47 |
|
using namespace std; |
48 |
|
namespace OpenMD { |
49 |
|
|
50 |
+ |
ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) { |
51 |
+ |
|
52 |
+ |
// In a parallel computation, row and colum scans must visit all |
53 |
+ |
// surrounding cells (not just the 14 upper triangular blocks that |
54 |
+ |
// are used when the processor can see all pairs) |
55 |
+ |
#ifdef IS_MPI |
56 |
+ |
cellOffsets_.clear(); |
57 |
+ |
cellOffsets_.push_back( Vector3i(-1,-1,-1) ); |
58 |
+ |
cellOffsets_.push_back( Vector3i( 0,-1,-1) ); |
59 |
+ |
cellOffsets_.push_back( Vector3i( 1,-1,-1) ); |
60 |
+ |
cellOffsets_.push_back( Vector3i(-1, 0,-1) ); |
61 |
+ |
cellOffsets_.push_back( Vector3i( 0, 0,-1) ); |
62 |
+ |
cellOffsets_.push_back( Vector3i( 1, 0,-1) ); |
63 |
+ |
cellOffsets_.push_back( Vector3i(-1, 1,-1) ); |
64 |
+ |
cellOffsets_.push_back( Vector3i( 0, 1,-1) ); |
65 |
+ |
cellOffsets_.push_back( Vector3i( 1, 1,-1) ); |
66 |
+ |
cellOffsets_.push_back( Vector3i(-1,-1, 0) ); |
67 |
+ |
cellOffsets_.push_back( Vector3i( 0,-1, 0) ); |
68 |
+ |
cellOffsets_.push_back( Vector3i( 1,-1, 0) ); |
69 |
+ |
cellOffsets_.push_back( Vector3i(-1, 0, 0) ); |
70 |
+ |
cellOffsets_.push_back( Vector3i( 0, 0, 0) ); |
71 |
+ |
cellOffsets_.push_back( Vector3i( 1, 0, 0) ); |
72 |
+ |
cellOffsets_.push_back( Vector3i(-1, 1, 0) ); |
73 |
+ |
cellOffsets_.push_back( Vector3i( 0, 1, 0) ); |
74 |
+ |
cellOffsets_.push_back( Vector3i( 1, 1, 0) ); |
75 |
+ |
cellOffsets_.push_back( Vector3i(-1,-1, 1) ); |
76 |
+ |
cellOffsets_.push_back( Vector3i( 0,-1, 1) ); |
77 |
+ |
cellOffsets_.push_back( Vector3i( 1,-1, 1) ); |
78 |
+ |
cellOffsets_.push_back( Vector3i(-1, 0, 1) ); |
79 |
+ |
cellOffsets_.push_back( Vector3i( 0, 0, 1) ); |
80 |
+ |
cellOffsets_.push_back( Vector3i( 1, 0, 1) ); |
81 |
+ |
cellOffsets_.push_back( Vector3i(-1, 1, 1) ); |
82 |
+ |
cellOffsets_.push_back( Vector3i( 0, 1, 1) ); |
83 |
+ |
cellOffsets_.push_back( Vector3i( 1, 1, 1) ); |
84 |
+ |
#endif |
85 |
+ |
} |
86 |
+ |
|
87 |
+ |
|
88 |
|
/** |
89 |
|
* distributeInitialData is essentially a copy of the older fortran |
90 |
|
* SimulationSetup |
91 |
|
*/ |
54 |
– |
|
92 |
|
void ForceMatrixDecomposition::distributeInitialData() { |
93 |
|
snap_ = sman_->getCurrentSnapshot(); |
94 |
|
storageLayout_ = sman_->getStorageLayout(); |
111 |
|
|
112 |
|
#ifdef IS_MPI |
113 |
|
|
114 |
< |
AtomCommIntRow = new Communicator<Row,int>(nLocal_); |
115 |
< |
AtomCommRealRow = new Communicator<Row,RealType>(nLocal_); |
79 |
< |
AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_); |
80 |
< |
AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_); |
81 |
< |
AtomCommPotRow = new Communicator<Row,potVec>(nLocal_); |
114 |
> |
MPI::Intracomm row = rowComm.getComm(); |
115 |
> |
MPI::Intracomm col = colComm.getComm(); |
116 |
|
|
117 |
< |
AtomCommIntColumn = new Communicator<Column,int>(nLocal_); |
118 |
< |
AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_); |
119 |
< |
AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_); |
120 |
< |
AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_); |
121 |
< |
AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_); |
117 |
> |
AtomPlanIntRow = new Plan<int>(row, nLocal_); |
118 |
> |
AtomPlanRealRow = new Plan<RealType>(row, nLocal_); |
