examples/msg/actions/actions.c

   1 /* Copyright (c) 2009, 2010. The SimGrid Team.
   2  * All rights reserved.                                                     */
   3
   4 /* This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the license (GNU LGPL) which comes with this package. */
   6
   7 #include <stdio.h>
   8 #include <stdlib.h>
   9 #include "msg/msg.h"            /* Yeah! If you want to use msg, you need to include msg/msg.h */
  10 #include "simix/simix.h"        /* semaphores for the barrier */
  11 #include "xbt.h"                /* calloc, printf */
  12 #include "simgrid_config.h"     /* getline */
  13 #include "instr/instr_private.h"
  14
  15 void SIMIX_ctx_raw_factory_init(smx_context_factory_t *factory);
  16
  17 XBT_LOG_NEW_DEFAULT_CATEGORY(actions,
  18                              "Messages specific for this msg example");
  19 int communicator_size = 0;
  20
  21 static void action_Isend(char*const* action);
  22
  23 typedef struct  {
  24   int last_Irecv_sender_id;
  25   int bcast_counter;
  26   int reduce_counter;
  27   int allReduce_counter;
  28   xbt_dynar_t isends; /* of msg_comm_t */
  29   /* Used to implement irecv+wait */
  30   xbt_dynar_t irecvs; /* of msg_comm_t */
  31   xbt_dynar_t tasks; /* of m_task_t */
  32 } s_process_globals_t, *process_globals_t;
  33
  34 /* Helper function */
  35 static double parse_double(const char *string)
  36 {
  37   double value;
  38   char *endptr;
  39
  40   value = strtod(string, &endptr);
  41   if (*endptr != '\0')
  42     THROW1(unknown_error, 0, "%s is not a double", string);
  43   return value;
  44 }
  45
  46 static int get_rank (const char *process_name)
  47 {
  48   return atoi(&(process_name[1]));
  49 }
  50
  51 static void asynchronous_cleanup(void) {
  52   process_globals_t globals = (process_globals_t) MSG_process_get_data(MSG_process_self());
  53
  54   /* Destroy any isend which correspond to completed communications */
  55   int found;
  56   msg_comm_t comm;
  57   while ((found = MSG_comm_testany(globals->isends)) != -1) {
  58     xbt_dynar_remove_at(globals->isends,found,&comm);
  59     MSG_comm_destroy(comm);
  60   }
  61 }
  62
  63 /* My actions */
  64 static void action_send(char*const* action)
  65 {
  66   char *name = NULL;
  67   char to[250];
  68   const char *size_str = action[3];
  69   double size=parse_double(size_str);
  70   double clock = MSG_get_clock(); /* this "call" is free thanks to inlining */
  71
  72   sprintf(to, "%s_%s", MSG_process_get_name(MSG_process_self()),action[2]);
  73
  74   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
  75     name = xbt_str_join_array(action, " ");
  76
  77 #ifdef HAVE_TRACING
  78   int rank = get_rank(MSG_process_get_name(MSG_process_self()));
  79   int dst_traced = get_rank(action[2]);
  80   TRACE_smpi_ptp_in(rank, rank, dst_traced, "send");
  81   TRACE_smpi_send(rank, rank, dst_traced);
  82 #endif
  83
  84   DEBUG2("Entering Send: %s (size: %lg)", name, size);
  85    if (size<65536) {
  86      action_Isend(action);
  87    } else {
  88      MSG_task_send(MSG_task_create(name, 0, size, NULL), to);
  89    }
  90
  91    VERB2("%s %f", name, MSG_get_clock() - clock);
  92
  93   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
  94     free(name);
  95
  96 #ifdef HAVE_TRACING
  97   TRACE_smpi_ptp_out(rank, rank, dst_traced, "send");
  98 #endif
  99
 100   asynchronous_cleanup();
 101 }
 102
 103 static void action_Isend(char*const* action)
 104 {
 105   char to[250];
 106   const char *size = action[3];
 107   double clock = MSG_get_clock();
