X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/44723299849907b8923bfac8f9b111bcc07f0987..a06e616da86a22a21a02fcc9ddaf192cecdef2bc:/src/simix/smx_network.c

diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c
index cdab3e1914..de7af8c0d9 100644
--- a/src/simix/smx_network.c
+++ b/src/simix/smx_network.c
@@ -118,11 +118,19 @@ SIMIX_rdv_get_count_waiting_comm(smx_rdv_t rdv, smx_host_t host)
  *  \param rdv The rendez-vous point
  *  \return The communication or NULL if empty
  */
-smx_comm_t SIMIX_rdv_get_head(smx_rdv_t rdv)
+XBT_INLINE smx_comm_t SIMIX_rdv_get_head(smx_rdv_t rdv)
 {
   return (smx_comm_t)xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
 }
 
+/** @brief adds some API-related data to the rendez-vous point */
+XBT_INLINE void SIMIX_rdv_set_data(smx_rdv_t rdv,void *data) {
+  rdv->data=data;
+}
+/** @brief gets API-related data from the rendez-vous point */
+XBT_INLINE void *SIMIX_rdv_get_data(smx_rdv_t rdv) {
+  return rdv->data;
+}
 
 /******************************************************************************/
 /*                           Communication Requests                           */
@@ -138,7 +146,7 @@ smx_comm_t SIMIX_communication_new(smx_comm_type_t type)
   /* alloc structures */
   smx_comm_t comm = xbt_new0(s_smx_comm_t, 1);
   comm->type = type;
-  comm->cond = SIMIX_cond_init();
+  comm->sem = SIMIX_sem_init(0);
   comm->refcount = 1;
   
   return comm;
@@ -154,16 +162,26 @@ void SIMIX_communication_destroy(smx_comm_t comm)
   if(comm->refcount > 0)
     return;
 
-  if(comm->cond){
-    SIMIX_cond_destroy(comm->cond);
-    comm->cond = NULL;
+  if(comm->sem){
+    SIMIX_sem_destroy(comm->sem);
+    comm->sem = NULL;
   }
   
   if(comm->act){
     SIMIX_action_destroy(comm->act);
     comm->act = NULL;
   }
-  
+
+  if(comm->src_timeout){
+    SIMIX_action_destroy(comm->src_timeout);
+    comm->src_timeout = NULL;
+  }
+
+  if(comm->dst_timeout){
+      SIMIX_action_destroy(comm->dst_timeout);
+      comm->dst_timeout = NULL;
+    }
+
   xbt_free(comm);
 }
 
@@ -194,7 +212,7 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
                                          comm->task_size, comm->rate);
 
     /* If any of the process is suspend, create the action but stop its execution,
-       it will be restart when the sender process resume */
+       it will be restarted when the sender process resume */
     if(SIMIX_process_is_suspended(comm->src_proc) || 
        SIMIX_process_is_suspended(comm->dst_proc)) {
       SIMIX_action_set_priority(comm->act, 0);
@@ -202,8 +220,11 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
     
     /* Add the communication as user data of the action */
     comm->act->data = comm;
-    
-    SIMIX_register_action_to_condition(comm->act, comm->cond);
+
+    /* The semaphore will only get signaled once, but since the first unlocked guy will
+     * release_forever() the semaphore, that will unlock the second (and any other)
+     * communication partner */
+    SIMIX_register_action_to_semaphore(comm->act, comm->sem);
   }
 }
 
@@ -219,39 +240,56 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
  */
 static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
 {
-  xbt_ex_t e;
+  smx_action_t act_sleep = NULL;
+  int src_timeout = 0;
+  int dst_timeout = 0;
 
   DEBUG1("Waiting for the completion of communication %p", comm);
   