119 |
> |
AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_); |
120 |
> |
AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_); |
121 |
> |
AtomPlanPotRow = new Plan<potVec>(row, nLocal_); |
122 |
|
|
123 |
< |
cgCommIntRow = new Communicator<Row,int>(nGroups_); |
124 |
< |
cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_); |
125 |
< |
cgCommIntColumn = new Communicator<Column,int>(nGroups_); |
126 |
< |
cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_); |
123 |
> |
AtomPlanIntColumn = new Plan<int>(col, nLocal_); |
124 |
> |
AtomPlanRealColumn = new Plan<RealType>(col, nLocal_); |
125 |
> |
AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_); |
126 |
> |
AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_); |
127 |
> |
AtomPlanPotColumn = new Plan<potVec>(col, nLocal_); |
128 |
|
|
129 |
< |
nAtomsInRow_ = AtomCommIntRow->getSize(); |
130 |
< |
nAtomsInCol_ = AtomCommIntColumn->getSize(); |
131 |
< |
nGroupsInRow_ = cgCommIntRow->getSize(); |
132 |
< |
nGroupsInCol_ = cgCommIntColumn->getSize(); |
129 |
> |
cgPlanIntRow = new Plan<int>(row, nGroups_); |
130 |
> |
cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_); |
131 |
> |
cgPlanIntColumn = new Plan<int>(col, nGroups_); |
132 |
> |
cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_); |
133 |
|
|
134 |
+ |
nAtomsInRow_ = AtomPlanIntRow->getSize(); |
135 |
+ |
nAtomsInCol_ = AtomPlanIntColumn->getSize(); |
136 |
+ |
nGroupsInRow_ = cgPlanIntRow->getSize(); |
137 |
+ |
nGroupsInCol_ = cgPlanIntColumn->getSize(); |
138 |
+ |
|
139 |
|
// Modify the data storage objects with the correct layouts and sizes: |
140 |
|
atomRowData.resize(nAtomsInRow_); |
141 |
|
atomRowData.setStorageLayout(storageLayout_); |
149 |
|
identsRow.resize(nAtomsInRow_); |
150 |
|
identsCol.resize(nAtomsInCol_); |
151 |
|
|
152 |
< |
AtomCommIntRow->gather(idents, identsRow); |
153 |
< |
AtomCommIntColumn->gather(idents, identsCol); |
152 |
> |
AtomPlanIntRow->gather(idents, identsRow); |
153 |
> |
AtomPlanIntColumn->gather(idents, identsCol); |
154 |
|
|
155 |
|
// allocate memory for the parallel objects |
156 |
+ |
atypesRow.resize(nAtomsInRow_); |
157 |
+ |
atypesCol.resize(nAtomsInCol_); |
158 |
+ |
|
159 |
+ |
for (int i = 0; i < nAtomsInRow_; i++) |
160 |
+ |
atypesRow[i] = ff_->getAtomType(identsRow[i]); |
161 |
+ |
for (int i = 0; i < nAtomsInCol_; i++) |
162 |
+ |
atypesCol[i] = ff_->getAtomType(identsCol[i]); |
163 |
+ |
|
164 |
+ |
pot_row.resize(nAtomsInRow_); |
165 |
+ |
pot_col.resize(nAtomsInCol_); |
166 |
+ |
|
167 |
|
AtomRowToGlobal.resize(nAtomsInRow_); |
168 |
|
AtomColToGlobal.resize(nAtomsInCol_); |
169 |
+ |
AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal); |
170 |
+ |
AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal); |
171 |
+ |
|
172 |
|
cgRowToGlobal.resize(nGroupsInRow_); |
173 |
|
cgColToGlobal.resize(nGroupsInCol_); |
174 |
+ |
cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal); |
175 |
+ |
cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal); |
176 |
+ |
|
177 |
|
massFactorsRow.resize(nAtomsInRow_); |
178 |
|
massFactorsCol.resize(nAtomsInCol_); |
179 |
< |
pot_row.resize(nAtomsInRow_); |
180 |
< |
pot_col.resize(nAtomsInCol_); |
179 |
> |
