src/s4u/s4u_Comm.cpp

   1 /* Copyright (c) 2006-2022. 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/msg/msg_private.hpp"
   7 //#include "xbt/log.h"
   8
   9 #include <simgrid/Exception.hpp>
  10 #include <simgrid/comm.h>
  11 #include <simgrid/s4u/Comm.hpp>
  12 #include <simgrid/s4u/Engine.hpp>
  13 #include <simgrid/s4u/Mailbox.hpp>
  14
  15 #include "src/kernel/activity/CommImpl.hpp"
  16 #include "src/kernel/actor/ActorImpl.hpp"
  17 #include "src/kernel/actor/SimcallObserver.hpp"
  18
  19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(s4u_comm, s4u_activity, "S4U asynchronous communications");
  20
  21 namespace simgrid {
  22 namespace s4u {
  23 xbt::signal<void(Comm const&)> Comm::on_send;
  24 xbt::signal<void(Comm const&)> Comm::on_recv;
  25 xbt::signal<void(Comm const&)> Comm::on_completion;
  26
  27 CommPtr Comm::set_copy_data_callback(void (*callback)(kernel::activity::CommImpl*, void*, size_t))
  28 {
  29   copy_data_function_ = callback;
  30   return this;
  31 }
  32
  33 void Comm::copy_buffer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
  34 {
  35   XBT_DEBUG("Copy the data over");
  36   memcpy(comm->dst_buff_, buff, buff_size);
  37   if (comm->is_detached()) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the
  38                              // original buffer available to the application ASAP
  39     xbt_free(buff);
  40     comm->src_buff_ = nullptr;
  41   }
  42 }
  43
  44 void Comm::copy_pointer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
  45 {
  46   xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
  47   *(void**)(comm->dst_buff_) = buff;
  48 }
  49
  50 Comm::~Comm()
  51 {
  52   if (state_ == State::STARTED && not detached_ &&
  53       (pimpl_ == nullptr || pimpl_->get_state() == kernel::activity::State::RUNNING)) {
  54     XBT_INFO("Comm %p freed before its completion. Did you forget to detach it? (state: %s)", this, get_state_str());
  55     if (pimpl_ != nullptr)
  56       XBT_INFO("pimpl_->state: %s", pimpl_->get_state_str());
  57     else
  58       XBT_INFO("pimpl_ is null");
  59     xbt_backtrace_display_current();
  60   }
  61 }
  62
  63 CommPtr Comm::sendto_init()
  64 {
  65   CommPtr res(new Comm());
  66   res->pimpl_ = kernel::activity::CommImplPtr(new kernel::activity::CommImpl());
  67   boost::static_pointer_cast<kernel::activity::CommImpl>(res->pimpl_)->detach();
  68   res->sender_ = kernel::actor::ActorImpl::self();
  69   return res;
  70 }
  71
  72 CommPtr Comm::sendto_init(Host* from, Host* to)
  73 {
  74   auto res = Comm::sendto_init()->set_source(from)->set_destination(to);
  75   res->set_state(State::STARTING);
  76   return res;
  77 }
  78
  79 CommPtr Comm::sendto_async(Host* from, Host* to, uint64_t simulated_size_in_bytes)
  80 {
  81   return Comm::sendto_init()->set_payload_size(simulated_size_in_bytes)->set_source(from)->set_destination(to);
  82 }
  83
  84 void Comm::sendto(Host* from, Host* to, uint64_t simulated_size_in_bytes)
  85 {
  86   sendto_async(from, to, simulated_size_in_bytes)->wait();
  87 }
  88
  89 CommPtr Comm::set_source(Host* from)
  90 {
  91   xbt_assert(state_ == State::INITED || state_ == State::STARTING,
  92              "Cannot change the source of a Comm once it's started (state: %s)", to_c_str(state_));
  93   boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_source(from);
  94   // Setting 'source' may allow to start the activity, let's try
  95   if (state_ == State::STARTING && remains_ <= 0)
  96     XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
  97   else
  98     vetoable_start();
  99
 100   return this;
 101 }
 102 Host* Comm::get_source() const
 103 {
 104   return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_source() : nullptr;
 105 }
 106
 107 CommPtr Comm::set_destination(Host* to)
 108 {
 109   xbt_assert(state_ == State::INITED || state_ == State::STARTING,
 110              "Cannot change the destination of a Comm once it's started (state: %s)", to_c_str(state_));
 111   boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_destination(to);
 112   // Setting 'destination' may allow to start the activity, let's try
 113   if (state_ == State::STARTING && remains_ <= 0)
 114     XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
 115   else
 116     vetoable_start();
