisRcut * thisRcut; |
468 | > | GrlistSq[i][j] = pow(thisRcut + skinThickness_, 2); |
469 | > | |
470 | > | // pair<int,int> key = make_pair(i,j); |
471 | > | // gTypeCutoffMap[key].first = thisRcut; |
472 | > | // gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2); |
473 | > | // sanity check |
474 | > | |
475 | > | if (userChoseCutoff_) { |
476 | > | if (abs(GrCut[i][j] - userCutoff_) > 0.0001) { |
477 | > | sprintf(painCave.errMsg, |
478 | > | "ForceMatrixDecomposition::createGtypeCutoffMap " |
479 | > | "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_); |
480 | > | painCave.severity = OPENMD_ERROR; |
481 | > | painCave.isFatal = 1; |
482 | > | simError(); |
483 | > | } |
484 | > | } |
485 | > | } |
486 | > | } |
487 | > | } |
488 | > | |
489 | > | void ForceMatrixDecomposition::getGroupCutoffs(int &cg1, int &cg2, RealType &rcut, RealType &rcutsq, RealType &rlistsq) { |
490 | > | int i, j; |
491 | > | #ifdef IS_MPI |
492 | > | i = groupRowToGtype[cg1]; |
493 | > | j = groupColToGtype[cg2]; |
494 | > | #else |
495 | > | i = groupToGtype[cg1]; |
496 | > | j = groupToGtype[cg2]; |
497 | > | #endif |
498 | > | rcut = GrCut[i][j]; |
499 | > | rcutsq = GrCutSq[i][j]; |
500 | > | rlistsq = GrlistSq[i][j]; |
501 | > | return; |
502 | > | //return gTypeCutoffMap[make_pair(i,j)]; |
503 | > | } |
504 | > | |
505 | > | int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) { |
506 | > | for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) { |
507 | > | if (toposForAtom[atom1][j] == atom2) |
508 | > | return topoDist[atom1][j]; |
509 | > | } |
510 | > | return 0; |
511 | > | } |
512 | > | |
513 | > | void ForceMatrixDecomposition::zeroWorkArrays() { |
514 | > | pairwisePot = 0.0; |
515 | > | embeddingPot = 0.0; |
516 | > | excludedPot = 0.0; |
517 | > | excludedSelfPot = 0.0; |
518 | > | |
519 | > | #ifdef IS_MPI |
520 | > | if (storageLayout_ & DataStorage::dslForce) { |
521 | > | fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero); |
522 | > | fill(atomColData.force.begin(), atomColData.force.end(), V3Zero); |
523 | > | } |
524 | > | |
525 | > | if (storageLayout_ & DataStorage::dslTorque) { |
526 | > | fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero); |
527 | > | fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero); |
528 | > | } |
529 | > | |
530 | > | fill(pot_row.begin(), pot_row.end(), |
531 | > | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); |
532 | > | |
533 | > | fill(pot_col.begin(), pot_col.end(), |
534 | > | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); |
535 | > | |
536 | > | fill(expot_row.begin(), expot_row.end(), |
537 | > | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); |
538 | > | |
539 | > | fill(expot_col.begin(), expot_col.end(), |
540 | > | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); |
541 | > | |
542 | > | if (storageLayout_ & DataStorage::dslParticlePot) { |
543 | > | fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), |
544 | > | 0.0); |
545 | > | fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), |
546 | > | 0.0); |
547 | > | } |
548 | > | |
549 | > | if (storageLayout_ & DataStorage::dslDensity) { |
550 | > | fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0); |
551 | > | fill(atomColData.density.begin(), atomColData.density.end(), 0.0); |
552 | > | } |
553 | > | |
554 | > | if (storageLayout_ & DataStorage::dslFunctional) { |
555 | > | fill(atomRowData.functional.begin(), atomRowData.functional.end(), |
556 | > | 0.0); |
557 | > | fill(atomColData.functional.begin(), atomColData.functional.end(), |
558 | > | 0.0); |
559 | > | } |
560 | > | |
561 | > | if (storageLayout_ & DataStorage::dslFunctionalDerivative) { |
562 | > | fill(atomRowData.functionalDerivative.begin(), |
563 | > | atomRowData.functionalDerivative.end(), 0.0); |
564 | > | fill(atomColData.functionalDerivative.begin(), |
565 | > | atomColData.functionalDerivative.end(), 0.0); |
566 | > | } |
567 | > | |
568 | > | if (storageLayout_ & DataStorage::dslSkippedCharge) { |
569 | > | fill(atomRowData.skippedCharge.begin(), |
570 | > | atomRowData.skippedCharge.end(), 0.0); |
571 | > | fill(atomColData.skippedCharge.begin(), |
572 | > | atomColData.skippedCharge.end(), 0.0); |
573 | > | } |
574 | > | |
575 | > | if (storageLayout_ & DataStorage::dslFlucQForce) { |
576 | > | fill(atomRowData.flucQFrc.begin(), |
577 | > | atomRowData.flucQFrc.end(), 0.0); |
578 | > | fill(atomColData.flucQFrc.begin(), |
579 | > | atomColData.flucQFrc.end(), 0.0); |
580 | > | } |
581 | > | |
582 | > | if (storageLayout_ & DataStorage::dslElectricField) { |
583 | > | fill(atomRowData.electricField.begin(), |
584 | > | atomRowData.electricField.end(), V3Zero); |
585 | > | fill(atomColData.electricField.begin(), |
586 | > | atomColData.electricField.end(), V3Zero); |
587 | > | } |
588 | > | |
589 | > | if (storageLayout_ & DataStorage::dslSitePotential) { |
590 | > | fill(atomRowData.sitePotential.begin(), |
591 | > | atomRowData.sitePotential.end(), 0.0); |
592 | > | fill(atomColData.sitePotential.begin(), |
593 | > | atomColData.sitePotential.end(), 0.0); |
594 | > | } |
595 | > | |
596 | > | #endif |
597 | > | // even in parallel, we need to zero out the local arrays: |
598 | > | |
599 | > | if (storageLayout_ & DataStorage::dslParticlePot) { |
600 | > | fill(snap_->atomData.particlePot.begin(), |
601 | > | snap_->atomData.particlePot.end(), 0.0); |
602 | > | } |
603 | > | |
604 | > | if (storageLayout_ & DataStorage::dslDensity) { |
605 | > | fill(snap_->atomData.density.begin(), |
606 | > | snap_->atomData.density.end(), 0.0); |
607 | > | } |
608 | > | |
609 | > | if (storageLayout_ & DataStorage::dslFunctional) { |
610 | > | fill(snap_->atomData.functional.begin(), |
611 | > | snap_->atomData.functional.end(), 0.0); |
612 | > | } |
613 | > | |
614 | > | if (storageLayout_ & DataStorage::dslFunctionalDerivative) { |
