--- branches/development/src/parallel/ForceMatrixDecomposition.cpp 2011/06/17 20:16:35 1584 +++ trunk/src/parallel/ForceMatrixDecomposition.cpp 2015/03/03 16:24:44 2061 @@ -35,8 +35,9 @@ * * [1] Meineke, et al., J. Comp. Chem. 26, 252-271 (2005). * [2] Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006). - * [3] Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008). - * [4] Vardeman & Gezelter, in progress (2009). + * [3] Sun, Lin & Gezelter, J. Chem. Phys. 128, 234107 (2008). + * [4] Kuang & Gezelter, J. Chem. Phys. 133, 164101 (2010). + * [5] Vardeman, Stocker & Gezelter, J. Chem. Theory Comput. 7, 834 (2011). */ #include "parallel/ForceMatrixDecomposition.hpp" #include "math/SquareMatrix3.hpp" @@ -47,55 +48,98 @@ namespace OpenMD { using namespace std; namespace OpenMD { + ForceMatrixDecomposition::ForceMatrixDecomposition(SimInfo* info, InteractionManager* iMan) : ForceDecomposition(info, iMan) { + + // Row and colum scans must visit all surrounding cells + cellOffsets_.clear(); + cellOffsets_.push_back( Vector3i(-1,-1,-1) ); + cellOffsets_.push_back( Vector3i( 0,-1,-1) ); + cellOffsets_.push_back( Vector3i( 1,-1,-1) ); + cellOffsets_.push_back( Vector3i(-1, 0,-1) ); + cellOffsets_.push_back( Vector3i( 0, 0,-1) ); + cellOffsets_.push_back( Vector3i( 1, 0,-1) ); + cellOffsets_.push_back( Vector3i(-1, 1,-1) ); + cellOffsets_.push_back( Vector3i( 0, 1,-1) ); + cellOffsets_.push_back( Vector3i( 1, 1,-1) ); + cellOffsets_.push_back( Vector3i(-1,-1, 0) ); + cellOffsets_.push_back( Vector3i( 0,-1, 0) ); + cellOffsets_.push_back( Vector3i( 1,-1, 0) ); + cellOffsets_.push_back( Vector3i(-1, 0, 0) ); + cellOffsets_.push_back( Vector3i( 0, 0, 0) ); + cellOffsets_.push_back( Vector3i( 1, 0, 0) ); + cellOffsets_.push_back( Vector3i(-1, 1, 0) ); + cellOffsets_.push_back( Vector3i( 0, 1, 0) ); + cellOffsets_.push_back( Vector3i( 1, 1, 0) ); + cellOffsets_.push_back( Vector3i(-1,-1, 1) ); + cellOffsets_.push_back( Vector3i( 0,-1, 1) ); + cellOffsets_.push_back( Vector3i( 1,-1, 1) ); + cellOffsets_.push_back( Vector3i(-1, 0, 1) ); + cellOffsets_.push_back( Vector3i( 0, 0, 1) ); + cellOffsets_.push_back( Vector3i( 1, 0, 1) ); + cellOffsets_.push_back( Vector3i(-1, 1, 1) ); + cellOffsets_.push_back( Vector3i( 0, 1, 1) ); + cellOffsets_.push_back( Vector3i( 1, 1, 1) ); + } + + /** * distributeInitialData is essentially a copy of the older fortran * SimulationSetup */ - void ForceMatrixDecomposition::distributeInitialData() { snap_ = sman_->getCurrentSnapshot(); storageLayout_ = sman_->getStorageLayout(); ff_ = info_->getForceField(); nLocal_ = snap_->getNumberOfAtoms(); - + nGroups_ = info_->getNLocalCutoffGroups(); - cerr << "in dId, nGroups = " << nGroups_ << "\n"; // gather the information for atomtype IDs (atids): idents = info_->getIdentArray(); + regions = info_->getRegions(); AtomLocalToGlobal = info_->getGlobalAtomIndices(); cgLocalToGlobal = info_->getGlobalGroupIndices(); vector globalGroupMembership = info_->getGlobalGroupMembership(); + massFactors = info_->getMassFactors(); - PairList excludes = info_->getExcludedInteractions(); - PairList oneTwo = info_->getOneTwoInteractions(); - PairList oneThree = info_->getOneThreeInteractions(); - PairList oneFour = info_->getOneFourInteractions(); - + PairList* excludes = info_->getExcludedInteractions(); + PairList* oneTwo = info_->getOneTwoInteractions(); + PairList* oneThree = info_->getOneThreeInteractions(); + PairList* oneFour = info_->getOneFourInteractions(); + + if (needVelocities_) + snap_->cgData.setStorageLayout(DataStorage::dslPosition | + DataStorage::dslVelocity); + else + snap_->cgData.setStorageLayout(DataStorage::dslPosition); + #ifdef IS_MPI - AtomCommIntRow = new Communicator(nLocal_); - AtomCommRealRow = new Communicator(nLocal_); - AtomCommVectorRow = new Communicator(nLocal_); - AtomCommMatrixRow = new Communicator(nLocal_); - AtomCommPotRow = new Communicator(nLocal_); + MPI_Comm row = rowComm.getComm(); + MPI_Comm col = colComm.getComm(); - AtomCommIntColumn = new Communicator(nLocal_); - AtomCommRealColumn = new Communicator(nLocal_); - AtomCommVectorColumn = new Communicator(nLocal_); - AtomCommMatrixColumn = new Communicator(nLocal_); - AtomCommPotColumn = new Communicator(nLocal_); + AtomPlanIntRow = new Plan(row, nLocal_); + AtomPlanRealRow = new Plan(row, nLocal_); + AtomPlanVectorRow = new Plan(row, nLocal_); + AtomPlanMatrixRow = new Plan(row, nLocal_); + AtomPlanPotRow = new Plan(row, nLocal_); - cgCommIntRow = new Communicator(nGroups_); - cgCommVectorRow = new Communicator(nGroups_); - cgCommIntColumn = new Communicator(nGroups_); - cgCommVectorColumn = new Communicator(nGroups_); + AtomPlanIntColumn = new Plan(col, nLocal_); + AtomPlanRealColumn = new Plan(col, nLocal_); + AtomPlanVectorColumn = new Plan(col, nLocal_); + AtomPlanMatrixColumn = new Plan(col, nLocal_); + AtomPlanPotColumn = new Plan(col, nLocal_); - nAtomsInRow_ = AtomCommIntRow->getSize(); - nAtomsInCol_ = AtomCommIntColumn->getSize(); - nGroupsInRow_ = cgCommIntRow->getSize(); - nGroupsInCol_ = cgCommIntColumn->getSize(); + cgPlanIntRow = new Plan(row, nGroups_); + cgPlanVectorRow = new Plan(row, nGroups_); + cgPlanIntColumn = new Plan(col, nGroups_); + cgPlanVectorColumn = new Plan(col, nGroups_); + nAtomsInRow_ = AtomPlanIntRow->getSize(); + nAtomsInCol_ = AtomPlanIntColumn->getSize(); + nGroupsInRow_ = cgPlanIntRow->getSize(); + nGroupsInCol_ = cgPlanIntColumn->getSize(); + // Modify the data storage objects with the correct layouts and sizes: atomRowData.resize(nAtomsInRow_); atomRowData.setStorageLayout(storageLayout_); @@ -104,23 +148,55 @@ namespace OpenMD { cgRowData.resize(nGroupsInRow_); cgRowData.setStorageLayout(DataStorage::dslPosition); cgColData.resize(nGroupsInCol_); - cgColData.setStorageLayout(DataStorage::dslPosition); - + if (needVelocities_) + // we only need column velocities if we need them. + cgColData.setStorageLayout(DataStorage::dslPosition | + DataStorage::dslVelocity); + else + cgColData.setStorageLayout(DataStorage::dslPosition); + identsRow.resize(nAtomsInRow_); identsCol.resize(nAtomsInCol_); - AtomCommIntRow->gather(idents, identsRow); - AtomCommIntColumn->gather(idents, identsCol); - - AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal); - AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal); + AtomPlanIntRow->gather(idents, identsRow); + AtomPlanIntColumn->gather(idents, identsCol); + + regionsRow.resize(nAtomsInRow_); + regionsCol.resize(nAtomsInCol_); - cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal); - cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal); + AtomPlanIntRow->gather(regions, regionsRow); + AtomPlanIntColumn->gather(regions, regionsCol); + + // allocate memory for the parallel objects + atypesRow.resize(nAtomsInRow_); + atypesCol.resize(nAtomsInCol_); - AtomCommRealRow->gather(massFactors, massFactorsRow); - AtomCommRealColumn->gather(massFactors, massFactorsCol); + for (int i = 0; i < nAtomsInRow_; i++) + atypesRow[i] = ff_->getAtomType(identsRow[i]); + for (int i = 0; i < nAtomsInCol_; i++) + atypesCol[i] = ff_->getAtomType(identsCol[i]); + pot_row.resize(nAtomsInRow_); + pot_col.resize(nAtomsInCol_); + + expot_row.resize(nAtomsInRow_); + expot_col.resize(nAtomsInCol_); + + AtomRowToGlobal.resize(nAtomsInRow_); + AtomColToGlobal.resize(nAtomsInCol_); + AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal); + AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal); + + cgRowToGlobal.resize(nGroupsInRow_); + cgColToGlobal.resize(nGroupsInCol_); + cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal); + cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal); + + massFactorsRow.resize(nAtomsInRow_); + massFactorsCol.resize(nAtomsInCol_); + AtomPlanRealRow->gather(massFactors, massFactorsRow); + AtomPlanRealColumn->gather(massFactors, massFactorsCol); + groupListRow_.clear(); groupListRow_.resize(nGroupsInRow_); for (int i = 0; i < nGroupsInRow_; i++) { @@ -143,8 +219,8 @@ namespace OpenMD { } } - skipsForAtom.clear(); - skipsForAtom.resize(nAtomsInRow_); + excludesForAtom.clear(); + excludesForAtom.resize(nAtomsInRow_); toposForAtom.clear(); toposForAtom.resize(nAtomsInRow_); topoDist.clear(); @@ -155,18 +231,18 @@ namespace OpenMD { for (int j = 0; j < nAtomsInCol_; j++) { int jglob = AtomColToGlobal[j]; - if (excludes.hasPair(iglob, jglob)) - skipsForAtom[i].push_back(j); + if (excludes->hasPair(iglob, jglob)) + excludesForAtom[i].push_back(j); - if (oneTwo.hasPair(iglob, jglob)) { + if (oneTwo->hasPair(iglob, jglob)) { toposForAtom[i].push_back(j); topoDist[i].push_back(1); } else { - if (oneThree.hasPair(iglob, jglob)) { + if (oneThree->hasPair(iglob, jglob)) { toposForAtom[i].push_back(j); topoDist[i].push_back(2); } else { - if (oneFour.hasPair(iglob, jglob)) { + if (oneFour->hasPair(iglob, jglob)) { toposForAtom[i].push_back(j); topoDist[i].push_back(3); } @@ -175,22 +251,9 @@ namespace OpenMD { } } -#endif - - groupList_.clear(); - groupList_.resize(nGroups_); - for (int i = 0; i < nGroups_; i++) { - int gid = cgLocalToGlobal[i]; - for (int j = 0; j < nLocal_; j++) { - int aid = AtomLocalToGlobal[j]; - if (globalGroupMembership[aid] == gid) { - groupList_[i].push_back(j); - } - } - } - - skipsForAtom.clear(); - skipsForAtom.resize(nLocal_); +#else + excludesForAtom.clear(); + excludesForAtom.resize(nLocal_); toposForAtom.clear(); toposForAtom.resize(nLocal_); topoDist.clear(); @@ -202,18 +265,18 @@ namespace OpenMD { for (int j = 0; j < nLocal_; j++) { int jglob = AtomLocalToGlobal[j]; - if (excludes.hasPair(iglob, jglob)) - skipsForAtom[i].push_back(j); + if (excludes->hasPair(iglob, jglob)) + excludesForAtom[i].push_back(j); - if (oneTwo.hasPair(iglob, jglob)) { + if (oneTwo->hasPair(iglob, jglob)) { toposForAtom[i].push_back(j); topoDist[i].push_back(1); } else { - if (oneThree.hasPair(iglob, jglob)) { + if (oneThree->hasPair(iglob, jglob)) { toposForAtom[i].push_back(j); topoDist[i].push_back(2); } else { - if (oneFour.hasPair(iglob, jglob)) { + if (oneFour->hasPair(iglob, jglob)) { toposForAtom[i].push_back(j); topoDist[i].push_back(3); } @@ -221,201 +284,40 @@ namespace OpenMD { } } } - - createGtypeCutoffMap(); - } - - void ForceMatrixDecomposition::createGtypeCutoffMap() { +#endif - RealType tol = 1e-6; - RealType rc; - int atid; - set atypes = info_->getSimulatedAtomTypes(); - vector atypeCutoff; - atypeCutoff.resize( atypes.size() ); - - for (set::iterator at = atypes.begin(); - at != atypes.end(); ++at){ - atid = (*at)->getIdent(); + // allocate memory for the parallel objects + atypesLocal.resize(nLocal_); - if (userChoseCutoff_) - atypeCutoff[atid] = userCutoff_; - else - atypeCutoff[atid] = interactionMan_->getSuggestedCutoffRadius(*at); - } + for (int i = 0; i < nLocal_; i++) + atypesLocal[i] = ff_->getAtomType(idents[i]); - vector gTypeCutoffs; - - // first we do a single loop over the cutoff groups to find the - // largest cutoff for any atypes present in this group. -#ifdef IS_MPI - vector groupCutoffRow(nGroupsInRow_, 0.0); - groupRowToGtype.resize(nGroupsInRow_); - for (int cg1 = 0; cg1 < nGroupsInRow_; cg1++) { - vector atomListRow = getAtomsInGroupRow(cg1); - for (vector::iterator ia = atomListRow.begin(); - ia != atomListRow.end(); ++ia) { - int atom1 = (*ia); - atid = identsRow[atom1]; - if (atypeCutoff[atid] > groupCutoffRow[cg1]) { - groupCutoffRow[cg1] = atypeCutoff[atid]; + groupList_.clear(); + groupList_.resize(nGroups_); + for (int i = 0; i < nGroups_; i++) { + int gid = cgLocalToGlobal[i]; + for (int j = 0; j < nLocal_; j++) { + int aid = AtomLocalToGlobal[j]; + if (globalGroupMembership[aid] == gid) { + groupList_[i].push_back(j); } - } - - bool gTypeFound = false; - for (int gt = 0; gt < gTypeCutoffs.size(); gt++) { - if (abs(groupCutoffRow[cg1] - gTypeCutoffs[gt]) < tol) { - groupRowToGtype[cg1] = gt; - gTypeFound = true; - } - } - if (!gTypeFound) { - gTypeCutoffs.push_back( groupCutoffRow[cg1] ); - groupRowToGtype[cg1] = gTypeCutoffs.size() - 1; - } - - } - vector groupCutoffCol(nGroupsInCol_, 0.0); - groupColToGtype.resize(nGroupsInCol_); - for (int cg2 = 0; cg2 < nGroupsInCol_; cg2++) { - vector atomListCol = getAtomsInGroupColumn(cg2); - for (vector::iterator jb = atomListCol.begin(); - jb != atomListCol.end(); ++jb) { - int atom2 = (*jb); - atid = identsCol[atom2]; - if (atypeCutoff[atid] > groupCutoffCol[cg2]) { - groupCutoffCol[cg2] = atypeCutoff[atid]; - } - } - bool gTypeFound = false; - for (int gt = 0; gt < gTypeCutoffs.size(); gt++) { - if (abs(groupCutoffCol[cg2] - gTypeCutoffs[gt]) < tol) { - groupColToGtype[cg2] = gt; - gTypeFound = true; - } - } - if (!gTypeFound) { - gTypeCutoffs.push_back( groupCutoffCol[cg2] ); - groupColToGtype[cg2] = gTypeCutoffs.size() - 1; - } - } -#else - - vector groupCutoff(nGroups_, 0.0); - groupToGtype.resize(nGroups_); - - cerr << "nGroups = " << nGroups_ << "\n"; - for (int cg1 = 0; cg1 < nGroups_; cg1++) { - - groupCutoff[cg1] = 0.0; - vector atomList = getAtomsInGroupRow(cg1); - - for (vector::iterator ia = atomList.begin(); - ia != atomList.end(); ++ia) { - int atom1 = (*ia); - atid = idents[atom1]; - if (atypeCutoff[atid] > groupCutoff[cg1]) { - groupCutoff[cg1] = atypeCutoff[atid]; - } - } - - bool gTypeFound = false; - for (int gt = 0; gt < gTypeCutoffs.size(); gt++) { - if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) { - groupToGtype[cg1] = gt; - gTypeFound = true; - } - } - if (!gTypeFound) { - gTypeCutoffs.push_back( groupCutoff[cg1] ); - groupToGtype[cg1] = gTypeCutoffs.size() - 1; } - } -#endif - - cerr << "gTypeCutoffs.size() = " << gTypeCutoffs.size() << "\n"; - // Now we find the maximum group cutoff value present in the simulation - - RealType groupMax = *max_element(gTypeCutoffs.begin(), gTypeCutoffs.end()); - -#ifdef IS_MPI - MPI::COMM_WORLD.Allreduce(&groupMax, &groupMax, 1, MPI::REALTYPE, MPI::MAX); -#endif - - RealType tradRcut = groupMax; - - for (int i = 0; i < gTypeCutoffs.size(); i++) { - for (int j = 0; j < gTypeCutoffs.size(); j++) { - RealType thisRcut; - switch(cutoffPolicy_) { - case TRADITIONAL: - thisRcut = tradRcut; - break; - case MIX: - thisRcut = 0.5 * (gTypeCutoffs[i] + gTypeCutoffs[j]); - break; - case MAX: - thisRcut = max(gTypeCutoffs[i], gTypeCutoffs[j]); - break; - default: - sprintf(painCave.errMsg, - "ForceMatrixDecomposition::createGtypeCutoffMap " - "hit an unknown cutoff policy!