src/parallel/ForceMatrixDecomposition.cpp

/*
 * Copyright (c) 2005 The University of Notre Dame. All Rights Reserved.
 *
 * The University of Notre Dame grants you ("Licensee") a
 * non-exclusive, royalty free, license to use, modify and
 * redistribute this software in source and binary code form, provided
 * that the following conditions are met:
 *
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 *
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the
 *    distribution.
 *
 * This software is provided "AS IS," without a warranty of any
 * kind. All express or implied conditions, representations and
 * warranties, including any implied warranty of merchantability,
 * fitness for a particular purpose or non-infringement, are hereby
 * excluded.  The University of Notre Dame and its licensors shall not
 * be liable for any damages suffered by licensee as a result of
 * using, modifying or distributing the software or its
 * derivatives. In no event will the University of Notre Dame or its
 * licensors be liable for any lost revenue, profit or data, or for
 * direct, indirect, special, consequential, incidental or punitive
 * damages, however caused and regardless of the theory of liability,
 * arising out of the use of or inability to use software, even if the
 * University of Notre Dame has been advised of the possibility of
 * such damages.
 *
 * SUPPORT OPEN SCIENCE!  If you use OpenMD or its source code in your
 * research, please cite the appropriate papers when you publish your
 * work.  Good starting points are:
 *                                                                      
 * [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).                        
 */
#include "parallel/ForceMatrixDecomposition.hpp"
#include "math/SquareMatrix3.hpp"
#include "nonbonded/NonBondedInteraction.hpp"
#include "brains/SnapshotManager.hpp"

using namespace std;
namespace OpenMD {

  /**
   * distributeInitialData is essentially a copy of the older fortran 
   * SimulationSetup
   */
  
  void ForceMatrixDecomposition::distributeInitialData() {
    snap_ = sman_->getCurrentSnapshot();
    storageLayout_ = sman_->getStorageLayout();
    nLocal_ = snap_->getNumberOfAtoms();
    nGroups_ = snap_->getNumberOfCutoffGroups();

#ifdef IS_MPI
 
    AtomCommIntRow = new Communicator<Row,int>(nLocal_);
    AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);

    AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
    AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
    AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
    AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);

    cgCommIntRow = new Communicator<Row,int>(nGroups_);
    cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
    cgCommIntColumn = new Communicator<Column,int>(nGroups_);
    cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);

    nAtomsInRow_ = AtomCommIntRow->getSize();
    nAtomsInCol_ = AtomCommIntColumn->getSize();
    nGroupsInRow_ = cgCommIntRow->getSize();
    nGroupsInCol_ = cgCommIntColumn->getSize();

    // Modify the data storage objects with the correct layouts and sizes:
    atomRowData.resize(nAtomsInRow_);
    atomRowData.setStorageLayout(storageLayout_);
    atomColData.resize(nAtomsInCol_);
    atomColData.setStorageLayout(storageLayout_);
    cgRowData.resize(nGroupsInRow_);
    cgRowData.setStorageLayout(DataStorage::dslPosition);
    cgColData.resize(nGroupsInCol_);
    cgColData.setStorageLayout(DataStorage::dslPosition);
    
    vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES, 
                                      vector<RealType> (nAtomsInRow_, 0.0));
    vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
                                      vector<RealType> (nAtomsInCol_, 0.0));


    vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
    
    // gather the information for atomtype IDs (atids):
    vector<int> identsLocal = info_->getIdentArray();
    identsRow.reserve(nAtomsInRow_);
    identsCol.reserve(nAtomsInCol_);
    
    AtomCommIntRow->gather(identsLocal, identsRow);
    AtomCommIntColumn->gather(identsLocal, identsCol);
    
    AtomLocalToGlobal = info_->getGlobalAtomIndices();
    AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
    AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
    
    cgLocalToGlobal = info_->getGlobalGroupIndices();
    cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
    cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);

    // still need:
    // topoDist
    // exclude
#endif
  }
    

  void ForceMatrixDecomposition::distributeData()  {
    snap_ = sman_->getCurrentSnapshot();
    storageLayout_ = sman_->getStorageLayout();
#ifdef IS_MPI
    
