| 1 | 
< | 
/** | 
| 2 | 
< | 
 * @file ForceDecomposition.cpp | 
| 3 | 
< | 
 * @author Charles Vardeman <cvardema.at.nd.edu> | 
| 4 | 
< | 
 * @date 08/18/2010 | 
| 5 | 
< | 
 * @time 11:56am | 
| 6 | 
< | 
 * @version 1.0 | 
| 1 | 
> | 
/* | 
| 2 | 
> | 
 * Copyright (c) 2005 The University of Notre Dame. All Rights Reserved. | 
| 3 | 
  | 
 * | 
| 8 | 
– | 
 * @section LICENSE | 
| 9 | 
– | 
 * Copyright (c) 2010 The University of Notre Dame. All Rights Reserved. | 
| 10 | 
– | 
 * | 
| 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 | 
| 38 | 
  | 
 * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).           | 
| 39 | 
  | 
 * [4]  Vardeman & Gezelter, in progress (2009).                         | 
| 40 | 
  | 
 */ | 
| 41 | 
+ | 
#include "parallel/ForceDecomposition.hpp" | 
| 42 | 
+ | 
#include "parallel/Communicator.hpp" | 
| 43 | 
+ | 
#include "math/SquareMatrix3.hpp" | 
| 44 | 
  | 
 | 
| 45 | 
+ | 
namespace OpenMD { | 
| 46 | 
  | 
 | 
| 47 | 
< | 
 | 
| 51 | 
< | 
/*  -*- c++ -*-  */ | 
| 52 | 
< | 
#include "config.h" | 
| 53 | 
< | 
#include <stdlib.h> | 
| 47 | 
> | 
  void ForceDecomposition::distributeInitialData() { | 
| 48 | 
  | 
#ifdef IS_MPI | 
| 55 | 
– | 
#include <mpi.h> | 
| 56 | 
– | 
#endif | 
| 49 | 
  | 
 | 
| 50 | 
< | 
#include <iostream> | 
| 51 | 
< | 
#include <vector> | 
| 60 | 
< | 
#include <algorithm> | 
| 61 | 
< | 
#include <cmath> | 
| 62 | 
< | 
#include "parallel/ForceDecomposition.hpp" | 
| 50 | 
> | 
    int nAtoms; | 
| 51 | 
> | 
    int nGroups; | 
| 52 | 
  | 
 | 
| 53 | 
+ | 
    AtomCommRealI = new Communicator<Row,RealType>(nAtoms); | 
| 54 | 
+ | 
    AtomCommVectorI = new Communicator<Row,Vector3d>(nAtoms); | 
| 55 | 
+ | 
    AtomCommMatrixI = new Communicator<Row,Mat3x3d>(nAtoms); | 
| 56 | 
  | 
 | 
| 57 | 
< | 
using namespace std; | 
| 58 | 
< | 
using namespace OpenMD; | 
| 57 | 
> | 
    AtomCommRealJ = new Communicator<Column,RealType>(nAtoms); | 
| 58 | 
> | 
    AtomCommVectorJ = new Communicator<Column,Vector3d>(nAtoms); | 
| 59 | 
> | 
    AtomCommMatrixJ = new Communicator<Column,Mat3x3d>(nAtoms); | 
| 60 | 
  | 
 | 
| 61 | 
< | 
//__static | 
| 62 | 
< | 
#ifdef IS_MPI | 
| 63 | 
< | 
static vector<MPI:Comm> communictors; | 
| 61 | 
> | 
    cgCommVectorI = new Communicator<Row,Vector3d>(nGroups); | 
| 62 | 
> | 
    cgCommVectorJ = new Communicator<Column,Vector3d>(nGroups); | 
| 63 | 
> | 
    // more to come | 
| 64 | 
  | 
#endif | 
| 65 | 
+ | 
  } | 
| 66 | 
+ | 
     | 
| 67 | 
  | 
 | 
| 73 | 
– | 
//____ MPITypeTraits | 
| 74 | 
– | 
template<typename T> | 
| 75 | 
– | 
struct MPITypeTraits; | 
| 68 | 
  | 
 | 
| 69 | 
+ | 
  void ForceDecomposition::distributeData()  { | 
| 70 | 
  | 
#ifdef IS_MPI | 
| 71 | 
< | 
template<> | 
| 72 | 
< | 
struct MPITypeTraits<RealType> { | 
| 73 | 
< | 
  static const MPI::Datatype datatype; | 
| 74 | 
< | 
}; | 
| 75 | 
< | 
const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL; | 
| 76 | 
< | 
 | 
| 77 | 
< | 
template<> | 
| 78 | 
< | 
struct MPITypeTraits<int> { | 
| 79 | 
< | 
  static const MPI::Datatype datatype; | 
| 80 | 
< | 
}; | 
| 81 | 
< | 
const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT; | 
| 82 | 
< | 
#endif | 
| 83 | 
< | 
 | 
| 84 | 
< | 
/** | 
| 85 | 
< | 
* Constructor for ForceDecomposition Parallel Decomposition Method | 
| 86 | 
< | 
* Will try to construct a symmetric grid of processors. Ideally, the | 
| 87 | 
< | 
