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/ActivitySet.hpp>
10 #include <simgrid/s4u/Comm.hpp>
11 #include <simgrid/s4u/Engine.hpp>
12 #include <simgrid/s4u/Mailbox.hpp>
14 #include "src/kernel/activity/CommImpl.hpp"
15 #include "src/kernel/actor/ActorImpl.hpp"
16 #include "src/kernel/actor/SimcallObserver.hpp"
17 #include "src/mc/mc.h"
18 #include "src/mc/mc_replay.hpp"
20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(s4u_comm, s4u_activity, "S4U asynchronous communications");
22 namespace simgrid::s4u {
23 xbt::signal<void(Comm const&)> Comm::on_send;
24 xbt::signal<void(Comm const&)> Comm::on_recv;
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_v338
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_v338
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),
83 src_buff_size, match_fun, nullptr, copy_data_fun, data,
85 comm = simgrid::kernel::actor::simcall_answered(
86 [&send_observer] { return simgrid::kernel::activity::CommImpl::isend(&send_observer); }, &send_observer);
88 if (simgrid::kernel::actor::ActivityWaitSimcall wait_observer{sender, comm.get(), timeout, "Wait"};
89 simgrid::kernel::actor::simcall_blocking(
91 wait_observer.get_activity()->wait_for(wait_observer.get_issuer(), wait_observer.get_timeout());
94 throw simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted");
98 simgrid::kernel::actor::CommIsendSimcall observer(sender, mbox->get_impl(), task_size, rate,
99 static_cast<unsigned char*>(src_buff), src_buff_size, match_fun,
100 nullptr, copy_data_fun, data, false, "Isend");
101 simgrid::kernel::actor::simcall_blocking([&observer, timeout] {
102 simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::isend(&observer);
103 comm->wait_for(observer.get_issuer(), timeout);
108 void Comm::recv(kernel::actor::ActorImpl* receiver, const Mailbox* mbox, void* dst_buff, size_t* dst_buff_size,
109 const std::function<bool(void*, void*, simgrid::kernel::activity::CommImpl*)>& match_fun,
110 const std::function<void(simgrid::kernel::activity::CommImpl*, void*, size_t)>& copy_data_fun,
111 void* data, double timeout, double rate)
113 xbt_assert(std::isfinite(timeout), "timeout is not finite!");
114 xbt_assert(mbox, "No rendez-vous point defined for recv");
116 if (MC_is_active() || MC_record_replay_is_active()) {
117 /* the model-checker wants two separate simcalls, and wants comm to be nullptr during the simcall */
118 simgrid::kernel::activity::ActivityImplPtr comm = nullptr;
120 simgrid::kernel::actor::CommIrecvSimcall observer{receiver,
122 static_cast<unsigned char*>(dst_buff),
129 comm = simgrid::kernel::actor::simcall_answered(
130 [&observer] { return simgrid::kernel::activity::CommImpl::irecv(&observer); }, &observer);
132 if (simgrid::kernel::actor::ActivityWaitSimcall wait_observer{receiver, comm.get(), timeout, "wait"};
133 simgrid::kernel::actor::simcall_blocking(
135 wait_observer.get_activity()->wait_for(wait_observer.get_issuer(), wait_observer.get_timeout());
138 throw simgrid::TimeoutException(XBT_THROW_POINT, "Timeouted");
142 simgrid::kernel::actor::CommIrecvSimcall observer(receiver, mbox->get_impl(), static_cast<unsigned char*>(dst_buff),
143 dst_buff_size, match_fun, copy_data_fun, data, rate, "Irecv");
144 simgrid::kernel::actor::simcall_blocking([&observer, timeout] {
145 simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::irecv(&observer);
146 comm->wait_for(observer.get_issuer(), timeout);
151 CommPtr Comm::sendto_init()
153 CommPtr res(new Comm());
154 res->pimpl_ = kernel::activity::CommImplPtr(new kernel::activity::CommImpl());
155 boost::static_pointer_cast<kernel::activity::CommImpl>(res->pimpl_)->detach();
156 res->sender_ = kernel::actor::ActorImpl::self();