 117
 118   return this;
 119 }
 120
 121 Host* Comm::get_destination() const
 122 {
 123   return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_destination() : nullptr;
 124 }
 125
 126 CommPtr Comm::set_rate(double rate)
 127 {
 128   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 129              __FUNCTION__);
 130   rate_ = rate;
 131   return this;
 132 }
 133
 134 CommPtr Comm::set_mailbox(Mailbox* mailbox)
 135 {
 136   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 137              __FUNCTION__);
 138   mailbox_ = mailbox;
 139   return this;
 140 }
 141
 142 CommPtr Comm::set_src_data(void* buff)
 143 {
 144   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 145              __FUNCTION__);
 146   xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
 147   src_buff_ = buff;
 148   return this;
 149 }
 150
 151 CommPtr Comm::set_src_data_size(size_t size)
 152 {
 153   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 154              __FUNCTION__);
 155   src_buff_size_ = size;
 156   return this;
 157 }
 158
 159 CommPtr Comm::set_src_data(void* buff, size_t size)
 160 {
 161   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 162              __FUNCTION__);
 163
 164   xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
 165   src_buff_      = buff;
 166   src_buff_size_ = size;
 167   return this;
 168 }
 169
 170 CommPtr Comm::set_dst_data(void** buff)
 171 {
 172   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 173              __FUNCTION__);
 174   xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
 175   dst_buff_ = buff;
 176   return this;
 177 }
 178
 179 CommPtr Comm::set_dst_data(void** buff, size_t size)
 180 {
 181   xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
 182              __FUNCTION__);
 183
 184   xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
 185   dst_buff_      = buff;
 186   dst_buff_size_ = size;
 187   return this;
 188 }
 189
 190 CommPtr Comm::set_payload_size(uint64_t bytes)
 191 {
 192   Activity::set_remaining(bytes);
 193   if (pimpl_) {
 194     boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_size(bytes);
 195   }
 196   return this;
 197 }
 198
 199 Actor* Comm::get_sender() const
 200 {
 201   kernel::actor::ActorImplPtr sender = nullptr;
 202   if (pimpl_)
 203     sender = boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->src_actor_;
 204   return sender ? sender->get_ciface() : nullptr;
 205 }
 206
 207 bool Comm::is_assigned() const
 208 {
 209   return (pimpl_ && boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->is_assigned()) ||
 210          mailbox_ != nullptr;
 211 }
 212
 213 Comm* Comm::start()
 214 {
 215   xbt_assert(get_state() == State::INITED || get_state() == State::STARTING,
 216              "You cannot use %s() once your communication started (not implemented)", __FUNCTION__);
 217   if (get_source() != nullptr || get_destination() != nullptr) {
 218     xbt_assert(is_assigned(), "When either from_ or to_ is specified, both must be.");
 219     xbt_assert(src_buff_ == nullptr && dst_buff_ == nullptr,
 220                "Direct host-to-host communications cannot carry any data.");
 221     XBT_DEBUG("host-to-host Comm. Pimpl already created and set, just start it.");
 222     kernel::actor::simcall_answered([this] {
 223       pimpl_->set_state(kernel::activity::State::READY);
 224       boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->start();
 225     });
 226   } else if (src_buff_ != nullptr) { // Sender side
 227     on_send(*this);
 228     kernel::actor::CommIsendSimcall observer{sender_,
 229                                              mailbox_->get_impl(),
 230                                              remains_,
 231                                              rate_,
 232                                              static_cast<unsigned char*>(src_buff_),
 233                                              src_buff_size_,
 234                                              match_fun_,
 235                                              clean_fun_,
 236                                              copy_data_function_,
 237                                              get_data<void>(),
 238                                              detached_};
 239     pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::isend(&observer); },
