src/mc/explo/CommunicationDeterminismChecker.cpp

   1 /* Copyright (c) 2008-2022. 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 "src/kernel/activity/MailboxImpl.hpp"
   7 #include "src/mc/explo/DFSExplorer.hpp"
   8 #include "src/mc/mc_config.hpp"
   9 #include "src/mc/mc_exit.hpp"
  10 #include "src/mc/mc_forward.hpp"
  11 #include "src/mc/mc_private.hpp"
  12 #include "src/mc/transition/TransitionAny.hpp"
  13 #include "src/mc/transition/TransitionComm.hpp"
  14
  15 #include <cstdint>
  16
  17 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
  18
  19 namespace simgrid::mc {
  20
  21 enum class CallType { NONE, SEND, RECV, WAIT, WAITANY };
  22 enum class CommPatternDifference { NONE, TYPE, MBOX, TAG, SRC_PROC, DST_PROC, DATA_SIZE };
  23 enum class PatternCommunicationType {
  24   none    = 0,
  25   send    = 1,
  26   receive = 2,
  27 };
  28
  29 class PatternCommunication {
  30 public:
  31   int num                       = 0;
  32   uintptr_t comm_addr           = 0;
  33   PatternCommunicationType type = PatternCommunicationType::send;
  34   unsigned long src_proc        = 0;
  35   unsigned long dst_proc        = 0;
  36   unsigned mbox                 = 0;
  37   unsigned size                 = 0;
  38   int tag                       = 0;
  39   int index                     = 0;
  40
  41   PatternCommunication dup() const
  42   {
  43     simgrid::mc::PatternCommunication res;
  44     // num?
  45     res.comm_addr = this->comm_addr;
  46     res.type      = this->type;
  47     res.src_proc  = this->src_proc;
  48     res.dst_proc  = this->dst_proc;
  49     res.mbox      = this->mbox;
  50     res.tag       = this->tag;
  51     res.index     = this->index;
  52     return res;
  53   }
  54 };
  55
  56 struct PatternCommunicationList {
  57   unsigned int index_comm = 0;
  58   std::vector<std::unique_ptr<simgrid::mc::PatternCommunication>> list;
  59 };
  60
  61 /********** Checker extension **********/
  62
  63 struct CommDetExtension {
  64   static simgrid::xbt::Extension<simgrid::mc::Exploration, CommDetExtension> EXTENSION_ID;
  65
  66   std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
  67   std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
  68
  69   bool initial_communications_pattern_done = false;
  70   bool recv_deterministic                  = true;
  71   bool send_deterministic                  = true;
  72   std::string send_diff;
  73   std::string recv_diff;
  74
  75   void exploration_start()
  76   {
  77     const unsigned long maxpid = Api::get().get_maxpid();
  78
  79     initial_communications_pattern.resize(maxpid);
  80     incomplete_communications_pattern.resize(maxpid);
  81   }
  82   void restore_communications_pattern(const simgrid::mc::State* state);
  83   void enforce_deterministic_pattern(aid_t process, const PatternCommunication* comm);
  84   void get_comm_pattern(const Transition* transition);
  85   void complete_comm_pattern(const CommWaitTransition* transition);
  86   void handle_comm_pattern(const Transition* transition);
  87 };
  88 simgrid::xbt::Extension<simgrid::mc::Exploration, CommDetExtension> CommDetExtension::EXTENSION_ID;
  89 /********** State Extension ***********/
  90
  91 class StateCommDet {
  92 public:
  93   std::vector<std::vector<simgrid::mc::PatternCommunication>> incomplete_comm_pattern_;
  94   std::vector<unsigned> communication_indices_;
  95
  96   static simgrid::xbt::Extension<simgrid::mc::State, StateCommDet> EXTENSION_ID;
  97   explicit StateCommDet(CommDetExtension* checker)
  98   {
  99     const unsigned long maxpid = Api::get().get_maxpid();
 100     for (unsigned long i = 0; i < maxpid; i++) {
 101       std::vector<simgrid::mc::PatternCommunication> res;
 102       for (auto const& comm : checker->incomplete_communications_pattern[i])
 103         res.push_back(comm->dup());
 104       incomplete_comm_pattern_.push_back(std::move(res));
 105     }
 106
 107     for (auto const& list_process_comm : checker->initial_communications_pattern)
 108       this->communication_indices_.push_back(list_process_comm.index_comm);
 109   }
 110 };
