Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
add support for CancelException
[simgrid.git] / src / simix / smx_network.cpp
index ed7189c..7916873 100644 (file)
@@ -7,64 +7,15 @@
 #include "simgrid/Exception.hpp"
 #include "src/kernel/activity/MailboxImpl.hpp"
 #include "src/mc/mc_replay.hpp"
-#include "src/simix/smx_network_private.hpp"
 #include "src/simix/smx_private.hpp"
 #include "src/surf/cpu_interface.hpp"
 #include "src/surf/network_interface.hpp"
 
-#include <boost/circular_buffer.hpp>
 #include <boost/range/algorithm.hpp>
 
 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix, "SIMIX network-related synchronization");
 
 static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall);
-static void SIMIX_comm_copy_data(smx_activity_t comm);
-static void SIMIX_comm_start(simgrid::kernel::activity::CommImplPtr synchro);
-
-/**
- *  @brief Checks if there is a communication activity queued in a deque matching our needs
- *  @param deque where to search into
- *  @param type The type of communication we are looking for (comm_send, comm_recv)
- *  @param match_fun the function to apply
- *  @param this_user_data additional parameter to the match_fun
- *  @param my_synchro what to compare against
- *  @param remove_matching whether or not to clean the found object from the queue
- *  @return The communication activity if found, nullptr otherwise
- */
-static simgrid::kernel::activity::CommImplPtr
-_find_matching_comm(boost::circular_buffer_space_optimized<smx_activity_t>* deque, e_smx_comm_type_t type,
-                    int (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), void* this_user_data,
-                    simgrid::kernel::activity::CommImplPtr my_synchro, bool remove_matching)
-{
-  void* other_user_data = nullptr;
-
-  for(auto it = deque->begin(); it != deque->end(); it++){
-    simgrid::kernel::activity::CommImplPtr comm =
-        boost::dynamic_pointer_cast<simgrid::kernel::activity::CommImpl>(std::move(*it));
-
-    if (comm->type == SIMIX_COMM_SEND) {
-      other_user_data = comm->src_data_;
-    } else if (comm->type == SIMIX_COMM_RECEIVE) {
-      other_user_data = comm->dst_data_;
-    }
-    if (comm->type == type && (match_fun == nullptr || match_fun(this_user_data, other_user_data, comm.get())) &&
-        (not comm->match_fun || comm->match_fun(other_user_data, this_user_data, my_synchro.get()))) {
-      XBT_DEBUG("Found a matching communication synchro %p", comm.get());
-      if (remove_matching)
-        deque->erase(it);
-#if SIMGRID_HAVE_MC
-      comm->mbox_cpy = comm->mbox;
-#endif
-      comm->mbox = nullptr;
-      return comm;
-    }
-    XBT_DEBUG("Sorry, communication synchro %p does not match our needs:"
-              " its type is %d but we are looking for a comm of type %d (or maybe the filtering didn't match)",
-              comm.get(), (int)comm->type, (int)type);
-  }
-  XBT_DEBUG("No matching communication synchro found");
-  return nullptr;
-}
 
 /******************************************************************************/
 /*                          Communication synchros                            */
@@ -81,6 +32,7 @@ XBT_PRIVATE void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_actor_t sr
   SIMCALL_SET_MC_VALUE(simcall, 0);
   simcall_HANDLER_comm_wait(simcall, comm, timeout);
 }
+
 XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend(
     smx_simcall_t /*simcall*/, smx_actor_t src_proc, smx_mailbox_t mbox, double task_size, double rate, void* src_buff,
     size_t src_buff_size, int (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
@@ -91,15 +43,16 @@ XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend(
   XBT_DEBUG("send from mailbox %p", mbox);
 
   /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
-  simgrid::kernel::activity::CommImplPtr this_comm =
-      simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl(SIMIX_COMM_SEND));
+  simgrid::kernel::activity::CommImplPtr this_comm = simgrid::kernel::activity::CommImplPtr(
+      new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::SEND));
 
   /* Look for communication synchro matching our needs. We also provide a description of
    * ourself so that the other side also gets a chance of choosing if it wants to match with us.
    *
    * If it is not found then push our communication into the rendez-vous point */
   simgrid::kernel::activity::CommImplPtr other_comm =
-      _find_matching_comm(&mbox->comm_queue_, SIMIX_COMM_RECEIVE, match_fun, data, this_comm, /*remove_matching*/ true);
+      mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::RECEIVE, match_fun, data, this_comm,
+                               /*done*/ false, /*remove_matching*/ true);
 
   if (not other_comm) {
     other_comm = std::move(this_comm);
@@ -118,7 +71,7 @@ XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend(
     XBT_DEBUG("Receive already pushed");
 
     other_comm->state_ = SIMIX_READY;
-    other_comm->type = SIMIX_COMM_READY;
+    other_comm->type   = simgrid::kernel::activity::CommImpl::Type::READY;
   }
   src_proc->comms.push_back(other_comm);
 
