ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/OpenMD/trunk/src/parallel/ForceMatrixDecomposition.cpp
Revision: 2057
Committed: Tue Mar 3 15:22:26 2015 UTC (10 years, 1 month ago) by gezelter
File size: 54573 byte(s)
Log Message:
Performance improvements, Removed CutoffPolicy

File Contents

# Content
1 /*
2 * Copyright (c) 2005 The University of Notre Dame. All Rights Reserved.
3 *
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
7 * that the following conditions are met:
8 *
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 *
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the
15 * distribution.
16 *
17 * This software is provided "AS IS," without a warranty of any
18 * kind. All express or implied conditions, representations and
19 * warranties, including any implied warranty of merchantability,
20 * fitness for a particular purpose or non-infringement, are hereby
21 * excluded. The University of Notre Dame and its licensors shall not
22 * be liable for any damages suffered by licensee as a result of
23 * using, modifying or distributing the software or its
24 * derivatives. In no event will the University of Notre Dame or its
25 * licensors be liable for any lost revenue, profit or data, or for
26 * direct, indirect, special, consequential, incidental or punitive
27 * damages, however caused and regardless of the theory of liability,
28 * arising out of the use of or inability to use software, even if the
29 * University of Notre Dame has been advised of the possibility of
30 * such damages.
31 *
32 * SUPPORT OPEN SCIENCE! If you use OpenMD or its source code in your
33 * research, please cite the appropriate papers when you publish your
34 * work. Good starting points are:
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, 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 // Row and colum scans must visit all surrounding cells
54 cellOffsets_.clear();
55 cellOffsets_.push_back( Vector3i(-1,-1,-1) );
56 cellOffsets_.push_back( Vector3i( 0,-1,-1) );
57 cellOffsets_.push_back( Vector3i( 1,-1,-1) );
58 cellOffsets_.push_back( Vector3i(-1, 0,-1) );
59 cellOffsets_.push_back( Vector3i( 0, 0,-1) );
60 cellOffsets_.push_back( Vector3i( 1, 0,-1) );
61 cellOffsets_.push_back( Vector3i(-1, 1,-1) );
62 cellOffsets_.push_back( Vector3i( 0, 1,-1) );
63 cellOffsets_.push_back( Vector3i( 1, 1,-1) );
64 cellOffsets_.push_back( Vector3i(-1,-1, 0) );
65 cellOffsets_.push_back( Vector3i( 0,-1, 0) );
66 cellOffsets_.push_back( Vector3i( 1,-1, 0) );
67 cellOffsets_.push_back( Vector3i(-1, 0, 0) );
68 cellOffsets_.push_back( Vector3i( 0, 0, 0) );
69 cellOffsets_.push_back( Vector3i( 1, 0, 0) );
70 cellOffsets_.push_back( Vector3i(-1, 1, 0) );
71 cellOffsets_.push_back( Vector3i( 0, 1, 0) );
72 cellOffsets_.push_back( Vector3i( 1, 1, 0) );
73 cellOffsets_.push_back( Vector3i(-1,-1, 1) );
74 cellOffsets_.push_back( Vector3i( 0,-1, 1) );
75 cellOffsets_.push_back( Vector3i( 1,-1, 1) );
76 cellOffsets_.push_back( Vector3i(-1, 0, 1) );
77 cellOffsets_.push_back( Vector3i( 0, 0, 1) );
78 cellOffsets_.push_back( Vector3i( 1, 0, 1) );
79 cellOffsets_.push_back( Vector3i(-1, 1, 1) );
80 cellOffsets_.push_back( Vector3i( 0, 1, 1) );
81 cellOffsets_.push_back( Vector3i( 1, 1, 1) );
82 }
83
84
85 /**
86 * distributeInitialData is essentially a copy of the older fortran
87 * SimulationSetup
88 */
89 void ForceMatrixDecomposition::distributeInitialData() {
90 snap_ = sman_->getCurrentSnapshot();
91 storageLayout_ = sman_->getStorageLayout();
92 ff_ = info_->getForceField();
93 nLocal_ = snap_->getNumberOfAtoms();
94
95 nGroups_ = info_->getNLocalCutoffGroups();
96 // gather the information for atomtype IDs (atids):
97 idents = info_->getIdentArray();
98 regions = info_->getRegions();
99 AtomLocalToGlobal = info_->getGlobalAtomIndices();
100 cgLocalToGlobal = info_->getGlobalGroupIndices();
101 vector<int> globalGroupMembership = info_->getGlobalGroupMembership();
102
103 massFactors = info_->getMassFactors();
104
105 PairList* excludes = info_->getExcludedInteractions();
106 PairList* oneTwo = info_->getOneTwoInteractions();
107 PairList* oneThree = info_->getOneThreeInteractions();
108 PairList* oneFour = info_->getOneFourInteractions();
109
110 if (needVelocities_)
111 snap_->cgData.setStorageLayout(DataStorage::dslPosition |
112 DataStorage::dslVelocity);
113 else
114 snap_->cgData.setStorageLayout(DataStorage::dslPosition);
115
116 #ifdef IS_MPI
117
118 MPI_Comm row = rowComm.getComm();
119 MPI_Comm col = colComm.getComm();
120
121 AtomPlanIntRow = new Plan<int>(row, nLocal_);
122 AtomPlanRealRow = new Plan<RealType>(row, nLocal_);
123 AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_);
124 AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_);
125 AtomPlanPotRow = new Plan<potVec>(row, nLocal_);
126
127 AtomPlanIntColumn = new Plan<int>(col, nLocal_);
128 AtomPlanRealColumn = new Plan<RealType>(col, nLocal_);
129 AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_);
130 AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_);
131 AtomPlanPotColumn = new Plan<potVec>(col, nLocal_);
132
133 cgPlanIntRow = new Plan<int>(row, nGroups_);
134 cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_);
135 cgPlanIntColumn = new Plan<int>(col, nGroups_);
136 cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_);
137
138 nAtomsInRow_ = AtomPlanIntRow->getSize();
139 nAtomsInCol_ = AtomPlanIntColumn->getSize();
140 nGroupsInRow_ = cgPlanIntRow->getSize();
141 nGroupsInCol_ = cgPlanIntColumn->getSize();
142
143 // Modify the data storage objects with the correct layouts and sizes:
144 atomRowData.resize(nAtomsInRow_);
145 atomRowData.setStorageLayout(storageLayout_);
146 atomColData.resize(nAtomsInCol_);
147 atomColData.setStorageLayout(storageLayout_);
148 cgRowData.resize(nGroupsInRow_);
149 cgRowData.setStorageLayout(DataStorage::dslPosition);
150 cgColData.resize(nGroupsInCol_);
151 if (needVelocities_)
152 // we only need column velocities if we need them.
