| 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 | < | vector<int>::iterator it; | 
| 156 | < | for (it = AtomLocalToGlobal.begin(); it != AtomLocalToGlobal.end(); ++it) { | 
| 157 | < | cerr << "my AtomLocalToGlobal = " << (*it) << "\n"; | 
| 118 | < | } | 
| 119 | < | AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal); | 
| 120 | < | AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal); | 
| 121 | < |  | 
| 122 | < | cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal); | 
| 123 | < | cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal); | 
| 155 | > | // allocate memory for the parallel objects | 
| 156 | > | atypesRow.resize(nAtomsInRow_); | 
| 157 | > | atypesCol.resize(nAtomsInCol_); | 
| 158 |  |  | 
| 159 | < | AtomCommRealRow->gather(massFactors, massFactorsRow); | 
| 160 | < | AtomCommRealColumn->gather(massFactors, massFactorsCol); | 
| 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 | + | AtomPlanRealRow->gather(massFactors, massFactorsRow); | 
| 180 | + | AtomPlanRealColumn->gather(massFactors, massFactorsCol); | 
| 181 | + |  | 
| 182 |  | groupListRow_.clear(); | 
| 183 |  | groupListRow_.resize(nGroupsInRow_); | 
| 184 |  | for (int i = 0; i < nGroupsInRow_; i++) { | 
| 235 |  |  | 
| 236 |  | #endif | 
| 237 |  |  | 
| 238 | + | // allocate memory for the parallel objects | 
| 239 | + | atypesLocal.resize(nLocal_); | 
| 240 | + |  | 
| 241 | + | for (int i = 0; i < nLocal_; i++) | 
| 242 | + | atypesLocal[i] = ff_->getAtomType(idents[i]); | 
| 243 | + |  | 
| 244 |  | groupList_.clear(); | 
| 245 |  | groupList_.resize(nGroups_); | 
| 246 |  | for (int i = 0; i < nGroups_; i++) { | 
| 293 |  | void ForceMatrixDecomposition::createGtypeCutoffMap() { | 
| 294 |  |  | 
| 295 |  | RealType tol = 1e-6; | 
| 296 | + | largestRcut_ = 0.0; | 
| 297 |  | RealType rc; | 
| 298 |  | int atid; | 
| 299 |  | set<AtomType*> atypes = info_->getSimulatedAtomTypes(); | 
| 300 | + |  | 
| 301 |  | map<int, RealType> atypeCutoff; | 
| 302 |  |  | 
| 303 |  | for (set<AtomType*>::iterator at = atypes.begin(); | 
| 305 |  | atid = (*at)->getIdent(); | 
| 306 |  | if (userChoseCutoff_) | 
| 307 |  | atypeCutoff[atid] = userCutoff_; | 
| 308 | < | else | 
| 308 | > | else | 
| 309 |  | atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at); | 
| 310 |  | } | 
| 311 | < |  | 
| 311 | > |  | 
| 312 |  | vector<RealType> gTypeCutoffs; | 
| 313 |  | // first we do a single loop over the cutoff groups to find the | 
| 314 |  | // largest cutoff for any atypes present in this group. | 
| 368 |  | vector<RealType> groupCutoff(nGroups_, 0.0); | 
| 369 |  | groupToGtype.resize(nGroups_); | 
| 370 |  | for (int cg1 = 0; cg1 < nGroups_; cg1++) { | 
| 309 | – |  | 
| 371 |  | groupCutoff[cg1] = 0.0; | 
| 372 |  | vector<int> atomList = getAtomsInGroupRow(cg1); | 
| 312 | – |  | 
| 373 |  | for (vector<int>::iterator ia = atomList.begin(); | 
| 374 |  | ia != atomList.end(); ++ia) { | 
| 375 |  | int atom1 = (*ia); | 
| 376 |  | atid = idents[atom1]; | 
