Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Remove the stateful model-checking from the archive. It's not working anymore
[simgrid.git] / src / mc / explo / CommunicationDeterminismChecker.cpp
1 /* Copyright (c) 2008-2023. 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 #include "xbt/string.hpp"
15
16 #include <cstdint>
17 #include <inttypes.h>
18
19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
20
21 namespace simgrid::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 class CommDetExtension {
66   Exploration& exploration_;
67
68 public:
69   static simgrid::xbt::Extension<simgrid::mc::Exploration, CommDetExtension> EXTENSION_ID;
70
71   explicit CommDetExtension(Exploration& explo) : exploration_(explo) {}
72
73   std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
74   std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
75
76   bool initial_communications_pattern_done = false;
77   bool recv_deterministic                  = true;
78   bool send_deterministic                  = true;
79   std::string send_diff;
80   std::string recv_diff;
81
82   void exploration_start()
83   {
84     const unsigned long maxpid = exploration_.get_remote_app().get_maxpid();
85
86     initial_communications_pattern.resize(maxpid);
87     incomplete_communications_pattern.resize(maxpid);
88   }
89   void restore_communications_pattern(const simgrid::mc::State* state, RemoteApp const& remote_app);
90   void enforce_deterministic_pattern(aid_t process, const PatternCommunication* comm);
91   void get_comm_pattern(const Transition* transition);
92   void complete_comm_pattern(const CommWaitTransition* transition);
93   void handle_comm_pattern(const Transition* transition);
94 };
95 simgrid::xbt::Extension<simgrid::mc::Exploration, CommDetExtension> CommDetExtension::EXTENSION_ID;
96 /********** State Extension ***********/
97
98 class StateCommDet {
99 public:
100   std::vector<std::vector<simgrid::mc::PatternCommunication>> incomplete_comm_pattern_;
101   std::vector<unsigned> communication_indices_;
102
103   static simgrid::xbt::Extension<simgrid::mc::State, StateCommDet> EXTENSION_ID;
104   explicit StateCommDet(CommDetExtension const& checker, RemoteApp const& remote_app)
105   {
106     const unsigned long maxpid = remote_app.get_maxpid();
107     for (unsigned long i = 0; i < maxpid; i++) {
108       std::vector<simgrid::mc::PatternCommunication> res;
109       for (auto const& comm : checker.incomplete_communications_pattern[i])
110         res.push_back(comm->dup());
111       incomplete_comm_pattern_.push_back(std::move(res));
112     }
113
114     for (auto const& list_process_comm : checker.initial_communications_pattern)
115       this->communication_indices_.push_back(list_process_comm.index_comm);
116   }
117 };
118 simgrid::xbt::Extension<simgrid::mc::State, StateCommDet> StateCommDet::EXTENSION_ID;
119
120 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
121                                                                const simgrid::mc::PatternCommunication* comm2)
122 {
123   using simgrid::mc::CommPatternDifference;
124   if (comm1->type != comm2->type)
125     return CommPatternDifference::TYPE;
126   if (comm1->mbox != comm2->mbox)
127     return CommPatternDifference::MBOX;
128   if (comm1->src_proc != comm2->src_proc)
129     return CommPatternDifference::SRC_PROC;
130   if (comm1->dst_proc != comm2->dst_proc)
131     return CommPatternDifference::DST_PROC;
132   if (comm1->tag != comm2->tag)
133     return CommPatternDifference::TAG;
134   if (comm1->size != comm2->size)
135     return CommPatternDifference::DATA_SIZE;
136   return CommPatternDifference::NONE;
137 }
138
139 void CommDetExtension::restore_communications_pattern(const simgrid::mc::State* state, RemoteApp const& remote_app)
140 {
141   for (size_t i = 0; i < initial_communications_pattern.size(); i++)
142     initial_communications_pattern[i].index_comm =
143         state->extension<simgrid::mc::StateCommDet>()->communication_indices_[i];
144
145   const unsigned long maxpid = remote_app.get_maxpid();
146   for (unsigned long i = 0; i < maxpid; i++) {
147     incomplete_communications_pattern[i].clear();
148     for (simgrid::mc::PatternCommunication const& comm :
149          state->extension<simgrid::mc::StateCommDet>()->incomplete_comm_pattern_[i])
150       incomplete_communications_pattern[i].push_back(new simgrid::mc::PatternCommunication(comm.dup()));
151   }
152 }
153
154 static std::string print_determinism_result(simgrid::mc::CommPatternDifference diff, aid_t actor,
155                                             const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
156 {
157   std::string type;
158   std::string res;
159
160   if (comm->type == simgrid::mc::PatternCommunicationType::send)
161     type = xbt::string_printf("The send communications pattern of the actor %ld is different!", actor - 1);
162   else
163     type = xbt::string_printf("The recv communications pattern of the actor %ld is different!", actor - 1);
164
165   using simgrid::mc::CommPatternDifference;
166   switch (diff) {
167     case CommPatternDifference::TYPE:
