| 1 | 
chuckv | 
1538 | 
/** | 
| 2 | 
  | 
  | 
 * @file Parallel.cpp | 
| 3 | 
  | 
  | 
 * @author Charles Vardeman <cvardema.at.nd.edu> | 
| 4 | 
  | 
  | 
 * @date 08/18/2010 | 
| 5 | 
  | 
  | 
 * @time 11:56am | 
| 6 | 
  | 
  | 
 * @version 1.0 | 
| 7 | 
  | 
  | 
 * | 
| 8 | 
  | 
  | 
 * @section LICENSE | 
| 9 | 
  | 
  | 
 * Copyright (C) 2010 The University of Notre Dame. All Rights Reserved. | 
| 10 | 
  | 
  | 
 * | 
| 11 | 
  | 
  | 
 * | 
| 12 | 
  | 
  | 
 * The University of Notre Dame grants you ("Licensee") a | 
| 13 | 
  | 
  | 
 * non-exclusive, royalty free, license to use, modify and | 
| 14 | 
  | 
  | 
 * redistribute this software in source and binary code form, provided | 
| 15 | 
  | 
  | 
 * that the following conditions are met: | 
| 16 | 
  | 
  | 
 * | 
| 17 | 
  | 
  | 
 * 1. Redistributions of source code must retain the above copyright | 
| 18 | 
  | 
  | 
 *    notice, this list of conditions and the following disclaimer. | 
| 19 | 
  | 
  | 
 * | 
| 20 | 
  | 
  | 
 * 2. Redistributions in binary form must reproduce the above copyright | 
| 21 | 
  | 
  | 
 *    notice, this list of conditions and the following disclaimer in the | 
| 22 | 
  | 
  | 
 *    documentation and/or other materials provided with the | 
| 23 | 
  | 
  | 
 *    distribution. | 
| 24 | 
  | 
  | 
 * | 
| 25 | 
  | 
  | 
 * This software is provided "AS IS," without a warranty of any | 
| 26 | 
  | 
  | 
 * kind. All express or implied conditions, representations and | 
| 27 | 
  | 
  | 
 * warranties, including any implied warranty of merchantability, | 
| 28 | 
  | 
  | 
 * fitness for a particular purpose or non-infringement, are hereby | 
| 29 | 
  | 
  | 
 * excluded.  The University of Notre Dame and its licensors shall not | 
| 30 | 
  | 
  | 
 * be liable for any damages suffered by licensee as a result of | 
| 31 | 
  | 
  | 
 * using, modifying or distributing the software or its | 
| 32 | 
  | 
  | 
 * derivatives. In no event will the University of Notre Dame or its | 
| 33 | 
  | 
  | 
 * licensors be liable for any lost revenue, profit or data, or for | 
| 34 | 
  | 
  | 
 * direct, indirect, special, consequential, incidental or punitive | 
| 35 | 
  | 
  | 
 * damages, however caused and regardless of the theory of liability, | 
| 36 | 
  | 
  | 
 * arising out of the use of or inability to use software, even if the | 
| 37 | 
  | 
  | 
 * University of Notre Dame has been advised of the possibility of | 
| 38 | 
  | 
  | 
 * such damages. | 
| 39 | 
  | 
  | 
 * | 
| 40 | 
  | 
  | 
 * SUPPORT OPEN SCIENCE!  If you use OpenMD or its source code in your | 
| 41 | 
  | 
  | 
 * research, please cite the appropriate papers when you publish your | 
| 42 | 
  | 
  | 
 * work.  Good starting points are: | 
| 43 | 
  | 
  | 
 *                                                                       | 
| 44 | 
  | 
  | 
 * [1]  Meineke, et al., J. Comp. Chem. 26, 252-271 (2005).              | 
| 45 | 
  | 
  | 
 * [2]  Fennell & Gezelter, J. Chem. Phys. 124, 234104 (2006).           | 
| 46 | 
  | 
  | 
 * [3]  Sun, Lin & Gezelter, J. Chem. Phys. 128, 24107 (2008).           | 
| 47 | 
  | 
  | 
 * [4]  Vardeman & Gezelter, in progress (2009).                   | 
| 48 | 
  | 
  | 
 */ | 
| 49 | 
  | 
  | 
 | 
| 50 | 
  | 
  | 
 | 
| 51 | 
  | 
  | 
#include <stdlib.h> | 
| 52 | 
  | 
  | 
#include "config.h" | 
| 53 | 
  | 
  | 
 | 
| 54 | 
  | 
  | 
#ifdef IS_MPI | 
| 55 | 
  | 
  | 
#include <mpi.h> | 
| 56 | 
  | 
  | 
#endif | 
| 57 | 
  | 
  | 
 | 
| 58 | 
  | 
  | 
#include <iostream> | 
| 59 | 
  | 
  | 
#include <vector> | 
| 60 | 
  | 
  | 
#include <algorithm> | 
| 61 | 
  | 
  | 
 | 
| 62 | 
  | 
  | 
#include "parallel/Parallel.hpp" | 
| 63 | 
  | 
  | 
 | 
| 64 | 
  | 
  | 
 | 
| 65 | 
  | 
  | 
using namespace std; | 
| 66 | 
  | 
  | 
using namespace OpenMD; | 
| 67 | 
  | 
  | 
 | 
| 68 | 
  | 
  | 
//#define DEBUG_PARALLEL | 
| 69 | 
  | 
  | 
 | 
| 70 | 
  | 
  | 
 | 
| 71 | 
  | 
  | 
#ifdef SINGLE_PRECISION | 
| 72 | 
  | 
  | 
#define MY_MPI_REAL MPI_FLOAT | 
| 73 | 
  | 
  | 
#else | 
| 74 | 
  | 
  | 
#define MY_MPI_REAL MPI_DOUBLE | 
| 75 | 
  | 
  | 
#endif | 
| 76 | 
  | 
  | 
 | 
| 77 | 
  | 
  | 
 | 
| 78 | 
  | 
  | 
 | 
| 79 | 
  | 
  | 
 | 
| 80 | 
  | 
  | 
//____ mpiAbort | 
| 81 | 
  | 
  | 
static void mpiAbort(); | 
| 82 | 
  | 
  | 
 | 
| 83 | 
  | 
  | 
void mpiAbort() { | 
| 84 | 
  | 
  | 
  if (Parallel::ok()) { | 
| 85 | 
  | 
  | 
#ifdef IS_MPI | 
| 86 | 
  | 
  | 
    MPI::Comm_world.Abort(1); | 
| 87 | 
  | 
  | 
#endif | 
| 88 | 
  | 
  | 
  } | 
| 89 | 
  | 
  | 
  exit(EXIT_FAILURE); | 
| 90 | 
  | 
  | 
} | 
| 91 | 
  | 
  | 
 | 
| 92 | 
  | 
  | 
//____ mpiExit | 
| 93 | 
  | 
  | 
static void mpiExit(); | 
| 94 | 
  | 
  | 
 | 
| 95 | 
  | 
  | 
void mpiExit() { | 
| 96 | 
  | 
  | 
  if (Parallel::ok()) | 
| 97 | 
  | 
  | 
    Parallel::finalize(); | 
| 98 | 
  | 
  | 
  exit(EXIT_SUCCESS); | 
| 99 | 
  | 
  | 
} | 
| 100 | 
  | 
  | 
 | 
| 101 | 
  | 
  | 
 | 
| 102 | 
  | 
  | 
//____ MPITypeTraits | 
| 103 | 
  | 
  | 
template<typename T> | 
| 104 | 
  | 
  | 
struct MPITypeTraits; | 
| 105 | 
  | 
  | 
 | 
| 106 | 
  | 
  | 
#ifdef IS_MPI | 
| 107 | 
  | 
  | 
template<> | 
| 108 | 
  | 
  | 
struct MPITypeTraits<RealType> { | 
| 109 | 
  | 
  | 
  static const MPI::Datatype datatype; | 
| 110 | 
  | 
  | 
}; | 
| 111 | 
  | 
  | 
const MPI_Datatype MPITypeTraits<RealType>::datatype = MY_MPI_REAL; | 
| 112 | 
  | 
  | 
 | 
| 113 | 
  | 
  | 
template<> | 
| 114 | 
  | 
  | 
struct MPITypeTraits<int> { | 
| 115 | 
  | 
  | 
  static const MPI::Datatype datatype; | 
| 116 | 
  | 
  | 
}; | 
| 117 | 
  | 
  | 
const MPI::Datatype MPITypeTraits<int>::datatype = MPI_INT; | 
| 118 | 
  | 
  | 
 | 
| 119 | 
  | 
  | 
 | 
| 120 | 
  | 
  | 
//____ allReduceScalar | 
| 121 | 
  | 
  | 
template<typename T> | 
| 122 | 
  | 
  | 
void allReduceScalar(T &begin) { | 
| 123 | 
  | 
  | 
  T tmp = begin; | 
| 124 | 
  | 
  | 
  /* | 
| 125 | 
  | 
  | 
  MPI_Allreduce(&tmp, &begin, 1, MPITypeTraits<T>::datatype, MPI_SUM, | 
| 126 | 
  | 
  | 
                (exludeMaster ? slaveComm : MPI_COMM_WORLD)); | 
