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 container_t smpi_container(int rank)
 152 {
 153   return simgrid::instr::Container::by_name(std::string("rank-") + std::to_string(rank));
 154 }
 155
 156 static std::string TRACE_smpi_put_key(int src, int dst, int tag, int send)
 157 {
 158   // get the deque for src#dst
 159   std::string aux =
 160       std::to_string(src) + "#" + std::to_string(dst) + "#" + std::to_string(tag) + "#" + std::to_string(send);
 161   auto it = keys.find(aux);
 162   std::deque<std::string>* d;
 163
 164   if (it == keys.end()) {
 165     d         = new std::deque<std::string>;
 166     keys[aux] = d;
 167   } else
 168     d = it->second;
 169
 170   //generate the key
 171   static unsigned long long counter = 0;
 172   counter++;
 173   std::string key =
 174       std::to_string(src) + "_" + std::to_string(dst) + "_" + std::to_string(tag) + "_" + std::to_string(counter);
 175
 176   //push it
 177   d->push_back(key);
 178
 179   return key;
 180 }
 181
 182 static std::string TRACE_smpi_get_key(int src, int dst, int tag, int send)
 183 {
 184   std::string key;
 185   std::string aux = std::to_string(src) + "#" + std::to_string(dst) + "#" + std::to_string(tag) + "#" +
 186                     std::to_string(send == 1 ? 0 : 1);
 187   auto it = keys.find(aux);
 188   if (it == keys.end()) {
 189     // first posted
 190     key = TRACE_smpi_put_key(src, dst, tag, send);
 191   } else {
 192     key = it->second->front();
 193     it->second->pop_front();
 194   }
 195   return key;
 196 }
 197
 198 static std::unordered_map<smx_actor_t, std::string> process_category;
 199
 200 void TRACE_internal_smpi_set_category (const char *category)
 201 {
 202   if (not TRACE_smpi_is_enabled())
 203     return;
 204
 205   //declare category
 206   TRACE_category (category);
 207
 208   if (category != nullptr)
 209     process_category[SIMIX_process_self()] = category;
 210 }
 211
 212 const char *TRACE_internal_smpi_get_category ()
 213 {
 214   if (not TRACE_smpi_is_enabled())
 215     return nullptr;
 216
 217   auto it = process_category.find(SIMIX_process_self());
 218   return (it == process_category.end()) ? nullptr : it->second.c_str();
 219 }
 220
 221 void TRACE_smpi_release()
 222 {
 223   for (auto const& elm : keys)
 224     delete elm.second;
 225 }
 226
 227 void TRACE_smpi_setup_container(int rank, sg_host_t host)
 228 {
 229   container_t father = simgrid::instr::Container::get_root();
 230   if (TRACE_smpi_is_grouped()) {
 231     father = simgrid::instr::Container::by_name_or_null(host->get_name());
 232     xbt_assert(father != nullptr, "Could not find a parent for mpi rank 'rank-%d' at function %s", rank, __func__);
 233   }
 234   father->create_child(std::string("rank-") + std::to_string(rank), "MPI"); // This container is of type MPI
 235 }
 236
 237 void TRACE_smpi_init(int rank)
 238 {
 239   if (not TRACE_smpi_is_enabled())
 240     return;
 241
 242   TRACE_smpi_setup_container(rank, sg_host_self());
 243 #if HAVE_PAPI
 244   container_t container   = simgrid::instr::Container::by_name(str);
 245   papi_counter_t counters = smpi_process()->papi_counters();
 246
 247   for (auto const& it : counters) {
 248     /**
 249      * Check whether this variable already exists or not. Otherwise, it will be created
 250      * multiple times but only the last one would be used...
