| 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_.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) ); | 
| 62 | + | cellOffsets_.push_back( Vector3i(-1,-1,-1) ); | 
| 63 | + | cellOffsets_.push_back( Vector3i( 0,-1,-1) ); | 
| 64 | + | cellOffsets_.push_back( Vector3i( 1,-1,-1) ); | 
| 65 | + | cellOffsets_.push_back( Vector3i( 1, 0,-1) ); | 
| 66 | + | cellOffsets_.push_back( Vector3i( 1, 1,-1) ); | 
| 67 | + | cellOffsets_.push_back( Vector3i( 0, 1,-1) ); | 
| 68 | + | cellOffsets_.push_back( Vector3i(-1, 1,-1) ); | 
| 69 | + | #endif | 
| 70 | + | } | 
| 71 | + |  | 
| 72 | + |  | 
| 73 |  | /** | 
| 74 |  | * distributeInitialData is essentially a copy of the older fortran | 
| 75 |  | * SimulationSetup | 
| 76 |  | */ | 
| 54 | – |  | 
| 77 |  | void ForceMatrixDecomposition::distributeInitialData() { | 
| 78 |  | snap_ = sman_->getCurrentSnapshot(); | 
| 79 |  | storageLayout_ = sman_->getStorageLayout(); | 
| 96 |  |  | 
| 97 |  | #ifdef IS_MPI | 
| 98 |  |  | 
| 99 | < | AtomCommIntRow = new Communicator<Row,int>(nLocal_); | 
| 100 | < | 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_); | 
| 99 | > | MPI::Intracomm row = rowComm.getComm(); | 
| 100 | > | MPI::Intracomm col = colComm.getComm(); | 
| 101 |  |  | 
| 102 | < | AtomCommIntColumn = new Communicator<Column,int>(nLocal_); | 
| 103 | < | AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_); | 
| 104 | < | AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_); | 
| 105 | < | AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_); | 
| 106 | < | AtomCommPotColumn = new Communicator<Column,potVec>(nLocal_); | 
| 102 | > | AtomPlanIntRow = new Plan<int>(row, nLocal_); | 
| 103 | > | AtomPlanRealRow = new Plan<RealType>(row, nLocal_); | 
| 104 | > | AtomPlanVectorRow = new Plan<Vector3d>(row, nLocal_); | 
| 105 | > | AtomPlanMatrixRow = new Plan<Mat3x3d>(row, nLocal_); | 
| 106 | > | AtomPlanPotRow = new Plan<potVec>(row, nLocal_); | 
| 107 |  |  | 
| 108 | < | cgCommIntRow = new Communicator<Row,int>(nGroups_); | 
| 109 | < | cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_); | 
| 110 | < | cgCommIntColumn = new Communicator<Column,int>(nGroups_); | 
| 111 | < | cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_); | 
| 108 | > | AtomPlanIntColumn = new Plan<int>(col, nLocal_); | 
| 109 | > | AtomPlanRealColumn = new Plan<RealType>(col, nLocal_); | 
| 110 | > | AtomPlanVectorColumn = new Plan<Vector3d>(col, nLocal_); | 
| 111 | > | AtomPlanMatrixColumn = new Plan<Mat3x3d>(col, nLocal_); | 
| 112 | > | AtomPlanPotColumn = new Plan<potVec>(col, nLocal_); | 
| 113 |  |  | 
| 114 | < | nAtomsInRow_ = AtomCommIntRow->getSize(); | 
| 115 | < | nAtomsInCol_ = AtomCommIntColumn->getSize(); | 
| 116 | < | nGroupsInRow_ = cgCommIntRow->getSize(); | 
| 117 | < | nGroupsInCol_ = cgCommIntColumn->getSize(); | 
| 114 | > | cgPlanIntRow = new Plan<int>(row, nGroups_); | 
| 115 | > | cgPlanVectorRow = new Plan<Vector3d>(row, nGroups_); | 