615 | > | fill(snap_->atomData.functionalDerivative.begin(), |
616 | > | snap_->atomData.functionalDerivative.end(), 0.0); |
617 | > | } |
618 | > | |
619 | > | if (storageLayout_ & DataStorage::dslSkippedCharge) { |
620 | > | fill(snap_->atomData.skippedCharge.begin(), |
621 | > | snap_->atomData.skippedCharge.end(), 0.0); |
622 | > | } |
623 | > | |
624 | > | if (storageLayout_ & DataStorage::dslElectricField) { |
625 | > | fill(snap_->atomData.electricField.begin(), |
626 | > | snap_->atomData.electricField.end(), V3Zero); |
627 | > | } |
628 | > | if (storageLayout_ & DataStorage::dslSitePotential) { |
629 | > | fill(snap_->atomData.sitePotential.begin(), |
630 | > | snap_->atomData.sitePotential.end(), 0.0); |
631 | > | } |
632 | > | } |
633 | > | |
634 | > | |
635 | > | void ForceMatrixDecomposition::distributeData() { |
636 | > | snap_ = sman_->getCurrentSnapshot(); |
637 | > | storageLayout_ = sman_->getStorageLayout(); |
638 | > | #ifdef IS_MPI |
639 | > | |
640 | > | // gather up the atomic positions |
641 | > | AtomPlanVectorRow->gather(snap_->atomData.position, |
642 | > | atomRowData.position); |
643 | > | AtomPlanVectorColumn->gather(snap_->atomData.position, |
644 | > | atomColData.position); |
645 | > | |
646 | > | // gather up the cutoff group positions |
647 | > | |
648 | > | cgPlanVectorRow->gather(snap_->cgData.position, |
649 | > | cgRowData.position); |
650 | > | |
651 | > | cgPlanVectorColumn->gather(snap_->cgData.position, |
652 | > | cgColData.position); |
653 | > | |
654 | > | |
655 | > | |
656 | > | if (needVelocities_) { |
657 | > | // gather up the atomic velocities |
658 | > | AtomPlanVectorColumn->gather(snap_->atomData.velocity, |
659 | > | atomColData.velocity); |
660 | > | |
661 | > | cgPlanVectorColumn->gather(snap_->cgData.velocity, |
662 | > | cgColData.velocity); |
663 | > | } |
664 | > | |
665 | > | |
666 | > | // if needed, gather the atomic rotation matrices |
667 | > | if (storageLayout_ & DataStorage::dslAmat) { |
668 | > | AtomPlanMatrixRow->gather(snap_->atomData.aMat, |
669 | > | atomRowData.aMat); |
670 | > | AtomPlanMatrixColumn->gather(snap_->atomData.aMat, |
671 | > | atomColData.aMat); |
672 | > | } |
673 | > | |
674 | > | // if needed, gather the atomic eletrostatic information |
675 | > | if (storageLayout_ & DataStorage::dslDipole) { |
676 | > | AtomPlanVectorRow->gather(snap_->atomData.dipole, |
677 | > | atomRowData.dipole); |
678 | > | AtomPlanVectorColumn->gather(snap_->atomData.dipole, |
679 | > | atomColData.dipole); |
680 | > | } |
681 | > | |
682 | > | if (storageLayout_ & DataStorage::dslQuadrupole) { |
683 | > | AtomPlanMatrixRow->gather(snap_->atomData.quadrupole, |
684 | > | atomRowData.quadrupole); |
685 | > | AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole, |
686 | > | atomColData.quadrupole); |
687 | > | } |
688 | > | |
689 | > | // if needed, gather the atomic fluctuating charge values |
690 | > | if (storageLayout_ & DataStorage::dslFlucQPosition) { |
691 | > | AtomPlanRealRow->gather(snap_->atomData.flucQPos, |
692 | > | atomRowData.flucQPos); |
693 | > | AtomPlanRealColumn->gather(snap_->atomData.flucQPos, |
694 | > | atomColData.flucQPos); |
695 | > | } |
696 | > | |
697 | > | #endif |
698 | > | } |
699 | > | |
700 | > | /* collects information obtained during the pre-pair loop onto local |
701 | > | * data structures. |
702 | > | */ |
703 | > | void ForceMatrixDecomposition::collectIntermediateData() { |
704 | > | snap_ = sman_->getCurrentSnapshot(); |
705 | > | storageLayout_ = sman_->getStorageLayout(); |
706 | > | #ifdef IS_MPI |
707 | > | |
708 | > | if (storageLayout_ & DataStorage::dslDensity) { |
709 | > | |
710 | > | AtomPlanRealRow->scatter(atomRowData.density, |
711 | > | snap_->atomData.density); |
712 | > | |
713 | > | int n = snap_->atomData.density.size(); |
714 | > | vector<RealType> rho_tmp(n, 0.0); |
715 | > | AtomPlanRealColumn->scatter(atomColData.density, rho_tmp); |
716 | > | for (int i = 0; i < n; i++) |
717 | > | snap_->atomData.density[i] += rho_tmp[i]; |
718 | > | } |
719 | > | |
720 | > | // this isn't necessary if we don't have polarizable atoms, but |
721 | > | // we'll leave it here for now. |
722 | > | if (storageLayout_ & DataStorage::dslElectricField) { |
723 | > | |
724 | > | AtomPlanVectorRow->scatter(atomRowData.electricField, |
725 | > | snap_->atomData.electricField); |
726 | > | |
727 | > | int n = snap_->atomData.electricField.size(); |
728 | > | vector<Vector3d> field_tmp(n, V3Zero); |
729 | > | AtomPlanVectorColumn->scatter(atomColData.electricField, |
730 | > | field_tmp); |
731 | > | for (int i = 0; i < n; i++) |
732 | > | snap_->atomData.electricField[i] += field_tmp[i]; |
733 | > | } |
734 | > | #endif |
735 | > | } |
736 | > | |
737 | > | /* |
738 | > | * redistributes information obtained during the pre-pair loop out to |
739 | > | * row and column-indexed data structures |
740 | > | */ |
741 | > | void ForceMatrixDecomposition::distributeIntermediateData() { |
742 | > | snap_ = sman_->getCurrentSnapshot(); |
743 | > | storageLayout_ = sman_->getStorageLayout(); |
744 | > | #ifdef IS_MPI |
745 | > | if (storageLayout_ & DataStorage::dslFunctional) { |
746 | > | AtomPlanRealRow->gather(snap_->atomData.functional, |
747 | > | atomRowData.functional); |
748 | > | AtomPlanRealColumn->gather(snap_->atomData.functional, |
749 | > | atomColData.functional); |
750 | > | } |
751 | > | |
752 | > | if (storageLayout_ & DataStorage::dslFunctionalDerivative) { |
753 | > | AtomPlanRealRow->gather(snap_->atomData.functionalDerivative, |
754 | > | atomRowData.functionalDerivative); |
755 | > | AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative, |
756 | > | atomColData.functionalDerivative); |
757 | > | } |
758 | > | #endif |
759 | > | } |
760 | > | |
761 | > | |
762 | > | void ForceMatrixDecomposition::collectData() { |