| 127 | 
  | 
  | 
  */ | 
| 128 | 
  | 
  | 
  MPI::Comm_world.Allreduce(&tmp,&begin,1, MPITypeTraits<T>::datatype, MPI::SUM); | 
| 129 | 
  | 
  | 
} | 
| 130 | 
  | 
  | 
 | 
| 131 | 
  | 
  | 
//____ allReduce | 
| 132 | 
  | 
  | 
template<typename T> | 
| 133 | 
  | 
  | 
void allReduce(T *begin, T *end) { | 
| 134 | 
  | 
  | 
  vector<T> tmp(end - begin); | 
| 135 | 
  | 
  | 
  copy(begin, end, tmp.begin()); | 
| 136 | 
  | 
  | 
  /*  MPI_Allreduce(&(tmp[0]), begin, (end - begin), MPITypeTraits<T>::datatype, | 
| 137 | 
  | 
  | 
                MPI_SUM, | 
| 138 | 
  | 
  | 
                (exludeMaster ? slaveComm : MPI_COMM_WORLD)); | 
| 139 | 
  | 
  | 
  */ | 
| 140 | 
  | 
  | 
   MPI::Comm_world.Allreduce(&tmp[0],&begin,1, MPITypeTraits<T>::datatype, MPI::SUM); | 
| 141 | 
  | 
  | 
} | 
| 142 | 
  | 
  | 
 | 
| 143 | 
  | 
  | 
template<bool exludeMaster, bool dobarrier> | 
| 144 | 
  | 
  | 
void allReduce(Vector3DBlock *coords) { | 
| 145 | 
  | 
  | 
  allReduce<exludeMaster, dobarrier>(&(coords->begin()->x), | 
| 146 | 
  | 
  | 
                                     &(coords->end()->x)); | 
| 147 | 
  | 
  | 
} | 
| 148 | 
  | 
  | 
 | 
| 149 | 
  | 
  | 
template<bool exludeMaster, bool dobarrier> | 
| 150 | 
  | 
  | 
void allReduce(ScalarStructure *energies) { | 
| 151 | 
  | 
  | 
  allReduce<exludeMaster, dobarrier>( | 
| 152 | 
  | 
  | 
    &((*energies)[ScalarStructure::FIRST]), | 
| 153 | 
  | 
  | 
    &((*energies)[ScalarStructure::LASTREDUCE])); | 
| 154 | 
  | 
  | 
} | 
| 155 | 
  | 
  | 
 | 
| 156 | 
  | 
  | 
//____ broadcastScalar | 
| 157 | 
  | 
  | 
template<bool exludeMaster, bool dobarrier, typename T> | 
| 158 | 
  | 
  | 
void broadcastScalar(T &begin) { | 
| 159 | 
  | 
  | 
  if (dobarrier) | 
| 160 | 
  | 
  | 
    doBarrier<exludeMaster>(); | 
| 161 | 
  | 
  | 
  MPI_Bcast(&begin, 1, MPITypeTraits<T>::datatype, master, | 
| 162 | 
  | 
  | 
            (exludeMaster ? slaveComm : MPI_COMM_WORLD)); | 
| 163 | 
  | 
  | 
} | 
| 164 | 
  | 
  | 
 | 
| 165 | 
  | 
  | 
//____ broadcast | 
| 166 | 
  | 
  | 
 | 
| 167 | 
  | 
  | 
template<bool exludeMaster, bool dobarrier, typename T> | 
| 168 | 
  | 
  | 
void broadcast(T *begin, T *end) { | 
| 169 | 
  | 
  | 
  if (dobarrier) | 
| 170 | 
  | 
  | 
    doBarrier<exludeMaster>(); | 
| 171 | 
  | 
  | 
  MPI_Bcast(begin, (end - begin), MPITypeTraits<T>::datatype, | 
| 172 | 
  | 
  | 
            (exludeMaster ? 0 : master), | 
| 173 | 
  | 
  | 
            (exludeMaster ? slaveComm : MPI_COMM_WORLD)); | 
| 174 | 
  | 
  | 
} | 
| 175 | 
  | 
  | 
 | 
| 176 | 
  | 
  | 
template<bool exludeMaster, bool dobarrier> | 
| 177 | 
  | 
  | 
void broadcast(Vector3DBlock *coords) { | 
| 178 | 
  | 
  | 
  broadcast<exludeMaster, dobarrier>(&(coords->begin()->x), | 
| 179 | 
  | 
  | 
                                     &(coords->end()->x)); | 
| 180 | 
  | 
  | 
} | 
| 181 | 
  | 
  | 
 | 
| 182 | 
  | 
  | 
template<bool exludeMaster, bool dobarrier> | 
| 183 | 
  | 
  | 
void broadcast(ScalarStructure *energies) { | 
| 184 | 
  | 
  | 
  broadcast<exludeMaster, dobarrier>( | 
| 185 | 
  | 
  | 
    &((*energies)[ScalarStructure::FIRST]), | 
| 186 | 
  | 
  | 
    &((*energies)[ScalarStructure::LASTREDUCE])); | 
| 187 | 
  | 
  | 
} | 
| 188 | 
  | 
  | 
 | 
| 189 | 
  | 
  | 
#endif | 
| 190 | 
  | 
  | 
 | 
| 191 | 
  | 
  | 
//____ Parallel | 
| 192 | 
  | 
  | 
 | 
| 193 | 
  | 
  | 
#ifdef IS_MPI | 
| 194 | 
  | 
  | 
const bool Parallel::isMPI = true; | 
| 195 | 
  | 
  | 
#else | 
| 196 | 
  | 
  | 
const bool Parallel::isMPI = false; | 
| 197 | 
  | 
  | 
#endif | 
| 198 | 
  | 
  | 
 | 
| 199 | 
  | 
  | 
bool Parallel::myInitialized = false; | 
| 200 | 
  | 
  | 
bool Parallel::myFinalized = false; | 
| 201 | 
  | 
  | 
int Parallel::myId = 0; | 
| 202 | 
  | 
  | 
int Parallel::myMasterId = master; | 
| 203 | 
  | 
  | 
int Parallel::myNum = 1; | 
| 204 | 
  | 
  | 
int Parallel::myAvailableId = 0; | 
| 205 | 
  | 
  | 
int Parallel::myAvailableNum = 1; | 
| 206 | 
  | 
  | 
bool Parallel::myIsParallel = Parallel::isMPI; | 
| 207 | 
  | 
  | 
bool Parallel::myIAmMaster = true; | 
| 208 | 
  | 
  | 
bool Parallel::myIAmSlave = true; | 
| 209 | 
  | 
  | 
ParallelType Parallel::myMode = ParallelType::STATIC; | 
| 210 | 
  | 
  | 
Parallel::WorkState Parallel::myWorkState = Parallel::SEQUENTIAL; | 
| 211 | 
  | 
  | 
 | 
| 212 | 
  | 
  | 
int Parallel::myPipeSize = 1; | 
| 213 | 
  | 
  | 
bool Parallel::myUseBarrier = false; | 
| 214 | 
  | 
  | 
int Parallel::myMaxPackages = -1; | 
| 215 | 
  | 
  | 
 | 
| 216 | 
  | 
  | 
int *Parallel::myBuffer = NULL; | 
| 217 | 
  | 
  | 
int Parallel::myNext = 0; | 
| 218 | 
  | 
  | 
int Parallel::myNextRange[2] = { | 
| 219 | 
  | 
  | 
  0, 0 | 
| 220 | 
  | 
  | 
}; | 
| 221 | 
  | 
  | 
vector<int>  Parallel::myDone; | 
| 222 | 
  | 
  | 
vector<int>  Parallel::myBlockList; | 
| 223 | 
  | 
  | 
 | 
| 224 | 
  | 
  | 
int Parallel::myRecv; | 
| 225 | 
  | 
  | 
int Parallel::myI; | 
| 226 | 
  | 
  | 
int Parallel::myP; | 
| 227 | 
  | 
  | 
 | 
| 228 | 
  | 
  | 
Parallel * Parallel::obj = NULL; | 
| 229 | 
  | 
  | 
 | 
| 230 | 
  | 
  | 
int Parallel::myOldId = 0; | 
| 231 | 
  | 
  | 
int Parallel::myOldNum = 1; | 
| 232 | 
  | 
  | 
ParallelType Parallel::myOldMode = ParallelType::STATIC; | 
| 233 | 
  | 
  | 
bool Parallel::myIsolated = false; | 
| 234 | 
  | 
  | 
 | 
| 235 | 
  | 
  | 
//____ Parallel | 
| 236 | 
  | 
  | 
 | 
| 237 | 
  | 
  | 
Parallel::~Parallel() { | 
| 238 | 
  | 
  | 
} | 
| 239 | 
  | 
  | 
 | 
| 240 | 
  | 
  | 
Parallel::Parallel() { | 
| 241 | 
  | 
  | 
  myInitialized = false; | 
| 242 | 
  | 
  | 
  myFinalized = false; | 
| 243 | 
  | 
  | 
  myId = 0; | 
| 244 | 
  | 
  | 
  myMasterId = master; | 
| 245 | 
  | 
  | 
  myNum = 1; | 
| 246 | 
  | 
  | 
  myAvailableId = 0; | 
| 247 | 
  | 
  | 
  myAvailableNum = 1; | 
| 248 | 
  | 
  | 
  myIsParallel = Parallel::isMPI; | 
| 249 | 
  | 
  | 
  myIAmMaster = true; | 
| 250 | 
  | 
  | 
  myIAmSlave = true; | 
| 251 | 
  | 
  | 
  myMode = ParallelType::DYNAMIC; | 
| 252 | 
  | 
  | 
 | 
| 253 | 
  | 
  | 