| 116 | > | cgPlanIntColumn = new Plan<int>(col, nGroups_); | 
| 117 | > | cgPlanVectorColumn = new Plan<Vector3d>(col, nGroups_); | 
| 118 |  |  | 
| 119 | + | nAtomsInRow_ = AtomPlanIntRow->getSize(); | 
| 120 | + | nAtomsInCol_ = AtomPlanIntColumn->getSize(); | 
| 121 | + | nGroupsInRow_ = cgPlanIntRow->getSize(); | 
| 122 | + | nGroupsInCol_ = cgPlanIntColumn->getSize(); | 
| 123 | + |  | 
| 124 |  | // Modify the data storage objects with the correct layouts and sizes: | 
| 125 |  | atomRowData.resize(nAtomsInRow_); | 
| 126 |  | atomRowData.setStorageLayout(storageLayout_); | 
| 134 |  | identsRow.resize(nAtomsInRow_); | 
| 135 |  | identsCol.resize(nAtomsInCol_); | 
| 136 |  |  | 
| 137 | < | AtomCommIntRow->gather(idents, identsRow); | 
| 138 | < | AtomCommIntColumn->gather(idents, identsCol); | 
| 137 | > | AtomPlanIntRow->gather(idents, identsRow); | 
| 138 | > | AtomPlanIntColumn->gather(idents, identsCol); | 
| 139 |  |  | 
| 140 |  | // allocate memory for the parallel objects | 
| 141 |  | atypesRow.resize(nAtomsInRow_); | 
| 151 |  |  | 
| 152 |  | AtomRowToGlobal.resize(nAtomsInRow_); | 
| 153 |  | AtomColToGlobal.resize(nAtomsInCol_); | 
| 154 | < | AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal); | 
| 155 | < | AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal); | 
| 156 | < |  | 
| 154 | > | AtomPlanIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal); | 
| 155 | > | AtomPlanIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal); | 
| 156 | > |  | 
| 157 |  | cgRowToGlobal.resize(nGroupsInRow_); | 
| 158 |  | cgColToGlobal.resize(nGroupsInCol_); | 
| 159 | < | cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal); | 
| 160 | < | cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal); | 
| 159 | > | cgPlanIntRow->gather(cgLocalToGlobal, cgRowToGlobal); | 
| 160 | > | cgPlanIntColumn->gather(cgLocalToGlobal, cgColToGlobal); | 
| 161 |  |  | 
| 162 |  | massFactorsRow.resize(nAtomsInRow_); | 
| 163 |  | massFactorsCol.resize(nAtomsInCol_); | 
| 164 | < | AtomCommRealRow->gather(massFactors, massFactorsRow); | 
| 165 | < | AtomCommRealColumn->gather(massFactors, massFactorsCol); | 
| 164 | > | AtomPlanRealRow->gather(massFactors, massFactorsRow); | 
| 165 | > | AtomPlanRealColumn->gather(massFactors, massFactorsCol); | 
| 166 |  |  | 
| 167 |  | groupListRow_.clear(); | 
| 168 |  | groupListRow_.resize(nGroupsInRow_); | 
| 542 |  | #ifdef IS_MPI | 
| 543 |  |  | 
| 544 |  | // gather up the atomic positions | 
| 545 | < | AtomCommVectorRow->gather(snap_->atomData.position, | 
| 545 | > | AtomPlanVectorRow->gather(snap_->atomData.position, | 
| 546 |  | atomRowData.position); | 
| 547 | < | AtomCommVectorColumn->gather(snap_->atomData.position, | 
| 547 | > | AtomPlanVectorColumn->gather(snap_->atomData.position, | 
| 548 |  | atomColData.position); | 
| 549 |  |  | 
| 550 |  | // gather up the cutoff group positions | 
| 551 | < | cgCommVectorRow->gather(snap_->cgData.position, | 
