X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/1747902010b7fc57b15c2e88473c5bb118d362d1..76af4f8d5d8f11687e093dd5f2c7f94c16e51387:/src/kernel/activity/CommImpl.cpp diff --git a/src/kernel/activity/CommImpl.cpp b/src/kernel/activity/CommImpl.cpp index f8de83ad91..fc67a922c8 100644 --- a/src/kernel/activity/CommImpl.cpp +++ b/src/kernel/activity/CommImpl.cpp @@ -3,20 +3,18 @@ /* 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 + #include "src/kernel/activity/CommImpl.hpp" -#include "simgrid/Exception.hpp" -#include "simgrid/kernel/resource/Action.hpp" -#include "simgrid/kernel/routing/NetPoint.hpp" -#include "simgrid/modelchecker.h" -#include "simgrid/s4u/Host.hpp" #include "src/kernel/activity/MailboxImpl.hpp" #include "src/kernel/context/Context.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, @@ -28,7 +26,7 @@ XBT_PRIVATE void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_actor_t sr 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); + comm->wait_for(simcall->issuer_, timeout); } XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr simcall_HANDLER_comm_isend( @@ -56,11 +54,11 @@ XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr simcall_HANDLER_comm_isen if (not other_comm) { other_comm = std::move(this_comm); - if (mbox->permanent_receiver_ != nullptr) { + if (mbox->is_permanent()) { // 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); + 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 { @@ -105,7 +103,7 @@ XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t re 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); + comm->wait_for(simcall->issuer_, timeout); } XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr @@ -120,7 +118,7 @@ simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_ 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()) { + 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(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data, @@ -151,7 +149,7 @@ simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_ /*remove_matching*/ true); if (other_comm == nullptr) { - XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue_.size()); + XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size()); other_comm = std::move(this_synchro); mbox->push(other_comm); } else { @@ -183,160 +181,35 @@ simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_ void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout) { - /* 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; - } + comm->wait_for(simcall->issuer_, timeout); } -void simcall_HANDLER_comm_test(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm) +bool simcall_HANDLER_comm_test(smx_simcall_t, simgrid::kernel::activity::CommImpl* comm) { - 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(); - } + return comm->test(); } -void simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count) +ssize_t 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); - - 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; - } - - 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(); + std::vector comms_vec(comms, comms + count); + return simgrid::kernel::activity::CommImpl::test_any(simcall->issuer_, comms_vec); } -static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall) -{ - 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; - } - } + std::vector comms_vec(comms, comms + count); + simgrid::kernel::activity::CommImpl::wait_any_for(simcall->issuer_, comms_vec, timeout); } /******************************************************************************/ /* SIMIX_comm_copy_data callbacks */ /******************************************************************************/ -static void (*SIMIX_comm_copy_data_callback)(simgrid::kernel::activity::CommImpl*, void*, - size_t) = &SIMIX_comm_copy_pointer_callback; +// XBT_ATTRIB_DEPRECATED_v333 +void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t)) +{ + simgrid::kernel::activity::CommImpl::set_copy_data_callback(callback); +} void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size) { @@ -349,11 +222,6 @@ void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, } } -void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t)) -{ - SIMIX_comm_copy_data_callback = callback; -} - void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size) { xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size); @@ -364,6 +232,13 @@ namespace simgrid { namespace kernel { namespace activity { +void (*CommImpl::copy_data_callback_)(CommImpl*, void*, size_t) = &SIMIX_comm_copy_pointer_callback; + +void CommImpl::set_copy_data_callback(void (*callback)(CommImpl*, void*, size_t)) +{ + copy_data_callback_ = callback; +} + CommImpl& CommImpl::set_size(double size) { size_ = size; @@ -431,10 +306,10 @@ CommImpl* CommImpl::start() from_ = from_ != nullptr ? from_ : src_actor_->get_host(); to_ = to_ != nullptr ? to_ : dst_actor_->get_host(); - /* FIXME[donassolo]: getting the network_model from the origin host - * Soon we 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(); + /* 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_ = net_model->communicate(from_, to_, size_, rate_); surf_action_->set_activity(this); @@ -453,8 +328,8 @@ CommImpl* CommImpl::start() } 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", @@ -480,12 +355,12 @@ 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; @@ -495,7 +370,7 @@ 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 */ @@ -503,6 +378,104 @@ void CommImpl::copy_data() copied_ = true; } +bool CommImpl::test() +{ + if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_) + state_ = State::DONE; + return ActivityImpl::test(); +} + +void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout) +{ + XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, to_c_str(state_)); + + /* Associate this simcall to the wait synchro */ + register_simcall(&issuer->simcall_); + + if (MC_is_active() || MC_record_replay_is_active()) { + int idx = issuer->simcall_.mc_value_; + if (idx == 0) { + state_ = 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; + state_ = (issuer == src_actor_ ? State::SRC_TIMEOUT : State::DST_TIMEOUT); + } + finish(); + return; + } + + /* If the synchro has already finish perform the error handling, */ + /* otherwise set up a waiting timeout on the right side */ + if (state_ != State::WAITING && state_ != State::RUNNING) { + finish(); + } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */ + resource::Action* sleep = issuer->get_host()->get_cpu()->sleep(timeout); + sleep->set_activity(this); + + if (issuer == src_actor_) + src_timeout_ = sleep; + else + dst_timeout_ = sleep; + } +} + +ssize_t CommImpl::test_any(const actor::ActorImpl* issuer, const std::vector& comms) +{ + if (MC_is_active() || MC_record_replay_is_active()) { + int idx = issuer->simcall_.mc_value_; + xbt_assert(idx == -1 || comms[idx]->test()); + return idx; + } + + for (std::size_t i = 0; i < comms.size(); ++i) { + if (comms[i]->test()) + return i; + } + return -1; +} + +void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector& comms, double timeout) +{ + if (MC_is_active() || MC_record_replay_is_active()) { + xbt_assert(timeout <= 0.0, "Timeout not implemented for waitany in the model-checker"); + int idx = issuer->simcall_.mc_value_; + auto* comm = comms[idx]; + comm->simcalls_.push_back(&issuer->simcall_); + simcall_comm_waitany__set__result(&issuer->simcall_, idx); + comm->state_ = State::DONE; + comm->finish(); + return; + } + + if (timeout < 0.0) { + issuer->simcall_.timeout_cb_ = nullptr; + } else { + issuer->simcall_.timeout_cb_ = timer::Timer::set(s4u::Engine::get_clock() + timeout, [issuer, comms]() { + // FIXME: Vector `comms' is copied here. Use a reference once its lifetime is extended (i.e. when the simcall is + // modernized). + issuer->simcall_.timeout_cb_ = nullptr; + for (auto* comm : comms) + comm->unregister_simcall(&issuer->simcall_); + simcall_comm_waitany__set__result(&issuer->simcall_, -1); + issuer->simcall_answer(); + }); + } + + for (auto* comm : comms) { + /* associate this simcall to the the synchro */ + comm->simcalls_.push_back(&issuer->simcall_); + + /* see if the synchro is already finished */ + if (comm->state_ != State::WAITING && comm->state_ != State::RUNNING) { + comm->finish(); + break; + } + } +} + void CommImpl::suspend() { /* FIXME: shall we suspend also the timeout synchro? */ @@ -577,6 +550,15 @@ void CommImpl::post() void CommImpl::finish() { + XBT_DEBUG("CommImpl::finish() in state %s", to_c_str(state_)); + + /* If the synchro is still in a rendez-vous point then remove from it */ + if (mbox_) + mbox_->remove(this); + + if (state_ == State::DONE) + copy_data(); + while (not simcalls_.empty()) { smx_simcall_t simcall = simcalls_.front(); simcalls_.pop_front(); @@ -588,61 +570,59 @@ void CommImpl::finish() 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); + 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++) + comms[i]->unregister_simcall(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; + auto element = std::find(comms, comms + count, this); + ssize_t 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); - - XBT_DEBUG("CommImpl::finish(): synchro state = %d", static_cast(state_)); - /* Check out for errors */ if (not simcall->issuer_->get_host()->is_on()) { simcall->issuer_->context_->set_wannadie(); } else { switch (state_) { - case State::DONE: - XBT_DEBUG("Communication %p complete!", this); - copy_data(); + case State::FAILED: + simcall->issuer_->exception_ = + std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); break; - case State::SRC_TIMEOUT: simcall->issuer_->exception_ = std::make_exception_ptr( - simgrid::TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender")); + 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")); + 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 + else { + state_ = kernel::activity::State::FAILED; simcall->issuer_->exception_ = - std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); + std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); + } break; case State::DST_HOST_FAILURE: if (simcall->issuer_ == dst_actor_) simcall->issuer_->context_->set_wannadie(); - else + else { + state_ = kernel::activity::State::FAILED; simcall->issuer_->exception_ = - std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); + std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); + } break; case State::LINK_FAILURE: @@ -658,21 +638,23 @@ void CommImpl::finish() } else { XBT_DEBUG("I'm neither source nor dest"); } + state_ = kernel::activity::State::FAILED; simcall->issuer_->throw_exception( - std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Link failure"))); + std::make_exception_ptr(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")); + simcall->issuer_->exception_ = + std::make_exception_ptr(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")); + simcall->issuer_->exception_ = + std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver")); break; default: - xbt_die("Unexpected synchro state in CommImpl::finish: %d", static_cast(state_)); + xbt_assert(state_ == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s", + to_c_str(state_)); } simcall->issuer_->simcall_answer(); } @@ -690,12 +672,12 @@ void CommImpl::finish() 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; + auto element = std::find(comms, comms + count, this); + ssize_t 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) { + } catch (Exception& e) { e.set_value(rank); } }