    // gather up the atomic positions
    AtomCommVectorRow->gather(snap_->atomData.position, 
                              atomRowData.position);
    AtomCommVectorColumn->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 needed, gather the atomic rotation matrices
    if (storageLayout_ & DataStorage::dslAmat) {
      AtomCommMatrixRow->gather(snap_->atomData.aMat, 
                                atomRowData.aMat);
      AtomCommMatrixColumn->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);
    }
#endif      
  }
  
  void ForceMatrixDecomposition::collectIntermediateData() {
    snap_ = sman_->getCurrentSnapshot();
    storageLayout_ = sman_->getStorageLayout();
#ifdef IS_MPI
    
    if (storageLayout_ & DataStorage::dslDensity) {
      
      AtomCommRealRow->scatter(atomRowData.density, 
                               snap_->atomData.density);
      
      int n = snap_->atomData.density.size();
      std::vector<RealType> rho_tmp(n, 0.0);
      AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
      for (int i = 0; i < n; i++)
        snap_->atomData.density[i] += rho_tmp[i];
    }
#endif
  }
  
  void ForceMatrixDecomposition::distributeIntermediateData() {
    snap_ = sman_->getCurrentSnapshot();
    storageLayout_ = sman_->getStorageLayout();
#ifdef IS_MPI
    if (storageLayout_ & DataStorage::dslFunctional) {
      AtomCommRealRow->gather(snap_->atomData.functional, 
                              atomRowData.functional);
      AtomCommRealColumn->gather(snap_->atomData.functional, 
                                 atomColData.functional);
    }
    
    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
      AtomCommRealRow->gather(snap_->atomData.functionalDerivative, 
                              atomRowData.functionalDerivative);
      AtomCommRealColumn->gather(snap_->atomData.functionalDerivative, 
                                 atomColData.functionalDerivative);
    }
#endif
  }
  
  
  void ForceMatrixDecomposition::collectData() {
    snap_ = sman_->getCurrentSnapshot();
    storageLayout_ = sman_->getStorageLayout();
#ifdef IS_MPI    
    int n = snap_->atomData.force.size();
    vector<Vector3d> frc_tmp(n, V3Zero);
    
    AtomCommVectorRow->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++)
      snap_->atomData.force[i] += frc_tmp[i];
    
    
    if (storageLayout_ & DataStorage::dslTorque) {

      int nt = snap_->atomData.force.size();
      vector<Vector3d> trq_tmp(nt, V3Zero);

      AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
      for (int i = 0; i < n; 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++)
        snap_->atomData.torque[i] += trq_tmp[i];
    }
    
    nLocal_ = snap_->getNumberOfAtoms();

    vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES, 
                                       vector<RealType> (nLocal_, 0.0));
    
    for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
      AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
      for (int ii = 0;  ii < pot_temp[i].size(); ii++ ) {
        pot_local[i] += pot_temp[i][ii];
      }
    }
#endif
  }

  
  Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
    Vector3d d;
    
#ifdef IS_MPI
    d = cgColData.position[cg2] - cgRowData.position[cg1];
#else
    d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
#endif
    
    snap_->wrapVector(d);
    return d;    
  }


  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){

    Vector3d d;
    
#ifdef IS_MPI
    d = cgRowData.position[cg1] - atomRowData.position[atom1];
#else
    d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
#endif

    snap_->wrapVector(d);
    return d;    
  }
  
  Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
    Vector3d d;
    
#ifdef IS_MPI
    d = cgColData.position[cg2] - atomColData.position[atom2];
#else
    d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
#endif
    
    snap_->wrapVector(d);
    return d;    
  }
    
  Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
    Vector3d d;
    
#ifdef IS_MPI
    d = atomColData.position[atom2] - atomRowData.position[atom1];
#else
    d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
#endif

    snap_->wrapVector(d);
    return d;    
  }

  void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
#ifdef IS_MPI
    atomRowData.force[atom1] += fg;
#else
    snap_->atomData.force[atom1] += fg;
#endif
  }

  void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
#ifdef IS_MPI
    atomColData.force[atom2] += fg;
#else
    snap_->atomData.force[atom2] += fg;
#endif
  }

    // filling interaction blocks with pointers
  InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {    
    InteractionData idat;

#ifdef IS_MPI
    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::dslDensity) {
      idat.rho1 = &(atomRowData.density[atom1]);
      idat.rho2 = &(atomColData.density[atom2]);
    }

    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
      idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
      idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
    }
#else
    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::dslDensity) {
      idat.rho1 = &(snap_->atomData.density[atom1]);
      idat.rho2 = &(snap_->atomData.density[atom2]);
    }

    if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
      idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
      idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
    }
#endif
    return idat;
  }

  InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){

    InteractionData idat;
#ifdef IS_MPI
    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::dslForce) {
      idat.t1 = &(atomRowData.force[atom1]);
      idat.t2 = &(atomColData.force[atom2]);
    }
#else
    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::dslForce) {
      idat.t1 = &(snap_->atomData.force[atom1]);
      idat.t2 = &(snap_->atomData.force[atom2]);
    }
#endif
    
  }


  /*
   * buildNeighborList
   *
   * first element of pair is row-indexed CutoffGroup
   * second element of pair is column-indexed CutoffGroup
   */
  vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
      
    vector<pair<int, int> > neighborList;
#ifdef IS_MPI
    cellListRow_.clear();
    cellListCol_.clear();
#else
    cellList_.clear();
#endif

    // dangerous to not do error checking.
    RealType rCut_;
 
    RealType rList_ = (rCut_ + 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);

    nCells_.x() = (int) ( Hx.length() )/ rList_;
    nCells_.y() = (int) ( Hy.length() )/ rList_;
    nCells_.z() = (int) ( Hz.length() )/ rList_;

    Mat3x3d invHmat = snap_->getInvHmat();
    Vector3d rs, scaled, dr;
    Vector3i whichCell;
    int cellIndex;

#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]);
     
      // 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]);

      // 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]);

      // 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;
      cellList_[cellIndex].push_back(i);
    }
#endif


    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<Vector3i>::iterator os = cellOffsets_.begin();
               os != cellOffsets_.end(); ++os) {
            
            Vector3i m2v = m1v + (*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<int>::iterator j1 = cellListRow_[m1].begin(); 
                 j1 != cellListRow_[m1].end(); ++j1) {
              for (vector<int>::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);
                  if (dr.lengthSquare() < rl2) {
                    neighborList.push_back(make_pair((*j1), (*j2)));
                  }
                }
              }
            }
#else
            for (vector<int>::iterator j1 = cellList_[m1].begin(); 
                 j1 != cellList_[m1].end(); ++j1) {
              for (vector<int>::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);
                  if (dr.lengthSquare() < rl2) {
                    neighborList.push_back(make_pair((*j1), (*j2)));
                  }
                }
              }
            }
#endif
          }
        }
      }
    }

    // save the local cutoff group positions for the check that is
    // done on each loop:
    saved_CG_positions_.clear();
    for (int i = 0; i < nGroups_; i++)
      saved_CG_positions_.push_back(snap_->cgData.position[i]);

