| 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. Acknowledgement of the program authors must be made in any | 
| 10 | *    publication of scientific results based in part on use of the | 
| 11 | *    program.  An acceptable form of acknowledgement is citation of | 
| 12 | *    the article in which the program was described (Matthew | 
| 13 | *    A. Meineke, Charles F. Vardeman II, Teng Lin, Christopher | 
| 14 | *    J. Fennell and J. Daniel Gezelter, "OOPSE: An Object-Oriented | 
| 15 | *    Parallel Simulation Engine for Molecular Dynamics," | 
| 16 | *    J. Comput. Chem. 26, pp. 252-271 (2005)) | 
| 17 | * | 
| 18 | * 2. Redistributions of source code must retain the above copyright | 
| 19 | *    notice, this list of conditions and the following disclaimer. | 
| 20 | * | 
| 21 | * 3. Redistributions in binary form must reproduce the above copyright | 
| 22 | *    notice, this list of conditions and the following disclaimer in the | 
| 23 | *    documentation and/or other materials provided with the | 
| 24 | *    distribution. | 
| 25 | * | 
| 26 | * This software is provided "AS IS," without a warranty of any | 
| 27 | * kind. All express or implied conditions, representations and | 
| 28 | * warranties, including any implied warranty of merchantability, | 
| 29 | * fitness for a particular purpose or non-infringement, are hereby | 
| 30 | * excluded.  The University of Notre Dame and its licensors shall not | 
| 31 | * be liable for any damages suffered by licensee as a result of | 
| 32 | * using, modifying or distributing the software or its | 
| 33 | * derivatives. In no event will the University of Notre Dame or its | 
| 34 | * licensors be liable for any lost revenue, profit or data, or for | 
| 35 | * direct, indirect, special, consequential, incidental or punitive | 
| 36 | * damages, however caused and regardless of the theory of liability, | 
| 37 | * arising out of the use of or inability to use software, even if the | 
| 38 | * University of Notre Dame has been advised of the possibility of | 
| 39 | * such damages. | 
| 40 | */ | 
| 41 |  | 
| 42 | #include "io/DumpWriter.hpp" | 
| 43 | #include "primitives/Molecule.hpp" | 
| 44 | #include "utils/simError.h" | 
| 45 | #include "io/basic_teebuf.hpp" | 
| 46 | #include "io/gzstream.hpp" | 
| 47 | #include "io/Globals.hpp" | 
| 48 |  | 
| 49 | #ifdef IS_MPI | 
| 50 | #include <mpi.h> | 
| 51 | #endif //is_mpi | 
| 52 |  | 
| 53 | namespace oopse { | 
| 54 |  | 
| 55 | DumpWriter::DumpWriter(SimInfo* info) | 
| 56 | : info_(info), filename_(info->getDumpFileName()), eorFilename_(info->getFinalConfigFileName()){ | 
| 57 |  | 
| 58 | Globals* simParams = info->getSimParams(); | 
| 59 | needCompression_ = simParams->getCompressDumpFile(); | 
| 60 | needForceVector_ = simParams->getOutputForceVector(); | 
| 61 |  | 
| 62 | #ifdef HAVE_LIBZ | 
| 63 | if (needCompression_) { | 
| 64 | filename_ += ".gz"; | 
| 65 | eorFilename_ += ".gz"; | 
| 66 | } | 
| 67 | #endif | 
| 68 |  | 
| 69 | #ifdef IS_MPI | 
| 70 |  | 
| 71 | if (worldRank == 0) { | 
| 72 | #endif // is_mpi | 
| 73 |  | 
| 74 |  | 
| 75 | dumpFile_ = createOStream(filename_); | 
| 76 |  | 
| 77 | if (!dumpFile_) { | 
