[490] | 1 | #include "globalScopeData.hpp" |
---|
[591] | 2 | #include "xios_spl.hpp" |
---|
[300] | 3 | #include "cxios.hpp" |
---|
[342] | 4 | #include "client.hpp" |
---|
[300] | 5 | #include <boost/functional/hash.hpp> |
---|
| 6 | #include "type.hpp" |
---|
| 7 | #include "context.hpp" |
---|
| 8 | #include "context_client.hpp" |
---|
| 9 | #include "oasis_cinterface.hpp" |
---|
[382] | 10 | #include "mpi.hpp" |
---|
[347] | 11 | #include "timer.hpp" |
---|
[2456] | 12 | #include "mem_checker.hpp" |
---|
[400] | 13 | #include "buffer_client.hpp" |
---|
[1587] | 14 | #include "string_tools.hpp" |
---|
[300] | 15 | |
---|
[335] | 16 | namespace xios |
---|
[490] | 17 | { |
---|
[300] | 18 | |
---|
[1639] | 19 | MPI_Comm CClient::intraComm ; |
---|
| 20 | MPI_Comm CClient::interComm ; |
---|
| 21 | std::list<MPI_Comm> CClient::contextInterComms; |
---|
[1158] | 22 | int CClient::serverLeader ; |
---|
[300] | 23 | bool CClient::is_MPI_Initialized ; |
---|
[1148] | 24 | int CClient::rank_ = INVALID_RANK; |
---|
[490] | 25 | StdOFStream CClient::m_infoStream; |
---|
[523] | 26 | StdOFStream CClient::m_errorStream; |
---|
[1639] | 27 | MPI_Comm& CClient::getInterComm(void) { return (interComm); } |
---|
[1587] | 28 | |
---|
[983] | 29 | ///--------------------------------------------------------------- |
---|
| 30 | /*! |
---|
| 31 | * \fn void CClient::initialize(const string& codeId, MPI_Comm& localComm, MPI_Comm& returnComm) |
---|
| 32 | * Function creates intraComm (CClient::intraComm) for client group with id=codeId and interComm (CClient::interComm) between client and server groups. |
---|
| 33 | * \param [in] codeId identity of context. |
---|
| 34 | * \param [in/out] localComm local communicator. |
---|
| 35 | * \param [in/out] returnComm (intra)communicator of client group. |
---|
| 36 | */ |
---|
| 37 | |
---|
[1639] | 38 | void CClient::initialize(const string& codeId, MPI_Comm& localComm, MPI_Comm& returnComm) |
---|
[300] | 39 | { |
---|
| 40 | int initialized ; |
---|
[1639] | 41 | MPI_Initialized(&initialized) ; |
---|
[300] | 42 | if (initialized) is_MPI_Initialized=true ; |
---|
| 43 | else is_MPI_Initialized=false ; |
---|
[1152] | 44 | int rank ; |
---|
[490] | 45 | |
---|
[300] | 46 | // don't use OASIS |
---|
| 47 | if (!CXios::usingOasis) |
---|
| 48 | { |
---|
[983] | 49 | // localComm isn't given |
---|
[1639] | 50 | if (localComm == MPI_COMM_NULL) |
---|
[300] | 51 | { |
---|
[490] | 52 | if (!is_MPI_Initialized) |
---|
[300] | 53 | { |
---|
[1639] | 54 | MPI_Init(NULL, NULL); |
---|
[300] | 55 | } |
---|
[359] | 56 | CTimer::get("XIOS").resume() ; |
---|
[1158] | 57 | CTimer::get("XIOS init/finalize").resume() ; |
---|
[490] | 58 | boost::hash<string> hashString ; |
---|
| 59 | |
---|
[1158] | 60 | unsigned long hashClient=hashString(codeId) ; |
---|
| 61 | unsigned long hashServer=hashString(CXios::xiosCodeId) ; |
---|
[300] | 62 | unsigned long* hashAll ; |
---|
| 63 | int size ; |
---|
| 64 | int myColor ; |
---|
| 65 | int i,c ; |
---|
[1639] | 66 | MPI_Comm newComm ; |
---|
[490] | 67 | |
---|