  myPipeSize = 1; | 
| 254 | 
  | 
  | 
  myUseBarrier = false; | 
| 255 | 
  | 
  | 
  myMaxPackages = -1; | 
| 256 | 
  | 
  | 
 | 
| 257 | 
  | 
  | 
  myBuffer = NULL; | 
| 258 | 
  | 
  | 
  myNext = 0; | 
| 259 | 
  | 
  | 
  myNextRange[0] = 0; | 
| 260 | 
  | 
  | 
  myNextRange[1] = 0; | 
| 261 | 
  | 
  | 
  myDone.clear(); | 
| 262 | 
  | 
  | 
  myBlockList.clear(); | 
| 263 | 
  | 
  | 
 | 
| 264 | 
  | 
  | 
  myOldId = 0; | 
| 265 | 
  | 
  | 
  myOldNum = 1; | 
| 266 | 
  | 
  | 
  myOldMode = ParallelType::STATIC; | 
| 267 | 
  | 
  | 
  myIsolated = false; | 
| 268 | 
  | 
  | 
} | 
| 269 | 
  | 
  | 
 | 
| 270 | 
  | 
  | 
Parallel &Parallel::instance() { | 
| 271 | 
  | 
  | 
  // We have to do it ourself ... M$ problem ... | 
| 272 | 
  | 
  | 
  if (obj == NULL) { | 
| 273 | 
  | 
  | 
    obj = new Parallel(); | 
| 274 | 
  | 
  | 
    atexit(kill); | 
| 275 | 
  | 
  | 
  } | 
| 276 | 
  | 
  | 
  return *obj; | 
| 277 | 
  | 
  | 
} | 
| 278 | 
  | 
  | 
 | 
| 279 | 
  | 
  | 
void Parallel::kill() { | 
| 280 | 
  | 
  | 
  Parallel *p = obj; | 
| 281 | 
  | 
  | 
  obj = NULL; | 
| 282 | 
  | 
  | 
  p->~Parallel(); | 
| 283 | 
  | 
  | 
} | 
| 284 | 
  | 
  | 
 | 
| 285 | 
  | 
  | 
void Parallel::init(int &argc, char ** &argv) { | 
| 286 | 
  | 
  | 
  instance(); | 
| 287 | 
  | 
  | 
  if (!myInitialized && !myFinalized) { | 
| 288 | 
  | 
  | 
#ifdef IS_MPI | 
| 289 | 
  | 
  | 
    MPI_Init(&argc, &argv); | 
| 290 | 
  | 
  | 
    MPI_Comm_size(MPI_COMM_WORLD, &myNum); | 
| 291 | 
  | 
  | 
    MPI_Comm_rank(MPI_COMM_WORLD, &myId); | 
| 292 | 
  | 
  | 
    MPI_Barrier(MPI_COMM_WORLD); | 
| 293 | 
  | 
  | 
    setOpenMDStartSerial(mpiStartSerial); | 
| 294 | 
  | 
  | 
    setOpenMDEndSerial(mpiEndSerial); | 
| 295 | 
  | 
  | 
#endif | 
| 296 | 
  | 
  | 
    setOpenMDAbort(mpiAbort); | 
| 297 | 
  | 
  | 
    setOpenMDExit(mpiExit); | 
| 298 | 
  | 
  | 
    myInitialized = true; | 
| 299 | 
  | 
  | 
    myIsParallel = (myNum > 1); | 
| 300 | 
  | 
  | 
    myIAmMaster = (myId == myMasterId); | 
| 301 | 
  | 
  | 
    report.setIAmMaster(iAmMaster()); | 
| 302 | 
  | 
  | 
    setMode(isParallel() ? ParallelType::DYNAMIC : ParallelType::STATIC); | 
| 303 | 
  | 
  | 
    setPipeSize(myPipeSize); | 
| 304 | 
  | 
  | 
    TimerStatistic::setParallel(myIsParallel); | 
| 305 | 
  | 
  | 
    if (iAmMaster() && isParallel()) | 
| 306 | 
  | 
  | 
      myDone.resize(myNum); | 
| 307 | 
  | 
  | 
 | 
| 308 | 
  | 
  | 
    report << plain << (isMPI ? "Using MPI." : "No MPI compilation.") << endr; | 
| 309 | 
  | 
  | 
  } else | 
| 310 | 
  | 
  | 
  if (iAmMaster()) | 
| 311 | 
  | 
  | 
    report << recoverable << "MPI is" << | 
| 312 | 
  | 
  | 
    ((myInitialized) ? " " : " not ") << "initialized and is" << | 
| 313 | 
  | 
  | 
    ((myFinalized) ? " " : " not ") << | 
| 314 | 
  | 
  | 
    "finalized. Called [Parallel::init].\n" << endr; | 
| 315 | 
  | 
  | 
} | 
| 316 | 
  | 
  | 
 | 
| 317 | 
  | 
  | 
void Parallel::finalize() { | 
| 318 | 
  | 
  | 
  if (ok("Called [Parallel::finalize].")) { | 
| 319 | 
  | 
  | 
    setOpenMDAbort(NULL); | 
| 320 | 
  | 
  | 
    setOpenMDExit(NULL); | 
| 321 | 
  | 
  | 
    setOpendMDStartSerial(NULL); | 
| 322 | 
  | 
  | 
    setOpenMDEndSerial(NULL); | 
| 323 | 
  | 
  | 
#ifdef IS_MPI | 
| 324 | 
  | 
  | 
    if (iAmSlave()) | 
| 325 | 
  | 
  | 
      MPI_Comm_free(&slaveComm); | 
| 326 | 
  | 
  | 
 | 
| 327 | 
  | 
  | 
    if (myBuffer != NULL) { | 
| 328 | 
  | 
  | 
      int size; | 
| 329 | 
  | 
  | 
      MPI_Buffer_detach(myBuffer, &size); | 
| 330 | 
  | 
  | 
      delete[] myBuffer; | 
| 331 | 
  | 
  | 
      myBuffer = NULL; | 
| 332 | 
  | 
  | 
    } | 
| 333 | 
  | 
  | 
    FFTComplex::FFTComplexMPIFinalize(); | 
| 334 | 
  | 
  | 
    MPI_Finalize(); | 
| 335 | 
  | 
  | 
#endif | 
| 336 | 
  | 
  | 
    myFinalized = true; | 
| 337 | 
  | 
  | 
  } | 
| 338 | 
  | 
  | 
} | 
| 339 | 
  | 
  | 
 | 
| 340 | 
  | 
  | 
bool Parallel::ok(const string &err) { | 
| 341 | 
  | 
  | 
  if (ok()) | 
| 342 | 
  | 
  | 
    return true; | 
| 343 | 
  | 
  | 
 | 
| 344 | 
  | 
  | 
  if (iAmMaster()) | 
| 345 | 
  | 
  | 
    report << recoverable << "MPI is" << ((myInitialized) ? " " : " not ") << | 
| 346 | 
  | 
  | 
    "initialized and is" << ((myFinalized) ? " " : " not ") << | 
| 347 | 
  | 
  | 
    "finalized. " << err << "\n" << endr; | 
| 348 | 
  | 
  | 
  return false; | 
| 349 | 
  | 
  | 
} | 
| 350 | 
  | 
  | 
 | 
| 351 | 
  | 
  | 
void Parallel::setMode(ParallelType mode) { | 
| 352 | 
  | 
  | 
  if (!ok("Called [Parallel::setMasterSlave]")) | 
| 353 | 
  | 
  | 
    return; | 
| 354 | 
  | 
  | 
#ifndef IS_MPI | 
| 355 | 
  | 
  | 
  mode = ParallelType::STATIC; | 
| 356 | 
  | 
  | 
#endif | 
| 357 | 
  | 
  | 
  if (!isParallel() || mode == ParallelType::UNDEFINED) | 
| 358 | 
  | 
  | 
    mode = ParallelType::STATIC; | 
| 359 | 
  | 
  | 
 | 
| 360 | 
  | 
  | 
  myMode = mode; | 
| 361 | 
  | 
  | 
  myAvailableNum = (mode == ParallelType::MASTERSLAVE ? myNum - 1  : myNum); | 
| 362 | 
  | 
  | 
  myIAmSlave = (mode == ParallelType::MASTERSLAVE ? (!myIAmMaster) : | 
| 363 | 
  | 
  | 
                true); | 
| 364 | 
  | 
  | 
  myAvailableId = | 
| 365 | 
  | 
  | 
    (myNum == myAvailableNum ? myId : | 
| 366 | 
  | 
  | 
     (myId == myMasterId ? -1 : (myId < myMasterId ? myId : myId - 1))); | 
| 367 | 
  | 
  | 
#ifdef IS_MPI | 
| 368 | 
  | 
  | 
  if (slaveComm != MPI_COMM_NULL) | 
| 369 | 
  | 
  | 
    MPI_Comm_free(&slaveComm); | 
| 370 | 
  | 
  | 
 | 
| 371 | 
  | 
  | 
  // Create intracommunicator only with slaves | 
| 372 | 
  | 
  | 
  MPI_Group worldGroup = MPI_GROUP_NULL; | 
| 373 | 
  | 
  | 
  MPI_Group slaveGroup = MPI_GROUP_NULL; | 
| 374 | 
  | 
  | 
  int excl[] = { | 
| 375 | 
  | 
  | 
    myMasterId | 
| 376 | 
  | 
  | 
  }; | 
| 377 | 
  | 
  | 
 | 
| 378 | 
  | 
  | 
  MPI_Comm_group(MPI_COMM_WORLD, &worldGroup); | 
| 379 | 
  | 
  | 
  MPI_Group_excl(worldGroup, myNum - myAvailableNum, excl, &slaveGroup); | 
| 380 | 
  | 
  | 
  MPI_Comm_create(MPI_COMM_WORLD, slaveGroup, &slaveComm); | 
| 381 | 
  | 
  | 