AtomPlanRealRow->gather(massFactors, massFactorsRow); |
180 |
> |
AtomPlanRealColumn->gather(massFactors, massFactorsCol); |
181 |
|
|
125 |
– |
AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal); |
126 |
– |
AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal); |
127 |
– |
|
128 |
– |
cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal); |
129 |
– |
cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal); |
130 |
– |
|
131 |
– |
AtomCommRealRow->gather(massFactors, massFactorsRow); |
132 |
– |
AtomCommRealColumn->gather(massFactors, massFactorsCol); |
133 |
– |
|
182 |
|
groupListRow_.clear(); |
183 |
|
groupListRow_.resize(nGroupsInRow_); |
184 |
|
for (int i = 0; i < nGroupsInRow_; i++) { |
233 |
|
} |
234 |
|
} |
235 |
|
|
236 |
< |
#endif |
189 |
< |
|
190 |
< |
groupList_.clear(); |
191 |
< |
groupList_.resize(nGroups_); |
192 |
< |
for (int i = 0; i < nGroups_; i++) { |
193 |
< |
int gid = cgLocalToGlobal[i]; |
194 |
< |
for (int j = 0; j < nLocal_; j++) { |
195 |
< |
int aid = AtomLocalToGlobal[j]; |
196 |
< |
if (globalGroupMembership[aid] == gid) { |
197 |
< |
groupList_[i].push_back(j); |
198 |
< |
} |
199 |
< |
} |
200 |
< |
} |
201 |
< |
|
236 |
> |
#else |
237 |
|
excludesForAtom.clear(); |
238 |
|
excludesForAtom.resize(nLocal_); |
239 |
|
toposForAtom.clear(); |
247 |
|
for (int j = 0; j < nLocal_; j++) { |
248 |
|
int jglob = AtomLocalToGlobal[j]; |
249 |
|
|
250 |
< |
if (excludes->hasPair(iglob, jglob)) |
250 |
> |
if (excludes->hasPair(iglob, jglob)) |
251 |
|
excludesForAtom[i].push_back(j); |
252 |
|
|
253 |
+ |
|
254 |
|
if (oneTwo->hasPair(iglob, jglob)) { |
255 |
|
toposForAtom[i].push_back(j); |
256 |
|
topoDist[i].push_back(1); |
267 |
|
} |
268 |
|
} |
269 |
|
} |
270 |
< |
|
270 |
> |
#endif |
271 |
> |
|
272 |
> |
// allocate memory for the parallel objects |
273 |
> |
atypesLocal.resize(nLocal_); |
274 |
> |
|
275 |
> |
for (int i = 0; i < nLocal_; i++) |
276 |
> |
atypesLocal[i] = ff_->getAtomType(idents[i]); |
277 |
> |
|
278 |
> |
groupList_.clear(); |
279 |
> |
groupList_.resize(nGroups_); |
280 |
> |
for (int i = 0; i < nGroups_; i++) { |
281 |
> |
int gid = cgLocalToGlobal[i]; |
282 |
> |
for (int j = 0; j < nLocal_; j++) { |
283 |
> |
int aid = AtomLocalToGlobal[j]; |
284 |
> |
if (globalGroupMembership[aid] == gid) { |
285 |
> |
groupList_[i].push_back(j); |
286 |
> |
} |
287 |
> |
} |
288 |
> |
} |
289 |
> |
|
290 |
> |
|
291 |
|
createGtypeCutoffMap(); |
292 |
|
|
293 |
|
} |
295 |
|
void ForceMatrixDecomposition::createGtypeCutoffMap() { |
296 |
|
|
297 |
|
RealType tol = 1e-6; |
298 |
+ |
largestRcut_ = 0.0; |
299 |
|
RealType rc; |
300 |
|
int atid; |
301 |
|
set<AtomType*> atypes = info_->getSimulatedAtomTypes(); |
302 |
+ |
|
303 |
|
map<int, RealType> atypeCutoff; |
304 |
|
|
305 |
|
for (set<AtomType*>::iterator at = atypes.begin(); |
307 |
|
atid = (*at)->getIdent(); |
308 |
|
if (userChoseCutoff_) |
309 |
|
atypeCutoff[atid] = userCutoff_; |
310 |
< |
else |
310 |
> |
else |
311 |
|
atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at); |
312 |
|
} |
313 |
< |
|
313 |
> |
|
314 |
|
vector<RealType> gTypeCutoffs; |
315 |
|
// first we do a single loop over the cutoff groups to find the |
316 |
|
// largest cutoff for any atypes present in this group. |
