Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Add the MPI tag to the Send/Recv Transitions
[simgrid.git] / src / mc / checker / 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/mc/checker/CommunicationDeterminismChecker.hpp"
7 #include "src/kernel/activity/MailboxImpl.hpp"
8 #include "src/mc/Session.hpp"
9 #include "src/mc/api/TransitionComm.hpp"
10 #include "src/mc/mc_config.hpp"
11 #include "src/mc/mc_exit.hpp"
12 #include "src/mc/mc_private.hpp"
13
14 #include <cstdint>
15
16 using api = simgrid::mc::Api;
17
18 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
19
20 namespace simgrid {
21 namespace mc {
22
23 enum class CallType { NONE, SEND, RECV, WAIT, WAITANY };
24 enum class CommPatternDifference { NONE, TYPE, MBOX, TAG, SRC_PROC, DST_PROC, DATA_SIZE };
25 enum class PatternCommunicationType {
26   none    = 0,
27   send    = 1,
28   receive = 2,
29 };
30
31 class PatternCommunication {
32 public:
33   int num = 0;
34   uintptr_t comm_addr           = 0;
35   PatternCommunicationType type = PatternCommunicationType::send;
36   unsigned long src_proc        = 0;
37   unsigned long dst_proc        = 0;
38   unsigned mbox                 = 0;
39   unsigned size                 = 0;
40   int tag   = 0;
41   int index = 0;
42
43   PatternCommunication dup() const
44   {
45     simgrid::mc::PatternCommunication res;
46     // num?
47     res.comm_addr = this->comm_addr;
48     res.type      = this->type;
49     res.src_proc  = this->src_proc;
50     res.dst_proc = this->dst_proc;
51     res.mbox     = this->mbox;
52     res.tag       = this->tag;
53     res.index = this->index;
54     return res;
55   }
56 };
57
58 struct PatternCommunicationList {
59   unsigned int index_comm = 0;
60   std::vector<std::unique_ptr<simgrid::mc::PatternCommunication>> list;
61 };
62
63 /********** Checker extension **********/
64
65 struct CommDetExtension {
66   static simgrid::xbt::Extension<simgrid::mc::Checker, CommDetExtension> EXTENSION_ID;
67
68   CommDetExtension()
69   {
70     const unsigned long maxpid = api::get().get_maxpid();
71
72     initial_communications_pattern.resize(maxpid);
73     incomplete_communications_pattern.resize(maxpid);
74   }
75
76   std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
77   std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
78
79   bool initial_communications_pattern_done = false;
80   bool recv_deterministic                  = true;
81   bool send_deterministic                  = true;
82   std::string send_diff;
83   std::string recv_diff;
84
85   void restore_communications_pattern(const simgrid::mc::State* state);
86   void deterministic_comm_pattern(aid_t process, const PatternCommunication* comm, bool backtracking);
87   void get_comm_pattern(const Transition* transition, bool backtracking);
88   void complete_comm_pattern(const CommWaitTransition* transition, bool backtracking);
89   void handle_comm_pattern(const Transition* transition, bool backtracking);
90 };
91 simgrid::xbt::Extension<simgrid::mc::Checker, CommDetExtension> CommDetExtension::EXTENSION_ID;
92 /********** State Extension ***********/
93
94 class StateCommDet {
95   CommDetExtension* checker_;
96
97 public:
98   std::vector<std::vector<simgrid::mc::PatternCommunication>> incomplete_comm_pattern_;
99   std::vector<unsigned> communication_indices_;
100
101   static simgrid::xbt::Extension<simgrid::mc::State, StateCommDet> EXTENSION_ID;
102   explicit StateCommDet(CommDetExtension* checker) : checker_(checker)
103   {
104     copy_incomplete_comm_pattern();
105     copy_index_comm_pattern();
106   }
107
108   void copy_incomplete_comm_pattern()
109   {
110     incomplete_comm_pattern_.clear();
111     const unsigned long maxpid = api::get().get_maxpid();
112     for (unsigned long i = 0; i < maxpid; i++) {
113       std::vector<simgrid::mc::PatternCommunication> res;
114       for (auto const& comm : checker_->incomplete_communications_pattern[i])
115         res.push_back(comm->dup());
116       incomplete_comm_pattern_.push_back(std::move(res));
117     }
118   }
119
120   void copy_index_comm_pattern()
121   {
122     communication_indices_.clear();
123     for (auto const& list_process_comm : checker_->initial_communications_pattern)
124       this->communication_indices_.push_back(list_process_comm.index_comm);
125   }
126 };
127 simgrid::xbt::Extension<simgrid::mc::State, StateCommDet> StateCommDet::EXTENSION_ID;
128
129 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
130                                                                const simgrid::mc::PatternCommunication* comm2)
131 {
132   using simgrid::mc::CommPatternDifference;
133   if (comm1->type != comm2->type)
134     return CommPatternDifference::TYPE;
135   if (comm1->mbox != comm2->mbox)
