1 /* Copyright (c) 2006-2023. The SimGrid Team. All rights reserved. */
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. */
7 #include <simgrid/Exception.hpp>
8 #include <simgrid/comm.h>
9 #include <simgrid/s4u/Comm.hpp>
10 #include <simgrid/s4u/Engine.hpp>
11 #include <simgrid/s4u/Mailbox.hpp>
13 #include "src/kernel/activity/CommImpl.hpp"
14 #include "src/kernel/actor/ActorImpl.hpp"
15 #include "src/kernel/actor/SimcallObserver.hpp"
16 #include "src/mc/mc.h"
17 #include "src/mc/mc_replay.hpp"
19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(s4u_comm, s4u_activity, "S4U asynchronous communications");
21 namespace simgrid::s4u {
22 xbt::signal<void(Comm const&)> Comm::on_send;
23 xbt::signal<void(Comm const&)> Comm::on_recv;
24 xbt::signal<void(Comm const&)> Comm::on_start;
26 CommPtr Comm::set_copy_data_callback(const std::function<void(kernel::activity::CommImpl*, void*, size_t)>& callback)
28 copy_data_function_ = callback;
32 void Comm::copy_buffer_callback(kernel::activity::CommImpl* comm, void* buff,
33 size_t buff_size) // XBT_ATTRIB_DEPRECATED_v337
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
40 comm->src_buff_ = nullptr;
44 void Comm::copy_pointer_callback(kernel::activity::CommImpl* comm, void* buff,
45 size_t buff_size) // XBT_ATTRIB_DEPRECATED_v337
47 xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
48 *(void**)(comm->dst_buff_) = buff;
53 if (state_ == State::STARTED && not detached_ &&
54 (pimpl_ == nullptr || pimpl_->get_state() == kernel::activity::State::RUNNING)) {
55 XBT_INFO("Comm %p freed before its completion. Did you forget to detach it? (state: %s)", this, get_state_str());
56 if (pimpl_ != nullptr)
57 XBT_INFO("pimpl_->state: %s", pimpl_->get_state_str());
59 XBT_INFO("pimpl_ is null");
60 xbt_backtrace_display_current();
64 void Comm::send(kernel::actor::ActorImpl* sender, const Mailbox* mbox, double task_size, double rate, void* src_buff,
66 const std::function<bool(void*, void*, simgrid::kernel::activity::CommImpl*)>& match_fun,
67 const std::function<void(simgrid::kernel::activity::CommImpl*, void*, size_t)>& copy_data_fun,
68 void* data, double timeout)
70 /* checking for infinite values */
71 xbt_assert(std::isfinite(task_size), "task_size is not finite!");
72 xbt_assert(std::isfinite(rate), "rate is not finite!");
73 xbt_assert(std::isfinite(timeout), "timeout is not finite!");
75 xbt_assert(mbox, "No rendez-vous point defined for send");
77 if (MC_is_active() || MC_record_replay_is_active()) {
78 /* the model-checker wants two separate simcalls, and wants comm to be nullptr during the simcall */
79 simgrid::kernel::activity::ActivityImplPtr comm = nullptr;
81 simgrid::kernel::actor::CommIsendSimcall send_observer{
82 sender, mbox->get_impl(), task_size, rate, static_cast<unsigned char*>(src_buff), src_buff_size, match_fun,
83 nullptr, copy_data_fun, data, false};
84 comm = simgrid::kernel::actor::simcall_answered(
85 [&send_observer] { return simgrid::kernel::activity::CommImpl::isend(&send_observer); }, &send_observer);
87 if (simgrid::kernel::actor::ActivityWaitSimcall wait_observer{sender, comm.get(), timeout};
88 simgrid::kernel::actor::simcall_blocking(
90 wait_observer.get_activity()->wait_for(wait_observer.get_issuer(), wait_observer.get_timeout());
93 throw simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted");
97 simgrid::kernel::actor::CommIsendSimcall observer(sender, mbox->get_impl(), task_size, rate,
98 static_cast<unsigned char*>(src_buff), src_buff_size, match_fun,
99 nullptr, copy_data_fun, data, false);
100 simgrid::kernel::actor::simcall_blocking([&observer, timeout] {
101 simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::isend(&observer);
102 comm->wait_for(observer.get_issuer(), timeout);
107 void Comm::recv(kernel::actor::ActorImpl* receiver, const Mailbox* mbox, void* dst_buff, size_t* dst_buff_size,
