Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Merge branch 'add_missing_comm_python_bindings' into 'master'
[simgrid.git] / 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 {
20 namespace mc {
21
22 enum class CallType { NONE, SEND, RECV, WAIT, WAITANY };
23 enum class CommPatternDifference { NONE, TYPE, MBOX, TAG, SRC_PROC, DST_PROC, DATA_SIZE };
24 enum class PatternCommunicationType {
25   none    = 0,
26   send    = 1,
27   receive = 2,
28 };
29
30 class PatternCommunication {
31 public:
32   int num                       = 0;
33   uintptr_t comm_addr           = 0;
34   PatternCommunicationType type = PatternCommunicationType::send;
35   unsigned long src_proc        = 0;
36   unsigned long dst_proc        = 0;
37   unsigned mbox                 = 0;
38   unsigned size                 = 0;
39   int tag                       = 0;
40   int index                     = 0;
41
42   PatternCommunication dup() const
43   {
44     simgrid::mc::PatternCommunication res;
45     // num?
46     res.comm_addr = this->comm_addr;
47     res.type      = this->type;
48     res.src_proc  = this->src_proc;
49     res.dst_proc  = this->dst_proc;
50     res.mbox      = this->mbox;
51     res.tag       = this->tag;
52     res.index     = this->index;
53     return res;
54   }
55 };
56
57 struct PatternCommunicationList {
58   unsigned int index_comm = 0;
59   std::vector<std::unique_ptr<simgrid::mc::PatternCommunication>> list;
60 };
61
62 /********** Checker extension **********/
63
64 struct CommDetExtension {
65   static simgrid::xbt::Extension<simgrid::mc::Exploration, CommDetExtension> EXTENSION_ID;
66
67   std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
68   std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
69
70   bool initial_communications_pattern_done = false;
71   bool recv_deterministic                  = true;
72   bool send_deterministic                  = true;
73   std::string send_diff;
74   std::string recv_diff;
75
76   void exploration_start()
77   {
78     const unsigned long maxpid = Api::get().get_maxpid();
79
80     initial_communications_pattern.resize(maxpid);
81     incomplete_communications_pattern.resize(maxpid);
82   }
83   void restore_communications_pattern(const simgrid::mc::State* state);
84   void enforce_deterministic_pattern(aid_t process, const PatternCommunication* comm);
85   void get_comm_pattern(const Transition* transition);
86   void complete_comm_pattern(const CommWaitTransition* transition);
87   void handle_comm_pattern(const Transition* transition);
88 };
89 simgrid::xbt::Extension<simgrid::mc::Exploration, CommDetExtension> CommDetExtension::EXTENSION_ID;
90 /********** State Extension ***********/
91
92 class StateCommDet {
93 public:
94   std::vector<std::vector<simgrid::mc::PatternCommunication>> incomplete_comm_pattern_;
95   std::vector<unsigned> communication_indices_;
96
97   static simgrid::xbt::Extension<simgrid::mc::State, StateCommDet> EXTENSION_ID;
98   explicit StateCommDet(CommDetExtension* checker)
99   {
100     const unsigned long maxpid = Api::get().get_maxpid();
101     for (unsigned long i = 0; i < maxpid; i++) {
102       std::vector<simgrid::mc::PatternCommunication> res;
103       for (auto const& comm : checker->incomplete_communications_pattern[i])
104         res.push_back(comm->dup());
105       incomplete_comm_pattern_.push_back(std::move(res));
106     }
107
108     for (auto const& list_process_comm : checker->initial_communications_pattern)
109       this->communication_indices_.push_back(list_process_comm.index_comm);
110   }
111 };
112 simgrid::xbt::Extension<simgrid::mc::State, StateCommDet> StateCommDet::EXTENSION_ID;
113
114 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
115                                                                const simgrid::mc::PatternCommunication* comm2)
116 {
117   using simgrid::mc::CommPatternDifference;
118   if (comm1->type != comm2->type)
119     return CommPatternDifference::TYPE;
120   if (comm1->mbox != comm2->mbox)
121     return CommPatternDifference::MBOX;
122   if (comm1->src_proc != comm2->src_proc)
123     return CommPatternDifference::SRC_PROC;
124   if (comm1->dst_proc != comm2->dst_proc)
125     return CommPatternDifference::DST_PROC;
126   if (comm1->tag != comm2->tag)
127     return CommPatternDifference::TAG;
128   if (comm1->size != comm2->size)
129     return CommPatternDifference::DATA_SIZE;
130   return CommPatternDifference::NONE;
131 }
132
133 void CommDetExtension::restore_communications_pattern(const simgrid::mc::State* state)
134 {
135   for (size_t i = 0; i < initial_communications_pattern.size(); i++)
136     initial_communications_pattern[i].index_comm =
137         state->extension<simgrid::mc::StateCommDet>()->communication_indices_[i];
138
139   const unsigned long maxpid = Api::get().get_maxpid();
140   for (unsigned long i = 0; i < maxpid; i++) {
141     incomplete_communications_pattern[i].clear();
142     for (simgrid::mc::PatternCommunication const& comm :
143          state->extension<simgrid::mc::StateCommDet>()->incomplete_comm_pattern_[i])
144       incomplete_communications_pattern[i].push_back(new simgrid::mc::PatternCommunication(comm.dup()));
145   }
146 }
147
148 static std::string print_determinism_result(simgrid::mc::CommPatternDifference diff, aid_t actor,
