src/smpi/internals/instr_smpi.cpp

   1 /* Copyright (c) 2010-2022. 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 "private.hpp"
   7 #include <boost/algorithm/string.hpp>
   8 #include <cctype>
   9 #include <cstdarg>
  10 #include <cwchar>
  11 #include <deque>
  12 #include <simgrid/host.h>
  13 #include <simgrid/s4u/Actor.hpp>
  14 #include <simgrid/s4u/Host.hpp>
  15 #include <simgrid/sg_config.hpp>
  16 #include <string>
  17 #include <vector>
  18
  19 #include "src/smpi/include/smpi_actor.hpp"
  20
  21 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_smpi, instr, "Tracing SMPI");
  22
  23 static std::unordered_map<std::string, std::deque<std::string>> keys;
  24
  25 static const std::map<std::string, std::string, std::less<>> smpi_colors = {{"recv", "1 0 0"},
  26                                                                             {"irecv", "1 0.52 0.52"},
  27                                                                             {"send", "0 0 1"},
  28                                                                             {"isend", "0.52 0.52 1"},
  29                                                                             {"sendrecv", "0 1 1"},
  30                                                                             {"wait", "1 1 0"},
  31                                                                             {"waitall", "0.78 0.78 0"},
  32                                                                             {"waitany", "0.78 0.78 0.58"},
  33                                                                             {"test", "0.52 0.52 0"},
  34
  35                                                                             {"allgather", "1 0 0"},
  36                                                                             {"allgatherv", "1 0.52 0.52"},
  37                                                                             {"allreduce", "1 0 1"},
  38                                                                             {"alltoall", "0.52 0 1"},
  39                                                                             {"alltoallv", "0.78 0.52 1"},
  40                                                                             {"barrier", "0 0.39 0.78"},
  41                                                                             {"bcast", "0 0.78 0.39"},
  42                                                                             {"gather", "1 1 0"},
  43                                                                             {"gatherv", "1 1 0.52"},
  44                                                                             {"reduce", "0 1 0"},
  45                                                                             {"reducescatter", "0.52 1 0.52"},
  46                                                                             {"scan", "1 0.58 0.23"},
  47                                                                             {"exscan", "1 0.54 0.25"},
  48                                                                             {"scatterv", "0.52 0 0.52"},
  49                                                                             {"scatter", "1 0.74 0.54"},
  50
  51                                                                             {"computing", "0 1 1"},
  52                                                                             {"sleeping", "0 0.5 0.5"},
  53
  54                                                                             {"init", "0 1 0"},
  55                                                                             {"finalize", "0 1 0"},
  56
  57                                                                             {"put", "0.3 1 0"},
  58                                                                             {"get", "0 1 0.3"},
  59                                                                             {"accumulate", "1 0.3 0"},
  60                                                                             {"rput", "0.3 1 0"},
  61                                                                             {"rget", "0 1 0.3"},
  62                                                                             {"raccumulate", "1 0.3 0"},
  63                                                                             {"compare_and_swap", "0.3 1 0"},
  64                                                                             {"get_accumulate", "0 1 0.3"},
  65                                                                             {"rget_accumulate", "1 0.3 0"},
  66                                                                             {"win_fence", "1 0 0.3"},
  67                                                                             {"win_post", "1 0 0.8"},
  68                                                                             {"win_wait", "1 0.8 0"},
  69                                                                             {"win_start", "0.8 0 1"},
  70                                                                             {"win_complete", "0.8 1 0"},
  71                                                                             {"win_lock", "1 0 0.3"},
  72                                                                             {"win_unlock", "1 0 0.3"},
  73                                                                             {"win_lock_all", "1 0 0.8"},
  74                                                                             {"win_unlock_all", "1 0.8 0"},
  75                                                                             {"win_flush", "1 0 0.3"},
  76                                                                             {"win_flush_local", "1 0 0.8"},
  77                                                                             {"win_flush_all", "1 0.8 0"},
  78                                                                             {"win_flush_local_all", "1 0 0.3"},
  79
  80                                                                             {"file_read", "1 1 0.3"}};
  81
  82 static const char* instr_find_color(const char* c_state)
  83 {
  84   std::string state(c_state);
  85   boost::algorithm::to_lower(state);
  86   if (state.substr(0, 5) == "pmpi_")
  87     state = state.substr(5, std::string::npos); // Remove pmpi_ to allow for exact matches
  88
  89   if (smpi_colors.find(state) != smpi_colors.end()) { // Exact match in the map?
  90     return smpi_colors.find(state)->second.c_str();
  91   }
  92   for (const auto& pair : smpi_colors) { // Is an entry of our map a substring of this state name?
  93     if (std::strstr(state.c_str(), pair.first.c_str()) != nullptr)
  94       return pair.second.c_str();
  95   }
  96
  97   return "0.5 0.5 0.5"; // Just in case we find nothing in the map ...