\n"); - painCave.severity = OPENMD_ERROR; - painCave.isFatal = 1; - simError(); - break; - } - - pair key = make_pair(i,j); - gTypeCutoffMap[key].first = thisRcut; - - if (thisRcut > largestRcut_) largestRcut_ = thisRcut; - - gTypeCutoffMap[key].second = thisRcut*thisRcut; - - gTypeCutoffMap[key].third = pow(thisRcut + skinThickness_, 2); - - // sanity check - - if (userChoseCutoff_) { - if (abs(gTypeCutoffMap[key].first - userCutoff_) > 0.0001) { - sprintf(painCave.errMsg, - "ForceMatrixDecomposition::createGtypeCutoffMap " - "user-specified rCut (%lf) does not match computed group Cutoff\n", userCutoff_); - painCave.severity = OPENMD_ERROR; - painCave.isFatal = 1; - simError(); - } - } - } - } + } } - - - groupCutoffs ForceMatrixDecomposition::getGroupCutoffs(int cg1, int cg2) { - int i, j; -#ifdef IS_MPI - i = groupRowToGtype[cg1]; - j = groupColToGtype[cg2]; -#else - i = groupToGtype[cg1]; - j = groupToGtype[cg2]; -#endif - return gTypeCutoffMap[make_pair(i,j)]; - } - + int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) { - for (int j = 0; j < toposForAtom[atom1].size(); j++) { + for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) { if (toposForAtom[atom1][j] == atom2) return topoDist[atom1][j]; - } + } return 0; } void ForceMatrixDecomposition::zeroWorkArrays() { pairwisePot = 0.0; embeddingPot = 0.0; + excludedPot = 0.0; + excludedSelfPot = 0.0; #ifdef IS_MPI if (storageLayout_ & DataStorage::dslForce) { @@ -434,9 +336,17 @@ namespace OpenMD { fill(pot_col.begin(), pot_col.end(), Vector (0.0)); + fill(expot_row.begin(), expot_row.end(), + Vector (0.0)); + + fill(expot_col.begin(), expot_col.end(), + Vector (0.0)); + if (storageLayout_ & DataStorage::dslParticlePot) { - fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), 0.0); - fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), 0.0); + fill(atomRowData.particlePot.begin(), atomRowData.particlePot.end(), + 0.0); + fill(atomColData.particlePot.begin(), atomColData.particlePot.end(), + 0.0); } if (storageLayout_ & DataStorage::dslDensity) { @@ -445,8 +355,10 @@ namespace OpenMD { } if (storageLayout_ & DataStorage::dslFunctional) { - fill(atomRowData.functional.begin(), atomRowData.functional.end(), 0.0); - fill(atomColData.functional.begin(), atomColData.functional.end(), 0.0); + fill(atomRowData.functional.begin(), atomRowData.functional.end(), + 0.0); + fill(atomColData.functional.begin(), atomColData.functional.end(), + 0.0); } if (storageLayout_ & DataStorage::dslFunctionalDerivative) { @@ -456,8 +368,37 @@ namespace OpenMD { atomColData.functionalDerivative.end(), 0.0); } -#else - + if (storageLayout_ & DataStorage::dslSkippedCharge) { + fill(atomRowData.skippedCharge.begin(), + atomRowData.skippedCharge.end(), 0.0); + fill(atomColData.skippedCharge.begin(), + atomColData.skippedCharge.end(), 0.0); + } + + if (storageLayout_ & DataStorage::dslFlucQForce) { + fill(atomRowData.flucQFrc.begin(), + atomRowData.flucQFrc.end(), 0.0); + fill(atomColData.flucQFrc.begin(), + atomColData.flucQFrc.end(), 0.0); + } + + if (storageLayout_ & DataStorage::dslElectricField) { + fill(atomRowData.electricField.begin(), + atomRowData.electricField.end(), V3Zero); + fill(atomColData.electricField.begin(), + atomColData.electricField.end(), V3Zero); + } + + if (storageLayout_ & DataStorage::dslSitePotential) { + fill(atomRowData.sitePotential.begin(), + atomRowData.sitePotential.end(), 0.0); + fill(atomColData.sitePotential.begin(), + atomColData.sitePotential.end(), 0.0); + } + +#endif + // even in parallel, we need to zero out the local arrays: + if (storageLayout_ & DataStorage::dslParticlePot) { fill(snap_->atomData.particlePot.begin(), snap_->atomData.particlePot.end(), 0.0); @@ -467,51 +408,104 @@ namespace OpenMD { fill(snap_->atomData.density.begin(), snap_->atomData.density.end(), 0.0); } + if (storageLayout_ & DataStorage::dslFunctional) { fill(snap_->atomData.functional.begin(), snap_->atomData.functional.end(), 0.0); } + if (storageLayout_ & DataStorage::dslFunctionalDerivative) { fill(snap_->atomData.functionalDerivative.begin(), snap_->atomData.functionalDerivative.end(), 0.0); } -#endif - + + if (storageLayout_ & DataStorage::dslSkippedCharge) { + fill(snap_->atomData.skippedCharge.begin(), + snap_->atomData.skippedCharge.end(), 0.0); + } + + if (storageLayout_ & DataStorage::dslElectricField) { + fill(snap_->atomData.electricField.begin(), + snap_->atomData.electricField.end(), V3Zero); + } + if (storageLayout_ & DataStorage::dslSitePotential) { + fill(snap_->atomData.sitePotential.begin(), + snap_->atomData.sitePotential.end(), 0.0); + } } void ForceMatrixDecomposition::distributeData() { + +#ifdef IS_MPI + snap_ = sman_->getCurrentSnapshot(); storageLayout_ = sman_->getStorageLayout(); -#ifdef IS_MPI + bool needsCG = true; + if(info_->getNCutoffGroups() != info_->getNAtoms()) + needsCG = false; + // gather up the atomic positions - AtomCommVectorRow->gather(snap_->atomData.position, + AtomPlanVectorRow->gather(snap_->atomData.position, atomRowData.position); - AtomCommVectorColumn->gather(snap_->atomData.position, + AtomPlanVectorColumn->gather(snap_->atomData.position, atomColData.position); // gather up the cutoff group positions - cgCommVectorRow->gather(snap_->cgData.position, - cgRowData.position); - cgCommVectorColumn->gather(snap_->cgData.position, - cgColData.position); + + if (needsCG) { + cgPlanVectorRow->gather(snap_->cgData.position, + cgRowData.position); + + cgPlanVectorColumn->gather(snap_->cgData.position, + cgColData.position); + } + + + if (needVelocities_) { + // gather up the atomic velocities + AtomPlanVectorColumn->gather(snap_->atomData.velocity, + atomColData.velocity); + + if (needsCG) { + cgPlanVectorColumn->gather(snap_->cgData.velocity, + cgColData.velocity); + } + } + // if needed, gather the atomic rotation matrices if (storageLayout_ & DataStorage::dslAmat) { - AtomCommMatrixRow->gather(snap_->atomData.aMat, + AtomPlanMatrixRow->gather(snap_->atomData.aMat, atomRowData.aMat); - AtomCommMatrixColumn->gather(snap_->atomData.aMat, + AtomPlanMatrixColumn->gather(snap_->atomData.aMat, atomColData.aMat); } - - // if needed, gather the atomic eletrostatic frames - if (storageLayout_ & DataStorage::dslElectroFrame) { - AtomCommMatrixRow->gather(snap_->atomData.electroFrame, - atomRowData.electroFrame); - AtomCommMatrixColumn->gather(snap_->atomData.electroFrame, - atomColData.electroFrame); + + // if needed, gather the atomic eletrostatic information + if (storageLayout_ & DataStorage::dslDipole) { + AtomPlanVectorRow->gather(snap_->atomData.dipole, + atomRowData.dipole); + AtomPlanVectorColumn->gather(snap_->atomData.dipole, + atomColData.dipole); } + + if (storageLayout_ & DataStorage::dslQuadrupole) { + AtomPlanMatrixRow->gather(snap_->atomData.quadrupole, + atomRowData.quadrupole); + AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole, + atomColData.quadrupole); + } + + // if needed, gather the atomic fluctuating charge values + if (storageLayout_ & DataStorage::dslFlucQPosition) { + AtomPlanRealRow->gather(snap_->atomData.flucQPos, + atomRowData.flucQPos); + AtomPlanRealColumn->gather(snap_->atomData.flucQPos, + atomColData.flucQPos); + } + #endif } @@ -525,15 +519,30 @@ namespace OpenMD { if (storageLayout_ & DataStorage::dslDensity) { - AtomCommRealRow->scatter(atomRowData.density, + AtomPlanRealRow->scatter(atomRowData.density, snap_->atomData.density); int n = snap_->atomData.density.size(); vector rho_tmp(n, 0.0); - AtomCommRealColumn->scatter(atomColData.density, rho_tmp); + AtomPlanRealColumn->scatter(atomColData.density, rho_tmp); for (int i = 0; i < n; i++) snap_->atomData.density[i] += rho_tmp[i]; } + + // this isn't necessary if we don't have polarizable atoms, but + // we'll leave it here for now. + if (storageLayout_ & DataStorage::dslElectricField) { + + AtomPlanVectorRow->scatter(atomRowData.electricField, + snap_->atomData.electricField); + + int n = snap_->atomData.electricField.size(); + vector field_tmp(n, V3Zero); + AtomPlanVectorColumn->scatter(atomColData.electricField, + field_tmp); + for (int i = 0; i < n; i++) + snap_->atomData.electricField[i] += field_tmp[i]; + } #endif } @@ -546,16 +555,16 @@ namespace OpenMD { storageLayout_ = sman_->getStorageLayout(); #ifdef IS_MPI if (storageLayout_ & DataStorage::dslFunctional) { - AtomCommRealRow->gather(snap_->atomData.functional, + AtomPlanRealRow->gather(snap_->atomData.functional, atomRowData.functional); - AtomCommRealColumn->gather(snap_->atomData.functional, + AtomPlanRealColumn->gather(snap_->atomData.functional, atomColData.functional); } if (storageLayout_ & DataStorage::dslFunctionalDerivative) { - AtomCommRealRow->gather(snap_->atomData.functionalDerivative, + AtomPlanRealRow->gather(snap_->atomData.functionalDerivative, atomRowData.functionalDerivative); - AtomCommRealColumn->gather(snap_->atomData.functionalDerivative, + AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative, atomColData.functionalDerivative); } #endif @@ -569,58 +578,234 @@ namespace OpenMD { int n = snap_->atomData.force.size(); vector frc_tmp(n, V3Zero); - AtomCommVectorRow->scatter(atomRowData.force, frc_tmp); + AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp); for (int i = 0; i < n; i++) { snap_->atomData.force[i] += frc_tmp[i]; frc_tmp[i] = 0.0; } - AtomCommVectorColumn->scatter(atomColData.force, frc_tmp); - for (int i = 0; i < n; i++) + AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp); + for (int i = 0; i < n; i++) { snap_->atomData.force[i] += frc_tmp[i]; - - + } + if (storageLayout_ & DataStorage::dslTorque) { - int nt = snap_->atomData.force.size(); + int nt = snap_->atomData.torque.size(); vector trq_tmp(nt, V3Zero); - AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp); - for (int i = 0; i < n; i++) { + AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp); + for (int i = 0; i < nt; i++) { snap_->atomData.torque[i] += trq_tmp[i]; trq_tmp[i] = 0.0; } - AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp); - for (int i = 0; i < n; i++) + AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp); + for (int i = 0; i < nt; i++) snap_->atomData.torque[i] += trq_tmp[i]; } + + if (storageLayout_ & DataStorage::dslSkippedCharge) { + + int ns = snap_->atomData.skippedCharge.size(); + vector skch_tmp(ns, 0.0); + + AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp); + for (int i = 0; i < ns; i++) { + snap_->atomData.skippedCharge[i] += skch_tmp[i]; + skch_tmp[i] = 0.0; + } + + AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp); + for (int i = 0; i < ns; i++) + snap_->atomData.skippedCharge[i] += skch_tmp[i]; + + } + if (storageLayout_ & DataStorage::dslFlucQForce) { + + int nq = snap_->atomData.flucQFrc.size(); + vector fqfrc_tmp(nq, 0.0); + + AtomPlanRealRow->scatter(atomRowData.flucQFrc, fqfrc_tmp); + for (int i = 0; i < nq; i++) { + snap_->atomData.flucQFrc[i] += fqfrc_tmp[i]; + fqfrc_tmp[i] = 0.0; + } + + AtomPlanRealColumn->scatter(atomColData.flucQFrc, fqfrc_tmp); + for (int i = 0; i < nq; i++) + snap_->atomData.flucQFrc[i] += fqfrc_tmp[i]; + + } + + if (storageLayout_ & DataStorage::dslElectricField) { + + int nef = snap_->atomData.electricField.size(); + vector efield_tmp(nef, V3Zero); + + AtomPlanVectorRow->scatter(atomRowData.electricField, efield_tmp); + for (int i = 0; i < nef; i++) { + snap_->atomData.electricField[i] += efield_tmp[i]; + efield_tmp[i] = 0.0; + } + + AtomPlanVectorColumn->scatter(atomColData.electricField, efield_tmp); + for (int i = 0; i < nef; i++) + snap_->atomData.electricField[i] += efield_tmp[i]; + } + + if (storageLayout_ & DataStorage::dslSitePotential) { + + int nsp = snap_->atomData.sitePotential.size(); + vector sp_tmp(nsp, 0.0); + + AtomPlanRealRow->scatter(atomRowData.sitePotential, sp_tmp); + for (int i = 0; i < nsp; i++) { + snap_->atomData.sitePotential[i] += sp_tmp[i]; + sp_tmp[i] = 0.0; + } + + AtomPlanRealColumn->scatter(atomColData.sitePotential, sp_tmp); + for (int i = 0; i < nsp; i++) + snap_->atomData.sitePotential[i] += sp_tmp[i]; + } + nLocal_ = snap_->getNumberOfAtoms(); vector pot_temp(nLocal_, Vector (0.0)); + vector expot_temp(nLocal_, + Vector (0.0)); // scatter/gather pot_row into the members of my column - AtomCommPotRow->scatter(pot_row, pot_temp); + AtomPlanPotRow->scatter(pot_row, pot_temp); + AtomPlanPotRow->scatter(expot_row, expot_temp); - for (int ii = 0; ii < pot_temp.size(); ii++ ) + for (int ii = 0; ii < pot_temp.size(); ii++ ) pairwisePot += pot_temp[ii]; - + + for (int ii = 0; ii < expot_temp.size(); ii++ ) + excludedPot += expot_temp[ii]; + + if (storageLayout_ & DataStorage::dslParticlePot) { + // This is the pairwise contribution to the particle pot. The + // embedding contribution is added in each of the low level + // non-bonded routines. In single processor, this is done in + // unpackInteractionData, not in collectData. + for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { + for (int i = 0; i < nLocal_; i++) { + // factor of two is because the total potential terms are divided + // by 2 in parallel due to row/ column scatter + snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii); + } + } + } + fill(pot_temp.begin(), pot_temp.end(), Vector (0.0)); + fill(expot_temp.begin(), expot_temp.end(), + Vector (0.0)); - AtomCommPotColumn->scatter(pot_col, pot_temp); + AtomPlanPotColumn->scatter(pot_col, pot_temp); + AtomPlanPotColumn->scatter(expot_col, expot_temp); for (int ii = 0; ii < pot_temp.size(); ii++ ) pairwisePot += pot_temp[ii]; + + for (int ii = 0; ii < expot_temp.size(); ii++ ) + excludedPot += expot_temp[ii]; + + if (storageLayout_ & DataStorage::dslParticlePot) { + // This is the pairwise contribution to the particle pot. The + // embedding contribution is added in each of the low level + // non-bonded routines. In single processor, this is done in + // unpackInteractionData, not in collectData. + for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { + for (int i = 0; i < nLocal_; i++) { + // factor of two is because the total potential terms are divided + // by 2 in parallel due to row/ column scatter + snap_->atomData.particlePot[i] += 2.0 * pot_temp[i](ii); + } + } + } + + if (storageLayout_ & DataStorage::dslParticlePot) { + int npp = snap_->atomData.particlePot.size(); + vector ppot_temp(npp, 0.0); + + // This is the direct or embedding contribution to the particle + // pot. + + AtomPlanRealRow->scatter(atomRowData.particlePot, ppot_temp); + for (int i = 0; i < npp; i++) { + snap_->atomData.particlePot[i] += ppot_temp[i]; + } + + fill(ppot_temp.begin(), ppot_temp.end(), 0.0); + + AtomPlanRealColumn->scatter(atomColData.particlePot, ppot_temp); + for (int i = 0; i < npp; i++) { + snap_->atomData.particlePot[i] += ppot_temp[i]; + } + } + + for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { + RealType ploc1 = pairwisePot[ii]; + RealType ploc2 = 0.0; + MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD); + pairwisePot[ii] = ploc2; + } + + for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { + RealType ploc1 = excludedPot[ii]; + RealType ploc2 = 0.0; + MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD); + excludedPot[ii] = ploc2; + } + + // Here be dragons. + MPI_Comm col = colComm.getComm(); + + MPI_Allreduce(MPI_IN_PLACE, + &snap_->frameData.conductiveHeatFlux[0], 3, + MPI_REALTYPE, MPI_SUM, col); + + #endif } - int ForceMatrixDecomposition::getNAtomsInRow() { + /** + * Collects information obtained during the post-pair (and embedding + * functional) loops onto local data structures. + */ + void ForceMatrixDecomposition::collectSelfData() { + snap_ = sman_->getCurrentSnapshot(); + storageLayout_ = sman_->getStorageLayout(); + #ifdef IS_MPI + for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { + RealType ploc1 = embeddingPot[ii]; + RealType ploc2 = 0.0; + MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD); + embeddingPot[ii] = ploc2; + } + for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { + RealType ploc1 = excludedSelfPot[ii]; + RealType ploc2 = 