168       res = xbt::string_printf("%s Different type for communication #%u", type.c_str(), cursor);
169       break;
170     case CommPatternDifference::MBOX:
171       res = xbt::string_printf("%s Different mailbox for communication #%u", type.c_str(), cursor);
172       break;
173     case CommPatternDifference::TAG:
174       res = xbt::string_printf("%s Different tag for communication #%u", type.c_str(), cursor);
175       break;
176     case CommPatternDifference::SRC_PROC:
177       res = xbt::string_printf("%s Different source for communication #%u", type.c_str(), cursor);
178       break;
179     case CommPatternDifference::DST_PROC:
180       res = xbt::string_printf("%s Different destination for communication #%u", type.c_str(), cursor);
181       break;
182     case CommPatternDifference::DATA_SIZE:
183       res = xbt::string_printf("%s Different data size for communication #%u", type.c_str(), cursor);
184       break;
185     default:
186       res = "";
187       break;
188   }
189
190   return res;
191 }
192
193 void CommDetExtension::enforce_deterministic_pattern(aid_t actor, const PatternCommunication* comm)
194 {
195   PatternCommunicationList& list = initial_communications_pattern[actor];
196   CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
197
198   if (diff != CommPatternDifference::NONE) {
199     if (comm->type == PatternCommunicationType::send) {
200       send_deterministic = false;
201       send_diff          = print_determinism_result(diff, actor, comm, list.index_comm + 1);
202     } else {
203       recv_deterministic = false;
204       if (recv_diff.empty())
205         recv_diff = print_determinism_result(diff, actor, comm, list.index_comm + 1);
206     }
207     if (_sg_mc_send_determinism && not send_deterministic) {
208       XBT_INFO("*********************************************************");
209       XBT_INFO("***** Non-send-deterministic communications pattern *****");
210       XBT_INFO("*********************************************************");
211       XBT_INFO("%s", send_diff.c_str());
212       exploration_.log_state();
213       throw McError(ExitStatus::NON_DETERMINISM);
214     } else if (_sg_mc_comms_determinism && (not send_deterministic && not recv_deterministic)) {
215       XBT_INFO("****************************************************");
216       XBT_INFO("***** Non-deterministic communications pattern *****");
217       XBT_INFO("****************************************************");
218       if (not send_diff.empty())
219         XBT_INFO("%s", send_diff.c_str());
220       if (not recv_diff.empty())
221         XBT_INFO("%s", recv_diff.c_str());
222       exploration_.log_state();
223       throw McError(ExitStatus::NON_DETERMINISM);
224     }
225   }
226 }
227
228 /********** Non Static functions ***********/
229
230 void CommDetExtension::get_comm_pattern(const Transition* transition)
231 {
232   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[transition->aid_];
233   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[transition->aid_];
234
235   auto pattern   = std::make_unique<PatternCommunication>();
236   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
237
238   if (transition->type_ == Transition::Type::COMM_ASYNC_SEND) {
239     const auto* send = static_cast<const CommSendTransition*>(transition);
240
241     pattern->type      = PatternCommunicationType::send;
242     pattern->comm_addr = send->get_comm();
243     pattern->tag       = send->get_tag();
244
245     // FIXME: Detached sends should be enforced when the receive is waited
246
247   } else if (transition->type_ == Transition::Type::COMM_ASYNC_RECV) {
248     const auto* recv = static_cast<const CommRecvTransition*>(transition);
249
250     pattern->type      = PatternCommunicationType::receive;
251     pattern->comm_addr = recv->get_comm();
252     pattern->tag       = recv->get_tag();
253   }
254
255   XBT_DEBUG("Insert incomplete comm pattern %p type:%d for process %ld (comm: %" PRIxPTR ")", pattern.get(),
256             (int)pattern->type, transition->aid_, pattern->comm_addr);
257   incomplete_communications_pattern[transition->aid_].push_back(pattern.release());
258 }
259
260 void CommDetExtension::complete_comm_pattern(const CommWaitTransition* transition)
261 {
262   /* Complete comm pattern */
263   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[transition->aid_];
264   uintptr_t comm_addr                                    = transition->get_comm();
265   auto current_comm_pattern =
266       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
267                    [&comm_addr](const PatternCommunication* comm) { return (comm->comm_addr == comm_addr); });
268   xbt_assert(current_comm_pattern != std::end(incomplete_pattern), "Corresponding communication not found!");
269   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
270
271   comm_pattern->src_proc = transition->get_sender();
272   comm_pattern->dst_proc = transition->get_receiver();
273   comm_pattern->mbox     = transition->get_mailbox();
274
275   XBT_DEBUG("Remove incomplete comm pattern for actor %ld at cursor %zd", transition->aid_,