| 377 | < | if (atypeCutoff[atid] > groupCutoff[cg1]) { | 
| 377 | > | if (atypeCutoff[atid] > groupCutoff[cg1]) | 
| 378 |  | groupCutoff[cg1] = atypeCutoff[atid]; | 
| 319 | – | } | 
| 379 |  | } | 
| 380 | < |  | 
| 380 | > |  | 
| 381 |  | bool gTypeFound = false; | 
| 382 |  | for (int gt = 0; gt < gTypeCutoffs.size(); gt++) { | 
| 383 |  | if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) { | 
| 385 |  | gTypeFound = true; | 
| 386 |  | } | 
| 387 |  | } | 
| 388 | < | if (!gTypeFound) { | 
| 388 | > | if (!gTypeFound) { | 
| 389 |  | gTypeCutoffs.push_back( groupCutoff[cg1] ); | 
| 390 |  | groupToGtype[cg1] = gTypeCutoffs.size() - 1; | 
| 391 |  | } | 
| 394 |  |  | 
| 395 |  | // Now we find the maximum group cutoff value present in the simulation | 
| 396 |  |  | 
| 397 | < | RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end()); | 
| 397 | > | RealType groupMax = *max_element(gTypeCutoffs.begin(), | 
| 398 | > | gTypeCutoffs.end()); | 
| 399 |  |  | 
| 400 |  | #ifdef IS_MPI | 
| 401 | < | MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX); | 
| 401 | > | MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, | 
| 402 | > | MPI::MAX); | 
| 403 |  | #endif | 
| 404 |  |  | 
| 405 |  | RealType tradRcut = groupMax; | 
| 429 |  |  | 
| 430 |  | pair<int,int> key = make_pair(i,j); | 
| 431 |  | gTypeCutoffMap[key].first = thisRcut; | 
| 371 | – |  | 
| 432 |  | if (thisRcut > largestRcut_) largestRcut_ = thisRcut; | 
| 373 | – |  | 
| 433 |  | gTypeCutoffMap[key].second = thisRcut*thisRcut; | 
| 375 | – |  | 
| 434 |  | gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2); | 
| 377 | – |  | 
| 435 |  | // sanity check | 
| 436 |  |  | 
| 437 |  | if (userChoseCutoff_) { | 
| 491 |  | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 492 |  |  | 
| 493 |  | if (storageLayout_ & DataStorage::dslParticlePot) { | 
| 494 | < | fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0); | 
| 495 | < | fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0); | 
| 494 | > | fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), | 
| 495 | > | 0.0); | 
| 496 | > | fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), | 
| 497 | > | 0.0); | 
| 498 |  | } | 
| 499 |  |  | 
| 500 |  | if (storageLayout_ & DataStorage::dslDensity) { | 
| 503 |  | } | 
| 504 |  |  | 
| 505 |  | if (storageLayout_ & DataStorage::dslFunctional) { | 
| 506 | < | fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0); | 
| 507 | < | fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0); | 
| 506 | > | fill(atomRowData.functional.begin(), atomRowData.functional.end(), | 
| 507 | > | 0.0); | 
| 508 | > | fill(atomColData.functional.begin(), atomColData.functional.end(), | 
| 509 | > | 0.0); | 
| 510 |  | } | 
| 511 |  |  | 
| 512 |  | if (storageLayout_ & DataStorage::dslFunctionalDerivative) { | 
| 523 |  | atomColData.skippedCharge.end(), 0.0); | 
| 524 |  | } | 
| 525 |  |  | 
| 526 | < | #else | 
| 527 | < |  | 
| 526 | > | #endif | 
| 527 | > | // even in parallel, we need to zero out the local arrays: | 
