| 175 |  |  | 
| 176 |  | pot_row.resize(nAtomsInRow_); | 
| 177 |  | pot_col.resize(nAtomsInCol_); | 
| 178 | + |  | 
| 179 | + | expot_row.resize(nAtomsInRow_); | 
| 180 | + | expot_col.resize(nAtomsInCol_); | 
| 181 |  |  | 
| 182 |  | AtomRowToGlobal.resize(nAtomsInRow_); | 
| 183 |  | AtomColToGlobal.resize(nAtomsInCol_); | 
| 310 |  |  | 
| 311 |  | RealType tol = 1e-6; | 
| 312 |  | largestRcut_ = 0.0; | 
| 310 | – | RealType rc; | 
| 313 |  | int atid; | 
| 314 |  | set<AtomType*> atypes = info_->getSimulatedAtomTypes(); | 
| 315 |  |  | 
| 394 |  | } | 
| 395 |  |  | 
| 396 |  | bool gTypeFound = false; | 
| 397 | < | for (int gt = 0; gt < gTypeCutoffs.size(); gt++) { | 
| 397 | > | for (unsigned int gt = 0; gt < gTypeCutoffs.size(); gt++) { | 
| 398 |  | if (abs(groupCutoff[cg1] - gTypeCutoffs[gt]) < tol) { | 
| 399 |  | groupToGtype[cg1] = gt; | 
| 400 |  | gTypeFound = true; | 
| 419 |  |  | 
| 420 |  | RealType tradRcut = groupMax; | 
| 421 |  |  | 
| 422 | < | for (int i = 0; i < gTypeCutoffs.size();  i++) { | 
| 423 | < | for (int j = 0; j < gTypeCutoffs.size();  j++) { | 
| 422 | > | for (unsigned int i = 0; i < gTypeCutoffs.size();  i++) { | 
| 423 | > | for (unsigned int j = 0; j < gTypeCutoffs.size();  j++) { | 
| 424 |  | RealType thisRcut; | 
| 425 |  | switch(cutoffPolicy_) { | 
| 426 |  | case TRADITIONAL: | 
| 476 |  | } | 
| 477 |  |  | 
| 478 |  | int ForceMatrixDecomposition::getTopologicalDistance(int atom1, int atom2) { | 
| 479 | < | for (int j = 0; j < toposForAtom[atom1].size(); j++) { | 
| 479 | > | for (unsigned int j = 0; j < toposForAtom[atom1].size(); j++) { | 
| 480 |  | if (toposForAtom[atom1][j] == atom2) | 
| 481 |  | return topoDist[atom1][j]; | 
| 482 |  | } | 
| 486 |  | void ForceMatrixDecomposition::zeroWorkArrays() { | 
| 487 |  | pairwisePot = 0.0; | 
| 488 |  | embeddingPot = 0.0; | 
| 489 | + | excludedPot = 0.0; | 
| 490 | + | excludedSelfPot = 0.0; | 
| 491 |  |  | 
| 492 |  | #ifdef IS_MPI | 
| 493 |  | if (storageLayout_ & DataStorage::dslForce) { | 
| 504 |  | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 505 |  |  | 
| 506 |  | fill(pot_col.begin(), pot_col.end(), | 
| 507 | + | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 508 | + |  | 
| 509 | + | fill(expot_row.begin(), expot_row.end(), | 
| 510 | + | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 511 | + |  | 
| 512 | + | fill(expot_col.begin(), expot_col.end(), | 
| 513 |  | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 514 |  |  | 
| 515 |  | if (storageLayout_ & DataStorage::dslParticlePot) { | 
| 559 |  | atomColData.electricField.end(), V3Zero); | 
| 560 |  | } | 
| 561 |  |  | 
| 552 | – | if (storageLayout_ & DataStorage::dslFlucQForce) { | 
| 553 | – | fill(atomRowData.flucQFrc.begin(), atomRowData.flucQFrc.end(), | 
| 554 | – | 0.0); | 
| 555 | – | fill(atomColData.flucQFrc.begin(), atomColData.flucQFrc.end(), | 
| 556 | – | 0.0); | 
| 557 | – | } | 
| 558 | – |  | 
| 562 |  | #endif | 
| 563 |  | // even in parallel, we need to zero out the local arrays: | 
| 564 |  |  | 
| 632 |  | AtomPlanMatrixColumn->gather(snap_->atomData.aMat, | 
| 633 |  | atomColData.aMat); | 
| 634 |  | } | 
| 635 | < |  | 
| 636 | < | // if needed, gather the atomic eletrostatic frames | 
