| 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_.push_back( Vector3i(-1, 0, 0) ); | 
| 57 | < | cellOffsets_.push_back( Vector3i(-1,-1, 0) ); | 
| 58 | < | cellOffsets_.push_back( Vector3i( 0,-1, 0) ); | 
| 59 | < | cellOffsets_.push_back( Vector3i( 1,-1, 0) ); | 
| 60 | < | cellOffsets_.push_back( Vector3i( 0, 0,-1) ); | 
| 61 | < | cellOffsets_.push_back( Vector3i(-1, 0, 1) ); | 
| 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) ); | 
| 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) ); | 
| 66 | – | cellOffsets_.push_back( Vector3i( 1, 1,-1) ); | 
| 67 | – | cellOffsets_.push_back( Vector3i( 0, 1,-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 |  |  | 
| 233 |  | } | 
| 234 |  | } | 
| 235 |  |  | 
| 236 | < | #endif | 
| 222 | < |  | 
| 223 | < | // allocate memory for the parallel objects | 
| 224 | < | atypesLocal.resize(nLocal_); | 
| 225 | < |  | 
| 226 | < | for (int i = 0; i < nLocal_; i++) | 
| 227 | < | atypesLocal[i] = ff_->getAtomType(idents[i]); | 
| 228 | < |  | 
| 229 | < | groupList_.clear(); | 
| 230 | < | groupList_.resize(nGroups_); | 
| 231 | < | for (int i = 0; i < nGroups_; i++) { | 
| 232 | < | int gid = cgLocalToGlobal[i]; | 
| 233 | < | for (int j = 0; j < nLocal_; j++) { | 
| 234 | < | int aid = AtomLocalToGlobal[j]; | 
| 235 | < | if (globalGroupMembership[aid] == gid) { | 
| 236 | < | groupList_[i].push_back(j); | 
| 237 | < | } | 
| 238 | < | } | 
| 239 | < | } | 
| 240 | < |  | 
| 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 |  | } | 
| 685 |  | } | 
| 686 |  |  | 
| 687 |  | AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp); | 
| 688 | < | for (int i = 0; i < ns; i++) | 
| 688 | > | for (int i = 0; i < ns; i++) | 
| 689 |  | snap_->atomData.skippedCharge[i] += skch_tmp[i]; | 
| 690 | + |  | 
| 691 |  | } | 
| 692 |  |  | 
| 693 |  | nLocal_ = snap_->getNumberOfAtoms(); | 
| 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 |  | } | 
| 865 |  | * field) must still be handled for these pairs. | 
| 866 |  | */ | 
| 867 |  | bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) { | 
| 868 | < | int unique_id_2; | 
| 869 | < | #ifdef IS_MPI | 
| 870 | < | // in MPI, we have to look up the unique IDs for the row atom. | 
| 846 | < | unique_id_2 = AtomColToGlobal[atom2]; | 
| 847 | < | #else | 
| 848 | < | // in the normal loop, the atom numbers are unique | 
| 849 | < | unique_id_2 = atom2; | 
| 850 | < | #endif | 
| 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 |  |  | 
| 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; | 
| 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]; | 
| 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 | < | // In parallel, we need to visit *all* pairs of row & | 
| 1182 | < | // column indicies and will truncate later on. | 
| 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) ); |