153 cgColData.setStorageLayout(DataStorage::dslPosition |
154 DataStorage::dslVelocity);
155 else
156 cgColData.setStorageLayout(DataStorage::dslPosition);
157
158 identsRow.resize(nAtomsInRow_);
159 identsCol.resize(nAtomsInCol_);
160
161 AtomPlanIntRow->gather(idents, identsRow);
162 AtomPlanIntColumn->gather(idents, identsCol);
163
164 regionsRow.resize(nAtomsInRow_);
165 regionsCol.resize(nAtomsInCol_);
166
167 AtomPlanIntRow->gather(regions, regionsRow);
168 AtomPlanIntColumn->gather(regions, regionsCol);
169
170 // allocate memory for the parallel objects
171 atypesRow.resize(nAtomsInRow_);
172 atypesCol.resize(nAtomsInCol_);
173
174 for (int i = 0; i < nAtomsInRow_; i++)
175 atypesRow[i] = ff_->getAtomType(identsRow[i]);
176 for (int i = 0; i < nAtomsInCol_; i++)
177 atypesCol[i] = ff_->getAtomType(identsCol[i]);
178
179 pot_row.resize(nAtomsInRow_);
180 pot_col.resize(nAtomsInCol_);
181
182 expot_row.resize(nAtomsInRow_);
183 expot_col.resize(nAtomsInCol_);
184
185 AtomRowToGlobal.resize(nAtomsInRow_);
186 AtomColToGlobal.resize(nAtomsInCol_);
187 AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
188 AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
189
190 cgRowToGlobal.resize(nGroupsInRow_);
191 cgColToGlobal.resize(nGroupsInCol_);
192 cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
193 cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
194
195 massFactorsRow.resize(nAtomsInRow_);
196 massFactorsCol.resize(nAtomsInCol_);
197 AtomPlanRealRow->gather(massFactors, massFactorsRow);
198 AtomPlanRealColumn->gather(massFactors, massFactorsCol);
199
200 groupListRow_.clear();
201 groupListRow_.resize(nGroupsInRow_);
202 for (int i = 0; i < nGroupsInRow_; i++) {
203 int gid = cgRowToGlobal[i];
204 for (int j = 0; j < nAtomsInRow_; j++) {
205 int aid = AtomRowToGlobal[j];
206 if (globalGroupMembership[aid] == gid)
207 groupListRow_[i].push_back(j);
208 }
209 }
210
211 groupListCol_.clear();
212 groupListCol_.resize(nGroupsInCol_);
213 for (int i = 0; i < nGroupsInCol_; i++) {
214 int gid = cgColToGlobal[i];
215 for (int j = 0; j < nAtomsInCol_; j++) {
216 int aid = AtomColToGlobal[j];
217 if (globalGroupMembership[aid] == gid)
218 groupListCol_[i].push_back(j);
219 }
220 }
221
222 excludesForAtom.clear();
223 excludesForAtom.resize(nAtomsInRow_);
224 toposForAtom.clear();
225 toposForAtom.resize(nAtomsInRow_);
226 topoDist.clear();
227 topoDist.resize(nAtomsInRow_);
228 for (int i = 0; i < nAtomsInRow_; i++) {
229 int iglob = AtomRowToGlobal[i];
230
231 for (int j = 0; j < nAtomsInCol_; j++) {
232 int jglob = AtomColToGlobal[j];
233
234 if (excludes->hasPair(iglob, jglob))
235 excludesForAtom[i].push_back(j);
236
237 if (oneTwo->hasPair(iglob, jglob)) {
238 toposForAtom[i].push_back(j);
239 topoDist[i].push_back(1);
240 } else {
241 if (oneThree->hasPair(iglob, jglob)) {
242 toposForAtom[i].push_back(j);
243 topoDist[i].push_back(2);
244 } else {
245 if (oneFour->hasPair(iglob, jglob)) {
246 toposForAtom[i].push_back(j);
247 topoDist[i].push_back(3);
248 }
249 }
250 }
251 }
252 }
253
254 #else
255 excludesForAtom.clear();
256 excludesForAtom.resize(nLocal_);
257 toposForAtom.clear();
258 toposForAtom.resize(nLocal_);
259 topoDist.clear();
260 topoDist.resize(nLocal_);
261
262 for (int i = 0; i < nLocal_; i++) {
263 int iglob = AtomLocalToGlobal[i];
264
265 for (int j = 0; j < nLocal_; j++) {
266 int jglob = AtomLocalToGlobal[j];
267
268 if (excludes->hasPair(iglob, jglob))
269 excludesForAtom[i].push_back(j);
270
271 if (oneTwo->hasPair(iglob, jglob)) {
272 toposForAtom[i].push_back(j);
273 topoDist[i].push_back(1);
274 } else {
275 if (oneThree->hasPair(iglob, jglob)) {
276 toposForAtom[i].push_back(j);
277 topoDist[i].push_back(2);
278 } else {
279 if (oneFour->hasPair(iglob, jglob)) {
280 toposForAtom[i].push_back(j);
281 topoDist[i].push_back(3);
282 }
283 }
284 }
285 }
286 }
287 #endif
288
289 // allocate memory for the parallel objects
290 atypesLocal.resize(nLocal_);
291
292 for (int i = 0; i < nLocal_; i++)
293 atypesLocal[i] = ff_->getAtomType(idents[i]);
294
295 groupList_.clear();
296 groupList_.resize(nGroups_);
297 for (int i = 0; i < nGroups_; i++) {
298 int gid = cgLocalToGlobal[i];
299 for (int j = 0; j < nLocal_; j++) {
300 int aid = AtomLocalToGlobal[j];
301 if (globalGroupMembership[aid] == gid) {
302 groupList_[i].push_back(j);
303 }
304 }
305 }
306 }
307
308 int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) {
309 for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) {
310 if (toposForAtom[atom1][j] == atom2)
311 return topoDist[atom1][j];
312 }
313 return 0;
314 }
315
316 void ForceMatrixDecomposition::zeroWorkArrays() {
317 pairwisePot = 0.0;
318 embeddingPot = 0.0;
319 excludedPot = 0.0;
320 excludedSelfPot = 0.0;
321
322 #ifdef IS_MPI
323 if (storageLayout_ & DataStorage::dslForce) {
324 fill(atomRowData.force.begin(), atomRowData.force.end(), V3Zero);
325 fill(atomColData.force.begin(), atomColData.force.end(), V3Zero);
326 }
327
328 if (storageLayout_ & DataStorage::dslTorque) {
329 fill(atomRowData.torque.begin(), atomRowData.torque.end(), V3Zero);
330 fill(atomColData.torque.begin(), atomColData.torque.end(), V3Zero);
331 }
332
333 fill(pot_row.begin(), pot_row.end(),
334 Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
335
336 fill(pot_col.begin(), pot_col.end(),
337 Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
338
339 fill(expot_row.begin(), expot_row.end(),
340 Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
341
342 fill(expot_col.begin(), expot_col.end(),
343 Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
344
345 if (storageLayout_ & DataStorage::dslParticlePot) {
346 fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(),
347 0.0);
348 fill(atomColData.particlePot.begin(), atomColData.particlePot.end(),
