1 /* Copyright (c) 2006-2022. 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. */
6 //#include "src/msg/msg_private.hpp"
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>
15 #include "src/kernel/activity/CommImpl.hpp"
16 #include "src/kernel/actor/ActorImpl.hpp"
17 #include "src/kernel/actor/SimcallObserver.hpp"
19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(s4u_comm, s4u_activity, "S4U asynchronous communications");
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;
29 if (state_ == State::STARTED && not detached_ &&
30 (pimpl_ == nullptr || pimpl_->get_state() == kernel::activity::State::RUNNING)) {
31 XBT_INFO("Comm %p freed before its completion. Did you forget to detach it? (state: %s)", this, get_state_str());
32 if (pimpl_ != nullptr)
33 XBT_INFO("pimpl_->state: %s", pimpl_->get_state_str());
35 XBT_INFO("pimpl_ is null");
36 xbt_backtrace_display_current();
40 ssize_t Comm::wait_any_for(const std::vector<CommPtr>& comms, double timeout)
42 std::vector<ActivityPtr> activities;
43 for (const auto& comm : comms)
44 activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
47 changed_pos = Activity::wait_any_for(activities, timeout);
48 } catch (const NetworkFailureException& e) {
50 for (auto c : comms) {
51 if (c->pimpl_->get_state() == kernel::activity::State::FAILED) {
52 c->complete(State::FAILED);
55 e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
60 void Comm::wait_all(const std::vector<CommPtr>& comms)
62 // TODO: this should be a simcall or something
63 for (auto& comm : comms)
67 size_t Comm::wait_all_for(const std::vector<CommPtr>& comms, double timeout)
74 double deadline = Engine::get_clock() + timeout;
75 std::vector<CommPtr> waited_comm(1, nullptr);
76 for (size_t i = 0; i < comms.size(); i++) {
77 double wait_timeout = std::max(0.0, deadline - Engine::get_clock());
78 waited_comm[0] = comms[i];
79 // Using wait_any_for() here (and not wait_for) because we don't want comms to be invalidated on timeout
80 if (wait_any_for(waited_comm, wait_timeout) == -1) {
81 XBT_DEBUG("Timeout (%g): i = %zu", wait_timeout, i);
88 CommPtr Comm::set_source(Host* from)
90 xbt_assert(state_ == State::INITED || state_ == State::STARTING,
91 "Cannot change the source of a Comm once it's started (state: %s)", to_c_str(state_));
93 // Setting 'from_' may allow to start the activity, let's try
99 CommPtr Comm::set_destination(Host* to)
101 xbt_assert(state_ == State::INITED || state_ == State::STARTING,
102 "Cannot change the destination of a Comm once it's started (state: %s)", to_c_str(state_));
104 // Setting 'to_' may allow to start the activity, let's try
110 CommPtr Comm::set_rate(double rate)
112 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
118 CommPtr Comm::set_src_data(void* buff)
120 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
122 xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
127 CommPtr Comm::set_src_data_size(size_t size)
129 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
131 src_buff_size_ = size;
135 CommPtr Comm::set_src_data(void* buff, size_t size)
137 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
140 xbt_assert(dst_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
142 src_buff_size_ = size;
146 CommPtr Comm::set_dst_data(void** buff)
148 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
150 xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
154 void* Comm::get_dst_data()
159 size_t Comm::get_dst_data_size() const
161 return dst_buff_size_;
163 CommPtr Comm::set_dst_data(void** buff, size_t size)
165 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication started (not implemented)",
