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