  MPI_Group_free(&worldGroup); | 
| 382 | 
  | 
  | 
  MPI_Group_free(&slaveGroup); | 
| 383 | 
  | 
  | 
  if (slaveComm != MPI_COMM_NULL) | 
| 384 | 
  | 
  | 
    MPI_Comm_rank(slaveComm, &myAvailableId); | 
| 385 | 
  | 
  | 
  FFTComplex::FFTComplexMPIInit( | 
| 386 | 
  | 
  | 
    mode == ParallelType::MASTERSLAVE ? myMasterId : -1); | 
| 387 | 
  | 
  | 
#endif | 
| 388 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 389 | 
  | 
  | 
  report << allnodesserial << "ParallelMode (" << getId << ") : " | 
| 390 | 
  | 
  | 
         << getMode().getString() << endr; | 
| 391 | 
  | 
  | 
#endif | 
| 392 | 
  | 
  | 
} | 
| 393 | 
  | 
  | 
 | 
| 394 | 
  | 
  | 
void Parallel::setPipeSize(int n) { | 
| 395 | 
  | 
  | 
  if (!ok("Called [Parallel::setPipeSize]")) | 
| 396 | 
  | 
  | 
    return; | 
| 397 | 
  | 
  | 
  myPipeSize = max(n, 0); | 
| 398 | 
  | 
  | 
 | 
| 399 | 
  | 
  | 
#ifdef IS_MPI | 
| 400 | 
  | 
  | 
  if (myBuffer != NULL) { | 
| 401 | 
  | 
  | 
    int size; | 
| 402 | 
  | 
  | 
    MPI_Buffer_detach(myBuffer, &size); | 
| 403 | 
  | 
  | 
    delete[] myBuffer; | 
| 404 | 
  | 
  | 
    myBuffer = NULL; | 
| 405 | 
  | 
  | 
  } | 
| 406 | 
  | 
  | 
 | 
| 407 | 
  | 
  | 
  // Allocate buffer for Bsend() | 
| 408 | 
  | 
  | 
  const int size = 3 * 2 * myNum * (myPipeSize + 1) + MPI_BSEND_OVERHEAD + 10; | 
| 409 | 
  | 
  | 
  myBuffer = new int[size]; | 
| 410 | 
  | 
  | 
  for (int i = 0; i < size; i++) | 
| 411 | 
  | 
  | 
    myBuffer[i] = 0; | 
| 412 | 
  | 
  | 
 | 
| 413 | 
  | 
  | 
  MPI_Buffer_attach(myBuffer, static_cast<int>(size * sizeof(int))); | 
| 414 | 
  | 
  | 
#endif | 
| 415 | 
  | 
  | 
} | 
| 416 | 
  | 
  | 
 | 
| 417 | 
  | 
  | 
void Parallel::setMaxPackages(int n) { | 
| 418 | 
  | 
  | 
  if (!ok("Called [Parallel::setMaxPackages]")) | 
| 419 | 
  | 
  | 
    return; | 
| 420 | 
  | 
  | 
  myMaxPackages = n < 0 ? (isDynamic() ? 3 : 0) : | 
| 421 | 
  | 
  | 
                  n; | 
| 422 | 
  | 
  | 
} | 
| 423 | 
  | 
  | 
 | 
| 424 | 
  | 
  | 
Parallel::WorkState Parallel::getWorkState() { | 
| 425 | 
  | 
  | 
  if (!isParallel()) | 
| 426 | 
  | 
  | 
    return SEQUENTIAL; | 
| 427 | 
  | 
  | 
  else if (myMode == ParallelType::STATIC) | 
| 428 | 
  | 
  | 
    return STATIC; | 
| 429 | 
  | 
  | 
  else if (iAmMaster() && myMode == ParallelType::DYNAMIC) | 
| 430 | 
  | 
  | 
    return MASTER; | 
| 431 | 
  | 
  | 
  else | 
| 432 | 
  | 
  | 
    return SLAVE; | 
| 433 | 
  | 
  | 
} | 
| 434 | 
  | 
  | 
 | 
| 435 | 
  | 
  | 
void Parallel::sync() { | 
| 436 | 
  | 
  | 
#ifdef IS_MPI | 
| 437 | 
  | 
  | 
  MPI_Barrier(MPI_COMM_WORLD); | 
| 438 | 
  | 
  | 
#endif | 
| 439 | 
  | 
  | 
} | 
| 440 | 
  | 
  | 
 | 
| 441 | 
  | 
  | 
void Parallel::syncSlave() { | 
| 442 | 
  | 
  | 
#ifdef IS_MPI | 
| 443 | 
  | 
  | 
  if (!iAmSlave() || !isParallel()) | 
| 444 | 
  | 
  | 
    return; | 
| 445 | 
  | 
  | 
  MPI_Barrier(slaveComm); | 
| 446 | 
  | 
  | 
#endif | 
| 447 | 
  | 
  | 
} | 
| 448 | 
  | 
  | 
 | 
| 449 | 
  | 
  | 
// Senda a vector 3DBlock over MPI as an array. | 
| 450 | 
  | 
  | 
#ifdef IS_MPI | 
| 451 | 
  | 
  | 
void Parallel::send(Vector3DBlock *vect, int address) { | 
| 452 | 
  | 
  | 
  /* Create a C-style array large enough to hold all the real values | 
| 453 | 
  | 
  | 
   * (3 / Vector3D) | 
| 454 | 
  | 
  | 
   */ | 
| 455 | 
  | 
  | 
  int size = vect->size(); | 
| 456 | 
  | 
  | 
  Real *vectArray = new Real[3 * size]; | 
| 457 | 
  | 
  | 
  if (vectArray == 0) { | 
| 458 | 
  | 
  | 
    cout << "Can't create Parallel::send() array! Quitting!" << endl; | 
| 459 | 
  | 
  | 
    MPI_Abort(MPI_COMM_WORLD, 1); | 
| 460 | 
  | 
  | 
  } | 
| 461 | 
  | 
  | 
 | 
| 462 | 
  | 
  | 
  /* Start dumping the Vector3D into the array.  This keeps us from having to | 
| 463 | 
  | 
  | 
   * create an MPI struct, pack it,     * and in general, mess with all that. | 
| 464 | 
  | 
  | 
   * It's inefficient, but it's at least a start. It's designed such that | 
| 465 | 
  | 
  | 
   * the array looks like {x1 y1 z1 x2 y2 z2 ... xN yN zN} for N Vector3D's */ | 
| 466 | 
  | 
  | 
  for (int i = 0; i < size; i++) { | 
| 467 | 
  | 
  | 
    vectArray[3 * i] = (*vect)[i][0]; | 
| 468 | 
  | 
  | 
    vectArray[3 * i + 1] = (*vect)[i][1]; | 
| 469 | 
  | 
  | 
    vectArray[3 * i + 2] = (*vect)[i][2]; | 
| 470 | 
  | 
  | 
  } | 
| 471 | 
  | 
  | 
 | 
| 472 | 
  | 
  | 
  /* Since it's an array of Reals, we can use the plain old Parallel::send | 
| 473 | 
  | 
  | 
   * routine since MPI can handle | 
| 474 | 
  | 
  | 
   * both single values and arrays with the same function call */ | 
| 475 | 
  | 
  | 
  send(vectArray, 3 * size, address); | 
| 476 | 
  | 
  | 
  delete[] vectArray; | 
| 477 | 
  | 
  | 
} | 
| 478 | 
  | 
  | 
 | 
| 479 | 
  | 
  | 
#else | 
| 480 | 
  | 
  | 
void Parallel::send(Vector3DBlock *, int) {} | 
| 481 | 
  | 
  | 
 | 
| 482 | 
  | 
  | 
#endif | 
| 483 | 
  | 
  | 
 | 
| 484 | 
  | 
  | 
// Same philosophy as Parallel::send, except we're receiving an array | 
| 485 | 
  | 
  | 
#ifdef IS_MPI | 
| 486 | 
  | 
  | 
void Parallel::recv(Vector3DBlock *vect, int address) { | 
| 487 | 
  | 
  | 
  /* Create an array of proper size. */ | 
| 488 | 
  | 
  | 
  int size = vect->size(); | 
| 489 | 
  | 
  | 
  Real *vectArray = new Real[3 * size]; | 
| 490 | 
  | 
  | 
  if (vectArray == 0) { | 
| 491 | 
  | 
  | 
    cout << "Can't create Parallel::recv() array!  Quitting!" << endl; | 
| 492 | 
  | 
  | 
    MPI_Abort(MPI_COMM_WORLD, 1); | 
| 493 | 
  | 
  | 
  } | 
| 494 | 
  | 
  | 
  /* Since it's an array, the Parallel::recv() call is sufficient (don't you | 
| 495 | 
  | 
  | 
   * just love function overloading?) */ | 
| 496 | 
  | 
  | 
  recv(vectArray, 3 * size, address); | 
| 497 | 
  | 
  | 
  /* Map the vector back onto an actual Vector3DBlock.  The array looks like | 
| 498 | 
  | 
  | 
   * this: {x1 y1 z1 x2 y2 z2 ... xN yN zN} */ | 
| 499 | 
  | 
  | 
  for (int i = 0; i < size; i++) { | 
| 500 | 
  | 
  | 
    (*vect)[i][0] = vectArray[3 * i]; | 
| 501 | 
  | 
  | 
    (*vect)[i][1] = vectArray[3 * i + 1]; | 
| 502 | 