    return neighborList;
  }
} //end namespace OpenMD
Revision:	1568
Committed:	Wed May 25 16:20:37 2011 UTC (14 years, 5 months ago) by gezelter
Original Path:	*branches/development/src/parallel/ForceMatrixDecomposition.cpp*
File size:	20524 byte(s)
Log Message:	Added neighbor list check, and migrated skinThickness into ForceDecomposition (and out of the InteractionManager). Removed a spurious inline.
#	Content
1	/*
2	* Copyright (c) 2005 The University of Notre Dame. All Rights Reserved.
3	*
4	* The University of Notre Dame grants you ("Licensee") a
5	* non-exclusive, royalty free, license to use, modify and
6	* redistribute this software in source and binary code form, provided
7	* that the following conditions are met:
8	*
9	* 1. Redistributions of source code must retain the above copyright
10	* notice, this list of conditions and the following disclaimer.
11	*
12	* 2. Redistributions in binary form must reproduce the above copyright
13	* notice, this list of conditions and the following disclaimer in the
14	* documentation and/or other materials provided with the
15	* distribution.
16	*
17	* This software is provided "AS IS," without a warranty of any
18	* kind. All express or implied conditions, representations and
19	* warranties, including any implied warranty of merchantability,
20	* fitness for a particular purpose or non-infringement, are hereby
21	* excluded. The University of Notre Dame and its licensors shall not
22	* be liable for any damages suffered by licensee as a result of
23	* using, modifying or distributing the software or its
24	* derivatives. In no event will the University of Notre Dame or its
25	* licensors be liable for any lost revenue, profit or data, or for
26	* direct, indirect, special, consequential, incidental or punitive
27	* damages, however caused and regardless of the theory of liability,
28	* arising out of the use of or inability to use software, even if the
29	* University of Notre Dame has been advised of the possibility of
30	* such damages.
31	*
32	* SUPPORT OPEN SCIENCE! If you use OpenMD or its source code in your
33	* research, please cite the appropriate papers when you publish your
34	* work. Good starting points are:
35	*
36	* [1] Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).
37	* [2] Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).
38	* [3] Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).
39	* [4] Vardeman & Gezelter, in progress (2009).
40	*/
41	#include "parallel/ForceMatrixDecomposition.hpp"
42	#include "math/SquareMatrix3.hpp"
43	#include "nonbonded/NonBondedInteraction.hpp"
44	#include "brains/SnapshotManager.hpp"
45
46	using namespace std;
47	namespace OpenMD {
48
49	/**
50	* distributeInitialData is essentially a copy of the older fortran
51	* SimulationSetup
52	*/
53
54	void ForceMatrixDecomposition::distributeInitialData() {
55	snap_ = sman_->getCurrentSnapshot();
56	storageLayout_ = sman_->getStorageLayout();
57	nLocal_ = snap_->getNumberOfAtoms();
58	nGroups_ = snap_->getNumberOfCutoffGroups();
59
60	#ifdef IS_MPI
61
62	AtomCommIntRow = new Communicator<Row,int>(nLocal_);
63	AtomCommRealRow = new Communicator<Row,RealType>(nLocal_);
64	AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal_);
65	AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal_);
66
67	AtomCommIntColumn = new Communicator<Column,int>(nLocal_);
68	AtomCommRealColumn = new Communicator<Column,RealType>(nLocal_);
69	AtomCommVectorColumn = new Communicator<Column,Vector3d>(nLocal_);
70	AtomCommMatrixColumn = new Communicator<Column,Mat3x3d>(nLocal_);
71
72	cgCommIntRow = new Communicator<Row,int>(nGroups_);
73	cgCommVectorRow = new Communicator<Row,Vector3d>(nGroups_);
74	cgCommIntColumn = new Communicator<Column,int>(nGroups_);
75	cgCommVectorColumn = new Communicator<Column,Vector3d>(nGroups_);
76
77	nAtomsInRow_ = AtomCommIntRow->getSize();
78	nAtomsInCol_ = AtomCommIntColumn->getSize();
79	nGroupsInRow_ = cgCommIntRow->getSize();