@@ -146,7 +99,8 @@ XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend(
     return (detached ? nullptr : other_comm);
   }
 
-  SIMIX_comm_start(other_comm);
+  other_comm->start();
+
   return (detached ? nullptr : other_comm);
 }
 
@@ -156,7 +110,8 @@ XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t re
                                            void (*copy_data_fun)(smx_activity_t, void*, size_t), void* data,
                                            double timeout, double rate)
 {
-  smx_activity_t comm = SIMIX_comm_irecv(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate);
+  smx_activity_t comm = simcall_HANDLER_comm_irecv(simcall, receiver, mbox, dst_buff, dst_buff_size, match_fun,
+                                                   copy_data_fun, data, rate);
   SIMCALL_SET_MC_VALUE(simcall, 0);
   simcall_HANDLER_comm_wait(simcall, comm, timeout);
 }
@@ -167,17 +122,8 @@ XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/,
                                                       void (*copy_data_fun)(smx_activity_t, void*, size_t), void* data,
                                                       double rate)
 {
-  return SIMIX_comm_irecv(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate);
-}
-
-smx_activity_t
-SIMIX_comm_irecv(smx_actor_t dst_proc, smx_mailbox_t mbox, void* dst_buff, size_t* dst_buff_size,
-                 int (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
-                 void (*copy_data_fun)(smx_activity_t, void*, size_t), // used to copy data if not default one
-                 void* data, double rate)
-{
-  simgrid::kernel::activity::CommImplPtr this_synchro =
-      simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl(SIMIX_COMM_RECEIVE));
+  simgrid::kernel::activity::CommImplPtr this_synchro = simgrid::kernel::activity::CommImplPtr(
+      new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::RECEIVE));
   XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
 
   simgrid::kernel::activity::CommImplPtr other_comm;
@@ -186,8 +132,9 @@ SIMIX_comm_irecv(smx_actor_t dst_proc, smx_mailbox_t mbox, void* dst_buff, size_
 
     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 = _find_matching_comm(&mbox->done_comm_queue_, SIMIX_COMM_SEND, match_fun, data, this_synchro,
-                                     /*remove_matching*/ true);
+    other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
+                                          this_synchro, /*done*/ true,
+                                          /*remove_matching*/ true);
     //if not found, assume the receiver came first, register it to the mailbox in the classical way
     if (not other_comm) {
       XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request into list");
@@ -197,7 +144,7 @@ SIMIX_comm_irecv(smx_actor_t dst_proc, smx_mailbox_t mbox, void* dst_buff, size_
       if (other_comm->surf_action_ && other_comm->remains() < 1e-12) {
         XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
         other_comm->state_ = SIMIX_DONE;
-        other_comm->type = SIMIX_COMM_DONE;
+        other_comm->type   = simgrid::kernel::activity::CommImpl::Type::DONE;
         other_comm->mbox = nullptr;
       }
     }
@@ -208,8 +155,9 @@ SIMIX_comm_irecv(smx_actor_t dst_proc, smx_mailbox_t mbox, void* dst_buff, size_
      * ourself so that the other side also gets a chance of choosing if it wants to match with us.
      *
      * If it is not found then push our communication into the rendez-vous point */
-    other_comm = _find_matching_comm(&mbox->comm_queue_, SIMIX_COMM_SEND, match_fun, data, this_synchro,
-                                     /*remove_matching*/ true);
+    other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
+                                          this_synchro, /*done*/ false,
+                                          /*remove_matching*/ true);
 
     if (other_comm == nullptr) {
       XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue_.size());
@@ -219,13 +167,13 @@ SIMIX_comm_irecv(smx_actor_t dst_proc, smx_mailbox_t mbox, void* dst_buff, size_
       XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
 
       other_comm->state_ = SIMIX_READY;
-      other_comm->type = SIMIX_COMM_READY;
+      other_comm->type   = simgrid::kernel::activity::CommImpl::Type::READY;
     }
-    dst_proc->comms.push_back(other_comm);
+    receiver->comms.push_back(other_comm);
   }
 