160 CommPtr Comm::sendto_init(Host* from, Host* to)
162 auto res = Comm::sendto_init()->set_source(from)->set_destination(to);
163 res->set_state(State::STARTING);
167 CommPtr Comm::sendto_async(Host* from, Host* to, uint64_t simulated_size_in_bytes)
169 return Comm::sendto_init()->set_payload_size(simulated_size_in_bytes)->set_source(from)->set_destination(to);
172 void Comm::sendto(Host* from, Host* to, uint64_t simulated_size_in_bytes)
174 sendto_async(from, to, simulated_size_in_bytes)->wait();
177 CommPtr Comm::set_source(Host* from)
179 xbt_assert(state_ == State::INITED || state_ == State::STARTING,
180 "Cannot change the source of a Comm once it's started (state: %s)", to_c_str(state_));
181 boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_source(from);
182 // Setting 'source' may allow to start the activity, let's try
183 if (state_ == State::STARTING && remains_ <= 0)
184 XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
190 Host* Comm::get_source() const
192 return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_source() : nullptr;
195 CommPtr Comm::set_destination(Host* to)
197 xbt_assert(state_ == State::INITED || state_ == State::STARTING,
198 "Cannot change the destination of a Comm once it's started (state: %s)", to_c_str(state_));
199 boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_destination(to);
200 // Setting 'destination' may allow to start the activity, let's try
201 if (state_ == State::STARTING && remains_ <= 0)
202 XBT_DEBUG("This communication has a payload size of 0 byte. It cannot start yet");
209 Host* Comm::get_destination() const
211 return pimpl_ ? boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->get_destination() : nullptr;
214 CommPtr Comm::set_rate(double rate)
216 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
222 CommPtr Comm::set_mailbox(Mailbox* mailbox)
224 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
230 CommPtr Comm::set_src_data(void* buff)
232 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
234 xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
239 CommPtr Comm::set_src_data_size(size_t size)
241 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
243 src_buff_size_ = size;
247 CommPtr Comm::set_src_data(void* buff, size_t size)
249 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
252 xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
254 src_buff_size_ = size;
258 CommPtr Comm::set_dst_data(void** buff)
260 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
262 xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
267 CommPtr Comm::set_dst_data(void** buff, size_t size)
269 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
272 xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
274 dst_buff_size_ = size;
278 CommPtr Comm::set_payload_size(uint64_t bytes)
280 set_remaining(bytes);
282 boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->set_size(bytes);
287 void* Comm::get_payload() const
289 xbt_assert(get_state() == State::FINISHED,
290 "You can only retrieve the payload of a communication that gracefully terminated, but its state is %s.",
292 return static_cast<kernel::activity::CommImpl*>(pimpl_.get())->payload_;
295 Actor* Comm::get_sender() const
297 kernel::actor::ActorImplPtr sender = nullptr;
299 sender = boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->src_actor_;
300 return sender ? sender->get_ciface() : nullptr;
303 Actor* Comm::get_receiver() const
305 kernel::actor::ActorImplPtr receiver = nullptr;
307 receiver = boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->dst_actor_;
308 return receiver ? receiver->get_ciface() : nullptr;
311 bool Comm::is_assigned() const