| 528 | > |  | 
| 529 |  | if (storageLayout_ & DataStorage::dslParticlePot) { | 
| 530 |  | fill(snap_->atomData.particlePot.begin(), | 
| 531 |  | snap_->atomData.particlePot.end(), 0.0); | 
| 547 |  | fill(snap_->atomData.skippedCharge.begin(), | 
| 548 |  | snap_->atomData.skippedCharge.end(), 0.0); | 
| 549 |  | } | 
| 488 | – | #endif | 
| 550 |  |  | 
| 551 |  | } | 
| 552 |  |  | 
| 557 |  | #ifdef IS_MPI | 
| 558 |  |  | 
| 559 |  | // gather up the atomic positions | 
| 560 | < | AtomCommVectorRow->gather(snap_->atomData.position, | 
| 560 | > | AtomPlanVectorRow->gather(snap_->atomData.position, | 
| 561 |  | atomRowData.position); | 
| 562 | < | AtomCommVectorColumn->gather(snap_->atomData.position, | 
| 562 | > | AtomPlanVectorColumn->gather(snap_->atomData.position, | 
| 563 |  | atomColData.position); | 
| 564 |  |  | 
| 565 |  | // gather up the cutoff group positions | 
| 566 | < | cgCommVectorRow->gather(snap_->cgData.position, | 
| 566 | > |  | 
| 567 | > | cgPlanVectorRow->gather(snap_->cgData.position, | 
| 568 |  | cgRowData.position); | 
| 569 | < | cgCommVectorColumn->gather(snap_->cgData.position, | 
| 569 | > |  | 
| 570 | > | cgPlanVectorColumn->gather(snap_->cgData.position, | 
| 571 |  | cgColData.position); | 
| 572 | + |  | 
| 573 |  |  | 
| 574 |  | // if needed, gather the atomic rotation matrices | 
| 575 |  | if (storageLayout_ & DataStorage::dslAmat) { | 
| 576 | < | AtomCommMatrixRow->gather(snap_->atomData.aMat, | 
| 576 | > | AtomPlanMatrixRow->gather(snap_->atomData.aMat, | 
| 577 |  | atomRowData.aMat); | 
| 578 | < | AtomCommMatrixColumn->gather(snap_->atomData.aMat, | 
| 578 | > | AtomPlanMatrixColumn->gather(snap_->atomData.aMat, | 
| 579 |  | atomColData.aMat); | 
| 580 |  | } | 
| 581 |  |  | 
| 582 |  | // if needed, gather the atomic eletrostatic frames | 
| 583 |  | if (storageLayout_ & DataStorage::dslElectroFrame) { | 
| 584 | < | AtomCommMatrixRow->gather(snap_->atomData.electroFrame, | 
| 584 | > | AtomPlanMatrixRow->gather(snap_->atomData.electroFrame, | 
| 585 |  | atomRowData.electroFrame); | 
| 586 | < | AtomCommMatrixColumn->gather(snap_->atomData.electroFrame, | 
| 586 | > | AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame, | 
| 587 |  | atomColData.electroFrame); | 
| 588 |  | } | 
| 589 | + |  | 
| 590 |  | #endif | 
| 591 |  | } | 
| 592 |  |  | 
| 600 |  |  | 
| 601 |  | if (storageLayout_ & DataStorage::dslDensity) { | 
| 602 |  |  | 
| 603 | < | AtomCommRealRow->scatter(atomRowData.density, | 
| 603 | > | AtomPlanRealRow->scatter(atomRowData.density, | 
| 604 |  | snap_->atomData.density); | 
| 605 |  |  | 
| 606 |  | int n = snap_->atomData.density.size(); | 
| 607 |  | vector<RealType> rho_tmp(n, 0.0); | 
| 608 | < | AtomCommRealColumn->scatter(atomColData.density, rho_tmp); | 
| 608 | > | AtomPlanRealColumn->scatter(atomColData.density, rho_tmp); | 
| 609 |  | for (int i = 0; i < n; i++) | 
| 610 |  | snap_->atomData.density[i] += rho_tmp[i]; | 
| 611 |  | } | 