349 0.0);
350 }
351
352 if (storageLayout_ & DataStorage::dslDensity) {
353 fill(atomRowData.density.begin(), atomRowData.density.end(), 0.0);
354 fill(atomColData.density.begin(), atomColData.density.end(), 0.0);
355 }
356
357 if (storageLayout_ & DataStorage::dslFunctional) {
358 fill(atomRowData.functional.begin(), atomRowData.functional.end(),
359 0.0);
360 fill(atomColData.functional.begin(), atomColData.functional.end(),
361 0.0);
362 }
363
364 if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
365 fill(atomRowData.functionalDerivative.begin(),
366 atomRowData.functionalDerivative.end(), 0.0);
367 fill(atomColData.functionalDerivative.begin(),
368 atomColData.functionalDerivative.end(), 0.0);
369 }
370
371 if (storageLayout_ & DataStorage::dslSkippedCharge) {
372 fill(atomRowData.skippedCharge.begin(),
373 atomRowData.skippedCharge.end(), 0.0);
374 fill(atomColData.skippedCharge.begin(),
375 atomColData.skippedCharge.end(), 0.0);
376 }
377
378 if (storageLayout_ & DataStorage::dslFlucQForce) {
379 fill(atomRowData.flucQFrc.begin(),
380 atomRowData.flucQFrc.end(), 0.0);
381 fill(atomColData.flucQFrc.begin(),
382 atomColData.flucQFrc.end(), 0.0);
383 }
384
385 if (storageLayout_ & DataStorage::dslElectricField) {
386 fill(atomRowData.electricField.begin(),
387 atomRowData.electricField.end(), V3Zero);
388 fill(atomColData.electricField.begin(),
389 atomColData.electricField.end(), V3Zero);
390 }
391
392 if (storageLayout_ & DataStorage::dslSitePotential) {
393 fill(atomRowData.sitePotential.begin(),
394 atomRowData.sitePotential.end(), 0.0);
395 fill(atomColData.sitePotential.begin(),
396 atomColData.sitePotential.end(), 0.0);
397 }
398
399 #endif
400 // even in parallel, we need to zero out the local arrays:
401
402 if (storageLayout_ & DataStorage::dslParticlePot) {
403 fill(snap_->atomData.particlePot.begin(),
404 snap_->atomData.particlePot.end(), 0.0);
405 }
406
407 if (storageLayout_ & DataStorage::dslDensity) {
408 fill(snap_->atomData.density.begin(),
409 snap_->atomData.density.end(), 0.0);
410 }
411
412 if (storageLayout_ & DataStorage::dslFunctional) {
413 fill(snap_->atomData.functional.begin(),
414 snap_->atomData.functional.end(), 0.0);
415 }
416
417 if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
418 fill(snap_->atomData.functionalDerivative.begin(),
419 snap_->atomData.functionalDerivative.end(), 0.0);
420 }
421
422 if (storageLayout_ & DataStorage::dslSkippedCharge) {
423 fill(snap_->atomData.skippedCharge.begin(),
424 snap_->atomData.skippedCharge.end(), 0.0);
425 }
426
427 if (storageLayout_ & DataStorage::dslElectricField) {
428 fill(snap_->atomData.electricField.begin(),
429 snap_->atomData.electricField.end(), V3Zero);
430 }
431 if (storageLayout_ & DataStorage::dslSitePotential) {
432 fill(snap_->atomData.sitePotential.begin(),
433 snap_->atomData.sitePotential.end(), 0.0);
434 }
435 }
436
437
438 void ForceMatrixDecomposition::distributeData() {
439 snap_ = sman_->getCurrentSnapshot();
440 storageLayout_ = sman_->getStorageLayout();
441
442 bool needsCG = true;
443 if(info_->getNCutoffGroups() != info_->getNAtoms())
444 needsCG = false;
445
446 #ifdef IS_MPI
447
448 // gather up the atomic positions
449 AtomPlanVectorRow->gather(snap_->atomData.position,
450 atomRowData.position);
451 AtomPlanVectorColumn->gather(snap_->atomData.position,
452 atomColData.position);
453
454 // gather up the cutoff group positions
455
456 if (needsCG) {
457 cgPlanVectorRow->gather(snap_->cgData.position,
458 cgRowData.position);
459
460 cgPlanVectorColumn->gather(snap_->cgData.position,
461 cgColData.position);
462 }
463
464
465 if (needVelocities_) {
466 // gather up the atomic velocities
467 AtomPlanVectorColumn->gather(snap_->atomData.velocity,
468 atomColData.velocity);
469
470 if (needsCG) {
471 cgPlanVectorColumn->gather(snap_->cgData.velocity,
472 cgColData.velocity);
473 }
474 }
475
476
477 // if needed, gather the atomic rotation matrices
478 if (storageLayout_ & DataStorage::dslAmat) {
479 AtomPlanMatrixRow->gather(snap_->atomData.aMat,
480 atomRowData.aMat);
481 AtomPlanMatrixColumn->gather(snap_->atomData.aMat,
482 atomColData.aMat);
483 }
484
485 // if needed, gather the atomic eletrostatic information
486 if (storageLayout_ & DataStorage::dslDipole) {
487 AtomPlanVectorRow->gather(snap_->atomData.dipole,
488 atomRowData.dipole);
489 AtomPlanVectorColumn->gather(snap_->atomData.dipole,
490 atomColData.dipole);
491 }
492
493 if (storageLayout_ & DataStorage::dslQuadrupole) {
494 AtomPlanMatrixRow->gather(snap_->atomData.quadrupole,
495 atomRowData.quadrupole);
496 AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole,
497 atomColData.quadrupole);
498 }
499
500 // if needed, gather the atomic fluctuating charge values
501 if (storageLayout_ & DataStorage::dslFlucQPosition) {
502 AtomPlanRealRow->gather(snap_->atomData.flucQPos,
503 atomRowData.flucQPos);
504 AtomPlanRealColumn->gather(snap_->atomData.flucQPos,
505 atomColData.flucQPos);
506 }
507
508 #endif
509 }
510
511 /* collects information obtained during the pre-pair loop onto local
512 * data structures.
513 */
514 void ForceMatrixDecomposition::collectIntermediateData() {
515 snap_ = sman_->getCurrentSnapshot();
516 storageLayout_ = sman_->getStorageLayout();
517 #ifdef IS_MPI
518
519 if (storageLayout_ & DataStorage::dslDensity) {
520
521 AtomPlanRealRow->scatter(atomRowData.density,
522 snap_->atomData.density);
523
524 int n = snap_->atomData.density.size();
525 vector<RealType> rho_tmp(n, 0.0);
526 AtomPlanRealColumn->scatter(atomColData.density, rho_tmp);
527 for (int i = 0; i < n; i++)
528 snap_->atomData.density[i] += rho_tmp[i];
529 }
530
531 // this isn't necessary if we don't have polarizable atoms, but
532 // we'll leave it here for now.