80	nGroupsInCol_ = cgCommIntColumn->getSize();
81
82	// Modify the data storage objects with the correct layouts and sizes:
83	atomRowData.resize(nAtomsInRow_);
84	atomRowData.setStorageLayout(storageLayout_);
85	atomColData.resize(nAtomsInCol_);
86	atomColData.setStorageLayout(storageLayout_);
87	cgRowData.resize(nGroupsInRow_);
88	cgRowData.setStorageLayout(DataStorage::dslPosition);
89	cgColData.resize(nGroupsInCol_);
90	cgColData.setStorageLayout(DataStorage::dslPosition);
91
92	vector<vector<RealType> > pot_row(N_INTERACTION_FAMILIES,
93	vector<RealType> (nAtomsInRow_, 0.0));
94	vector<vector<RealType> > pot_col(N_INTERACTION_FAMILIES,
95	vector<RealType> (nAtomsInCol_, 0.0));
96
97
98	vector<RealType> pot_local(N_INTERACTION_FAMILIES, 0.0);
99
100	// gather the information for atomtype IDs (atids):
101	vector<int> identsLocal = info_->getIdentArray();
102	identsRow.reserve(nAtomsInRow_);
103	identsCol.reserve(nAtomsInCol_);
104
105	AtomCommIntRow->gather(identsLocal, identsRow);
106	AtomCommIntColumn->gather(identsLocal, identsCol);
107
108	AtomLocalToGlobal = info_->getGlobalAtomIndices();
109	AtomCommIntRow->gather(AtomLocalToGlobal, AtomRowToGlobal);
110	AtomCommIntColumn->gather(AtomLocalToGlobal, AtomColToGlobal);
111
112	cgLocalToGlobal = info_->getGlobalGroupIndices();
113	cgCommIntRow->gather(cgLocalToGlobal, cgRowToGlobal);
114	cgCommIntColumn->gather(cgLocalToGlobal, cgColToGlobal);
115
116	// still need:
117	// topoDist
118	// exclude
119	#endif
120	}
121
122
123
124	void ForceMatrixDecomposition::distributeData() {
125	snap_ = sman_->getCurrentSnapshot();
126	storageLayout_ = sman_->getStorageLayout();
127	#ifdef IS_MPI
128
129	// gather up the atomic positions
130	AtomCommVectorRow->gather(snap_->atomData.position,
131	atomRowData.position);
132	AtomCommVectorColumn->gather(snap_->atomData.position,
133	atomColData.position);
134
135	// gather up the cutoff group positions
136	cgCommVectorRow->gather(snap_->cgData.position,
137	cgRowData.position);
138	cgCommVectorColumn->gather(snap_->cgData.position,
139	cgColData.position);
140
141	// if needed, gather the atomic rotation matrices
142	if (storageLayout_ & DataStorage::dslAmat) {
143	AtomCommMatrixRow->gather(snap_->atomData.aMat,
144	atomRowData.aMat);
145	AtomCommMatrixColumn->gather(snap_->atomData.aMat,
146	atomColData.aMat);
147	}
148
149	// if needed, gather the atomic eletrostatic frames
150	if (storageLayout_ & DataStorage::dslElectroFrame) {
151	AtomCommMatrixRow->gather(snap_->atomData.electroFrame,
152	atomRowData.electroFrame);
153	AtomCommMatrixColumn->gather(snap_->atomData.electroFrame,
154	atomColData.electroFrame);
155	}
156	#endif
157	}
158
159	void ForceMatrixDecomposition::collectIntermediateData() {
160	snap_ = sman_->getCurrentSnapshot();
161	storageLayout_ = sman_->getStorageLayout();
162	#ifdef IS_MPI
163
164	if (storageLayout_ & DataStorage::dslDensity) {
165
166	AtomCommRealRow->scatter(atomRowData.density,
167	snap_->atomData.density);
168
169	int n = snap_->atomData.density.size();
170	std::vector<RealType> rho_tmp(n, 0.0);
171	AtomCommRealColumn->scatter(atomColData.density, rho_tmp);
172	for (int i = 0; i < n; i++)
173	snap_->atomData.density[i] += rho_tmp[i];
174	}
175	#endif
176	}
177
178	void ForceMatrixDecomposition::distributeIntermediateData() {
179	snap_ = sman_->getCurrentSnapshot();
180	storageLayout_ = sman_->getStorageLayout();
181	#ifdef IS_MPI
182	if (storageLayout_ & DataStorage::dslFunctional) {
183	AtomCommRealRow->gather(snap_->atomData.functional,
184	atomRowData.functional);
185	AtomCommRealColumn->gather(snap_->atomData.functional,
186	atomColData.functional);
187	}
188
189	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
190	AtomCommRealRow->gather(snap_->atomData.functionalDerivative,
191	atomRowData.functionalDerivative);
192	AtomCommRealColumn->gather(snap_->atomData.functionalDerivative,
193	atomColData.functionalDerivative);