  | 
  | 
    (*vect)[i][2] = vectArray[3 * i + 2]; | 
| 503 | 
  | 
  | 
  } | 
| 504 | 
  | 
  | 
 | 
| 505 | 
  | 
  | 
  delete[] vectArray; | 
| 506 | 
  | 
  | 
} | 
| 507 | 
  | 
  | 
 | 
| 508 | 
  | 
  | 
#else | 
| 509 | 
  | 
  | 
void Parallel::recv(Vector3DBlock *, int) {} | 
| 510 | 
  | 
  | 
 | 
| 511 | 
  | 
  | 
#endif | 
| 512 | 
  | 
  | 
 | 
| 513 | 
  | 
  | 
// Overwrites the Vector3D with a new one after sending it to another node | 
| 514 | 
  | 
  | 
#ifdef IS_MPI | 
| 515 | 
  | 
  | 
void Parallel::sendrecv_replace(Vector3DBlock *vect, int sendaddr, | 
| 516 | 
  | 
  | 
                                int recvaddr) { | 
| 517 | 
  | 
  | 
  int size = vect->size(); | 
| 518 | 
  | 
  | 
  Real *vectArray = new Real[3 * size]; | 
| 519 | 
  | 
  | 
  if (vectArray == 0) { | 
| 520 | 
  | 
  | 
    cout << "Can't create Parallel::send() array! Quitting!" << endl; | 
| 521 | 
  | 
  | 
    MPI_Abort(MPI_COMM_WORLD, 1); | 
| 522 | 
  | 
  | 
  } | 
| 523 | 
  | 
  | 
  // Familiar mapping strategy... | 
| 524 | 
  | 
  | 
  for (int i = 0; i < size; i++) { | 
| 525 | 
  | 
  | 
    vectArray[3 * i] = (*vect)[i][0]; | 
| 526 | 
  | 
  | 
    vectArray[3 * i + 1] = (*vect)[i][1]; | 
| 527 | 
  | 
  | 
    vectArray[3 * i + 2] = (*vect)[i][2]; | 
| 528 | 
  | 
  | 
  } | 
| 529 | 
  | 
  | 
 | 
| 530 | 
  | 
  | 
  // Reuse of function calls to actually handle the MPI calls | 
| 531 | 
  | 
  | 
  sendrecv_replace(vectArray, 3 * size, sendaddr, recvaddr); | 
| 532 | 
  | 
  | 
  // Map it back in to the Vector3D and nobody will ever know we mucked with | 
| 533 | 
  | 
  | 
  // it... 8-) | 
| 534 | 
  | 
  | 
  for (int i = 0; i < size; i++) { | 
| 535 | 
  | 
  | 
    (*vect)[i][0] = vectArray[3 * i]; | 
| 536 | 
  | 
  | 
    (*vect)[i][1] = vectArray[3 * i + 1]; | 
| 537 | 
  | 
  | 
    (*vect)[i][2] = vectArray[3 * i + 2]; | 
| 538 | 
  | 
  | 
  } | 
| 539 | 
  | 
  | 
 | 
| 540 | 
  | 
  | 
  delete[] vectArray; | 
| 541 | 
  | 
  | 
} | 
| 542 | 
  | 
  | 
 | 
| 543 | 
  | 
  | 
#else | 
| 544 | 
  | 
  | 
void Parallel::sendrecv_replace(Vector3DBlock *, int, int) {} | 
| 545 | 
  | 
  | 
 | 
| 546 | 
  | 
  | 
#endif | 
| 547 | 
  | 
  | 
 | 
| 548 | 
  | 
  | 
#ifdef IS_MPI | 
| 549 | 
  | 
  | 
void Parallel::send(Real *data, int num, int address) { | 
| 550 | 
  | 
  | 
  // Just a nice wrapper that automatically selects the MPI datatype for you | 
| 551 | 
  | 
  | 
  // and handles all the annoying things | 
| 552 | 
  | 
  | 
  MPI_Send(data, num, MPITypeTraits<Real>::datatype, address, 0, | 
| 553 | 
  | 
  | 
           MPI_COMM_WORLD); | 
| 554 | 
  | 
  | 
} | 
| 555 | 
  | 
  | 
 | 
| 556 | 
  | 
  | 
#else | 
| 557 | 
  | 
  | 
void Parallel::send(Real *, int, int) {} | 
| 558 | 
  | 
  | 
 | 
| 559 | 
  | 
  | 
#endif | 
| 560 | 
  | 
  | 
 | 
| 561 | 
  | 
  | 
#ifdef IS_MPI | 
| 562 | 
  | 
  | 
void Parallel::recv(Real *data, int num, int address) { | 
| 563 | 
  | 
  | 
  // Another MPI wrapper function.... | 
| 564 | 
  | 
  | 
  MPI_Status status; | 
| 565 | 
  | 
  | 
  MPI_Recv(data, num, MPITypeTraits<Real>::datatype, address, 0, | 
| 566 | 
  | 
  | 
           MPI_COMM_WORLD, | 
| 567 | 
  | 
  | 
           &status); | 
| 568 | 
  | 
  | 
} | 
| 569 | 
  | 
  | 
 | 
| 570 | 
  | 
  | 
#else | 
| 571 | 
  | 
  | 
void Parallel::recv(Real *, int, int) {} | 
| 572 | 
  | 
  | 
 | 
| 573 | 
  | 
  | 
#endif | 
| 574 | 
  | 
  | 
 | 
| 575 | 
  | 
  | 
#ifdef IS_MPI | 
| 576 | 
  | 
  | 
void Parallel::sendrecv(Real *senddata, int sendnum, int sendaddr, | 
| 577 | 
  | 
  | 
                        Real *recvdata, int recvnum, | 
| 578 | 
  | 
  | 
                        int recvaddr) { | 
| 579 | 
  | 
  | 
  MPI_Status status; | 
| 580 | 
  | 
  | 
  MPI_Sendrecv(senddata, sendnum, MPITypeTraits<Real>::datatype, sendaddr, 0, | 
| 581 | 
  | 
  | 
               recvdata, recvnum, MPITypeTraits<Real>::datatype, recvaddr, 0, | 
| 582 | 
  | 
  | 
               MPI_COMM_WORLD, | 
| 583 | 
  | 
  | 
               &status); | 
| 584 | 
  | 
  | 
} | 
| 585 | 
  | 
  | 
 | 
| 586 | 
  | 
  | 
#else | 
| 587 | 
  | 
  | 
void Parallel::sendrecv(Real *, int, int, Real *, int, int) {} | 
| 588 | 
  | 
  | 
 | 
| 589 | 
  | 
  | 
#endif | 
| 590 | 
  | 
  | 
 | 
| 591 | 
  | 
  | 
#ifdef IS_MPI | 
| 592 | 
  | 
  | 
void Parallel::sendrecv_replace(Real *data, int num, int sendaddr, | 
| 593 | 
  | 
  | 
                                int recvaddr) { | 
| 594 | 
  | 
  | 
  MPI_Status status; | 
| 595 | 
  | 
  | 
  MPI_Sendrecv_replace(data, num, MPITypeTraits<Real>::datatype, sendaddr, 0, | 
| 596 | 
  | 
  | 
                       recvaddr, 0, MPI_COMM_WORLD, | 
| 597 | 
  | 
  | 
                       &status); | 
| 598 | 
  | 
  | 
} | 
| 599 | 
  | 
  | 
 | 
| 600 | 
  | 
  | 
#else | 
| 601 | 
  | 
  | 
void Parallel::sendrecv_replace(Real *, int, int, int) {} | 
| 602 | 
  | 
  | 
 | 
| 603 | 
  | 
  | 
#endif | 
| 604 | 
  | 
  | 
 | 
| 605 | 
  | 
  | 
#ifdef IS_MPI | 
| 606 | 
  | 
  | 
void Parallel::gather(Real *data, int num, Real *data_array, int address) { | 
| 607 | 
  | 
  | 
  MPI_Gather(data, num, MPITypeTraits<Real>::datatype, data_array, num, | 
| 608 | 
  | 
  | 
             MPITypeTraits<Real>::datatype, address, | 
| 609 | 
  | 
  | 
             MPI_COMM_WORLD); | 
| 610 | 
  | 
  | 
} | 
| 611 | 
  | 
  | 
 | 
| 612 | 
  | 
  | 
#else | 
| 613 | 
  | 
  | 
void Parallel::gather(Real *, int, Real *, int) {} | 
| 614 | 
  | 
  | 
 | 
| 615 | 
  | 
  | 
#endif | 
| 616 | 
  | 
  | 
 | 
| 617 | 
  | 
  | 
#ifdef IS_MPI | 
| 618 | 
  | 
  | 
void Parallel::allgather(Real *data, int num, Real *data_array) { | 
| 619 | 
  | 
  | 
  MPI_Allgather(data, num, MPITypeTraits<Real>::datatype, data_array, num, | 
| 620 | 
  | 
  | 
                MPITypeTraits<Real>::datatype, | 
| 621 | 
  | 
  | 
                MPI_COMM_WORLD); | 
| 622 | 
  | 
  | 
} | 
| 623 | 
  | 
  | 
 | 
| 624 | 
  | 
  | 
#else | 
| 625 | 
  | 
  | 
void Parallel::allgather(Real *, int, Real *) {} | 
| 626 | 
  | 
  | 
 | 
| 627 | 
  | 
  | 
#endif | 
| 628 | 
  | 
  | 
 | 
| 629 | 
  | 
  | 
#ifdef IS_MPI | 
| 630 | 
  | 