763 | > | snap_ = sman_->getCurrentSnapshot(); |
764 | > | storageLayout_ = sman_->getStorageLayout(); |
765 | > | #ifdef IS_MPI |
766 | > | int n = snap_->atomData.force.size(); |
767 | > | vector<Vector3d> frc_tmp(n, V3Zero); |
768 | > | |
769 | > | AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp); |
770 | > | for (int i = 0; i < n; i++) { |
771 | > | snap_->atomData.force[i] += frc_tmp[i]; |
772 | > | frc_tmp[i] = 0.0; |
773 | > | } |
774 | > | |
775 | > | AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp); |
776 | > | for (int i = 0; i < n; i++) { |
777 | > | snap_->atomData.force[i] += frc_tmp[i]; |
778 | > | } |
779 | > | |
780 | > | if (storageLayout_ & DataStorage::dslTorque) { |
781 | > | |
782 | > | int nt = snap_->atomData.torque.size(); |
783 | > | vector<Vector3d> trq_tmp(nt, V3Zero); |
784 | > | |
785 | > | AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp); |
786 | > | for (int i = 0; i < nt; i++) { |
787 | > | snap_->atomData.torque[i] += trq_tmp[i]; |
788 | > | trq_tmp[i] = 0.0; |
789 | > | } |
790 | > | |
791 | > | AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp); |
792 | > | for (int i = 0; i < nt; i++) |
793 | > | snap_->atomData.torque[i] += trq_tmp[i]; |
794 | > | } |
795 | > | |
796 | > | if (storageLayout_ & DataStorage::dslSkippedCharge) { |
797 | > | |
798 | > | int ns = snap_->atomData.skippedCharge.size(); |
799 | > | vector<RealType> skch_tmp(ns, 0.0); |
800 | > | |
801 | > | AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp); |
802 | > | for (int i = 0; i < ns; i++) { |
803 | > | snap_->atomData.skippedCharge[i] += skch_tmp[i]; |
804 | > | skch_tmp[i] = 0.0; |
805 | > | } |
806 | > | |
807 | > | AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp); |
808 | > | for (int i = 0; i < ns; i++) |
809 | > | snap_->atomData.skippedCharge[i] += skch_tmp[i]; |
810 | > | |
811 | > | } |
812 | > | |
813 | > | if (storageLayout_ & DataStorage::dslFlucQForce) { |
814 | > | |
815 | > | int nq = snap_->atomData.flucQFrc.size(); |
816 | > | vector<RealType> fqfrc_tmp(nq, 0.0); |
817 | > | |
818 | > | AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp); |
819 | > | for (int i = 0; i < nq; i++) { |
820 | > | snap_->atomData.flucQFrc[i] += fqfrc_tmp[i]; |
821 | > | fqfrc_tmp[i] = 0.0; |
822 | > | } |
823 | > | |
824 | > | AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp); |
825 | > | for (int i = 0; i < nq; i++) |
826 | > | snap_->atomData.flucQFrc[i] += fqfrc_tmp[i]; |
827 | > | |
828 | > | } |
829 | > | |
830 | > | if (storageLayout_ & DataStorage::dslElectricField) { |
831 | > | |
832 | > | int nef = snap_->atomData.electricField.size(); |
833 | > | vector<Vector3d> efield_tmp(nef, V3Zero); |
834 | ||
835 | + | AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp); |
836 | + | for (int i = 0; i < nef; i++) { |
837 | + | snap_->atomData.electricField[i] += efield_tmp[i]; |
838 | + | efield_tmp[i] = 0.0; |
839 | + | } |
840 | + | |
841 | + | AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp); |
842 | + | for (int i = 0; i < nef; i++) |
843 | + | snap_->atomData.electricField[i] += efield_tmp[i]; |
844 | + | } |
845 | ||
846 | + | if (storageLayout_ & DataStorage::dslSitePotential) { |
847 | + | |
848 | + | int nsp = snap_->atomData.sitePotential.size(); |
849 | + | vector<RealType> sp_tmp(nsp, 0.0); |
850 | + | |
851 | + | AtomPlanRealRow->scatter(atomRowData.sitePotential, sp_tmp); |
852 | + | for (int i = 0; i < nsp; i++) { |
853 | + | snap_->atomData.sitePotential[i] += sp_tmp[i]; |
854 | + | sp_tmp[i] = 0.0; |
855 | + | } |
856 | + | |
857 | + | AtomPlanRealColumn->scatter(atomColData.sitePotential, sp_tmp); |
858 | + | for (int i = 0; i < nsp; i++) |
859 | + | snap_->atomData.sitePotential[i] += sp_tmp[i]; |
860 | + | } |
861 | + | |
862 | + | nLocal_ = snap_->getNumberOfAtoms(); |
863 | + | |
864 | + | vector<potVec> pot_temp(nLocal_, |
865 | + | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); |
866 | + | vector<potVec> expot_temp(nLocal_, |
867 | + | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); |
868 | + | |
869 | + | // scatter/gather pot_row into the members of my column |
870 | + | |
871 | + | AtomPlanPotRow->scatter(pot_row, pot_temp); |
872 | + | AtomPlanPotRow->scatter(expot_row, expot_temp); |
873 | + | |
874 | + | for (int ii = 0; ii < pot_temp.size(); ii++ ) |
875 | + | pairwisePot += pot_temp[ii]; |
876 | + | |
877 | + | for (int ii = 0; ii < expot_temp.size(); ii++ ) |
878 | + | excludedPot += expot_temp[ii]; |
879 | + | |
880 | + | if (storageLayout_ & DataStorage::dslParticlePot) { |
881 | + | // This is the pairwise contribution to the particle pot. The |
882 | + | // embedding contribution is added in each of the low level |
883 | + | // non-bonded routines. In single processor, this is done in |
884 | + | // unpackInteractionData, not in collectData. |
885 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { |
886 | + | for (int i = 0; i < nLocal_; i++) { |
887 | + | // factor of two is because the total potential terms are divided |
888 | + | // by 2 in parallel due to row/ column scatter |
889 | + | snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii); |
890 | + | } |
891 | + | } |
892 | + | } |
893 | + | |
894 | + | fill(pot_temp.begin(), pot_temp.end(), |
895 | + | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); |
896 | + | fill(expot_temp.begin(), expot_temp.end(), |
897 | + | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); |
898 | + | |
899 | + | AtomPlanPotColumn->scatter(pot_col, pot_temp); |
900 | + | AtomPlanPotColumn->scatter(expot_col, expot_temp); |
901 | + | |
902 | + | for (int ii = 0; ii < pot_temp.size(); ii++ ) |
903 | + | pairwisePot += pot_temp[ii]; |
904 | + | |
905 | + | for (int ii = 0; ii < expot_temp.size(); ii++ ) |
906 | + | excludedPot += expot_temp[ii]; |
907 | + | |