168 xbt_assert(src_buff_ == nullptr, "Cannot set the src and dst buffers at the same time");
170 dst_buff_size_ = size;
173 CommPtr Comm::set_payload_size(uint64_t bytes)
175 Activity::set_remaining(bytes);
179 CommPtr Comm::sendto_init()
181 CommPtr res(new Comm());
182 res->sender_ = kernel::actor::ActorImpl::self();
186 CommPtr Comm::sendto_init(Host* from, Host* to)
188 auto res = Comm::sendto_init();
195 CommPtr Comm::sendto_async(Host* from, Host* to, uint64_t simulated_size_in_bytes)
197 auto res = Comm::sendto_init(from, to)->set_payload_size(simulated_size_in_bytes);
198 res->vetoable_start();
202 void Comm::sendto(Host* from, Host* to, uint64_t simulated_size_in_bytes)
204 sendto_async(from, to, simulated_size_in_bytes)->wait();
209 xbt_assert(get_state() == State::INITED || get_state() == State::STARTING,
210 "You cannot use %s() once your communication started (not implemented)", __FUNCTION__);
211 if (from_ != nullptr || to_ != nullptr) {
212 xbt_assert(from_ != nullptr && to_ != nullptr, "When either from_ or to_ is specified, both must be.");
213 xbt_assert(src_buff_ == nullptr && dst_buff_ == nullptr,
214 "Direct host-to-host communications cannot carry any data.");
215 pimpl_ = kernel::actor::simcall([this] {
216 kernel::activity::CommImplPtr res(new kernel::activity::CommImpl(this->from_, this->to_, this->get_remaining()));
221 } else if (src_buff_ != nullptr) { // Sender side
223 kernel::actor::CommIsendSimcall observer{sender_,
224 mailbox_->get_impl(),
227 static_cast<unsigned char*>(src_buff_),
234 pimpl_ = kernel::actor::simcall([&observer] { return kernel::activity::CommImpl::isend(&observer); }, &observer);
235 } else if (dst_buff_ != nullptr) { // Receiver side
236 xbt_assert(not detached_, "Receive cannot be detached");
238 kernel::actor::CommIrecvSimcall observer{receiver_,
239 mailbox_->get_impl(),
240 static_cast<unsigned char*>(dst_buff_),
246 pimpl_ = kernel::actor::simcall([&observer] { return kernel::activity::CommImpl::irecv(&observer); }, &observer);
248 xbt_die("Cannot start a communication before specifying whether we are the sender or the receiver");
255 pimpl_->set_iface(this);
256 pimpl_->set_actor(sender_);
259 state_ = State::STARTED;
263 /** @brief Block the calling actor until the communication is finished, or until timeout
265 * On timeout, an exception is thrown and the communication is invalidated.
267 * @param timeout the amount of seconds to wait for the comm termination.
268 * Negative values denote infinite wait times. 0 as a timeout returns immediately. */
269 Comm* Comm::wait_for(double timeout)
271 XBT_DEBUG("Calling Comm::wait_for with state %s", get_state_str());
272 kernel::actor::ActorImpl* issuer = nullptr;
274 case State::FINISHED:
277 throw NetworkFailureException(XBT_THROW_POINT, "Cannot wait for a failed communication");
279 case State::STARTING: // It's not started yet. Do it in one simcall if it's a regular communication
280 if (from_ != nullptr || to_ != nullptr) {
281 return vetoable_start()->wait_for(timeout); // In the case of host2host comm, do it in two simcalls
282 } else if (src_buff_ != nullptr) {
284 simcall_comm_send(sender_, mailbox_->get_impl(), remains_, rate_, src_buff_, src_buff_size_, match_fun_,
285 copy_data_function_, get_data<void>(), timeout);
289 simcall_comm_recv(receiver_, mailbox_->get_impl(), dst_buff_, &dst_buff_size_, match_fun_, copy_data_function_,
290 get_data<void>(), timeout, rate_);
295 issuer = kernel::actor::ActorImpl::self();
296 kernel::actor::ActivityWaitSimcall observer{issuer, pimpl_.get(), timeout};
297 if (kernel::actor::simcall_blocking(
298 [&observer] { observer.get_activity()->wait_for(observer.get_issuer(), observer.get_timeout()); },