 240                                              &observer);
 241   } else if (dst_buff_ != nullptr) { // Receiver side
 242     xbt_assert(not detached_, "Receive cannot be detached");
 243     on_recv(*this);
 244     kernel::actor::CommIrecvSimcall observer{receiver_,
 245                                              mailbox_->get_impl(),
 246                                              static_cast<unsigned char*>(dst_buff_),
 247                                              &dst_buff_size_,
 248                                              match_fun_,
 249                                              copy_data_function_,
 250                                              get_data<void>(),
 251                                              rate_};
 252     pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::irecv(&observer); },
 253                                              &observer);
 254   } else {
 255     xbt_die("Cannot start a communication before specifying whether we are the sender or the receiver");
 256   }
 257
 258   if (suspended_)
 259     pimpl_->suspend();
 260
 261   if (not detached_) {
 262     pimpl_->set_iface(this);
 263     pimpl_->set_actor(sender_);
 264   }
 265
 266   state_ = State::STARTED;
 267   return this;
 268 }
 269
 270 Comm* Comm::detach()
 271 {
 272   xbt_assert(state_ == State::INITED || state_ == State::STARTING,
 273              "You cannot use %s() once your communication is %s (not implemented)", __FUNCTION__, get_state_str());
 274   xbt_assert(dst_buff_ == nullptr && dst_buff_size_ == 0, "You can only detach sends, not recvs");
 275   detached_ = true;
 276   vetoable_start();
 277   return this;
 278 }
 279
 280 ssize_t Comm::test_any(const std::vector<CommPtr>& comms)
 281 {
 282   std::vector<ActivityPtr> activities;
 283   for (const auto& comm : comms)
 284     activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
 285   return Activity::test_any(activities);
 286 }
 287
 288 /** @brief Block the calling actor until the communication is finished, or until timeout
 289  *
 290  * On timeout, an exception is thrown and the communication is invalidated.
 291  *
 292  * @param timeout the amount of seconds to wait for the comm termination.
 293  *                Negative values denote infinite wait times. 0 as a timeout returns immediately. */
 294 Comm* Comm::wait_for(double timeout)
 295 {
 296   XBT_DEBUG("Calling Comm::wait_for with state %s", get_state_str());
 297   kernel::actor::ActorImpl* issuer = nullptr;
 298   switch (state_) {
 299     case State::FINISHED:
 300       break;
 301     case State::FAILED:
 302       throw NetworkFailureException(XBT_THROW_POINT, "Cannot wait for a failed communication");
 303     case State::INITED:
 304     case State::STARTING: // It's not started yet. Do it in one simcall if it's a regular communication
 305       if (get_source() != nullptr || get_destination() != nullptr) {
 306         return vetoable_start()->wait_for(timeout); // In the case of host2host comm, do it in two simcalls
 307       } else if (src_buff_ != nullptr) {
 308         on_send(*this);
 309         simcall_comm_send(sender_, mailbox_->get_impl(), remains_, rate_, src_buff_, src_buff_size_, match_fun_,
 310                           copy_data_function_, get_data<void>(), timeout);
 311
 312       } else { // Receiver
 313         on_recv(*this);
 314         simcall_comm_recv(receiver_, mailbox_->get_impl(), dst_buff_, &dst_buff_size_, match_fun_, copy_data_function_,
 315                           get_data<void>(), timeout, rate_);
 316       }
 317       break;
 318     case State::STARTED:
 319       try {
 320         issuer = kernel::actor::ActorImpl::self();
 321         kernel::actor::ActivityWaitSimcall observer{issuer, pimpl_.get(), timeout};
 322         if (kernel::actor::simcall_blocking(
 323                 [&observer] { observer.get_activity()->wait_for(observer.get_issuer(), observer.get_timeout()); },
 324                 &observer)) {
 325           throw TimeoutException(XBT_THROW_POINT, "Timeouted");
 326         }
 327       } catch (const NetworkFailureException& e) {
 328         issuer->simcall_.observer_ = nullptr; // Comm failed on network failure, reset the observer to nullptr
 329         complete(State::FAILED);