108 const std::function<bool(void*, void*, simgrid::kernel::activity::CommImpl*)>& match_fun,
109 const std::function<void(simgrid::kernel::activity::CommImpl*, void*, size_t)>& copy_data_fun,
110 void* data, double timeout, double rate)
112 xbt_assert(std::isfinite(timeout), "timeout is not finite!");
113 xbt_assert(mbox, "No rendez-vous point defined for recv");
115 if (MC_is_active() || MC_record_replay_is_active()) {
116 /* the model-checker wants two separate simcalls, and wants comm to be nullptr during the simcall */
117 simgrid::kernel::activity::ActivityImplPtr comm = nullptr;
119 simgrid::kernel::actor::CommIrecvSimcall observer{receiver,
121 static_cast<unsigned char*>(dst_buff),
127 comm = simgrid::kernel::actor::simcall_answered(
128 [&observer] { return simgrid::kernel::activity::CommImpl::irecv(&observer); }, &observer);
130 if (simgrid::kernel::actor::ActivityWaitSimcall wait_observer{receiver, comm.get(), timeout};
131 simgrid::kernel::actor::simcall_blocking(
133 wait_observer.get_activity()->wait_for(wait_observer.get_issuer(), wait_observer.get_timeout());
136 throw simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted");
140 simgrid::kernel::actor::CommIrecvSimcall observer(receiver, mbox->get_impl(), static_cast<unsigned char*>(dst_buff),
141 dst_buff_size, match_fun, copy_data_fun, data, rate);
142 simgrid::kernel::actor::simcall_blocking([&observer, timeout] {
143 simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::irecv(&observer);
144 comm->wait_for(observer.get_issuer(), timeout);
149 CommPtr Comm::sendto_init()
151 CommPtr res(new Comm());
152 res->pimpl_ = kernel::activity::CommImplPtr(new kernel::activity::CommImpl());
153 boost::static_pointer_cast<kernel::activity::CommImpl>(res->pimpl_)->detach();
154 res->sender_ = kernel::actor::ActorImpl::self();
158 CommPtr Comm::sendto_init(Host* from, Host* to)
160 auto res = Comm::sendto_init()->set_source(from)->set_destination(to);
161 res->set_state(State::STARTING);
165 CommPtr Comm::sendto_async(Host* from, Host* to, uint64_t simulated_size_in_bytes)
167 return Comm::sendto_init()->set_payload_size(simulated_size_in_bytes)->set_source(from)->set_destination(to);
170 void Comm::sendto(Host* from, Host* to, uint64_t simulated_size_in_bytes)
172 sendto_async(from, to, simulated_size_in_bytes)->wait();
175 CommPtr Comm::set_source(Host* from)
177 xbt_assert(state_ == State::INITED || state_ == State::STARTING,
178 "Cannot change the source of a Comm once it's started (state: %s)", to_c_str(state_));
179 boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_source(from);
180 // Setting 'source' may allow to start the activity, let's try
181 if (state_ == State::STARTING && remains_ <= 0)
182 XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
188 Host* Comm::get_source() const
190 return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_source() : nullptr;
193 CommPtr Comm::set_destination(Host* to)
195 xbt_assert(state_ == State::INITED || state_ == State::STARTING,
196 "Cannot change the destination of a Comm once it's started (state: %s)", to_c_str(state_));
197 boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_destination(to);
198 // Setting 'destination' may allow to start the activity, let's try
199 if (state_ == State::STARTING && remains_ <= 0)
200 XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
207 Host* Comm::get_destination() const
209 return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_destination() : nullptr;
212 CommPtr Comm::set_rate(double rate)
214 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
220 CommPtr Comm::set_mailbox(Mailbox* mailbox)
222 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
228 CommPtr Comm::set_src_data(void* buff)
230 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
232 xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