[1639] | 68 | MPI_Comm_size(CXios::globalComm,&size) ; |
---|
| 69 | MPI_Comm_rank(CXios::globalComm,&rank_); |
---|
[490] | 70 | |
---|
[300] | 71 | hashAll=new unsigned long[size] ; |
---|
[490] | 72 | |
---|
[1639] | 73 | MPI_Allgather(&hashClient,1,MPI_LONG,hashAll,1,MPI_LONG,CXios::globalComm) ; |
---|
[300] | 74 | |
---|
| 75 | map<unsigned long, int> colors ; |
---|
| 76 | map<unsigned long, int> leaders ; |
---|
[490] | 77 | |
---|
[300] | 78 | for(i=0,c=0;i<size;i++) |
---|
| 79 | { |
---|
| 80 | if (colors.find(hashAll[i])==colors.end()) |
---|
| 81 | { |
---|
| 82 | colors[hashAll[i]] =c ; |
---|
| 83 | leaders[hashAll[i]]=i ; |
---|
| 84 | c++ ; |
---|
| 85 | } |
---|
| 86 | } |
---|
[490] | 87 | |
---|
[491] | 88 | // Verify whether we are on server mode or not |
---|
| 89 | CXios::setNotUsingServer(); |
---|
| 90 | for (i=0; i < size; ++i) |
---|
| 91 | { |
---|
[1158] | 92 | if (hashServer == hashAll[i]) |
---|
[491] | 93 | { |
---|
| 94 | CXios::setUsingServer(); |
---|
| 95 | break; |
---|
| 96 | } |
---|
| 97 | } |
---|
| 98 | |
---|
[1243] | 99 | myColor=colors[hashClient]; |
---|
[1639] | 100 | MPI_Comm_split(CXios::globalComm,myColor,rank_,&intraComm) ; |
---|
[300] | 101 | |
---|
| 102 | if (CXios::usingServer) |
---|
[490] | 103 | { |
---|
[300] | 104 | int clientLeader=leaders[hashClient] ; |
---|
[1148] | 105 | serverLeader=leaders[hashServer] ; |
---|
[493] | 106 | int intraCommSize, intraCommRank ; |
---|
[1639] | 107 | MPI_Comm_size(intraComm,&intraCommSize) ; |
---|
| 108 | MPI_Comm_rank(intraComm,&intraCommRank) ; |
---|
[1148] | 109 | info(50)<<"intercommCreate::client "<<rank_<<" intraCommSize : "<<intraCommSize |
---|
| 110 | <<" intraCommRank :"<<intraCommRank<<" clientLeader "<< serverLeader<<endl ; |
---|
[1639] | 111 | MPI_Intercomm_create(intraComm, 0, CXios::globalComm, serverLeader, 0, &interComm) ; |
---|
[1152] | 112 | //rank_ = intraCommRank; |
---|
[300] | 113 | } |
---|
| 114 | else |
---|
| 115 | { |
---|
[1639] | 116 | MPI_Comm_dup(intraComm,&interComm) ; |
---|
[300] | 117 | } |
---|
| 118 | delete [] hashAll ; |
---|
| 119 | } |
---|
| 120 | // localComm argument is given |
---|
[490] | 121 | else |
---|
[300] | 122 | { |
---|
| 123 | if (CXios::usingServer) |
---|
[490] | 124 | { |
---|
[300] | 125 | //ERROR("void CClient::initialize(const string& codeId,MPI_Comm& localComm,MPI_Comm& returnComm)", << " giving a local communictor is not compatible with using server mode") ; |
---|
| 126 | } |
---|
| 127 | else |
---|
| 128 | { |
---|
[1639] | 129 | MPI_Comm_dup(localComm,&intraComm) ; |
---|
| 130 | MPI_Comm_dup(intraComm,&interComm) ; |
---|
[300] | 131 | } |
---|
| 132 | } |
---|
| 133 | } |
---|
| 134 | // using OASIS |
---|
| 135 | else |
---|
| 136 | { |
---|
[983] | 137 | // localComm isn't given |
---|
[1639] | 138 | if (localComm == MPI_COMM_NULL) |
---|
[300] | 139 | { |
---|
| 140 | if (!is_MPI_Initialized) oasis_init(codeId) ; |
---|
[655] | 141 | oasis_get_localcomm(localComm) ; |
---|
[300] | 142 | } |
---|
[1639] | 143 | MPI_Comm_dup(localComm,&intraComm) ; |
---|