908 | + | if (storageLayout_ & DataStorage::dslParticlePot) { |
909 | + | // This is the pairwise contribution to the particle pot. The |
910 | + | // embedding contribution is added in each of the low level |
911 | + | // non-bonded routines. In single processor, this is done in |
912 | + | // unpackInteractionData, not in collectData. |
913 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { |
914 | + | for (int i = 0; i < nLocal_; i++) { |
915 | + | // factor of two is because the total potential terms are divided |
916 | + | // by 2 in parallel due to row/ column scatter |
917 | + | snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii); |
918 | + | } |
919 | + | } |
920 | + | } |
921 | + | |
922 | + | if (storageLayout_ & DataStorage::dslParticlePot) { |
923 | + | int npp = snap_->atomData.particlePot.size(); |
924 | + | vector<RealType> ppot_temp(npp, 0.0); |
925 | + | |
926 | + | // This is the direct or embedding contribution to the particle |
927 | + | // pot. |
928 | + | |
929 | + | AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp); |
930 | + | for (int i = 0; i < npp; i++) { |
931 | + | snap_->atomData.particlePot[i] += ppot_temp[i]; |
932 | + | } |
933 | + | |
934 | + | fill(ppot_temp.begin(), ppot_temp.end(), 0.0); |
935 | + | |
936 | + | AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp); |
937 | + | for (int i = 0; i < npp; i++) { |
938 | + | snap_->atomData.particlePot[i] += ppot_temp[i]; |
939 | + | } |
940 | + | } |
941 | + | |
942 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { |
943 | + | RealType ploc1 = pairwisePot[ii]; |
944 | + | RealType ploc2 = 0.0; |
945 | + | MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD); |
946 | + | pairwisePot[ii] = ploc2; |
947 | + | } |
948 | + | |
949 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { |
950 | + | RealType ploc1 = excludedPot[ii]; |
951 | + | RealType ploc2 = 0.0; |
952 | + | MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD); |
953 | + | excludedPot[ii] = ploc2; |
954 | + | } |
955 | + | |
956 | + | // Here be dragons. |
957 | + | MPI_Comm col = colComm.getComm(); |
958 | + | |
959 | + | MPI_Allreduce(MPI_IN_PLACE, |
960 | + | &snap_->frameData.conductiveHeatFlux[0], 3, |
961 | + | MPI_REALTYPE, MPI_SUM, col); |
962 | + | |
963 | + | |
964 | + | #endif |
965 | + | |
966 | + | } |
967 | + | |
968 | + | /** |
969 | + | * Collects information obtained during the post-pair (and embedding |
970 | + | * functional) loops onto local data structures. |
971 | + | */ |
972 | + | void ForceMatrixDecomposition::collectSelfData() { |
973 | + | snap_ = sman_->getCurrentSnapshot(); |
974 | + | storageLayout_ = sman_->getStorageLayout(); |
975 | + | |
976 | + | #ifdef IS_MPI |
977 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { |
978 | + | RealType ploc1 = embeddingPot[ii]; |
979 | + | RealType ploc2 = 0.0; |
980 | + | MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD); |
981 | + | embeddingPot[ii] = ploc2; |
982 | + | } |
983 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { |
984 | + | RealType ploc1 = excludedSelfPot[ii]; |
985 | + | RealType ploc2 = 0.0; |
986 | + | MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD); |
987 | + | excludedSelfPot[ii] = ploc2; |
988 | + | } |
989 | + | #endif |
990 | + | |
991 | + | } |
992 | + | |
993 | + | |
994 | + | |
995 | + | int& ForceMatrixDecomposition::getNAtomsInRow() { |
996 | + | #ifdef IS_MPI |
997 | + | return nAtomsInRow_; |
998 | + | #else |
999 | + | return nLocal_; |
1000 | + | #endif |
1001 | + | } |
1002 | + | |
1003 | + | /** |
1004 | + | * returns the list of atoms belonging to this group. |
1005 | + | */ |
1006 | + | vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){ |
1007 | + | #ifdef IS_MPI |
1008 | + | return groupListRow_[cg1]; |
1009 | + | #else |
1010 | + | return groupList_[cg1]; |
1011 | + | #endif |
1012 | + | } |
1013 | + | |
1014 | + | vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){ |
1015 | + | #ifdef IS_MPI |
1016 | + | return groupListCol_[cg2]; |
1017 | + | #else |
1018 | + | return groupList_[cg2]; |
1019 | + | #endif |
1020 | + | } |
1021 | + | |
1022 | + | Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){ |
1023 | + | Vector3d d; |
1024 | + | |
1025 | + | #ifdef IS_MPI |
1026 | + | d = cgColData.position[cg2] - cgRowData.position[cg1]; |
1027 | + | #else |
1028 | + | d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1]; |
1029 | + | #endif |
1030 | + | |
1031 | + | if (usePeriodicBoundaryConditions_) { |
1032 | + | snap_->wrapVector(d); |
1033 | + | } |
1034 | + | return d; |
1035 | + | } |
1036 | + | |
1037 | + | Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){ |
1038 | + | #ifdef IS_MPI |
1039 | + | return cgColData.velocity[cg2]; |
1040 | + | #else |
1041 | + | return snap_->cgData.velocity[cg2]; |
1042 | + | #endif |
1043 | + | } |
1044 | + | |
1045 | + | Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){ |
1046 | + | #ifdef IS_MPI |
1047 | + | return atomColData.velocity[atom2]; |
1048 | + | #else |
1049 | + | return snap_->atomData.velocity[atom2]; |
1050 | + | #endif |
1051 | + | } |
1052 | + | |
1053 | + | |
1054 | + | Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){ |
1055 | + | |
1056 | + | Vector3d d; |
1057 | + | |
1058 | + | #ifdef IS_MPI |
1059 | + | d = cgRowData.position[cg1] - atomRowData.position[atom1]; |
1060 | + | #else |
1061 | + | d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1]; |
1062 | + | #endif |
1063 | + | if (usePeriodicBoundaryConditions_) { |
1064 | + | snap_->wrapVector(d); |
1065 | + | } |
1066 | + | return d; |
1067 | + | } |
1068 | + | |
1069 | + | Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){ |
1070 | + | Vector3d d; |
1071 | + | |
1072 | + | #ifdef IS_MPI |
1073 | + | d = cgColData.position[cg2] - atomColData.position[atom2]; |
1074 | + | #else |
1075 | + | d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2]; |
