-/* 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/Exception.hpp>
+#include <simgrid/kernel/routing/NetPoint.hpp>
+#include <simgrid/modelchecker.h>
+#include <simgrid/s4u/Engine.hpp>
+
#include "src/kernel/activity/ExecImpl.hpp"
-#include "simgrid/Exception.hpp"
-#include "simgrid/modelchecker.h"
+#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"
-#include "src/surf/HostImpl.hpp"
-#include "src/surf/cpu_interface.hpp"
-#include "src/surf/surf_interface.hpp"
-
-#include "simgrid/s4u/Host.hpp"
-
-XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
-
-void simcall_HANDLER_execution_wait(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro)
-{
- XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state_);
-
- /* Associate this simcall to the synchro */
- synchro->simcalls_.push_back(simcall);
- simcall->issuer->waiting_synchro = synchro;
- /* set surf's synchro */
- if (MC_is_active() || MC_record_replay_is_active()) {
- synchro->state_ = SIMIX_DONE;
- synchro->finish();
- return;
- }
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_cpu, kernel, "Kernel cpu-related synchronization");
- /* If the synchro is already finished then perform the error handling */
- if (synchro->state_ != SIMIX_RUNNING)
- synchro->finish();
-}
+namespace simgrid::kernel::activity {
-void simcall_HANDLER_execution_test(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro)
+ExecImpl::ExecImpl()
{
- int res = (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING);
- if (res) {
- synchro->simcalls_.push_back(simcall);
- synchro->finish();
- } else {
- SIMIX_simcall_answer(simcall);
+ piface_ = new s4u::Exec(this);
+ actor::ActorImpl* self = actor::ActorImpl::self();
+ if (self) {
+ set_actor(self);
+ self->activities_.insert(this);
}
- simcall_execution_test__set__result(simcall, res);
}
-namespace simgrid {
-namespace kernel {
-namespace activity {
-
-ExecImpl::~ExecImpl()
+ExecImpl& ExecImpl::set_host(s4u::Host* host)
{
- if (timeout_detector_)
- timeout_detector_->unref();
- XBT_DEBUG("Destroy exec %p", this);
+ ActivityImpl::set_hosts({host});
+ return *this;
}
-ExecImpl& ExecImpl::set_host(s4u::Host* host)
+ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
{
- host_ = host;
+ ActivityImpl::set_hosts(hosts);
return *this;
}
-ExecImpl& ExecImpl::set_name(const std::string& name)
+ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
{
- ActivityImpl::set_name(name);
+ flops_amounts_.assign(1, flops_amount);
return *this;
}
-ExecImpl& ExecImpl::set_tracing_category(const std::string& category)
+ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
{
- ActivityImpl::set_category(category);
+ flops_amounts_ = flops_amounts;
return *this;
}
-ExecImpl& ExecImpl::set_timeout(double timeout)
+ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
{
- if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) {
- timeout_detector_ = host_->pimpl_cpu->sleep(timeout);
- timeout_detector_->set_data(this);
- }
+ bytes_amounts_ = bytes_amounts;
+
return *this;
}
-
-ExecImpl* ExecImpl::start(double flops_amount, double priority, double bound)
+ExecImpl& ExecImpl::set_thread_count(int thread_count)
{
- state_ = SIMIX_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);
- }
+ thread_count_ = thread_count;
- XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
- ExecImpl::on_creation(*this);
- return this;
+ return *this;
}
-ExecImpl* ExecImpl::start(const std::vector<s4u::Host*>& hosts, const std::vector<double>& flops_amounts,
- const std::vector<double>& bytes_amounts)
+ExecImpl* ExecImpl::start()
{
- state_ = SIMIX_RUNNING;
- /* set surf's synchro */
+ set_state(State::RUNNING);
if (not MC_is_active() && not MC_record_replay_is_active()) {
- surf_action_ = surf_host_model->execute_parallel(hosts, flops_amounts.data(), bytes_amounts.data(), -1);
- if (surf_action_ != nullptr) {
- surf_action_->set_data(this);
+ 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 parallel execute synchro %p", this);
- ExecImpl::on_creation(*this);
+
+ XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
return this;
}
-void ExecImpl::cancel()
-{
- XBT_VERB("This exec %p is canceled", this);
- if (surf_action_ != nullptr)
- surf_action_->cancel();
-}
double ExecImpl::get_remaining() const
{