 111 simgrid::xbt::Extension<simgrid::mc::State, StateCommDet> StateCommDet::EXTENSION_ID;
 112
 113 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
 114                                                                const simgrid::mc::PatternCommunication* comm2)
 115 {
 116   using simgrid::mc::CommPatternDifference;
 117   if (comm1->type != comm2->type)
 118     return CommPatternDifference::TYPE;
 119   if (comm1->mbox != comm2->mbox)
 120     return CommPatternDifference::MBOX;
 121   if (comm1->src_proc != comm2->src_proc)
 122     return CommPatternDifference::SRC_PROC;
 123   if (comm1->dst_proc != comm2->dst_proc)
 124     return CommPatternDifference::DST_PROC;
 125   if (comm1->tag != comm2->tag)
 126     return CommPatternDifference::TAG;
 127   if (comm1->size != comm2->size)
 128     return CommPatternDifference::DATA_SIZE;
 129   return CommPatternDifference::NONE;
 130 }
 131
 132 void CommDetExtension::restore_communications_pattern(const simgrid::mc::State* state)
 133 {
 134   for (size_t i = 0; i < initial_communications_pattern.size(); i++)
 135     initial_communications_pattern[i].index_comm =
 136         state->extension<simgrid::mc::StateCommDet>()->communication_indices_[i];
 137
 138   const unsigned long maxpid = Api::get().get_maxpid();
 139   for (unsigned long i = 0; i < maxpid; i++) {
 140     incomplete_communications_pattern[i].clear();
 141     for (simgrid::mc::PatternCommunication const& comm :
 142          state->extension<simgrid::mc::StateCommDet>()->incomplete_comm_pattern_[i])
 143       incomplete_communications_pattern[i].push_back(new simgrid::mc::PatternCommunication(comm.dup()));
 144   }
 145 }
 146
 147 static std::string print_determinism_result(simgrid::mc::CommPatternDifference diff, aid_t actor,
 148                                             const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
 149 {
 150   std::string type;
 151   std::string res;
 152
 153   if (comm->type == simgrid::mc::PatternCommunicationType::send)
 154     type = xbt::string_printf("The send communications pattern of the actor %ld is different!", actor - 1);
 155   else
 156     type = xbt::string_printf("The recv communications pattern of the actor %ld is different!", actor - 1);
 157
 158   using simgrid::mc::CommPatternDifference;
 159   switch (diff) {
 160     case CommPatternDifference::TYPE:
 161       res = xbt::string_printf("%s Different type for communication #%u", type.c_str(), cursor);
 162       break;
 163     case CommPatternDifference::MBOX:
 164       res = xbt::string_printf("%s Different mailbox for communication #%u", type.c_str(), cursor);
 165       break;
 166     case CommPatternDifference::TAG:
 167       res = xbt::string_printf("%s Different tag for communication #%u", type.c_str(), cursor);
 168       break;
 169     case CommPatternDifference::SRC_PROC:
 170       res = xbt::string_printf("%s Different source for communication #%u", type.c_str(), cursor);
 171       break;
 172     case CommPatternDifference::DST_PROC:
 173       res = xbt::string_printf("%s Different destination for communication #%u", type.c_str(), cursor);
 174       break;
 175     case CommPatternDifference::DATA_SIZE:
 176       res = xbt::string_printf("%s Different data size for communication #%u", type.c_str(), cursor);
 177       break;
 178     default:
 179       res = "";
 180       break;
 181   }
 182
 183   return res;
 184 }
 185
 186 void CommDetExtension::enforce_deterministic_pattern(aid_t actor, const PatternCommunication* comm)
 187 {
 188   PatternCommunicationList& list = initial_communications_pattern[actor];
 189   CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
 190
 191   if (diff != CommPatternDifference::NONE) {
 192     if (comm->type == PatternCommunicationType::send) {
 193       send_deterministic = false;
 194       send_diff          = print_determinism_result(diff, actor, comm, list.index_comm + 1);
 195     } else {
 196       recv_deterministic = false;
 197       if (recv_diff.empty())
 198         recv_diff = print_determinism_result(diff, actor, comm, list.index_comm + 1);
 199     }
 200     if (_sg_mc_send_determinism && not send_deterministic) {