300 throw TimeoutException(XBT_THROW_POINT, "Timeouted");
302 } catch (const NetworkFailureException& e) {
303 issuer->simcall_.observer_ = nullptr; // Comm failed on network failure, reset the observer to nullptr
304 complete(State::FAILED);
305 e.rethrow_nested(XBT_THROW_POINT, boost::core::demangle(typeid(e).name()) + " raised in kernel mode.");
309 case State::CANCELED:
310 throw CancelException(XBT_THROW_POINT, "Communication canceled");
315 complete(State::FINISHED);
319 ssize_t Comm::test_any(const std::vector<CommPtr>& comms)
321 std::vector<ActivityPtr> activities;
322 for (const auto& comm : comms)
323 activities.push_back(boost::dynamic_pointer_cast<Activity>(comm));
324 return Activity::test_any(activities);
329 xbt_assert(state_ == State::INITED, "You cannot use %s() once your communication is %s (not implemented)",
330 __FUNCTION__, get_state_str());
331 xbt_assert(dst_buff_ == nullptr && dst_buff_size_ == 0, "You can only detach sends, not recvs");
337 Mailbox* Comm::get_mailbox() const
342 Actor* Comm::get_sender() const
344 kernel::actor::ActorImplPtr sender = nullptr;
346 sender = boost::static_pointer_cast<kernel::activity::CommImpl>(pimpl_)->src_actor_;
347 return sender ? sender->get_ciface() : nullptr;
350 CommPtr Comm::set_copy_data_callback(void (*callback)(kernel::activity::CommImpl*, void*, size_t))
352 copy_data_function_ = callback;
355 void Comm::copy_buffer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
357 XBT_DEBUG("Copy the data over");
358 memcpy(comm->dst_buff_, buff, buff_size);
359 if (comm->detached()) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the
360 // original buffer available to the application ASAP
362 comm->src_buff_ = nullptr;
366 void Comm::copy_pointer_callback(kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
368 xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
369 *(void**)(comm->dst_buff_) = buff;
373 } // namespace simgrid
374 /* **************************** Public C interface *************************** */
375 void sg_comm_detach(sg_comm_t comm, void (*clean_function)(void*))
377 comm->detach(clean_function);
380 void sg_comm_unref(sg_comm_t comm)
384 int sg_comm_test(sg_comm_t comm)
386 bool finished = comm->test();
392 sg_error_t sg_comm_wait(sg_comm_t comm)
394 return sg_comm_wait_for(comm, -1);
397 sg_error_t sg_comm_wait_for(sg_comm_t comm, double timeout)
399 sg_error_t status = SG_OK;
401 simgrid::s4u::CommPtr s4u_comm(comm, false);
403 s4u_comm->wait_for(timeout);
404 } catch (const simgrid::TimeoutException&) {
405 status = SG_ERROR_TIMEOUT;
406 } catch (const simgrid::CancelException&) {
407 status = SG_ERROR_CANCELED;
408 } catch (const simgrid::NetworkFailureException&) {
409 status = SG_ERROR_NETWORK;
414 void sg_comm_wait_all(sg_comm_t* comms, size_t count)
416 sg_comm_wait_all_for(comms, count, -1);
419 size_t sg_comm_wait_all_for(sg_comm_t* comms, size_t count, double timeout)
421 std::vector<simgrid::s4u::CommPtr> s4u_comms;
422 for (size_t i = 0; i < count; i++)
423 s4u_comms.emplace_back(comms[i], false);
425 size_t pos = simgrid::s4u::Comm::wait_all_for(s4u_comms, timeout);
426 for (size_t i = pos; i < count; i++)
427 s4u_comms[i]->add_ref();
431 ssize_t sg_comm_wait_any(sg_comm_t* comms, size_t count)
433 return sg_comm_wait_any_for(comms, count, -1);
436 ssize_t sg_comm_wait_any_for(sg_comm_t* comms, size_t count, double timeout)
438 std::vector<simgrid::s4u::CommPtr> s4u_comms;
439 for (size_t i = 0; i < count; i++)
440 s4u_comms.emplace_back(comms[i], false);
442 ssize_t pos = simgrid::s4u::Comm::wait_any_for(s4u_comms, timeout);
443 for (size_t i = 0; i < count; i++) {
444 if (pos != -1 && static_cast<size_t>(pos) != i)
445 s4u_comms[i]->add_ref();