| 552 | < | cgRowData.position); | 
| 553 | < | cgCommVectorColumn->gather(snap_->cgData.position, | 
| 551 | > |  | 
| 552 | > | cgPlanVectorRow->gather(snap_->cgData.position, | 
| 553 | > | cgRowData.position); | 
| 554 | > |  | 
| 555 | > | cgPlanVectorColumn->gather(snap_->cgData.position, | 
| 556 |  | cgColData.position); | 
| 557 | + |  | 
| 558 |  |  | 
| 559 |  | // if needed, gather the atomic rotation matrices | 
| 560 |  | if (storageLayout_ & DataStorage::dslAmat) { | 
| 561 | < | AtomCommMatrixRow->gather(snap_->atomData.aMat, | 
| 561 | > | AtomPlanMatrixRow->gather(snap_->atomData.aMat, | 
| 562 |  | atomRowData.aMat); | 
| 563 | < | AtomCommMatrixColumn->gather(snap_->atomData.aMat, | 
| 563 | > | AtomPlanMatrixColumn->gather(snap_->atomData.aMat, | 
| 564 |  | atomColData.aMat); | 
| 565 |  | } | 
| 566 |  |  | 
| 567 |  | // if needed, gather the atomic eletrostatic frames | 
| 568 |  | if (storageLayout_ & DataStorage::dslElectroFrame) { | 
| 569 | < | AtomCommMatrixRow->gather(snap_->atomData.electroFrame, | 
| 569 | > | AtomPlanMatrixRow->gather(snap_->atomData.electroFrame, | 
| 570 |  | atomRowData.electroFrame); | 
| 571 | < | AtomCommMatrixColumn->gather(snap_->atomData.electroFrame, | 
| 571 | > | AtomPlanMatrixColumn->gather(snap_->atomData.electroFrame, | 
| 572 |  | atomColData.electroFrame); | 
| 573 |  | } | 
| 574 |  |  | 
| 585 |  |  | 
| 586 |  | if (storageLayout_ & DataStorage::dslDensity) { | 
| 587 |  |  | 
| 588 | < | AtomCommRealRow->scatter(atomRowData.density, | 
| 588 | > | AtomPlanRealRow->scatter(atomRowData.density, | 
| 589 |  | snap_->atomData.density); | 
| 590 |  |  | 
| 591 |  | int n = snap_->atomData.density.size(); | 
| 592 |  | vector<RealType> rho_tmp(n, 0.0); | 
| 593 | < | AtomCommRealColumn->scatter(atomColData.density, rho_tmp); | 
| 593 | > | AtomPlanRealColumn->scatter(atomColData.density, rho_tmp); | 
| 594 |  | for (int i = 0; i < n; i++) | 
| 595 |  | snap_->atomData.density[i] += rho_tmp[i]; | 
| 596 |  | } | 
| 606 |  | storageLayout_ = sman_->getStorageLayout(); | 
| 607 |  | #ifdef IS_MPI | 
| 608 |  | if (storageLayout_ & DataStorage::dslFunctional) { | 
| 609 | < | AtomCommRealRow->gather(snap_->atomData.functional, | 
| 609 | > | AtomPlanRealRow->gather(snap_->atomData.functional, | 
| 610 |  | atomRowData.functional); | 
| 611 | < | AtomCommRealColumn->gather(snap_->atomData.functional, | 
| 611 | > | AtomPlanRealColumn->gather(snap_->atomData.functional, | 
| 612 |  | atomColData.functional); | 
| 613 |  | } | 
| 614 |  |  | 
| 615 |  | if (storageLayout_ & DataStorage::dslFunctionalDerivative) { | 
| 616 | < | AtomCommRealRow->gather(snap_->atomData.functionalDerivative, | 
| 616 | > | AtomPlanRealRow->gather(snap_->atomData.functionalDerivative, | 
| 617 |  | atomRowData.functionalDerivative); | 
| 618 | < | AtomCommRealColumn->gather(snap_->atomData.functionalDerivative, | 
| 618 | > | AtomPlanRealColumn->gather(snap_->atomData.functionalDerivative, | 