533 if (storageLayout_ & DataStorage::dslElectricField) {
534
535 AtomPlanVectorRow->scatter(atomRowData.electricField,
536 snap_->atomData.electricField);
537
538 int n = snap_->atomData.electricField.size();
539 vector<Vector3d> field_tmp(n, V3Zero);
540 AtomPlanVectorColumn->scatter(atomColData.electricField,
541 field_tmp);
542 for (int i = 0; i < n; i++)
543 snap_->atomData.electricField[i] += field_tmp[i];
544 }
545 #endif
546 }
547
548 /*
549 * redistributes information obtained during the pre-pair loop out to
550 * row and column-indexed data structures
551 */
552 void ForceMatrixDecomposition::distributeIntermediateData() {
553 snap_ = sman_->getCurrentSnapshot();
554 storageLayout_ = sman_->getStorageLayout();
555 #ifdef IS_MPI
556 if (storageLayout_ & DataStorage::dslFunctional) {
557 AtomPlanRealRow->gather(snap_->atomData.functional,
558 atomRowData.functional);
559 AtomPlanRealColumn->gather(snap_->atomData.functional,
560 atomColData.functional);
561 }
562
563 if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
564 AtomPlanRealRow->gather(snap_->atomData.functionalDerivative,
565 atomRowData.functionalDerivative);
566 AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative,
567 atomColData.functionalDerivative);
568 }
569 #endif
570 }
571
572
573 void ForceMatrixDecomposition::collectData() {
574 snap_ = sman_->getCurrentSnapshot();
575 storageLayout_ = sman_->getStorageLayout();
576 #ifdef IS_MPI
577 int n = snap_->atomData.force.size();
578 vector<Vector3d> frc_tmp(n, V3Zero);
579
580 AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp);
581 for (int i = 0; i < n; i++) {
582 snap_->atomData.force[i] += frc_tmp[i];
583 frc_tmp[i] = 0.0;
584 }
585
586 AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp);
587 for (int i = 0; i < n; i++) {
588 snap_->atomData.force[i] += frc_tmp[i];
589 }
590
591 if (storageLayout_ & DataStorage::dslTorque) {
592
593 int nt = snap_->atomData.torque.size();
594 vector<Vector3d> trq_tmp(nt, V3Zero);
595
596 AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp);
597 for (int i = 0; i < nt; i++) {
598 snap_->atomData.torque[i] += trq_tmp[i];
599 trq_tmp[i] = 0.0;
600 }
601
602 AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp);
603 for (int i = 0; i < nt; i++)
604 snap_->atomData.torque[i] += trq_tmp[i];
605 }
606
607 if (storageLayout_ & DataStorage::dslSkippedCharge) {
608
609 int ns = snap_->atomData.skippedCharge.size();
610 vector<RealType> skch_tmp(ns, 0.0);
611
612 AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp);
613 for (int i = 0; i < ns; i++) {
614 snap_->atomData.skippedCharge[i] += skch_tmp[i];
615 skch_tmp[i] = 0.0;
616 }
617
618 AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp);
619 for (int i = 0; i < ns; i++)
620 snap_->atomData.skippedCharge[i] += skch_tmp[i];
621
622 }
623
624 if (storageLayout_ & DataStorage::dslFlucQForce) {
625
626 int nq = snap_->atomData.flucQFrc.size();
627 vector<RealType> fqfrc_tmp(nq, 0.0);
628
629 AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp);
630 for (int i = 0; i < nq; i++) {
631 snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
632 fqfrc_tmp[i] = 0.0;
633 }
634
635 AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp);
636 for (int i = 0; i < nq; i++)
637 snap_->atomData.flucQFrc[i] += fqfrc_tmp[i];
638
639 }
640
641 if (storageLayout_ & DataStorage::dslElectricField) {
642
643 int nef = snap_->atomData.electricField.size();
644 vector<Vector3d> efield_tmp(nef, V3Zero);
645
646 AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp);
647 for (int i = 0; i < nef; i++) {
648 snap_->atomData.electricField[i] += efield_tmp[i];
649 efield_tmp[i] = 0.0;
650 }
651
652 AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp);
653 for (int i = 0; i < nef; i++)
654 snap_->atomData.electricField[i] += efield_tmp[i];
655 }
656
657 if (storageLayout_ & DataStorage::dslSitePotential) {
658
659 int nsp = snap_->atomData.sitePotential.size();
660 vector<RealType> sp_tmp(nsp, 0.0);
661
662 AtomPlanRealRow->scatter(atomRowData.sitePotential, sp_tmp);
663 for (int i = 0; i < nsp; i++) {
664 snap_->atomData.sitePotential[i] += sp_tmp[i];
665 sp_tmp[i] = 0.0;
666 }
667
668 AtomPlanRealColumn->scatter(atomColData.sitePotential, sp_tmp);
669 for (int i = 0; i < nsp; i++)
670 snap_->atomData.sitePotential[i] += sp_tmp[i];
671 }
672
673 nLocal_ = snap_->getNumberOfAtoms();
674
675 vector<potVec> pot_temp(nLocal_,
676 Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
677 vector<potVec> expot_temp(nLocal_,
678 Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
679
680 // scatter/gather pot_row into the members of my column
681
682 AtomPlanPotRow->scatter(pot_row, pot_temp);
683 AtomPlanPotRow->scatter(expot_row, expot_temp);
684
685 for (int ii = 0; ii < pot_temp.size(); ii++ )
686 pairwisePot += pot_temp[ii];
687
688 for (int ii = 0; ii < expot_temp.size(); ii++ )
689 excludedPot += expot_temp[ii];
690
691 if (storageLayout_ & DataStorage::dslParticlePot) {
692 // This is the pairwise contribution to the particle pot. The
693 // embedding contribution is added in each of the low level
694 // non-bonded routines. In single processor, this is done in
695 // unpackInteractionData, not in collectData.
696 for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
697 for (int i = 0; i < nLocal_; i++) {
698 // factor of two is because the total potential terms are divided
699 // by 2 in parallel due to row/ column scatter
700 snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
701 }
702 }
703 }
704
705 fill(pot_temp.begin(), pot_temp.end(),
706 Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
707 fill(expot_temp.begin(), expot_temp.end(),
708 Vector<RealType, N_INTERACTION_FAMILIES> (0.0));
709
710 AtomPlanPotColumn->scatter(pot_col, pot_temp);
711 AtomPlanPotColumn->scatter(expot_col, expot_temp);
712
713 for (int ii = 0; ii < pot_temp.size(); ii++ )
714 pairwisePot += pot_temp[ii];
715
716 for (int ii = 0; ii < expot_temp.size(); ii++ )
717 excludedPot += expot_temp[ii];
718
719 if (storageLayout_ & DataStorage::dslParticlePot) {
720 // This is the pairwise contribution to the particle pot. The
721 // embedding contribution is added in each of the low level
722 // non-bonded routines. In single processor, this is done in
723 // unpackInteractionData, not in collectData.
724 for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
725 for (int i = 0; i < nLocal_; i++) {
726 // factor of two is because the total potential terms are divided
727 // by 2 in parallel due to row/ column scatter
728 snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii);
729 }
730 }
731 }
732
733 if (storageLayout_ & DataStorage::dslParticlePot) {
734 int npp = snap_->atomData.particlePot.size();
735 vector<RealType> ppot_temp(npp, 0.0);
736
737 // This is the direct or embedding contribution to the particle
738 // pot.
739
740 AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp);
741 for (int i = 0; i < npp; i++) {
742 snap_->atomData.particlePot[i] += ppot_temp[i];
743 }
744
745 fill(ppot_temp.begin(), ppot_temp.end(), 0.0);
746
747 AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp);
748 for (int i = 0; i < npp; i++) {
749 snap_->atomData.particlePot[i] += ppot_temp[i];
750 }
751 }
752
753 for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
754 RealType ploc1 = pairwisePot[ii];
755 RealType ploc2 = 0.0;
756 MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
757 pairwisePot[ii] = ploc2;
758 }
759
760 for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
761 RealType ploc1 = excludedPot[ii];
762 RealType ploc2 = 0.0;
763 MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
764 excludedPot[ii] = ploc2;
765 }
766
767 // Here be dragons.