| 637 | < | if (storageLayout_ & DataStorage::dslElectroFrame) { | 
| 638 | < | AtomPlanMatrixRow->gather(snap_->atomData.electroFrame, | 
| 639 | < | atomRowData.electroFrame); | 
| 640 | < | AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame, | 
| 641 | < | atomColData.electroFrame); | 
| 635 | > |  | 
| 636 | > | // if needed, gather the atomic eletrostatic information | 
| 637 | > | if (storageLayout_ & DataStorage::dslDipole) { | 
| 638 | > | AtomPlanVectorRow->gather(snap_->atomData.dipole, | 
| 639 | > | atomRowData.dipole); | 
| 640 | > | AtomPlanVectorColumn->gather(snap_->atomData.dipole, | 
| 641 | > | atomColData.dipole); | 
| 642 |  | } | 
| 643 |  |  | 
| 644 | + | if (storageLayout_ & DataStorage::dslQuadrupole) { | 
| 645 | + | AtomPlanMatrixRow->gather(snap_->atomData.quadrupole, | 
| 646 | + | atomRowData.quadrupole); | 
| 647 | + | AtomPlanMatrixColumn->gather(snap_->atomData.quadrupole, | 
| 648 | + | atomColData.quadrupole); | 
| 649 | + | } | 
| 650 | + |  | 
| 651 |  | // if needed, gather the atomic fluctuating charge values | 
| 652 |  | if (storageLayout_ & DataStorage::dslFlucQPosition) { | 
| 653 |  | AtomPlanRealRow->gather(snap_->atomData.flucQPos, | 
| 686 |  |  | 
| 687 |  | int n = snap_->atomData.electricField.size(); | 
| 688 |  | vector<Vector3d> field_tmp(n, V3Zero); | 
| 689 | < | AtomPlanVectorColumn->scatter(atomColData.electricField, field_tmp); | 
| 689 | > | AtomPlanVectorColumn->scatter(atomColData.electricField, | 
| 690 | > | field_tmp); | 
| 691 |  | for (int i = 0; i < n; i++) | 
| 692 |  | snap_->atomData.electricField[i] += field_tmp[i]; | 
| 693 |  | } | 
| 791 |  |  | 
| 792 |  | vector<potVec> pot_temp(nLocal_, | 
| 793 |  | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 794 | + | vector<potVec> expot_temp(nLocal_, | 
| 795 | + | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 796 |  |  | 
| 797 |  | // scatter/gather pot_row into the members of my column | 
| 798 |  |  | 
| 799 |  | AtomPlanPotRow->scatter(pot_row, pot_temp); | 
| 800 | + | AtomPlanPotRow->scatter(expot_row, expot_temp); | 
| 801 |  |  | 
| 802 | < | for (int ii = 0;  ii < pot_temp.size(); ii++ ) | 
| 802 | > | for (int ii = 0;  ii < pot_temp.size(); ii++ ) | 
| 803 |  | pairwisePot += pot_temp[ii]; | 
| 804 | + |  | 
| 805 | + | for (int ii = 0;  ii < expot_temp.size(); ii++ ) | 
| 806 | + | excludedPot += expot_temp[ii]; | 
| 807 |  |  | 
| 808 |  | if (storageLayout_ & DataStorage::dslParticlePot) { | 
| 809 |  | // This is the pairwise contribution to the particle pot.  The | 
| 821 |  |  | 
| 822 |  | fill(pot_temp.begin(), pot_temp.end(), | 
| 823 |  | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 824 | + | fill(expot_temp.begin(), expot_temp.end(), | 
| 825 | + | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 826 |  |  | 
| 827 |  | AtomPlanPotColumn->scatter(pot_col, pot_temp); | 
| 828 | + | AtomPlanPotColumn->scatter(expot_col, expot_temp); | 
| 829 |  |  | 
| 830 |  | for (int ii = 0;  ii < pot_temp.size(); ii++ ) | 
| 831 |  | pairwisePot += pot_temp[ii]; | 
| 832 |  |  | 
| 833 | + | for (int ii = 0;  ii < expot_temp.size(); ii++ ) | 
| 834 | + | excludedPot += expot_temp[ii]; | 
| 835 | + |  | 
| 836 |  | if (storageLayout_ & DataStorage::dslParticlePot) { | 
| 837 |  | // This is the pairwise contribution to the particle pot.  The | 