| 619 |  | atomColData.functionalDerivative); | 
| 620 |  | } | 
| 621 |  | #endif | 
| 629 |  | int n = snap_->atomData.force.size(); | 
| 630 |  | vector<Vector3d> frc_tmp(n, V3Zero); | 
| 631 |  |  | 
| 632 | < | AtomCommVectorRow->scatter(atomRowData.force, frc_tmp); | 
| 632 | > | AtomPlanVectorRow->scatter(atomRowData.force, frc_tmp); | 
| 633 |  | for (int i = 0; i < n; i++) { | 
| 634 |  | snap_->atomData.force[i] += frc_tmp[i]; | 
| 635 |  | frc_tmp[i] = 0.0; | 
| 636 |  | } | 
| 637 |  |  | 
| 638 | < | AtomCommVectorColumn->scatter(atomColData.force, frc_tmp); | 
| 639 | < | for (int i = 0; i < n; i++) | 
| 638 | > | AtomPlanVectorColumn->scatter(atomColData.force, frc_tmp); | 
| 639 | > | for (int i = 0; i < n; i++) { | 
| 640 |  | snap_->atomData.force[i] += frc_tmp[i]; | 
| 641 | + | } | 
| 642 |  |  | 
| 643 |  | if (storageLayout_ & DataStorage::dslTorque) { | 
| 644 |  |  | 
| 645 |  | int nt = snap_->atomData.torque.size(); | 
| 646 |  | vector<Vector3d> trq_tmp(nt, V3Zero); | 
| 647 |  |  | 
| 648 | < | AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp); | 
| 648 | > | AtomPlanVectorRow->scatter(atomRowData.torque, trq_tmp); | 
| 649 |  | for (int i = 0; i < nt; i++) { | 
| 650 |  | snap_->atomData.torque[i] += trq_tmp[i]; | 
| 651 |  | trq_tmp[i] = 0.0; | 
| 652 |  | } | 
| 653 |  |  | 
| 654 | < | AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp); | 
| 654 | > | AtomPlanVectorColumn->scatter(atomColData.torque, trq_tmp); | 
| 655 |  | for (int i = 0; i < nt; i++) | 
| 656 |  | snap_->atomData.torque[i] += trq_tmp[i]; | 
| 657 |  | } | 
| 661 |  | int ns = snap_->atomData.skippedCharge.size(); | 
| 662 |  | vector<RealType> skch_tmp(ns, 0.0); | 
| 663 |  |  | 
| 664 | < | AtomCommRealRow->scatter(atomRowData.skippedCharge, skch_tmp); | 
| 664 | > | AtomPlanRealRow->scatter(atomRowData.skippedCharge, skch_tmp); | 
| 665 |  | for (int i = 0; i < ns; i++) { | 
| 666 |  | snap_->atomData.skippedCharge[i] += skch_tmp[i]; | 
| 667 |  | skch_tmp[i] = 0.0; | 
| 668 |  | } | 
| 669 |  |  | 
| 670 | < | AtomCommRealColumn->scatter(atomColData.skippedCharge, skch_tmp); | 
| 670 | > | AtomPlanRealColumn->scatter(atomColData.skippedCharge, skch_tmp); | 
| 671 |  | for (int i = 0; i < ns; i++) | 
| 672 |  | snap_->atomData.skippedCharge[i] += skch_tmp[i]; | 
| 673 |  | } | 
| 679 |  |  | 
| 680 |  | // scatter/gather pot_row into the members of my column | 
| 681 |  |  | 
| 682 | < | AtomCommPotRow->scatter(pot_row, pot_temp); | 
| 682 | > | AtomPlanPotRow->scatter(pot_row, pot_temp); | 
| 683 |  |  | 
| 684 |  | for (int ii = 0;  ii < pot_temp.size(); ii++ ) | 
| 685 |  | pairwisePot += pot_temp[ii]; | 
| 687 |  | fill(pot_temp.begin(), pot_temp.end(), | 
| 688 |  | Vector<RealType, N_INTERACTION_FAMILIES> (0.0)); | 
| 689 |  |  | 
| 690 | < | AtomCommPotColumn->scatter(pot_col, pot_temp); | 
| 690 | > | AtomPlanPotColumn->scatter(pot_col, pot_temp); | 
| 691 |  |  | 
| 692 |  | for (int ii = 0;  ii < pot_temp.size(); ii++ ) | 