768 MPI_Comm col = colComm.getComm();
769
770 MPI_Allreduce(MPI_IN_PLACE,
771 &snap_->frameData.conductiveHeatFlux[0], 3,
772 MPI_REALTYPE, MPI_SUM, col);
773
774
775 #endif
776
777 }
778
779 /**
780 * Collects information obtained during the post-pair (and embedding
781 * functional) loops onto local data structures.
782 */
783 void ForceMatrixDecomposition::collectSelfData() {
784 snap_ = sman_->getCurrentSnapshot();
785 storageLayout_ = sman_->getStorageLayout();
786
787 #ifdef IS_MPI
788 for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
789 RealType ploc1 = embeddingPot[ii];
790 RealType ploc2 = 0.0;
791 MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
792 embeddingPot[ii] = ploc2;
793 }
794 for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) {
795 RealType ploc1 = excludedSelfPot[ii];
796 RealType ploc2 = 0.0;
797 MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD);
798 excludedSelfPot[ii] = ploc2;
799 }
800 #endif
801
802 }
803
804
805
806 int& ForceMatrixDecomposition::getNAtomsInRow() {
807 #ifdef IS_MPI
808 return nAtomsInRow_;
809 #else
810 return nLocal_;
811 #endif
812 }
813
814 /**
815 * returns the list of atoms belonging to this group.
816 */
817 vector<int>& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){
818 #ifdef IS_MPI
819 return groupListRow_[cg1];
820 #else
821 return groupList_[cg1];
822 #endif
823 }
824
825 vector<int>& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){
826 #ifdef IS_MPI
827 return groupListCol_[cg2];
828 #else
829 return groupList_[cg2];
830 #endif
831 }
832
833 Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
834 Vector3d d;
835
836 #ifdef IS_MPI
837 d = cgColData.position[cg2] - cgRowData.position[cg1];
838 #else
839 d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
840 #endif
841
842 if (usePeriodicBoundaryConditions_) {
843 snap_->wrapVector(d);
844 }
845 return d;
846 }
847
848 Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){
849 #ifdef IS_MPI
850 return cgColData.velocity[cg2];
851 #else
852 return snap_->cgData.velocity[cg2];
853 #endif
854 }
855
856 Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){
857 #ifdef IS_MPI
858 return atomColData.velocity[atom2];
859 #else
860 return snap_->atomData.velocity[atom2];
861 #endif
862 }
863
864
865 Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
866
867 Vector3d d;
868
869 #ifdef IS_MPI
870 d = cgRowData.position[cg1] - atomRowData.position[atom1];
871 #else
872 d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
873 #endif
874 if (usePeriodicBoundaryConditions_) {
875 snap_->wrapVector(d);
876 }
877 return d;
878 }
879
880 Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
881 Vector3d d;
882
883 #ifdef IS_MPI
884 d = cgColData.position[cg2] - atomColData.position[atom2];
885 #else
886 d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
887 #endif
888 if (usePeriodicBoundaryConditions_) {
889 snap_->wrapVector(d);
890 }
891 return d;
892 }
893
894 RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) {
895 #ifdef IS_MPI
896 return massFactorsRow[atom1];
897 #else
898 return massFactors[atom1];
899 #endif
900 }
901
902 RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) {
903 #ifdef IS_MPI
904 return massFactorsCol[atom2];
905 #else
906 return massFactors[atom2];
907 #endif
908
909 }
910
911 Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
912 Vector3d d;
913
914 #ifdef IS_MPI
915 d = atomColData.position[atom2] - atomRowData.position[atom1];
916 #else
917 d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
918 #endif
919 if (usePeriodicBoundaryConditions_) {
920 snap_->wrapVector(d);
921 }
922 return d;
923 }
924
925 vector<int>& ForceMatrixDecomposition::getExcludesForAtom(int atom1) {
926 return excludesForAtom[atom1];
927 }
928
929 /**
930 * We need to exclude some overcounted interactions that result from
931 * the parallel decomposition.
932 */
933 bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) {
934 int unique_id_1, unique_id_2;
935
936 #ifdef IS_MPI
937 // in MPI, we have to look up the unique IDs for each atom
938 unique_id_1 = AtomRowToGlobal[atom1];
939 unique_id_2 = AtomColToGlobal[atom2];
940 // group1 = cgRowToGlobal[cg1];
941 // group2 = cgColToGlobal[cg2];
942 #else
943 unique_id_1 = AtomLocalToGlobal[atom1];
944 unique_id_2 = AtomLocalToGlobal[atom2];
945 int group1 = cgLocalToGlobal[cg1];
946 int group2 = cgLocalToGlobal[cg2];
947 #endif
948
949 if (unique_id_1 == unique_id_2) return true;
950
951 #ifdef IS_MPI
952 // this prevents us from doing the pair on multiple processors
953 if (unique_id_1 < unique_id_2) {
954 if ((unique_id_1 + unique_id_2) % 2 == 0) return true;
955 } else {
956 if ((unique_id_1 + unique_id_2) % 2 == 1) return true;
957 }
958 #endif
959
960 #ifndef IS_MPI
961 if (group1 == group2) {
962 if (unique_id_1 < unique_id_2) return true;
963 }
964 #endif
965
966 return false;
967 }
968
969 /**
970 * We need to handle the interactions for atoms who are involved in
971 * the same rigid body as well as some short range interactions
972 * (bonds, bends, torsions) differently from other interactions.
973 * We'll still visit the pairwise routines, but with a flag that
974 * tells those routines to exclude the pair from direct long range
975 * interactions. Some indirect interactions (notably reaction
976 * field) must still be handled for these pairs.