1076 | + | #endif |
1077 | + | if (usePeriodicBoundaryConditions_) { |
1078 | + | snap_->wrapVector(d); |
1079 | + | } |
1080 | + | return d; |
1081 | + | } |
1082 | + | |
1083 | + | RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) { |
1084 | + | #ifdef IS_MPI |
1085 | + | return massFactorsRow[atom1]; |
1086 | + | #else |
1087 | + | return massFactors[atom1]; |
1088 | + | #endif |
1089 | + | } |
1090 | + | |
1091 | + | RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) { |
1092 | + | #ifdef IS_MPI |
1093 | + | return massFactorsCol[atom2]; |
1094 | + | #else |
1095 | + | return massFactors[atom2]; |
1096 | + | #endif |
1097 | + | |
1098 | + | } |
1099 | + | |
1100 | + | Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){ |
1101 | + | Vector3d d; |
1102 | + | |
1103 | + | #ifdef IS_MPI |
1104 | + | d = atomColData.position[atom2] - atomRowData.position[atom1]; |
1105 | + | #else |
1106 | + | d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1]; |
1107 | + | #endif |
1108 | + | if (usePeriodicBoundaryConditions_) { |
1109 | + | snap_->wrapVector(d); |
1110 | + | } |
1111 | + | return d; |
1112 | + | } |
1113 | + | |
1114 | + | vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) { |
1115 | + | return excludesForAtom[atom1]; |
1116 | + | } |
1117 | + | |
1118 | + | /** |
1119 | + | * We need to exclude some overcounted interactions that result from |
1120 | + | * the parallel decomposition. |
1121 | + | */ |
1122 | + | bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) { |
1123 | + | int unique_id_1, unique_id_2; |
1124 | + | |
1125 | + | #ifdef IS_MPI |
1126 | + | // in MPI, we have to look up the unique IDs for each atom |
1127 | + | unique_id_1 = AtomRowToGlobal[atom1]; |
1128 | + | unique_id_2 = AtomColToGlobal[atom2]; |
1129 | + | // group1 = cgRowToGlobal[cg1]; |
1130 | + | // group2 = cgColToGlobal[cg2]; |
1131 | + | #else |
1132 | + | unique_id_1 = AtomLocalToGlobal[atom1]; |
1133 | + | unique_id_2 = AtomLocalToGlobal[atom2]; |
1134 | + | int group1 = cgLocalToGlobal[cg1]; |
1135 | + | int group2 = cgLocalToGlobal[cg2]; |
1136 | + | #endif |
1137 | + | |
1138 | + | if (unique_id_1 == unique_id_2) return true; |
1139 | + | |
1140 | + | #ifdef IS_MPI |
1141 | + | // this prevents us from doing the pair on multiple processors |
1142 | + | if (unique_id_1 < unique_id_2) { |
1143 | + | if ((unique_id_1 + unique_id_2) % 2 == 0) return true; |
1144 | + | } else { |
1145 | + | if ((unique_id_1 + unique_id_2) % 2 == 1) return true; |
1146 | + | } |
1147 | + | #endif |
1148 | + | |
1149 | + | #ifndef IS_MPI |
1150 | + | if (group1 == group2) { |
1151 | + | if (unique_id_1 < unique_id_2) return true; |
1152 | + | } |
1153 | + | #endif |
1154 | + | |
1155 | + | return false; |
1156 | + | } |
1157 | + | |
1158 | + | /** |
1159 | + | * We need to handle the interactions for atoms who are involved in |
1160 | + | * the same rigid body as well as some short range interactions |
1161 | + | * (bonds, bends, torsions) differently from other interactions. |
1162 | + | * We'll still visit the pairwise routines, but with a flag that |
1163 | + | * tells those routines to exclude the pair from direct long range |
1164 | + | * interactions. Some indirect interactions (notably reaction |
1165 | + | * field) must still be handled for these pairs. |
1166 | + | */ |
1167 | + | bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) { |
1168 | + | |
1169 | + | // excludesForAtom was constructed to use row/column indices in the MPI |
1170 | + | // version, and to use local IDs in the non-MPI version: |
1171 | + | |
1172 | + | for (vector<int>::iterator i = excludesForAtom[atom1].begin(); |
1173 | + | i != excludesForAtom[atom1].end(); ++i) { |
1174 | + | if ( (*i) == atom2 ) return true; |
1175 | + | } |
1176 | + | |
1177 | + | return false; |
1178 | + | } |
1179 | + | |
1180 | + | |
1181 | + | void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){ |
1182 | + | #ifdef IS_MPI |
1183 | + | atomRowData.force[atom1] += fg; |
1184 | + | #else |
1185 | + | snap_->atomData.force[atom1] += fg; |
1186 | + | #endif |
1187 | + | } |
1188 | + | |
1189 | + | void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){ |
1190 | + | #ifdef IS_MPI |
1191 | + | atomColData.force[atom2] += fg; |
1192 | + | #else |
1193 | + | snap_->atomData.force[atom2] += fg; |
1194 | + | #endif |
1195 | + | } |
1196 | + | |
1197 | + | // filling interaction blocks with pointers |
1198 | + | void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat, |
1199 | + | int atom1, int atom2) { |
1200 | + | |
1201 | + | idat.excluded = excludeAtomPair(atom1, atom2); |
1202 | + | |
1203 | + | #ifdef IS_MPI |
1204 | + | //idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]); |
1205 | + | idat.atid1 = identsRow[atom1]; |
1206 | + | idat.atid2 = identsCol[atom2]; |
1207 | + | |
1208 | + | if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) { |
1209 | + | idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]); |
1210 | + | } else { |
1211 | + | idat.sameRegion = false; |
1212 | + | } |
1213 | + | |
1214 | + | if (storageLayout_ & DataStorage::dslAmat) { |
1215 | + | idat.A1 = &(atomRowData.aMat[atom1]); |
1216 | + | idat.A2 = &(atomColData.aMat[atom2]); |
1217 | + | } |
1218 | + | |
1219 | + | if (storageLayout_ & DataStorage::dslTorque) { |
1220 | + | idat.t1 = &(atomRowData.torque[atom1]); |
1221 | + | idat.t2 = &(atomColData.torque[atom2]); |
1222 | + | } |
1223 | + | |
1224 | + | if (storageLayout_ & DataStorage::dslDipole) { |
1225 | + | idat.dipole1 = &(atomRowData.dipole[atom1]); |
1226 | + | idat.dipole2 = &(atomColData.dipole[atom2]); |
1227 | + | } |
1228 | + | |
1229 | + | if (storageLayout_ & DataStorage::dslQuadrupole) { |
1230 | + | idat.quadrupole1 = &(atomRowData.quadrupole[atom1]); |
1231 | + | idat.quadrupole2 = &(atomColData.quadrupole[atom2]); |