136     return CommPatternDifference::MBOX;
137   if (comm1->src_proc != comm2->src_proc)
138     return CommPatternDifference::SRC_PROC;
139   if (comm1->dst_proc != comm2->dst_proc)
140     return CommPatternDifference::DST_PROC;
141   if (comm1->tag != comm2->tag)
142     return CommPatternDifference::TAG;
143   if (comm1->size != comm2->size)
144     return CommPatternDifference::DATA_SIZE;
145   return CommPatternDifference::NONE;
146 }
147
148 void CommDetExtension::restore_communications_pattern(const simgrid::mc::State* state)
149 {
150   for (size_t i = 0; i < initial_communications_pattern.size(); i++)
151     initial_communications_pattern[i].index_comm =
152         state->extension<simgrid::mc::StateCommDet>()->communication_indices_[i];
153
154   const unsigned long maxpid = api::get().get_maxpid();
155   for (unsigned long i = 0; i < maxpid; i++) {
156     incomplete_communications_pattern[i].clear();
157     for (simgrid::mc::PatternCommunication const& comm :
158          state->extension<simgrid::mc::StateCommDet>()->incomplete_comm_pattern_[i])
159       incomplete_communications_pattern[i].push_back(new simgrid::mc::PatternCommunication(comm.dup()));
160   }
161 }
162
163 static std::string print_determinism_result(simgrid::mc::CommPatternDifference diff, aid_t process,
164                                             const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
165 {
166   std::string type;
167   std::string res;
168
169   if (comm->type == simgrid::mc::PatternCommunicationType::send)
170     type = xbt::string_printf("The send communications pattern of the process %ld is different!", process - 1);
171   else
172     type = xbt::string_printf("The recv communications pattern of the process %ld is different!", process - 1);
173
174   using simgrid::mc::CommPatternDifference;
175   switch (diff) {
176     case CommPatternDifference::TYPE:
177       res = xbt::string_printf("%s Different type for communication #%u", type.c_str(), cursor);
178       break;
179     case CommPatternDifference::MBOX:
180       res = xbt::string_printf("%s Different mailbox for communication #%u", type.c_str(), cursor);
181       break;
182     case CommPatternDifference::TAG:
183       res = xbt::string_printf("%s Different tag for communication #%u", type.c_str(), cursor);
184       break;
185     case CommPatternDifference::SRC_PROC:
186       res = xbt::string_printf("%s Different source for communication #%u", type.c_str(), cursor);
187       break;
188     case CommPatternDifference::DST_PROC:
189       res = xbt::string_printf("%s Different destination for communication #%u", type.c_str(), cursor);
190       break;
191     case CommPatternDifference::DATA_SIZE:
192       res = xbt::string_printf("%s Different data size for communication #%u", type.c_str(), cursor);
193       break;
194     default:
195       res = "";
196       break;
197   }
198
199   return res;
200 }
201
202 void CommDetExtension::deterministic_comm_pattern(aid_t actor, const PatternCommunication* comm, bool backtracking)
203 {
204   if (not backtracking) {
205     PatternCommunicationList& list = initial_communications_pattern[actor];
206     CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
207
208     if (diff != CommPatternDifference::NONE) {
209       if (comm->type == PatternCommunicationType::send) {
210         send_deterministic = false;
211         send_diff          = print_determinism_result(diff, actor, comm, list.index_comm + 1);
212       } else {
213         recv_deterministic = false;
214         recv_diff          = print_determinism_result(diff, actor, comm, list.index_comm + 1);
215       }
216       if (_sg_mc_send_determinism && not send_deterministic) {
217         XBT_INFO("*********************************************************");
218         XBT_INFO("***** Non-send-deterministic communications pattern *****");
219         XBT_INFO("*********************************************************");
220         XBT_INFO("%s", send_diff.c_str());
221         api::get().log_state();
222         api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
223       } else if (_sg_mc_comms_determinism && (not send_deterministic && not recv_deterministic)) {
224         XBT_INFO("****************************************************");
225         XBT_INFO("***** Non-deterministic communications pattern *****");
226         XBT_INFO("****************************************************");
227         if (not send_diff.empty())
228           XBT_INFO("%s", send_diff.c_str());
229         if (not recv_diff.empty())
230           XBT_INFO("%s", recv_diff.c_str());
231         api::get().log_state();
232         api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