 330         e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
 331       }
 332       break;
 333
 334     case State::CANCELED:
 335       throw CancelException(XBT_THROW_POINT, "Communication canceled");
 336
 337     default:
 338       THROW_IMPOSSIBLE;
 339   }
 340   complete(State::FINISHED);
 341   return this;
 342 }
 343
 344 ssize_t Comm::wait_any_for(const std::vector<CommPtr>& comms, double timeout)
 345 {
 346   std::vector<ActivityPtr> activities;
 347   for (const auto& comm : comms)
 348     activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
 349   ssize_t changed_pos;
 350   try {
 351     changed_pos = Activity::wait_any_for(activities, timeout);
 352   } catch (const NetworkFailureException& e) {
 353     changed_pos = -1;
 354     for (auto c : comms) {
 355       if (c->pimpl_->get_state() == kernel::activity::State::FAILED) {
 356         c->complete(State::FAILED);
 357       }
 358     }
 359     e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
 360   }
 361   return changed_pos;
 362 }
 363
 364 void Comm::wait_all(const std::vector<CommPtr>& comms)
 365 {
 366   // TODO: this should be a simcall or something
 367   for (auto& comm : comms)
 368     comm->wait();
 369 }
 370
 371 size_t Comm::wait_all_for(const std::vector<CommPtr>& comms, double timeout)
 372 {
 373   if (timeout < 0.0) {
 374     wait_all(comms);
 375     return comms.size();
 376   }
 377
 378   double deadline = Engine::get_clock() + timeout;
 379   std::vector<CommPtr> waited_comm(1, nullptr);
 380   for (size_t i = 0; i < comms.size(); i++) {
 381     double wait_timeout = std::max(0.0, deadline - Engine::get_clock());
 382     waited_comm[0]      = comms[i];
 383     // Using wait_any_for() here (and not wait_for) because we don't want comms to be invalidated on timeout
 384     if (wait_any_for(waited_comm, wait_timeout) == -1) {
 385       XBT_DEBUG("Timeout (%g): i = %zu", wait_timeout, i);
 386       return i;
 387     }
 388   }
 389   return comms.size();
 390 }
 391 } // namespace s4u
 392 } // namespace simgrid
 393 /* **************************** Public C interface *************************** */
 394 void sg_comm_detach(sg_comm_t comm, void (*clean_function)(void*))
 395 {
 396   comm->detach(clean_function);
 397   comm->unref();
 398 }
 399 void sg_comm_unref(sg_comm_t comm)
 400 {
 401   comm->unref();
 402 }
 403 int sg_comm_test(sg_comm_t comm)
 404 {
 405   bool finished = comm->test();
 406   if (finished)
 407     comm->unref();
 408   return finished;
 409 }
 410
 411 sg_error_t sg_comm_wait(sg_comm_t comm)
 412 {
 413   return sg_comm_wait_for(comm, -1);
 414 }
 415
 416 sg_error_t sg_comm_wait_for(sg_comm_t comm, double timeout)
 417 {
 418   sg_error_t status = SG_OK;
 419
 420   simgrid::s4u::CommPtr s4u_comm(comm, false);
 421   try {
 422     s4u_comm->wait_for(timeout);
 423   } catch (const simgrid::TimeoutException&) {
 424     status = SG_ERROR_TIMEOUT;
 425   } catch (const simgrid::CancelException&) {
 426     status = SG_ERROR_CANCELED;
 427   } catch (const simgrid::NetworkFailureException&) {
 428     status = SG_ERROR_NETWORK;
 429   }
 430   return status;
 431 }
 432
 433 void sg_comm_wait_all(sg_comm_t* comms, size_t count)
 434 {
 435   sg_comm_wait_all_for(comms, count, -1);
 436 }
 437
 438 size_t sg_comm_wait_all_for(sg_comm_t* comms, size_t count, double timeout)
 439 {
 440   std::vector<simgrid::s4u::CommPtr> s4u_comms;
 441   for (size_t i = 0; i < count; i++)
 442     s4u_comms.emplace_back(comms[i], false);
 443
 444   size_t pos = simgrid::s4u::Comm::wait_all_for(s4u_comms, timeout);
 445   for (size_t i = pos; i < count; i++)
 446     s4u_comms[i]->add_ref();
 447   return pos;
 448 }
 449
 450 ssize_t sg_comm_wait_any(sg_comm_t* comms, size_t count)
 451 {
 452   return sg_comm_wait_any_for(comms, count, -1);
 453 }
 454
 455 ssize_t sg_comm_wait_any_for(sg_comm_t* comms, size_t count, double timeout)
 456 {
 457   std::vector<simgrid::s4u::CommPtr> s4u_comms;
 458   for (size_t i = 0; i < count; i++)
 459     s4u_comms.emplace_back(comms[i], false);
 460
 461   ssize_t pos = simgrid::s4u::Comm::wait_any_for(s4u_comms, timeout);
 462   for (size_t i = 0; i < count; i++) {
 463     if (pos != -1 && static_cast<size_t>(pos) != i)
 464       s4u_comms[i]->add_ref();
 465   }
 466   return pos;
 467 }