--- branches/development/src/parallel/ForceDecomposition.cpp	2011/01/11 18:58:12	1538
+++ branches/development/src/parallel/ForceMatrixDecomposition.cpp	2011/04/27 18:38:15	1549
@@ -1,13 +1,6 @@
-/**
- * @file ForceDecomposition.cpp
- * @author Charles Vardeman <cvardema.at.nd.edu>
- * @date 08/18/2010
- * @time 11:56am
- * @version 1.0
+/*
+ * Copyright (c) 2005 The University of Notre Dame. All Rights Reserved.
  *
- * @section LICENSE
- * Copyright (c) 2010 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
@@ -45,93 +38,194 @@
  * [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() {
+#ifdef IS_MPI    
+    Snapshot* snap = sman_->getCurrentSnapshot();
+    int nLocal = snap->getNumberOfAtoms();
+    int nGroups = snap->getNumberOfCutoffGroups();
 
-/*  -*- c++ -*-  */
-#include "config.h"
-#include <stdlib.h>
-#ifdef IS_MPI
-#include <mpi.h>
-#endif
+    AtomCommIntRow = new Communicator<Row,int>(nLocal);
+    AtomCommRealRow = new Communicator<Row,RealType>(nLocal);
+    AtomCommVectorRow = new Communicator<Row,Vector3d>(nLocal);
+    AtomCommMatrixRow = new Communicator<Row,Mat3x3d>(nLocal);
 
-#include <iostream>
-#include <vector>
-#include <algorithm>
-#include <cmath>
-#include "parallel/ForceDecomposition.hpp"
+    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);
 
-using namespace std;
-using namespace OpenMD;
+    int nAtomsInRow = AtomCommIntRow->getSize();
+    int nAtomsInCol = AtomCommIntColumn->getSize();
+    int nGroupsInRow = cgCommIntRow->getSize();
+    int nGroupsInCol = cgCommIntColumn->getSize();
+    
+    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);
 
-//__static
-#ifdef IS_MPI
-static vector<MPI:Comm> communictors;
+    // still need:
+    // topoDist
+    // exclude
 #endif
+  }
+    
 
-//____ MPITypeTraits
-template<typename T>
-struct MPITypeTraits;
 
+  void ForceMatrixDecomposition::distributeData()  {
 #ifdef IS_MPI
-template<>
-struct MPITypeTraits<RealType> {
-  static const MPI::Datatype datatype;
-};
-const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL;
+    Snapshot* snap = sman_->getCurrentSnapshot();
+    
+    // gather up the atomic positions
+    AtomCommVectorRow->gather(snap->atomData.position, 
+                            snap->atomIData.position);
+    AtomCommVectorColumn->gather(snap->atomData.position, 
+                            snap->atomJData.position);
+    
+    // gather up the cutoff group positions
+    cgCommVectorRow->gather(snap->cgData.position, 
+                          snap->cgIData.position);
+    cgCommVectorColumn->gather(snap->cgData.position, 
+                          snap->cgJData.position);
+    
+    // if needed, gather the atomic rotation matrices
+    if (snap->atomData.getStorageLayout() & DataStorage::dslAmat) {
+      AtomCommMatrixRow->gather(snap->atomData.aMat, 
+                              snap->atomIData.aMat);
+      AtomCommMatrixColumn->gather(snap->atomData.aMat, 
+                              snap->atomJData.aMat);
+    }
+    
+    // if needed, gather the atomic eletrostatic frames
+    if (snap->atomData.getStorageLayout() & DataStorage::dslElectroFrame) {
+      AtomCommMatrixRow->gather(snap->atomData.electroFrame, 
+                              snap->atomIData.electroFrame);
+      AtomCommMatrixColumn->gather(snap->atomData.electroFrame, 
+                              snap->atomJData.electroFrame);
+    }
+#endif      
+  }
+  
+  void ForceMatrixDecomposition::collectIntermediateData() {
+#ifdef IS_MPI
+    Snapshot* snap = sman_->getCurrentSnapshot();
+    
+    if (snap->atomData.getStorageLayout() & DataStorage::dslDensity) {
 
-template<>
-struct MPITypeTraits<int> {
-  static const MPI::Datatype datatype;
-};
-const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT;
-#endif
+      AtomCommRealRow->scatter(snap->atomIData.density, 
+                             snap->atomData.density);
 