0.0; + MPI_Allreduce(&ploc1, &ploc2, 1, MPI_REALTYPE, MPI_SUM, MPI_COMM_WORLD); + excludedSelfPot[ii] = ploc2; + } +#endif + + } + + + + int& ForceMatrixDecomposition::getNAtomsInRow() { +#ifdef IS_MPI return nAtomsInRow_; #else return nLocal_; @@ -630,7 +815,7 @@ namespace OpenMD { /** * returns the list of atoms belonging to this group. */ - vector ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){ + vector& ForceMatrixDecomposition::getAtomsInGroupRow(int cg1){ #ifdef IS_MPI return groupListRow_[cg1]; #else @@ -638,7 +823,7 @@ namespace OpenMD { #endif } - vector ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){ + vector& ForceMatrixDecomposition::getAtomsInGroupColumn(int cg2){ #ifdef IS_MPI return groupListCol_[cg2]; #else @@ -655,11 +840,29 @@ namespace OpenMD { d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1]; #endif - snap_->wrapVector(d); + if (usePeriodicBoundaryConditions_) { + snap_->wrapVector(d); + } return d; } + Vector3d& ForceMatrixDecomposition::getGroupVelocityColumn(int cg2){ +#ifdef IS_MPI + return cgColData.velocity[cg2]; +#else + return snap_->cgData.velocity[cg2]; +#endif + } + Vector3d& ForceMatrixDecomposition::getAtomVelocityColumn(int atom2){ +#ifdef IS_MPI + return atomColData.velocity[atom2]; +#else + return snap_->atomData.velocity[atom2]; +#endif + } + + Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){ Vector3d d; @@ -669,8 +872,9 @@ namespace OpenMD { #else d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1]; #endif - - snap_->wrapVector(d); + if (usePeriodicBoundaryConditions_) { + snap_->wrapVector(d); + } return d; } @@ -682,21 +886,21 @@ namespace OpenMD { #else d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2]; #endif - - snap_->wrapVector(d); + if (usePeriodicBoundaryConditions_) { + snap_->wrapVector(d); + } return d; } - RealType ForceMatrixDecomposition::getMassFactorRow(int atom1) { + RealType& ForceMatrixDecomposition::getMassFactorRow(int atom1) { #ifdef IS_MPI return massFactorsRow[atom1]; #else - cerr << "mfs = " << massFactors.size() << " atom1 = " << atom1 << "\n"; return massFactors[atom1]; #endif } - RealType ForceMatrixDecomposition::getMassFactorColumn(int atom2) { + RealType& ForceMatrixDecomposition::getMassFactorColumn(int atom2) { #ifdef IS_MPI return massFactorsCol[atom2]; #else @@ -713,48 +917,73 @@ namespace OpenMD { #else d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1]; #endif - - snap_->wrapVector(d); + if (usePeriodicBoundaryConditions_) { + snap_->wrapVector(d); + } return d; } - vector ForceMatrixDecomposition::getSkipsForAtom(int atom1) { - return skipsForAtom[atom1]; + vector& ForceMatrixDecomposition::getExcludesForAtom(int atom1) { + return excludesForAtom[atom1]; } /** - * There are a number of reasons to skip a pair or a - * particle. Mostly we do this to exclude atoms who are involved in - * short range interactions (bonds, bends, torsions), but we also - * need to exclude some overcounted interactions that result from + * We need to exclude some overcounted interactions that result from * the parallel decomposition. */ - bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) { + bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) { int unique_id_1, unique_id_2; - + #ifdef IS_MPI // in MPI, we have to look up the unique IDs for each atom unique_id_1 = AtomRowToGlobal[atom1]; unique_id_2 = AtomColToGlobal[atom2]; - - // this situation should only arise in MPI simulations + // group1 = cgRowToGlobal[cg1]; + // group2 = cgColToGlobal[cg2]; +#else + unique_id_1 = AtomLocalToGlobal[atom1]; + unique_id_2 = AtomLocalToGlobal[atom2]; + int group1 = cgLocalToGlobal[cg1]; + int group2 = cgLocalToGlobal[cg2]; +#endif + if (unique_id_1 == unique_id_2) return true; - + +#ifdef IS_MPI // this prevents us from doing the pair on multiple processors if (unique_id_1 < unique_id_2) { if ((unique_id_1 + unique_id_2) % 2 == 0) return true; } else { - if ((unique_id_1 + unique_id_2) % 2 == 1) return true; + if ((unique_id_1 + unique_id_2) % 2 == 1) return true; } -#else - // in the normal loop, the atom numbers are unique - unique_id_1 = atom1; - unique_id_2 = atom2; +#endif + +#ifndef IS_MPI + if (group1 == group2) { + if (unique_id_1 < unique_id_2) return true; + } #endif - for (vector::iterator i = skipsForAtom[atom1].begin(); - i != skipsForAtom[atom1].end(); ++i) { - if ( (*i) == unique_id_2 ) return true; + return false; + } + + /** + * We need to handle the interactions for atoms who are involved in + * the same rigid body as well as some short range interactions + * (bonds, bends, torsions) differently from other interactions. + * We'll still visit the pairwise routines, but with a flag that + * tells those routines to exclude the pair from direct long range + * interactions. Some indirect interactions (notably reaction + * field) must still be handled for these pairs. + */ + bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) { + + // excludesForAtom was constructed to use row/column indices in the MPI + // version, and to use local IDs in the non-MPI version: + + for (vector::iterator i = excludesForAtom[atom1].begin(); + i != excludesForAtom[atom1].end(); ++i) { + if ( (*i) == atom2 ) return true; } return false; @@ -779,364 +1008,622 @@ namespace OpenMD { // filling interaction blocks with pointers void ForceMatrixDecomposition::fillInteractionData(InteractionData &idat, - int atom1, int atom2) { + int atom1, int atom2, + bool newAtom1) { + + idat.excluded = excludeAtomPair(atom1, atom2); + + if (newAtom1) { + #ifdef IS_MPI - - idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]), - ff_->getAtomType(identsCol[atom2]) ); - + idat.atid1 = identsRow[atom1]; + idat.atid2 = identsCol[atom2]; + + if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) { + idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]); + } else { + idat.sameRegion = false; + } + + if (storageLayout_ & DataStorage::dslAmat) { + idat.A1 = &(atomRowData.aMat[atom1]); + idat.A2 = &(atomColData.aMat[atom2]); + } + + if (storageLayout_ & DataStorage::dslTorque) { + idat.t1 = &(atomRowData.torque[atom1]); + idat.t2 = &(atomColData.torque[atom2]); + } + + if (storageLayout_ & DataStorage::dslDipole) { + idat.dipole1 = &(atomRowData.dipole[atom1]); + idat.dipole2 = &(atomColData.dipole[atom2]); + } + + if (storageLayout_ & DataStorage::dslQuadrupole) { + idat.quadrupole1 = &(atomRowData.quadrupole[atom1]); + idat.quadrupole2 = &(atomColData.quadrupole[atom2]); + } + + if (storageLayout_ & DataStorage::dslDensity) { + idat.rho1 = &(atomRowData.density[atom1]); + idat.rho2 = &(atomColData.density[atom2]); + } + + if (storageLayout_ & DataStorage::dslFunctional) { + idat.frho1 = &(atomRowData.functional[atom1]); + idat.frho2 = &(atomColData.functional[atom2]); + } + + if (storageLayout_ & DataStorage::dslFunctionalDerivative) { + idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]); + idat.dfrho2 = &(atomColData.functionalDerivative[atom2]); + } + + if (storageLayout_ & DataStorage::dslParticlePot) { + idat.particlePot1 = &(atomRowData.particlePot[atom1]); + idat.particlePot2 = &(atomColData.particlePot[atom2]); + } + + if (storageLayout_ & DataStorage::dslSkippedCharge) { + idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]); + idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]); + } + + if (storageLayout_ & DataStorage::dslFlucQPosition) { + idat.flucQ1 = &(atomRowData.flucQPos[atom1]); + idat.flucQ2 = &(atomColData.flucQPos[atom2]); + } + +#else + + idat.atid1 = idents[atom1]; + idat.atid2 = idents[atom2]; + + if (regions[atom1] >= 0 && regions[atom2] >= 