src/smpi/internals/instr_smpi.cpp

   1 /* Copyright (c) 2010-2018. 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 "private.hpp"
   8 #include <boost/algorithm/string.hpp>
   9 #include <simgrid/s4u/Actor.hpp>
  10 #include <cctype>
  11 #include <cstdarg>
  12 #include <cwchar>
  13 #include <deque>
  14 #include <simgrid/sg_config.hpp>
  15 #include <simgrid/s4u/Host.hpp>
  16 #include <string>
  17 #include <vector>
  18
  19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_smpi, instr, "Tracing SMPI");
  20
  21 static std::unordered_map<std::string, std::deque<std::string>*> keys;
  22
  23 static const char* smpi_colors[] = {"recv",
  24                                     "1 0 0",
  25                                     "irecv",
  26                                     "1 0.52 0.52",
  27                                     "send",
  28                                     "0 0 1",
  29                                     "isend",
  30                                     "0.52 0.52 1",
  31                                     "sendrecv",
  32                                     "0 1 1",
  33                                     "wait",
  34                                     "1 1 0",
  35                                     "waitall",
  36                                     "0.78 0.78 0",
  37                                     "waitany",
  38                                     "0.78 0.78 0.58",
  39                                     "test",
  40                                     "0.52 0.52 0",
  41
  42                                     "allgather",
  43                                     "1 0 0",
  44                                     "allgatherv",
  45                                     "1 0.52 0.52",
  46                                     "allreduce",
  47                                     "1 0 1",
  48                                     "alltoall",
  49                                     "0.52 0 1",
  50                                     "alltoallv",
  51                                     "0.78 0.52 1",
  52                                     "barrier",
  53                                     "0 0.78 0.78",
  54                                     "bcast",
  55                                     "0 0.78 0.39",
  56                                     "gather",
  57                                     "1 1 0",
  58                                     "gatherv",
  59                                     "1 1 0.52",
  60                                     "reduce",
  61                                     "0 1 0",
  62                                     "reducescatter",
  63                                     "0.52 1 0.52",
  64                                     "scan",
  65                                     "1 0.58 0.23",
  66                                     "exscan",
  67                                     "1 0.54 0.25",
  68                                     "scatterv",
  69                                     "0.52 0 0.52",
  70                                     "scatter",
  71                                     "1 0.74 0.54",
  72
  73                                     "computing",
  74                                     "0 1 1",
  75                                     "sleeping",
  76                                     "0 0.5 0.5",
  77
  78                                     "init",
  79                                     "0 1 0",
  80                                     "finalize",
  81                                     "0 1 0",
  82
  83                                     "put",
  84                                     "0.3 1 0",
  85                                     "get",
  86                                     "0 1 0.3",
  87                                     "accumulate",
  88                                     "1 0.3 0",
  89                                     "migration",
  90                                     "0.2 0.5 0.2",
  91                                     "rput",
  92                                     "0.3 1 0",
  93                                     "rget",
  94                                     "0 1 0.3",
  95                                     "raccumulate",
  96                                     "1 0.3 0",
  97                                     "compare_and_swap",
  98                                     "0.3 1 0",
  99                                     "get_accumulate",
 100                                     "0 1 0.3",
 101                                     "rget_accumulate",
 102                                     "1 0.3 0",
 103                                     "win_fence",
 104                                     "1 0 0.3",
 105                                     "win_post",
 106                                     "1 0 0.8",
 107                                     "win_wait",
 108                                     "1 0.8 0",
 109                                     "win_start",
 110                                     "0.8 0 1",
 111                                     "win_complete",