149                                             const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
150 {
151   std::string type;
152   std::string res;
153
154   if (comm->type == simgrid::mc::PatternCommunicationType::send)
155     type = xbt::string_printf("The send communications pattern of the actor %ld is different!", actor - 1);
156   else
157     type = xbt::string_printf("The recv communications pattern of the actor %ld is different!", actor - 1);
158
159   using simgrid::mc::CommPatternDifference;
160   switch (diff) {
161     case CommPatternDifference::TYPE:
162       res = xbt::string_printf("%s Different type for communication #%u", type.c_str(), cursor);
163       break;
164     case CommPatternDifference::MBOX:
165       res = xbt::string_printf("%s Different mailbox for communication #%u", type.c_str(), cursor);
166       break;
167     case CommPatternDifference::TAG:
168       res = xbt::string_printf("%s Different tag for communication #%u", type.c_str(), cursor);
169       break;
170     case CommPatternDifference::SRC_PROC:
171       res = xbt::string_printf("%s Different source for communication #%u", type.c_str(), cursor);
172       break;
173     case CommPatternDifference::DST_PROC:
174       res = xbt::string_printf("%s Different destination for communication #%u", type.c_str(), cursor);
175       break;
176     case CommPatternDifference::DATA_SIZE:
177       res = xbt::string_printf("%s Different data size for communication #%u", type.c_str(), cursor);
178       break;
179     default:
180       res = "";
181       break;
182   }
183
184   return res;
185 }
186
187 void CommDetExtension::enforce_deterministic_pattern(aid_t actor, const PatternCommunication* comm)
188 {
189   PatternCommunicationList& list = initial_communications_pattern[actor];
190   CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
191
192   if (diff != CommPatternDifference::NONE) {
193     if (comm->type == PatternCommunicationType::send) {
194       send_deterministic = false;
195       send_diff          = print_determinism_result(diff, actor, comm, list.index_comm + 1);
196     } else {
197       recv_deterministic = false;
198       if (recv_diff.empty())
199         recv_diff = print_determinism_result(diff, actor, comm, list.index_comm + 1);
200     }
201     if (_sg_mc_send_determinism && not send_deterministic) {
202       XBT_INFO("*********************************************************");
203       XBT_INFO("***** Non-send-deterministic communications pattern *****");
204       XBT_INFO("*********************************************************");
205       XBT_INFO("%s", send_diff.c_str());
206       Api::get().get_session().log_state();
207       Api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
208     } else if (_sg_mc_comms_determinism && (not send_deterministic && not recv_deterministic)) {
209       XBT_INFO("****************************************************");
210       XBT_INFO("***** Non-deterministic communications pattern *****");
211       XBT_INFO("****************************************************");
212       if (not send_diff.empty())
213         XBT_INFO("%s", send_diff.c_str());
214       if (not recv_diff.empty())
215         XBT_INFO("%s", recv_diff.c_str());
216       Api::get().get_session().log_state();
217       Api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
218     }
219   }
220 }
221
222 /********** Non Static functions ***********/
223
224 void CommDetExtension::get_comm_pattern(const Transition* transition)
225 {
226   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[transition->aid_];
227   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[transition->aid_];
228
229   auto pattern   = std::make_unique<PatternCommunication>();
230   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
231
232   if (transition->type_ == Transition::Type::COMM_SEND) {
233     auto* send = static_cast<const CommSendTransition*>(transition);
234
235     pattern->type      = PatternCommunicationType::send;
236     pattern->comm_addr = send->get_comm();
237     pattern->tag       = send->get_tag();
238
239     // FIXME: Detached sends should be enforced when the receive is waited
240
241   } else if (transition->type_ == Transition::Type::COMM_RECV) {
242     auto* recv = static_cast<const CommRecvTransition*>(transition);
243
244     pattern->type      = PatternCommunicationType::receive;
245     pattern->comm_addr = recv->get_comm();
246     pattern->tag       = recv->get_tag();
247   }
248
249   XBT_DEBUG("Insert incomplete comm pattern %p type:%d for process %ld (comm: %" PRIxPTR ")", pattern.get(),
250             (int)pattern->type, transition->aid_, pattern->comm_addr);
251   incomplete_communications_pattern[transition->aid_].push_back(pattern.release());
252 }
253
254 void CommDetExtension::complete_comm_pattern(const CommWaitTransition* transition)
255 {
256   /* Complete comm pattern */
257   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[transition->aid_];
258   uintptr_t comm_addr                                    = transition->get_comm();
259   auto current_comm_pattern =
260       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
261                    [&comm_addr](const PatternCommunication* comm) { return (comm->comm_addr == comm_addr); });
