X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/b875e185e4ea9d8e5aec292f061354e2edc46c34..9a6b1a23ce86f475df4c3dd63310c3484cebd956:/src/kernel/activity/CommImpl.cpp diff --git a/src/kernel/activity/CommImpl.cpp b/src/kernel/activity/CommImpl.cpp index 0191ebf65d..143ebf80b6 100644 --- a/src/kernel/activity/CommImpl.cpp +++ b/src/kernel/activity/CommImpl.cpp @@ -1,368 +1,81 @@ -/* Copyright (c) 2007-2021. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2022. The SimGrid Team. All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ +#include +#include +#include +#include + +#define SIMIX_H_NO_DEPRECATED_WARNING // avoid deprecation warning on include (remove with XBT_ATTRIB_DEPRECATED_v333) +#include + +#include "src/kernel/EngineImpl.hpp" #include "src/kernel/activity/CommImpl.hpp" -#include "simgrid/Exception.hpp" -#include "simgrid/kernel/resource/Action.hpp" -#include "simgrid/modelchecker.h" -#include "simgrid/s4u/Host.hpp" #include "src/kernel/activity/MailboxImpl.hpp" -#include "src/kernel/context/Context.hpp" +#include "src/kernel/actor/SimcallObserver.hpp" +#include "src/kernel/resource/CpuImpl.hpp" +#include "src/kernel/resource/NetworkModel.hpp" +#include "src/kernel/resource/StandardLinkImpl.hpp" #include "src/mc/mc_replay.hpp" -#include "src/surf/cpu_interface.hpp" -#include "src/surf/network_interface.hpp" -#include "src/surf/surf_interface.hpp" - -#include -XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix, "SIMIX network-related synchronization"); - -XBT_PRIVATE void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_actor_t src, smx_mailbox_t mbox, double task_size, - double rate, unsigned char* src_buff, size_t src_buff_size, - bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), - void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), - void* data, double timeout) -{ - simgrid::kernel::activity::ActivityImplPtr comm = simcall_HANDLER_comm_isend( - simcall, src, mbox, task_size, rate, src_buff, src_buff_size, match_fun, nullptr, copy_data_fun, data, false); - simcall->mc_value_ = 0; - simcall_HANDLER_comm_wait(simcall, static_cast(comm.get()), timeout); -} - -XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr simcall_HANDLER_comm_isend( - smx_simcall_t /*simcall*/, smx_actor_t src_proc, smx_mailbox_t mbox, double task_size, double rate, - unsigned char* src_buff, size_t src_buff_size, - bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), - void (*clean_fun)(void*), // used to free the synchro in case of problem after a detached send - void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), // used to copy data if not default one - void* data, bool detached) -{ - XBT_DEBUG("send from mailbox %p", mbox); - - /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */ - simgrid::kernel::activity::CommImplPtr this_comm( - new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::SEND)); - - /* Look for communication synchro matching our needs. We also provide a description of - * ourself so that the other side also gets a chance of choosing if it wants to match with us. - * - * If it is not found then push our communication into the rendez-vous point */ - simgrid::kernel::activity::CommImplPtr other_comm = - mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::RECEIVE, match_fun, data, this_comm, - /*done*/ false, /*remove_matching*/ true); - - if (not other_comm) { - other_comm = std::move(this_comm); - - if (mbox->permanent_receiver_ != nullptr) { - // this mailbox is for small messages, which have to be sent right now - other_comm->state_ = simgrid::kernel::activity::State::READY; - other_comm->dst_actor_ = mbox->permanent_receiver_.get(); - mbox->done_comm_queue_.push_back(other_comm); - XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get()); - - } else { - mbox->push(other_comm); - } - } else { - XBT_DEBUG("Receive already pushed"); - - other_comm->state_ = simgrid::kernel::activity::State::READY; - } - - if (detached) { - other_comm->detach(); - other_comm->clean_fun = clean_fun; - } else { - other_comm->clean_fun = nullptr; - src_proc->activities_.emplace_back(other_comm); - } - - /* Setup the communication synchro */ - other_comm->src_actor_ = src_proc; - other_comm->src_data_ = data; - (*other_comm).set_src_buff(src_buff, src_buff_size).set_size(task_size).set_rate(rate); - - other_comm->match_fun = match_fun; - other_comm->copy_data_fun = copy_data_fun; - if (MC_is_active() || MC_record_replay_is_active()) - other_comm->state_ = simgrid::kernel::activity::State::RUNNING; - else - other_comm->start(); +XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_network, kernel, "Kernel network-related synchronization"); - return (detached ? nullptr : other_comm); -} - -XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox, - unsigned char* dst_buff, size_t* dst_buff_size, - bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), - void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), - void* data, double timeout, double rate) +/******************************************************************************/ +/* SIMIX_comm_copy_data callbacks */ +/******************************************************************************/ +// XBT_ATTRIB_DEPRECATED_v333 +void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t)) { - simgrid::kernel::activity::ActivityImplPtr comm = simcall_HANDLER_comm_irecv( - simcall, receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate); - simcall->mc_value_ = 0; - simcall_HANDLER_comm_wait(simcall, static_cast(comm.get()), timeout); + simgrid::kernel::activity::CommImpl::set_copy_data_callback(callback); } -XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr -simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, unsigned char* dst_buff, - size_t* dst_buff_size, bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), - void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), void* data, - double rate) -{ - simgrid::kernel::activity::CommImplPtr this_synchro( - new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::RECEIVE)); - XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get()); - - simgrid::kernel::activity::CommImplPtr other_comm; - // communication already done, get it inside the list of completed comms - if (mbox->permanent_receiver_ != nullptr && not mbox->done_comm_queue_.empty()) { - XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication"); - // find a match in the list of already received comms - other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data, - this_synchro, /*done*/ true, - /*remove_matching*/ true); - // if not found, assume the receiver came first, register it to the mailbox in the classical way - if (not other_comm) { - XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request " - "into list"); - other_comm = std::move(this_synchro); - mbox->push(other_comm); - } else { - if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) { - XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get()); - other_comm->state_ = simgrid::kernel::activity::State::DONE; - other_comm->set_mailbox(nullptr); - } - } - } else { - /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */ - - /* Look for communication activity matching our needs. We also provide a description of - * ourself so that the other side also gets a chance of choosing if it wants to match with us. - * - * If it is not found then push our communication into the rendez-vous point */ - other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data, - this_synchro, /*done*/ false, - /*remove_matching*/ true); - - if (other_comm == nullptr) { - XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue_.size()); - other_comm = std::move(this_synchro); - mbox->push(other_comm); - } else { - XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get()); - - other_comm->state_ = simgrid::kernel::activity::State::READY; - } - receiver->activities_.emplace_back(other_comm); - } - - /* Setup communication synchro */ - other_comm->dst_actor_ = receiver; - other_comm->dst_data_ = data; - other_comm->set_dst_buff(dst_buff, dst_buff_size); - - if (rate > -1.0 && (other_comm->get_rate() < 0.0 || rate < other_comm->get_rate())) - other_comm->set_rate(rate); - - other_comm->match_fun = match_fun; - other_comm->copy_data_fun = copy_data_fun; - - if (MC_is_active() || MC_record_replay_is_active()) { - other_comm->state_ = simgrid::kernel::activity::State::RUNNING; - return