| 78 | sprintf(painCave.errMsg, "Could not open \"%s\" for dump output.\n", | 
| 79 | filename_.c_str()); | 
| 80 | painCave.isFatal = 1; | 
| 81 | simError(); | 
| 82 | } | 
| 83 |  | 
| 84 | #ifdef IS_MPI | 
| 85 |  | 
| 86 | } | 
| 87 |  | 
| 88 | sprintf(checkPointMsg, "Sucessfully opened output file for dumping.\n"); | 
| 89 | MPIcheckPoint(); | 
| 90 |  | 
| 91 | #endif // is_mpi | 
| 92 |  | 
| 93 | } | 
| 94 |  | 
| 95 |  | 
| 96 | DumpWriter::DumpWriter(SimInfo* info, const std::string& filename) | 
| 97 | : info_(info), filename_(filename){ | 
| 98 |  | 
| 99 | Globals* simParams = info->getSimParams(); | 
| 100 | eorFilename_ = filename_.substr(0, filename_.rfind(".")) + ".eor"; | 
| 101 |  | 
| 102 | needCompression_ = simParams->getCompressDumpFile(); | 
| 103 | needForceVector_ = simParams->getOutputForceVector(); | 
| 104 |  | 
| 105 | #ifdef HAVE_LIBZ | 
| 106 | if (needCompression_) { | 
| 107 | filename_ += ".gz"; | 
| 108 | eorFilename_ += ".gz"; | 
| 109 | } | 
| 110 | #endif | 
| 111 |  | 
| 112 | #ifdef IS_MPI | 
| 113 |  | 
| 114 | if (worldRank == 0) { | 
| 115 | #endif // is_mpi | 
| 116 |  | 
| 117 |  | 
| 118 | dumpFile_ = createOStream(filename_); | 
| 119 |  | 
| 120 | if (!dumpFile_) { | 
| 121 | sprintf(painCave.errMsg, "Could not open \"%s\" for dump output.\n", | 
| 122 | filename_.c_str()); | 
| 123 | painCave.isFatal = 1; | 
| 124 | simError(); | 
| 125 | } | 
| 126 |  | 
| 127 | #ifdef IS_MPI | 
| 128 |  | 
| 129 | } | 
| 130 |  | 
| 131 | sprintf(checkPointMsg, "Sucessfully opened output file for dumping.\n"); | 
| 132 | MPIcheckPoint(); | 
| 133 |  | 
| 134 | #endif // is_mpi | 
| 135 |  | 
| 136 | } | 
| 137 |  | 
| 138 | DumpWriter::~DumpWriter() { | 
| 139 |  | 
| 140 | #ifdef IS_MPI | 
| 141 |  | 
| 142 | if (worldRank == 0) { | 
| 143 | #endif // is_mpi | 
| 144 |  | 
| 145 | delete dumpFile_; | 
| 146 |  | 
| 147 | #ifdef IS_MPI | 
| 148 |  | 
| 149 | } | 
| 150 |  | 
| 151 | #endif // is_mpi | 
| 152 |  | 
| 153 | } | 
| 154 |  | 
| 155 | void DumpWriter::writeCommentLine(std::ostream& os, Snapshot* s) { | 
| 156 |  | 
| 157 | double currentTime; | 
| 158 | Mat3x3d hmat; | 
| 159 | double chi; | 
| 160 | double integralOfChiDt; | 
| 161 | Mat3x3d eta; | 
| 162 |  | 
| 163 | currentTime = s->getTime(); | 
| 164 | hmat = s->getHmat(); | 
| 165 | chi = s->getChi(); | 
| 166 | integralOfChiDt = s->getIntegralOfChiDt(); | 
| 167 | eta = s->getEta(); | 
| 168 |  | 
| 169 | os << currentTime << ";\t" | 
| 170 | << hmat(0, 0) << "\t" << hmat(1, 0) << "\t" << hmat(2, 0) << ";\t" | 
| 171 | << hmat(0, 1) << "\t" << hmat(1, 1) << "\t" << hmat(2, 1) << ";\t" | 
| 172 | << hmat(0, 2) << "\t" << hmat(1, 2) << "\t" << hmat(2, 2) << ";\t"; | 
| 173 |  | 
| 174 | //write out additional parameters, such as chi and eta | 
| 175 |  | 
| 176 | os << chi << "\t" << integralOfChiDt << "\t;"; | 
| 177 |  | 
| 178 | os << eta(0, 0) << "\t" << eta(1, 0) << "\t" << eta(2, 0) << ";\t" | 
| 179 | << eta(0, 1) << "\t" << eta(1, 1) << "\t" << eta(2, 1) << ";\t" | 
| 180 | << eta(0, 2) << "\t" << eta(1, 2) << "\t" << eta(2, 2) << ";"; | 
| 181 |  | 
| 182 | os << "\n"; | 
| 183 | } | 
| 184 |  | 
| 185 | void DumpWriter::writeFrame(std::ostream& os) { | 
| 186 | const int BUFFERSIZE = 2000; | 