   /* Setup communication synchro */
-  other_comm->dst_actor_     = dst_proc;
+  other_comm->dst_actor_     = receiver;
   other_comm->dst_buff_      = dst_buff;
   other_comm->dst_buff_size_ = dst_buff_size;
   other_comm->dst_data_      = data;
@@ -240,45 +188,10 @@ SIMIX_comm_irecv(smx_actor_t dst_proc, smx_mailbox_t mbox, void* dst_buff, size_
     other_comm->state_ = SIMIX_RUNNING;
     return other_comm;
   }
-
-  SIMIX_comm_start(other_comm);
+  other_comm->start();
   return other_comm;
 }
 
-smx_activity_t simcall_HANDLER_comm_iprobe(smx_simcall_t simcall, smx_mailbox_t mbox, int type,
-                                           simix_match_func_t match_fun, void* data)
-{
-  return SIMIX_comm_iprobe(simcall->issuer, mbox, type, match_fun, data);
-}
-
-smx_activity_t SIMIX_comm_iprobe(smx_actor_t dst_proc, smx_mailbox_t mbox, int type, simix_match_func_t match_fun,
-                                 void* data)
-{
-  XBT_DEBUG("iprobe from %p %p", mbox, &mbox->comm_queue_);
-  simgrid::kernel::activity::CommImplPtr this_comm;
-  int smx_type;
-  if(type == 1){
-    this_comm = simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl(SIMIX_COMM_SEND));
-    smx_type = SIMIX_COMM_RECEIVE;
-  } else{
-    this_comm = simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl(SIMIX_COMM_RECEIVE));
-    smx_type = SIMIX_COMM_SEND;
-  }
-  smx_activity_t other_synchro=nullptr;
-  if (mbox->permanent_receiver_ != nullptr && not mbox->done_comm_queue_.empty()) {
-    XBT_DEBUG("first check in the permanent recv mailbox, to see if we already got something");
-    other_synchro = _find_matching_comm(&mbox->done_comm_queue_, (e_smx_comm_type_t)smx_type, match_fun, data,
-                                        this_comm, /*remove_matching*/ false);
-  }
-  if (not other_synchro) {
-    XBT_DEBUG("check if we have more luck in the normal mailbox");
-    other_synchro = _find_matching_comm(&mbox->comm_queue_, (e_smx_comm_type_t)smx_type, match_fun, data, this_comm,
-                                        /*remove_matching*/ false);
-  }
-
-  return other_synchro;
-}
-
 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_activity_t synchro, double timeout)
 {
   /* Associate this simcall to the wait synchro */
@@ -438,50 +351,6 @@ void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall)
   }
 }
 
-/**
- *  @brief Starts the simulation of a communication synchro.
- *  @param comm the communication that will be started
- */
-static inline void SIMIX_comm_start(simgrid::kernel::activity::CommImplPtr comm)
-{
-  /* If both the sender and the receiver are already there, start the communication */
-  if (comm->state_ == SIMIX_READY) {
-
-    simgrid::s4u::Host* sender   = comm->src_actor_->host_;
-    simgrid::s4u::Host* receiver = comm->dst_actor_->host_;
-
-    comm->surf_action_ = surf_network_model->communicate(sender, receiver, comm->task_size_, comm->rate_);
-    comm->surf_action_->set_data(comm.get());
-    comm->state_ = SIMIX_RUNNING;
-
-    XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p)", comm.get(), sender->get_cname(),
-              receiver->get_cname(), comm->surf_action_);
-
-    /* If a link is failed, detect it immediately */
-    if (comm->surf_action_->get_state() == simgrid::kernel::resource::Action::State::FAILED) {
-      XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", sender->get_cname(),
-                receiver->get_cname());
-      comm->state_ = SIMIX_LINK_FAILURE;
-      comm->cleanupSurf();
-    }
-
-    /* If any of the process is suspended, create the synchro but stop its execution,
-       it will be restarted when the sender process resume */
-    if (comm->src_actor_->is_suspended() || comm->dst_actor_->is_suspended()) {
-      if (comm->src_actor_->is_suspended())
-        XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
-                  "communication",
-                  comm->src_actor_->get_cname(), comm->src_actor_->host_->get_cname());
-      else
-        XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
-                  "communication",
-                  comm->dst_actor_->get_cname(), comm->dst_actor_->host_->get_cname());
-
-      comm->surf_action_->suspend();
-    }
-  }
-}
-
 /**
  * @brief Answers the SIMIX simcalls associated to a communication synchro.
  * @param synchro a finished communication synchro
@@ -520,7 +389,7 @@ void SIMIX_comm_finish(smx_activity_t synchro)
 