| 621 |  | storageLayout_ = sman_->getStorageLayout(); | 
| 622 |  | #ifdef IS_MPI | 
| 623 |  | if (storageLayout_ & DataStorage::dslFunctional) { | 
| 624 | < | AtomCommRealRow->gather(snap_->atomData.functional, | 
| 624 | > | AtomPlanRealRow->gather(snap_->atomData.functional, | 
| 625 |  | atomRowData.functional); | 
| 626 | < | AtomCommRealColumn->gather(snap_->atomData.functional, | 
| 626 | > | AtomPlanRealColumn->gather(snap_->atomData.functional, | 
| 627 |  | atomColData.functional); | 
| 628 |  | } | 
| 629 |  |  | 
| 630 |  | if (storageLayout_ & DataStorage::dslFunctionalDerivative) { | 
| 631 | < | AtomCommRealRow->gather(snap_->atomData.functionalDerivative, | 
| 631 | > | AtomPlanRealRow->gather(snap_->atomData.functionalDerivative, | 
| 632 |  | atomRowData.functionalDerivative); | 
| 633 | < | AtomCommRealColumn->gather(snap_->atomData.functionalDerivative, | 
| 633 | > | AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative, | 
| 634 |  | atomColData.functionalDerivative); | 
| 635 |  | } | 
| 636 |  | #endif | 
| 644 |  | int n = snap_->atomData.force.size(); | 
| 645 |  | vector<Vector3d> frc_tmp(n, V3Zero); | 
| 646 |  |  | 
| 647 | < | AtomCommVectorRow->scatter(atomRowData.force, frc_tmp); | 
| 647 | > | AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp); | 
| 648 |  | for (int i = 0; i < n; i++) { | 
| 649 |  | snap_->atomData.force[i] += frc_tmp[i]; | 
| 650 |  | frc_tmp[i] = 0.0; | 
| 651 |  | } | 
| 652 |  |  | 
| 653 | < | AtomCommVectorColumn->scatter(atomColData.force, frc_tmp); | 
| 654 | < | for (int i = 0; i < n; i++) | 
| 653 | > | AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp); | 
| 654 | > | for (int i = 0; i < n; i++) { | 
| 655 |  | snap_->atomData.force[i] += frc_tmp[i]; | 
| 656 | < |  | 
| 657 | < |  | 
| 656 | > | } | 
| 657 | > |  | 
| 658 |  | if (storageLayout_ & DataStorage::dslTorque) { | 
| 659 |  |  | 
| 660 |  | int nt = snap_->atomData.torque.size(); | 
| 661 |  | vector<Vector3d> trq_tmp(nt, V3Zero); | 
| 662 |  |  | 
| 663 | < | AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp); | 
| 663 | > | AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp); | 
| 664 |  | for (int i = 0; i < nt; i++) { | 
| 665 |  | snap_->atomData.torque[i] += trq_tmp[i]; | 
| 666 |  | trq_tmp[i] = 0.0; | 
| 667 |  | } | 
| 668 |  |  | 
| 669 | < | AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp); | 
| 669 | > | AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp); | 
| 670 |  | for (int i = 0; i < nt; i++) | 
| 671 |  | snap_->atomData.torque[i] += trq_tmp[i]; | 
| 672 |  | } | 
| 676 |  | int ns = snap_->atomData.skippedCharge.size(); | 
| 677 |  | vector<RealType> skch_tmp(ns, 0.0); | 
| 678 |  |  | 
| 679 | < | AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp); | 
| 679 | > | AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp); | 
| 680 |  | for (int i = 0; i < ns; i++) { | 
| 681 | < | snap_->atomData.skippedCharge[i] = skch_tmp[i]; | 
| 681 | > | snap_->atomData.skippedCharge[i] += skch_tmp[i]; | 