other_comm; - } - other_comm->start(); - return other_comm; -} - -void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout) +// XBT_ATTRIB_DEPRECATED_v333 +void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size) { - /* Associate this simcall to the wait synchro */ - XBT_DEBUG("simcall_HANDLER_comm_wait, %p", comm); - - comm->register_simcall(simcall); - - if (MC_is_active() || MC_record_replay_is_active()) { - int idx = simcall->mc_value_; - if (idx == 0) { - comm->state_ = simgrid::kernel::activity::State::DONE; - } else { - /* If we reached this point, the wait simcall must have a timeout */ - /* Otherwise it shouldn't be enabled and executed by the MC */ - if (timeout < 0.0) - THROW_IMPOSSIBLE; - - if (comm->src_actor_ == simcall->issuer_) - comm->state_ = simgrid::kernel::activity::State::SRC_TIMEOUT; - else - comm->state_ = simgrid::kernel::activity::State::DST_TIMEOUT; - } - - comm->finish(); - return; - } - - /* If the synchro has already finish perform the error handling, */ - /* otherwise set up a waiting timeout on the right side */ - if (comm->state_ != simgrid::kernel::activity::State::WAITING && - comm->state_ != simgrid::kernel::activity::State::RUNNING) { - comm->finish(); - } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */ - simgrid::kernel::resource::Action* sleep = simcall->issuer_->get_host()->pimpl_cpu->sleep(timeout); - sleep->set_activity(comm); - - if (simcall->issuer_ == comm->src_actor_) - comm->src_timeout_ = sleep; - else - comm->dst_timeout_ = sleep; - } + simgrid::s4u::Comm::copy_buffer_callback(comm, buff, buff_size); } -void simcall_HANDLER_comm_test(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm) +// XBT_ATTRIB_DEPRECATED_v333 +void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size) { - bool res; - - if (MC_is_active() || MC_record_replay_is_active()) { - res = comm->src_actor_ && comm->dst_actor_; - if (res) - comm->state_ = simgrid::kernel::activity::State::DONE; - } else { - res = comm->state_ != simgrid::kernel::activity::State::WAITING && - comm->state_ != simgrid::kernel::activity::State::RUNNING; - } - - simcall_comm_test__set__result(simcall, res); - if (res) { - comm->simcalls_.push_back(simcall); - comm->finish(); - } else { - simcall->issuer_->simcall_answer(); - } + simgrid::s4u::Comm::copy_pointer_callback(comm, buff, buff_size); } -void simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count) -{ - // The default result is -1 -- this means, "nothing is ready". - // It can be changed below, but only if something matches. - simcall_comm_testany__set__result(simcall, -1); +namespace simgrid::kernel::activity { +xbt::signal CommImpl::on_start; +xbt::signal CommImpl::on_completion; - if (MC_is_active() || MC_record_replay_is_active()) { - int idx = simcall->mc_value_; - if (idx == -1) { - simcall->issuer_->simcall_answer(); - } else { - simgrid::kernel::activity::CommImpl* comm = comms[idx]; - simcall_comm_testany__set__result(simcall, idx); - comm->simcalls_.push_back(simcall); - comm->state_ = simgrid::kernel::activity::State::DONE; - comm->finish(); - } - return; - } +std::function CommImpl::copy_data_callback_ = &s4u::Comm::copy_pointer_callback; - for (std::size_t i = 0; i != count; ++i) { - simgrid::kernel::activity::CommImpl* comm = comms[i]; - if (comm->state_ != simgrid::kernel::activity::State::WAITING && - comm->state_ != simgrid::kernel::activity::State::RUNNING) { - simcall_comm_testany__set__result(simcall, i); - comm->simcalls_.push_back(simcall); - comm->finish(); - return; - } - } - simcall->issuer_->simcall_answer(); -} - -static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall) +void CommImpl::set_copy_data_callback(const std::function& callback) { - simgrid::kernel::activity::CommImpl** comms = simcall_comm_waitany__get__comms(simcall); - size_t count = simcall_comm_waitany__get__count(simcall); - - for (size_t i = 0; i < count; i++) { - // Remove the first occurrence of simcall: - auto* comm = comms[i]; - auto j = boost::range::find(comm->simcalls_, simcall); - if (j != comm->simcalls_.end()) - comm->simcalls_.erase(j); - } -} -void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count, - double timeout) -{ - if (MC_is_active() || MC_record_replay_is_active()) { - if (timeout > 0.0) - xbt_die("Timeout not implemented for waitany in the model-checker"); - int idx = simcall->mc_value_; - auto* comm = comms[idx]; - comm->simcalls_.push_back(simcall); - simcall_comm_waitany__set__result(simcall, idx); - comm->state_ = simgrid::kernel::activity::State::DONE; - comm->finish(); - return; - } - - if (timeout < 0.0) { - simcall->timeout_cb_ = nullptr; - } else { - simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall]() { - simcall->timeout_cb_ = nullptr; - SIMIX_waitany_remove_simcall_from_actions(simcall); - simcall_comm_waitany__set__result(simcall, -1); - simcall->issuer_->simcall_answer(); - }); - } - - for (size_t i = 0; i < count; i++) { - /* associate this simcall to the the synchro */ - auto* comm = comms[i]; - comm->simcalls_.push_back(simcall); - - /* see if the synchro is already finished */ - if (comm->state_ != simgrid::kernel::activity::State::WAITING && - comm->state_ != simgrid::kernel::activity::State::RUNNING) { - comm->finish(); - break; - } - } + copy_data_callback_ = callback; } -/******************************************************************************/ -/* SIMIX_comm_copy_data callbacks */ -/******************************************************************************/ -static void (*SIMIX_comm_copy_data_callback)(simgrid::kernel::activity::CommImpl*, void*, - size_t) = &SIMIX_comm_copy_pointer_callback; - -void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size) +CommImpl& CommImpl::set_type(CommImplType type) { - XBT_DEBUG("Copy the data over"); - memcpy(comm->dst_buff_, buff, buff_size); - if (comm->detached()) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the - // original buffer available to the application ASAP - xbt_free(buff); - comm->src_buff_ = nullptr; - } + type_ = type; + return *this; } -void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t)) +CommImpl& CommImpl::set_source(s4u::Host* from) { - SIMIX_comm_copy_data_callback = callback; + xbt_assert( from_ == nullptr ); + from_ = from; + hosts_.push_back(from); + return *this; } -void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size) +CommImpl& CommImpl::set_destination(s4u::Host* to) { - xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size); - *(void**)(comm->dst_buff_) = buff; + xbt_assert( to_ == nullptr ); + to_ = to; + hosts_.push_back(to_); + return *this; } -namespace simgrid { -namespace kernel { -namespace activity { - CommImpl& CommImpl::set_size(double size) { size_ = size; @@ -376,6 +89,8 @@ CommImpl& CommImpl::set_rate(double rate) } CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox) { + if (mbox != nullptr) + mbox_id_ = mbox->get_id(); mbox_ = mbox; return *this; } @@ -400,18 +115,13 @@ CommImpl& CommImpl::detach() return *this; } -CommImpl::CommImpl(s4u::Host* from, s4u::Host* to, double bytes) : size_(bytes), detached_(true), from_(from), to_(to) -{ - state_ = State::READY; -} - CommImpl::~CommImpl() { XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_); cleanup_surf(); - if (detached_ && state_ != State::DONE) { + if (detached_ && get_state() != State::DONE) { /* the communication has failed and was detached: * we have to free the buffer */ if (clean_fun) @@ -426,14 +136,23 @@ CommImpl::~CommImpl() CommImpl* CommImpl::start() { /* If both the sender and the receiver are already there, start the communication */ - if (state_ == State::READY) { + if (get_state() == State::READY) { from_ = from_ != nullptr ? from_ : src_actor_->get_host(); + xbt_assert(from_->is_on()); to_ = to_ != nullptr ? to_ : dst_actor_->get_host(); + xbt_assert(to_->is_on()); + + /* Getting the network_model from the origin host + * Valid while we have a single network model, otherwise we would need to change this function to first get the + * routes and later create the respective surf actions */ + auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model(); - surf_action_ = surf_network_model->communicate(from_, to_, size_, rate_); + surf_action_ = net_model->communicate(from_, to_, size_, rate_); surf_action_->set_activity(this); surf_action_->set_category(get_tracing_category()); - state_ = State::RUNNING; + set_start_time(surf_action_->get_start_time()); + set_state(State::RUNNING); + on_start(*this); XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(), to_->get_cname(), surf_action_, get_state_str()); @@ -442,13 +161,13 @@ CommImpl* CommImpl::start() if (surf_action_->get_state() == resource::Action::State::FAILED) { XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(), to_->get_cname()); - state_ = State::LINK_FAILURE; + set_state(State::LINK_FAILURE); post(); } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) || (dst_actor_ != nullptr && dst_actor_->is_suspended())) { - /* If any of the process is suspended, create the synchro but stop its execution, - it will be restarted when the sender process resume */ + /* If any of the actor is suspended, create the synchro but stop its execution, + it will be restarted when the sender actor resume */ if (src_actor_->is_suspended()) XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the " "communication", @@ -465,6 +184,16 @@ CommImpl* CommImpl::start() return this; } +std::vector CommImpl::get_traversed_links() const +{ + xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __FUNCTION__, + get_state_str()); + std::vector vlinks; + XBT_ATTRIB_UNUSED double res = 0; + from_->route_to(to_, vlinks, &res); + return vlinks; +} + /** @brief Copy the communication data from the sender's buffer to the receiver's one */ void CommImpl::copy_data() { @@ -474,12 +203,12 @@ void CommImpl::copy_data() return; XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this, - src_actor_ ? src_actor_->get_host()->get_cname() : "a finished process", src_buff_, - dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished process", dst_buff_, buff_size); + src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_, + dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size); /* Copy at most dst_buff_size bytes of the message to receiver's buffer */ if (dst_buff_size_) { - buff_size = std::min(buff_size, *(dst_buff_size_)); + buff_size = std::min(buff_size, *dst_buff_size_); /* Update the receiver's buffer size to the copied amount */ *dst_buff_size_ = buff_size; @@ -489,7 +218,7 @@ void CommImpl::copy_data() if (copy_data_fun) copy_data_fun(this, src_buff_, buff_size); else - SIMIX_comm_copy_data_callback(this, src_buff_, buff_size); + copy_data_callback_(this, src_buff_, buff_size); } /* Set the copied flag so we copy data only once */ @@ -497,6 +226,171 @@ void CommImpl::copy_data() copied_ = true; } +ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer) +{ + auto* mbox = observer->get_mailbox(); + XBT_DEBUG("send from mailbox %p", mbox); + + /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */ + CommImplPtr this_comm(new CommImpl()); + this_comm->set_type(CommImplType::SEND); + + /* Look for communication synchro matching our needs. We also provide a description of + * ourself so that the other side also gets a chance of choosing if it wants to match with us. + * + * If it is not found then push our communication into the rendez-vous point */ + CommImplPtr other_comm = + mbox->find_matching_comm(CommImplType::RECEIVE, observer->get_match_fun(), observer->get_payload(), this_comm, + /*done*/ false, /*remove_matching*/ true); + + if (not other_comm) { + other_comm = std::move(this_comm); + + if (mbox->is_permanent()) { + // this mailbox is for small messages, which have to be sent right now + other_comm->set_state(State::READY); + other_comm->dst_actor_ = mbox->get_permanent_receiver().get(); + mbox->push_done(other_comm); + XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get()); + + } else { + mbox->push(other_comm); + } + } else { + XBT_DEBUG("Receive already pushed"); + + other_comm->set_state(State::READY); + } + observer->set_comm(other_comm.get()); + + if (observer->is_detached()) { + other_comm->detach(); + other_comm->clean_fun = observer->get_clean_fun(); + EngineImpl::get_instance()->get_maestro()->activities_.emplace_back(other_comm); + } else { + other_comm->clean_fun = nullptr; + observer->get_issuer()->activities_.emplace_back(other_comm); + } + + /* Setup the communication synchro */ + other_comm->src_actor_ = observer->get_issuer(); + other_comm->src_data_ = observer->get_payload(); + (*other_comm) + .set_src_buff(observer->get_src_buff(), observer->get_src_buff_size()) + .set_size(observer->get_payload_size()) + .set_rate(observer->get_rate()); + + other_comm->match_fun = observer->get_match_fun(); + other_comm->copy_data_fun = observer->get_copy_data_fun(); + + if (MC_is_active() || MC_record_replay_is_active()) + other_comm->set_state(simgrid::kernel::activity::State::RUNNING); + else + other_comm->start(); + + return (observer->is_detached() ? nullptr : other_comm); +} + +ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer) +{ + CommImplPtr this_synchro(new CommImpl()); + this_synchro->set_type(CommImplType::RECEIVE); + + auto* mbox = observer->get_mailbox(); + XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get()); + + CommImplPtr other_comm; + // communication already done, get it inside the list of completed comms + if (mbox->is_permanent() && mbox->has_some_done_comm()) { + XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication"); + // find a match in the list of already received comms + other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(), + this_synchro, /*done*/ true, /*remove_matching*/ true); + if (other_comm && other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) { + XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get()); + other_comm->set_state(State::DONE); + other_comm->set_mailbox(nullptr); + } else { + // if not found, assume the receiver came first, register it to the mailbox in the classical way + if (not other_comm) { + XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request " + "into list"); + other_comm = std::move(this_synchro); + mbox->push(other_comm); + } + observer->get_issuer()->activities_.emplace_back(other_comm); + } + } else { + /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */ + + /* Look for communication activity matching our needs. We also provide a description of + * ourself so that the other side also gets a chance of choosing if it wants to match with us. + * + * If it is not found then push our communication into the rendez-vous point */ + other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(), + this_synchro, /*done*/ false, /*remove_matching*/ true); + + if (other_comm == nullptr) { + XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size()); + other_comm = std::move(this_synchro); + mbox->push(other_comm); + } else { + XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get()); + + other_comm->set_state(simgrid::kernel::activity::State::READY); + } + observer->get_issuer()->activities_.emplace_back(other_comm); + } + observer->set_comm(other_comm.get()); + + /* Setup communication synchro */ + other_comm->dst_actor_ = observer->get_issuer(); + other_comm->dst_data_ = observer->get_payload(); + other_comm->set_dst_buff(observer->get_dst_buff(), observer->get_dst_buff_size()); + + if (observer->get_rate() > -1.0 && (other_comm->get_rate() < 0.0 || observer->get_rate() < other_comm->get_rate())) + other_comm->set_rate(observer->get_rate()); + + other_comm->match_fun = observer->get_match_fun(); + other_comm->copy_data_fun = observer->get_copy_data_fun(); + + if (MC_is_active() || MC_record_replay_is_active()) { + other_comm->set_state(State::RUNNING); + return other_comm; + } + other_comm->start(); + + return other_comm; +} + +bool CommImpl::test(actor::ActorImpl* issuer) +{ + if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_) + set_state(State::DONE); + return ActivityImpl::test(issuer); +} + +void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout) +{ + XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str()); + + /* Associate this simcall to the wait synchro */ + register_simcall(&issuer->simcall_); + if (MC_is_active() || MC_record_replay_is_active()) { + // FIXME: what about timeouts? + set_state(State::DONE); + finish(); + return; + } + ActivityImpl::wait_for(issuer, timeout); +} + +void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector& comms, double timeout) +{ + std::vector activities(comms.begin(), comms.end()); + ActivityImpl::wait_any_for(issuer, activities, timeout); +} + void CommImpl::suspend() { /* FIXME: shall we suspend also the timeout synchro? */ @@ -516,13 +410,13 @@ void CommImpl::resume() void CommImpl::cancel() { /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */ - if (state_ == State::WAITING) { + if (get_state() == State::WAITING) { if (not detached_) { mbox_->remove(this); - state_ = State::CANCELED; + set_state(State::CANCELED); } } else if (not MC_is_active() /* when running the MC there are no surf actions */ - && not MC_record_replay_is_active() && (state_ == State::READY || state_ == State::RUNNING)) { + && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) { surf_action_->cancel(); } } @@ -545,19 +439,24 @@ void CommImpl::cleanup_surf() void CommImpl::post() { + on_completion(*this); + /* Update synchro state */ if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED) - state_ = State::SRC_TIMEOUT; + set_state(State::SRC_TIMEOUT); else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED) - state_ = State::DST_TIMEOUT; - else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED) - state_ = State::SRC_HOST_FAILURE; - else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED) - state_ = State::DST_HOST_FAILURE; + set_state(State::DST_TIMEOUT); + else if ((from_ && not from_->is_on()) || (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)) + set_state(State::SRC_HOST_FAILURE); + else if ((to_ && not to_->is_on()) || (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)) + set_state(State::DST_HOST_FAILURE); else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) { - state_ = State::LINK_FAILURE; - } else - state_ = State::DONE; + set_state(State::LINK_FAILURE); + } else if (get_state() == State::RUNNING) { + xbt_assert(from_ && from_->is_on()); + xbt_assert(to_ && to_->is_on()); + set_state(State::DONE); + } XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(), src_actor_.get(), dst_actor_.get(), detached_); @@ -568,129 +467,103 @@ void CommImpl::post() /* Answer all simcalls associated with the synchro */ finish(); } +void CommImpl::set_exception(actor::ActorImpl* issuer) +{ + switch (get_state()) { + case State::FAILED: + issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); + break; + case State::SRC_TIMEOUT: + issuer->exception_ = + std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender")); + break; + + case State::DST_TIMEOUT: + issuer->exception_ = + std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver")); + break; + + case State::SRC_HOST_FAILURE: + if (issuer == src_actor_) + issuer->set_wannadie(); + else { + set_state(State::FAILED); + issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); + } + break; + + case State::DST_HOST_FAILURE: + if (issuer == dst_actor_) + issuer->set_wannadie(); + else { + set_state(State::FAILED); + issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); + } + break; + + case State::LINK_FAILURE: + XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) " + "detached:%d", + this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr, + dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_); + if (src_actor_ == issuer) { + XBT_DEBUG("I'm source"); + } else if (dst_actor_ == issuer) { + XBT_DEBUG("I'm dest"); + } else { + XBT_DEBUG("I'm neither source nor dest"); + } + set_state(State::FAILED); + issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure"))); + break; + + case State::CANCELED: + if (issuer == dst_actor_) + issuer->exception_ = + std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender")); + else + issuer->exception_ = + std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver")); + break; + + default: + xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s", + get_state_str()); + } +} void CommImpl::finish() { + XBT_DEBUG("CommImpl::finish() in state %s", get_state_str()); + /* If the synchro is still in a rendez-vous point then remove from it */ + if (mbox_) + mbox_->remove(this); + + if (get_state() == State::DONE) + copy_data(); + while (not simcalls_.empty()) { - smx_simcall_t simcall = simcalls_.front(); + actor::Simcall* simcall = simcalls_.front(); simcalls_.pop_front(); /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the * simcall */ - if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case - continue; // if actor handling comm is killed - if (simcall->call_ == simix::Simcall::COMM_WAITANY) { - SIMIX_waitany_remove_simcall_from_actions(simcall); - if (simcall->timeout_cb_) { - simcall->timeout_cb_->remove(); - simcall->timeout_cb_ = nullptr; - } - if (not MC_is_active() && not MC_record_replay_is_active()) { - CommImpl** comms = simcall_comm_waitany__get__comms(simcall); - size_t count = simcall_comm_waitany__get__count(simcall); - CommImpl** element = std::find(comms, comms + count, this); - int rank = (element != comms + count) ? element - comms : -1; - simcall_comm_waitany__set__result(simcall, rank); - } - } - - /* If the synchro is still in a rendez-vous point then remove from it */ - if (mbox_) - mbox_->remove(this); + if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case + continue; // if actor handling comm is killed - XBT_DEBUG("CommImpl::finish(): synchro state = %d", static_cast(state_)); + handle_activity_waitany(simcall); /* Check out for errors */ if (not simcall->issuer_->get_host()->is_on()) { - simcall->issuer_->context_->set_wannadie(); + simcall->issuer_->set_wannadie(); } else { - switch (state_) { - case State::DONE: - XBT_DEBUG("Communication %p complete!", this); - copy_data(); - break; - - case State::SRC_TIMEOUT: - simcall->issuer_->exception_ = std::make_exception_ptr( - simgrid::TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender")); - break; - - case State::DST_TIMEOUT: - simcall->issuer_->exception_ = std::make_exception_ptr( - simgrid::TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver")); - break; - - case State::SRC_HOST_FAILURE: - if (simcall->issuer_ == src_actor_) - simcall->issuer_->context_->set_wannadie(); - else - simcall->issuer_->exception_ = - std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); - break; - - case State::DST_HOST_FAILURE: - if (simcall->issuer_ == dst_actor_) - simcall->issuer_->context_->set_wannadie(); - else - simcall->issuer_->exception_ = - std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); - break; - - case State::LINK_FAILURE: - XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) " - "detached:%d", - this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr, - dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, simcall->issuer_->get_cname(), - simcall->issuer_, detached_); - if (src_actor_ == simcall->issuer_) { - XBT_DEBUG("I'm source"); - } else if (dst_actor_ == simcall->issuer_) { - XBT_DEBUG("I'm dest"); - } else { - XBT_DEBUG("I'm neither source nor dest"); - } - simcall->issuer_->throw_exception( - std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Link failure"))); - break; - - case State::CANCELED: - if (simcall->issuer_ == dst_actor_) - simcall->issuer_->exception_ = std::make_exception_ptr( - simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the sender")); - else - simcall->issuer_->exception_ = std::make_exception_ptr( - simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the receiver")); - break; - - default: - xbt_die("Unexpected synchro state in CommImpl::finish: %d", static_cast(state_)); - } - simcall->issuer_->simcall_answer(); - } - /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */ - if (simcall->issuer_->exception_ && - (simcall->call_ == simix::Simcall::COMM_WAITANY || simcall->call_ == simix::Simcall::COMM_TESTANY)) { - // First retrieve the rank of our failing synchro - CommImpl** comms; - size_t count; - if (simcall->call_ == simix::Simcall::COMM_WAITANY) { - comms = simcall_comm_waitany__get__comms(simcall); - count = simcall_comm_waitany__get__count(simcall); - } else { - /* simcall->call_ == simix::Simcall::COMM_TESTANY */ - comms = simcall_comm_testany__get__comms(simcall); - count = simcall_comm_testany__get__count(simcall); - } - CommImpl** element = std::find(comms, comms + count, this); - int rank = (element != comms + count) ? element - comms : -1; - // In order to modify the exception we have to rethrow it: - try { - std::rethrow_exception(simcall->issuer_->exception_); - } catch (simgrid::Exception& e) { - e.set_value(rank); + // Do not answer to dying actors + if (not simcall->issuer_->wannadie()) { + set_exception(simcall->issuer_); + simcall->issuer_->simcall_answer(); } } @@ -705,6 +578,4 @@ void CommImpl::finish() } } -} // namespace activity -} // namespace kernel -} // namespace simgrid +} // namespace simgrid::kernel::activity