src/smpi/internals/smpi_actor.cpp

   1 /* Copyright (c) 2009-2019. The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #include "src/smpi/include/smpi_actor.hpp"
   7 #include "mc/mc.h"
   8 #include "smpi_comm.hpp"
   9 #include "src/mc/mc_replay.hpp"
  10 #include "src/simix/smx_private.hpp"
  11
  12 #if HAVE_PAPI
  13 #include "papi.h"
  14 extern std::string papi_default_config_name;
  15 #endif
  16
  17 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_process, smpi, "Logging specific to SMPI (kernel)");
  18
  19 namespace simgrid {
  20 namespace smpi {
  21
  22 using simgrid::s4u::Actor;
  23 using simgrid::s4u::ActorPtr;
  24
  25 ActorExt::ActorExt(ActorPtr actor, simgrid::s4u::Barrier* finalization_barrier)
  26     : finalization_barrier_(finalization_barrier), actor_(actor)
  27 {
  28   mailbox_         = simgrid::s4u::Mailbox::by_name("SMPI-" + std::to_string(actor_->get_pid()));
  29   mailbox_small_   = simgrid::s4u::Mailbox::by_name("small-" + std::to_string(actor_->get_pid()));
  30   mailboxes_mutex_ = xbt_mutex_init();
  31   timer_           = xbt_os_timer_new();
  32   state_           = SmpiProcessState::UNINITIALIZED;
  33   if (MC_is_active())
  34     MC_ignore_heap(timer_, xbt_os_timer_size());
  35
  36 #if HAVE_PAPI
  37   if (not simgrid::config::get_value<std::string>("smpi/papi-events").empty()) {
  38     // TODO: Implement host/process/thread based counters. This implementation
  39     // just always takes the values passed via "default", like this:
  40     // "default:COUNTER1:COUNTER2:COUNTER3;".
  41     auto it = units2papi_setup.find(papi_default_config_name);
  42     if (it != units2papi_setup.end()) {
  43       papi_event_set_    = it->second.event_set;
  44       papi_counter_data_ = it->second.counter_data;
  45       XBT_DEBUG("Setting PAPI set for process %li", actor->get_pid());
  46     } else {
  47       papi_event_set_ = PAPI_NULL;
  48       XBT_DEBUG("No PAPI set for process %li", actor->get_pid());
  49     }
  50   }
  51 #endif
  52 }
  53
  54 ActorExt::~ActorExt()
  55 {
  56   if (comm_self_ != MPI_COMM_NULL)
  57     simgrid::smpi::Comm::destroy(comm_self_);
  58   if (comm_intra_ != MPI_COMM_NULL)
  59     simgrid::smpi::Comm::destroy(comm_intra_);
  60   xbt_os_timer_free(timer_);
  61   xbt_mutex_destroy(mailboxes_mutex_);
  62 }
  63
  64 void ActorExt::set_data(const char* instance_id)
  65 {
  66   instance_id_                   = std::string(instance_id);
  67   comm_world_                    = smpi_deployment_comm_world(instance_id_);
  68   simgrid::s4u::Barrier* barrier = smpi_deployment_finalization_barrier(instance_id_);
  69   if (barrier != nullptr) // don't overwrite the current one if the instance has none
  70     finalization_barrier_ = barrier;
  71
  72   // set the process attached to the mailbox
  73   mailbox_small_->set_receiver(actor_);
  74   XBT_DEBUG("<%ld> SMPI process has been initialized: %p", actor_->get_pid(), actor_.get());
  75 }
  76
  77 /** @brief Prepares the current process for termination. */
  78 void ActorExt::finalize()
  79 {
  80   state_ = SmpiProcessState::FINALIZED;
  81   XBT_DEBUG("<%ld> Process left the game", actor_->get_pid());
  82
  83   // This leads to an explosion of the search graph which cannot be reduced:
  84   if (MC_is_active() || MC_record_replay_is_active())
  85     return;
  86   // wait for all pending asynchronous comms to finish
  87   finalization_barrier_->wait();
  88 }
  89
  90 /** @brief Check if a process is finalized */
  91 int ActorExt::finalized()
  92 {
  93   return (state_ == SmpiProcessState::FINALIZED);
  94 }
  95
  96 /** @brief Check if a process is partially initialized already */
  97 int ActorExt::initializing()
  98 {
  99   return (state_ == SmpiProcessState::INITIALIZING);
 100 }
 101
 102 /** @brief Check if a process is initialized */
 103 int ActorExt::initialized()
 104 {
 105   // TODO cheinrich: Check if we still need this. This should be a global condition, not for a
 106   // single process ... ?