 112                                     "0.8 1 0",
 113                                     "win_lock",
 114                                     "1 0 0.3",
 115                                     "win_unlock",
 116                                     "1 0 0.3",
 117                                     "win_lock_all",
 118                                     "1 0 0.8",
 119                                     "win_unlock_all",
 120                                     "1 0.8 0",
 121                                     "win_flush",
 122                                     "1 0 0.3",
 123                                     "win_flush_local",
 124                                     "1 0 0.8",
 125                                     "win_flush_all",
 126                                     "1 0.8 0",
 127                                     "win_flush_local_all",
 128                                     "1 0 0.3",
 129                                     "",
 130                                     ""};
 131
 132 static const char* instr_find_color(const char* state)
 133 {
 134   std::string target = std::string(state);
 135   boost::algorithm::to_lower(target);
 136   const char* ret     = nullptr;
 137   unsigned int i      = 0;
 138   const char* current = smpi_colors[i];
 139   while (current != nullptr) {
 140     if (target == current                          // exact match
 141         || strstr(target.c_str(), current) != 0) { // as substring
 142       ret = smpi_colors[i + 1];
 143       break;
 144     }
 145     i+=2;
 146     current = smpi_colors[i];
 147   }
 148   return ret;
 149 }
 150
 151 XBT_PRIVATE std::string smpi_container_key(int rank)
 152 {
 153   return std::string("rank-") + std::to_string(rank);
 154 }
 155
 156 XBT_PRIVATE container_t smpi_container(int rank)
 157 {
 158   return simgrid::instr::Container::byName(smpi_container_key(rank));
 159 }
 160
 161 static std::string TRACE_smpi_put_key(int src, int dst, int tag, int send)
 162 {
 163   // get the deque for src#dst
 164   std::string aux =
 165       std::to_string(src) + "#" + std::to_string(dst) + "#" + std::to_string(tag) + "#" + std::to_string(send);
 166   auto it = keys.find(aux);
 167   std::deque<std::string>* d;
 168
 169   if (it == keys.end()) {
 170     d         = new std::deque<std::string>;
 171     keys[aux] = d;
 172   } else
 173     d = it->second;
 174
 175   //generate the key
 176   static unsigned long long counter = 0;
 177   counter++;
 178   std::string key =
 179       std::to_string(src) + "_" + std::to_string(dst) + "_" + std::to_string(tag) + "_" + std::to_string(counter);
 180
 181   //push it
 182   d->push_back(key);
 183
 184   return key;
 185 }
 186
 187 static std::string TRACE_smpi_get_key(int src, int dst, int tag, int send)
 188 {
 189   std::string key;
 190   std::string aux = std::to_string(src) + "#" + std::to_string(dst) + "#" + std::to_string(tag) + "#" +
 191                     std::to_string(send == 1 ? 0 : 1);
 192   auto it = keys.find(aux);
 193   if (it == keys.end()) {
 194     // first posted
 195     key = TRACE_smpi_put_key(src, dst, tag, send);
 196   } else {
 197     key = it->second->front();
 198     it->second->pop_front();
 199   }
 200   return key;
 201 }
 202
 203 static std::unordered_map<smx_actor_t, std::string> process_category;
 204
 205 void TRACE_internal_smpi_set_category (const char *category)
 206 {
 207   if (not TRACE_smpi_is_enabled())
 208     return;
 209
 210   //declare category
 211   TRACE_category (category);
 212
 213   if (category != nullptr)
 214     process_category[SIMIX_process_self()] = category;
 215 }
 216
 217 const char *TRACE_internal_smpi_get_category ()
 218 {
 219   if (not TRACE_smpi_is_enabled())
 220     return nullptr;
 221
 222   auto it = process_category.find(SIMIX_process_self());
 223   return (it == process_category.end()) ? nullptr : it->second.c_str();
 224 }
 225
 226 void TRACE_smpi_release()
 227 {
 228   for (auto const& elm : keys)
 229     delete elm.second;
 230 }
 231
 232 void TRACE_smpi_setup_container(int rank, sg_host_t host)
 233 {
 234   std::string str = smpi_container_key(rank);
 235
 236   container_t father;
 237   if (TRACE_smpi_is_grouped()){
 238     father = simgrid::instr::Container::byNameOrNull(host->get_name());
 239   }else{
 240     father = simgrid::instr::Container::getRoot();
 241   }
 242   xbt_assert(father != nullptr, "Could not find a parent for mpi rank %s at function %s", str.c_str(), __func__);
 243   father->createChild(str, "MPI"); // This container is of type MPI
 244 }
 245
 246 void TRACE_smpi_init(int rank)
 247 {
 248   if (not TRACE_smpi_is_enabled())
 249     return;
 250
 251   TRACE_smpi_setup_container(rank, sg_host_self());
 252 #if HAVE_PAPI
 253   container_t container   = simgrid::instr::Container::byName(str);
 254   papi_counter_t counters = smpi_process()->papi_counters();
 255
 256   for (auto const& it : counters) {
 257     /**
 258      * Check whether this variable already exists or not. Otherwise, it will be created
 259      * multiple times but only the last one would be used...