* number of processors will be a square ex: 4, 9, 16, 25. | 
| 88 | 
< | 
* | 
| 89 | 
< | 
*/ | 
| 90 | 
< | 
 | 
| 91 | 
< | 
ForceDecomposition::ForceDecomposition() { | 
| 92 | 
< | 
 | 
| 71 | 
> | 
    Snapshot* snap = sman_->getCurrentSnapshot(); | 
| 72 | 
> | 
     | 
| 73 | 
> | 
    // gather up the atomic positions | 
| 74 | 
> | 
    AtomCommVectorI->gather(snap->atomData.position,  | 
| 75 | 
> | 
                            snap->atomIData.position); | 
| 76 | 
> | 
    AtomCommVectorJ->gather(snap->atomData.position,  | 
| 77 | 
> | 
                            snap->atomJData.position); | 
| 78 | 
> | 
     | 
| 79 | 
> | 
    // gather up the cutoff group positions | 
| 80 | 
> | 
    cgCommVectorI->gather(snap->cgData.position,  | 
| 81 | 
> | 
                          snap->cgIData.position); | 
| 82 | 
> | 
    cgCommVectorJ->gather(snap->cgData.position,  | 
| 83 | 
> | 
                          snap->cgJData.position); | 
| 84 | 
> | 
     | 
| 85 | 
> | 
    // if needed, gather the atomic rotation matrices | 
| 86 | 
> | 
    if (snap->atomData.getStorageLayout() & DataStorage::dslAmat) { | 
| 87 | 
> | 
      AtomCommMatrixI->gather(snap->atomData.aMat,  | 
| 88 | 
> | 
                              snap->atomIData.aMat); | 
| 89 | 
> | 
      AtomCommMatrixJ->gather(snap->atomData.aMat,  | 
| 90 | 
> | 
                              snap->atomJData.aMat); | 
| 91 | 
> | 
    } | 
| 92 | 
> | 
     | 
| 93 | 
> | 
    // if needed, gather the atomic eletrostatic frames | 
| 94 | 
> | 
    if (snap->atomData.getStorageLayout() & DataStorage::dslElectroFrame) { | 
| 95 | 
> | 
      AtomCommMatrixI->gather(snap->atomData.electroFrame,  | 
| 96 | 
> | 
                              snap->atomIData.electroFrame); | 
| 97 | 
> | 
      AtomCommMatrixJ->gather(snap->atomData.electroFrame,  | 
| 98 | 
> | 
                              snap->atomJData.electroFrame); | 
| 99 | 
> | 
    } | 
| 100 | 
> | 
#endif       | 
| 101 | 
> | 
  } | 
| 102 | 
> | 
   | 
| 103 | 
> | 
  void ForceDecomposition::collectIntermediateData() { | 
| 104 | 
  | 
#ifdef IS_MPI | 
| 105 | 
< | 
  int nProcs = MPI::COMM_WORLD.Get_size(); | 
| 106 | 
< | 
  int worldRank = MPI::COMM_WORLD.Get_rank(); | 
| 105 | 
> | 
    Snapshot* snap = sman_->getCurrentSnapshot(); | 
| 106 | 
> | 
     | 
| 107 | 
> | 
    if (snap->atomData.getStorageLayout() & DataStorage::dslDensity) { | 
| 108 | 
> | 
      AtomCommRealI->scatter(snap->atomIData.density,  | 
| 109 | 
> | 
                             snap->atomData.density); | 
| 110 | 
> | 
      std::vector<RealType> rho_tmp; | 
| 111 | 
> | 
      int n = snap->getNumberOfAtoms(); | 
| 112 | 
> | 
      rho_tmp.reserve( n ); | 
| 113 | 
> | 
      AtomCommRealJ->scatter(snap->atomJData.density, rho_tmp); | 
| 114 | 
> | 
      for (int i = 0; i < n; i++) | 
| 115 | 
> | 
        snap->atomData.density[i] += rho_tmp[i]; | 
| 116 | 
> | 
    } | 
| 117 | 
  | 
#endif | 
| 118 | 
< | 
 | 
| 119 | 
< | 
  // First time through, construct column stride. | 
| 120 | 
< | 
  if (communicators.size() == 0) | 
| 121 | 
< | 
  { | 
| 122 | 
< | 
    int nColumnsMax = (int) round(sqrt((float) nProcs)); | 
| 123 | 
< | 
    for (int i = 0; i < nProcs; ++i) | 
| 124 | 
< | 
    { | 
| 125 | 
< | 
      if (nProcs%i==0) nColumns=i; | 
| 118 | 
> | 
  } | 
| 119 | 
> | 
   | 
| 120 | 
> | 
  void ForceDecomposition::distributeIntermediateData() { | 
| 121 | 
> | 