370 |
|
vector<RealType> groupCutoff(nGroups_, 0.0); |
371 |
|
groupToGtype.resize(nGroups_); |
372 |
|
for (int cg1 = 0; cg1 < nGroups_; cg1++) { |
315 |
– |
|
373 |
|
groupCutoff[cg1] = 0.0; |
374 |
|
vector<int> atomList = getAtomsInGroupRow(cg1); |
318 |
– |
|
375 |
|
for (vector<int>::iterator ia = atomList.begin(); |
376 |
|
ia != atomList.end(); ++ia) { |
377 |
|
int atom1 = (*ia); |
378 |
|
atid = idents[atom1]; |
379 |
< |
if (atypeCutoff[atid] > groupCutoff[cg1]) { |
379 |
> |
if (atypeCutoff[atid] > groupCutoff[cg1]) |
380 |
|
groupCutoff[cg1] = atypeCutoff[atid]; |
325 |
– |
} |
381 |
|
} |
382 |
< |
|
382 |
> |
|
383 |
|
bool gTypeFound = false; |
384 |
|
for (int gt = 0; gt < gTypeCutoffs.size(); gt++) { |
385 |
|
if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) { |
387 |
|
gTypeFound = true; |
388 |
|
} |
389 |
|
} |
390 |
< |
if (!gTypeFound) { |
390 |
> |
if (!gTypeFound) { |
391 |
|
gTypeCutoffs.push_back( groupCutoff[cg1] ); |
392 |
|
groupToGtype[cg1] = gTypeCutoffs.size() - 1; |
393 |
|
} |
431 |
|
|
432 |
|
pair<int,int> key = make_pair(i,j); |
433 |
|
gTypeCutoffMap[key].first = thisRcut; |
379 |
– |
|
434 |
|
if (thisRcut > largestRcut_) largestRcut_ = thisRcut; |
381 |
– |
|
435 |
|
gTypeCutoffMap[key].second = thisRcut*thisRcut; |
383 |
– |
|
436 |
|
gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2); |
385 |
– |
|
437 |
|
// sanity check |
438 |
|
|
439 |
|
if (userChoseCutoff_) { |
559 |
|
#ifdef IS_MPI |
560 |
|
|
561 |
|
// gather up the atomic positions |
562 |
< |
AtomCommVectorRow->gather(snap_->atomData.position, |
562 |
> |
AtomPlanVectorRow->gather(snap_->atomData.position, |
563 |
|
atomRowData.position); |
564 |
< |
AtomCommVectorColumn->gather(snap_->atomData.position, |
564 |
> |
AtomPlanVectorColumn->gather(snap_->atomData.position, |
565 |
|
atomColData.position); |
566 |
|
|
567 |
|
// gather up the cutoff group positions |
568 |
< |
cgCommVectorRow->gather(snap_->cgData.position, |
568 |
> |
|
569 |
> |
cgPlanVectorRow->gather(snap_->cgData.position, |
570 |
|
cgRowData.position); |
571 |
< |
cgCommVectorColumn->gather(snap_->cgData.position, |
571 |
> |
|
572 |
> |
cgPlanVectorColumn->gather(snap_->cgData.position, |
573 |
|
cgColData.position); |
574 |
+ |
|
575 |
|
|
576 |
|
// if needed, gather the atomic rotation matrices |
577 |
|
if (storageLayout_ & DataStorage::dslAmat) { |
578 |
< |
AtomCommMatrixRow->gather(snap_->atomData.aMat, |
578 |
> |
AtomPlanMatrixRow->gather(snap_->atomData.aMat, |
579 |
|
atomRowData.aMat); |
580 |
< |
AtomCommMatrixColumn->gather(snap_->atomData.aMat, |
580 |
> |
AtomPlanMatrixColumn->gather(snap_->atomData.aMat, |
581 |
|
atomColData.aMat); |
582 |
|
} |
583 |
|
|
584 |
|
// if needed, gather the atomic eletrostatic frames |
585 |
|
if (storageLayout_ & DataStorage::dslElectroFrame) { |
586 |
< |
AtomCommMatrixRow->gather(snap_->atomData.electroFrame, |
586 |
> |
AtomPlanMatrixRow->gather(snap_->atomData.electroFrame, |
587 |
|
atomRowData.electroFrame); |
588 |
< |
AtomCommMatrixColumn->gather(snap_->atomData.electroFrame, |
588 |
> |
AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame, |
589 |
|
atomColData.electroFrame); |
590 |
|
} |
591 |
|
|
602 |
|
|
603 |
|