[655] | 144 | |
---|
[359] | 145 | CTimer::get("XIOS").resume() ; |
---|
[1158] | 146 | CTimer::get("XIOS init/finalize").resume() ; |
---|
[511] | 147 | |
---|
| 148 | if (CXios::usingServer) |
---|
[300] | 149 | { |
---|
[1639] | 150 | MPI_Status status ; |
---|
| 151 | MPI_Comm_rank(intraComm,&rank_) ; |
---|
[506] | 152 | |
---|
[300] | 153 | oasis_get_intercomm(interComm,CXios::xiosCodeId) ; |
---|
[1639] | 154 | if (rank_==0) MPI_Recv(&serverLeader,1, MPI_INT, 0, 0, interComm, &status) ; |
---|
| 155 | MPI_Bcast(&serverLeader,1,MPI_INT,0,intraComm) ; |
---|
[300] | 156 | } |
---|
[1639] | 157 | else MPI_Comm_dup(intraComm,&interComm) ; |
---|
[300] | 158 | } |
---|
[490] | 159 | |
---|
[1639] | 160 | MPI_Comm_dup(intraComm,&returnComm) ; |
---|
[300] | 161 | } |
---|
[490] | 162 | |
---|
[983] | 163 | ///--------------------------------------------------------------- |
---|
| 164 | /*! |
---|
| 165 | * \fn void CClient::registerContext(const string& id, MPI_Comm contextComm) |
---|
[1021] | 166 | * \brief Sends a request to create a context to server. Creates client/server contexts. |
---|
[983] | 167 | * \param [in] id id of context. |
---|
| 168 | * \param [in] contextComm. |
---|
[1021] | 169 | * Function is only called by client. |
---|
[983] | 170 | */ |
---|
[1639] | 171 | void CClient::registerContext(const string& id, MPI_Comm contextComm) |
---|
[300] | 172 | { |
---|
[346] | 173 | CContext::setCurrent(id) ; |
---|
[511] | 174 | CContext* context=CContext::create(id); |
---|
| 175 | StdString idServer(id); |
---|
| 176 | idServer += "_server"; |
---|
[490] | 177 | |
---|
[983] | 178 | if (CXios::isServer && !context->hasServer) |
---|
| 179 | // Attached mode |
---|
[300] | 180 | { |
---|
[1639] | 181 | MPI_Comm contextInterComm ; |
---|
| 182 | MPI_Comm_dup(contextComm,&contextInterComm) ; |
---|
[983] | 183 | CContext* contextServer = CContext::create(idServer); |
---|
| 184 | |
---|
| 185 | // Firstly, initialize context on client side |
---|
| 186 | context->initClient(contextComm,contextInterComm, contextServer); |
---|
| 187 | |
---|
| 188 | // Secondly, initialize context on server side |
---|
| 189 | contextServer->initServer(contextComm,contextInterComm, context); |
---|
| 190 | |
---|
| 191 | // Finally, we should return current context to context client |
---|
| 192 | CContext::setCurrent(id); |
---|
| 193 | |
---|
| 194 | contextInterComms.push_back(contextInterComm); |
---|
| 195 | } |
---|
| 196 | else |
---|
| 197 | { |
---|
[300] | 198 | int size,rank,globalRank ; |
---|
| 199 | size_t message_size ; |
---|
| 200 | int leaderRank ; |
---|
[1639] | 201 | MPI_Comm contextInterComm ; |
---|
[490] | 202 | |
---|
[1639] | 203 | MPI_Comm_size(contextComm,&size) ; |
---|
| 204 | MPI_Comm_rank(contextComm,&rank) ; |
---|
| 205 | MPI_Comm_rank(CXios::globalComm,&globalRank) ; |
---|
[300] | 206 | if (rank!=0) globalRank=0 ; |
---|
[490] | 207 | |
---|
[300] | 208 | CMessage msg ; |
---|
[511] | 209 | msg<<idServer<<size<<globalRank ; |
---|
| 210 | // msg<<id<<size<<globalRank ; |
---|
[300] | 211 | |
---|