| 838 |  | // embedding contribution is added in each of the low level | 
| 874 |  | pairwisePot[ii] = ploc2; | 
| 875 |  | } | 
| 876 |  |  | 
| 877 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { | 
| 878 | + | RealType ploc1 = excludedPot[ii]; | 
| 879 | + | RealType ploc2 = 0.0; | 
| 880 | + | MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM); | 
| 881 | + | excludedPot[ii] = ploc2; | 
| 882 | + | } | 
| 883 | + |  | 
| 884 |  | // Here be dragons. | 
| 885 |  | MPI::Intracomm col = colComm.getComm(); | 
| 886 |  |  | 
| 908 |  | MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM); | 
| 909 |  | embeddingPot[ii] = ploc2; | 
| 910 |  | } | 
| 911 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { | 
| 912 | + | RealType ploc1 = excludedSelfPot[ii]; | 
| 913 | + | RealType ploc2 = 0.0; | 
| 914 | + | MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM); | 
| 915 | + | excludedSelfPot[ii] = ploc2; | 
| 916 | + | } | 
| 917 |  | #endif | 
| 918 |  |  | 
| 919 |  | } | 
| 1043 |  | * the parallel decomposition. | 
| 1044 |  | */ | 
| 1045 |  | bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2, int cg1, int cg2) { | 
| 1046 | < | int unique_id_1, unique_id_2, group1, group2; | 
| 1046 | > | int unique_id_1, unique_id_2; | 
| 1047 |  |  | 
| 1048 |  | #ifdef IS_MPI | 
| 1049 |  | // in MPI, we have to look up the unique IDs for each atom | 
| 1050 |  | unique_id_1 = AtomRowToGlobal[atom1]; | 
| 1051 |  | unique_id_2 = AtomColToGlobal[atom2]; | 
| 1052 | < | group1 = cgRowToGlobal[cg1]; | 
| 1053 | < | group2 = cgColToGlobal[cg2]; | 
| 1052 | > | // group1 = cgRowToGlobal[cg1]; | 
| 1053 | > | // group2 = cgColToGlobal[cg2]; | 
| 1054 |  | #else | 
| 1055 |  | unique_id_1 = AtomLocalToGlobal[atom1]; | 
| 1056 |  | unique_id_2 = AtomLocalToGlobal[atom2]; | 
| 1057 | < | group1 = cgLocalToGlobal[cg1]; | 
| 1058 | < | group2 = cgLocalToGlobal[cg2]; | 
| 1057 | > | int group1 = cgLocalToGlobal[cg1]; | 
| 1058 | > | int group2 = cgLocalToGlobal[cg2]; | 
| 1059 |  | #endif | 
| 1060 |  |  | 
| 1061 |  | if (unique_id_1 == unique_id_2) return true; | 
| 1133 |  | idat.A2 = &(atomColData.aMat[atom2]); | 
| 1134 |  | } | 
| 1135 |  |  | 
| 1100 | – | if (storageLayout_ & DataStorage::dslElectroFrame) { | 
| 1101 | – | idat.eFrame1 = &(atomRowData.electroFrame[atom1]); | 
| 1102 | – | idat.eFrame2 = &(atomColData.electroFrame[atom2]); | 
| 1103 | – | } | 
| 1104 | – |  | 
| 1136 |  | if (storageLayout_ & DataStorage::dslTorque) { | 
| 1137 |  | idat.t1 = &(atomRowData.torque[atom1]); | 
| 1138 |  | idat.t2 = &(atomColData.torque[atom2]); | 
| 1139 |  | } | 
| 1140 |  |  | 
| 1141 | + | if (storageLayout_ & DataStorage::dslDipole) { | 
| 1142 | + | idat.dipole1 = &(atomRowData.dipole[atom1]); | 
| 1143 | + | idat.dipole2 = &(atomColData.dipole[atom2]); | 
| 1144 | + | } | 
| 1145 | + |  | 
| 1146 | + | if (storageLayout_ & DataStorage::dslQuadrupole) { | 
| 1147 | + | idat.quadrupole1 = &(atomRowData.quadrupole[atom1]); | 
| 1148 | + | idat.quadrupole2 = &(atomColData.quadrupole[atom2]); | 
| 1149 | + | } | 
| 1150 | + |  | 
| 1151 |  | if (storageLayout_ & DataStorage::dslDensity) { | 
| 1152 |  | idat.rho1 = &(atomRowData.density[atom1]); | 
| 1153 |  | idat.rho2 = &(atomColData.density[atom2]); | 