if (storageLayout_ & DataStorage::dslDensity) { |
604 |
|
|
605 |
< |
AtomCommRealRow->scatter(atomRowData.density, |
605 |
> |
AtomPlanRealRow->scatter(atomRowData.density, |
606 |
|
snap_->atomData.density); |
607 |
|
|
608 |
|
int n = snap_->atomData.density.size(); |
609 |
|
vector<RealType> rho_tmp(n, 0.0); |
610 |
< |
AtomCommRealColumn->scatter(atomColData.density, rho_tmp); |
610 |
> |
AtomPlanRealColumn->scatter(atomColData.density, rho_tmp); |
611 |
|
for (int i = 0; i < n; i++) |
612 |
|
snap_->atomData.density[i] += rho_tmp[i]; |
613 |
|
} |
623 |
|
storageLayout_ = sman_->getStorageLayout(); |
624 |
|
#ifdef IS_MPI |
625 |
|
if (storageLayout_ & DataStorage::dslFunctional) { |
626 |
< |
AtomCommRealRow->gather(snap_->atomData.functional, |
626 |
> |
AtomPlanRealRow->gather(snap_->atomData.functional, |
627 |
|
atomRowData.functional); |
628 |
< |
AtomCommRealColumn->gather(snap_->atomData.functional, |
628 |
> |
AtomPlanRealColumn->gather(snap_->atomData.functional, |
629 |
|
atomColData.functional); |
630 |
|
} |
631 |
|
|
632 |
|
if (storageLayout_ & DataStorage::dslFunctionalDerivative) { |
633 |
< |
AtomCommRealRow->gather(snap_->atomData.functionalDerivative, |
633 |
> |
AtomPlanRealRow->gather(snap_->atomData.functionalDerivative, |
634 |
|
atomRowData.functionalDerivative); |
635 |
< |
AtomCommRealColumn->gather(snap_->atomData.functionalDerivative, |
635 |
> |
AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative, |
636 |
|
atomColData.functionalDerivative); |
637 |
|
} |
638 |
|
#endif |
646 |
|
int n = snap_->atomData.force.size(); |
647 |
|
vector<Vector3d> frc_tmp(n, V3Zero); |
648 |
|
|
649 |
< |
AtomCommVectorRow->scatter(atomRowData.force, frc_tmp); |
649 |
> |
AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp); |
650 |
|
for (int i = 0; i < n; i++) { |
651 |
|
snap_->atomData.force[i] += frc_tmp[i]; |
652 |
|
frc_tmp[i] = 0.0; |
653 |
|
} |
654 |
|
|
655 |
< |
AtomCommVectorColumn->scatter(atomColData.force, frc_tmp); |
656 |
< |
for (int i = 0; i < n; i++) |
655 |
> |
AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp); |
656 |
> |
for (int i = 0; i < n; i++) { |
657 |
|
snap_->atomData.force[i] += frc_tmp[i]; |
658 |
+ |
} |
659 |
|
|
660 |
|
if (storageLayout_ & DataStorage::dslTorque) { |
661 |
|
|
662 |
|
int nt = snap_->atomData.torque.size(); |
663 |
|
vector<Vector3d> trq_tmp(nt, V3Zero); |
664 |
|
|
665 |
< |
AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp); |
665 |
> |
AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp); |
666 |
|
for (int i = 0; i < nt; i++) { |
667 |
|
snap_->atomData.torque[i] += trq_tmp[i]; |
668 |
|
trq_tmp[i] = 0.0; |
669 |
|
} |
670 |
|
|
671 |
< |
AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp); |
671 |
> |
AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp); |
672 |
|
for (int i = 0; i < nt; i++) |
673 |
|
snap_->atomData.torque[i] += trq_tmp[i]; |
674 |
|
} |
678 |
|
int ns = snap_->atomData.skippedCharge.size(); |
679 |
|
vector<RealType> skch_tmp(ns, 0.0); |
680 |
|
|
681 |
< |
AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp); |
681 |
> |
AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp); |
682 |
|
for (int i = 0; i < ns; i++) { |
683 |
|
snap_->atomData.skippedCharge[i] += skch_tmp[i]; |
684 |
|