| 212 | int messageSize=msg.size() ; |
---|
[1158] | 213 | char * buff = new char[messageSize] ; |
---|
| 214 | CBufferOut buffer((void*)buff,messageSize) ; |
---|
[300] | 215 | buffer<<msg ; |
---|
[490] | 216 | |
---|
[1639] | 217 | MPI_Send((void*)buff,buffer.count(),MPI_CHAR,serverLeader,1,CXios::globalComm) ; |
---|
[1158] | 218 | |
---|
[1639] | 219 | MPI_Intercomm_create(contextComm,0,CXios::globalComm,serverLeader,10+globalRank,&contextInterComm) ; |
---|
[1148] | 220 | info(10)<<"Register new Context : "<<id<<endl ; |
---|
[1639] | 221 | MPI_Comm inter ; |
---|
| 222 | MPI_Intercomm_merge(contextInterComm,0,&inter) ; |
---|
| 223 | MPI_Barrier(inter) ; |
---|
[1858] | 224 | MPI_Comm_free(&inter); |
---|
[490] | 225 | |
---|
[1148] | 226 | context->initClient(contextComm,contextInterComm) ; |
---|
[490] | 227 | |
---|
[1148] | 228 | contextInterComms.push_back(contextInterComm); |
---|
[1009] | 229 | delete [] buff ; |
---|
[300] | 230 | |
---|
| 231 | } |
---|
| 232 | } |
---|
[490] | 233 | |
---|
[1587] | 234 | /*! |
---|
| 235 | * \fn void CClient::callOasisEnddef(void) |
---|
| 236 | * \brief Send the order to the servers to call "oasis_enddef". It must be done by each compound of models before calling oasis_enddef on client side |
---|
| 237 | * Function is only called by client. |
---|
| 238 | */ |
---|
| 239 | void CClient::callOasisEnddef(void) |
---|
| 240 | { |
---|
| 241 | bool oasisEnddef=CXios::getin<bool>("call_oasis_enddef",true) ; |
---|
| 242 | if (!oasisEnddef) ERROR("void CClient::callOasisEnddef(void)", <<"Function xios_oasis_enddef called but variable <call_oasis_enddef> is set to false."<<endl |
---|
| 243 | <<"Variable <call_oasis_enddef> must be set to true"<<endl) ; |
---|
| 244 | if (CXios::isServer) |
---|
| 245 | // Attached mode |
---|
| 246 | { |
---|
| 247 | // nothing to do |
---|
| 248 | } |
---|
| 249 | else |
---|
| 250 | { |
---|
| 251 | int rank ; |
---|
| 252 | int msg=0 ; |
---|
| 253 | |
---|
[1639] | 254 | MPI_Comm_rank(intraComm,&rank) ; |
---|
[1587] | 255 | if (rank==0) |
---|
| 256 | { |
---|
[1639] | 257 | MPI_Send(&msg,1,MPI_INT,0,5,interComm) ; // tags oasis_endded = 5 |
---|
[1587] | 258 | } |
---|
| 259 | |
---|
| 260 | } |
---|
| 261 | } |
---|
| 262 | |
---|
| 263 | |
---|
[300] | 264 | void CClient::finalize(void) |
---|
| 265 | { |
---|
| 266 | int rank ; |
---|
| 267 | int msg=0 ; |
---|
[697] | 268 | |
---|
[1639] | 269 | MPI_Comm_rank(intraComm,&rank) ; |
---|
[697] | 270 | |
---|
[992] | 271 | if (!CXios::isServer) |
---|
[300] | 272 | { |
---|
[1639] | 273 | MPI_Comm_rank(intraComm,&rank) ; |
---|
[490] | 274 | if (rank==0) |
---|
[332] | 275 | { |
---|
[1639] | 276 | MPI_Send(&msg,1,MPI_INT,0,0,interComm) ; |
---|
[332] | 277 | } |
---|
[300] | 278 | } |
---|
[490] | 279 | |
---|
[1866] | 280 | |
---|
[2029] | 281 | /* MPI_Comm_free(&interComm); */ // WARNING remove freeing communicator !! --> deadlock raised, to be checked |
---|
[1639] | 282 | for (std::list<MPI_Comm>::iterator it = contextInterComms.begin(); it != contextInterComms.end(); it++) |
---|