1232 | + | } |
1233 | + | |
1234 | + | if (storageLayout_ & DataStorage::dslDensity) { |
1235 | + | idat.rho1 = &(atomRowData.density[atom1]); |
1236 | + | idat.rho2 = &(atomColData.density[atom2]); |
1237 | + | } |
1238 | + | |
1239 | + | if (storageLayout_ & DataStorage::dslFunctional) { |
1240 | + | idat.frho1 = &(atomRowData.functional[atom1]); |
1241 | + | idat.frho2 = &(atomColData.functional[atom2]); |
1242 | + | } |
1243 | + | |
1244 | + | if (storageLayout_ & DataStorage::dslFunctionalDerivative) { |
1245 | + | idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]); |
1246 | + | idat.dfrho2 = &(atomColData.functionalDerivative[atom2]); |
1247 | + | } |
1248 | + | |
1249 | + | if (storageLayout_ & DataStorage::dslParticlePot) { |
1250 | + | idat.particlePot1 = &(atomRowData.particlePot[atom1]); |
1251 | + | idat.particlePot2 = &(atomColData.particlePot[atom2]); |
1252 | + | } |
1253 | + | |
1254 | + | if (storageLayout_ & DataStorage::dslSkippedCharge) { |
1255 | + | idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]); |
1256 | + | idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]); |
1257 | + | } |
1258 | + | |
1259 | + | if (storageLayout_ & DataStorage::dslFlucQPosition) { |
1260 | + | idat.flucQ1 = &(atomRowData.flucQPos[atom1]); |
1261 | + | idat.flucQ2 = &(atomColData.flucQPos[atom2]); |
1262 | + | } |
1263 | + | |
1264 | + | #else |
1265 | + | |
1266 | + | //idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]); |
1267 | + | idat.atid1 = idents[atom1]; |
1268 | + | idat.atid2 = idents[atom2]; |
1269 | + | |
1270 | + | if (regions[atom1] >= 0 && regions[atom2] >= 0) { |
1271 | + | idat.sameRegion = (regions[atom1] == regions[atom2]); |
1272 | + | } else { |
1273 | + | idat.sameRegion = false; |
1274 | + | } |
1275 | + | |
1276 | + | if (storageLayout_ & DataStorage::dslAmat) { |
1277 | + | idat.A1 = &(snap_->atomData.aMat[atom1]); |
1278 | + | idat.A2 = &(snap_->atomData.aMat[atom2]); |
1279 | + | } |
1280 | + | |
1281 | + | if (storageLayout_ & DataStorage::dslTorque) { |
1282 | + | idat.t1 = &(snap_->atomData.torque[atom1]); |
1283 | + | idat.t2 = &(snap_->atomData.torque[atom2]); |
1284 | + | } |
1285 | + | |
1286 | + | if (storageLayout_ & DataStorage::dslDipole) { |
1287 | + | idat.dipole1 = &(snap_->atomData.dipole[atom1]); |
1288 | + | idat.dipole2 = &(snap_->atomData.dipole[atom2]); |
1289 | + | } |
1290 | + | |
1291 | + | if (storageLayout_ & DataStorage::dslQuadrupole) { |
1292 | + | idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]); |
1293 | + | idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]); |
1294 | + | } |
1295 | + | |
1296 | + | if (storageLayout_ & DataStorage::dslDensity) { |
1297 | + | idat.rho1 = &(snap_->atomData.density[atom1]); |
1298 | + | idat.rho2 = &(snap_->atomData.density[atom2]); |
1299 | + | } |
1300 | + | |
1301 | + | if (storageLayout_ & DataStorage::dslFunctional) { |
1302 | + | idat.frho1 = &(snap_->atomData.functional[atom1]); |
1303 | + | idat.frho2 = &(snap_->atomData.functional[atom2]); |
1304 | + | } |
1305 | + | |
1306 | + | if (storageLayout_ & DataStorage::dslFunctionalDerivative) { |
1307 | + | idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]); |
1308 | + | idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]); |
1309 | + | } |
1310 | + | |
1311 | + | if (storageLayout_ & DataStorage::dslParticlePot) { |
1312 | + | idat.particlePot1 = &(snap_->atomData.particlePot[atom1]); |
1313 | + | idat.particlePot2 = &(snap_->atomData.particlePot[atom2]); |
1314 | + | } |
1315 | + | |
1316 | + | if (storageLayout_ & DataStorage::dslSkippedCharge) { |
1317 | + | idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]); |
1318 | + | idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]); |
1319 | + | } |
1320 | + | |
1321 | + | if (storageLayout_ & DataStorage::dslFlucQPosition) { |
1322 | + | idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]); |
1323 | + | idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]); |
1324 | + | } |
1325 | + | |
1326 | + | #endif |
1327 | + | } |
1328 | + | |
1329 | + | |
1330 | + | void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) { |
1331 | + | #ifdef IS_MPI |
1332 | + | pot_row[atom1] += RealType(0.5) * *(idat.pot); |
1333 | + | pot_col[atom2] += RealType(0.5) * *(idat.pot); |
1334 | + | expot_row[atom1] += RealType(0.5) * *(idat.excludedPot); |
1335 | + | expot_col[atom2] += RealType(0.5) * *(idat.excludedPot); |
1336 | + | |
1337 | + | atomRowData.force[atom1] += *(idat.f1); |
1338 | + | atomColData.force[atom2] -= *(idat.f1); |
1339 | + | |
1340 | + | if (storageLayout_ & DataStorage::dslFlucQForce) { |
1341 | + | atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1); |
1342 | + | atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2); |
1343 | + | } |
1344 | + | |
1345 | + | if (storageLayout_ & DataStorage::dslElectricField) { |
1346 | + | atomRowData.electricField[atom1] += *(idat.eField1); |
1347 | + | atomColData.electricField[atom2] += *(idat.eField2); |
1348 | + | } |
1349 | + | |
1350 | + | if (storageLayout_ & DataStorage::dslSitePotential) { |
1351 | + | atomRowData.sitePotential[atom1] += *(idat.sPot1); |
1352 | + | atomColData.sitePotential[atom2] += *(idat.sPot2); |
1353 | + | } |
1354 | + | |
1355 | + | #else |
1356 | + | pairwisePot += *(idat.pot); |
1357 | + | excludedPot += *(idat.excludedPot); |
1358 | + | |
1359 | + | snap_->atomData.force[atom1] += *(idat.f1); |
1360 | + | snap_->atomData.force[atom2] -= *(idat.f1); |
1361 | + | |
1362 | + | if (idat.doParticlePot) { |
1363 | + | // This is the pairwise contribution to the particle pot. The |
1364 | + | // embedding contribution is added in each of the low level |
1365 | + | // non-bonded routines. In parallel, this calculation is done |
1366 | + | // in collectData, not in unpackInteractionData. |