313 return (pimpl_ && boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->is_assigned()) ||
317 Comm* Comm::do_start()
319 xbt_assert(get_state() == State::INITED || get_state() == State::STARTING,
320 "You cannot use %s() once your communication started (not implemented)", __func__);
322 auto myself = kernel::actor::ActorImpl::self();
324 if (get_source() != nullptr || get_destination() != nullptr) {
325 xbt_assert(is_assigned(), "When either from_ or to_ is specified, both must be.");
326 xbt_assert(src_buff_ == nullptr && dst_buff_ == nullptr,
327 "Direct host-to-host communications cannot carry any data.");
328 XBT_DEBUG("host-to-host Comm. Pimpl already created and set, just start it.");
329 kernel::actor::simcall_answered([this] {
330 pimpl_->set_state(kernel::activity::State::READY);
331 boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->start();
334 fire_on_this_start();
335 } else if (myself == sender_) {
338 kernel::actor::CommIsendSimcall observer{sender_,
339 mailbox_->get_impl(),
342 static_cast<unsigned char*>(src_buff_),
350 pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::isend(&observer); },
352 } else if (myself == receiver_) {
353 xbt_assert(not detached_, "Receive cannot be detached");
356 kernel::actor::CommIrecvSimcall observer{receiver_,
357 mailbox_->get_impl(),
358 static_cast<unsigned char*>(dst_buff_),
365 pimpl_ = kernel::actor::simcall_answered([&observer] { return kernel::activity::CommImpl::irecv(&observer); },
368 xbt_die("Cannot start a communication before specifying whether we are the sender or the receiver");
375 pimpl_->set_iface(this);
376 pimpl_->set_actor(sender_);
377 // Only throw the signal when both sides are here and the status is READY
378 if (pimpl_->get_state() != kernel::activity::State::WAITING) {
380 fire_on_this_start();
384 state_ = State::STARTED;
390 xbt_assert(state_ == State::INITED || state_ == State::STARTING,
391 "You cannot use %s() once your communication is %s (not implemented)", __func__, get_state_str());
392 xbt_assert(dst_buff_ == nullptr && dst_buff_size_ == 0, "You can only detach sends, not recvs");
398 ssize_t Comm::test_any(const std::vector<CommPtr>& comms) // XBT_ATTRIB_DEPRECATED_v339
400 std::vector<kernel::activity::ActivityImpl*> ractivities(comms.size());
401 std::transform(begin(comms), end(comms), begin(ractivities), [](const CommPtr& act) { return act->pimpl_.get(); });
403 kernel::actor::ActorImpl* issuer = kernel::actor::ActorImpl::self();
404 kernel::actor::ActivityTestanySimcall observer{issuer, ractivities, "test_any"};
405 ssize_t changed_pos = kernel::actor::simcall_answered(
407 return kernel::activity::ActivityImpl::test_any(observer.get_issuer(), observer.get_activities());
410 if (changed_pos != -1)
411 comms.at(changed_pos)->complete(State::FINISHED);
415 /** @brief Block the calling actor until the communication is finished, or until timeout
417 * On timeout, an exception is thrown and the communication is invalidated.
419 * @param timeout the amount of seconds to wait for the comm termination.
420 * Negative values denote infinite wait times. 0 as a timeout returns immediately. */
421 Comm* Comm::wait_for(double timeout)
423 XBT_DEBUG("Calling Comm::wait_for with state %s", get_state_str());
424 kernel::actor::ActorImpl* issuer = nullptr;
426 case State::FINISHED:
429 throw NetworkFailureException(XBT_THROW_POINT, "Cannot wait for a failed communication");
431 case State::STARTING: // It's not started yet. Do it in one simcall if it's a regular communication
432 if (get_source() != nullptr || get_destination() != nullptr) {
433 return start()->wait_for(timeout); // In the case of host2host comm, do it in two simcalls
434 } else if (src_buff_ != nullptr) {