| 1187 |  | idat.A2 = &(snap_->atomData.aMat[atom2]); | 
| 1188 |  | } | 
| 1189 |  |  | 
| 1190 | < | if (storageLayout_ & DataStorage::dslElectroFrame) { | 
| 1150 | < | idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]); | 
| 1151 | < | idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]); | 
| 1152 | < | } | 
| 1190 | > | RealType ct = dot(idat.A1->getColumn(2), idat.A2->getColumn(2)); | 
| 1191 |  |  | 
| 1192 |  | if (storageLayout_ & DataStorage::dslTorque) { | 
| 1193 |  | idat.t1 = &(snap_->atomData.torque[atom1]); | 
| 1194 |  | idat.t2 = &(snap_->atomData.torque[atom2]); | 
| 1195 | + | } | 
| 1196 | + |  | 
| 1197 | + | if (storageLayout_ & DataStorage::dslDipole) { | 
| 1198 | + | idat.dipole1 = &(snap_->atomData.dipole[atom1]); | 
| 1199 | + | idat.dipole2 = &(snap_->atomData.dipole[atom2]); | 
| 1200 | + | } | 
| 1201 | + |  | 
| 1202 | + | if (storageLayout_ & DataStorage::dslQuadrupole) { | 
| 1203 | + | idat.quadrupole1 = &(snap_->atomData.quadrupole[atom1]); | 
| 1204 | + | idat.quadrupole2 = &(snap_->atomData.quadrupole[atom2]); | 
| 1205 |  | } | 
| 1206 |  |  | 
| 1207 |  | if (storageLayout_ & DataStorage::dslDensity) { | 
| 1242 |  | #ifdef IS_MPI | 
| 1243 |  | pot_row[atom1] += RealType(0.5) *  *(idat.pot); | 
| 1244 |  | pot_col[atom2] += RealType(0.5) *  *(idat.pot); | 
| 1245 | + | expot_row[atom1] += RealType(0.5) *  *(idat.excludedPot); | 
| 1246 | + | expot_col[atom2] += RealType(0.5) *  *(idat.excludedPot); | 
| 1247 |  |  | 
| 1248 |  | atomRowData.force[atom1] += *(idat.f1); | 
| 1249 |  | atomColData.force[atom2] -= *(idat.f1); | 
| 1260 |  |  | 
| 1261 |  | #else | 
| 1262 |  | pairwisePot += *(idat.pot); | 
| 1263 | + | excludedPot += *(idat.excludedPot); | 
| 1264 |  |  | 
| 1265 |  | snap_->atomData.force[atom1] += *(idat.f1); | 
| 1266 |  | snap_->atomData.force[atom2] -= *(idat.f1); | 
| 1352 |  | for (int j = 0; j < 3; j++) { | 
| 1353 |  | scaled[j] -= roundMe(scaled[j]); | 
| 1354 |  | scaled[j] += 0.5; | 
| 1355 | + | // Handle the special case when an object is exactly on the | 
| 1356 | + | // boundary (a scaled coordinate of 1.0 is the same as | 
| 1357 | + | // scaled coordinate of 0.0) | 
| 1358 | + | if (scaled[j] >= 1.0) scaled[j] -= 1.0; | 
| 1359 |  | } | 
| 1360 |  |  | 
| 1361 |  | // find xyz-indices of cell that cutoffGroup is in. | 
| 1380 |  | for (int j = 0; j < 3; j++) { | 
| 1381 |  | scaled[j] -= roundMe(scaled[j]); | 
| 1382 |  | scaled[j] += 0.5; | 
| 1383 | + | // Handle the special case when an object is exactly on the | 
| 1384 | + | // boundary (a scaled coordinate of 1.0 is the same as | 
| 1385 | + | // scaled coordinate of 0.0) | 
| 1386 | + | if (scaled[j] >= 1.0) scaled[j] -= 1.0; | 
| 1387 |  | } | 
| 1388 |  |  | 
| 1389 |  | // find xyz-indices of cell that cutoffGroup is in. | 
| 1410 |  | for (int j = 0; j < 3; j++) { | 
| 1411 |  | scaled[j] -= roundMe(scaled[j]); | 
| 1412 |  | scaled[j] += 0.5; | 
| 1413 | + | // Handle the special case when an object is exactly on the | 
| 1414 | + | // boundary (a scaled coordinate of 1.0 is the same as | 
| 1415 | + | // scaled coordinate of 0.0) | 
| 1416 | + | if (scaled[j] >= 1.0) scaled[j] -= 1.0; | 
| 1417 |  | } | 
| 1418 |  |  | 
| 1419 |  | // find xyz-indices of cell that cutoffGroup is in. |