#ifdef IS_MPI | 
| 122 | 
> | 
    Snapshot* snap = sman_->getCurrentSnapshot(); | 
| 123 | 
> | 
    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctional) { | 
| 124 | 
> | 
      AtomCommRealI->gather(snap->atomData.functional,  | 
| 125 | 
> | 
                            snap->atomIData.functional); | 
| 126 | 
> | 
      AtomCommRealJ->gather(snap->atomData.functional,  | 
| 127 | 
> | 
                            snap->atomJData.functional); | 
| 128 | 
  | 
    } | 
| 129 | 
< | 
 | 
| 130 | 
< | 
    int nRows = nProcs/nColumns;     | 
| 131 | 
< | 
    myRank_ = (int) worldRank%nColumns; | 
| 129 | 
> | 
     | 
| 130 | 
> | 
    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctionalDerivative) { | 
| 131 | 
> | 
      AtomCommRealI->gather(snap->atomData.functionalDerivative,  | 
| 132 | 
> | 
                            snap->atomIData.functionalDerivative); | 
| 133 | 
> | 
      AtomCommRealJ->gather(snap->atomData.functionalDerivative,  | 
| 134 | 
> | 
                            snap->atomJData.functionalDerivative); | 
| 135 | 
> | 
    } | 
| 136 | 
> | 
#endif | 
| 137 | 
  | 
  } | 
| 117 | 
– | 
  else | 
| 118 | 
– | 
  { | 
| 119 | 
– | 
    myRank_ = myRank/nColumns; | 
| 120 | 
– | 
  } | 
| 121 | 
– | 
  MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0); | 
| 138 | 
  | 
   | 
| 123 | 
– | 
  isColumn_ = false; | 
| 139 | 
  | 
   | 
| 140 | 
< | 
} | 
| 140 | 
> | 
  void ForceDecomposition::collectData() { | 
| 141 | 
> | 
#ifdef IS_MPI | 
| 142 | 
> | 
    Snapshot* snap = sman_->getCurrentSnapshot(); | 
| 143 | 
> | 
    int n = snap->getNumberOfAtoms(); | 
| 144 | 
  | 
 | 
| 145 | 
< | 
ForceDecomposition::gather(sendbuf, receivebuf){ | 
| 146 | 
< | 
  communicators(myIndex_).Allgatherv(); | 
| 147 | 
< | 
} | 
| 145 | 
> | 
    std::vector<Vector3d> frc_tmp; | 
| 146 | 
> | 
    frc_tmp.reserve( n ); | 
| 147 | 
> | 
     | 
| 148 | 
> | 
    AtomCommVectorI->scatter(snap->atomIData.force, frc_tmp); | 
| 149 | 
> | 
    for (int i = 0; i < n; i++) | 
| 150 | 
> | 
      snap->atomData.force[i] += frc_tmp[i]; | 
| 151 | 
> | 
     | 
| 152 | 
> | 
    AtomCommVectorJ->scatter(snap->atomJData.force, frc_tmp); | 
| 153 | 
> | 
    for (int i = 0; i < n; i++) | 
| 154 | 
> | 
      snap->atomData.force[i] += frc_tmp[i]; | 
| 155 | 
> | 
     | 
| 156 | 
> | 
     | 
| 157 | 
> | 
    if (snap->atomData.getStorageLayout() & DataStorage::dslTorque) { | 
| 158 | 
> | 
      std::vector<Vector3d> trq_tmp; | 
| 159 | 
> | 
      trq_tmp.reserve( n ); | 
| 160 | 
> | 
       | 
| 161 | 
> | 
      AtomCommVectorI->scatter(snap->atomIData.torque, trq_tmp); | 
| 162 | 
> | 
      for (int i = 0; i < n; i++) | 
| 163 | 
> | 
        snap->atomData.torque[i] += trq_tmp[i]; | 
| 164 | 
> | 
       | 
| 165 | 
> | 
      AtomCommVectorJ->scatter(snap->atomJData.torque, trq_tmp); | 
| 166 | 
> | 
      for (int i = 0; i < n; i++) | 
| 167 | 
> | 
        snap->atomData.torque[i] += trq_tmp[i]; | 
| 168 | 
> | 
    } | 
| 169 | 
> | 
     | 
| 170 | 
> | 
    // Still need pot! | 
| 171 | 
> | 
     | 
| 172 | 
  | 
 | 
| 173 | 
  | 
 | 
| 174 | 
< | 
 | 
| 175 | 
< | 
ForceDecomposition::scatter(sbuffer, rbuffer){ | 
| 176 | 
< | 
  communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM); | 
| 177 | 
< | 
} | 
| 136 | 
< | 
 | 
| 137 | 
< | 
 | 
| 174 | 
> | 
#endif | 
| 175 | 
> | 
  } | 
| 176 | 
> | 
   | 
| 177 | 
> | 
} //end namespace OpenMD |