 260      */
 261     if (s_type::getOrNull(it.first.c_str(), container->type_) == nullptr) {
 262       Type::variableNew(it.first.c_str(), "", container->type_);
 263     }
 264   }
 265 #endif
 266 }
 267
 268 void TRACE_smpi_finalize(int rank)
 269 {
 270   if (not TRACE_smpi_is_enabled())
 271     return;
 272
 273   smpi_container(rank)->removeFromParent();
 274 }
 275
 276 void TRACE_smpi_computing_init(int rank)
 277 {
 278  //first use, initialize the color in the trace
 279  if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing())
 280    smpi_container(rank)->getState("MPI_STATE")->addEntityValue("computing", instr_find_color("computing"));
 281 }
 282
 283 void TRACE_smpi_computing_in(int rank, double amount)
 284 {
 285   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing())
 286     smpi_container(rank)
 287         ->getState("MPI_STATE")
 288         ->pushEvent("computing", new simgrid::instr::CpuTIData("compute", amount));
 289 }
 290
 291 void TRACE_smpi_computing_out(int rank)
 292 {
 293   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing())
 294     smpi_container(rank)->getState("MPI_STATE")->popEvent();
 295 }
 296
 297 void TRACE_smpi_sleeping_in(int rank, double duration)
 298 {
 299   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_sleeping())
 300     smpi_container(rank)
 301         ->getState("MPI_STATE")
 302         ->pushEvent("sleeping", new simgrid::instr::CpuTIData("sleep", duration));
 303 }
 304
 305 void TRACE_smpi_sleeping_out(int rank)
 306 {
 307   if (TRACE_smpi_is_enabled() && not TRACE_smpi_is_sleeping())
 308     smpi_container(rank)->getState("MPI_STATE")->popEvent();
 309 }
 310
 311 void TRACE_smpi_testing_in(int rank)
 312 {
 313   //do not forget to set the color first, otherwise this will explode
 314   if (not TRACE_smpi_is_enabled())
 315     return;
 316
 317   simgrid::instr::StateType* state = smpi_container(rank)->getState("MPI_STATE");
 318   state->addEntityValue("test");
 319   state->pushEvent("test", new simgrid::instr::NoOpTIData("test"));
 320 }
 321
 322 void TRACE_smpi_testing_out(int rank)
 323 {
 324   if (TRACE_smpi_is_enabled())
 325     smpi_container(rank)->getState("MPI_STATE")->popEvent();
 326 }
 327
 328 void TRACE_smpi_comm_in(int rank, const char* operation, simgrid::instr::TIData* extra)
 329 {
 330   if (not TRACE_smpi_is_enabled()) {
 331     delete extra;
 332     return;
 333   }
 334
 335   simgrid::instr::StateType* state = smpi_container(rank)->getState("MPI_STATE");
 336   state->addEntityValue(operation, instr_find_color(operation));
 337   state->pushEvent(operation, extra);
 338 }
 339
 340 void TRACE_smpi_comm_out(int rank)
 341 {
 342   if (TRACE_smpi_is_enabled())
 343     smpi_container(rank)->getState("MPI_STATE")->popEvent();
 344 }
 345
 346 void TRACE_smpi_send(int rank, int src, int dst, int tag, int size)
 347 {
 348   if (not TRACE_smpi_is_enabled())
 349     return;
 350
 351   std::string key = TRACE_smpi_get_key(src, dst, tag, 1);
 352
 353   XBT_DEBUG("Send tracing from %d to %d, tag %d, with key %s", src, dst, tag, key.c_str());
 354   simgrid::instr::Container::getRoot()->getLink("MPI_LINK")->startEvent(smpi_container(rank), "PTP", key, size);
 355 }
 356
 357 void TRACE_smpi_recv(int src, int dst, int tag)
 358 {
 359   if (not TRACE_smpi_is_enabled())
 360     return;
 361
 362   std::string key = TRACE_smpi_get_key(src, dst, tag, 0);
 363
 364   XBT_DEBUG("Recv tracing from %d to %d, tag %d, with key %s", src, dst, tag, key.c_str());
 365   simgrid::instr::Container::getRoot()->getLink("MPI_LINK")->endEvent(smpi_container(dst), "PTP", key);
 366 }
 367
 368 /**************** Functions to trace the migration of tasks. *****************/
 369 void TRACE_smpi_send_process_data_in(int rank)
 370 {
 371   if (!TRACE_smpi_is_enabled()) return;
 372
 373   smpi_container(rank)->getState("MIGRATE_STATE")->addEntityValue("migration", instr_find_color("migration"));
 374   smpi_container(rank)->getState("MIGRATE_STATE")->pushEvent("migration");
 375 }
 376
 377 void TRACE_smpi_send_process_data_out(int rank)
 378 {
 379   if (!TRACE_smpi_is_enabled()) return;
 380
 381   /* Clean the process state. */
 382   smpi_container(rank)->getState("MIGRATE_STATE")->popEvent();
 383 }
 384
 385 void TRACE_smpi_process_change_host(int rank, sg_host_t new_host)
 386 {
 387   if (!TRACE_smpi_is_enabled()) return;
 388
 389   /** The key is (most likely) used to match the events in the trace */
 390   static long long int counter = 0;
 391   std::string key              = std::to_string(counter);
 392   counter++;
 393
 394   // start link (= tell the trace that this rank moves from A to B)
 395   container_t cont = smpi_container(rank);
 396   simgrid::instr::Container::getRoot()->getLink("MIGRATE_LINK")->startEvent(cont, "M", key);
 397
 398   // Destroy container of this rank on this host
 399   cont->removeFromParent();
 400
 401   // Setup container on new host
 402   TRACE_smpi_setup_container(rank, new_host);
 403
 404   // end link
 405   cont = smpi_container(rank); // This points to the newly created container
 406   simgrid::instr::Container::getRoot()->getLink("MIGRATE_LINK")->endEvent(cont, "M", key);
 407 }