233       }
234     }
235   }
236 }
237
238 /********** Non Static functions ***********/
239
240 void CommDetExtension::get_comm_pattern(const Transition* transition, bool backtracking)
241 {
242   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[transition->aid_];
243   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[transition->aid_];
244
245   auto pattern   = std::make_unique<PatternCommunication>();
246   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
247
248   if (transition->type_ == Transition::Type::COMM_SEND) {
249     auto* send = dynamic_cast<const CommSendTransition*>(transition);
250     /* Create comm pattern */
251     pattern->type      = PatternCommunicationType::send;
252     pattern->comm_addr = send->get_comm();
253     pattern->mbox      = send->get_mailbox();
254     pattern->src_proc  = send->aid_;
255     pattern->tag       = send->get_tag();
256
257 #if HAVE_SMPI
258     // auto send_detached = api::get().check_send_request_detached(request);
259     if (false) { // send_detached) {
260       if (initial_communications_pattern_done) {
261         /* Evaluate comm determinism */
262         deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
263         initial_communications_pattern[pattern->src_proc].index_comm++;
264       } else {
265         /* Store comm pattern */
266         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
267       }
268       return;
269     }
270 #endif
271   } else if (transition->type_ == Transition::Type::COMM_RECV) {
272     auto* recv = static_cast<const CommRecvTransition*>(transition);
273
274     pattern->type = PatternCommunicationType::receive;
275     pattern->comm_addr = recv->get_comm();
276     pattern->tag       = recv->get_tag();
277     pattern->mbox     = recv->get_mailbox();
278     pattern->dst_proc = recv->aid_;
279   }
280
281   XBT_DEBUG("Insert incomplete comm pattern %p type:%d for process %ld (comm: %" PRIxPTR ")", pattern.get(),
282             (int)pattern->type, transition->aid_, pattern->comm_addr);
283   incomplete_communications_pattern[transition->aid_].push_back(pattern.release());
284 }
285
286 void CommDetExtension::complete_comm_pattern(const CommWaitTransition* transition, bool backtracking)
287 {
288   /* Complete comm pattern */
289   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[transition->aid_];
290   uintptr_t comm_addr                                    = transition->get_comm();
291   auto current_comm_pattern =
292       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
293                    [&comm_addr](const PatternCommunication* comm) { return (comm->comm_addr == comm_addr); });
294   xbt_assert(current_comm_pattern != std::end(incomplete_pattern), "Corresponding communication not found!");
295   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
296
297   comm_pattern->src_proc = transition->get_sender();
298   comm_pattern->dst_proc = transition->get_receiver();
299
300   XBT_DEBUG("Remove incomplete comm pattern for actor %ld at cursor %zd", transition->aid_,
301             std::distance(begin(incomplete_pattern), current_comm_pattern));
302   incomplete_pattern.erase(current_comm_pattern);
303
304   if (initial_communications_pattern_done) {
305     /* Evaluate comm determinism */
306     deterministic_comm_pattern(transition->aid_, comm_pattern.get(), backtracking);
307     initial_communications_pattern[transition->aid_].index_comm++;
308   } else {
309     /* Store comm pattern */
310     initial_communications_pattern[transition->aid_].list.push_back(std::move(comm_pattern));
311   }
312 }
313
314 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session* session) : Checker(session)
315 {
316   CommDetExtension::EXTENSION_ID = simgrid::mc::Checker::extension_create<CommDetExtension>();
317   StateCommDet::EXTENSION_ID = simgrid::mc::State::extension_create<StateCommDet>();
318 }
319
320 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
321
322 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
323 {
324   RecordTrace res;
325   for (auto const& state : stack_)
326     res.push_back(state->get_transition());
327   return res;
328 }
329
330 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
331 {
332   std::vector<std::string> trace;
333   for (auto const& state : stack_)
334     trace.push_back(state->get_transition()->to_string());
335   return trace;
336 }
337
338 void CommunicationDeterminismChecker::log_state() // override
339 {
340   if (_sg_mc_comms_determinism) {
341     if (extension<CommDetExtension>()->send_deterministic && not extension<CommDetExtension>()->recv_deterministic) {
342       XBT_INFO("*******************************************************");
343       XBT_INFO("**** Only-send-deterministic communication pattern ****");
