]> AND Public Git Repository - simgrid.git/blobdiff - src/kernel/activity/ExecImpl.cpp
Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
try to match Activity design to Resource design
[simgrid.git] / src / kernel / activity / ExecImpl.cpp
index 80127ed919515376d995656e599b5a9930806686..319d8b02b6a9e0ca47052c0181f9fdc7e101f1d2 100644 (file)
@@ -1,11 +1,14 @@
-/* Copyright (c) 2007-2020. The SimGrid Team. All rights reserved.          */
+/* Copyright (c) 2007-2021. 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 "src/kernel/activity/ExecImpl.hpp"
 #include "simgrid/Exception.hpp"
+#include "simgrid/kernel/routing/NetPoint.hpp"
 #include "simgrid/modelchecker.h"
+#include "simgrid/s4u/Exec.hpp"
+#include "src/kernel/actor/SimcallObserver.hpp"
 #include "src/mc/mc_replay.hpp"
 #include "src/surf/HostImpl.hpp"
 #include "src/surf/cpu_interface.hpp"
 
 #include "simgrid/s4u/Host.hpp"
 
-#include <boost/range/algorithm.hpp>
-
 XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(simix_process);
 
-void simcall_HANDLER_execution_waitany_for(smx_simcall_t simcall, simgrid::kernel::activity::ExecImpl* execs[],
-                                           size_t count, double timeout)
-{
-  if (timeout < 0.0) {
-    simcall->timeout_cb_ = nullptr;
-  } else {
-    simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall, execs, count]() {
-      simcall->timeout_cb_ = nullptr;
-      for (size_t i = 0; i < count; i++) {
-        // Remove the first occurrence of simcall:
-        auto* exec = execs[i];
-        auto j     = boost::range::find(exec->simcalls_, simcall);
-        if (j != exec->simcalls_.end())
-          exec->simcalls_.erase(j);
-      }
-      simcall_execution_waitany_for__set__result(simcall, -1);
-      simcall->issuer_->simcall_answer();
-    });
-  }
-
-  for (size_t i = 0; i < count; i++) {
-    /* associate this simcall to the the synchro */
-    auto* exec = execs[i];
-    exec->simcalls_.push_back(simcall);
-
-    /* see if the synchro is already finished */
-    if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
-        exec->state_ != simgrid::kernel::activity::State::RUNNING) {
-      exec->finish();
-      break;
-    }
-  }
-}
-
 namespace simgrid {
 namespace kernel {
 namespace activity {
 
+ExecImpl::ExecImpl()
+{
+  piface_                = new s4u::Exec(this);
+  actor::ActorImpl* self = actor::ActorImpl::self();
+  if (self) {
+    actor_ = self;
+    self->activities_.emplace_back(this);
+  }
+}
+
 ExecImpl& ExecImpl::set_host(s4u::Host* host)
 {
   hosts_.assign(1, host);
@@ -69,7 +46,7 @@ ExecImpl& ExecImpl::set_hosts(const std::vector<s4u::Host*>& hosts)
 
 ExecImpl& ExecImpl::set_timeout(double timeout)
 {
-  if (timeout > 0 && not MC_is_active() && not MC_record_replay_is_active()) {
+  if (timeout >= 0 && not MC_is_active() && not MC_record_replay_is_active()) {
     timeout_detector_.reset(hosts_.front()->pimpl_cpu->sleep(timeout));
     timeout_detector_->set_activity(this);
   }
@@ -104,10 +81,14 @@ ExecImpl* ExecImpl::start()
       surf_action_->set_sharing_penalty(sharing_penalty_);
       surf_action_->set_category(get_tracing_category());
 
-      if (bound_ > 0)
+      if (bound_ > 0) {
         surf_action_->set_bound(bound_);
+        surf_action_->set_user_bound(bound_);
+      }
     } else {
-      surf_action_ = surf_host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
+      // FIXME[donassolo]: verify if all hosts belongs to the same netZone?
+      auto host_model = hosts_.front()->get_netpoint()->get_englobing_zone()->get_host_model();
+      surf_action_    = host_model->execute_parallel(hosts_, flops_amounts_.data(), bytes_amounts_.data(), -1);
     }
     surf_action_->set_activity(this);
   }
@@ -141,21 +122,33 @@ ExecImpl& ExecImpl::set_sharing_penalty(double sharing_penalty)
 
 void ExecImpl::post()
 {
-  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
+  xbt_assert(surf_action_ != nullptr);
+  if (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 */
     state_ = State::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 */
+  } else if (surf_action_->get_state() == resource::Action::State::FAILED) {
+    /* If all the hosts are running the synchro didn't fail, then the synchro was canceled */
     state_ = State::CANCELED;
   } else if (timeout_detector_ && timeout_detector_->get_state() == resource::Action::State::FINISHED) {
-    state_ = State::TIMEOUT;
+    if (surf_action_->get_remains() > 0.0) {
+      surf_action_->set_state(resource::Action::State::FAILED);
+      state_ = State::TIMEOUT;
+    } else {
+      state_ = State::DONE;
+    }
   } else {
     state_ = State::DONE;
   }
 