 201       XBT_INFO("*********************************************************");
 202       XBT_INFO("***** Non-send-deterministic communications pattern *****");
 203       XBT_INFO("*********************************************************");
 204       XBT_INFO("%s", send_diff.c_str());
 205       Api::get().get_session().log_state();
 206       Api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 207     } else if (_sg_mc_comms_determinism && (not send_deterministic && not recv_deterministic)) {
 208       XBT_INFO("****************************************************");
 209       XBT_INFO("***** Non-deterministic communications pattern *****");
 210       XBT_INFO("****************************************************");
 211       if (not send_diff.empty())
 212         XBT_INFO("%s", send_diff.c_str());
 213       if (not recv_diff.empty())
 214         XBT_INFO("%s", recv_diff.c_str());
 215       Api::get().get_session().log_state();
 216       Api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 217     }
 218   }
 219 }
 220
 221 /********** Non Static functions ***********/
 222
 223 void CommDetExtension::get_comm_pattern(const Transition* transition)
 224 {
 225   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[transition->aid_];
 226   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[transition->aid_];
 227
 228   auto pattern   = std::make_unique<PatternCommunication>();
 229   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
 230
 231   if (transition->type_ == Transition::Type::COMM_SEND) {
 232     auto* send = static_cast<const CommSendTransition*>(transition);
 233
 234     pattern->type      = PatternCommunicationType::send;
 235     pattern->comm_addr = send->get_comm();
 236     pattern->tag       = send->get_tag();
 237
 238     // FIXME: Detached sends should be enforced when the receive is waited
 239
 240   } else if (transition->type_ == Transition::Type::COMM_RECV) {
 241     auto* recv = static_cast<const CommRecvTransition*>(transition);
 242
 243     pattern->type      = PatternCommunicationType::receive;
 244     pattern->comm_addr = recv->get_comm();
 245     pattern->tag       = recv->get_tag();
 246   }
 247
 248   XBT_DEBUG("Insert incomplete comm pattern %p type:%d for process %ld (comm: %" PRIxPTR ")", pattern.get(),
 249             (int)pattern->type, transition->aid_, pattern->comm_addr);
 250   incomplete_communications_pattern[transition->aid_].push_back(pattern.release());
 251 }
 252
 253 void CommDetExtension::complete_comm_pattern(const CommWaitTransition* transition)
 254 {
 255   /* Complete comm pattern */
 256   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[transition->aid_];
 257   uintptr_t comm_addr                                    = transition->get_comm();
 258   auto current_comm_pattern =
 259       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
 260                    [&comm_addr](const PatternCommunication* comm) { return (comm->comm_addr == comm_addr); });
 261   xbt_assert(current_comm_pattern != std::end(incomplete_pattern), "Corresponding communication not found!");
 262   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
 263
 264   comm_pattern->src_proc = transition->get_sender();
 265   comm_pattern->dst_proc = transition->get_receiver();
 266   comm_pattern->mbox     = transition->get_mailbox();
 267
 268   XBT_DEBUG("Remove incomplete comm pattern for actor %ld at cursor %zd", transition->aid_,
 269             std::distance(begin(incomplete_pattern), current_comm_pattern));
 270   incomplete_pattern.erase(current_comm_pattern);
 271
 272   if (initial_communications_pattern_done) {
 273     /* Evaluate comm determinism */
 274     enforce_deterministic_pattern(transition->aid_, comm_pattern.get());
 275     initial_communications_pattern[transition->aid_].index_comm++;
 276   } else {
 277     /* Store comm pattern */
 278     initial_communications_pattern[transition->aid_].list.push_back(std::move(comm_pattern));
 279   }
 280 }
 281
 282 void CommDetExtension::handle_comm_pattern(const Transition* transition)
 283 {
 284   if (not _sg_mc_comms_determinism && not _sg_mc_send_determinism)
 285     return;
 286