262   xbt_assert(current_comm_pattern != std::end(incomplete_pattern), "Corresponding communication not found!");
263   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
264
265   comm_pattern->src_proc = transition->get_sender();
266   comm_pattern->dst_proc = transition->get_receiver();
267   comm_pattern->mbox     = transition->get_mailbox();
268
269   XBT_DEBUG("Remove incomplete comm pattern for actor %ld at cursor %zd", transition->aid_,
270             std::distance(begin(incomplete_pattern), current_comm_pattern));
271   incomplete_pattern.erase(current_comm_pattern);
272
273   if (initial_communications_pattern_done) {
274     /* Evaluate comm determinism */
275     enforce_deterministic_pattern(transition->aid_, comm_pattern.get());
276     initial_communications_pattern[transition->aid_].index_comm++;
277   } else {
278     /* Store comm pattern */
279     initial_communications_pattern[transition->aid_].list.push_back(std::move(comm_pattern));
280   }
281 }
282
283 void CommDetExtension::handle_comm_pattern(const Transition* transition)
284 {
285   if (not _sg_mc_comms_determinism && not _sg_mc_send_determinism)
286     return;
287
288   switch (transition->type_) {
289     case Transition::Type::COMM_SEND:
290       get_comm_pattern(transition);
291       break;
292     case Transition::Type::COMM_RECV:
293       get_comm_pattern(transition);
294       break;
295     case Transition::Type::COMM_WAIT:
296       complete_comm_pattern(static_cast<const CommWaitTransition*>(transition));
297       break;
298     case Transition::Type::WAITANY:
299       // Ignore wait on non-comm
300       if (auto const* wait = dynamic_cast<const CommWaitTransition*>(
301               static_cast<const WaitAnyTransition*>(transition)->get_current_transition()))
302         complete_comm_pattern(wait);
303       break;
304     default: /* Ignore unhandled transition types */
305       break;
306   }
307 }
308
309 /* FIXME: CommDet probably don't play nicely with stateful exploration
310
311       bool all_communications_are_finished = true;
312       for (size_t current_actor = 1; all_communications_are_finished && current_actor < maxpid; current_actor++) {
313         if (not extension->incomplete_communications_pattern[current_actor].empty()) {
314           XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
315           all_communications_are_finished = false;
316         }
317       }
318  */
319
320 Exploration* create_communication_determinism_checker(Session* session)
321 {
322   CommDetExtension::EXTENSION_ID = simgrid::mc::Exploration::extension_create<CommDetExtension>();
323   StateCommDet::EXTENSION_ID     = simgrid::mc::State::extension_create<StateCommDet>();
324
325   XBT_DEBUG("********* Start communication determinism verification *********");
326
327   auto extension = new CommDetExtension();
328
329   DFSExplorer::on_exploration_start([extension]() {
330     XBT_INFO("Check communication determinism");
331     extension->exploration_start();
332   });
333   DFSExplorer::on_backtracking([extension]() { extension->initial_communications_pattern_done = true; });
334   DFSExplorer::on_state_creation([extension](State* state) { state->extension_set(new StateCommDet(extension)); });
335
336   DFSExplorer::on_restore_system_state([extension](State* state) { extension->restore_communications_pattern(state); });
337
338   DFSExplorer::on_restore_initial_state([extension]() {
339     const unsigned long maxpid = Api::get().get_maxpid();
340     assert(maxpid == extension->incomplete_communications_pattern.size());
341     assert(maxpid == extension->initial_communications_pattern.size());
342     for (unsigned long j = 0; j < maxpid; j++) {
343       extension->incomplete_communications_pattern[j].clear();
344       extension->initial_communications_pattern[j].index_comm = 0;
345     }
346   });
347
348   DFSExplorer::on_transition_replay([extension](Transition* t) { extension->handle_comm_pattern(t); });
349   DFSExplorer::on_transition_execute([extension](Transition* t) { extension->handle_comm_pattern(t); });
350
351   DFSExplorer::on_log_state([extension]() {
352     if (_sg_mc_comms_determinism) {
353       if (extension->send_deterministic && not extension->recv_deterministic) {
354         XBT_INFO("*******************************************************");
355         XBT_INFO("**** Only-send-deterministic communication pattern ****");
356         XBT_INFO("*******************************************************");
357         XBT_INFO("%s", extension->recv_diff.c_str());
358       }
359       if (not extension->send_deterministic && extension->recv_deterministic) {
360         XBT_INFO("*******************************************************");
361         XBT_INFO("**** Only-recv-deterministic communication pattern ****");
362         XBT_INFO("*******************************************************");
363         XBT_INFO("%s", extension->send_diff.c_str());
364       }
365     }
366     XBT_INFO("Send-deterministic : %s", extension->send_deterministic ? "Yes" : "No");
367     if (_sg_mc_comms_determinism)
368       XBT_INFO("Recv-deterministic : %s", extension->recv_deterministic ? "Yes" : "No");
369     delete extension;
370   });
371
372   return new DFSExplorer(session);
373 }
374
375 } // namespace mc
376 } // namespace simgrid