194	}
195	#endif
196	}
197
198
199	void ForceMatrixDecomposition::collectData() {
200	snap_ = sman_->getCurrentSnapshot();
201	storageLayout_ = sman_->getStorageLayout();
202	#ifdef IS_MPI
203	int n = snap_->atomData.force.size();
204	vector<Vector3d> frc_tmp(n, V3Zero);
205
206	AtomCommVectorRow->scatter(atomRowData.force, frc_tmp);
207	for (int i = 0; i < n; i++) {
208	snap_->atomData.force[i] += frc_tmp[i];
209	frc_tmp[i] = 0.0;
210	}
211
212	AtomCommVectorColumn->scatter(atomColData.force, frc_tmp);
213	for (int i = 0; i < n; i++)
214	snap_->atomData.force[i] += frc_tmp[i];
215
216
217	if (storageLayout_ & DataStorage::dslTorque) {
218
219	int nt = snap_->atomData.force.size();
220	vector<Vector3d> trq_tmp(nt, V3Zero);
221
222	AtomCommVectorRow->scatter(atomRowData.torque, trq_tmp);
223	for (int i = 0; i < n; i++) {
224	snap_->atomData.torque[i] += trq_tmp[i];
225	trq_tmp[i] = 0.0;
226	}
227
228	AtomCommVectorColumn->scatter(atomColData.torque, trq_tmp);
229	for (int i = 0; i < n; i++)
230	snap_->atomData.torque[i] += trq_tmp[i];
231	}
232
233	nLocal_ = snap_->getNumberOfAtoms();
234
235	vector<vector<RealType> > pot_temp(N_INTERACTION_FAMILIES,
236	vector<RealType> (nLocal_, 0.0));
237
238	for (int i = 0; i < N_INTERACTION_FAMILIES; i++) {
239	AtomCommRealRow->scatter(pot_row[i], pot_temp[i]);
240	for (int ii = 0; ii < pot_temp[i].size(); ii++ ) {
241	pot_local[i] += pot_temp[i][ii];
242	}
243	}
244	#endif
245	}
246
247
248	Vector3d ForceMatrixDecomposition::getIntergroupVector(int cg1, int cg2){
249	Vector3d d;
250
251	#ifdef IS_MPI
252	d = cgColData.position[cg2] - cgRowData.position[cg1];
253	#else
254	d = snap_->cgData.position[cg2] - snap_->cgData.position[cg1];
255	#endif
256
257	snap_->wrapVector(d);
258	return d;
259	}
260
261
262	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorRow(int atom1, int cg1){
263
264	Vector3d d;
265
266	#ifdef IS_MPI
267	d = cgRowData.position[cg1] - atomRowData.position[atom1];
268	#else
269	d = snap_->cgData.position[cg1] - snap_->atomData.position[atom1];
270	#endif
271
272	snap_->wrapVector(d);
273	return d;
274	}
275
276	Vector3d ForceMatrixDecomposition::getAtomToGroupVectorColumn(int atom2, int cg2){
277	Vector3d d;
278
279	#ifdef IS_MPI
280	d = cgColData.position[cg2] - atomColData.position[atom2];
281	#else
282	d = snap_->cgData.position[cg2] - snap_->atomData.position[atom2];
283	#endif
284
285	snap_->wrapVector(d);
286	return d;
287	}
288
289	Vector3d ForceMatrixDecomposition::getInteratomicVector(int atom1, int atom2){
290	Vector3d d;
291
292	#ifdef IS_MPI
293	d = atomColData.position[atom2] - atomRowData.position[atom1];
294	#else
295	d = snap_->atomData.position[atom2] - snap_->atomData.position[atom1];
296	#endif
297
298	snap_->wrapVector(d);
299	return d;
300	}
301
302	void ForceMatrixDecomposition::addForceToAtomRow(int atom1, Vector3d fg){
303	#ifdef IS_MPI
304	atomRowData.force[atom1] += fg;
305	#else
306	snap_->atomData.force[atom1] += fg;
307	#endif
308	}
309
310	void ForceMatrixDecomposition::addForceToAtomColumn(int atom2, Vector3d fg){
311	#ifdef IS_MPI
312	atomColData.force[atom2] += fg;
313	#else
314	snap_->atomData.force[atom2] += fg;
315	#endif
316	}
317
318	// filling interaction blocks with pointers
319	InteractionData ForceMatrixDecomposition::fillInteractionData(int atom1, int atom2) {
320	InteractionData idat;
321
322	#ifdef IS_MPI
323	if (storageLayout_ & DataStorage::dslAmat) {
324	idat.A1 = &(atomRowData.aMat[atom1]);
325	idat.A2 = &(atomColData.aMat[atom2]);
326	}
327
328	if (storageLayout_ & DataStorage::dslElectroFrame) {
329	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
330	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
331	}
332
333	if (storageLayout_ & DataStorage::dslTorque) {
334	idat.t1 = &(atomRowData.torque[atom1]);
335	idat.t2 = &(atomColData.torque[atom2]);
336	}
337
338	if (storageLayout_ & DataStorage::dslDensity) {
339	idat.rho1 = &(atomRowData.density[atom1]);