     /* Check out for errors */
 
-    if (simcall->issuer->host_->is_off()) {
+    if (not simcall->issuer->host_->is_on()) {
       simcall->issuer->context_->iwannadie = true;
       simcall->issuer->exception =
           std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
@@ -529,7 +398,7 @@ void SIMIX_comm_finish(smx_activity_t synchro)
 
         case SIMIX_DONE:
           XBT_DEBUG("Communication %p complete!", synchro.get());
-          SIMIX_comm_copy_data(synchro);
+          comm->copy_data();
           break;
 
         case SIMIX_SRC_TIMEOUT:
@@ -577,9 +446,11 @@ void SIMIX_comm_finish(smx_activity_t synchro)
 
         case SIMIX_CANCELED:
           if (simcall->issuer == comm->dst_actor_)
-            SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Communication canceled by the sender");
+            simcall->issuer->exception = std::make_exception_ptr(
+                simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
           else
-            SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Communication canceled by the receiver");
+            simcall->issuer->exception = std::make_exception_ptr(
+                simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
           break;
 
         default:
@@ -614,13 +485,9 @@ void SIMIX_comm_finish(smx_activity_t synchro)
       } catch (simgrid::NetworkFailureException& e) {
         e.value                    = rank;
         simcall->issuer->exception = std::make_exception_ptr(e);
-      } catch (xbt_ex& e) {
-        if (e.category == cancel_error) {
-          e.value                    = rank;
-          simcall->issuer->exception = std::make_exception_ptr(e);
-        } else {
-          xbt_die("Unexpected xbt_ex(%s). Please enhance this code", xbt_ex_catname(e.category));
-        }
+      } catch (simgrid::CancelException& e) {
+        e.value                    = rank;
+        simcall->issuer->exception = std::make_exception_ptr(e);
       }
     }
 
@@ -640,32 +507,13 @@ void SIMIX_comm_finish(smx_activity_t synchro)
       }
     }
 
-    if (simcall->issuer->host_->is_off())
-      simcall->issuer->context_->iwannadie = true;
-    else
+    if (simcall->issuer->host_->is_on())
       SIMIX_simcall_answer(simcall);
+    else
+      simcall->issuer->context_->iwannadie = true;
   }
 }
 
-/******************************************************************************/
-/*                    SIMIX_comm_copy_data callbacks                       */
-/******************************************************************************/
-static void (*SIMIX_comm_copy_data_callback) (smx_activity_t, void*, size_t) = &SIMIX_comm_copy_pointer_callback;
-
-void SIMIX_comm_set_copy_data_callback(void (*callback) (smx_activity_t, void*, size_t))
-{
-  SIMIX_comm_copy_data_callback = callback;
-}
-
-void SIMIX_comm_copy_pointer_callback(smx_activity_t synchro, void* buff, size_t buff_size)
-{
-  simgrid::kernel::activity::CommImplPtr comm =
-      boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
-
-  xbt_assert((buff_size == sizeof(void *)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
-  *(void**)(comm->dst_buff_) = buff;
-}
-
 void SIMIX_comm_copy_buffer_callback(smx_activity_t synchro, void* buff, size_t buff_size)
 {
   simgrid::kernel::activity::CommImplPtr comm =
@@ -678,41 +526,3 @@ void SIMIX_comm_copy_buffer_callback(smx_activity_t synchro, void* buff, size_t
     comm->src_buff_ = nullptr;
   }
 }
-
-/**
- *  @brief Copy the communication data from the sender's buffer to the receiver's one
- *  @param synchro The communication
- */
-void SIMIX_comm_copy_data(smx_activity_t synchro)
-{
-  simgrid::kernel::activity::CommImplPtr comm =
-      boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
-
-  size_t buff_size = comm->src_buff_size_;
-  /* If there is no data to copy then return */
-  if (not comm->src_buff_ || not comm->dst_buff_ || comm->copied)
-    return;
-
-  XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", comm.get(),
-            comm->src_actor_ ? comm->src_actor_->host_->get_cname() : "a finished process", comm->src_buff_,
-            comm->dst_actor_ ? comm->dst_actor_->host_->get_cname() : "a finished process", comm->dst_buff_, buff_size);
-
-  /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
-  if (comm->dst_buff_size_)
-    buff_size = std::min(buff_size, *(comm->dst_buff_size_));
-
-  /* Update the receiver's buffer size to the copied amount */
-  if (comm->dst_buff_size_)
-    *comm->dst_buff_size_ = buff_size;
-
-  if (buff_size > 0){
-      if(comm->copy_data_fun)
-        comm->copy_data_fun(comm, comm->src_buff_, buff_size);
-      else
-        SIMIX_comm_copy_data_callback(comm, comm->src_buff_, buff_size);
-  }
-
-  /* Set the copied flag so we copy data only once */
-  /* (this function might be called from both communication ends) */
-  comm->copied = 1;
-}