  | 
void Parallel::reduceSlaves(Real *begin, Real *end) { | 
| 631 | 
  | 
  | 
  if (!iAmSlave() || !isParallel()) | 
| 632 | 
  | 
  | 
    return; | 
| 633 | 
  | 
  | 
  TimerStatistic::timer[TimerStatistic::COMMUNICATION].start(); | 
| 634 | 
  | 
  | 
 | 
| 635 | 
  | 
  | 
  allReduce<true, true>(begin, end); | 
| 636 | 
  | 
  | 
 | 
| 637 | 
  | 
  | 
  TimerStatistic::timer[TimerStatistic::COMMUNICATION].stop(); | 
| 638 | 
  | 
  | 
} | 
| 639 | 
  | 
  | 
 | 
| 640 | 
  | 
  | 
#else | 
| 641 | 
  | 
  | 
void Parallel::reduceSlaves(Real *, Real *) {} | 
| 642 | 
  | 
  | 
 | 
| 643 | 
  | 
  | 
#endif | 
| 644 | 
  | 
  | 
 | 
| 645 | 
  | 
  | 
#ifdef IS_MPI | 
| 646 | 
  | 
  | 
void Parallel::distribute(ScalarStructure *energies, Vector3DBlock *coords) { | 
| 647 | 
  | 
  | 
  if (!iAmMaster() && !energies->distributed()) { | 
| 648 | 
  | 
  | 
    energies->clear(); | 
| 649 | 
  | 
  | 
    coords->zero(); | 
| 650 | 
  | 
  | 
  }- | 
| 651 | 
  | 
  | 
  energies->distribute(); | 
| 652 | 
  | 
  | 
  coords->distribute(); | 
| 653 | 
  | 
  | 
} | 
| 654 | 
  | 
  | 
 | 
| 655 | 
  | 
  | 
#else | 
| 656 | 
  | 
  | 
void Parallel::distribute(ScalarStructure *, Vector3DBlock *) {} | 
| 657 | 
  | 
  | 
 | 
| 658 | 
  | 
  | 
#endif | 
| 659 | 
  | 
  | 
 | 
| 660 | 
  | 
  | 
#ifdef IS_MPI | 
| 661 | 
  | 
  | 
void Parallel::reduce(ScalarStructure *energies, Vector3DBlock *coords) { | 
| 662 | 
  | 
  | 
  energies->reduce(); | 
| 663 | 
  | 
  | 
  coords->reduce(); | 
| 664 | 
  | 
  | 
  if (!isParallel() || energies->distributed()) | 
| 665 | 
  | 
  | 
    return; | 
| 666 | 
  | 
  | 
  TimerStatistic::timer[TimerStatistic::COMMUNICATION].start(); | 
| 667 | 
  | 
  | 
 | 
| 668 | 
  | 
  | 
  allReduce<false, true>(coords); | 
| 669 | 
  | 
  | 
  allReduce<false, false>(energies); | 
| 670 | 
  | 
  | 
 | 
| 671 | 
  | 
  | 
  TimerStatistic::timer[TimerStatistic::COMMUNICATION].stop(); | 
| 672 | 
  | 
  | 
} | 
| 673 | 
  | 
  | 
 | 
| 674 | 
  | 
  | 
#else | 
| 675 | 
  | 
  | 
void Parallel::reduce(ScalarStructure *, Vector3DBlock *) {} | 
| 676 | 
  | 
  | 
 | 
| 677 | 
  | 
  | 
#endif | 
| 678 | 
  | 
  | 
 | 
| 679 | 
  | 
  | 
#ifdef IS_MPI | 
| 680 | 
  | 
  | 
void Parallel::bcast(Vector3DBlock *coords) { | 
| 681 | 
  | 
  | 
  if (!isParallel()) | 
| 682 | 
  | 
  | 
    return; | 
| 683 | 
  | 
  | 
  broadcast<false, true>(coords); | 
| 684 | 
  | 
  | 
} | 
| 685 | 
  | 
  | 
 | 
| 686 | 
  | 
  | 
#else | 
| 687 | 
  | 
  | 
void Parallel::bcast(Vector3DBlock *) {} | 
| 688 | 
  | 
  | 
 | 
| 689 | 
  | 
  | 
#endif | 
| 690 | 
  | 
  | 
 | 
| 691 | 
  | 
  | 
#ifdef IS_MPI | 
| 692 | 
  | 
  | 
void Parallel::bcast(int &n) { | 
| 693 | 
  | 
  | 
  if (!isParallel()) | 
| 694 | 
  | 
  | 
    return; | 
| 695 | 
  | 
  | 
  broadcastScalar<false, true>(n); | 
| 696 | 
  | 
  | 
} | 
| 697 | 
  | 
  | 
 | 
| 698 | 
  | 
  | 
#else | 
| 699 | 
  | 
  | 
void Parallel::bcast(int &) {} | 
| 700 | 
  | 
  | 
 | 
| 701 | 
  | 
  | 
#endif | 
| 702 | 
  | 
  | 
 | 
| 703 | 
  | 
  | 
#ifdef IS_MPI | 
| 704 | 
  | 
  | 
void Parallel::bcastSlaves(Real *begin, Real *end) { | 
| 705 | 
  | 
  | 
  if (!iAmSlave() || !isParallel()) | 
| 706 | 
  | 
  | 
    return; | 
| 707 | 
  | 
  | 
  TimerStatistic::timer[TimerStatistic::COMMUNICATION].start(); | 
| 708 | 
  | 
  | 
 | 
| 709 | 
  | 
  | 
  broadcast<true, true>(begin, end); | 
| 710 | 
  | 
  | 
 | 
| 711 | 
  | 
  | 
  TimerStatistic::timer[TimerStatistic::COMMUNICATION].stop(); | 
| 712 | 
  | 
  | 
} | 
| 713 | 
  | 
  | 
 | 
| 714 | 
  | 
  | 
#else | 
| 715 | 
  | 
  | 
void Parallel::bcastSlaves(Real *, Real *) {} | 
| 716 | 
  | 
  | 
 | 
| 717 | 
  | 
  | 
#endif | 
| 718 | 
  | 
  | 
 | 
| 719 | 
  | 
  | 
unsigned int Parallel::getNumberOfPackages(unsigned int n) { | 
| 720 | 
  | 
  | 
  if (getMaxPackages() < 1 || | 
| 721 | 
  | 
  | 
      static_cast<unsigned int>(getAvailableNum()) >= n) | 
| 722 | 
  | 
  | 
    return n; | 
| 723 | 
  | 
  | 
  return min(n / getAvailableNum(), | 
| 724 | 
  | 
  | 
             static_cast<unsigned int>(getMaxPackages())) | 
| 725 | 
  | 
  | 
         * static_cast<unsigned int>(getAvailableNum()); | 
| 726 | 
  | 
  | 
} | 
| 727 | 
  | 
  | 
 | 
| 728 | 
  | 
  | 
void Parallel::resetNext(const vector<int> &blocks) { | 
| 729 | 
  | 
  | 
  resetNext(); | 
| 730 | 
  | 
  | 
  if (!iAmMaster()) | 
| 731 | 
  | 
  | 
    return; | 
| 732 | 
  | 
  | 
#ifdef IS_MPI | 
| 733 | 
  | 
  | 
  if (blocks.empty()) | 
| 734 | 
  | 
  | 
    return; | 
| 735 | 
  | 
  | 
 | 
| 736 | 
  | 
  | 
  // Vector of ranges (n0,n1,n3, ... ,nM, -1, -1, ..., -1) | 
| 737 | 
  | 
  | 
  // n0,n1,n3, ... ,nM are the numbers a slave will call next() | 
| 738 | 
  | 
  | 
  // for one given force | 
| 739 | 
  | 
  | 
  // A slave will call next() and check if the | 
| 740 | 
  | 
  | 
  // actual number of next()-calls is inside the | 
| 741 | 
  | 
  | 
  // range received by the master and compute if true. | 
| 742 | 
  | 
  | 
  int n = 0; | 
| 743 | 
  | 
  | 
  for (unsigned int i = 0; i < blocks.size(); i++) | 
| 744 | 
  | 
  | 
    n += blocks[i]; | 
| 745 | 
  | 
  | 
 | 
| 746 | 
  | 
  | 
  if (n < 1) | 
| 747 | 
  | 
  | 
    return; | 
| 748 | 
  | 
  | 
 | 
| 749 | 
  | 
  | 
  myBlockList.resize(n + 1 + myAvailableNum + 1); | 
| 750 | 
  | 
  | 
  for (unsigned int i = 0; static_cast<int>(i) <= n; i++) | 
| 751 | 
  | 
  | 
    myBlockList[i] = i; | 
| 752 | 
  | 
  | 
 | 
| 753 | 
  | 
  | 
  // Adding ranges (-1,-1) to indicate that there is no more | 
| 754 | 
  | 
  | 
  // to compute. | 
| 755 | 
  | 
  | 
  for (unsigned int i = n + 1; i < myBlockList.size(); i++) | 
| 756 | 
  | 
  | 
    myBlockList[i] = -1; | 
| 757 | 
  | 
  | 
 | 
| 758 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 759 | 
  | 
  | 
  report << allnodes << plain << "Blocks :"; | 
| 760 | 
  | 
  | 
  for (unsigned int i = 0; i < myBlockList.size(); i++) | 
| 761 | 
  | 
  | 
    report << myBlockList[i] << " "; | 
| 762 | 
  | 
  | 
 | 
| 763 | 
  | 
  | 
  report << endr; | 
| 764 | 
  | 
  | 
#endif | 
| 765 | 
  | 
  | 