977 */
978 bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) {
979
980 // excludesForAtom was constructed to use row/column indices in the MPI
981 // version, and to use local IDs in the non-MPI version:
982
983 for (vector<int>::iterator i = excludesForAtom[atom1].begin();
984 i != excludesForAtom[atom1].end(); ++i) {
985 if ( (*i) == atom2 ) return true;
986 }
987
988 return false;
989 }
990
991
992 void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
993 #ifdef IS_MPI
994 atomRowData.force[atom1] += fg;
995 #else
996 snap_->atomData.force[atom1] += fg;
997 #endif
998 }
999
1000 void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
1001 #ifdef IS_MPI
1002 atomColData.force[atom2] += fg;
1003 #else
1004 snap_->atomData.force[atom2] += fg;
1005 #endif
1006 }
1007
1008 // filling interaction blocks with pointers
1009 void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat,
1010 int atom1, int atom2,
1011 bool newAtom1) {
1012
1013 idat.excluded = excludeAtomPair(atom1, atom2);
1014
1015 if (newAtom1) {
1016
1017 #ifdef IS_MPI
1018 idat.atid1 = identsRow[atom1];
1019 idat.atid2 = identsCol[atom2];
1020
1021 if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1022 idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1023 } else {
1024 idat.sameRegion = false;
1025 }
1026
1027 if (storageLayout_ & DataStorage::dslAmat) {
1028 idat.A1 = &(atomRowData.aMat[atom1]);
1029 idat.A2 = &(atomColData.aMat[atom2]);
1030 }
1031
1032 if (storageLayout_ & DataStorage::dslTorque) {
1033 idat.t1 = &(atomRowData.torque[atom1]);
1034 idat.t2 = &(atomColData.torque[atom2]);
1035 }
1036
1037 if (storageLayout_ & DataStorage::dslDipole) {
1038 idat.dipole1 = &(atomRowData.dipole[atom1]);
1039 idat.dipole2 = &(atomColData.dipole[atom2]);
1040 }
1041
1042 if (storageLayout_ & DataStorage::dslQuadrupole) {
1043 idat.quadrupole1 = &(atomRowData.quadrupole[atom1]);
1044 idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1045 }
1046
1047 if (storageLayout_ & DataStorage::dslDensity) {
1048 idat.rho1 = &(atomRowData.density[atom1]);
1049 idat.rho2 = &(atomColData.density[atom2]);
1050 }
1051
1052 if (storageLayout_ & DataStorage::dslFunctional) {
1053 idat.frho1 = &(atomRowData.functional[atom1]);
1054 idat.frho2 = &(atomColData.functional[atom2]);
1055 }
1056
1057 if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1058 idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
1059 idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1060 }
1061
1062 if (storageLayout_ & DataStorage::dslParticlePot) {
1063 idat.particlePot1 = &(atomRowData.particlePot[atom1]);
1064 idat.particlePot2 = &(atomColData.particlePot[atom2]);
1065 }
1066
1067 if (storageLayout_ & DataStorage::dslSkippedCharge) {
1068 idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]);
1069 idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1070 }
1071
1072 if (storageLayout_ & DataStorage::dslFlucQPosition) {
1073 idat.flucQ1 = &(atomRowData.flucQPos[atom1]);
1074 idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1075 }
1076
1077 #else
1078
1079 idat.atid1 = idents[atom1];
1080 idat.atid2 = idents[atom2];
1081
1082 if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1083 idat.sameRegion = (regions[atom1] == regions[atom2]);
1084 } else {
1085 idat.sameRegion = false;
1086 }
1087
1088 if (storageLayout_ & DataStorage::dslAmat) {
1089 idat.A1 = &(snap_->atomData.aMat[atom1]);
1090 idat.A2 = &(snap_->atomData.aMat[atom2]);
1091 }
1092
1093 if (storageLayout_ & DataStorage::dslTorque) {
1094 idat.t1 = &(snap_->atomData.torque[atom1]);
1095 idat.t2 = &(snap_->atomData.torque[atom2]);
1096 }
1097
1098 if (storageLayout_ & DataStorage::dslDipole) {
1099 idat.dipole1 = &(snap_->atomData.dipole[atom1]);
1100 idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1101 }
1102
1103 if (storageLayout_ & DataStorage::dslQuadrupole) {
1104 idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]);
1105 idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1106 }
1107
1108 if (storageLayout_ & DataStorage::dslDensity) {
1109 idat.rho1 = &(snap_->atomData.density[atom1]);
1110 idat.rho2 = &(snap_->atomData.density[atom2]);
1111 }
1112
1113 if (storageLayout_ & DataStorage::dslFunctional) {
1114 idat.frho1 = &(snap_->atomData.functional[atom1]);
1115 idat.frho2 = &(snap_->atomData.functional[atom2]);
1116 }
1117
1118 if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1119 idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
1120 idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1121 }
1122
1123 if (storageLayout_ & DataStorage::dslParticlePot) {
1124 idat.particlePot1 = &(snap_->atomData.particlePot[atom1]);
1125 idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1126 }
1127
1128 if (storageLayout_ & DataStorage::dslSkippedCharge) {
1129 idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]);
1130 idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1131 }
1132
1133 if (storageLayout_ & DataStorage::dslFlucQPosition) {
1134 idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]);
1135 idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1136 }
1137 #endif
1138
1139 } else {
1140 // atom1 is not new, so don't bother updating properties of that atom:
1141 #ifdef IS_MPI
1142 idat.atid2 = identsCol[atom2];
1143
1144 if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) {
1145 idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]);
1146 } else {
1147 idat.sameRegion = false;
1148 }
1149
1150 if (storageLayout_ & DataStorage::dslAmat) {
1151 idat.A2 = &(atomColData.aMat[atom2]);
1152 }
1153
1154 if (storageLayout_ & DataStorage::dslTorque) {
1155 idat.t2 = &(atomColData.torque[atom2]);
1156 }
1157
1158 if (storageLayout_ & DataStorage::dslDipole) {
1159 idat.dipole2 = &(atomColData.dipole[atom2]);
1160 }
1161
1162 if (storageLayout_ & DataStorage::dslQuadrupole) {
1163 idat.quadrupole2 = &(atomColData.quadrupole[atom2]);
1164 }
1165
1166 if (storageLayout_ & DataStorage::dslDensity) {
1167 idat.rho2 = &(atomColData.density[atom2]);
1168 }
1169
1170 if (storageLayout_ & DataStorage::dslFunctional) {
1171 idat.frho2 = &(atomColData.functional[atom2]);
1172 }
1173
1174 if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1175 idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
1176 }
1177
1178 if (storageLayout_ & DataStorage::dslParticlePot) {
1179 idat.particlePot2 = &(atomColData.particlePot[atom2]);
1180 }
1181
1182 if (storageLayout_ & DataStorage::dslSkippedCharge) {
1183 idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]);
1184 }
1185
1186 if (storageLayout_ & DataStorage::dslFlucQPosition) {
1187 idat.flucQ2 = &(atomColData.flucQPos[atom2]);
1188 }
1189
1190 #else
1191 idat.atid2 = idents[atom2];
1192
1193 if (regions[atom1] >= 0 && regions[atom2] >= 0) {
1194 idat.sameRegion = (regions[atom1] == regions[atom2]);
1195 } else {
1196 idat.sameRegion = false;
1197 }
1198
1199 if (storageLayout_ & DataStorage::dslAmat) {
1200 idat.A2 = &(snap_->atomData.aMat[atom2]);
1201 }
1202
1203 if (storageLayout_ & DataStorage::dslTorque) {
1204 idat.t2 = &(snap_->atomData.torque[atom2]);
1205 }
1206
1207 if (storageLayout_ & DataStorage::dslDipole) {
1208 idat.dipole2 = &(snap_->atomData.dipole[atom2]);
1209 }
1210
1211 if (storageLayout_ & DataStorage::dslQuadrupole) {
1212 idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]);
1213 }
1214
1215 if (storageLayout_ & DataStorage::dslDensity) {
1216 idat.rho2 = &(snap_->atomData.density[atom2]);
1217 }
1218
1219 if (storageLayout_ & DataStorage::dslFunctional) {
1220 idat.frho2 = &(snap_->atomData.functional[atom2]);
1221 }
1222
1223 if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
1224 idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
1225 }
1226
1227 if (storageLayout_ & DataStorage::dslParticlePot) {
1228 idat.particlePot2 = &(snap_->atomData.particlePot[atom2]);
1229 }
1230
1231 if (storageLayout_ & DataStorage::dslSkippedCharge) {
1232 idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]);
1233 }
1234
1235 if (storageLayout_ & DataStorage::dslFlucQPosition) {
1236 idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]);
1237 }
1238
1239 #endif
1240 }
1241 }
1242
1243 void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat,
1244 int atom1, int atom2) {
1245 #ifdef IS_MPI
1246 pot_row[atom1] += RealType(0.5) * *(idat.pot);
1247 pot_col[atom2] += RealType(0.5) * *(idat.pot);
1248 expot_row[atom1] += RealType(0.5) * *(idat.excludedPot);
1249 expot_col[atom2] += RealType(0.5) * *(idat.excludedPot);
1250
1251 atomRowData.force[atom1] += *(idat.f1);
1252 atomColData.force[atom2] -= *(idat.f1);
1253
1254 if (storageLayout_ & DataStorage::dslFlucQForce) {
1255 atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1256 atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1257 }
1258
1259 if (storageLayout_ & DataStorage::dslElectricField) {
1260 atomRowData.electricField[atom1] += *(idat.eField1);
1261 atomColData.electricField[atom2] += *(idat.eField2);
1262 }
1263
1264 if (storageLayout_ & DataStorage::dslSitePotential) {
1265 atomRowData.sitePotential[atom1] += *(idat.sPot1);
1266 atomColData.sitePotential[atom2] += *(idat.sPot2);
1267 }
1268
1269 #else
1270 pairwisePot += *(idat.pot);
1271 excludedPot += *(idat.excludedPot);
1272
1273 snap_->atomData.force[atom1] += *(idat.f1);
1274 snap_->atomData.force[atom2] -= *(idat.f1);
1275
1276 if (idat.doParticlePot) {
1277 // This is the pairwise contribution to the particle pot. The
1278 // embedding contribution is added in each of the low level
1279 // non-bonded routines. In parallel, this calculation is done
1280 // in collectData, not in unpackInteractionData.