237 CommPtr Comm::set_src_data_size(size_t size)
239 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
241 src_buff_size_ = size;
245 CommPtr Comm::set_src_data(void* buff, size_t size)
247 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
250 xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
252 src_buff_size_ = size;
256 CommPtr Comm::set_dst_data(void** buff)
258 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
260 xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
265 CommPtr Comm::set_dst_data(void** buff, size_t size)
267 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
270 xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
272 dst_buff_size_ = size;
276 CommPtr Comm::set_payload_size(uint64_t bytes)
278 set_remaining(bytes);
280 boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_size(bytes);
285 Actor* Comm::get_sender() const
287 kernel::actor::ActorImplPtr sender = nullptr;
289 sender = boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->src_actor_;
290 return sender ? sender->get_ciface() : nullptr;
293 Actor* Comm::get_receiver() const
295 kernel::actor::ActorImplPtr receiver = nullptr;
297 receiver = boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->dst_actor_;
298 return receiver ? receiver->get_ciface() : nullptr;
301 bool Comm::is_assigned() const
303 return (pimpl_ && boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->is_assigned()) ||
307 Comm* Comm::do_start()
309 xbt_assert(get_state() == State::INITED || get_state() == State::STARTING,
310 "You cannot use %s() once your communication started (not implemented)", __FUNCTION__);
311 if (get_source() != nullptr || get_destination() != nullptr) {
312 xbt_assert(is_assigned(), "When either from_ or to_ is specified, both must be.");
313 xbt_assert(src_buff_ == nullptr && dst_buff_ == nullptr,
314 "Direct host-to-host communications cannot carry any data.");
315 XBT_DEBUG("host-to-host Comm. Pimpl already created and set, just start it.");
318 kernel::actor::simcall_answered([this] {
319 pimpl_->set_state(kernel::activity::State::READY);
320 boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->start();
322 } else if (src_buff_ != nullptr) { // Sender side
324 kernel::actor::CommIsendSimcall observer{sender_,
325 mailbox_->get_impl(),
328 static_cast<unsigned char*>(src_buff_),
335 pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::isend(&observer); },
337 } else if (dst_buff_ != nullptr) { // Receiver side
338 xbt_assert(not detached_, "Receive cannot be detached");
340 kernel::actor::CommIrecvSimcall observer{receiver_,
341 mailbox_->get_impl(),
342 static_cast<unsigned char*>(dst_buff_),
348 pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::irecv(&observer); },
351 xbt_die("Cannot start a communication before specifying whether we are the sender or the receiver");
358 pimpl_->set_iface(this);
359 pimpl_->set_actor(sender_);
360 // Only throw the signal when both sides are here and the status is READY
361 if (pimpl_->get_state() != kernel::activity::State::WAITING)
365 state_ = State::STARTED;
371 xbt_assert(state_ == State::INITED || state_ == State::STARTING,
372 "You cannot use %s() once your communication is %s (not implemented)", __FUNCTION__, get_state_str());
373 xbt_assert(dst_buff_ == nullptr && dst_buff_size_ == 0, "You can only detach sends, not recvs");
379 ssize_t Comm::test_any(const std::vector<CommPtr>& comms)
381 std::vector<ActivityPtr> activities;
382 for (const auto& comm : comms)
383 activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
384 return Activity::test_any(activities);
387 /** @brief Block the calling actor until the communication is finished, or until timeout
389 * On timeout, an exception is thrown and the communication is invalidated.
391 * @param timeout the amount of seconds to wait for the comm termination.