[2029] | 283 | /* MPI_Comm_free(&(*it)) */ ; // WARNING remove freeing communicator !! --> deadlock raised, to be checked |
---|
[1639] | 284 | MPI_Comm_free(&intraComm); |
---|
[361] | 285 | |
---|
[1158] | 286 | CTimer::get("XIOS init/finalize").suspend() ; |
---|
[655] | 287 | CTimer::get("XIOS").suspend() ; |
---|
| 288 | |
---|
[300] | 289 | if (!is_MPI_Initialized) |
---|
| 290 | { |
---|
| 291 | if (CXios::usingOasis) oasis_finalize(); |
---|
[1639] | 292 | else MPI_Finalize() ; |
---|
[300] | 293 | } |
---|
[697] | 294 | |
---|
[300] | 295 | info(20) << "Client side context is finalized"<<endl ; |
---|
[1158] | 296 | report(0) <<" Performance report : Whole time from XIOS init and finalize: "<< CTimer::get("XIOS init/finalize").getCumulatedTime()<<" s"<<endl ; |
---|
[490] | 297 | report(0) <<" Performance report : total time spent for XIOS : "<< CTimer::get("XIOS").getCumulatedTime()<<" s"<<endl ; |
---|
[347] | 298 | report(0)<< " Performance report : time spent for waiting free buffer : "<< CTimer::get("Blocking time").getCumulatedTime()<<" s"<<endl ; |
---|
[1158] | 299 | report(0)<< " Performance report : Ratio : "<< CTimer::get("Blocking time").getCumulatedTime()/CTimer::get("XIOS init/finalize").getCumulatedTime()*100.<<" %"<<endl ; |
---|
[347] | 300 | report(0)<< " Performance report : This ratio must be close to zero. Otherwise it may be usefull to increase buffer size or numbers of server"<<endl ; |
---|
[511] | 301 | // report(0)<< " Memory report : Current buffer_size : "<<CXios::bufferSize<<endl ; |
---|
[732] | 302 | report(0)<< " Memory report : Minimum buffer size required : " << CClientBuffer::maxRequestSize << " bytes" << endl ; |
---|
[400] | 303 | report(0)<< " Memory report : increasing it by a factor will increase performance, depending of the volume of data wrote in file at each time step of the file"<<endl ; |
---|
[1158] | 304 | report(100)<<CTimer::getAllCumulatedTime()<<endl ; |
---|
[2503] | 305 | if (CXios::reportMemory) |
---|
| 306 | { |
---|
| 307 | report(100)<<CMemChecker::getAllCumulatedMem()<<endl ; |
---|
| 308 | } |
---|
[400] | 309 | } |
---|
[490] | 310 | |
---|
[1148] | 311 | /*! |
---|
[1243] | 312 | * Return global rank without oasis and current rank in model intraComm in case of oasis |
---|
[1148] | 313 | */ |
---|
[490] | 314 | int CClient::getRank() |
---|
| 315 | { |
---|
[1148] | 316 | return rank_; |
---|
[490] | 317 | } |
---|
| 318 | |
---|
[523] | 319 | /*! |
---|
| 320 | * Open a file specified by a suffix and an extension and use it for the given file buffer. |
---|
| 321 | * The file name will be suffix+rank+extension. |
---|
| 322 | * |
---|
| 323 | * \param fileName[in] protype file name |
---|
| 324 | * \param ext [in] extension of the file |
---|
| 325 | * \param fb [in/out] the file buffer |
---|
| 326 | */ |
---|
| 327 | void CClient::openStream(const StdString& fileName, const StdString& ext, std::filebuf* fb) |
---|
| 328 | { |
---|
| 329 | StdStringStream fileNameClient; |
---|
| 330 | int numDigit = 0; |
---|
| 331 | int size = 0; |
---|
[1233] | 332 | int rank; |
---|