0) { + idat.sameRegion = (regions[atom1] == regions[atom2]); + } else { + idat.sameRegion = false; + } + + if (storageLayout_ & DataStorage::dslAmat) { + idat.A1 = &(snap_->atomData.aMat[atom1]); + idat.A2 = &(snap_->atomData.aMat[atom2]); + } + + if (storageLayout_ & DataStorage::dslTorque) { + idat.t1 = &(snap_->atomData.torque[atom1]); + idat.t2 = &(snap_->atomData.torque[atom2]); + } + + if (storageLayout_ & DataStorage::dslDipole) { + idat.dipole1 = &(snap_->atomData.dipole[atom1]); + idat.dipole2 = &(snap_->atomData.dipole[atom2]); + } + + if (storageLayout_ & DataStorage::dslQuadrupole) { + idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]); + idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]); + } + + if (storageLayout_ & DataStorage::dslDensity) { + idat.rho1 = &(snap_->atomData.density[atom1]); + idat.rho2 = &(snap_->atomData.density[atom2]); + } + + if (storageLayout_ & DataStorage::dslFunctional) { + idat.frho1 = &(snap_->atomData.functional[atom1]); + idat.frho2 = &(snap_->atomData.functional[atom2]); + } + + if (storageLayout_ & DataStorage::dslFunctionalDerivative) { + idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]); + idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]); + } + + if (storageLayout_ & DataStorage::dslParticlePot) { + idat.particlePot1 = &(snap_->atomData.particlePot[atom1]); + idat.particlePot2 = &(snap_->atomData.particlePot[atom2]); + } + + if (storageLayout_ & DataStorage::dslSkippedCharge) { + idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]); + idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]); + } + + if (storageLayout_ & DataStorage::dslFlucQPosition) { + idat.flucQ1 = &(snap_->atomData.flucQPos[atom1]); + idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]); + } +#endif + + } else { + // atom1 is not new, so don't bother updating properties of that atom: +#ifdef IS_MPI + idat.atid2 = identsCol[atom2]; + + if (regionsRow[atom1] >= 0 && regionsCol[atom2] >= 0) { + idat.sameRegion = (regionsRow[atom1] == regionsCol[atom2]); + } else { + idat.sameRegion = false; + } + if (storageLayout_ & DataStorage::dslAmat) { - idat.A1 = &(atomRowData.aMat[atom1]); idat.A2 = &(atomColData.aMat[atom2]); } - if (storageLayout_ & DataStorage::dslElectroFrame) { - idat.eFrame1 = &(atomRowData.electroFrame[atom1]); - idat.eFrame2 = &(atomColData.electroFrame[atom2]); - } - if (storageLayout_ & DataStorage::dslTorque) { - idat.t1 = &(atomRowData.torque[atom1]); idat.t2 = &(atomColData.torque[atom2]); } + if (storageLayout_ & DataStorage::dslDipole) { + idat.dipole2 = &(atomColData.dipole[atom2]); + } + + if (storageLayout_ & DataStorage::dslQuadrupole) { + idat.quadrupole2 = &(atomColData.quadrupole[atom2]); + } + if (storageLayout_ & DataStorage::dslDensity) { - idat.rho1 = &(atomRowData.density[atom1]); idat.rho2 = &(atomColData.density[atom2]); } if (storageLayout_ & DataStorage::dslFunctional) { - idat.frho1 = &(atomRowData.functional[atom1]); idat.frho2 = &(atomColData.functional[atom2]); } if (storageLayout_ & DataStorage::dslFunctionalDerivative) { - idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]); idat.dfrho2 = &(atomColData.functionalDerivative[atom2]); } if (storageLayout_ & DataStorage::dslParticlePot) { - idat.particlePot1 = &(atomRowData.particlePot[atom1]); idat.particlePot2 = &(atomColData.particlePot[atom2]); } -#else + if (storageLayout_ & DataStorage::dslSkippedCharge) { + idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]); + } - idat.atypes = make_pair( ff_->getAtomType(idents[atom1]), - ff_->getAtomType(idents[atom2]) ); + if (storageLayout_ & DataStorage::dslFlucQPosition) { + idat.flucQ2 = &(atomColData.flucQPos[atom2]); + } +#else + idat.atid2 = idents[atom2]; + + if (regions[atom1] >= 0 && regions[atom2] >= 0) { + idat.sameRegion = (regions[atom1] == regions[atom2]); + } else { + idat.sameRegion = false; + } + if (storageLayout_ & DataStorage::dslAmat) { - idat.A1 = &(snap_->atomData.aMat[atom1]); idat.A2 = &(snap_->atomData.aMat[atom2]); } - if (storageLayout_ & DataStorage::dslElectroFrame) { - idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]); - idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]); - } - if (storageLayout_ & DataStorage::dslTorque) { - idat.t1 = &(snap_->atomData.torque[atom1]); idat.t2 = &(snap_->atomData.torque[atom2]); } + if (storageLayout_ & DataStorage::dslDipole) { + idat.dipole2 = &(snap_->atomData.dipole[atom2]); + } + + if (storageLayout_ & DataStorage::dslQuadrupole) { + idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]); + } + if (storageLayout_ & DataStorage::dslDensity) { - idat.rho1 = &(snap_->atomData.density[atom1]); idat.rho2 = &(snap_->atomData.density[atom2]); } if (storageLayout_ & DataStorage::dslFunctional) { - idat.frho1 = &(snap_->atomData.functional[atom1]); idat.frho2 = &(snap_->atomData.functional[atom2]); } if (storageLayout_ & DataStorage::dslFunctionalDerivative) { - idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]); idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]); } if (storageLayout_ & DataStorage::dslParticlePot) { - idat.particlePot1 = &(snap_->atomData.particlePot[atom1]); idat.particlePot2 = &(snap_->atomData.particlePot[atom2]); } + if (storageLayout_ & DataStorage::dslSkippedCharge) { + idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]); + } + + if (storageLayout_ & DataStorage::dslFlucQPosition) { + idat.flucQ2 = &(snap_->atomData.flucQPos[atom2]); + } + #endif + } } - - void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, int atom1, int atom2) { + void ForceMatrixDecomposition::unpackInteractionData(InteractionData &idat, + int atom1, int atom2) { #ifdef IS_MPI - pot_row[atom1] += 0.5 * *(idat.pot); - pot_col[atom2] += 0.5 * *(idat.pot); + pot_row[atom1] += RealType(0.5) * *(idat.pot); + pot_col[atom2] += RealType(0.5) * *(idat.pot); + expot_row[atom1] += RealType(0.5) * *(idat.excludedPot); + expot_col[atom2] += RealType(0.5) * *(idat.excludedPot); atomRowData.force[atom1] += *(idat.f1); atomColData.force[atom2] -= *(idat.f1); -#else - pairwisePot += *(idat.pot); - snap_->atomData.force[atom1] += *(idat.f1); - snap_->atomData.force[atom2] -= *(idat.f1); -#endif - - } - - - void ForceMatrixDecomposition::fillSkipData(InteractionData &idat, - int atom1, int atom2) { -#ifdef IS_MPI - idat.atypes = make_pair( ff_->getAtomType(identsRow[atom1]), - ff_->getAtomType(identsCol[atom2]) ); - - if (storageLayout_ & DataStorage::dslElectroFrame) { - idat.eFrame1 = &(atomRowData.electroFrame[atom1]); - idat.eFrame2 = &(atomColData.electroFrame[atom2]); + if (storageLayout_ & DataStorage::dslFlucQForce) { + atomRowData.flucQFrc[atom1] -= *(idat.dVdFQ1); + atomColData.flucQFrc[atom2] -= *(idat.dVdFQ2); } - if (storageLayout_ & DataStorage::dslTorque) { - idat.t1 = &(atomRowData.torque[atom1]); - idat.t2 = &(atomColData.torque[atom2]); + if (storageLayout_ & DataStorage::dslElectricField) { + atomRowData.electricField[atom1] += *(idat.eField1); + atomColData.electricField[atom2] += *(idat.eField2); } - if (storageLayout_ & DataStorage::dslSkippedCharge) { - idat.skippedCharge1 = &(atomRowData.skippedCharge[atom1]); - idat.skippedCharge2 = &(atomColData.skippedCharge[atom2]); + if (storageLayout_ & DataStorage::dslSitePotential) { + atomRowData.sitePotential[atom1] += *(idat.sPot1); + atomColData.sitePotential[atom2] += *(idat.sPot2); } + #else - idat.atypes = make_pair( ff_->getAtomType(idents[atom1]), - ff_->getAtomType(idents[atom2]) ); + pairwisePot += *(idat.pot); + excludedPot += *(idat.excludedPot); - if (storageLayout_ & DataStorage::dslElectroFrame) { - idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]); - idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]); + snap_->atomData.force[atom1] += *(idat.f1); + snap_->atomData.force[atom2] -= *(idat.f1); + + if (idat.doParticlePot) { + // This is the pairwise contribution to the particle pot. The + // embedding contribution is added in each of the low level + // non-bonded routines. In parallel, this calculation is done + // in collectData, not in unpackInteractionData. + snap_->atomData.particlePot[atom1] += *(idat.vpair) * *(idat.sw); + snap_->atomData.particlePot[atom2] += *(idat.vpair) * *(idat.sw); } + + if (storageLayout_ & DataStorage::dslFlucQForce) { + snap_->atomData.flucQFrc[atom1] -= *(idat.dVdFQ1); + snap_->atomData.flucQFrc[atom2] -= *(idat.dVdFQ2); + } - if (storageLayout_ & DataStorage::dslTorque) { - idat.t1 = &(snap_->atomData.torque[atom1]); - idat.t2 = &(snap_->atomData.torque[atom2]); + if (storageLayout_ & DataStorage::dslElectricField) { + snap_->atomData.electricField[atom1] += *(idat.eField1); + snap_->atomData.electricField[atom2] += *(idat.eField2); } - if (storageLayout_ & DataStorage::dslSkippedCharge) { - idat.skippedCharge1 = &(snap_->atomData.skippedCharge[atom1]); - idat.skippedCharge2 = &(snap_->atomData.skippedCharge[atom2]); + if (storageLayout_ & DataStorage::dslSitePotential) { + snap_->atomData.sitePotential[atom1] += *(idat.sPot1); + snap_->atomData.sitePotential[atom2] += *(idat.sPot2); } -#endif - } - - void ForceMatrixDecomposition::unpackSkipData(InteractionData &idat, int atom1, int atom2) { -#ifdef IS_MPI - pot_row[atom1] += 0.5 * *(idat.pot); - pot_col[atom2] += 0.5 * *(idat.pot); -#else - pairwisePot += *(idat.pot); #endif - + } - /* * buildNeighborList * - * first element of pair is row-indexed CutoffGroup - * second element of pair is column-indexed CutoffGroup + * Constructs the Verlet neighbor list for a force-matrix + * decomposition. In this case, each processor is responsible for + * row-site interactions with column-sites. + * + * neighborList is returned as a packed array of neighboring + * column-ordered CutoffGroups. The starting position in + * neighborList for each row-ordered CutoffGroup is given by the + * returned vector point. */ - vector > ForceMatrixDecomposition::buildNeighborList() { - - vector > neighborList; - groupCutoffs cuts; + void ForceMatrixDecomposition::buildNeighborList(vector& neighborList, + vector& point) { + neighborList.clear(); + point.clear(); + int len = 0; + + bool doAllPairs = false; + + Snapshot* snap_ = sman_->getCurrentSnapshot(); + Mat3x3d box; + Mat3x3d invBox; + + Vector3d rs, scaled, dr; + Vector3i whichCell; + int cellIndex; + #ifdef IS_MPI cellListRow_.clear(); cellListCol_.clear(); + point.resize(nGroupsInRow_+1); #else cellList_.clear(); + point.resize(nGroups_+1); #endif + + if (!usePeriodicBoundaryConditions_) { + box = snap_->getBoundingBox(); + invBox = snap_->getInvBoundingBox(); + } else { + box = snap_->getHmat(); + invBox = snap_->getInvHmat(); + } + + Vector3d A = box.getColumn(0); + Vector3d B = box.getColumn(1); + Vector3d C = box.getColumn(2); - RealType rList_ = (largestRcut_ + skinThickness_); - RealType rl2 = rList_ * rList_; - Snapshot* snap_ = sman_->getCurrentSnapshot(); - Mat3x3d Hmat = snap_->getHmat(); - Vector3d Hx = Hmat.getColumn(0); - Vector3d Hy = Hmat.getColumn(1); - Vector3d Hz = Hmat.getColumn(2); + // Required for triclinic cells + Vector3d AxB = cross(A, B); + Vector3d BxC = cross(B, C); + Vector3d CxA = cross(C, A); - nCells_.x() = (int) ( Hx.length() )/ rList_; - nCells_.y() = (int) ( Hy.length() )/ rList_; - nCells_.z() = (int) ( Hz.length() )/ rList_; + // unit vectors perpendicular to the faces of the triclinic cell: + AxB.normalize(); + BxC.normalize(); + CxA.normalize(); - Mat3x3d invHmat = snap_->getInvHmat(); - Vector3d rs, scaled, dr; - Vector3i whichCell; - int cellIndex; + // A set of perpendicular lengths in triclinic cells: + RealType Wa = abs(dot(A, BxC)); + RealType Wb = abs(dot(B, CxA)); + RealType Wc = abs(dot(C, AxB)); + + nCells_.x() = int( Wa / rList_ ); + nCells_.y() = int( Wb / rList_ ); + nCells_.z() = int( Wc / rList_ ); + + // handle small boxes where the cell offsets can end up repeating cells + if (nCells_.x() < 3) doAllPairs = true; + if (nCells_.y() < 3) doAllPairs = true; + if (nCells_.z() < 3) doAllPairs = true; + int nCtot = nCells_.x() * nCells_.y() * nCells_.z(); - + #ifdef IS_MPI cellListRow_.resize(nCtot); cellListCol_.resize(nCtot); #else cellList_.resize(nCtot); #endif - + + if (!doAllPairs) { + #ifdef IS_MPI - for (int i = 0; i < nGroupsInRow_; i++) { - rs = cgRowData.position[i]; - - // scaled positions relative to the box vectors - scaled = invHmat * rs; - - // wrap the vector back into the unit box by subtracting integer box - // numbers - for (int j = 0; j < 3; j++) { - scaled[j] -= roundMe(scaled[j]); - scaled[j] += 0.5; + + for (int i = 0; i < nGroupsInRow_; i++) { + rs = cgRowData.position[i]; + + // scaled positions relative to the box vectors + scaled = invBox * rs; + + // wrap the vector back into the unit box by subtracting integer box + // numbers + for (int j = 0; j < 3; j++) { + scaled[j] -= roundMe(scaled[j]); + scaled[j] += 0.5; + // Handle the special case when an object is exactly on the + // boundary (a scaled coordinate of 1.0 is the same as + // scaled coordinate of 0.0) + if (scaled[j] >= 1.0) scaled[j] -= 1.0; + } + + // find xyz-indices of cell that cutoffGroup is in. + whichCell.x() = nCells_.x() * scaled.x(); + whichCell.y() = nCells_.y() * scaled.y(); + whichCell.z() = nCells_.z() * scaled.z(); + + // find single index of this cell: + cellIndex = Vlinear(whichCell, nCells_); + + // add this cutoff group to the list of groups in this cell; + cellListRow_[cellIndex].push_back(i); } - - // find xyz-indices of cell that cutoffGroup is in. - whichCell.x() = nCells_.x() * scaled.x(); - whichCell.y() = nCells_.y() * scaled.y(); - whichCell.z() = nCells_.z() * scaled.z(); - - // find single index of this cell: - cellIndex = Vlinear(whichCell, nCells_); - - // add this cutoff group to the list of groups in this cell; - cellListRow_[cellIndex].push_back(i); - } - - for (int i = 0; i < nGroupsInCol_; i++) { - rs = cgColData.position[i]; - - // scaled positions relative to the box vectors - scaled = invHmat * rs; - - // wrap the vector back into the unit box by subtracting integer box - // numbers - for (int j = 0; j < 3; j++) { - scaled[j] -= roundMe(scaled[j]); - scaled[j] += 0.5; + for (int i = 0; i < nGroupsInCol_; i++) { + rs = cgColData.position[i]; + + // scaled positions relative to the box vectors + scaled = invBox * rs; + + // wrap the vector back into the unit box by subtracting integer box + // numbers + for (int j = 0; j < 3; j++) { + scaled[j] -= roundMe(scaled[j]); + scaled[j] += 0.5; + // Handle the special case when an object is exactly on the + // boundary (a scaled coordinate of 1.0 is the same as + // scaled coordinate of 0.0) + if (scaled[j] >= 1.0) scaled[j] -= 1.0; + } + + // find xyz-indices of cell that cutoffGroup is in. + whichCell.x() = nCells_.x() * scaled.x(); + whichCell.y() = nCells_.y() * scaled.y(); + whichCell.z() = nCells_.z() * scaled.z(); + + // find single index of this cell: + cellIndex = Vlinear(whichCell, nCells_); + + // add this cutoff group to the list of groups in this cell; + cellListCol_[cellIndex].push_back(i); } - - // find xyz-indices of cell that cutoffGroup is in. - whichCell.x() = nCells_.x() * scaled.x(); - whichCell.y() = nCells_.y() * scaled.y(); - whichCell.z() = nCells_.z() * scaled.z(); - - // find