Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
factor simcall/activity binding
[simgrid.git] / src / kernel / activity / ExecImpl.cpp
index 5da186d..8e1c3e8 100644 (file)
 /* 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 "src/kernel/activity/ExecImpl.hpp"
+#include "simgrid/Exception.hpp"
 #include "simgrid/modelchecker.h"
 #include "src/mc/mc_replay.hpp"
-
-#include "src/kernel/activity/ExecImpl.hpp"
-#include "src/simix/smx_host_private.hpp"
-#include "src/surf/surf_interface.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);
-namespace simgrid {
-namespace kernel {
-namespace activity {
 
-ExecImpl::ExecImpl(std::string name, std::string tracing_category, resource::Action* timeout_detector, s4u::Host* host)
-    : ActivityImpl(name), host_(host), timeout_detector_(timeout_detector)
+void simcall_HANDLER_execution_wait(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro)
 {
-  this->state_ = SIMIX_RUNNING;
-  this->set_category(tracing_category);
+  XBT_DEBUG("Wait for execution of synchro %p, state %d", synchro, (int)synchro->state_);
+
+  /* Associate this simcall to the synchro */
+  synchro->register_simcall(simcall);
 
-  if (timeout_detector != nullptr)
-    timeout_detector_->set_data(this);
+  /* set surf's synchro */
+  if (MC_is_active() || MC_record_replay_is_active()) {
+    synchro->state_ = SIMIX_DONE;
+    synchro->finish();
+    return;
+  }
 
-  XBT_DEBUG("Create exec %p", this);
+  /* If the synchro is already finished then perform the error handling */
+  if (synchro->state_ != SIMIX_RUNNING)
+    synchro->finish();
+}
+
+void simcall_HANDLER_execution_test(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* synchro)
+{
+  bool res = (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING);
+  if (res) {
+    synchro->simcalls_.push_back(simcall);
+    synchro->finish();
+  } else {
+    SIMIX_simcall_answer(simcall);
+  }
+  simcall_execution_test__set__result(simcall, res);
 }
 
+namespace simgrid {
+namespace kernel {
+namespace activity {
+
 ExecImpl::~ExecImpl()
 {
-  if (surf_action_)
-    surf_action_->unref();
   if (timeout_detector_)
     timeout_detector_->unref();
   XBT_DEBUG("Destroy exec %p", this);
 }
 
-ExecImpl* ExecImpl::start(double flops_amount, double priority, double bound)
+ExecImpl& ExecImpl::set_host(s4u::Host* host)
 {
-  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 (not hosts_.empty())
+    hosts_.clear();
+  hosts_.push_back(host);
+  return *this;
+}
+
+ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
+{
+  hosts_ = hosts;
+  return *this;
+}
+
+ExecImpl& ExecImpl::set_timeout(double timeout)
+{
+  if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) {
+    timeout_detector_ = hosts_.front()->pimpl_cpu->sleep(timeout);
+    timeout_detector_->set_activity(this);
   }
+  return *this;
+}
 
-  XBT_DEBUG("Create execute synchro %p: %s", this, name_.c_str());
-  ExecImpl::on_creation(this);
-  return this;
+ExecImpl& ExecImpl::set_flops_amount(double flops_amount)
+{
+  if (not flops_amounts_.empty())
+    flops_amounts_.clear();
+  flops_amounts_.push_back(flops_amount);
+  return *this;
+}
+
+ExecImpl& ExecImpl::set_flops_amounts(const std::vector<double>& flops_amounts)
+{
+  flops_amounts_ = flops_amounts;
+  return *this;
+}
+
+ExecImpl& ExecImpl::set_bytes_amounts(const std::vector<double>& bytes_amounts)
+{
+  bytes_amounts_ = bytes_amounts;
+
+  return *this;
 }
 
-void ExecImpl::cancel()
+ExecImpl* ExecImpl::start()
 {
-  XBT_VERB("This exec %p is canceled", this);
-  if (surf_action_ != nullptr)
-    surf_action_->cancel();
+  state_ = SIMIX_RUNNING;
+  if (not MC_is_active() && not MC_record_replay_is_active()) {
+    if (hosts_.size() == 1) {
+      surf_action_ = hosts_.front()->pimpl_cpu->execution_start(flops_amounts_.front());
+      surf_action_->set_priority(priority_);
+      surf_action_->set_category(get_tracing_category());
+
+      if (bound_ > 0)
+        surf_action_->set_bound(bound_);
+    } else {
+      surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
+    }
+    surf_action_->set_activity(this);
+  }
+
+  XBT_DEBUG("Create execute synchro %p: %s", this, get_cname());
+  ExecImpl::on_creation(*this);
+  return this;
 }
 
-double ExecImpl::get_remaining()
+double ExecImpl::get_seq_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;
+  return (surf_action_ == nullptr) ? 0 : surf_action_->get_remains() / surf_action_->get_cost();
 }
 
-double ExecImpl::get_remaining_ratio()
+double ExecImpl::get_par_remaining_ratio()
 {
-  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();
+  // parallel task: their remain is already between 0 and 1
+  return (surf_action_ == nullptr) ? 0 : surf_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_priority(double priority)
 {
-  if (surf_action_)
-    surf_action_->set_priority(priority);
+  priority_ = priority;
+  return *this;
 }
 
 void ExecImpl::post()
 {
-  if (host_ && host_->is_off()) { /* FIXME: handle resource failure for parallel tasks too */
+  if (hosts_.size() == 1 && not hosts_.front()->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;
@@ -104,12 +161,10 @@ void ExecImpl::post()
     state_ = SIMIX_DONE;
   }
 
-  on_completion(this);
+  on_completion(*this);
+
+  clean_action();
 
-  if (surf_action_) {
-    surf_action_->unref();
-    surf_action_ = nullptr;
-  }
   if (timeout_detector_) {
     timeout_detector_->unref();
     timeout_detector_ = nullptr;
@@ -117,37 +172,84 @@ void ExecImpl::post()
 
   /* If there are simcalls associated with the synchro, then answer them */
   if (not simcalls_.empty())
-    SIMIX_execution_finish(this);
+    finish();
+}
+
+void ExecImpl::finish()
+{
+  while (not simcalls_.empty()) {
+    smx_simcall_t 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_);
+
+    /* Fail the process if the host is down */
+    if (simcall->issuer->get_host()->is_on())
+      SIMIX_simcall_answer(simcall);
+    else
+      simcall->issuer->context_->iwannadie = true;
+  }
 }
 
-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());
     new_action->set_remains(old_action->get_remains());
-    new_action->set_data(this);
+    new_action->set_activity(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.
 
-    old_action->set_data(nullptr);
+    old_action->set_activity(nullptr);
     old_action->cancel();
     old_action->unref();
     this->surf_action_ = new_action;
   }
 
-  on_migration(this, to);
+  on_migration(*this, to);
   return this;
 }
 
 /*************
  * Callbacks *
  *************/
-xbt::signal<void(ExecImplPtr)> ExecImpl::on_creation;
-xbt::signal<void(ExecImplPtr)> ExecImpl::on_completion;
-xbt::signal<void(ExecImplPtr, s4u::Host*)> ExecImpl::on_migration;
+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