skch_tmp[i] = 0.0; |
685 |
|
} |
686 |
|
|
687 |
< |
AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp); |
688 |
< |
for (int i = 0; i < ns; i++) |
687 |
> |
AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp); |
688 |
> |
for (int i = 0; i < ns; i++) |
689 |
|
snap_->atomData.skippedCharge[i] += skch_tmp[i]; |
690 |
+ |
|
691 |
|
} |
692 |
|
|
693 |
|
nLocal_ = snap_->getNumberOfAtoms(); |
697 |
|
|
698 |
|
// scatter/gather pot_row into the members of my column |
699 |
|
|
700 |
< |
AtomCommPotRow->scatter(pot_row, pot_temp); |
700 |
> |
AtomPlanPotRow->scatter(pot_row, pot_temp); |
701 |
|
|
702 |
|
for (int ii = 0; ii < pot_temp.size(); ii++ ) |
703 |
|
pairwisePot += pot_temp[ii]; |
705 |
|
fill(pot_temp.begin(), pot_temp.end(), |
706 |
|
Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); |
707 |
|
|
708 |
< |
AtomCommPotColumn->scatter(pot_col, pot_temp); |
708 |
> |
AtomPlanPotColumn->scatter(pot_col, pot_temp); |
709 |
|
|
710 |
|
for (int ii = 0; ii < pot_temp.size(); ii++ ) |
711 |
|
pairwisePot += pot_temp[ii]; |
712 |
+ |
|
713 |
+ |
for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { |
714 |
+ |
RealType ploc1 = pairwisePot[ii]; |
715 |
+ |
RealType ploc2 = 0.0; |
716 |
+ |
MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM); |
717 |
+ |
pairwisePot[ii] = ploc2; |
718 |
+ |
} |
719 |
+ |
|
720 |
+ |
for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { |
721 |
+ |
RealType ploc1 = embeddingPot[ii]; |
722 |
+ |
RealType ploc2 = 0.0; |
723 |
+ |
MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM); |
724 |
+ |
embeddingPot[ii] = ploc2; |
725 |
+ |
} |
726 |
+ |
|
727 |
|
#endif |
728 |
|
|
729 |
|
} |
836 |
|
*/ |
837 |
|
bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) { |
838 |
|
int unique_id_1, unique_id_2; |
839 |
< |
|
839 |
> |
|
840 |
|
#ifdef IS_MPI |
841 |
|
// in MPI, we have to look up the unique IDs for each atom |
842 |
|
unique_id_1 = AtomRowToGlobal[atom1]; |
865 |
|
* field) must still be handled for these pairs. |
866 |
|
*/ |
867 |
|
bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) { |
868 |
< |
int unique_id_2; |
868 |
> |
|
869 |
> |
// excludesForAtom was constructed to use row/column indices in the MPI |
870 |
> |
// version, and to use local IDs in the non-MPI version: |
871 |
|
|
799 |
– |
#ifdef IS_MPI |
800 |
– |
// in MPI, we have to look up the unique IDs for the row atom. |
801 |
– |
unique_id_2 = AtomColToGlobal[atom2]; |
802 |
– |
#else |
803 |
– |
// in the normal loop, the atom numbers are unique |
804 |
– |
unique_id_2 = atom2; |
805 |
– |
#endif |
806 |
– |
|
872 |
|
for (vector<int>::iterator i = excludesForAtom[atom1].begin(); |
873 |
|
i != excludesForAtom[atom1].end(); ++i) { |
874 |
< |
if ( (*i) == unique_id_2 ) return true; |
874 |
> |
if ( (*i) == atom2 ) return true; |
875 |
|
} |
876 |
|
|
877 |
|
return false; |
901 |
|
idat.excluded = excludeAtomPair(atom1, atom2); |
902 |
|
|
903 |
|
#ifdef IS_MPI |
904 |
+ |
idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]); |
905 |
+ |
//idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]), |
906 |
+ |
// ff_->getAtomType(identsCol[atom2]) ); |
907 |
|
|
840 |
– |
idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]), |
841 |
– |