276             std::distance(begin(incomplete_pattern), current_comm_pattern));
277   incomplete_pattern.erase(current_comm_pattern);
278
279   if (initial_communications_pattern_done) {
280     /* Evaluate comm determinism */
281     enforce_deterministic_pattern(transition->aid_, comm_pattern.get());
282     initial_communications_pattern[transition->aid_].index_comm++;
283   } else {
284     /* Store comm pattern */
285     initial_communications_pattern[transition->aid_].list.push_back(std::move(comm_pattern));
286   }
287 }
288
289 void CommDetExtension::handle_comm_pattern(const Transition* transition)
290 {
291   if (not _sg_mc_comms_determinism && not _sg_mc_send_determinism)
292     return;
293
294   switch (transition->type_) {
295     case Transition::Type::COMM_ASYNC_SEND:
296     case Transition::Type::COMM_ASYNC_RECV:
297       get_comm_pattern(transition);
298       break;
299     case Transition::Type::COMM_WAIT:
300       complete_comm_pattern(static_cast<const CommWaitTransition*>(transition));
301       break;
302     case Transition::Type::WAITANY:
303       // Ignore wait on non-comm
304       if (auto const* wait = dynamic_cast<const CommWaitTransition*>(
305               static_cast<const WaitAnyTransition*>(transition)->get_current_transition()))
306         complete_comm_pattern(wait);
307       break;
308     default: /* Ignore unhandled transition types */
309       break;
310   }
311 }
312
313 /* FIXME: CommDet probably don't play nicely with stateful exploration
314
315       bool all_communications_are_finished = true;
316       for (size_t current_actor = 1; all_communications_are_finished && current_actor < maxpid; current_actor++) {
317         if (not extension->incomplete_communications_pattern[current_actor].empty()) {
318           XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
319           all_communications_are_finished = false;
320         }
321       }
322  */
323
324 Exploration* create_communication_determinism_checker(const std::vector<char*>& args, ReductionMode mode)
325 {
326   CommDetExtension::EXTENSION_ID = simgrid::mc::Exploration::extension_create<CommDetExtension>();
327   StateCommDet::EXTENSION_ID     = simgrid::mc::State::extension_create<StateCommDet>();
328
329   XBT_DEBUG("********* Start communication determinism verification *********");
330
331   auto* base      = new DFSExplorer(args, mode);
332   auto* extension = new CommDetExtension(*base);
333
334   DFSExplorer::on_exploration_start([extension](RemoteApp const&) {
335     XBT_INFO("Check communication determinism");
336     extension->exploration_start();
337   });
338   DFSExplorer::on_backtracking(
339       [extension](RemoteApp const&) { extension->initial_communications_pattern_done = true; });
340   DFSExplorer::on_state_creation([extension](State* state, RemoteApp const& remote_app) {
341     state->extension_set(new StateCommDet(*extension, remote_app));
342   });
343
344   DFSExplorer::on_restore_system_state([extension](State const* state, RemoteApp const& remote_app) {
345     extension->restore_communications_pattern(state, remote_app);
346   });
347
348   DFSExplorer::on_restore_initial_state([extension](RemoteApp const& remote_app) {
349     const unsigned long maxpid = remote_app.get_maxpid();
350     assert(maxpid == extension->incomplete_communications_pattern.size());
351     assert(maxpid == extension->initial_communications_pattern.size());
352     for (unsigned long j = 0; j < maxpid; j++) {
353       extension->incomplete_communications_pattern[j].clear();
354       extension->initial_communications_pattern[j].index_comm = 0;
355     }
356   });
357
358   DFSExplorer::on_transition_replay(
359       [extension](Transition const* t, RemoteApp const&) { extension->handle_comm_pattern(t); });
360   DFSExplorer::on_transition_execute(
361       [extension](Transition const* t, RemoteApp const&) { extension->handle_comm_pattern(t); });
362
363   DFSExplorer::on_log_state([extension](RemoteApp const&) {
364     if (_sg_mc_comms_determinism) {
365       if (extension->send_deterministic && not extension->recv_deterministic) {
366         XBT_INFO("*******************************************************");
367         XBT_INFO("**** Only-send-deterministic communication pattern ****");
368         XBT_INFO("*******************************************************");
369         XBT_INFO("%s", extension->recv_diff.c_str());
370       }
371       if (not extension->send_deterministic && extension->recv_deterministic) {
372         XBT_INFO("*******************************************************");
373         XBT_INFO("**** Only-recv-deterministic communication pattern ****");
374         XBT_INFO("*******************************************************");
375         XBT_INFO("%s", extension->send_diff.c_str());
376       }
377     }
378     XBT_INFO("Send-deterministic : %s", extension->send_deterministic ? "Yes" : "No");
379     if (_sg_mc_comms_determinism)
380       XBT_INFO("Recv-deterministic : %s", extension->recv_deterministic ? "Yes" : "No");
381     delete extension;
382   });
383
384   return base;
385 }
386
387 } // namespace simgrid::mc