X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/db47dd48806619184080b0b9d9e655d7d9d1ed63..4d58af5e9f29128ea5d5cbb677884eae5ba1bf81:/src/kernel/activity/ExecImpl.cpp diff --git a/src/kernel/activity/ExecImpl.cpp b/src/kernel/activity/ExecImpl.cpp index 5da186d396..a05e00c29c 100644 --- a/src/kernel/activity/ExecImpl.cpp +++ b/src/kernel/activity/ExecImpl.cpp @@ -1,154 +1,234 @@ -/* Copyright (c) 2007-2019. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2023. 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 "simgrid/modelchecker.h" -#include "src/mc/mc_replay.hpp" +#include +#include +#include +#include #include "src/kernel/activity/ExecImpl.hpp" -#include "src/simix/smx_host_private.hpp" -#include "src/surf/surf_interface.hpp" -#include "src/surf/cpu_interface.hpp" +#include "src/kernel/actor/ActorImpl.hpp" +#include "src/kernel/actor/SimcallObserver.hpp" +#include "src/kernel/resource/CpuImpl.hpp" +#include "src/kernel/resource/HostImpl.hpp" +#include "src/mc/mc_replay.hpp" + +XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_cpu, kernel, "Kernel cpu-related synchronization"); + +namespace simgrid::kernel::activity { -#include "simgrid/s4u/Host.hpp" +ExecImpl::ExecImpl() +{ + piface_ = new s4u::Exec(this); + actor::ActorImpl* self = actor::ActorImpl::self(); + if (self) { + set_actor(self); + self->activities_.insert(this); + } +} -XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process); -namespace simgrid { -namespace kernel { -namespace activity { +ExecImpl& ExecImpl::set_host(s4u::Host* host) +{ + ActivityImpl::set_hosts({host}); + return *this; +} -ExecImpl::ExecImpl(std::string name, std::string tracing_category, resource::Action* timeout_detector, s4u::Host* host) - : ActivityImpl(name), host_(host), timeout_detector_(timeout_detector) +ExecImpl& ExecImpl::set_hosts(const std::vector& hosts) { - this->state_ = SIMIX_RUNNING; - this->set_category(tracing_category); + ActivityImpl::set_hosts(hosts); + return *this; +} - if (timeout_detector != nullptr) - timeout_detector_->set_data(this); +ExecImpl& ExecImpl::set_flops_amount(double flops_amount) +{ + flops_amounts_.assign(1, flops_amount); + return *this; +} - XBT_DEBUG("Create exec %p", this); +ExecImpl& ExecImpl::set_flops_amounts(const std::vector& flops_amounts) +{ + flops_amounts_ = flops_amounts; + return *this; } -ExecImpl::~ExecImpl() +ExecImpl& ExecImpl::set_bytes_amounts(const std::vector& bytes_amounts) +{ + bytes_amounts_ = bytes_amounts; + + return *this; +} +ExecImpl& ExecImpl::set_thread_count(int thread_count) { - if (surf_action_) - surf_action_->unref(); - if (timeout_detector_) - timeout_detector_->unref(); - XBT_DEBUG("Destroy exec %p", this); + thread_count_ = thread_count; + + return *this; } -ExecImpl* ExecImpl::start(double flops_amount, double priority, double bound) +ExecImpl* ExecImpl::start() { + set_state(State::RUNNING); if (not MC_is_active() && not MC_record_replay_is_active()) { - surf_action_ = host_->pimpl_cpu->execution_start(flops_amount); - surf_action_->set_data(this); - surf_action_->set_priority(priority); - if (bound > 0) - surf_action_->set_bound(bound); + if (get_hosts().size() == 1) { + xbt_assert(not flops_amounts_.empty(), "Cannot start Exec: no flops_amount defined."); + if (thread_count_ == 1) { + model_action_ = get_host()->get_cpu()->execution_start(flops_amounts_.front(), bound_); + model_action_->set_sharing_penalty(sharing_penalty_); + } else { + auto host_model = get_host()->get_netpoint()->get_englobing_zone()->get_host_model(); + model_action_ = host_model->execute_thread(get_host(), flops_amounts_.front(), thread_count_); + } + model_action_->set_category(get_tracing_category()); + } else { + // get the model from first host since we have only 1 by now + auto host_model = get_host()->get_netpoint()->get_englobing_zone()->get_host_model(); + model_action_ = host_model->execute_parallel(get_hosts(), flops_amounts_.data(), bytes_amounts_.data(), -1); + } + model_action_->set_activity(this); + set_start_time(model_action_->get_start_time()); } - XBT_DEBUG("Create execute synchro %p: %s", this, name_.c_str()); - ExecImpl::on_creation(this); + XBT_DEBUG("Create execute synchro %p: %s", this, get_cname()); return this; } -void ExecImpl::cancel() +double ExecImpl::get_remaining() const +{ + if (get_state() == State::WAITING || get_state() == State::FAILED) + return flops_amounts_.front(); + return ActivityImpl::get_remaining(); +} + +double ExecImpl::get_seq_remaining_ratio() { - XBT_VERB("This exec %p is canceled", this); - if (surf_action_ != nullptr) - surf_action_->cancel(); + if (get_state() == State::WAITING) + return 1; + return (model_action_ == nullptr) ? 0 : model_action_->get_remains() / model_action_->get_cost(); } -double ExecImpl::get_remaining() +double ExecImpl::get_par_remaining_ratio() { - xbt_assert(host_ != nullptr, "Calling remains() on a parallel execution is not allowed. " - "We would need to return a vector instead of a scalar. " - "Did you mean remainingRatio() instead?"); - return surf_action_ ? surf_action_->get_remains() : 0; + // parallel task: their remain is already between 0 and 1 + if (get_state() == State::WAITING) + return 1; + return (model_action_ == nullptr) ? 