344       XBT_INFO("*******************************************************");
345       XBT_INFO("%s", extension<CommDetExtension>()->recv_diff.c_str());
346     }
347     if (not extension<CommDetExtension>()->send_deterministic && extension<CommDetExtension>()->recv_deterministic) {
348       XBT_INFO("*******************************************************");
349       XBT_INFO("**** Only-recv-deterministic communication pattern ****");
350       XBT_INFO("*******************************************************");
351       XBT_INFO("%s", extension<CommDetExtension>()->send_diff.c_str());
352     }
353   }
354   XBT_INFO("Expanded states = %ld", State::get_expanded_states());
355   XBT_INFO("Visited states = %lu", api::get().mc_get_visited_states());
356   XBT_INFO("Executed transitions = %lu", Transition::get_executed_transitions());
357   XBT_INFO("Send-deterministic : %s", extension<CommDetExtension>()->send_deterministic ? "Yes" : "No");
358   if (_sg_mc_comms_determinism)
359     XBT_INFO("Recv-deterministic : %s", extension<CommDetExtension>()->recv_deterministic ? "Yes" : "No");
360 }
361
362 void CommunicationDeterminismChecker::prepare()
363 {
364   auto initial_state = std::make_unique<State>();
365   extension_set(new CommDetExtension());
366   initial_state->extension_set(new StateCommDet(extension<CommDetExtension>()));
367
368   XBT_DEBUG("********* Start communication determinism verification *********");
369
370   /* Add all enabled actors to the interleave set of the initial state */
371   for (auto& act : api::get().get_actors()) {
372     auto actor = act.copy.get_buffer();
373     if (get_session().actor_is_enabled(actor->get_pid()))
374       initial_state->mark_todo(actor->get_pid());
375   }
376
377   stack_.push_back(std::move(initial_state));
378 }
379
380 void CommunicationDeterminismChecker::restoreState()
381 {
382   auto extension = this->extension<CommDetExtension>();
383
384   /* If asked to rollback on a state that has a snapshot, restore it */
385   State* last_state = stack_.back().get();
386   if (last_state->system_state_) {
387     api::get().restore_state(last_state->system_state_);
388     extension->restore_communications_pattern(last_state);
389     return;
390   }
391
392   /* if no snapshot, we need to restore the initial state and replay the transitions */
393   get_session().restore_initial_state();
394
395   const unsigned long maxpid = api::get().get_maxpid();
396   assert(maxpid == extension->incomplete_communications_pattern.size());
397   assert(maxpid == extension->initial_communications_pattern.size());
398   for (unsigned long j = 0; j < maxpid; j++) {
399     extension->incomplete_communications_pattern[j].clear();
400     extension->initial_communications_pattern[j].index_comm = 0;
401   }
402
403   /* Traverse the stack from the state at position start and re-execute the transitions */
404   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
405     if (state == stack_.back())
406       break;
407
408     auto* transition = state->get_transition();
409     transition->replay();
410     extension->handle_comm_pattern(transition, true);
411
412     /* Update statistics */
413     api::get().mc_inc_visited_states();
414   }
415 }
416
417 void CommDetExtension::handle_comm_pattern(const Transition* transition, bool backtracking)
418 {
419   if (not _sg_mc_comms_determinism && not _sg_mc_send_determinism)
420     return;
421
422   switch (transition->type_) {
423     case Transition::Type::COMM_SEND:
424       get_comm_pattern(transition, backtracking);
425       break;
426     case Transition::Type::COMM_RECV:
427       get_comm_pattern(transition, backtracking);
428       break;
429     case Transition::Type::COMM_WAIT: {
430       complete_comm_pattern(static_cast<const CommWaitTransition*>(transition), backtracking);
431       break;
432     }
433     case Transition::Type::WAITANY: {
434       auto const* t    = static_cast<const WaitAnyTransition*>(transition)->get_current_transition();
435       auto const* wait = dynamic_cast<const CommWaitTransition*>(t);
436       if (wait != nullptr) // Ignore wait on non-comm
437         complete_comm_pattern(wait, backtracking);
438     } break;
439     default: /* Ignore unhandled transition types */
440       break;
441   }
442 }
443
444 void CommunicationDeterminismChecker::real_run()
445 {
446   auto extension = this->extension<CommDetExtension>();
447
448   std::unique_ptr<VisitedState> visited_state = nullptr;
449   const unsigned long maxpid                  = api::get().get_maxpid();
450
451   while (not stack_.empty()) {
452     /* Get current state */
453     State* cur_state = stack_.back().get();
454
455     XBT_DEBUG("**************************************************");