437 send(sender_, mailbox_, remains_, rate_, src_buff_, src_buff_size_, match_fun_, copy_data_function_,
438 get_data<void>(), timeout);
443 recv(receiver_, mailbox_, dst_buff_, &dst_buff_size_, match_fun_, copy_data_function_, get_data<void>(),
449 issuer = kernel::actor::ActorImpl::self();
450 kernel::actor::ActivityWaitSimcall observer{issuer, pimpl_.get(), timeout, "Wait"};
451 if (kernel::actor::simcall_blocking(
452 [&observer] { observer.get_activity()->wait_for(observer.get_issuer(), observer.get_timeout()); },
454 throw TimeoutException(XBT_THROW_POINT, "Timeouted");
456 } catch (const NetworkFailureException& e) {
457 issuer->simcall_.observer_ = nullptr; // Comm failed on network failure, reset the observer to nullptr
458 complete(State::FAILED);
459 e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
463 case State::CANCELED:
464 throw CancelException(XBT_THROW_POINT, "Communication canceled");
469 complete(State::FINISHED);
473 ssize_t Comm::deprecated_wait_any_for(const std::vector<CommPtr>& comms, double timeout) // XBT_ATTRIB_DEPRECATED_v339
478 for (const auto& comm : comms)
481 auto* ret = set.wait_any_for(timeout).get();
482 for (size_t i = 0; i < comms.size(); i++)
483 if (comms[i].get() == ret)
486 } catch (TimeoutException& e) {
488 } catch (const NetworkFailureException& e) {
490 if (c->pimpl_->get_state() == kernel::activity::State::FAILED)
491 c->complete(State::FAILED);
493 e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
498 void Comm::wait_all(const std::vector<CommPtr>& comms) // XBT_ATTRIB_DEPRECATED_v339
500 // TODO: this should be a simcall or something
501 for (const auto& comm : comms)
505 size_t Comm::wait_all_for(const std::vector<CommPtr>& comms, double timeout) // XBT_ATTRIB_DEPRECATED_v339
508 for (const auto& comm : comms)
514 for (auto comm : comms)
516 set.wait_all_for(timeout);
520 } // namespace simgrid::s4u
521 /* **************************** Public C interface *************************** */
522 int sg_comm_isinstance(sg_activity_t acti)
524 return dynamic_cast<simgrid::s4u::Comm*>(acti) != nullptr;
527 void sg_comm_detach(sg_comm_t comm, void (*clean_function)(void*))
529 comm->detach(clean_function);
532 void sg_comm_unref(sg_comm_t comm)
536 int sg_comm_test(sg_comm_t comm)
538 bool finished = comm->test();
544 sg_error_t sg_comm_wait(sg_comm_t comm)
546 return sg_comm_wait_for(comm, -1);
549 sg_error_t sg_comm_wait_for(sg_comm_t comm, double timeout)
551 sg_error_t status = SG_OK;
553 simgrid::s4u::CommPtr s4u_comm(comm, false);
555 s4u_comm->wait_for(timeout);
556 } catch (const simgrid::TimeoutException&) {
557 status = SG_ERROR_TIMEOUT;
558 } catch (const simgrid::CancelException&) {
559 status = SG_ERROR_CANCELED;
560 } catch (const simgrid::NetworkFailureException&) {
561 status = SG_ERROR_NETWORK;
566 void sg_comm_wait_all(sg_comm_t* comms, size_t count)
568 simgrid::s4u::ActivitySet as;
569 for (size_t i = 0; i < count; i++)
575 ssize_t sg_comm_wait_any(sg_comm_t* comms, size_t count)
577 std::vector<simgrid::s4u::CommPtr> s4u_comms;
578 for (size_t i = 0; i < count; i++)
579 s4u_comms.emplace_back(comms[i], false);
581 ssize_t pos = simgrid::s4u::Comm::deprecated_wait_any_for(s4u_comms, -1);
582 for (size_t i = 0; i < count; i++) {
583 if (pos != -1 && static_cast<size_t>(pos) != i)
584 s4u_comms[i]->add_ref();
589 ssize_t sg_comm_wait_any_for(sg_comm_t* comms, size_t count, double timeout)
591 std::vector<simgrid::s4u::CommPtr> s4u_comms;
592 for (size_t i = 0; i < count; i++)
593 s4u_comms.emplace_back(comms[i], false);
595 ssize_t pos = simgrid::s4u::Comm::deprecated_wait_any_for(s4u_comms, timeout);
596 for (size_t i = 0; i < count; i++) {
597 if (pos != -1 && static_cast<size_t>(pos) != i)
598 s4u_comms[i]->add_ref();