392 * Negative values denote infinite wait times. 0 as a timeout returns immediately. */
393 Comm* Comm::wait_for(double timeout)
395 XBT_DEBUG("Calling Comm::wait_for with state %s", get_state_str());
396 kernel::actor::ActorImpl* issuer = nullptr;
398 case State::FINISHED:
401 throw NetworkFailureException(XBT_THROW_POINT, "Cannot wait for a failed communication");
403 case State::STARTING: // It's not started yet. Do it in one simcall if it's a regular communication
404 if (get_source() != nullptr || get_destination() != nullptr) {
405 return start()->wait_for(timeout); // In the case of host2host comm, do it in two simcalls
406 } else if (src_buff_ != nullptr) {
408 send(sender_, mailbox_, remains_, rate_, src_buff_, src_buff_size_, match_fun_, copy_data_function_,
409 get_data<void>(), timeout);
413 recv(receiver_, mailbox_, dst_buff_, &dst_buff_size_, match_fun_, copy_data_function_, get_data<void>(),
419 issuer = kernel::actor::ActorImpl::self();
420 kernel::actor::ActivityWaitSimcall observer{issuer, pimpl_.get(), timeout};
421 if (kernel::actor::simcall_blocking(
422 [&observer] { observer.get_activity()->wait_for(observer.get_issuer(), observer.get_timeout()); },
424 throw TimeoutException(XBT_THROW_POINT, "Timeouted");
426 } catch (const NetworkFailureException& e) {
427 issuer->simcall_.observer_ = nullptr; // Comm failed on network failure, reset the observer to nullptr
428 complete(State::FAILED);
429 e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
433 case State::CANCELED:
434 throw CancelException(XBT_THROW_POINT, "Communication canceled");
439 complete(State::FINISHED);
443 ssize_t Comm::wait_any_for(const std::vector<CommPtr>& comms, double timeout)
445 std::vector<ActivityPtr> activities;
446 for (const auto& comm : comms)
447 activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
450 changed_pos = Activity::wait_any_for(activities, timeout);
451 } catch (const NetworkFailureException& e) {
453 for (auto c : comms) {
454 if (c->pimpl_->get_state() == kernel::activity::State::FAILED) {
455 c->complete(State::FAILED);
458 e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
463 void Comm::wait_all(const std::vector<CommPtr>& comms)
465 // TODO: this should be a simcall or something
466 for (auto& comm : comms)
470 size_t Comm::wait_all_for(const std::vector<CommPtr>& comms, double timeout)
477 double deadline = Engine::get_clock() + timeout;
478 std::vector<CommPtr> waited_comm(1, nullptr);
479 for (size_t i = 0; i < comms.size(); i++) {
480 double wait_timeout = std::max(0.0, deadline - Engine::get_clock());
481 waited_comm[0] = comms[i];
482 // Using wait_any_for() here (and not wait_for) because we don't want comms to be invalidated on timeout
483 if (wait_any_for(waited_comm, wait_timeout) == -1) {
484 XBT_DEBUG("Timeout (%g): i = %zu", wait_timeout, i);
490 } // namespace simgrid::s4u
491 /* **************************** Public C interface *************************** */
492 void sg_comm_detach(sg_comm_t comm, void (*clean_function)(void*))
494 comm->detach(clean_function);
497 void sg_comm_unref(sg_comm_t comm)
501 int sg_comm_test(sg_comm_t comm)
503 bool finished = comm->test();
509 sg_error_t sg_comm_wait(sg_comm_t comm)
511 return sg_comm_wait_for(comm, -1);
514 sg_error_t sg_comm_wait_for(sg_comm_t comm, double timeout)
516 sg_error_t status = SG_OK;
518 simgrid::s4u::CommPtr s4u_comm(comm, false);
520 s4u_comm->wait_for(timeout);
521 } catch (const simgrid::TimeoutException&) {
522 status = SG_ERROR_TIMEOUT;
523 } catch (const simgrid::CancelException&) {
524 status = SG_ERROR_CANCELED;
525 } catch (const simgrid::NetworkFailureException&) {
526 status = SG_ERROR_NETWORK;
531 void sg_comm_wait_all(sg_comm_t* comms, size_t count)
533 sg_comm_wait_all_for(comms, count, -1);
536 size_t sg_comm_wait_all_for(sg_comm_t* comms, size_t count, double timeout)
538 std::vector<simgrid::s4u::CommPtr> s4u_comms;
539 for (size_t i = 0; i < count; i++)
540 s4u_comms.emplace_back(comms[i], false);
542 size_t pos = simgrid::s4u::Comm::wait_all_for(s4u_comms, timeout);
543 for (size_t i = pos; i < count; i++)
544 s4u_comms[i]->add_ref();
548 ssize_t sg_comm_wait_any(sg_comm_t* comms, size_t count)
550 return sg_comm_wait_any_for(comms, count, -1);
553 ssize_t sg_comm_wait_any_for(sg_comm_t* comms, size_t count, double timeout)
555 std::vector<simgrid::s4u::CommPtr> s4u_comms;
556 for (size_t i = 0; i < count; i++)
557 s4u_comms.emplace_back(comms[i], false);
559 ssize_t pos = simgrid::s4u::Comm::wait_any_for(s4u_comms, timeout);
560 for (size_t i = 0; i < count; i++) {
561 if (pos != -1 && static_cast<size_t>(pos) != i)
562 s4u_comms[i]->add_ref();