 251      */
 252     if (s_type::getOrNull(it.first.c_str(), container->type_) == nullptr) {
 253       Type::variableNew(it.first.c_str(), "", container->type_);
 254     }
 255   }
 256 #endif
 257 }
 258
 259 void TRACE_smpi_finalize(int rank)
 260 {
 261   if (not TRACE_smpi_is_enabled())
 262     return;
 263
 264   smpi_container(rank)->remove_from_parent();
 265 }
 266
 267 void TRACE_smpi_computing_init(int rank)
 268 {
 269  //first use, initialize the color in the trace
 270  if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing())
 271    smpi_container(rank)->get_state("MPI_STATE")->add_entity_value("computing", instr_find_color("computing"));
 272 }
 273
 274 void TRACE_smpi_computing_in(int rank, double amount)
 275 {
 276   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing())
 277     smpi_container(rank)
 278         ->get_state("MPI_STATE")
 279         ->push_event("computing", new simgrid::instr::CpuTIData("compute", amount));
 280 }
 281
 282 void TRACE_smpi_computing_out(int rank)
 283 {
 284   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing())
 285     smpi_container(rank)->get_state("MPI_STATE")->pop_event();
 286 }
 287
 288 void TRACE_smpi_sleeping_in(int rank, double duration)
 289 {
 290   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_sleeping())
 291     smpi_container(rank)
 292         ->get_state("MPI_STATE")
 293         ->push_event("sleeping", new simgrid::instr::CpuTIData("sleep", duration));
 294 }
 295
 296 void TRACE_smpi_sleeping_out(int rank)
 297 {
 298   if (TRACE_smpi_is_enabled() && not TRACE_smpi_is_sleeping())
 299     smpi_container(rank)->get_state("MPI_STATE")->pop_event();
 300 }
 301
 302 void TRACE_smpi_testing_in(int rank)
 303 {
 304   //do not forget to set the color first, otherwise this will explode
 305   if (not TRACE_smpi_is_enabled())
 306     return;
 307
 308   simgrid::instr::StateType* state = smpi_container(rank)->get_state("MPI_STATE");
 309   state->add_entity_value("test");
 310   state->push_event("test", new simgrid::instr::NoOpTIData("test"));
 311 }
 312
 313 void TRACE_smpi_testing_out(int rank)
 314 {
 315   if (TRACE_smpi_is_enabled())
 316     smpi_container(rank)->get_state("MPI_STATE")->pop_event();
 317 }
 318
 319 void TRACE_smpi_comm_in(int rank, const char* operation, simgrid::instr::TIData* extra)
 320 {
 321   if (not TRACE_smpi_is_enabled()) {
 322     delete extra;
 323     return;
 324   }
 325
 326   simgrid::instr::StateType* state = smpi_container(rank)->get_state("MPI_STATE");
 327   state->add_entity_value(operation, instr_find_color(operation));
 328   state->push_event(operation, extra);
 329 }
 330
 331 void TRACE_smpi_comm_out(int rank)
 332 {
 333   if (TRACE_smpi_is_enabled())
 334     smpi_container(rank)->get_state("MPI_STATE")->pop_event();
 335 }
 336
 337 void TRACE_smpi_send(int rank, int src, int dst, int tag, int size)
 338 {
 339   if (not TRACE_smpi_is_enabled())
 340     return;
 341
 342   std::string key = TRACE_smpi_get_key(src, dst, tag, 1);
 343
 344   XBT_DEBUG("Send tracing from %d to %d, tag %d, with key %s", src, dst, tag, key.c_str());
 345   simgrid::instr::Container::get_root()->get_link("MPI_LINK")->start_event(smpi_container(rank), "PTP", key, size);
 346 }
 347
 348 void TRACE_smpi_recv(int src, int dst, int tag)
 349 {
 350   if (not TRACE_smpi_is_enabled())
 351     return;
 352
 353   std::string key = TRACE_smpi_get_key(src, dst, tag, 0);
 354
 355   XBT_DEBUG("Recv tracing from %d to %d, tag %d, with key %s", src, dst, tag, key.c_str());
 356   simgrid::instr::Container::get_root()->get_link("MPI_LINK")->end_event(smpi_container(dst), "PTP", key);
 357 }
 358
 359 /**************** Functions to trace the migration of tasks. *****************/
 360 void TRACE_smpi_send_process_data_in(int rank)
 361 {
 362   if (!TRACE_smpi_is_enabled()) return;
 363
 364   smpi_container(rank)->get_state("MIGRATE_STATE")->add_entity_value("migration", instr_find_color("migration"));
 365   smpi_container(rank)->get_state("MIGRATE_STATE")->push_event("migration");
 366 }
 367
 368 void TRACE_smpi_send_process_data_out(int rank)
 369 {
 370   if (!TRACE_smpi_is_enabled()) return;
 371
 372   /* Clean the process state. */
 373   smpi_container(rank)->get_state("MIGRATE_STATE")->pop_event();
 374 }
 375
 376 void TRACE_smpi_process_change_host(int rank, sg_host_t new_host)
 377 {
 378   if (!TRACE_smpi_is_enabled()) return;
 379
 380   /** The key is (most likely) used to match the events in the trace */
 381   static long long int counter = 0;
 382   std::string key              = std::to_string(counter);
 383   counter++;
 384
 385   // start link (= tell the trace that this rank moves from A to B)
 386   container_t cont = smpi_container(rank);
 387   simgrid::instr::Container::get_root()->get_link("MIGRATE_LINK")->start_event(cont, "M", key);
 388
 389   // Destroy container of this rank on this host
 390   cont->remove_from_parent();
 391
 392   // Setup container on new host
 393   TRACE_smpi_setup_container(rank, new_host);
 394
 395   // end link
 396   cont = smpi_container(rank); // This points to the newly created container
 397   simgrid::instr::Container::get_root()->get_link("MIGRATE_LINK")->end_event(cont, "M", key);
 398 }