| 682 |  | skch_tmp[i] = 0.0; | 
| 683 |  | } | 
| 684 |  |  | 
| 685 | < | AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp); | 
| 685 | > | AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp); | 
| 686 |  | for (int i = 0; i < ns; i++) | 
| 687 |  | snap_->atomData.skippedCharge[i] += skch_tmp[i]; | 
| 688 |  | } | 
| 694 |  |  | 
| 695 |  | // scatter/gather pot_row into the members of my column | 
| 696 |  |  | 
| 697 | < | AtomCommPotRow->scatter(pot_row, pot_temp); | 
| 697 | > | AtomPlanPotRow->scatter(pot_row, pot_temp); | 
| 698 |  |  | 
| 699 |  | for (int ii = 0;  ii < pot_temp.size(); ii++ ) | 
| 700 |  | pairwisePot += pot_temp[ii]; | 
| 702 |  | fill(pot_temp.begin(), pot_temp.end(), | 
| 703 |  | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 704 |  |  | 
| 705 | < | AtomCommPotColumn->scatter(pot_col, pot_temp); | 
| 705 | > | AtomPlanPotColumn->scatter(pot_col, pot_temp); | 
| 706 |  |  | 
| 707 |  | for (int ii = 0;  ii < pot_temp.size(); ii++ ) | 
| 708 |  | pairwisePot += pot_temp[ii]; | 
| 709 | + |  | 
| 710 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { | 
| 711 | + | RealType ploc1 = pairwisePot[ii]; | 
| 712 | + | RealType ploc2 = 0.0; | 
| 713 | + | MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM); | 
| 714 | + | pairwisePot[ii] = ploc2; | 
| 715 | + | } | 
| 716 | + |  | 
| 717 |  | #endif | 
| 718 |  |  | 
| 719 |  | } | 
| 826 |  | */ | 
| 827 |  | bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) { | 
| 828 |  | int unique_id_1, unique_id_2; | 
| 829 | < |  | 
| 829 | > |  | 
| 830 |  | #ifdef IS_MPI | 
| 831 |  | // in MPI, we have to look up the unique IDs for each atom | 
| 832 |  | unique_id_1 = AtomRowToGlobal[atom1]; | 
| 856 |  | */ | 
| 857 |  | bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) { | 
| 858 |  | int unique_id_2; | 
| 786 | – |  | 
| 859 |  | #ifdef IS_MPI | 
| 860 |  | // in MPI, we have to look up the unique IDs for the row atom. | 
| 861 |  | unique_id_2 = AtomColToGlobal[atom2]; | 
| 896 |  | idat.excluded = excludeAtomPair(atom1, atom2); | 
| 897 |  |  | 
| 898 |  | #ifdef IS_MPI | 
| 899 | + | idat.atypes = make_pair( atypesRow[atom1], atypesCol[atom2]); | 
| 900 | + | //idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]), | 
| 901 | + | //                         ff_->getAtomType(identsCol[atom2]) ); | 
| 902 |  |  | 
| 828 | – | idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]), | 
| 829 | – | ff_->getAtomType(identsCol[atom2]) ); | 
| 830 | – |  | 
| 903 |  | if (storageLayout_ & DataStorage::dslAmat) { | 
| 904 |  | idat.A1 = &(atomRowData.aMat[atom1]); | 
| 905 |  | idat.A2 = &(atomColData.aMat[atom2]); | 
| 942 |  |  | 
| 943 |  | #else | 
| 944 |  |  | 
| 945 | < | idat.atypes = make_pair( ff_->getAtomType(idents[atom1]), | 
| 946 | < | ff_->getAtomType(idents[atom2]) ); | 
| 945 | > | idat.atypes = make_pair( atypesLocal[atom1], atypesLocal[atom2]); | 
| 946 | > | //idat.atypes = make_pair( ff_->getAtomType(idents[atom1]), | 
| 947 | > | //                         ff_->getAtomType(idents[atom2]) ); | 