0 : model_action_->get_remains(); } -double ExecImpl::get_remaining_ratio() +ExecImpl& ExecImpl::set_bound(double bound) { - if (host_ == - nullptr) // parallel task: their remain is already between 0 and 1 (see comment in ExecImpl::get_remaining()) - return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains(); - else // Actually compute the ratio for sequential tasks - return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost(); + bound_ = bound; + return *this; } -void ExecImpl::set_bound(double bound) +ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty) { - if (surf_action_) - surf_action_->set_bound(bound); + sharing_penalty_ = sharing_penalty; + return *this; } -void ExecImpl::set_priority(double priority) + +ExecImpl& ExecImpl::update_sharing_penalty(double sharing_penalty) { - if (surf_action_) - surf_action_->set_priority(priority); + sharing_penalty_ = sharing_penalty; + model_action_->set_sharing_penalty(sharing_penalty); + return *this; } -void ExecImpl::post() +void ExecImpl::set_exception(actor::ActorImpl* issuer) +{ + switch (get_state()) { + case State::FAILED: + static_cast(get_iface())->complete(s4u::Activity::State::FAILED); + if (issuer->get_host()->is_on()) + issuer->exception_ = std::make_exception_ptr(HostFailureException(XBT_THROW_POINT, "Host failed")); + else /* else, the actor will be killed with no possibility to survive */ + issuer->set_wannadie(); + break; + + case State::CANCELED: + issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled")); + break; + + case State::TIMEOUT: + issuer->exception_ = std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Timeouted")); + break; + + default: + xbt_assert(get_state() == State::DONE, "Internal error in ExecImpl::finish(): unexpected synchro state %s", + get_state_str()); + } +} +void ExecImpl::finish() { - if (host_ && host_->is_off()) { /* FIXME: handle resource failure for parallel tasks too */ - /* If the host running the synchro failed, notice it. This way, the asking - * process can be killed if it runs on that host itself */ - state_ = SIMIX_FAILED; - } else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) { - /* If the host running the synchro didn't fail, then the synchro was canceled */ - state_ = SIMIX_CANCELED; - } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) { - state_ = SIMIX_TIMEOUT; - } else { - state_ = SIMIX_DONE; + XBT_DEBUG("ExecImpl::finish() in state %s", get_state_str()); + if (model_action_ != nullptr) { + if (auto const& hosts = get_hosts(); + std::any_of(hosts.begin(), hosts.end(), [](const s4u::Host* host) { return not host->is_on(); })) { + /* If one of the hosts running the synchro failed, notice it. This way, the asking + * process can be killed if it runs on that host itself */ + set_state(State::FAILED); + } else if (model_action_->get_state() == resource::Action::State::FAILED) { + /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */ + set_state(State::CANCELED); + } else { + set_state(State::DONE); + } + + clean_action(); } - on_completion(this); + if (get_actor() != nullptr) + get_actor()->activities_.erase(this); - if (surf_action_) { - surf_action_->unref(); - surf_action_ = nullptr; - } - if (timeout_detector_) { - timeout_detector_->unref(); - timeout_detector_ = nullptr; + while (not simcalls_.empty()) { + actor::Simcall* simcall = simcalls_.front(); + simcalls_.pop_front(); + + if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case + continue; // if process handling comm is killed + + handle_activity_waitany(simcall); + + set_exception(simcall->issuer_); + + simcall->issuer_->waiting_synchro_ = nullptr; + /* Fail the process if the host is down */ + if (simcall->issuer_->get_host()->is_on()) + simcall->issuer_->simcall_answer(); + else + simcall->issuer_->set_wannadie(); } +} - /* If there are simcalls associated with the synchro, then answer them */ - if (not simcalls_.empty()) - SIMIX_execution_finish(this); +void ExecImpl::reset() +{ + clear_hosts(); + set_start_time(-1.0); } -ActivityImpl* ExecImpl::migrate(simgrid::s4u::Host* to) +ActivityImpl* ExecImpl::migrate(s4u::Host* to) { if (not MC_is_active() && not MC_record_replay_is_active()) { - resource::Action* old_action = this->surf_action_; - resource::Action* new_action = to->pimpl_cpu->execution_start(old_action->get_cost()); + resource::Action* old_action = this->model_action_; + resource::Action* new_action = to->get_cpu()->execution_start(old_action->get_cost(), old_action->get_user_bound()); new_action->set_remains(old_action->get_remains()); - new_action->set_data(this); - new_action->set_priority(old_action->get_priority()); - - // FIXME: the user-defined bound seem to not be kept by LMM, that seem to overwrite it for the multi-core modeling. - // I hope that the user did not provide any. + new_action->set_activity(this); + new_action->set_sharing_penalty(old_action->get_sharing_penalty()); + new_action->set_user_bound(old_action->get_user_bound()); - old_action->set_data(nullptr); + old_action->set_activity(nullptr); old_action->cancel(); old_action->unref(); - this->surf_action_ = new_action; + this->model_action_ = new_action; } - on_migration(this, to); + on_migration(*this, to); return this; } /************* * Callbacks * *************/ -xbt::signal ExecImpl::on_creation; -xbt::signal ExecImpl::on_completion; -xbt::signal ExecImpl::on_migration; +xbt::signal ExecImpl::on_migration; -} // namespace activity -} // namespace kernel -} // namespace simgrid +} // namespace simgrid::kernel::activity