 107   return (state_ == SmpiProcessState::INITIALIZED);
 108 }
 109
 110 /** @brief Mark a process as initialized (=MPI_Init called) */
 111 void ActorExt::mark_as_initialized()
 112 {
 113   if (state_ != SmpiProcessState::FINALIZED)
 114     state_ = SmpiProcessState::INITIALIZED;
 115 }
 116
 117 void ActorExt::set_replaying(bool value)
 118 {
 119   if (state_ != SmpiProcessState::FINALIZED)
 120     replaying_ = value;
 121 }
 122
 123 bool ActorExt::replaying()
 124 {
 125   return replaying_;
 126 }
 127
 128 ActorPtr ActorExt::get_actor()
 129 {
 130   return actor_;
 131 }
 132
 133 /**
 134  * @brief Returns a structure that stores the location (filename + linenumber) of the last calls to MPI_* functions.
 135  *
 136  * @see smpi_trace_set_call_location
 137  */
 138 smpi_trace_call_location_t* ActorExt::call_location()
 139 {
 140   return &trace_call_loc_;
 141 }
 142
 143 void ActorExt::set_privatized_region(smpi_privatization_region_t region)
 144 {
 145   privatized_region_ = region;
 146 }
 147
 148 smpi_privatization_region_t ActorExt::privatized_region()
 149 {
 150   return privatized_region_;
 151 }
 152
 153 MPI_Comm ActorExt::comm_world()
 154 {
 155   return comm_world_ == nullptr ? MPI_COMM_NULL : *comm_world_;
 156 }
 157
 158 xbt_mutex_t ActorExt::mailboxes_mutex()
 159 {
 160   return mailboxes_mutex_;
 161 }
 162
 163 #if HAVE_PAPI
 164 int ActorExt::papi_event_set()
 165 {
 166   return papi_event_set_;
 167 }
 168
 169 papi_counter_t& ActorExt::papi_counters()
 170 {
 171   return papi_counter_data_;
 172 }
 173 #endif
 174
 175 xbt_os_timer_t ActorExt::timer()
 176 {
 177   return timer_;
 178 }
 179
 180 void ActorExt::simulated_start()
 181 {
 182   simulated_ = SIMIX_get_clock();
 183 }
 184
 185 double ActorExt::simulated_elapsed()
 186 {
 187   return SIMIX_get_clock() - simulated_;
 188 }
 189
 190 MPI_Comm ActorExt::comm_self()
 191 {
 192   if (comm_self_ == MPI_COMM_NULL) {
 193     MPI_Group group = new Group(1);
 194     comm_self_      = new Comm(group, nullptr);
 195     group->set_mapping(actor_, 0);
 196   }
 197   return comm_self_;
 198 }
 199
 200 MPI_Comm ActorExt::comm_intra()
 201 {
 202   return comm_intra_;
 203 }
 204
 205 void ActorExt::set_comm_intra(MPI_Comm comm)
 206 {
 207   comm_intra_ = comm;
 208 }
 209
 210 void ActorExt::set_sampling(int s)
 211 {
 212   sampling_ = s;
 213 }
 214
 215 int ActorExt::sampling()
 216 {
 217   return sampling_;
 218 }
 219
 220 void ActorExt::init()
 221 {
 222   if (smpi_process_count() == 0) {
 223     xbt_die("SimGrid was not initialized properly before entering MPI_Init. Aborting, please check compilation process "
 224             "and use smpirun\n");
 225   }
 226
 227   simgrid::s4u::ActorPtr proc = simgrid::s4u::Actor::self();
 228   // cheinrich: I'm not sure what the impact of the SMPI_switch_data_segment on this call is. I moved
 229   // this up here so that I can set the privatized region before the switch.
 230   ActorExt* process = smpi_process_remote(proc);
 231   // if we are in MPI_Init and argc handling has already been done.
 232   if (process->initialized())
 233     return;
 234
 235   if (smpi_privatize_global_variables == SmpiPrivStrategies::MMAP) {
 236     /* Now using the segment index of this process  */
 237     process->set_privatized_region(smpi_init_global_memory_segment_process());
 238     /* Done at the process's creation */
 239     SMPI_switch_data_segment(proc);
 240   }
 241
 242   const char* instance_id = simgrid::s4u::Actor::self()->get_property("instance_id");
 243   const int rank          = xbt_str_parse_int(simgrid::s4u::Actor::self()->get_property("rank"), "Cannot parse rank");
 244
 245   process->state_ = SmpiProcessState::INITIALIZING;
 246   smpi_deployment_register_process(instance_id, rank, proc);
 247
 248   process->set_data(instance_id);
 249 }
 250
 251 int ActorExt::get_optind()
 252 {
 253   return optind;
 254 }
 255 void ActorExt::set_optind(int new_optind)
 256 {
 257   optind = new_optind;
 258 }
 259
 260 } // namespace smpi
 261 } // namespace simgrid