 287   switch (transition->type_) {
 288     case Transition::Type::COMM_SEND:
 289     case Transition::Type::COMM_RECV:
 290       get_comm_pattern(transition);
 291       break;
 292     case Transition::Type::COMM_WAIT:
 293       complete_comm_pattern(static_cast<const CommWaitTransition*>(transition));
 294       break;
 295     case Transition::Type::WAITANY:
 296       // Ignore wait on non-comm
 297       if (auto const* wait = dynamic_cast<const CommWaitTransition*>(
 298               static_cast<const WaitAnyTransition*>(transition)->get_current_transition()))
 299         complete_comm_pattern(wait);
 300       break;
 301     default: /* Ignore unhandled transition types */
 302       break;
 303   }
 304 }
 305
 306 /* FIXME: CommDet probably don't play nicely with stateful exploration
 307
 308       bool all_communications_are_finished = true;
 309       for (size_t current_actor = 1; all_communications_are_finished && current_actor < maxpid; current_actor++) {
 310         if (not extension->incomplete_communications_pattern[current_actor].empty()) {
 311           XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
 312           all_communications_are_finished = false;
 313         }
 314       }
 315  */
 316
 317 Exploration* create_communication_determinism_checker(Session* session)
 318 {
 319   CommDetExtension::EXTENSION_ID = simgrid::mc::Exploration::extension_create<CommDetExtension>();
 320   StateCommDet::EXTENSION_ID     = simgrid::mc::State::extension_create<StateCommDet>();
 321
 322   XBT_DEBUG("********* Start communication determinism verification *********");
 323
 324   auto extension = new CommDetExtension();
 325
 326   DFSExplorer::on_exploration_start([extension]() {
 327     XBT_INFO("Check communication determinism");
 328     extension->exploration_start();
 329   });
 330   DFSExplorer::on_backtracking([extension]() { extension->initial_communications_pattern_done = true; });
 331   DFSExplorer::on_state_creation([extension](State* state) { state->extension_set(new StateCommDet(extension)); });
 332
 333   DFSExplorer::on_restore_system_state([extension](State* state) { extension->restore_communications_pattern(state); });
 334
 335   DFSExplorer::on_restore_initial_state([extension]() {
 336     const unsigned long maxpid = Api::get().get_maxpid();
 337     assert(maxpid == extension->incomplete_communications_pattern.size());
 338     assert(maxpid == extension->initial_communications_pattern.size());
 339     for (unsigned long j = 0; j < maxpid; j++) {
 340       extension->incomplete_communications_pattern[j].clear();
 341       extension->initial_communications_pattern[j].index_comm = 0;
 342     }
 343   });
 344
 345   DFSExplorer::on_transition_replay([extension](Transition* t) { extension->handle_comm_pattern(t); });
 346   DFSExplorer::on_transition_execute([extension](Transition* t) { extension->handle_comm_pattern(t); });
 347
 348   DFSExplorer::on_log_state([extension]() {
 349     if (_sg_mc_comms_determinism) {
 350       if (extension->send_deterministic && not extension->recv_deterministic) {
 351         XBT_INFO("*******************************************************");
 352         XBT_INFO("**** Only-send-deterministic communication pattern ****");
 353         XBT_INFO("*******************************************************");
 354         XBT_INFO("%s", extension->recv_diff.c_str());
 355       }
 356       if (not extension->send_deterministic && extension->recv_deterministic) {
 357         XBT_INFO("*******************************************************");
 358         XBT_INFO("**** Only-recv-deterministic communication pattern ****");
 359         XBT_INFO("*******************************************************");
 360         XBT_INFO("%s", extension->send_diff.c_str());
 361       }
 362     }
 363     XBT_INFO("Send-deterministic : %s", extension->send_deterministic ? "Yes" : "No");
 364     if (_sg_mc_comms_determinism)
 365       XBT_INFO("Recv-deterministic : %s", extension->recv_deterministic ? "Yes" : "No");
 366     delete extension;
 367   });
 368
 369   return new DFSExplorer(session);
 370 }
 371
 372 } // namespace simgrid::mc