  // Fill up the pipe ... | 
| 766 | 
  | 
  | 
  myI = 0;          // Actual range index | 
| 767 | 
  | 
  | 
  myRecv = 1;     // Number of pending recieves | 
| 768 | 
  | 
  | 
  // One more since one slave will send a request but not call | 
| 769 | 
  | 
  | 
  // recv for the last range | 
| 770 | 
  | 
  | 
  int stop = 0;     // | 
| 771 | 
  | 
  | 
  myP = 0; | 
| 772 | 
  | 
  | 
  myDone[myId] = 0; | 
| 773 | 
  | 
  | 
  for (int j = 0; j < (myPipeSize + 1) * myNum; j++) { | 
| 774 | 
  | 
  | 
    if (stop == myNum - 1) | 
| 775 | 
  | 
  | 
      break; | 
| 776 | 
  | 
  | 
 | 
| 777 | 
  | 
  | 
    int myP = j % myNum; | 
| 778 | 
  | 
  | 
    if (myP == myMasterId) | 
| 779 | 
  | 
  | 
      continue; | 
| 780 | 
  | 
  | 
 | 
| 781 | 
  | 
  | 
    MPI_Bsend(&(myBlockList[myI]), 2, MPI_INT, myP, SEND_RANGE, | 
| 782 | 
  | 
  | 
              MPI_COMM_WORLD); | 
| 783 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 784 | 
  | 
  | 
    report << allnodes << plain << "Block[" << myI << "] = [" << | 
| 785 | 
  | 
  | 
    myBlockList[myI] << "," << | 
| 786 | 
  | 
  | 
    myBlockList[myI + 1] << "] to " << myP << "." << endr; | 
| 787 | 
  | 
  | 
#endif | 
| 788 | 
  | 
  | 
    myDone[myP] = (myBlockList[myI + 2] < 0 ? 1 : 0); | 
| 789 | 
  | 
  | 
    if (myDone[myP] != 0) | 
| 790 | 
  | 
  | 
      stop++; | 
| 791 | 
  | 
  | 
    if (myDone[myP] == 0) | 
| 792 | 
  | 
  | 
      myRecv++; | 
| 793 | 
  | 
  | 
    myI++; | 
| 794 | 
  | 
  | 
  } | 
| 795 | 
  | 
  | 
 | 
| 796 | 
  | 
  | 
  if (myMode == ParallelType::MASTERSLAVE) | 
| 797 | 
  | 
  | 
    nextMaster(); | 
| 798 | 
  | 
  | 
#endif | 
| 799 | 
  | 
  | 
} | 
| 800 | 
  | 
  | 
 | 
| 801 | 
  | 
  | 
void Parallel::nextMaster() { | 
| 802 | 
  | 
  | 
#ifdef IS_MPI | 
| 803 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 804 | 
  | 
  | 
  report << allnodes << plain << "Parallel::nextMaster Recv " << myRecv << | 
| 805 | 
  | 
  | 
  "." << endr; | 
| 806 | 
  | 
  | 
#endif | 
| 807 | 
  | 
  | 
  while (myRecv > 0) { | 
| 808 | 
  | 
  | 
    int test = 0; | 
| 809 | 
  | 
  | 
    MPI_Status status; | 
| 810 | 
  | 
  | 
    char tmp[1]; | 
| 811 | 
  | 
  | 
 | 
| 812 | 
  | 
  | 
    // Receiving the request for a new range from a slave. | 
| 813 | 
  | 
  | 
    // We wait until we got a msg. Note that waiting for ANY_SOURCE | 
| 814 | 
  | 
  | 
    // could lead to a starving of some nodes. | 
| 815 | 
  | 
  | 
    while (!test) { | 
| 816 | 
  | 
  | 
      myP = (1 + myP) % myNum; | 
| 817 | 
  | 
  | 
      if (myP != myMasterId) | 
| 818 | 
  | 
  | 
        MPI_Iprobe(myP, NEED_RANGE, MPI_COMM_WORLD, &test, &status); | 
| 819 | 
  | 
  | 
    } | 
| 820 | 
  | 
  | 
 | 
| 821 | 
  | 
  | 
    MPI_Recv(tmp, 0, MPI_CHAR, myP, NEED_RANGE, MPI_COMM_WORLD, &status); | 
| 822 | 
  | 
  | 
    myRecv--; | 
| 823 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 824 | 
  | 
  | 
    report << allnodes << plain << "Recieve from " << myP << "." << endr; | 
| 825 | 
  | 
  | 
    report << allnodes << plain << "Recv " << myRecv << "." << endr; | 
| 826 | 
  | 
  | 
#endif | 
| 827 | 
  | 
  | 
 | 
| 828 | 
  | 
  | 
    // We skip to send a further range if the node got the last | 
| 829 | 
  | 
  | 
    // range since the that node will terminate without | 
| 830 | 
  | 
  | 
    // receiving any messages. | 
| 831 | 
  | 
  | 
    if (myDone[myP] == 0) { | 
| 832 | 
  | 
  | 
      MPI_Bsend(&(myBlockList[myI]), 2, MPI_INT, myP, SEND_RANGE, | 
| 833 | 
  | 
  | 
                MPI_COMM_WORLD); | 
| 834 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 835 | 
  | 
  | 
      report << allnodes << plain << "Block[" << myI << "] = [" << | 
| 836 | 
  | 
  | 
      myBlockList[myI] << "," << | 
| 837 | 
  | 
  | 
      myBlockList[myI + 1] << "] to " << myP << "." << endr; | 
| 838 | 
  | 
  | 
#endif | 
| 839 | 
  | 
  | 
      myDone[myP] = (myBlockList[myI + 2] < 0 ? 1 : 0); | 
| 840 | 
  | 
  | 
      if (myDone[myP] == 0) | 
| 841 | 
  | 
  | 
        myRecv++; | 
| 842 | 
  | 
  | 
      myI++; | 
| 843 | 
  | 
  | 
    } | 
| 844 | 
  | 
  | 
  } | 
| 845 | 
  | 
  | 
#endif | 
| 846 | 
  | 
  | 
 | 
| 847 | 
  | 
  | 
} | 
| 848 | 
  | 
  | 
 | 
| 849 | 
  | 
  | 
bool Parallel::next() { | 
| 850 | 
  | 
  | 
  bool doNext = true; | 
| 851 | 
  | 
  | 
   | 
| 852 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 853 | 
  | 
  | 
  int oldNext = myNext; | 
| 854 | 
  | 
  | 
#endif | 
| 855 | 
  | 
  | 
  switch (myWorkState) { | 
| 856 | 
  | 
  | 
#ifdef IS_MPI | 
| 857 | 
  | 
  | 
    case MASTER: | 
| 858 | 
  | 
  | 
       | 
| 859 | 
  | 
  | 
    do | 
| 860 | 
  | 
  | 
    { | 
| 861 | 
  | 
  | 
       | 
| 862 | 
  | 
  | 
    } while (condition);Next = false; | 
| 863 | 
  | 
  | 
    do { | 
| 864 | 
  | 
  | 
      for (int k = 0; k < myNum; ++k) { | 
| 865 | 
  | 
  | 
        myP = (1 + myP) % myNum; | 
| 866 | 
  | 
  | 
        if (myP == myMasterId) | 
| 867 | 
  | 
  | 
          continue; | 
| 868 | 
  | 
  | 
         | 
| 869 | 
  | 
  | 
        int test = 0; | 
| 870 | 
  | 
  | 
        MPI_Status status; | 
| 871 | 
  | 
  | 
        char tmp[1]; | 
| 872 | 
  | 
  | 
         | 
| 873 | 
  | 
  | 
        MPI_Iprobe(myP, NEED_RANGE, MPI_COMM_WORLD, &test, &status); | 
| 874 | 
  | 
  | 
         | 
| 875 | 
  | 
  | 
        if (test) { | 
| 876 | 
  | 
  | 
          MPI_Recv(tmp, 0, MPI_CHAR, myP, NEED_RANGE, MPI_COMM_WORLD, &status); | 
| 877 | 
  | 
  | 
          myRecv--; | 
| 878 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 879 | 
  | 
  | 
          report << allnodes << plain << "Recieve from " << myP << "." << | 
| 880 | 
  | 
  | 
              endr; | 
| 881 | 
  | 
  | 
#endif | 
| 882 | 
  | 
  | 
           | 
| 883 | 
  | 
  | 
          // We skip to send a further range if the node got the last | 
| 884 | 
  | 
  | 
          // range since the that node will terminate without | 
| 885 | 
  | 
  | 
          // receiving any messages. | 
| 886 | 
  | 
  | 
          if (myDone[myP] == 0) { | 
| 887 | 
  | 
  | 
            MPI_Bsend(&(myBlockList[myI]), 2, MPI_INT, myP, SEND_RANGE, | 