  98 }
  99
 100 XBT_PRIVATE simgrid::instr::Container* smpi_container(aid_t pid)
 101 {
 102   return simgrid::instr::Container::by_name(std::string("rank-") + std::to_string(pid));
 103 }
 104
 105 static std::string TRACE_smpi_put_key(aid_t src, aid_t dst, int tag, int send)
 106 {
 107   //generate the key
 108   static unsigned long long counter = 0;
 109   counter++;
 110   std::string key =
 111       std::to_string(src) + "_" + std::to_string(dst) + "_" + std::to_string(tag) + "_" + std::to_string(counter);
 112
 113   //push it
 114   std::string aux =
 115       std::to_string(src) + "#" + std::to_string(dst) + "#" + std::to_string(tag) + "#" + std::to_string(send);
 116   keys[aux].push_back(key);
 117
 118   return key;
 119 }
 120
 121 static std::string TRACE_smpi_get_key(aid_t src, aid_t dst, int tag, int send)
 122 {
 123   std::string key;
 124   std::string aux = std::to_string(src) + "#" + std::to_string(dst) + "#" + std::to_string(tag) + "#" +
 125                     std::to_string(send == 1 ? 0 : 1);
 126   auto it = keys.find(aux);
 127   if (it == keys.end()) {
 128     // first posted
 129     key = TRACE_smpi_put_key(src, dst, tag, send);
 130   } else {
 131     key = it->second.front();
 132     it->second.pop_front();
 133     if (it->second.empty())
 134       keys.erase(it);
 135   }
 136   return key;
 137 }
 138
 139 void TRACE_smpi_setup_container(aid_t pid, const_sg_host_t host)
 140 {
 141   auto* parent = simgrid::instr::Container::get_root();
 142   if (TRACE_smpi_is_grouped()) {
 143     parent = simgrid::instr::Container::by_name_or_null(host->get_name());
 144     xbt_assert(parent != nullptr, "Could not find a parent for mpi rank 'rank-%ld' at function %s", pid, __func__);
 145   }
 146   parent->create_child(std::string("rank-") + std::to_string(pid), "MPI"); // This container is of type MPI
 147 }
 148
 149 void TRACE_smpi_init(aid_t pid, const std::string& calling_func)
 150 {
 151   if (not TRACE_smpi_is_enabled())
 152     return;
 153
 154   auto self = simgrid::s4u::Actor::self();
 155
 156   TRACE_smpi_setup_container(pid, sg_host_self());
 157   simgrid::s4u::this_actor::on_exit([self](bool) { smpi_container(self->get_pid())->remove_from_parent(); });
 158
 159   simgrid::instr::StateType* state = smpi_container(pid)->get_state("MPI_STATE");
 160
 161   state->add_entity_value(calling_func, instr_find_color(calling_func.c_str()));
 162   state->push_event(calling_func, new simgrid::instr::NoOpTIData("init"));
 163   state->pop_event();
 164   if (TRACE_smpi_is_computing())
 165     state->add_entity_value("computing", instr_find_color("computing"));
 166   if (TRACE_smpi_is_sleeping())
 167     state->add_entity_value("sleeping", instr_find_color("sleeping"));
 168
 169 #if HAVE_PAPI
 170   const simgrid::instr::Container* container = smpi_container(pid);
 171   papi_counter_t counters = smpi_process()->papi_counters();
 172
 173   for (auto const& it : counters) {
 174     /**
 175      * Check whether this variable already exists or not. Otherwise, it will be created
 176      * multiple times but only the last one would be used...
 177      */
 178     container->get_type()->by_name_or_create(it.first, "");
 179   }
 180 #endif
 181 }
 182
 183 void TRACE_smpi_sleeping_in(aid_t pid, double duration)
 184 {
 185   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_sleeping())
 186     smpi_container(pid)
 187         ->get_state("MPI_STATE")
 188         ->push_event("sleeping", new simgrid::instr::CpuTIData("sleep", duration));
 189 }
 190
 191 void TRACE_smpi_sleeping_out(aid_t pid)
 192 {
 193   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_sleeping())
 194     smpi_container(pid)->get_state("MPI_STATE")->pop_event();
 195 }
 196
 197 void TRACE_smpi_comm_in(aid_t pid, const char* operation, simgrid::instr::TIData* extra)
 198 {
 199   if (not TRACE_smpi_is_enabled()) {
 200     delete extra;
 201     return;
 202   }
 203
 204   simgrid::instr::StateType* state = smpi_container(pid)->get_state("MPI_STATE");
 205   state->add_entity_value(operation, instr_find_color(operation));
 206   state->push_event(operation, extra);
 207 }
 208
 209 void TRACE_smpi_comm_out(aid_t pid)
 210 {
 211   if (TRACE_smpi_is_enabled())
 212     smpi_container(pid)->get_state("MPI_STATE")->pop_event();
 213 }
 214
 215 void TRACE_smpi_send(aid_t rank, aid_t src, aid_t dst, int tag, size_t size)
 216 {
 217   if (not TRACE_smpi_is_enabled())
 218     return;
 219
 220   std::string key = TRACE_smpi_get_key(src, dst, tag, 1);
 221
 222   XBT_DEBUG("Send tracing from %ld to %ld, tag %d, with key %s", src, dst, tag, key.c_str());
 223   simgrid::instr::Container::get_root()->get_link("MPI_LINK")->start_event(smpi_container(rank), "PTP", key, size);
 224 }
 225
 226 void TRACE_smpi_recv(aid_t src, aid_t dst, int tag)
 227 {
 228   if (not TRACE_smpi_is_enabled())
 229     return;
 230
 231   std::string key = TRACE_smpi_get_key(src, dst, tag, 0);
 232
 233   XBT_DEBUG("Recv tracing from %ld to %ld, tag %d, with key %s", src, dst, tag, key.c_str());
 234   simgrid::instr::Container::get_root()->get_link("MPI_LINK")->end_event(smpi_container(dst), "PTP", key);
 235 }