1367 | + | snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw); |
1368 | + | snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw); |
1369 | + | } |
1370 | + | |
1371 | + | if (storageLayout_ & DataStorage::dslFlucQForce) { |
1372 | + | snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1); |
1373 | + | snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2); |
1374 | + | } |
1375 | + | |
1376 | + | if (storageLayout_ & DataStorage::dslElectricField) { |
1377 | + | snap_->atomData.electricField[atom1] += *(idat.eField1); |
1378 | + | snap_->atomData.electricField[atom2] += *(idat.eField2); |
1379 | + | } |
1380 | + | |
1381 | + | if (storageLayout_ & DataStorage::dslSitePotential) { |
1382 | + | snap_->atomData.sitePotential[atom1] += *(idat.sPot1); |
1383 | + | snap_->atomData.sitePotential[atom2] += *(idat.sPot2); |
1384 | + | } |
1385 | + | |
1386 | + | #endif |
1387 | + | |
1388 | + | } |
1389 | + | |
1390 | + | /* |
1391 | + | * buildNeighborList |
1392 | + | * |
1393 | + | * first element of pair is row-indexed CutoffGroup |
1394 | + | * second element of pair is column-indexed CutoffGroup |
1395 | + | */ |
1396 | + | void ForceMatrixDecomposition::buildNeighborList(vector<pair<int,int> >& neighborList) { |
1397 | + | |
1398 | + | neighborList.clear(); |
1399 | + | groupCutoffs cuts; |
1400 | + | bool doAllPairs = false; |
1401 | + | |
1402 | + | RealType rList_ = (largestRcut_ + skinThickness_); |
1403 | + | RealType rcut, rcutsq, rlistsq; |
1404 | + | Snapshot* snap_ = sman_->getCurrentSnapshot(); |
1405 | + | Mat3x3d box; |
1406 | + | Mat3x3d invBox; |
1407 | + | |
1408 | + | Vector3d rs, scaled, dr; |
1409 | + | Vector3i whichCell; |
1410 | + | int cellIndex; |
1411 | + | |
1412 | + | #ifdef IS_MPI |
1413 | + | cellListRow_.clear(); |
1414 | + | cellListCol_.clear(); |
1415 | + | #else |
1416 | + | cellList_.clear(); |
1417 | + | #endif |
1418 | + | |
1419 | + | if (!usePeriodicBoundaryConditions_) { |
1420 | + | box = snap_->getBoundingBox(); |
1421 | + | invBox = snap_->getInvBoundingBox(); |
1422 | + | } else { |
1423 | + | box = snap_->getHmat(); |
1424 | + | invBox = snap_->getInvHmat(); |
1425 | + | } |
1426 | + | |
1427 | + | Vector3d boxX = box.getColumn(0); |
1428 | + | Vector3d boxY = box.getColumn(1); |
1429 | + | Vector3d boxZ = box.getColumn(2); |
1430 | + | |
1431 | + | nCells_.x() = int( boxX.length() / rList_ ); |
1432 | + | nCells_.y() = int( boxY.length() / rList_ ); |
1433 | + | nCells_.z() = int( boxZ.length() / rList_ ); |
1434 | + | |
1435 | + | // handle small boxes where the cell offsets can end up repeating cells |
1436 | + | |
1437 | + | if (nCells_.x() < 3) doAllPairs = true; |
1438 | + | if (nCells_.y() < 3) doAllPairs = true; |
1439 | + | if (nCells_.z() < 3) doAllPairs = true; |
1440 | + | |
1441 | + | int nCtot = nCells_.x() * nCells_.y() * nCells_.z(); |
1442 | + | |
1443 | + | #ifdef IS_MPI |
1444 | + | cellListRow_.resize(nCtot); |
1445 | + | cellListCol_.resize(nCtot); |
1446 | + | #else |
1447 | + | cellList_.resize(nCtot); |
1448 | + | #endif |
1449 | + | |
1450 | + | if (!doAllPairs) { |
1451 | + | #ifdef IS_MPI |
1452 | + | |
1453 | + | for (int i = 0; i < nGroupsInRow_; i++) { |
1454 | + | rs = cgRowData.position[i]; |
1455 | + | |
1456 | + | // scaled positions relative to the box vectors |
1457 | + | scaled = invBox * rs; |
1458 | + | |
1459 | + | // wrap the vector back into the unit box by subtracting integer box |
1460 | + | // numbers |
1461 | + | for (int j = 0; j < 3; j++) { |
1462 | + | scaled[j] -= roundMe(scaled[j]); |
1463 | + | scaled[j] += 0.5; |
1464 | + | // Handle the special case when an object is exactly on the |
1465 | + | // boundary (a scaled coordinate of 1.0 is the same as |
1466 | + | // scaled coordinate of 0.0) |
1467 | + | if (scaled[j] >= 1.0) scaled[j] -= 1.0; |
1468 | + | } |
1469 | + | |
1470 | + | // find xyz-indices of cell that cutoffGroup is in. |
1471 | + | whichCell.x() = nCells_.x() * scaled.x(); |
1472 | + | whichCell.y() = nCells_.y() * scaled.y(); |
1473 | + | whichCell.z() = nCells_.z() * scaled.z(); |
1474 | + | |
1475 | + | // find single index of this cell: |
1476 | + | cellIndex = Vlinear(whichCell, nCells_); |
1477 | + | |
1478 | + | // add this cutoff group to the list of groups in this cell; |
1479 | + | cellListRow_[cellIndex].push_back(i); |
1480 | + | } |
1481 | + | for (int i = 0; i < nGroupsInCol_; i++) { |
1482 | + | rs = cgColData.position[i]; |
1483 | + | |
1484 | + | // scaled positions relative to the box vectors |
1485 | + | scaled = invBox * rs; |
1486 | + | |
1487 | + | // wrap the vector back into the unit box by subtracting integer box |
1488 | + | // numbers |
1489 | + | for (int j = 0; j < 3; j++) { |
1490 | + | scaled[j] -= roundMe(scaled[j]); |
1491 | + | scaled[j] += 0.5; |
1492 | + | // Handle the special case when an object is exactly on the |
1493 | + | // boundary (a scaled coordinate of 1.0 is the same as |
1494 | + | // scaled coordinate of 0.0) |
1495 | + | if (scaled[j] >= 1.0) scaled[j] -= 1.0; |
1496 | + | } |
1497 | + | |
1498 | + | // find xyz-indices of cell that cutoffGroup is in. |
1499 | + | whichCell.x() = nCells_.x() * scaled.x(); |
1500 | + | whichCell.y() = nCells_.y() * scaled.y(); |
1501 | + | whichCell.z() = nCells_.z() * scaled.z(); |
1502 | + | |
1503 | + | // find single index of this cell: |
1504 | + | cellIndex = Vlinear(whichCell, nCells_); |
1505 | + | |
1506 | + | // add this cutoff group to the list of groups in this cell; |
1507 | + | cellListCol_[cellIndex].push_back(i); |
1508 | + | } |
1509 | + | |
1510 | + | #else |
1511 | + | for (int i = 0; i < nGroups_; i++) { |
1512 | + | rs = snap_->cgData.position[i]; |
1513 | + | |
1514 | + | // scaled positions relative to the box vectors |
1515 | + | scaled = invBox * rs; |
1516 | + | |
1517 | + | // wrap the vector back into the unit box by subtracting integer box |