| 187 | const int MINIBUFFERSIZE = 100; | 
| 188 |  | 
| 189 | char tempBuffer[BUFFERSIZE]; | 
| 190 | char writeLine[BUFFERSIZE]; | 
| 191 |  | 
| 192 | Quat4d q; | 
| 193 | Vector3d ji; | 
| 194 | Vector3d pos; | 
| 195 | Vector3d vel; | 
| 196 | Vector3d frc; | 
| 197 | Vector3d trq; | 
| 198 |  | 
| 199 | Molecule* mol; | 
| 200 | StuntDouble* integrableObject; | 
| 201 | SimInfo::MoleculeIterator mi; | 
| 202 | Molecule::IntegrableObjectIterator ii; | 
| 203 |  | 
| 204 | int nTotObjects; | 
| 205 | nTotObjects = info_->getNGlobalIntegrableObjects(); | 
| 206 |  | 
| 207 | #ifndef IS_MPI | 
| 208 |  | 
| 209 |  | 
| 210 | os << nTotObjects << "\n"; | 
| 211 |  | 
| 212 | writeCommentLine(os, info_->getSnapshotManager()->getCurrentSnapshot()); | 
| 213 |  | 
| 214 | for (mol = info_->beginMolecule(mi); mol != NULL; mol = info_->nextMolecule(mi)) { | 
| 215 |  | 
| 216 | for (integrableObject = mol->beginIntegrableObject(ii); integrableObject != NULL; | 
| 217 | integrableObject = mol->nextIntegrableObject(ii)) { | 
| 218 |  | 
| 219 |  | 
| 220 | pos = integrableObject->getPos(); | 
| 221 | vel = integrableObject->getVel(); | 
| 222 |  | 
| 223 | sprintf(tempBuffer, "%s\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t", | 
| 224 | integrableObject->getType().c_str(), | 
| 225 | pos[0], pos[1], pos[2], | 
| 226 | vel[0], vel[1], vel[2]); | 
| 227 |  | 
| 228 | strcpy(writeLine, tempBuffer); | 
| 229 |  | 
| 230 | if (integrableObject->isDirectional()) { | 
| 231 | q = integrableObject->getQ(); | 
| 232 | ji = integrableObject->getJ(); | 
| 233 |  | 
| 234 | sprintf(tempBuffer, "%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf", | 
| 235 | q[0], q[1], q[2], q[3], | 
| 236 | ji[0], ji[1], ji[2]); | 
| 237 | strcat(writeLine, tempBuffer); | 
| 238 | } else { | 
| 239 | strcat(writeLine, "0.0\t0.0\t0.0\t0.0\t0.0\t0.0\t0.0"); | 
| 240 | } | 
| 241 |  | 
| 242 | if (needForceVector_) { | 
| 243 | frc = integrableObject->getFrc(); | 
| 244 | trq = integrableObject->getTrq(); | 
| 245 |  | 
| 246 | sprintf(tempBuffer, "\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf", | 
| 247 | frc[0], frc[1], frc[2], | 
| 248 | trq[0], trq[1], trq[2]); | 
| 249 | strcat(writeLine, tempBuffer); | 
| 250 | } | 
| 251 |  | 
| 252 | strcat(writeLine, "\n"); | 
| 253 | os << writeLine; | 
| 254 |  | 
| 255 | } | 
| 256 | } | 
| 257 |  | 
| 258 | os.flush(); | 
| 259 | #else // is_mpi | 
| 260 | /********************************************************************* | 
| 261 | * Documentation?  You want DOCUMENTATION? | 
| 262 | * | 
| 263 | * Why all the potatoes below? | 
| 264 | * | 
| 265 | * To make a long story short, the original version of DumpWriter | 
| 266 | * worked in the most inefficient way possible.  Node 0 would | 
| 267 | * poke each of the node for an individual atom's formatted data | 
| 268 | * as node 0 worked its way down the global index. This was particularly | 
| 269 | * inefficient since the method blocked all processors at every atom | 
| 270 | * (and did it twice!). | 
| 271 | * | 
| 272 | * An intermediate version of DumpWriter could be described from Node | 
| 273 | * zero's perspective as follows: | 
| 274 | * | 