+  get_iface()->set_finish_time(surf_action_->get_finish_time());
+
   clean_action();
   timeout_detector_.reset();
+  if (actor_) {
+    actor_->activities_.remove(this);
+    actor_ = nullptr;
+  }
   /* Answer all simcalls associated with the synchro */
   finish();
 }
@@ -170,18 +163,14 @@ void ExecImpl::finish()
      * 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_ == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
-      continue;                        // if process handling comm is killed
-    if (simcall->call_ == SIMCALL_EXECUTION_WAITANY_FOR) {
-      simgrid::kernel::activity::ExecImpl** execs = simcall_execution_waitany_for__get__execs(simcall);
-      size_t count                                = simcall_execution_waitany_for__get__count(simcall);
+    if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
+      continue;                                 // if process handling comm is killed
+    if (auto* observer =
+            dynamic_cast<kernel::actor::ExecutionWaitanySimcall*>(simcall->observer_)) { // simcall is a wait_any?
+      const auto* execs = observer->get_execs();
 
-      for (size_t i = 0; i < count; i++) {
-        // Remove the first occurrence of simcall:
-        auto* exec = execs[i];
-        auto j     = boost::range::find(exec->simcalls_, simcall);
-        if (j != exec->simcalls_.end())
-          exec->simcalls_.erase(j);
+      for (auto* exec : *execs) {
+        exec->unregister_simcall(simcall);
 
         if (simcall->timeout_cb_) {
           simcall->timeout_cb_->remove();
@@ -190,12 +179,11 @@ void ExecImpl::finish()
       }
 
       if (not MC_is_active() && not MC_record_replay_is_active()) {
-        ExecImpl** element = std::find(execs, execs + count, this);
-        int rank           = (element != execs + count) ? element - execs : -1;
-        simcall_execution_waitany_for__set__result(simcall, rank);
+        auto element = std::find(execs->begin(), execs->end(), this);
+        int rank     = element != execs->end() ? static_cast<int>(std::distance(execs->begin(), element)) : -1;
+        observer->set_result(rank);
       }
     }
-
     switch (state_) {
       case State::DONE:
         /* do nothing, synchro done */
@@ -226,7 +214,7 @@ void ExecImpl::finish()
         xbt_die("Internal error in ExecImpl::finish(): unexpected synchro state %d", static_cast<int>(state_));
     }
 
-    simcall->issuer_->waiting_synchro = nullptr;
+    simcall->issuer_->waiting_synchro_ = nullptr;
     /* Fail the process if the host is down */
     if (simcall->issuer_->get_host()->is_on())
       simcall->issuer_->simcall_answer();
@@ -243,9 +231,7 @@ ActivityImpl* ExecImpl::migrate(s4u::Host* to)
     new_action->set_remains(old_action->get_remains());
     new_action->set_activity(this);
     new_action->set_sharing_penalty(old_action->get_sharing_penalty());
-
-    // 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_user_bound(old_action->get_user_bound());
 
     old_action->set_activity(nullptr);
     old_action->cancel();
@@ -257,6 +243,33 @@ ActivityImpl* ExecImpl::migrate(s4u::Host* to)
   return this;
 }
 
+void ExecImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<ExecImpl*>* execs, double timeout)
+{
+  if (timeout < 0.0) {
+    issuer->simcall_.timeout_cb_ = nullptr;
+  } else {
+    issuer->simcall_.timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [issuer, execs]() {
+      issuer->simcall_.timeout_cb_ = nullptr;
+      for (auto* exec : *execs)
+        exec->unregister_simcall(&issuer->simcall_);
+      // default result (-1) is set in mc::ExecutionWaitanySimcall
+      issuer->simcall_answer();
+    });
+  }
+
+  for (auto* exec : *execs) {
+    /* associate this simcall to the the synchro */
+    exec->simcalls_.push_back(&issuer->simcall_);
+
+    /* see if the synchro is already finished */
+    if (exec->state_ != simgrid::kernel::activity::State::WAITING &&
+        exec->state_ != simgrid::kernel::activity::State::RUNNING) {
+      exec->finish();
+      break;
+    }
+  }
+}
+
 /*************
  * Callbacks *
  *************/