 108   process_globals_t globals = (process_globals_t) MSG_process_get_data(MSG_process_self());
 109
 110
 111   sprintf(to, "%s_%s", MSG_process_get_name(MSG_process_self()),action[2]);
 112   msg_comm_t comm =
 113       MSG_task_isend( MSG_task_create(to,0,parse_double(size),NULL), to);
 114   xbt_dynar_push(globals->isends,&comm);
 115
 116   DEBUG1("Isend on %s", MSG_process_get_name(MSG_process_self()));
 117   VERB2("%s %f", xbt_str_join_array(action, " "), MSG_get_clock() - clock);
 118
 119   asynchronous_cleanup();
 120 }
 121
 122
 123 static void action_recv(char*const* action)
 124 {
 125   char *name = NULL;
 126   char mailbox_name[250];
 127   m_task_t task = NULL;
 128   double clock = MSG_get_clock();
 129
 130   sprintf(mailbox_name, "%s_%s", action[2],
 131           MSG_process_get_name(MSG_process_self()));
 132
 133   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 134     name = xbt_str_join_array(action, " ");
 135
 136 #ifdef HAVE_TRACING
 137   int rank = get_rank(MSG_process_get_name(MSG_process_self()));
 138   int src_traced = get_rank(action[2]);
 139   TRACE_smpi_ptp_in(rank, src_traced, rank, "recv");
 140 #endif
 141
 142   DEBUG1("Receiving: %s", name);
 143   MSG_task_receive(&task, mailbox_name);
 144   //  MSG_task_receive(&task, MSG_process_get_name(MSG_process_self()));
 145   VERB2("%s %f", name, MSG_get_clock() - clock);
 146   MSG_task_destroy(task);
 147
 148   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 149     free(name);
 150 #ifdef HAVE_TRACING
 151   TRACE_smpi_ptp_out(rank, src_traced, rank, "recv");
 152   TRACE_smpi_recv(rank, src_traced, rank);
 153 #endif
 154
 155   asynchronous_cleanup();
 156 }
 157
 158 static void action_Irecv(char*const* action)
 159 {
 160   char mailbox[250];
 161   double clock = MSG_get_clock();
 162   process_globals_t globals = (process_globals_t) MSG_process_get_data(MSG_process_self());
 163
 164   DEBUG1("Irecv on %s", MSG_process_get_name(MSG_process_self()));
 165 #ifdef HAVE_TRACING
 166   int rank = get_rank(MSG_process_get_name(MSG_process_self()));
 167   int src_traced = get_rank(action[2]);
 168   globals->last_Irecv_sender_id = src_traced;
 169   MSG_process_set_data(MSG_process_self(), (void *) globals);
 170
 171   TRACE_smpi_ptp_in(rank, src_traced, rank, "Irecv");
 172 #endif
 173
 174   sprintf(mailbox, "%s_%s", action[2],
 175           MSG_process_get_name(MSG_process_self()));
 176   m_task_t t=NULL;
 177   xbt_dynar_push(globals->tasks,&t);
 178   msg_comm_t c =
 179       MSG_task_irecv(
 180           xbt_dynar_get_ptr(globals->tasks, xbt_dynar_length(globals->tasks)-1),
 181           mailbox);
 182   xbt_dynar_push(globals->irecvs,&c);
 183
 184   VERB2("%s %f", xbt_str_join_array(action, " "), MSG_get_clock() - clock);
 185
 186 #ifdef HAVE_TRACING
 187   TRACE_smpi_ptp_out(rank, src_traced, rank, "Irecv");
 188 #endif
 189
 190   asynchronous_cleanup();
 191 }
 192
 193
 194 static void action_wait(char*const* action)
 195 {
 196   char *name = NULL;
 197   m_task_t task = NULL;
 198   msg_comm_t comm;
 199   double clock = MSG_get_clock();
 200   process_globals_t globals = (process_globals_t) MSG_process_get_data(MSG_process_self());
 201
 202   xbt_assert1(xbt_dynar_length(globals->irecvs),
 203       "action wait not preceded by any irecv: %s", xbt_str_join_array(action," "));
 204
 205   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 206     name = xbt_str_join_array(action, " ");
 207 #ifdef HAVE_TRACING
 208   process_globals_t counters = (process_globals_t) MSG_process_get_data(MSG_process_self());