340	idat.rho2 = &(atomColData.density[atom2]);
341	}
342
343	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
344	idat.dfrho1 = &(atomRowData.functionalDerivative[atom1]);
345	idat.dfrho2 = &(atomColData.functionalDerivative[atom2]);
346	}
347	#else
348	if (storageLayout_ & DataStorage::dslAmat) {
349	idat.A1 = &(snap_->atomData.aMat[atom1]);
350	idat.A2 = &(snap_->atomData.aMat[atom2]);
351	}
352
353	if (storageLayout_ & DataStorage::dslElectroFrame) {
354	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
355	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
356	}
357
358	if (storageLayout_ & DataStorage::dslTorque) {
359	idat.t1 = &(snap_->atomData.torque[atom1]);
360	idat.t2 = &(snap_->atomData.torque[atom2]);
361	}
362
363	if (storageLayout_ & DataStorage::dslDensity) {
364	idat.rho1 = &(snap_->atomData.density[atom1]);
365	idat.rho2 = &(snap_->atomData.density[atom2]);
366	}
367
368	if (storageLayout_ & DataStorage::dslFunctionalDerivative) {
369	idat.dfrho1 = &(snap_->atomData.functionalDerivative[atom1]);
370	idat.dfrho2 = &(snap_->atomData.functionalDerivative[atom2]);
371	}
372	#endif
373	return idat;
374	}
375
376	InteractionData ForceMatrixDecomposition::fillSkipData(int atom1, int atom2){
377
378	InteractionData idat;
379	#ifdef IS_MPI
380	if (storageLayout_ & DataStorage::dslElectroFrame) {
381	idat.eFrame1 = &(atomRowData.electroFrame[atom1]);
382	idat.eFrame2 = &(atomColData.electroFrame[atom2]);
383	}
384	if (storageLayout_ & DataStorage::dslTorque) {
385	idat.t1 = &(atomRowData.torque[atom1]);
386	idat.t2 = &(atomColData.torque[atom2]);
387	}
388	if (storageLayout_ & DataStorage::dslForce) {
389	idat.t1 = &(atomRowData.force[atom1]);
390	idat.t2 = &(atomColData.force[atom2]);
391	}
392	#else
393	if (storageLayout_ & DataStorage::dslElectroFrame) {
394	idat.eFrame1 = &(snap_->atomData.electroFrame[atom1]);
395	idat.eFrame2 = &(snap_->atomData.electroFrame[atom2]);
396	}
397	if (storageLayout_ & DataStorage::dslTorque) {
398	idat.t1 = &(snap_->atomData.torque[atom1]);
399	idat.t2 = &(snap_->atomData.torque[atom2]);
400	}
401	if (storageLayout_ & DataStorage::dslForce) {
402	idat.t1 = &(snap_->atomData.force[atom1]);
403	idat.t2 = &(snap_->atomData.force[atom2]);
404	}
405	#endif
406
407	}
408
409
410
411
412	/*
413	* buildNeighborList
414	*
415	* first element of pair is row-indexed CutoffGroup
416	* second element of pair is column-indexed CutoffGroup
417	*/
418	vector<pair<int, int> > ForceMatrixDecomposition::buildNeighborList() {
419
420	vector<pair<int, int> > neighborList;
421	#ifdef IS_MPI
422	cellListRow_.clear();
423	cellListCol_.clear();
424	#else
425	cellList_.clear();
426	#endif
427
428	// dangerous to not do error checking.
429	RealType rCut_;
430
431	RealType rList_ = (rCut_ + skinThickness_);
432	RealType rl2 = rList_ * rList_;
433	Snapshot* snap_ = sman_->getCurrentSnapshot();
434	Mat3x3d Hmat = snap_->getHmat();
435	Vector3d Hx = Hmat.getColumn(0);
436	Vector3d Hy = Hmat.getColumn(1);
437	Vector3d Hz = Hmat.getColumn(2);
438
439	nCells_.x() = (int) ( Hx.length() )/ rList_;
440	nCells_.y() = (int) ( Hy.length() )/ rList_;
441	nCells_.z() = (int) ( Hz.length() )/ rList_;
442
443	Mat3x3d invHmat = snap_->getInvHmat();
444	Vector3d rs, scaled, dr;
445	Vector3i whichCell;
446	int cellIndex;
447
448	#ifdef IS_MPI
449	for (int i = 0; i < nGroupsInRow_; i++) {
450	rs = cgRowData.position[i];
451	// scaled positions relative to the box vectors
452	scaled = invHmat * rs;
453	// wrap the vector back into the unit box by subtracting integer box
454	// numbers
455	for (int j = 0; j < 3; j++)
456	scaled[j] -= roundMe(scaled[j]);
457
458	// find xyz-indices of cell that cutoffGroup is in.
459	whichCell.x() = nCells_.x() * scaled.x();
460	whichCell.y() = nCells_.y() * scaled.y();
461	whichCell.z() = nCells_.z() * scaled.z();
462
463	// find single index of this cell:
464	cellIndex = Vlinear(whichCell, nCells_);
