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 std::string smpi_container_key(int rank)
 147 {
 148   return std::string("rank-") + std::to_string(rank);
 149 }
 150
 151 XBT_PRIVATE container_t smpi_container(int rank)
 152 {
 153   return simgrid::instr::Container::byName(smpi_container_key(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   std::string str = smpi_container_key(rank);
 230
 231   container_t father;
 232   if (TRACE_smpi_is_grouped()){
 233     father = simgrid::instr::Container::byNameOrNull(host->get_name());
 234   }else{
 235     father = simgrid::instr::Container::getRoot();
 236   }
 237   xbt_assert(father != nullptr, "Could not find a parent for mpi rank %s at function %s", str.c_str(), __func__);
 238   father->createChild(str, "MPI"); // This container is of type MPI
 239 }
 240
 241 void TRACE_smpi_init(int rank)
 242 {
 243   if (not TRACE_smpi_is_enabled())
 244     return;
 245
 246   TRACE_smpi_setup_container(rank, sg_host_self());
 247 #if HAVE_PAPI
 248   container_t container   = simgrid::instr::Container::byName(str);
 249   papi_counter_t counters = smpi_process()->papi_counters();
 250
 251   for (auto const& it : counters) {
 252     /**
 253      * Check whether this variable already exists or not. Otherwise, it will be created
 254      * multiple times but only the last one would be used...
 255      */
 256     if (s_type::getOrNull(it.first.c_str(), container->type_) == nullptr) {
 257       Type::variableNew(it.first.c_str(), "", container->type_);
 258     }
 259   }
 260 #endif
 261 }
 262
 263 void TRACE_smpi_finalize(int rank)
 264 {
 265   if (not TRACE_smpi_is_enabled())
 266     return;
 267
 268   smpi_container(rank)->removeFromParent();
 269 }
 270
 271 void TRACE_smpi_computing_init(int rank)
 272 {
 273  //first use, initialize the color in the trace
 274  if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing())
 275    smpi_container(rank)->getState("MPI_STATE")->addEntityValue("computing", instr_find_color("computing"));
 276 }
 277
 278 void TRACE_smpi_computing_in(int rank, double amount)
 279 {
 280   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing())
 281     smpi_container(rank)
 282         ->getState("MPI_STATE")
 283         ->pushEvent("computing", new simgrid::instr::CpuTIData("compute", amount));
 284 }
 285
 286 void TRACE_smpi_computing_out(int rank)
 287 {
 288   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing())
 289     smpi_container(rank)->getState("MPI_STATE")->popEvent();
 290 }
 291
 292 void TRACE_smpi_sleeping_in(int rank, double duration)
 293 {
 294   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_sleeping())
 295     smpi_container(rank)
 296         ->getState("MPI_STATE")
 297         ->pushEvent("sleeping", new simgrid::instr::CpuTIData("sleep", duration));
 298 }
 299
 300 void TRACE_smpi_sleeping_out(int rank)
 301 {
 302   if (TRACE_smpi_is_enabled() && not TRACE_smpi_is_sleeping())
 303     smpi_container(rank)->getState("MPI_STATE")->popEvent();
 304 }
 305
 306 void TRACE_smpi_testing_in(int rank)
 307 {
 308   //do not forget to set the color first, otherwise this will explode
 309   if (not TRACE_smpi_is_enabled())
 310     return;
 311
 312   simgrid::instr::StateType* state = smpi_container(rank)->getState("MPI_STATE");
 313   state->addEntityValue("test");
 314   state->pushEvent("test", new simgrid::instr::NoOpTIData("test"));
 315 }
 316
 317 void TRACE_smpi_testing_out(int rank)
 318 {
 319   if (TRACE_smpi_is_enabled())
 320     smpi_container(rank)->getState("MPI_STATE")->popEvent();
 321 }
 322
 323 void TRACE_smpi_comm_in(int rank, const char* operation, simgrid::instr::TIData* extra)
 324 {
 325   if (not TRACE_smpi_is_enabled()) {
 326     delete extra;
 327     return;
 328   }
 329
 330   simgrid::instr::StateType* state = smpi_container(rank)->getState("MPI_STATE");
 331   state->addEntityValue(operation, instr_find_color(operation));
 332   state->pushEvent(operation, extra);
 333 }
 334
 335 void TRACE_smpi_comm_out(int rank)
 336 {
 337   if (TRACE_smpi_is_enabled())
 338     smpi_container(rank)->getState("MPI_STATE")->popEvent();
 339 }
 340
 341 void TRACE_smpi_send(int rank, int src, int dst, int tag, int size)
 342 {
 343   if (not TRACE_smpi_is_enabled())
 344     return;
 345
 346   std::string key = TRACE_smpi_get_key(src, dst, tag, 1);
 347
 348   XBT_DEBUG("Send tracing from %d to %d, tag %d, with key %s", src, dst, tag, key.c_str());
 349   simgrid::instr::Container::getRoot()->getLink("MPI_LINK")->startEvent(smpi_container(rank), "PTP", key, size);
 350 }
 351
 352 void TRACE_smpi_recv(int src, int dst, int tag)
 353 {
 354   if (not TRACE_smpi_is_enabled())
 355     return;
 356
 357   std::string key = TRACE_smpi_get_key(src, dst, tag, 0);
 358
 359   XBT_DEBUG("Recv tracing from %d to %d, tag %d, with key %s", src, dst, tag, key.c_str());
 360   simgrid::instr::Container::getRoot()->getLink("MPI_LINK")->endEvent(smpi_container(dst), "PTP", key);
 361 }