 209   int src_traced = counters->last_Irecv_sender_id;
 210   int rank = get_rank(MSG_process_get_name(MSG_process_self()));
 211   TRACE_smpi_ptp_in(rank, src_traced, rank, "wait");
 212 #endif
 213
 214   DEBUG1("Entering %s", name);
 215   comm = xbt_dynar_pop_as(globals->irecvs,msg_comm_t);
 216   MSG_comm_wait(comm,-1);
 217   task = xbt_dynar_pop_as(globals->tasks,m_task_t);
 218   MSG_comm_destroy(comm);
 219   MSG_task_destroy(task);
 220
 221   VERB2("%s %f", name, MSG_get_clock() - clock);
 222   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 223     free(name);
 224 #ifdef HAVE_TRACING
 225   TRACE_smpi_ptp_out(rank, src_traced, rank, "wait");
 226   TRACE_smpi_recv(rank, src_traced, rank);
 227 #endif
 228
 229 }
 230
 231 /* FIXME: that's a poor man's implementation: we should take the message exchanges into account */
 232 static void action_barrier(char*const* action)
 233 {
 234   char *name = NULL;
 235   static smx_mutex_t mutex = NULL;
 236   static smx_cond_t cond = NULL;
 237   static int processes_arrived_sofar=0;
 238
 239   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 240     name = xbt_str_join_array(action, " ");
 241
 242   if (mutex == NULL) {       // first arriving on the barrier
 243     mutex = SIMIX_req_mutex_init();
 244     cond = SIMIX_req_cond_init();
 245     processes_arrived_sofar=0;
 246   }
 247   DEBUG2("Entering barrier: %s (%d already there)", name,processes_arrived_sofar);
 248
 249   SIMIX_req_mutex_lock(mutex);
 250   if (++processes_arrived_sofar == communicator_size) {
 251     SIMIX_req_cond_broadcast(cond);
 252     SIMIX_req_mutex_unlock(mutex);
 253   } else {
 254     SIMIX_req_cond_wait(cond,mutex);
 255     SIMIX_req_mutex_unlock(mutex);
 256   }
 257
 258   DEBUG1("Exiting barrier: %s", name);
 259
 260   processes_arrived_sofar--;
 261   if (!processes_arrived_sofar) {
 262     SIMIX_req_cond_destroy(cond);
 263     SIMIX_req_mutex_destroy(mutex);
 264     mutex=NULL;
 265   }
 266
 267   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 268     free(name);
 269
 270 }
 271
 272 static void action_reduce(char*const* action)
 273 {
 274         int i;
 275         char *reduce_identifier;
 276         char mailbox[80];
 277         double comm_size = parse_double(action[2]);
 278         double comp_size = parse_double(action[3]);
 279         m_task_t comp_task = NULL;
 280         const char *process_name;
 281         double clock = MSG_get_clock();
 282
 283         process_globals_t counters = (process_globals_t) MSG_process_get_data(MSG_process_self());
 284
 285         xbt_assert0(communicator_size, "Size of Communicator is not defined, "
 286                         "can't use collective operations");
 287
 288         process_name = MSG_process_get_name(MSG_process_self());
 289
 290         reduce_identifier = bprintf("reduce_%d", counters->reduce_counter++);
 291
 292         if (!strcmp(process_name, "p0")) {
 293                 DEBUG2("%s: %s is the Root", reduce_identifier, process_name);
 294
 295                 msg_comm_t *comms = xbt_new0(msg_comm_t,communicator_size-1);
 296             m_task_t *tasks = xbt_new0(m_task_t,communicator_size-1);
 297             for (i = 1; i < communicator_size; i++) {
 298               sprintf(mailbox, "%s_p%d_p0", reduce_identifier, i);
 299               comms[i-1] = MSG_task_irecv(&(tasks[i-1]),mailbox);
 300             }
 301             MSG_comm_waitall(comms,communicator_size-1,-1);
 302             for (i = 1; i < communicator_size; i++) {