-/**
-* Constructor for ForceDecomposition Parallel Decomposition Method
-* Will try to construct a symmetric grid of processors. Ideally, the
-* number of processors will be a square ex: 4, 9, 16, 25.
-*
-*/
-
-ForceDecomposition::ForceDecomposition() {
-
-#ifdef IS_MPI
-  int nProcs = MPI::COMM_WORLD.Get_size();
-  int worldRank = MPI::COMM_WORLD.Get_rank();
+      int n = snap->atomData.density.size();
+      std::vector<RealType> rho_tmp(n, 0.0);
+      AtomCommRealColumn->scatter(snap->atomJData.density, rho_tmp);
+      for (int i = 0; i < n; i++)
+        snap->atomData.density[i] += rho_tmp[i];
+    }
 #endif
-
-  // First time through, construct column stride.
-  if (communicators.size() == 0)
-  {
-    int nColumnsMax = (int) round(sqrt((float) nProcs));
-    for (int i = 0; i < nProcs; ++i)
-    {
-      if (nProcs%i==0) nColumns=i;
+  }
+  
+  void ForceMatrixDecomposition::distributeIntermediateData() {
+#ifdef IS_MPI
+    Snapshot* snap = sman_->getCurrentSnapshot();
+    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctional) {
+      AtomCommRealRow->gather(snap->atomData.functional, 
+                            snap->atomIData.functional);
+      AtomCommRealColumn->gather(snap->atomData.functional, 
+                            snap->atomJData.functional);
     }
-
-    int nRows = nProcs/nColumns;    
-    myRank_ = (int) worldRank%nColumns;
+    
+    if (snap->atomData.getStorageLayout() & DataStorage::dslFunctionalDerivative) {
+      AtomCommRealRow->gather(snap->atomData.functionalDerivative, 
+                            snap->atomIData.functionalDerivative);
+      AtomCommRealColumn->gather(snap->atomData.functionalDerivative, 
+                            snap->atomJData.functionalDerivative);
+    }
+#endif
   }
-  else
-  {
-    myRank_ = myRank/nColumns;
-  }
-  MPI::Comm newComm = MPI:COMM_WORLD.Split(myRank_,0);
   
-  isColumn_ = false;
   
-}
+  void ForceMatrixDecomposition::collectData() {
+#ifdef IS_MPI
+    Snapshot* snap = sman_->getCurrentSnapshot();
+    
+    int n = snap->atomData.force.size();
+    vector<Vector3d> frc_tmp(n, V3Zero);
+    
+    AtomCommVectorRow->scatter(snap->atomIData.force, frc_tmp);
+    for (int i = 0; i < n; i++) {
+      snap->atomData.force[i] += frc_tmp[i];
+      frc_tmp[i] = 0.0;
+    }
+    
+    AtomCommVectorColumn->scatter(snap->atomJData.force, frc_tmp);
+    for (int i = 0; i < n; i++)
+      snap->atomData.force[i] += frc_tmp[i];
+    
+    
+    if (snap->atomData.getStorageLayout() & DataStorage::dslTorque) {
 
-ForceDecomposition::gather(sendbuf, receivebuf){
-  communicators(myIndex_).Allgatherv();
-}
+      int nt = snap->atomData.force.size();
+      vector<Vector3d> trq_tmp(nt, V3Zero);
 
+      AtomCommVectorRow->scatter(snap->atomIData.torque, trq_tmp);
+      for (int i = 0; i < n; i++) {
+        snap->atomData.torque[i] += trq_tmp[i];
+        trq_tmp[i] = 0.0;
+      }
+      
+      AtomCommVectorColumn->scatter(snap->atomJData.torque, trq_tmp);
+      for (int i = 0; i < n; i++)
+        snap->atomData.torque[i] += trq_tmp[i];
+    }
+    
+    int nLocal = snap->getNumberOfAtoms();
 
-
-ForceDecomposition::scatter(sbuffer, rbuffer){
-  communicators(myIndex_).Reduce_scatter(sbuffer, recevbuf. recvcounts, MPI::DOUBLE, MPI::SUM);
-}
-
-
+    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
+  }
+  
+} //end namespace OpenMD