| 275 | *  1) Have 100 of your friends stand in a circle. | 
| 276 | *  2) When you say go, have all of them start tossing potatoes at | 
| 277 | *     you (one at a time). | 
| 278 | *  3) Catch the potatoes. | 
| 279 | * | 
| 280 | * It was an improvement, but MPI has buffers and caches that could | 
| 281 | * best be described in this analogy as "potato nets", so there's no | 
| 282 | * need to block the processors atom-by-atom. | 
| 283 | * | 
| 284 | * This new and improved DumpWriter works in an even more efficient | 
| 285 | * way: | 
| 286 | * | 
| 287 | *  1) Have 100 of your friend stand in a circle. | 
| 288 | *  2) When you say go, have them start tossing 5-pound bags of | 
| 289 | *     potatoes at you. | 
| 290 | *  3) Once you've caught a friend's bag of potatoes, | 
| 291 | *     toss them a spud to let them know they can toss another bag. | 
| 292 | * | 
| 293 | * How's THAT for documentation? | 
| 294 | * | 
| 295 | *********************************************************************/ | 
| 296 | const int masterNode = 0; | 
| 297 |  | 
| 298 | int * potatoes; | 
| 299 | int myPotato; | 
| 300 | int nProc; | 
| 301 | int which_node; | 
| 302 | double atomData[19]; | 
| 303 | int isDirectional; | 
| 304 | char MPIatomTypeString[MINIBUFFERSIZE]; | 
| 305 | int msgLen; // the length of message actually recieved at master nodes | 
| 306 | int haveError; | 
| 307 | MPI_Status istatus; | 
| 308 | int nCurObj; | 
| 309 |  | 
| 310 | // code to find maximum tag value | 
| 311 | int * tagub; | 
| 312 | int flag; | 
| 313 | int MAXTAG; | 
| 314 | MPI_Attr_get(MPI_COMM_WORLD, MPI_TAG_UB, &tagub, &flag); | 
| 315 |  | 
| 316 | if (flag) { | 
| 317 | MAXTAG = *tagub; | 
| 318 | } else { | 
| 319 | MAXTAG = 32767; | 
| 320 | } | 
| 321 |  | 
| 322 | if (worldRank == masterNode) { //master node (node 0) is responsible for writing the dump file | 
| 323 |  | 
| 324 | // Node 0 needs a list of the magic potatoes for each processor; | 
| 325 |  | 
| 326 | MPI_Comm_size(MPI_COMM_WORLD, &nProc); | 
| 327 | potatoes = new int[nProc]; | 
| 328 |  | 
| 329 | //write out the comment lines | 
| 330 | for(int i = 0; i < nProc; i++) { | 
| 331 | potatoes[i] = 0; | 
| 332 | } | 
| 333 |  | 
| 334 |  | 
| 335 | os << nTotObjects << "\n"; | 
| 336 | writeCommentLine(os, info_->getSnapshotManager()->getCurrentSnapshot()); | 
| 337 |  | 
| 338 | for(int i = 0; i < info_->getNGlobalMolecules(); i++) { | 
| 339 |  | 
| 340 | // Get the Node number which has this atom; | 
| 341 |  | 
| 342 | which_node = info_->getMolToProc(i); | 
| 343 |  | 
| 344 | if (which_node != masterNode) { //current molecule is in slave node | 
| 345 | if (potatoes[which_node] + 1 >= MAXTAG) { | 
| 346 | // The potato was going to exceed the maximum value, | 
| 347 | // so wrap this processor potato back to 0: | 
| 348 |  | 
| 349 | potatoes[which_node] = 0; | 
| 350 | MPI_Send(&potatoes[which_node], 1, MPI_INT, which_node, 0, | 
| 351 | MPI_COMM_WORLD); | 
| 352 | } | 
| 353 |  | 
| 354 | myPotato = potatoes[which_node]; | 
| 355 |  | 
| 356 | //recieve the number of integrableObject in current molecule | 
| 357 | MPI_Recv(&nCurObj, 1, MPI_INT, which_node, myPotato, | 
| 358 | MPI_COMM_WORLD, &istatus); | 