 303                 MSG_comm_destroy(comms[i-1]);
 304                 MSG_task_destroy(tasks[i-1]);
 305             }
 306             free(tasks);
 307
 308             comp_task = MSG_task_create("reduce_comp", comp_size, 0, NULL);
 309             DEBUG1("%s: computing 'reduce_comp'", reduce_identifier);
 310             MSG_task_execute(comp_task);
 311             MSG_task_destroy(comp_task);
 312             DEBUG1("%s: computed", reduce_identifier);
 313
 314         } else {
 315                 DEBUG2("%s: %s sends", reduce_identifier, process_name);
 316                 sprintf(mailbox, "%s_%s_p0", reduce_identifier, process_name);
 317             DEBUG1("put on %s", mailbox);
 318             MSG_task_send(MSG_task_create(reduce_identifier, 0, comm_size, NULL),
 319                           mailbox);
 320         }
 321
 322         VERB2("%s %f", xbt_str_join_array(action, " "), MSG_get_clock() - clock);
 323         free(reduce_identifier);
 324 }
 325
 326 static void action_bcast(char*const* action)
 327 {
 328         int i;
 329         char *bcast_identifier;
 330         char mailbox[80];
 331         double comm_size = parse_double(action[2]);
 332         m_task_t task = NULL;
 333         const char *process_name;
 334         double clock = MSG_get_clock();
 335
 336         process_globals_t counters = (process_globals_t) MSG_process_get_data(MSG_process_self());
 337
 338         xbt_assert0(communicator_size, "Size of Communicator is not defined, "
 339                         "can't use collective operations");
 340
 341         process_name = MSG_process_get_name(MSG_process_self());
 342
 343         bcast_identifier = bprintf("bcast_%d", counters->bcast_counter++);
 344
 345         if (!strcmp(process_name, "p0")) {
 346                 DEBUG2("%s: %s is the Root", bcast_identifier, process_name);
 347
 348             msg_comm_t *comms = xbt_new0(msg_comm_t,communicator_size-1);
 349
 350             for (i = 1; i < communicator_size; i++) {
 351               sprintf(mailbox, "%s_p0_p%d", bcast_identifier, i);
 352               comms[i-1] =
 353                   MSG_task_isend(MSG_task_create(mailbox,0,comm_size,NULL),
 354                       mailbox);
 355             }
 356             MSG_comm_waitall(comms,communicator_size-1,-1);
 357                 for (i = 1; i < communicator_size; i++)
 358                MSG_comm_destroy(comms[i-1]);
 359             free(comms);
 360
 361             DEBUG2("%s: all messages sent by %s have been received",
 362                    bcast_identifier, process_name);
 363
 364         } else {
 365             sprintf(mailbox, "%s_p0_%s", bcast_identifier, process_name);
 366             MSG_task_receive(&task, mailbox);
 367             MSG_task_destroy(task);
 368             DEBUG2("%s: %s has received", bcast_identifier, process_name);
 369         }
 370
 371         VERB2("%s %f", xbt_str_join_array(action, " "), MSG_get_clock() - clock);
 372         free(bcast_identifier);
 373 }
 374
 375
 376 static void action_sleep(char*const* action)
 377 {
 378   char *name = NULL;
 379   const char *duration = action[2];
 380   double clock = MSG_get_clock();
 381
 382   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 383     name = xbt_str_join_array(action, " ");
 384
 385   DEBUG1("Entering %s", name);
 386   MSG_process_sleep(parse_double(duration));
 387   VERB2("%s %f ", name, MSG_get_clock() - clock);
 388
 389   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 390     free(name);
 391 }
 392
 393 static void action_allReduce(char*const* action) {