465	// add this cutoff group to the list of groups in this cell;
466	cellListRow_[cellIndex].push_back(i);
467	}
468
469	for (int i = 0; i < nGroupsInCol_; i++) {
470	rs = cgColData.position[i];
471	// scaled positions relative to the box vectors
472	scaled = invHmat * rs;
473	// wrap the vector back into the unit box by subtracting integer box
474	// numbers
475	for (int j = 0; j < 3; j++)
476	scaled[j] -= roundMe(scaled[j]);
477
478	// find xyz-indices of cell that cutoffGroup is in.
479	whichCell.x() = nCells_.x() * scaled.x();
480	whichCell.y() = nCells_.y() * scaled.y();
481	whichCell.z() = nCells_.z() * scaled.z();
482
483	// find single index of this cell:
484	cellIndex = Vlinear(whichCell, nCells_);
485	// add this cutoff group to the list of groups in this cell;
486	cellListCol_[cellIndex].push_back(i);
487	}
488	#else
489	for (int i = 0; i < nGroups_; i++) {
490	rs = snap_->cgData.position[i];
491	// scaled positions relative to the box vectors
492	scaled = invHmat * rs;
493	// wrap the vector back into the unit box by subtracting integer box
494	// numbers
495	for (int j = 0; j < 3; j++)
496	scaled[j] -= roundMe(scaled[j]);
497
498	// find xyz-indices of cell that cutoffGroup is in.
499	whichCell.x() = nCells_.x() * scaled.x();
500	whichCell.y() = nCells_.y() * scaled.y();
501	whichCell.z() = nCells_.z() * scaled.z();
502
503	// find single index of this cell:
504	cellIndex = Vlinear(whichCell, nCells_);
505	// add this cutoff group to the list of groups in this cell;
506	cellList_[cellIndex].push_back(i);
507	}
508	#endif
509
510
511
512	for (int m1z = 0; m1z < nCells_.z(); m1z++) {
513	for (int m1y = 0; m1y < nCells_.y(); m1y++) {
514	for (int m1x = 0; m1x < nCells_.x(); m1x++) {
515	Vector3i m1v(m1x, m1y, m1z);
516	int m1 = Vlinear(m1v, nCells_);
517
518	for (vector<Vector3i>::iterator os = cellOffsets_.begin();
519	os != cellOffsets_.end(); ++os) {
520
521	Vector3i m2v = m1v + (*os);
522
523	if (m2v.x() >= nCells_.x()) {
524	m2v.x() = 0;
525	} else if (m2v.x() < 0) {
526	m2v.x() = nCells_.x() - 1;
527	}
528
529	if (m2v.y() >= nCells_.y()) {
530	m2v.y() = 0;
531	} else if (m2v.y() < 0) {
532	m2v.y() = nCells_.y() - 1;
533	}
534
535	if (m2v.z() >= nCells_.z()) {
536	m2v.z() = 0;
537	} else if (m2v.z() < 0) {
538	m2v.z() = nCells_.z() - 1;
539	}
540
541	int m2 = Vlinear (m2v, nCells_);
542
543	#ifdef IS_MPI
544	for (vector<int>::iterator j1 = cellListRow_[m1].begin();
545	j1 != cellListRow_[m1].end(); ++j1) {
546	for (vector<int>::iterator j2 = cellListCol_[m2].begin();
547	j2 != cellListCol_[m2].end(); ++j2) {
548
549	// Always do this if we're in different cells or if
550	// we're in the same cell and the global index of the
551	// j2 cutoff group is less than the j1 cutoff group
552
553	if (m2 != m1 \|\| cgColToGlobal[(j2)] < cgRowToGlobal[(j1)]) {
554	dr = cgColData.position[(j2)] - cgRowData.position[(j1)];
555	snap_->wrapVector(dr);
556	if (dr.lengthSquare() < rl2) {
557	neighborList.push_back(make_pair((j1), (j2)));
558	}
559	}
560	}
561	}
562	#else
563	for (vector<int>::iterator j1 = cellList_[m1].begin();
564	j1 != cellList_[m1].end(); ++j1) {
565	for (vector<int>::iterator j2 = cellList_[m2].begin();
566	j2 != cellList_[m2].end(); ++j2) {
567
568	// Always do this if we're in different cells or if
569	// we're in the same cell and the global index of the
570	// j2 cutoff group is less than the j1 cutoff group
571
572	if (m2 != m1 \|\| (j2) < (j1)) {
573	dr = snap_->cgData.position[(j2)] - snap_->cgData.position[(j1)];
574	snap_->wrapVector(dr);
575	if (dr.lengthSquare() < rl2) {
576	neighborList.push_back(make_pair((j1), (j2)));
577	}
578	}
579	}
580	}
581	#endif
582	}
583	}
584	}
585	}
586
587	// save the local cutoff group positions for the check that is
588	// done on each loop:
589	saved_CG_positions_.clear();
590	for (int i = 0; i < nGroups_; i++)
591	saved_CG_positions_.push_back(snap_->cgData.position[i]);
592
593	return neighborList;
594	}
595	} //end namespace OpenMD