[1639] | 333 | MPI_Comm_size(CXios::globalComm, &size); |
---|
[523] | 334 | while (size) |
---|
| 335 | { |
---|
| 336 | size /= 10; |
---|
| 337 | ++numDigit; |
---|
| 338 | } |
---|
[497] | 339 | |
---|
[1233] | 340 | if (CXios::usingOasis) |
---|
| 341 | { |
---|
[1639] | 342 | MPI_Comm_rank(CXios::globalComm,&rank); |
---|
[1233] | 343 | fileNameClient << fileName << "_" << std::setfill('0') << std::setw(numDigit) << rank << ext; |
---|
| 344 | } |
---|
| 345 | else |
---|
| 346 | fileNameClient << fileName << "_" << std::setfill('0') << std::setw(numDigit) << getRank() << ext; |
---|
| 347 | |
---|
| 348 | |
---|
[523] | 349 | fb->open(fileNameClient.str().c_str(), std::ios::out); |
---|
| 350 | if (!fb->is_open()) |
---|
| 351 | ERROR("void CClient::openStream(const StdString& fileName, const StdString& ext, std::filebuf* fb)", |
---|
[1542] | 352 | << std::endl << "Can not open <" << fileNameClient.str() << "> file to write the client log(s)."); |
---|
[523] | 353 | } |
---|
[490] | 354 | |
---|
[523] | 355 | /*! |
---|
| 356 | * \brief Open a file stream to write the info logs |
---|
| 357 | * Open a file stream with a specific file name suffix+rank |
---|
| 358 | * to write the info logs. |
---|
| 359 | * \param fileName [in] protype file name |
---|
| 360 | */ |
---|
| 361 | void CClient::openInfoStream(const StdString& fileName) |
---|
| 362 | { |
---|
| 363 | std::filebuf* fb = m_infoStream.rdbuf(); |
---|
| 364 | openStream(fileName, ".out", fb); |
---|
[490] | 365 | |
---|
[523] | 366 | info.write2File(fb); |
---|
| 367 | report.write2File(fb); |
---|
| 368 | } |
---|
[490] | 369 | |
---|
[523] | 370 | //! Write the info logs to standard output |
---|
| 371 | void CClient::openInfoStream() |
---|
| 372 | { |
---|
| 373 | info.write2StdOut(); |
---|
| 374 | report.write2StdOut(); |
---|
| 375 | } |
---|
[490] | 376 | |
---|
[523] | 377 | //! Close the info logs file if it opens |
---|
| 378 | void CClient::closeInfoStream() |
---|
| 379 | { |
---|
| 380 | if (m_infoStream.is_open()) m_infoStream.close(); |
---|
| 381 | } |
---|
[490] | 382 | |
---|
[523] | 383 | /*! |
---|
| 384 | * \brief Open a file stream to write the error log |
---|
| 385 | * Open a file stream with a specific file name suffix+rank |
---|
| 386 | * to write the error log. |
---|
| 387 | * \param fileName [in] protype file name |
---|
| 388 | */ |
---|
| 389 | void CClient::openErrorStream(const StdString& fileName) |
---|
| 390 | { |
---|
| 391 | std::filebuf* fb = m_errorStream.rdbuf(); |
---|
| 392 | openStream(fileName, ".err", fb); |
---|
| 393 | |
---|
| 394 | error.write2File(fb); |
---|
| 395 | } |
---|
| 396 | |
---|
| 397 | //! Write the error log to standard error output |
---|
| 398 | void CClient::openErrorStream() |
---|
| 399 | { |
---|
| 400 | error.write2StdErr(); |
---|
| 401 | } |
---|
| 402 | |
---|
| 403 | //! Close the error log file if it opens |
---|
| 404 | void CClient::closeErrorStream() |
---|
| 405 | { |
---|
| 406 | if (m_errorStream.is_open()) m_errorStream.close(); |
---|
| 407 | } |
---|
[300] | 408 | } |
---|