| 359 | myPotato++; | 
| 360 |  | 
| 361 | for(int l = 0; l < nCurObj; l++) { | 
| 362 | if (potatoes[which_node] + 2 >= MAXTAG) { | 
| 363 | // The potato was going to exceed the maximum value, | 
| 364 | // so wrap this processor potato back to 0: | 
| 365 |  | 
| 366 | potatoes[which_node] = 0; | 
| 367 | MPI_Send(&potatoes[which_node], 1, MPI_INT, which_node, | 
| 368 | 0, MPI_COMM_WORLD); | 
| 369 | } | 
| 370 |  | 
| 371 | MPI_Recv(MPIatomTypeString, MINIBUFFERSIZE, MPI_CHAR, | 
| 372 | which_node, myPotato, MPI_COMM_WORLD, | 
| 373 | &istatus); | 
| 374 |  | 
| 375 | myPotato++; | 
| 376 |  | 
| 377 | MPI_Recv(atomData, 19, MPI_DOUBLE, which_node, myPotato, | 
| 378 | MPI_COMM_WORLD, &istatus); | 
| 379 | myPotato++; | 
| 380 |  | 
| 381 | MPI_Get_count(&istatus, MPI_DOUBLE, &msgLen); | 
| 382 |  | 
| 383 | if (msgLen == 13 || msgLen == 19) | 
| 384 | isDirectional = 1; | 
| 385 | else | 
| 386 | isDirectional = 0; | 
| 387 |  | 
| 388 | // If we've survived to here, format the line: | 
| 389 |  | 
| 390 | if (!isDirectional) { | 
| 391 | sprintf(writeLine, "%s\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t", | 
| 392 | MPIatomTypeString, atomData[0], | 
| 393 | atomData[1], atomData[2], | 
| 394 | atomData[3], atomData[4], | 
| 395 | atomData[5]); | 
| 396 |  | 
| 397 | strcat(writeLine, | 
| 398 | "0.0\t0.0\t0.0\t0.0\t0.0\t0.0\t0.0"); | 
| 399 | } else { | 
| 400 | sprintf(writeLine, | 
| 401 | "%s\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf", | 
| 402 | MPIatomTypeString, | 
| 403 | atomData[0], | 
| 404 | atomData[1], | 
| 405 | atomData[2], | 
| 406 | atomData[3], | 
| 407 | atomData[4], | 
| 408 | atomData[5], | 
| 409 | atomData[6], | 
| 410 | atomData[7], | 
| 411 | atomData[8], | 
| 412 | atomData[9], | 
| 413 | atomData[10], | 
| 414 | atomData[11], | 
| 415 | atomData[12]); | 
| 416 | } | 
| 417 |  | 
| 418 | if (needForceVector_) { | 
| 419 | if (!isDirectional) { | 
| 420 | sprintf(writeLine, "\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf", | 
| 421 | atomData[6], | 
| 422 | atomData[7], | 
| 423 | atomData[8], | 
| 424 | atomData[9], | 
| 425 | atomData[10], | 
| 426 | atomData[11]); | 
| 427 | } else { | 
| 428 | sprintf(writeLine, "\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf", | 
| 429 | atomData[13], | 
| 430 | atomData[14], | 
| 431 | atomData[15], | 
| 432 | atomData[16], | 
| 433 | atomData[17], | 
| 434 | atomData[18]); | 
| 435 | } | 
| 436 | } | 
| 437 |  | 
| 438 | sprintf(writeLine, "\n"); | 
| 439 | os << writeLine; | 
| 440 |  | 
| 441 | } // end for(int l =0) | 
| 442 |  | 
| 443 | potatoes[which_node] = myPotato; | 
| 444 | } else { //master node has current molecule | 
| 445 |  | 
| 446 | mol = info_->getMoleculeByGlobalIndex(i); | 
| 447 |  | 
| 448 | if (mol == NULL) { | 
| 449 | sprintf(painCave.errMsg, "Molecule not found on node %d!", worldRank); | 
| 450 | painCave.isFatal = 1; | 
| 451 | simError(); | 
| 452 | } | 
| 453 |  | 
| 454 | for (integrableObject = mol->beginIntegrableObject(ii); integrableObject != NULL; | 
| 455 | integrableObject = mol->nextIntegrableObject(ii)) { | 
| 456 |  | 
| 457 | pos = integrableObject->getPos(); | 
| 458 | vel = integrableObject->getVel(); | 
| 459 |  | 