 394   int i;
 395   char *allreduce_identifier;
 396   char mailbox[80];
 397   double comm_size = parse_double(action[2]);
 398   double comp_size = parse_double(action[3]);
 399   m_task_t task = NULL, comp_task = NULL;
 400   const char *process_name;
 401   double clock = MSG_get_clock();
 402
 403   process_globals_t counters = (process_globals_t) MSG_process_get_data(MSG_process_self());
 404
 405   xbt_assert0(communicator_size, "Size of Communicator is not defined, "
 406               "can't use collective operations");
 407
 408   process_name = MSG_process_get_name(MSG_process_self());
 409
 410   allreduce_identifier = bprintf("allReduce_%d", counters->allReduce_counter++);
 411
 412   if (!strcmp(process_name, "p0")) {
 413     DEBUG2("%s: %s is the Root", allreduce_identifier, process_name);
 414
 415     msg_comm_t *comms = xbt_new0(msg_comm_t,communicator_size-1);
 416     m_task_t *tasks = xbt_new0(m_task_t,communicator_size-1);
 417     for (i = 1; i < communicator_size; i++) {
 418       sprintf(mailbox, "%s_p%d_p0", allreduce_identifier, i);
 419       comms[i-1] = MSG_task_irecv(&(tasks[i-1]),mailbox);
 420     }
 421     MSG_comm_waitall(comms,communicator_size-1,-1);
 422     for (i = 1; i < communicator_size; i++) {
 423       MSG_comm_destroy(comms[i-1]);
 424       MSG_task_destroy(tasks[i-1]);
 425     }
 426     free(tasks);
 427
 428     comp_task = MSG_task_create("allReduce_comp", comp_size, 0, NULL);
 429     DEBUG1("%s: computing 'reduce_comp'", allreduce_identifier);
 430     MSG_task_execute(comp_task);
 431     MSG_task_destroy(comp_task);
 432     DEBUG1("%s: computed", allreduce_identifier);
 433
 434     for (i = 1; i < communicator_size; i++) {
 435       sprintf(mailbox, "%s_p0_p%d", allreduce_identifier, i);
 436       comms[i-1] =
 437           MSG_task_isend(MSG_task_create(mailbox,0,comm_size,NULL),
 438               mailbox);
 439     }
 440     MSG_comm_waitall(comms,communicator_size-1,-1);
 441     for (i = 1; i < communicator_size; i++)
 442        MSG_comm_destroy(comms[i-1]);
 443     free(comms);
 444
 445     DEBUG2("%s: all messages sent by %s have been received",
 446            allreduce_identifier, process_name);
 447
 448   } else {
 449     DEBUG2("%s: %s sends", allreduce_identifier, process_name);
 450     sprintf(mailbox, "%s_%s_p0", allreduce_identifier, process_name);
 451     DEBUG1("put on %s", mailbox);
 452     MSG_task_send(MSG_task_create(allreduce_identifier, 0, comm_size, NULL),
 453                   mailbox);
 454
 455     sprintf(mailbox, "%s_p0_%s", allreduce_identifier, process_name);
 456     MSG_task_receive(&task, mailbox);
 457     MSG_task_destroy(task);
 458     DEBUG2("%s: %s has received", allreduce_identifier, process_name);
 459   }
 460
 461   VERB2("%s %f", xbt_str_join_array(action, " "), MSG_get_clock() - clock);
 462   free(allreduce_identifier);
 463 }
 464
 465 static void action_comm_size(char*const* action)
 466 {
 467   char *name = NULL;
 468   const char *size = action[2];
 469   double clock = MSG_get_clock();
 470
 471   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 472     name = xbt_str_join_array(action, " ");
 473   communicator_size = parse_double(size);
 474   VERB2("%s %f", name, MSG_get_clock() - clock);
 475   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 476     free(name);
 477 }
 478
 479 static void action_compute(char*const* action)
 480 {
 481   char *name = NULL;
 482   const char *amout = action[2];
 483   m_task_t task = MSG_task_create(name, parse_double(amout), 0, NULL);
 484   double clock = MSG_get_clock();