| 888 | 
  | 
  | 
                      MPI_COMM_WORLD); | 
| 889 | 
  | 
  | 
#if defined (DEBUG_PARALLEL) | 
| 890 | 
  | 
  | 
            report << allnodes << plain << "Block[" << myI << "] = [" << | 
| 891 | 
  | 
  | 
                myBlockList[myI] << "," << | 
| 892 | 
  | 
  | 
                myBlockList[myI + 1] << "] to " << myP << "." << endr; | 
| 893 | 
  | 
  | 
#endif | 
| 894 | 
  | 
  | 
            myDone[myP] = (myBlockList[myI + 2] < 0 ? 1 : 0); | 
| 895 | 
  | 
  | 
            if (myDone[myP] == 0) | 
| 896 | 
  | 
  | 
              myRecv++; | 
| 897 | 
  | 
  | 
            myI++; | 
| 898 | 
  | 
  | 
          } | 
| 899 | 
  | 
  | 
        } | 
| 900 | 
  | 
  | 
      } | 
| 901 | 
  | 
  | 
    } while (myBlockList[myI + 2] < 0 && myBlockList[myI + 0] >= 0 && | 
| 902 | 
  | 
  | 
             myBlockList[myI + 1] >= 0); | 
| 903 | 
  | 
  | 
 | 
| 904 | 
  | 
  | 
    if (myNextRange[1] == myNext) { | 
| 905 | 
  | 
  | 
      myNextRange[0] = myBlockList[myI]; | 
| 906 | 
  | 
  | 
      myNextRange[1] = myBlockList[myI + 1]; | 
| 907 | 
  | 
  | 
      myDone[myId] = (myBlockList[myI + 2] < 0 ? 1 : 0); | 
| 908 | 
  | 
  | 
#if defined (DEBUG_PARALLEL) | 
| 909 | 
  | 
  | 
      report << allnodes << plain << "Block[" << myI << "] = [" << | 
| 910 | 
  | 
  | 
      myBlockList[myI] << "," << | 
| 911 | 
  | 
  | 
      myBlockList[myI + 1] << "] to " << getId() << "." << endr; | 
| 912 | 
  | 
  | 
#endif | 
| 913 | 
  | 
  | 
      myI++; | 
| 914 | 
  | 
  | 
    } | 
| 915 | 
  | 
  | 
    doNext = !(myNext < myNextRange[0] || myNextRange[1] < 0); | 
| 916 | 
  | 
  | 
    myNext++; | 
| 917 | 
  | 
  | 
 | 
| 918 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 919 | 
  | 
  | 
    report << allnodes << plain << "Recv " << myRecv << "." << endr; | 
| 920 | 
  | 
  | 
#endif | 
| 921 | 
  | 
  | 
    if (myDone[myId] != 0) | 
| 922 | 
  | 
  | 
      nextMaster(); | 
| 923 | 
  | 
  | 
 | 
| 924 | 
  | 
  | 
    break; | 
| 925 | 
  | 
  | 
 | 
| 926 | 
  | 
  | 
  case Parallel::SLAVE: | 
| 927 | 
  | 
  | 
    if (myNextRange[1] == myNext) { | 
| 928 | 
  | 
  | 
      // We need a new range from the master | 
| 929 | 
  | 
  | 
      char tmp[1]; | 
| 930 | 
  | 
  | 
      MPI_Status status; | 
| 931 | 
  | 
  | 
      TimerStatistic::timer[TimerStatistic::IDLE].start(); | 
| 932 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 933 | 
  | 
  | 
      report << allnodes << plain << "Recv new block " << getId() << "." << | 
| 934 | 
  | 
  | 
      endr; | 
| 935 | 
  | 
  | 
#endif | 
| 936 | 
  | 
  | 
      MPI_Recv(myNextRange, 2, MPI_INT, myMasterId, SEND_RANGE, | 
| 937 | 
  | 
  | 
               MPI_COMM_WORLD, | 
| 938 | 
  | 
  | 
               &status); | 
| 939 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 940 | 
  | 
  | 
      report << allnodes << plain << "Recv new block " << getId() << " [" << | 
| 941 | 
  | 
  | 
      myNextRange[0] << "," << myNextRange[1] << "]." << endr; | 
| 942 | 
  | 
  | 
#endif | 
| 943 | 
  | 
  | 
      // Asking for the next range such we have it when we need it next time. | 
| 944 | 
  | 
  | 
      TimerStatistic::timer[TimerStatistic::IDLE].stop(); | 
| 945 | 
  | 
  | 
      if (myNextRange[1] >= 0) { | 
| 946 | 
  | 
  | 
        MPI_Bsend(tmp, 0, MPI_CHAR, myMasterId, NEED_RANGE, MPI_COMM_WORLD); | 
| 947 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 948 | 
  | 
  | 
        report << allnodes << plain << "Ask new block " << getId() << "." << | 
| 949 | 
  | 
  | 
        endr; | 
| 950 | 
  | 
  | 
#endif | 
| 951 | 
  | 
  | 
      } | 
| 952 | 
  | 
  | 
      if (myNextRange[1] < 0) | 
| 953 | 
  | 
  | 
        myWorkState = DONE; | 
| 954 | 
  | 
  | 
    } | 
| 955 | 
  | 
  | 
 | 
| 956 | 
  | 
  | 
    doNext = !(myNext < myNextRange[0] || myNextRange[1] < 0); | 
| 957 | 
  | 
  | 
    myNext++; | 
| 958 | 
  | 
  | 
 | 
| 959 | 
  | 
  | 
    break; | 
| 960 | 
  | 
  | 
#endif | 
| 961 | 
  | 
  | 
 | 
| 962 | 
  | 
  | 
  case DONE: | 
| 963 | 
  | 
  | 
    doNext = false; | 
| 964 | 
  | 
  | 
    break; | 
| 965 | 
  | 
  | 
 | 
| 966 | 
  | 
  | 
  case STATIC: | 
| 967 | 
  | 
  | 
    doNext = (myNext == myId); | 
| 968 | 
  | 
  | 
    myNext = (myNext + 1) % myNum; | 
| 969 | 
  | 
  | 
    break; | 
| 970 | 
  | 
  | 
 | 
| 971 | 
  | 
  | 
  case SEQUENTIAL: | 
| 972 | 
  | 
  | 
  default: | 
| 973 | 
  | 
  | 
    break; | 
| 974 | 
  | 
  | 
  } | 
| 975 | 
  | 
  | 
 | 
| 976 | 
  | 
  | 
#ifdef DEBUG_PARALLEL | 
| 977 | 
  | 
  | 
  report << allnodes << plain << "Next (" << getId() << ") : " | 
| 978 | 
  | 
  | 
         << (bool)doNext << ", " << myWorkState << ", " << oldNext << ", [" | 
| 979 | 
  | 
  | 
         << myNextRange[0] << "," << myNextRange[1] << "]" << endr; | 
| 980 | 
  | 
  | 
#endif | 
| 981 | 
  | 
  | 
  return doNext; | 
| 982 | 
  | 
  | 
} | 
| 983 | 
  | 
  | 
 | 
| 984 | 
  | 
  | 
void Parallel::isolateNode() { | 
| 985 | 
  | 
  | 
  if (myIsolated == false) { | 
| 986 | 
  | 
  | 
    myOldId = myId; | 
| 987 | 
  | 
  | 
    myOldNum = myNum; | 
| 988 | 
  | 
  | 
    myOldMode = myMode; | 
| 989 | 
  | 
  | 
    myIsolated = true; | 
| 990 | 
  | 
  | 
 | 
| 991 | 
  | 
  | 
    myId = 0; | 
| 992 | 
  | 
  | 
    myNum = 1; | 
| 993 | 
  | 
  | 
    myIsParallel = false; | 
| 994 | 
  | 
  | 
    myIAmMaster = (myId == myMasterId); | 
| 995 | 
  | 
  | 
  } | 
| 996 | 
  | 
  | 
} | 
| 997 | 
  | 
  | 
 | 
| 998 | 
  | 
  | 
void Parallel::integrateNode() { | 
| 999 | 
  | 
  | 
  if (myIsolated == true) { | 
| 1000 | 
  | 
  | 
    myId = myOldId; | 
| 1001 | 
  | 
  | 
    myNum = myOldNum; | 
| 1002 | 
  | 
  | 
    myIsolated = false; | 
| 1003 | 
  | 
  | 
 | 
| 1004 | 
  | 
  | 
    myIsParallel = (myNum > 1); | 
| 1005 | 
  | 
  | 
    myIAmMaster = (myId == myMasterId); | 
| 1006 | 
  | 
  | 
  } | 
| 1007 | 
  | 
  | 
} | 
| 1008 | 
  | 
  | 
 | 
| 1009 | 
  | 
  | 
 | 
| 1010 | 
  | 
  | 
 | 
| 1011 | 
  | 
  | 
 | 
| 1012 | 
  | 
  | 
 | 
| 1013 | 
  | 
  | 
 | 
| 1014 | 
  | 
  | 
 | 
| 1015 | 
  | 
  | 
 | 
| 1016 | 
  | 
  | 
 | 
| 1017 | 
  | 
  | 
 |