| 460 | atomData[0] = pos[0]; | 
| 461 | atomData[1] = pos[1]; | 
| 462 | atomData[2] = pos[2]; | 
| 463 |  | 
| 464 | atomData[3] = vel[0]; | 
| 465 | atomData[4] = vel[1]; | 
| 466 | atomData[5] = vel[2]; | 
| 467 |  | 
| 468 | isDirectional = 0; | 
| 469 |  | 
| 470 | if (integrableObject->isDirectional()) { | 
| 471 | isDirectional = 1; | 
| 472 |  | 
| 473 | q = integrableObject->getQ(); | 
| 474 | ji = integrableObject->getJ(); | 
| 475 |  | 
| 476 | for(int j = 0; j < 6; j++) { | 
| 477 | atomData[j] = atomData[j]; | 
| 478 | } | 
| 479 |  | 
| 480 | atomData[6] = q[0]; | 
| 481 | atomData[7] = q[1]; | 
| 482 | atomData[8] = q[2]; | 
| 483 | atomData[9] = q[3]; | 
| 484 |  | 
| 485 | atomData[10] = ji[0]; | 
| 486 | atomData[11] = ji[1]; | 
| 487 | atomData[12] = ji[2]; | 
| 488 | } | 
| 489 |  | 
| 490 | if (needForceVector_) { | 
| 491 | frc = integrableObject->getFrc(); | 
| 492 | trq = integrableObject->getTrq(); | 
| 493 |  | 
| 494 | if (!isDirectional) { | 
| 495 | atomData[6] = frc[0]; | 
| 496 | atomData[7] = frc[1]; | 
| 497 | atomData[8] = frc[2]; | 
| 498 | atomData[9] = trq[0]; | 
| 499 | atomData[10] = trq[1]; | 
| 500 | atomData[11] = trq[2]; | 
| 501 | } else { | 
| 502 | atomData[13] = frc[0]; | 
| 503 | atomData[14] = frc[1]; | 
| 504 | atomData[15] = frc[2]; | 
| 505 | atomData[16] = trq[0]; | 
| 506 | atomData[17] = trq[1]; | 
| 507 | atomData[18] = trq[2]; | 
| 508 | } | 
| 509 | } | 
| 510 |  | 
| 511 | // If we've survived to here, format the line: | 
| 512 |  | 
| 513 | if (!isDirectional) { | 
| 514 | sprintf(writeLine, "%s\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t", | 
| 515 | integrableObject->getType().c_str(), atomData[0], | 
| 516 | atomData[1], atomData[2], | 
| 517 | atomData[3], atomData[4], | 
| 518 | atomData[5]); | 
| 519 |  | 
| 520 | strcat(writeLine, | 
| 521 | "0.0\t0.0\t0.0\t0.0\t0.0\t0.0\t0.0"); | 
| 522 | } else { | 
| 523 | sprintf(writeLine, | 
| 524 | "%s\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf", | 
| 525 | integrableObject->getType().c_str(), | 
| 526 | atomData[0], | 
| 527 | atomData[1], | 
| 528 | atomData[2], | 
| 529 | atomData[3], | 
| 530 | atomData[4], | 
| 531 | atomData[5], | 
| 532 | atomData[6], | 
| 533 | atomData[7], | 
| 534 | atomData[8], | 
| 535 | atomData[9], | 
| 536 | atomData[10], | 
| 537 | atomData[11], | 
| 538 | atomData[12]); | 
| 539 | } | 
| 540 |  | 
| 541 | if (needForceVector_) { | 
| 542 | if (!isDirectional) { | 
| 543 | sprintf(writeLine, "\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf", | 
| 544 | atomData[6], | 
| 545 | atomData[7], | 
| 546 | atomData[8], | 
| 547 | atomData[9], | 
| 548 | atomData[10], | 
| 549 | atomData[11]); | 
| 550 | } else { | 
| 551 | sprintf(writeLine, "\t%lf\t%lf\t%lf\t%lf\t%lf\t%lf", | 
| 552 | atomData[13], | 
| 553 | atomData[14], | 
| 554 | atomData[15], | 
| 555 | atomData[16], | 
| 556 | atomData[17], | 
| 557 | atomData[18]); | 
| 558 | } | 
| 559 | } | 
| 560 |  | 
| 561 | sprintf(writeLine, "\n"); | 
| 562 | os << writeLine; | 
| 563 |  | 
| 564 | } //end for(iter = integrableObject.begin()) | 
| 565 | } | 
| 566 | } //end for(i = 0; i < mpiSim->getNmol()) | 
| 567 |  | 
| 568 | os.flush(); | 
| 569 |  | 