 485
 486   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 487     name = xbt_str_join_array(action, " ");
 488   DEBUG1("Entering %s", name);
 489   MSG_task_execute(task);
 490   MSG_task_destroy(task);
 491   VERB2("%s %f", name, MSG_get_clock() - clock);
 492   if (XBT_LOG_ISENABLED(actions, xbt_log_priority_verbose))
 493     free(name);
 494 }
 495
 496 static void action_init(char*const* action)
 497 {
 498 #ifdef HAVE_TRACING
 499   TRACE_smpi_init(get_rank(MSG_process_get_name(MSG_process_self())));
 500 #endif
 501   DEBUG0("Initialize the counters");
 502   process_globals_t globals = (process_globals_t) calloc(1, sizeof(s_process_globals_t));
 503   globals->isends = xbt_dynar_new(sizeof(msg_comm_t),NULL);
 504   globals->irecvs = xbt_dynar_new(sizeof(msg_comm_t),NULL);
 505   globals->tasks  = xbt_dynar_new(sizeof(m_task_t),NULL);
 506   MSG_process_set_data(MSG_process_self(),globals);
 507
 508 }
 509
 510 static void action_finalize(char*const* action)
 511 {
 512 #ifdef HAVE_TRACING
 513   TRACE_smpi_finalize(get_rank(MSG_process_get_name(MSG_process_self())));
 514 #endif
 515   process_globals_t globals = (process_globals_t) MSG_process_get_data(MSG_process_self());
 516   if (globals){
 517     xbt_dynar_free_container(&(globals->isends));
 518     xbt_dynar_free_container(&(globals->irecvs));
 519     xbt_dynar_free_container(&(globals->tasks));
 520     free(globals);
 521   }
 522 }
 523
 524 /** Main function */
 525 int main(int argc, char *argv[])
 526 {
 527   MSG_error_t res = MSG_OK;
 528
 529   smx_factory_initializer_to_use = SIMIX_ctx_raw_factory_init;
 530
 531   /* Check the given arguments */
 532   MSG_global_init(&argc, argv);
 533   if (argc < 3) {
 534     printf("Usage: %s platform_file deployment_file [action_files]\n",
 535            argv[0]);
 536     printf
 537         ("example: %s msg_platform.xml msg_deployment.xml actions # if all actions are in the same file\n",
 538          argv[0]);
 539     printf
 540         ("example: %s msg_platform.xml msg_deployment.xml # if actions are in separate files, specified in deployment\n",
 541          argv[0]);
 542     exit(1);
 543   }
 544
 545   /*  Simulation setting */
 546   MSG_create_environment(argv[1]);
 547
 548   /* No need to register functions as in classical MSG programs: the actions get started anyway */
 549   MSG_launch_application(argv[2]);
 550
 551   /*   Action registration */
 552   MSG_action_register("init",     action_init);
 553   MSG_action_register("finalize", action_finalize);
 554   MSG_action_register("comm_size",action_comm_size);
 555   MSG_action_register("send",     action_send);
 556   MSG_action_register("Isend",    action_Isend);
 557   MSG_action_register("recv",     action_recv);
 558   MSG_action_register("Irecv",    action_Irecv);
 559   MSG_action_register("wait",     action_wait);
 560   MSG_action_register("barrier",  action_barrier);
 561   MSG_action_register("bcast",    action_bcast);
 562   MSG_action_register("reduce",   action_reduce);
 563   MSG_action_register("allReduce",action_allReduce);
 564   MSG_action_register("sleep",    action_sleep);
 565   MSG_action_register("compute",  action_compute);
 566
 567
 568   /* Actually do the simulation using MSG_action_trace_run */
 569   res = MSG_action_trace_run(argv[3]);  // it's ok to pass a NULL argument here
 570
 571   INFO1("Simulation time %g", MSG_get_clock());
 572   MSG_clean();
 573
 574   if (res == MSG_OK)
 575     return 0;
 576   else
 577     return 1;
 578 }                               /* end_of_main */