1281 snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw);
1282 snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw);
1283 }
1284
1285 if (storageLayout_ & DataStorage::dslFlucQForce) {
1286 snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1);
1287 snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2);
1288 }
1289
1290 if (storageLayout_ & DataStorage::dslElectricField) {
1291 snap_->atomData.electricField[atom1] += *(idat.eField1);
1292 snap_->atomData.electricField[atom2] += *(idat.eField2);
1293 }
1294
1295 if (storageLayout_ & DataStorage::dslSitePotential) {
1296 snap_->atomData.sitePotential[atom1] += *(idat.sPot1);
1297 snap_->atomData.sitePotential[atom2] += *(idat.sPot2);
1298 }
1299
1300 #endif
1301
1302 }
1303
1304 /*
1305 * buildNeighborList
1306 *
1307 * Constructs the Verlet neighbor list for a force-matrix
1308 * decomposition. In this case, each processor is responsible for
1309 * row-site interactions with column-sites.
1310 *
1311 * neighborList is returned as a packed array of neighboring
1312 * column-ordered CutoffGroups. The starting position in
1313 * neighborList for each row-ordered CutoffGroup is given by the
1314 * returned vector point.
1315 */
1316 void ForceMatrixDecomposition::buildNeighborList(vector<int>& neighborList,
1317 vector<int>& point) {
1318 neighborList.clear();
1319 point.clear();
1320 int len = 0;
1321
1322 bool doAllPairs = false;
1323
1324 Snapshot* snap_ = sman_->getCurrentSnapshot();
1325 Mat3x3d box;
1326 Mat3x3d invBox;
1327
1328 Vector3d rs, scaled, dr;
1329 Vector3i whichCell;
1330 int cellIndex;
1331
1332 #ifdef IS_MPI
1333 cellListRow_.clear();
1334 cellListCol_.clear();
1335 point.resize(nGroupsInRow_+1);
1336 #else
1337 cellList_.clear();
1338 point.resize(nGroups_+1);
1339 #endif
1340
1341 if (!usePeriodicBoundaryConditions_) {
1342 box = snap_->getBoundingBox();
1343 invBox = snap_->getInvBoundingBox();
1344 } else {
1345 box = snap_->getHmat();
1346 invBox = snap_->getInvHmat();
1347 }
1348
1349 Vector3d A = box.getColumn(0);
1350 Vector3d B = box.getColumn(1);
1351 Vector3d C = box.getColumn(2);
1352
1353 // Required for triclinic cells
1354 Vector3d AxB = cross(A, B);
1355 Vector3d BxC = cross(B, C);
1356 Vector3d CxA = cross(C, A);
1357
1358 // unit vectors perpendicular to the faces of the triclinic cell:
1359 AxB.normalize();
1360 BxC.normalize();
1361 CxA.normalize();
1362
1363 // A set of perpendicular lengths in triclinic cells:
1364 RealType Wa = abs(dot(A, BxC));
1365 RealType Wb = abs(dot(B, CxA));
1366 RealType Wc = abs(dot(C, AxB));
1367
1368 nCells_.x() = int( Wa / rList_ );
1369 nCells_.y() = int( Wb / rList_ );
1370 nCells_.z() = int( Wc / rList_ );
1371
1372 // handle small boxes where the cell offsets can end up repeating cells
1373 if (nCells_.x() < 3) doAllPairs = true;
1374 if (nCells_.y() < 3) doAllPairs = true;
1375 if (nCells_.z() < 3) doAllPairs = true;
1376
1377 int nCtot = nCells_.x() * nCells_.y() * nCells_.z();
1378
1379 #ifdef IS_MPI
1380 cellListRow_.resize(nCtot);
1381 cellListCol_.resize(nCtot);
1382 #else
1383 cellList_.resize(nCtot);
1384 #endif
1385
1386 if (!doAllPairs) {
1387
1388 #ifdef IS_MPI
1389
1390 for (int i = 0; i < nGroupsInRow_; i++) {
1391 rs = cgRowData.position[i];
1392
1393 // scaled positions relative to the box vectors
1394 scaled = invBox * rs;
1395
1396 // wrap the vector back into the unit box by subtracting integer box
1397 // numbers
1398 for (int j = 0; j < 3; j++) {
1399 scaled[j] -= roundMe(scaled[j]);
1400 scaled[j] += 0.5;
1401 // Handle the special case when an object is exactly on the
1402 // boundary (a scaled coordinate of 1.0 is the same as
1403 // scaled coordinate of 0.0)
1404 if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1405 }
1406
1407 // find xyz-indices of cell that cutoffGroup is in.
1408 whichCell.x() = nCells_.x() * scaled.x();
1409 whichCell.y() = nCells_.y() * scaled.y();
1410 whichCell.z() = nCells_.z() * scaled.z();
1411
1412 // find single index of this cell:
1413 cellIndex = Vlinear(whichCell, nCells_);
1414
1415 // add this cutoff group to the list of groups in this cell;
1416 cellListRow_[cellIndex].push_back(i);
1417 }
1418 for (int i = 0; i < nGroupsInCol_; i++) {
1419 rs = cgColData.position[i];
1420
1421 // scaled positions relative to the box vectors
1422 scaled = invBox * rs;
1423
1424 // wrap the vector back into the unit box by subtracting integer box
1425 // numbers
1426 for (int j = 0; j < 3; j++) {
1427 scaled[j] -= roundMe(scaled[j]);
1428 scaled[j] += 0.5;
1429 // Handle the special case when an object is exactly on the
1430 // boundary (a scaled coordinate of 1.0 is the same as
1431 // scaled coordinate of 0.0)
1432 if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1433 }
1434
1435 // find xyz-indices of cell that cutoffGroup is in.