| 570 | sprintf(checkPointMsg, "Sucessfully took a dump.\n"); | 
| 571 | MPIcheckPoint(); | 
| 572 |  | 
| 573 | delete [] potatoes; | 
| 574 | } else { | 
| 575 |  | 
| 576 | // worldRank != 0, so I'm a remote node. | 
| 577 |  | 
| 578 | // Set my magic potato to 0: | 
| 579 |  | 
| 580 | myPotato = 0; | 
| 581 |  | 
| 582 | for(int i = 0; i < info_->getNGlobalMolecules(); i++) { | 
| 583 |  | 
| 584 | // Am I the node which has this integrableObject? | 
| 585 | int whichNode = info_->getMolToProc(i); | 
| 586 | if (whichNode == worldRank) { | 
| 587 | if (myPotato + 1 >= MAXTAG) { | 
| 588 |  | 
| 589 | // The potato was going to exceed the maximum value, | 
| 590 | // so wrap this processor potato back to 0 (and block until | 
| 591 | // node 0 says we can go: | 
| 592 |  | 
| 593 | MPI_Recv(&myPotato, 1, MPI_INT, 0, 0, MPI_COMM_WORLD, | 
| 594 | &istatus); | 
| 595 | } | 
| 596 |  | 
| 597 | mol = info_->getMoleculeByGlobalIndex(i); | 
| 598 |  | 
| 599 |  | 
| 600 | nCurObj = mol->getNIntegrableObjects(); | 
| 601 |  | 
| 602 | MPI_Send(&nCurObj, 1, MPI_INT, 0, myPotato, MPI_COMM_WORLD); | 
| 603 | myPotato++; | 
| 604 |  | 
| 605 | for (integrableObject = mol->beginIntegrableObject(ii); integrableObject != NULL; | 
| 606 | integrableObject = mol->nextIntegrableObject(ii)) { | 
| 607 |  | 
| 608 | if (myPotato + 2 >= MAXTAG) { | 
| 609 |  | 
| 610 | // The potato was going to exceed the maximum value, | 
| 611 | // so wrap this processor potato back to 0 (and block until | 
| 612 | // node 0 says we can go: | 
| 613 |  | 
| 614 | MPI_Recv(&myPotato, 1, MPI_INT, 0, 0, MPI_COMM_WORLD, | 
| 615 | &istatus); | 
| 616 | } | 
| 617 |  | 
| 618 | pos = integrableObject->getPos(); | 
| 619 | vel = integrableObject->getVel(); | 
| 620 |  | 
| 621 | atomData[0] = pos[0]; | 
| 622 | atomData[1] = pos[1]; | 
| 623 | atomData[2] = pos[2]; | 
| 624 |  | 
| 625 | atomData[3] = vel[0]; | 
| 626 | atomData[4] = vel[1]; | 
| 627 | atomData[5] = vel[2]; | 
| 628 |  | 
| 629 | isDirectional = 0; | 
| 630 |  | 
| 631 | if (integrableObject->isDirectional()) { | 
| 632 | isDirectional = 1; | 
| 633 |  | 
| 634 | q = integrableObject->getQ(); | 
| 635 | ji = integrableObject->getJ(); | 
| 636 |  | 
| 637 | atomData[6] = q[0]; | 
| 638 | atomData[7] = q[1]; | 
| 639 | atomData[8] = q[2]; | 
| 640 | atomData[9] = q[3]; | 
| 641 |  | 
| 642 | atomData[10] = ji[0]; | 
| 643 | atomData[11] = ji[1]; | 
| 644 | atomData[12] = ji[2]; | 
| 645 | } | 
| 646 |  | 
| 647 | if (needForceVector_) { | 
| 648 | frc = integrableObject->getFrc(); | 
| 649 | trq = integrableObject->getTrq(); | 
| 650 |  | 
| 651 | if (!isDirectional) { | 
| 652 | atomData[6] = frc[0]; | 
| 653 | atomData[7] = frc[1]; | 
| 654 | atomData[8] = frc[2]; | 
| 655 |  | 
| 656 | atomData[9] = trq[0]; | 
| 657 | atomData[10] = trq[1]; | 
| 658 | atomData[11] = trq[2]; | 
| 659 | } else { | 
| 660 | atomData[13] = frc[0]; | 
| 661 | atomData[14] = frc[1]; | 
| 662 | atomData[15] = frc[2]; | 
| 663 |  | 
| 664 | atomData[16] = trq[0]; | 
| 665 | atomData[17] = trq[1]; | 
| 666 | atomData[18] = trq[2]; | 
| 667 | } | 
| 668 | } | 
| 669 |  | 
| 670 | strncpy(MPIatomTypeString, integrableObject->getType().c_str(), MINIBUFFERSIZE); | 