1518 | + | // numbers |
1519 | + | for (int j = 0; j < 3; j++) { |
1520 | + | scaled[j] -= roundMe(scaled[j]); |
1521 | + | scaled[j] += 0.5; |
1522 | + | // Handle the special case when an object is exactly on the |
1523 | + | // boundary (a scaled coordinate of 1.0 is the same as |
1524 | + | // scaled coordinate of 0.0) |
1525 | + | if (scaled[j] >= 1.0) scaled[j] -= 1.0; |
1526 | + | } |
1527 | + | |
1528 | + | // find xyz-indices of cell that cutoffGroup is in. |
1529 | + | whichCell.x() = int(nCells_.x() * scaled.x()); |
1530 | + | whichCell.y() = int(nCells_.y() * scaled.y()); |
1531 | + | whichCell.z() = int(nCells_.z() * scaled.z()); |
1532 | + | |
1533 | + | // find single index of this cell: |
1534 | + | cellIndex = Vlinear(whichCell, nCells_); |
1535 | + | |
1536 | + | // add this cutoff group to the list of groups in this cell; |
1537 | + | cellList_[cellIndex].push_back(i); |
1538 | + | } |
1539 | + | |
1540 | + | #endif |
1541 | + | |
1542 | + | for (int m1z = 0; m1z < nCells_.z(); m1z++) { |
1543 | + | for (int m1y = 0; m1y < nCells_.y(); m1y++) { |
1544 | + | for (int m1x = 0; m1x < nCells_.x(); m1x++) { |
1545 | + | Vector3i m1v(m1x, m1y, m1z); |
1546 | + | int m1 = Vlinear(m1v, nCells_); |
1547 | + | |
1548 | + | for (vector<Vector3i>::iterator os = cellOffsets_.begin(); |
1549 | + | os != cellOffsets_.end(); ++os) { |
1550 | + | |
1551 | + | Vector3i m2v = m1v + (*os); |
1552 | + | |
1553 | + | |
1554 | + | if (m2v.x() >= nCells_.x()) { |
1555 | + | m2v.x() = 0; |
1556 | + | } else if (m2v.x() < 0) { |
1557 | + | m2v.x() = nCells_.x() - 1; |
1558 | + | } |
1559 | + | |
1560 | + | if (m2v.y() >= nCells_.y()) { |
1561 | + | m2v.y() = 0; |
1562 | + | } else if (m2v.y() < 0) { |
1563 | + | m2v.y() = nCells_.y() - 1; |
1564 | + | } |
1565 | + | |
1566 | + | if (m2v.z() >= nCells_.z()) { |
1567 | + | m2v.z() = 0; |
1568 | + | } else if (m2v.z() < 0) { |
1569 | + | m2v.z() = nCells_.z() - 1; |
1570 | + | } |
1571 | + | |
1572 | + | int m2 = Vlinear (m2v, nCells_); |
1573 | + | |
1574 | + | #ifdef IS_MPI |
1575 | + | for (vector<int>::iterator j1 = cellListRow_[m1].begin(); |
1576 | + | j1 != cellListRow_[m1].end(); ++j1) { |
1577 | + | for (vector<int>::iterator j2 = cellListCol_[m2].begin(); |
1578 | + | j2 != cellListCol_[m2].end(); ++j2) { |
1579 | + | |
1580 | + | // In parallel, we need to visit *all* pairs of row |
1581 | + | // & column indicies and will divide labor in the |
1582 | + | // force evaluation later. |
1583 | + | dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)]; |
1584 | + | if (usePeriodicBoundaryConditions_) { |
1585 | + | snap_->wrapVector(dr); |
1586 | + | } |
1587 | + | getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq ); |
1588 | + | if (dr.lengthSquare() < rlistsq) { |
1589 | + | neighborList.push_back(make_pair((*j1), (*j2))); |
1590 | + | } |
1591 | + | } |
1592 | + | } |
1593 | + | #else |
1594 | + | for (vector<int>::iterator j1 = cellList_[m1].begin(); |
1595 | + | j1 != cellList_[m1].end(); ++j1) { |
1596 | + | for (vector<int>::iterator j2 = cellList_[m2].begin(); |
1597 | + | j2 != cellList_[m2].end(); ++j2) { |
1598 | + | |
1599 | + | // Always do this if we're in different cells or if |
1600 | + | // we're in the same cell and the global index of |
1601 | + | // the j2 cutoff group is greater than or equal to |
1602 | + | // the j1 cutoff group. Note that Rappaport's code |
1603 | + | // has a "less than" conditional here, but that |
1604 | + | // deals with atom-by-atom computation. OpenMD |
1605 | + | // allows atoms within a single cutoff group to |
1606 | + | // interact with each other. |
1607 | + | |
1608 | + | if (m2 != m1 || (*j2) >= (*j1) ) { |
1609 | + | |
1610 | + | dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)]; |
1611 | + | if (usePeriodicBoundaryConditions_) { |
1612 | + | snap_->wrapVector(dr); |
1613 | + | } |
1614 | + | getGroupCutoffs( (*j1), (*j2), rcut, rcutsq, rlistsq ); |
1615 | + | if (dr.lengthSquare() < rlistsq) { |
1616 | + | neighborList.push_back(make_pair((*j1), (*j2))); |
1617 | + | } |
1618 | + | } |
1619 | + | } |
1620 | + | } |
1621 | + | #endif |
1622 | + | } |
1623 | + | } |
1624 | + | } |
1625 | + | } |
1626 | + | } else { |
1627 | + | // branch to do all cutoff group pairs |
1628 | + | #ifdef IS_MPI |
1629 | + | for (int j1 = 0; j1 < nGroupsInRow_; j1++) { |
1630 | + | for (int j2 = 0; j2 < nGroupsInCol_; j2++) { |
1631 | + | dr = cgColData.position[j2] - cgRowData.position[j1]; |
1632 | + | if (usePeriodicBoundaryConditions_) { |
1633 | + | snap_->wrapVector(dr); |
1634 | + | } |
1635 | + | getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq); |
1636 | + | if (dr.lengthSquare() < rlistsq) { |
1637 | + | neighborList.push_back(make_pair(j1, j2)); |
1638 | + | } |
1639 | + | } |
1640 | + | } |
1641 | + | #else |
1642 | + | // include all groups here. |
1643 | + | for (int j1 = 0; j1 < nGroups_; j1++) { |
1644 | + | // include self group interactions j2 == j1 |
1645 | + | for (int j2 = j1; j2 < nGroups_; j2++) { |
1646 | + | dr = snap_->cgData.position[j2] - snap_->cgData.position[j1]; |
1647 | + | if (usePeriodicBoundaryConditions_) { |
1648 | + | snap_->wrapVector(dr); |
1649 | + | } |
1650 | + | getGroupCutoffs( j1, j2, rcut, rcutsq, rlistsq ); |
1651 | + | if (dr.lengthSquare() < rlistsq) { |
1652 | + | neighborList.push_back(make_pair(j1, j2)); |
1653 | + | } |
1654 | + | } |
1655 | + | } |
1656 | + | #endif |
1657 | + | } |
1658 | + | |
1659 | + | // save the local cutoff group positions for the check that is |
1660 | + | // done on each loop: |
1661 | + | saved_CG_positions_.clear(); |
1662 | + | for (int i = 0; i < nGroups_; i++) |
1663 | + | saved_CG_positions_.push_back(snap_->cgData.position[i]); |
1664 | + | } |
1665 | + | } //end namespace OpenMD |
– | Removed lines |
+ | Added lines |
< | Changed lines |
> | Changed lines |