1436 whichCell.x() = nCells_.x() * scaled.x();
1437 whichCell.y() = nCells_.y() * scaled.y();
1438 whichCell.z() = nCells_.z() * scaled.z();
1439
1440 // find single index of this cell:
1441 cellIndex = Vlinear(whichCell, nCells_);
1442
1443 // add this cutoff group to the list of groups in this cell;
1444 cellListCol_[cellIndex].push_back(i);
1445 }
1446
1447 #else
1448 for (int i = 0; i < nGroups_; i++) {
1449 rs = snap_->cgData.position[i];
1450
1451 // scaled positions relative to the box vectors
1452 scaled = invBox * rs;
1453
1454 // wrap the vector back into the unit box by subtracting integer box
1455 // numbers
1456 for (int j = 0; j < 3; j++) {
1457 scaled[j] -= roundMe(scaled[j]);
1458 scaled[j] += 0.5;
1459 // Handle the special case when an object is exactly on the
1460 // boundary (a scaled coordinate of 1.0 is the same as
1461 // scaled coordinate of 0.0)
1462 if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1463 }
1464
1465 // find xyz-indices of cell that cutoffGroup is in.
1466 whichCell.x() = int(nCells_.x() * scaled.x());
1467 whichCell.y() = int(nCells_.y() * scaled.y());
1468 whichCell.z() = int(nCells_.z() * scaled.z());
1469
1470 // find single index of this cell:
1471 cellIndex = Vlinear(whichCell, nCells_);
1472
1473 // add this cutoff group to the list of groups in this cell;
1474 cellList_[cellIndex].push_back(i);
1475 }
1476
1477 #endif
1478
1479 #ifdef IS_MPI
1480 for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1481 rs = cgRowData.position[j1];
1482 #else
1483
1484 for (int j1 = 0; j1 < nGroups_; j1++) {
1485 rs = snap_->cgData.position[j1];
1486 #endif
1487 point[j1] = len;
1488
1489 // scaled positions relative to the box vectors
1490 scaled = invBox * rs;
1491
1492 // wrap the vector back into the unit box by subtracting integer box
1493 // numbers
1494 for (int j = 0; j < 3; j++) {
1495 scaled[j] -= roundMe(scaled[j]);
1496 scaled[j] += 0.5;
1497 // Handle the special case when an object is exactly on the
1498 // boundary (a scaled coordinate of 1.0 is the same as
1499 // scaled coordinate of 0.0)
1500 if (scaled[j] >= 1.0) scaled[j] -= 1.0;
1501 }
1502
1503 // find xyz-indices of cell that cutoffGroup is in.
1504 whichCell.x() = nCells_.x() * scaled.x();
1505 whichCell.y() = nCells_.y() * scaled.y();
1506 whichCell.z() = nCells_.z() * scaled.z();
1507
1508 // find single index of this cell:
1509 int m1 = Vlinear(whichCell, nCells_);
1510
1511 for (vector<Vector3i>::iterator os = cellOffsets_.begin();
1512 os != cellOffsets_.end(); ++os) {
1513
1514 Vector3i m2v = whichCell + (*os);
1515
1516 if (m2v.x() >= nCells_.x()) {
1517 m2v.x() = 0;
1518 } else if (m2v.x() < 0) {
1519 m2v.x() = nCells_.x() - 1;
1520 }
1521
1522 if (m2v.y() >= nCells_.y()) {
1523 m2v.y() = 0;
1524 } else if (m2v.y() < 0) {
1525 m2v.y() = nCells_.y() - 1;
1526 }
1527
1528 if (m2v.z() >= nCells_.z()) {
1529 m2v.z() = 0;
1530 } else if (m2v.z() < 0) {
1531 m2v.z() = nCells_.z() - 1;
1532 }
1533 int m2 = Vlinear (m2v, nCells_);
1534 #ifdef IS_MPI
1535 for (vector<int>::iterator j2 = cellListCol_[m2].begin();
1536 j2 != cellListCol_[m2].end(); ++j2) {
1537
1538 // In parallel, we need to visit *all* pairs of row
1539 // & column indicies and will divide labor in the
1540 // force evaluation later.
1541 dr = cgColData.position[(*j2)] - rs;
1542 if (usePeriodicBoundaryConditions_) {
1543 snap_->wrapVector(dr);
1544 }
1545 if (dr.lengthSquare() < rListSq_) {
1546 neighborList.push_back( (*j2) );
1547 ++len;
1548 }
1549 }
1550 #else
1551 for (vector<int>::iterator j2 = cellList_[m2].begin();
1552 j2 != cellList_[m2].end(); ++j2) {
1553
1554 // Always do this if we're in different cells or if
1555 // we're in the same cell and the global index of
1556 // the j2 cutoff group is greater than or equal to
1557 // the j1 cutoff group. Note that Rappaport's code
1558 // has a "less than" conditional here, but that
1559 // deals with atom-by-atom computation. OpenMD
1560 // allows atoms within a single cutoff group to
1561 // interact with each other.
1562
1563 if ( (*j2) >= j1 ) {
1564
1565 dr = snap_->cgData.position[(*j2)] - rs;
1566 if (usePeriodicBoundaryConditions_) {
1567 snap_->wrapVector(dr);
1568 }
1569 if ( dr.lengthSquare() < rListSq_) {
1570 neighborList.push_back( (*j2) );
1571 ++len;
1572 }
1573 }
1574 }
1575 #endif
1576 }
1577 }
1578 } else {
1579 // branch to do all cutoff group pairs
1580 #ifdef IS_MPI
1581 for (int j1 = 0; j1 < nGroupsInRow_; j1++) {
1582 point[j1] = len;
1583 rs = cgRowData.position[j1];
1584 for (int j2 = 0; j2 < nGroupsInCol_; j2++) {
1585 dr = cgColData.position[j2] - rs;
1586 if (usePeriodicBoundaryConditions_) {
1587 snap_->wrapVector(dr);
1588 }
1589 if (dr.lengthSquare() < rListSq_) {
1590 neighborList.push_back( j2 );
1591 ++len;
1592 }
1593 }
1594 }
1595 #else
1596 // include all groups here.
1597 for (int j1 = 0; j1 < nGroups_; j1++) {
1598 point[j1] = len;
1599 rs = snap_->cgData.position[j1];
1600 // include self group interactions j2 == j1
1601 for (int j2 = j1; j2 < nGroups_; j2++) {
1602 dr = snap_->cgData.position[j2] - rs;
1603 if (usePeriodicBoundaryConditions_) {
1604 snap_->wrapVector(dr);
1605 }
1606 if (dr.lengthSquare() < rListSq_) {
1607 neighborList.push_back( j2 );
1608 ++len;
1609 }
1610 }
1611 }
1612 #endif
1613 }
1614
1615 #ifdef IS_MPI
1616 point[nGroupsInRow_] = len;
1617 #else
1618 point[nGroups_] = len;
1619 #endif
1620
1621 // save the local cutoff group positions for the check that is
1622 // done on each loop:
1623 saved_CG_positions_.clear();
1624 saved_CG_positions_.reserve(nGroups_);
1625 for (int i = 0; i < nGroups_; i++)
1626 saved_CG_positions_.push_back(snap_->cgData.position[i]);
1627 }
1628 } //end namespace OpenMD