| 671 |  | 
| 672 | // null terminate the  std::string before sending (just in case): | 
| 673 | MPIatomTypeString[MINIBUFFERSIZE - 1] = '\0'; | 
| 674 |  | 
| 675 | MPI_Send(MPIatomTypeString, MINIBUFFERSIZE, MPI_CHAR, 0, | 
| 676 | myPotato, MPI_COMM_WORLD); | 
| 677 |  | 
| 678 | myPotato++; | 
| 679 |  | 
| 680 | if (isDirectional && needForceVector_) { | 
| 681 | MPI_Send(atomData, 19, MPI_DOUBLE, 0, myPotato, | 
| 682 | MPI_COMM_WORLD); | 
| 683 | } else if (isDirectional) { | 
| 684 | MPI_Send(atomData, 13, MPI_DOUBLE, 0, myPotato, | 
| 685 | MPI_COMM_WORLD); | 
| 686 | } else if (needForceVector_) { | 
| 687 | MPI_Send(atomData, 12, MPI_DOUBLE, 0, myPotato, | 
| 688 | MPI_COMM_WORLD); | 
| 689 | } else { | 
| 690 | MPI_Send(atomData, 6, MPI_DOUBLE, 0, myPotato, | 
| 691 | MPI_COMM_WORLD); | 
| 692 | } | 
| 693 |  | 
| 694 | myPotato++; | 
| 695 | } | 
| 696 |  | 
| 697 | } | 
| 698 |  | 
| 699 | } | 
| 700 | sprintf(checkPointMsg, "Sucessfully took a dump.\n"); | 
| 701 | MPIcheckPoint(); | 
| 702 | } | 
| 703 |  | 
| 704 | #endif // is_mpi | 
| 705 |  | 
| 706 | } | 
| 707 |  | 
| 708 | void DumpWriter::writeDump() { | 
| 709 | writeFrame(*dumpFile_); | 
| 710 | } | 
| 711 |  | 
| 712 | void DumpWriter::writeEor() { | 
| 713 | std::ostream* eorStream; | 
| 714 |  | 
| 715 | #ifdef IS_MPI | 
| 716 | if (worldRank == 0) { | 
| 717 | #endif // is_mpi | 
| 718 |  | 
| 719 | eorStream = createOStream(eorFilename_); | 
| 720 |  | 
| 721 | #ifdef IS_MPI | 
| 722 | } | 
| 723 | #endif // is_mpi | 
| 724 |  | 
| 725 | writeFrame(*eorStream); | 
| 726 |  | 
| 727 | #ifdef IS_MPI | 
| 728 | if (worldRank == 0) { | 
| 729 | #endif // is_mpi | 
| 730 | delete eorStream; | 
| 731 |  | 
| 732 | #ifdef IS_MPI | 
| 733 | } | 
| 734 | #endif // is_mpi | 
| 735 |  | 
| 736 | } | 
| 737 |  | 
| 738 |  | 
| 739 | void DumpWriter::writeDumpAndEor() { | 
| 740 | std::vector<std::streambuf*> buffers; | 
| 741 | std::ostream* eorStream; | 
| 742 | #ifdef IS_MPI | 
| 743 | if (worldRank == 0) { | 
| 744 | #endif // is_mpi | 
| 745 |  | 
| 746 | buffers.push_back(dumpFile_->rdbuf()); | 
| 747 |  | 
| 748 | eorStream = createOStream(eorFilename_); | 
| 749 |  | 
| 750 | buffers.push_back(eorStream->rdbuf()); | 
| 751 |  | 
| 752 | #ifdef IS_MPI | 
| 753 | } | 
| 754 | #endif // is_mpi | 
| 755 |  | 
| 756 | TeeBuf tbuf(buffers.begin(), buffers.end()); | 
| 757 | std::ostream os(&tbuf); | 
| 758 |  | 
| 759 | writeFrame(os); | 
| 760 |  | 
| 761 | #ifdef IS_MPI | 
| 762 | if (worldRank == 0) { | 
| 763 | #endif // is_mpi | 
| 764 | delete eorStream; | 
| 765 |  | 
| 766 | #ifdef IS_MPI | 
| 767 | } | 
| 768 | #endif // is_mpi | 
| 769 |  | 
| 770 | } | 
| 771 |  | 
| 772 | std::ostream* DumpWriter::createOStream(const std::string& filename) { | 
| 773 |  | 
| 774 | std::ostream* newOStream; | 
| 775 | #ifdef HAVE_LIBZ | 
| 776 | if (needCompression_) { | 
| 777 | newOStream = new ogzstream(filename.c_str()); | 
| 778 | } else { | 
| 779 | newOStream = new std::ofstream(filename.c_str()); | 
| 780 | } | 
| 781 | #else | 
| 782 | newOStream = new std::ofstream(filename.c_str()); | 
| 783 | #endif | 
| 784 | return newOStream; | 
| 785 | } | 
| 786 |  | 
| 787 | }//end namespace oopse |