single index of this cell: - cellIndex = Vlinear(whichCell, nCells_); - - // add this cutoff group to the list of groups in this cell; - cellListCol_[cellIndex].push_back(i); - } + #else - for (int i = 0; i < nGroups_; i++) { - rs = snap_->cgData.position[i]; - - // scaled positions relative to the box vectors - scaled = invHmat * rs; - - // wrap the vector back into the unit box by subtracting integer box - // numbers - for (int j = 0; j < 3; j++) { - scaled[j] -= roundMe(scaled[j]); - scaled[j] += 0.5; + for (int i = 0; i < nGroups_; i++) { + rs = snap_->cgData.position[i]; + + // scaled positions relative to the box vectors + scaled = invBox * rs; + + // wrap the vector back into the unit box by subtracting integer box + // numbers + for (int j = 0; j < 3; j++) { + scaled[j] -= roundMe(scaled[j]); + scaled[j] += 0.5; + // Handle the special case when an object is exactly on the + // boundary (a scaled coordinate of 1.0 is the same as + // scaled coordinate of 0.0) + if (scaled[j] >= 1.0) scaled[j] -= 1.0; + } + + // find xyz-indices of cell that cutoffGroup is in. + whichCell.x() = int(nCells_.x() * scaled.x()); + whichCell.y() = int(nCells_.y() * scaled.y()); + whichCell.z() = int(nCells_.z() * scaled.z()); + + // find single index of this cell: + cellIndex = Vlinear(whichCell, nCells_); + + // add this cutoff group to the list of groups in this cell; + cellList_[cellIndex].push_back(i); } - // find xyz-indices of cell that cutoffGroup is in. - whichCell.x() = nCells_.x() * scaled.x(); - whichCell.y() = nCells_.y() * scaled.y(); - whichCell.z() = nCells_.z() * scaled.z(); +#endif - // find single index of this cell: - cellIndex = Vlinear(whichCell, nCells_); +#ifdef IS_MPI + for (int j1 = 0; j1 < nGroupsInRow_; j1++) { + rs = cgRowData.position[j1]; +#else - // add this cutoff group to the list of groups in this cell; - cellList_[cellIndex].push_back(i); - } + for (int j1 = 0; j1 < nGroups_; j1++) { + rs = snap_->cgData.position[j1]; #endif + point[j1] = len; + + // scaled positions relative to the box vectors + scaled = invBox * rs; + + // wrap the vector back into the unit box by subtracting integer box + // numbers + for (int j = 0; j < 3; j++) { + scaled[j] -= roundMe(scaled[j]); + scaled[j] += 0.5; + // Handle the special case when an object is exactly on the + // boundary (a scaled coordinate of 1.0 is the same as + // scaled coordinate of 0.0) + if (scaled[j] >= 1.0) scaled[j] -= 1.0; + } + + // find xyz-indices of cell that cutoffGroup is in. + whichCell.x() = nCells_.x() * scaled.x(); + whichCell.y() = nCells_.y() * scaled.y(); + whichCell.z() = nCells_.z() * scaled.z(); + + // find single index of this cell: + int m1 = Vlinear(whichCell, nCells_); - for (int m1z = 0; m1z < nCells_.z(); m1z++) { - for (int m1y = 0; m1y < nCells_.y(); m1y++) { - for (int m1x = 0; m1x < nCells_.x(); m1x++) { - Vector3i m1v(m1x, m1y, m1z); - int m1 = Vlinear(m1v, nCells_); + for (vector::iterator os = cellOffsets_.begin(); + os != cellOffsets_.end(); ++os) { + + Vector3i m2v = whichCell + (*os); - for (vector::iterator os = cellOffsets_.begin(); - os != cellOffsets_.end(); ++os) { + if (m2v.x() >= nCells_.x()) { + m2v.x() = 0; + } else if (m2v.x() < 0) { + m2v.x() = nCells_.x() - 1; + } + + if (m2v.y() >= nCells_.y()) { + m2v.y() = 0; + } else if (m2v.y() < 0) { + m2v.y() = nCells_.y() - 1; + } + + if (m2v.z() >= nCells_.z()) { + m2v.z() = 0; + } else if (m2v.z() < 0) { + m2v.z() = nCells_.z() - 1; + } + int m2 = Vlinear (m2v, nCells_); +#ifdef IS_MPI + for (vector::iterator j2 = cellListCol_[m2].begin(); + j2 != cellListCol_[m2].end(); ++j2) { - Vector3i m2v = m1v + (*os); - - if (m2v.x() >= nCells_.x()) { - m2v.x() = 0; - } else if (m2v.x() < 0) { - m2v.x() = nCells_.x() - 1; + // In parallel, we need to visit *all* pairs of row + // & column indicies and will divide labor in the + // force evaluation later. + dr = cgColData.position[(*j2)] - rs; + if (usePeriodicBoundaryConditions_) { + snap_->wrapVector(dr); } + if (dr.lengthSquare() < rListSq_) { + neighborList.push_back( (*j2) ); + ++len; + } + } +#else + for (vector::iterator j2 = cellList_[m2].begin(); + j2 != cellList_[m2].end(); ++j2) { + + // Always do this if we're in different cells or if + // we're in the same cell and the global index of + // the j2 cutoff group is greater than or equal to + // the j1 cutoff group. Note that Rappaport's code + // has a "less than" conditional here, but that + // deals with atom-by-atom computation. OpenMD + // allows atoms within a single cutoff group to + // interact with each other. - if (m2v.y() >= nCells_.y()) { - m2v.y() = 0; - } else if (m2v.y() < 0) { - m2v.y() = nCells_.y() - 1; - } - - if (m2v.z() >= nCells_.z()) { - m2v.z() = 0; - } else if (m2v.z() < 0) { - m2v.z() = nCells_.z() - 1; - } - - int m2 = Vlinear (m2v, nCells_); - -#ifdef IS_MPI - for (vector::iterator j1 = cellListRow_[m1].begin(); - j1 != cellListRow_[m1].end(); ++j1) { - for (vector::iterator j2 = cellListCol_[m2].begin(); - j2 != cellListCol_[m2].end(); ++j2) { - - // Always do this if we're in different cells or if - // we're in the same cell and the global index of the - // j2 cutoff group is less than the j1 cutoff group - - if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) { - dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)]; - snap_->wrapVector(dr); - cuts = getGroupCutoffs( (*j1), (*j2) ); - if (dr.lengthSquare() < cuts.third) { - neighborList.push_back(make_pair((*j1), (*j2))); - } - } + if ( (*j2) >= j1 ) { + + dr = snap_->cgData.position[(*j2)] - rs; + if (usePeriodicBoundaryConditions_) { + snap_->wrapVector(dr); } - } -#else - - for (vector::iterator j1 = cellList_[m1].begin(); - j1 != cellList_[m1].end(); ++j1) { - for (vector::iterator j2 = cellList_[m2].begin(); - j2 != cellList_[m2].end(); ++j2) { - - // Always do this if we're in different cells or if - // we're in the same cell and the global index of the - // j2 cutoff group is less than the j1 cutoff group - - if (m2 != m1 || (*j2) < (*j1)) { - dr = snap_->cgData.position[(*j2)] - snap_->cgData.position[(*j1)]; - snap_->wrapVector(dr); - cuts = getGroupCutoffs( (*j1), (*j2) ); - if (dr.lengthSquare() < cuts.third) { - neighborList.push_back(make_pair((*j1), (*j2))); - } - } + if ( dr.lengthSquare() < rListSq_) { + neighborList.push_back( (*j2) ); + ++len; } } + } #endif + } + } + } else { + // branch to do all cutoff group pairs +#ifdef IS_MPI + for (int j1 = 0; j1 < nGroupsInRow_; j1++) { + point[j1] = len; + rs = cgRowData.position[j1]; + for (int j2 = 0; j2 < nGroupsInCol_; j2++) { + dr = cgColData.position[j2] - rs; + if (usePeriodicBoundaryConditions_) { + snap_->wrapVector(dr); } + if (dr.lengthSquare() < rListSq_) { + neighborList.push_back( j2 ); + ++len; + } } + } +#else + // include all groups here. + for (int j1 = 0; j1 < nGroups_; j1++) { + point[j1] = len; + rs = snap_->cgData.position[j1]; + // include self group interactions j2 == j1 + for (int j2 = j1; j2 < nGroups_; j2++) { + dr = snap_->cgData.position[j2] - rs; + if (usePeriodicBoundaryConditions_) { + snap_->wrapVector(dr); + } + if (dr.lengthSquare() < rListSq_) { + neighborList.push_back( j2 ); + ++len; + } + } } +#endif } - + +#ifdef IS_MPI + point[nGroupsInRow_] = len; +#else + point[nGroups_] = len; +#endif + // save the local cutoff group positions for the check that is // done on each loop: saved_CG_positions_.clear(); + saved_CG_positions_.reserve(nGroups_); for (int i = 0; i < nGroups_; i++) saved_CG_positions_.push_back(snap_->cgData.position[i]); - - return neighborList; } } //end namespace OpenMD