| 693 |  | pairwisePot += pot_temp[ii]; | 
| 694 | + |  | 
| 695 | + | for (int ii = 0; ii < N_INTERACTION_FAMILIES; ii++) { | 
| 696 | + | RealType ploc1 = pairwisePot[ii]; | 
| 697 | + | RealType ploc2 = 0.0; | 
| 698 | + | MPI::COMM_WORLD.Allreduce(&ploc1, &ploc2, 1, MPI::REALTYPE, MPI::SUM); | 
| 699 | + | pairwisePot[ii] = ploc2; | 
| 700 | + | } | 
| 701 | + |  | 
| 702 |  | #endif | 
| 703 |  |  | 
| 704 |  | } | 
| 811 |  | */ | 
| 812 |  | bool ForceMatrixDecomposition::skipAtomPair(int atom1, int atom2) { | 
| 813 |  | int unique_id_1, unique_id_2; | 
| 814 | < |  | 
| 814 | > |  | 
| 815 |  | #ifdef IS_MPI | 
| 816 |  | // in MPI, we have to look up the unique IDs for each atom | 
| 817 |  | unique_id_1 = AtomRowToGlobal[atom1]; | 
| 841 |  | */ | 
| 842 |  | bool ForceMatrixDecomposition::excludeAtomPair(int atom1, int atom2) { | 
| 843 |  | int unique_id_2; | 
| 807 | – |  | 
| 844 |  | #ifdef IS_MPI | 
| 845 |  | // in MPI, we have to look up the unique IDs for the row atom. | 
| 846 |  | unique_id_2 = AtomColToGlobal[atom2]; | 
| 1067 |  | // add this cutoff group to the list of groups in this cell; | 
| 1068 |  | cellListRow_[cellIndex].push_back(i); | 
| 1069 |  | } | 
| 1034 | – |  | 
| 1070 |  | for (int i = 0; i < nGroupsInCol_; i++) { | 
| 1071 |  | rs = cgColData.position[i]; | 
| 1072 |  |  | 
| 1111 |  | whichCell.z() = nCells_.z() * scaled.z(); | 
| 1112 |  |  | 
| 1113 |  | // find single index of this cell: | 
| 1114 | < | cellIndex = Vlinear(whichCell, nCells_); | 
| 1114 | > | cellIndex = Vlinear(whichCell, nCells_); | 
| 1115 |  |  | 
| 1116 |  | // add this cutoff group to the list of groups in this cell; | 
| 1117 |  | cellList_[cellIndex].push_back(i); | 
| 1155 |  | for (vector<int>::iterator j2 = cellListCol_[m2].begin(); | 
| 1156 |  | j2 != cellListCol_[m2].end(); ++j2) { | 
| 1157 |  |  | 
| 1158 | < | // Always do this if we're in different cells or if | 
| 1159 | < | // we're in the same cell and the global index of the | 
| 1160 | < | // j2 cutoff group is less than the j1 cutoff group | 
| 1161 | < |  | 
| 1162 | < | if (m2 != m1 || cgColToGlobal[(*j2)] < cgRowToGlobal[(*j1)]) { | 
| 1163 | < | dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)]; | 
| 1164 | < | snap_->wrapVector(dr); | 
| 1165 | < | cuts = getGroupCutoffs( (*j1), (*j2) ); | 
| 1131 | < | if (dr.lengthSquare() < cuts.third) { | 
| 1132 | < | neighborList.push_back(make_pair((*j1), (*j2))); | 
| 1133 | < | } | 
| 1134 | < | } | 
| 1158 | > | // In parallel, we need to visit *all* pairs of row & | 
| 1159 | > | // column indicies and will truncate later on. | 
| 1160 | > | dr = cgColData.position[(*j2)] - cgRowData.position[(*j1)]; | 
| 1161 | > | snap_->wrapVector(dr); | 
| 1162 | > | cuts = getGroupCutoffs( (*j1), (*j2) ); | 
| 1163 | > | if (dr.lengthSquare() < cuts.third) { | 
| 1164 | > | neighborList.push_back(make_pair((*j1), (*j2))); | 
| 1165 | > | } | 
| 1166 |  | } | 
| 1167 |  | } | 
| 1168 |  | #else |