| 948 |  |  | 
| 949 |  | if (storageLayout_ & DataStorage::dslAmat) { | 
| 950 |  | idat.A1 = &(snap_->atomData.aMat[atom1]); | 
| 1082 |  | // add this cutoff group to the list of groups in this cell; | 
| 1083 |  | cellListRow_[cellIndex].push_back(i); | 
| 1084 |  | } | 
| 1012 | – |  | 
| 1085 |  | for (int i = 0; i < nGroupsInCol_; i++) { | 
| 1086 |  | rs = cgColData.position[i]; | 
| 1087 |  |  | 
| 1106 |  | // add this cutoff group to the list of groups in this cell; | 
| 1107 |  | cellListCol_[cellIndex].push_back(i); | 
| 1108 |  | } | 
| 1109 | + |  | 
| 1110 |  | #else | 
| 1111 |  | for (int i = 0; i < nGroups_; i++) { | 
| 1112 |  | rs = snap_->cgData.position[i]; | 
| 1127 |  | whichCell.z() = nCells_.z() * scaled.z(); | 
| 1128 |  |  | 
| 1129 |  | // find single index of this cell: | 
| 1130 | < | cellIndex = Vlinear(whichCell, nCells_); | 
| 1130 | > | cellIndex = Vlinear(whichCell, nCells_); | 
| 1131 |  |  | 
| 1132 |  | // add this cutoff group to the list of groups in this cell; | 
| 1133 |  | cellList_[cellIndex].push_back(i); | 
| 1134 |  | } | 
| 1135 | + |  | 
| 1136 |  | #endif | 
| 1137 |  |  | 
| 1138 |  | for (int m1z = 0; m1z < nCells_.z(); m1z++) { | 
| 1145 |  | os != cellOffsets_.end(); ++os) { | 
| 1146 |  |  | 
| 1147 |  | Vector3i m2v = m1v + (*os); | 
| 1148 | < |  | 
| 1148 | > |  | 
| 1149 | > |  | 
| 1150 |  | if (m2v.x() >= nCells_.x()) { | 
| 1151 |  | m2v.x() = 0; | 
| 1152 |  | } else if (m2v.x() < 0) { | 
| 1164 |  | } else if (m2v.z() < 0) { | 
| 1165 |  | m2v.z() = nCells_.z() - 1; | 
| 1166 |  | } | 
| 1167 | < |  | 
| 1167 | > |  | 
| 1168 |  | int m2 = Vlinear (m2v, nCells_); | 
| 1169 |  |  | 
| 1170 |  | #ifdef IS_MPI | 
| 1173 |  | for (vector<int>::iterator j2 = cellListCol_[m2].begin(); | 
| 1174 |  | j2 != cellListCol_[m2].end(); ++j2) { | 
| 1175 |  |  | 
| 1176 | < | // Always do this if we're in different cells or if | 
| 1177 | < | // we're in the same cell and the global index of the | 
| 1178 | < | // j2 cutoff group is less than the j1 cutoff group | 
| 1179 | < |  | 
| 1180 | < | if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) { | 
| 1181 | < | dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)]; | 
| 1182 | < | snap_->wrapVector(dr); | 
| 1183 | < | cuts = getGroupCutoffs( (*j1), (*j2) ); | 
| 1184 | < | if (dr.lengthSquare() < cuts.third) { | 
| 1110 | < | neighborList.push_back(make_pair((*j1), (*j2))); | 
| 1111 | < | } | 
| 1112 | < | } | 
| 1176 | > | // In parallel, we need to visit *all* pairs of row | 
| 1177 | > | // & column indicies and will divide labor in the | 
| 1178 | > | // force evaluation later. | 
| 1179 | > | dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)]; | 
| 1180 | > | snap_->wrapVector(dr); | 
| 1181 | > | cuts = getGroupCutoffs( (*j1), (*j2) ); | 
| 1182 | > | if (dr.lengthSquare() < cuts.third) { | 
| 1183 | > | neighborList.push_back(make_pair((*j1), (*j2))); | 
| 1184 | > | } | 
| 1185 |  | } | 
| 1186 |  | } | 
| 1187 |  | #else |