456     XBT_DEBUG("Exploration depth = %zu (state = %ld, interleaved processes = %zu)", stack_.size(), cur_state->num_,
457               cur_state->count_todo());
458
459     /* Update statistics */
460     api::get().mc_inc_visited_states();
461
462     int next_transition = -1;
463     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
464       next_transition = cur_state->next_transition();
465
466     if (next_transition >= 0 && visited_state == nullptr) {
467       cur_state->execute_next(next_transition);
468
469       auto* transition  = cur_state->get_transition();
470       XBT_DEBUG("Execute: %s", transition->to_string().c_str());
471
472       std::string req_str;
473       if (dot_output != nullptr)
474         req_str = api::get().request_get_dot_output(transition);
475
476       extension->handle_comm_pattern(transition, false);
477
478       /* Create the new expanded state */
479       auto next_state = std::make_unique<State>();
480       next_state->extension_set(new StateCommDet(extension));
481
482       bool all_communications_are_finished = true;
483       for (size_t current_actor = 1; all_communications_are_finished && current_actor < maxpid; current_actor++) {
484         if (not extension->incomplete_communications_pattern[current_actor].empty()) {
485           XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
486           all_communications_are_finished = false;
487         }
488       }
489
490       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
491        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
492        * with the initial pattern. */
493       bool compare_snapshots = extension->initial_communications_pattern_done && all_communications_are_finished;
494
495       if (_sg_mc_max_visited_states != 0)
496         visited_state = visited_states_.addVisitedState(next_state->num_, next_state.get(), compare_snapshots);
497       else
498         visited_state = nullptr;
499
500       if (visited_state == nullptr) {
501         /* Add all enabled actors to the interleave set of the next state */
502         for (auto& act : api::get().get_actors()) {
503           auto actor = act.copy.get_buffer();
504           if (get_session().actor_is_enabled(actor->get_pid()))
505             next_state->mark_todo(actor->get_pid());
506         }
507
508         if (dot_output != nullptr)
509           fprintf(dot_output, "\"%ld\" -> \"%ld\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
510
511       } else if (dot_output != nullptr)
512         fprintf(dot_output, "\"%ld\" -> \"%ld\" [%s];\n", cur_state->num_,
513                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
514
515       stack_.push_back(std::move(next_state));
516     } else {
517       if (stack_.size() > (std::size_t)_sg_mc_max_depth)
518         XBT_WARN("/!\\ Max depth reached! /!\\ ");
519       else if (visited_state != nullptr)
520         XBT_DEBUG("State already visited (equal to state %ld), exploration stopped on this path.",
521                   visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
522       else
523         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
524
525       extension->initial_communications_pattern_done = true;
526
527       /* Trash the current state, no longer needed */
528       XBT_DEBUG("Delete state %ld at depth %zu", cur_state->num_, stack_.size());
529       stack_.pop_back();
530
531       visited_state = nullptr;
532
533       api::get().mc_check_deadlock();
534
535       while (not stack_.empty()) {
536         std::unique_ptr<State> state(std::move(stack_.back()));
537         stack_.pop_back();
538         if (state->count_todo() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
539           /* We found a back-tracking point, let's loop */
540           XBT_DEBUG("Back-tracking to state %ld at depth %zu", state->num_, stack_.size() + 1);
541           stack_.push_back(std::move(state));
542
543           this->restoreState();
544
545           XBT_DEBUG("Back-tracking to state %ld at depth %zu done", stack_.back()->num_, stack_.size());
546
547           break;
548         } else {
549           XBT_DEBUG("Delete state %ld at depth %zu", state->num_, stack_.size() + 1);
550         }
551       }
552     }
553   }
554
555   api::get().log_state();
556 }
557
558 void CommunicationDeterminismChecker::run()
559 {
560   XBT_INFO("Check communication determinism");
561   get_session().take_initial_snapshot();
562
563   this->prepare();
564   this->real_run();
565 }
566
567 Checker* create_communication_determinism_checker(Session* session)
568 {
569   return new CommunicationDeterminismChecker(session);
570 }
571
572 } // namespace mc
573 } // namespace simgrid