- return surf_action_ ? surf_action_->get_remains() : 0;
+ if (get_state() == State::WAITING || get_state() == State::FAILED)
+ return flops_amounts_.front();
+ return ActivityImpl::get_remaining();
}
double ExecImpl::get_seq_remaining_ratio()
{
- return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
+ if (get_state() == State::WAITING)
+ return 1;
+ return (model_action_ == nullptr) ? 0 : model_action_->get_remains() / model_action_->get_cost();
}
double ExecImpl::get_par_remaining_ratio()
{
// parallel task: their remain is already between 0 and 1
- return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains();
+ if (get_state() == State::WAITING)
+ return 1;
+ return (model_action_ == nullptr) ? 0 : model_action_->get_remains();
}
-void ExecImpl::set_bound(double bound)
+ExecImpl& ExecImpl::set_bound(double bound)
{
- if (surf_action_)
- surf_action_->set_bound(bound);
+ bound_ = bound;
+ return *this;
}
-void ExecImpl::set_priority(double priority)
+
+ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
{
- if (surf_action_)
- surf_action_->set_priority(priority);
+ sharing_penalty_ = sharing_penalty;
+ return *this;
}
-void ExecImpl::post()
+ExecImpl& ExecImpl::update_sharing_penalty(double sharing_penalty)
{
- if (host_ && not host_->is_on()) { /* 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;
- }
+ sharing_penalty_ = sharing_penalty;
+ model_action_->set_sharing_penalty(sharing_penalty);
+ return *this;
+}
- on_completion(*this);
+void ExecImpl::set_exception(actor::ActorImpl* issuer)
+{
+ switch (get_state()) {
+ case State::FAILED:
+ static_cast<s4u::Exec*>(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;
- if (surf_action_) {
- surf_action_->unref();
- surf_action_ = nullptr;
- }
- if (timeout_detector_) {
- timeout_detector_->unref();
- timeout_detector_ = nullptr;
- }
+ case State::CANCELED:
+ issuer->exception_ = std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Execution Canceled"));
+ break;
- /* If there are simcalls associated with the synchro, then answer them */
- if (not simcalls_.empty())
- finish();
-}
+ 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()
{
+ 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();
+ }
+
+ if (get_actor() != nullptr)
+ get_actor()->activities_.erase(this);
+
while (not simcalls_.empty()) {
- smx_simcall_t simcall = simcalls_.front();
+ actor::Simcall* simcall = simcalls_.front();
simcalls_.pop_front();
- switch (state_) {
-
- case SIMIX_DONE:
- /* do nothing, synchro done */
- XBT_DEBUG("ExecImpl::finish(): execution successful");
- break;
-
- case SIMIX_FAILED:
- XBT_DEBUG("ExecImpl::finish(): host '%s' failed", simcall->issuer->get_host()->get_cname());
- simcall->issuer->context_->iwannadie = true;
- if (simcall->issuer->get_host()->is_on())
- simcall->issuer->exception_ =
- std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
- /* else, the actor will be killed with no possibility to survive */
- break;
-
- case SIMIX_CANCELED:
- XBT_DEBUG("ExecImpl::finish(): execution canceled");
- simcall->issuer->exception_ =
- std::make_exception_ptr(simgrid::CancelException(XBT_THROW_POINT, "Execution Canceled"));
- break;
-
- case SIMIX_TIMEOUT:
- XBT_DEBUG("ExecImpl::finish(): execution timeouted");
- simcall->issuer->exception_ = std::make_exception_ptr(simgrid::TimeoutError(XBT_THROW_POINT, "Timeouted"));
- break;
-
- default:
- xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
- }
- simcall->issuer->waiting_synchro = nullptr;
- simcall_execution_wait__set__result(simcall, state_);
+ 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())
- SIMIX_simcall_answer(simcall);
+ if (simcall->issuer_->get_host()->is_on())
+ simcall->issuer_->simcall_answer();
else
- simcall->issuer->context_->iwannadie = true;
+ simcall->issuer_->set_wannadie();
}
}
+void ExecImpl::reset()
+{
+ clear_hosts();
+ set_start_time(-1.0);
+}
+
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);
/*************
* Callbacks *
*************/
-xbt::signal<void(ExecImpl&)> ExecImpl::on_creation;
-xbt::signal<void(ExecImpl const&)> ExecImpl::on_completion;
xbt::signal<void(ExecImpl const&, s4u::Host*)> ExecImpl::on_migration;
-} // namespace activity
-} // namespace kernel
-} // namespace simgrid
+} // namespace simgrid::kernel::activity