ff_->getAtomType(identsCol[atom2]) ); |
842 |
– |
|
908 |
|
if (storageLayout_ & DataStorage::dslAmat) { |
909 |
|
idat.A1 = &(atomRowData.aMat[atom1]); |
910 |
|
idat.A2 = &(atomColData.aMat[atom2]); |
947 |
|
|
948 |
|
#else |
949 |
|
|
950 |
< |
idat.atypes = make_pair( ff_->getAtomType(idents[atom1]), |
951 |
< |
ff_->getAtomType(idents[atom2]) ); |
950 |
> |
idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]); |
951 |
> |
//idat.atypes = make_pair( ff_->getAtomType(idents[atom1]), |
952 |
> |
// ff_->getAtomType(idents[atom2]) ); |
953 |
|
|
954 |
|
if (storageLayout_ & DataStorage::dslAmat) { |
955 |
|
idat.A1 = &(snap_->atomData.aMat[atom1]); |
1087 |
|
// add this cutoff group to the list of groups in this cell; |
1088 |
|
cellListRow_[cellIndex].push_back(i); |
1089 |
|
} |
1024 |
– |
|
1090 |
|
for (int i = 0; i < nGroupsInCol_; i++) { |
1091 |
|
rs = cgColData.position[i]; |
1092 |
|
|
1111 |
|
// add this cutoff group to the list of groups in this cell; |
1112 |
|
cellListCol_[cellIndex].push_back(i); |
1113 |
|
} |
1114 |
+ |
|
1115 |
|
#else |
1116 |
|
for (int i = 0; i < nGroups_; i++) { |
1117 |
|
rs = snap_->cgData.position[i]; |
1132 |
|
whichCell.z() = nCells_.z() * scaled.z(); |
1133 |
|
|
1134 |
|
// find single index of this cell: |
1135 |
< |
cellIndex = Vlinear(whichCell, nCells_); |
1135 |
> |
cellIndex = Vlinear(whichCell, nCells_); |
1136 |
|
|
1137 |
|
// add this cutoff group to the list of groups in this cell; |
1138 |
|
cellList_[cellIndex].push_back(i); |
1139 |
|
} |
1140 |
+ |
|
1141 |
|
#endif |
1142 |
|
|
1143 |
|
for (int m1z = 0; m1z < nCells_.z(); m1z++) { |
1150 |
|
os != cellOffsets_.end(); ++os) { |
1151 |
|
|
1152 |
|
Vector3i m2v = m1v + (*os); |
1153 |
< |
|
1153 |
> |
|
1154 |
> |
|
1155 |
|
if (m2v.x() >= nCells_.x()) { |
1156 |
|
m2v.x() = 0; |
1157 |
|
} else if (m2v.x() < 0) { |
1169 |
|
} else if (m2v.z() < 0) { |
1170 |
|
m2v.z() = nCells_.z() - 1; |
1171 |
|
} |
1172 |
< |
|
1172 |
> |
|
1173 |
|
int m2 = Vlinear (m2v, nCells_); |
1174 |
|
|
1175 |
|
#ifdef IS_MPI |
1178 |
|
for (vector<int>::iterator j2 = cellListCol_[m2].begin(); |
1179 |
|
j2 != cellListCol_[m2].end(); ++j2) { |
1180 |
|
|
1181 |
< |
// Always do this if we're in different cells or if |
1182 |
< |
// we're in the same cell and the global index of the |
1183 |
< |
// j2 cutoff group is less than the j1 cutoff group |
1184 |
< |
|
1185 |
< |
if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) { |
1186 |
< |
dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)]; |
1187 |
< |
snap_->wrapVector(dr); |
1188 |
< |
cuts = getGroupCutoffs( (*j1), (*j2) ); |
1189 |
< |
if (dr.lengthSquare() < cuts.third) { |
1122 |
< |
neighborList.push_back(make_pair((*j1), (*j2))); |
1123 |
< |
} |
1124 |
< |
} |
1181 |
> |
// In parallel, we need to visit *all* pairs of row |
1182 |
> |
// & column indicies and will divide labor in the |
1183 |
> |
// force evaluation later. |
1184 |
> |
dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)]; |
1185 |
> |
snap_->wrapVector(dr); |
1186 |
> |
cuts = getGroupCutoffs( (*j1), (*j2) ); |
1187 |
> |
if (dr.lengthSquare() < cuts.third) { |
1188 |
> |
neighborList.push_back(make_pair((*j1), (*j2))); |
1189 |
> |
} |
1190 |
|
} |
1191 |
|
} |
1192 |
|
#else |