-  if(timeout > 0){
-    TRY{
-      SIMIX_cond_wait_timeout(comm->cond, NULL, timeout);
-    }
-    CATCH(e){
-      /* If there is a timeout then cancel the communication if it is running or 
-         remove it from the rendez-vous otherwise. Then signal the other peer,
-         destroy the communication and retrow the exception. */
-      if(e.category == timeout_error){
-        DEBUG1("Communication timeout! %p", comm);
-        if(comm->act && SIMIX_action_get_state(comm->act) == SURF_ACTION_RUNNING)
-          SIMIX_communication_cancel(comm);
-        else
-          SIMIX_rdv_remove(comm->rdv, comm);
-
-        SIMIX_cond_signal(comm->cond);
-        SIMIX_communication_destroy(comm);
-      }
-      RETHROW;
-    }
-  }else{
-    SIMIX_cond_wait(comm->cond, NULL);
+  if (timeout >= 0) {
+    act_sleep = SIMIX_action_sleep(SIMIX_host_self(), timeout);
+		if(SIMIX_process_self()==comm->src_proc)
+			comm->src_timeout = act_sleep;
+		else
+			comm->dst_timeout = act_sleep;
+    SIMIX_action_set_name(act_sleep,bprintf("Timeout for comm %p and wait on semaphore %p (max_duration:%f)", comm, comm->sem,timeout));
+    SIMIX_register_action_to_semaphore(act_sleep, comm->sem);
+    SIMIX_process_self()->waiting_action = act_sleep;
+    SIMIX_sem_block_onto(comm->sem);
+    SIMIX_process_self()->waiting_action = NULL;
+    SIMIX_unregister_action_to_semaphore(act_sleep, comm->sem);
+  } else {
+    SIMIX_sem_acquire(comm->sem);
+  }
+
+  /* Check for timeouts */
+  if ((src_timeout = ((comm->src_timeout) && (SIMIX_action_get_state(comm->src_timeout) == SURF_ACTION_DONE))) ||
+      (dst_timeout = ((comm->dst_timeout) && (SIMIX_action_get_state(comm->dst_timeout) == SURF_ACTION_DONE))) ) {
+			/* Somebody did a timeout! */
+    if (src_timeout) DEBUG1("Communication timeout from the src! %p", comm);
+    if (dst_timeout) DEBUG1("Communication timeout from the dst! %p", comm);
+
+    if(comm->act && SIMIX_action_get_state(comm->act) == SURF_ACTION_RUNNING)
+      SIMIX_communication_cancel(comm);
+    else if (comm->rdv)
+      SIMIX_rdv_remove(comm->rdv, comm);
+
+    /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
+    SIMIX_sem_release_forever(comm->sem);
+    SIMIX_communication_destroy(comm);
+
+    THROW1(timeout_error, 0, "Communication timeouted because of %s",src_timeout?"the source":"the destination");
   }
 
   DEBUG1("Communication %p complete! Let's check for errors", comm);
+
+  /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
+  SIMIX_sem_release_forever(comm->sem);
   
   /* Check for errors other than timeouts (they are catched above) */
   if(!SIMIX_host_get_state(SIMIX_host_self())){
-    SIMIX_rdv_remove(comm->rdv, comm);
+    if(comm->rdv)
+      SIMIX_rdv_remove(comm->rdv, comm);
     SIMIX_communication_destroy(comm);
     THROW0(host_error, 0, "Host failed");
   } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
@@ -264,16 +302,17 @@ static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, doub
  *  \brief Cancels a communication
  *  \brief comm The communication to cancel
  */
-void SIMIX_communication_cancel(smx_comm_t comm)
+XBT_INLINE void SIMIX_communication_cancel(smx_comm_t comm)
 {
-  SIMIX_action_cancel(comm->act);
+  if (comm->act)
+    SIMIX_action_cancel(comm->act);
 }
 
 /**
  *  \brief get the amount remaining from the communication
  *  \param comm The communication
  */
-double SIMIX_communication_get_remains(smx_comm_t comm)
+XBT_INLINE double SIMIX_communication_get_remains(smx_comm_t comm)
 {
   return SIMIX_action_get_remains(comm->act);
 }  
@@ -288,23 +327,33 @@ void SIMIX_network_copy_data(smx_comm_t comm)
   if(!comm->src_buff || !comm->dst_buff)
     return;
   
-  size_t src_buff_size = comm->src_buff_size;
-  size_t dst_buff_size = *comm->dst_buff_size;
-  
   /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
-  dst_buff_size = MIN(dst_buff_size, src_buff_size);
+  size_t buff_size = comm->src_buff_size;
+  if (comm->dst_buff_size)
+    buff_size = MIN(buff_size,*(comm->dst_buff_size));
   
   /* Update the receiver's buffer size to the copied amount */
-  *comm->dst_buff_size = dst_buff_size;
+  if (comm->dst_buff_size)
+    *comm->dst_buff_size = buff_size;
 
-  if(dst_buff_size == 0)
+  if(buff_size == 0)
     return;
 
-  memcpy(comm->dst_buff, comm->src_buff, dst_buff_size);
-
-  DEBUG4("Copying comm %p data from %s -> %s (%zu bytes)", 
-         comm, comm->src_proc->smx_host->name, comm->dst_proc->smx_host->name,
-         dst_buff_size);
+#ifdef HAVE_RUBY /* FIXME: KILLME */
+  INFO6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
+      comm,
+      comm->src_proc->smx_host->name, comm->src_buff,
+      comm->dst_proc->smx_host->name, comm->dst_buff,
+      buff_size);
+#else
+  DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
+      comm,
+      comm->src_proc->smx_host->name, comm->src_buff,
+      comm->dst_proc->smx_host->name, comm->dst_buff,
+      buff_size);
+#endif
+
+  memcpy(comm->dst_buff, comm->src_buff, buff_size);
 }
 
 /**
@@ -312,7 +361,7 @@ void SIMIX_network_copy_data(smx_comm_t comm)
  *  \param comm The communication
  *  \return the user data
  */
-void *SIMIX_communication_get_data(smx_comm_t comm)
+XBT_INLINE void *SIMIX_communication_get_data(smx_comm_t comm)
 {
   return comm->data;
 }
@@ -329,20 +378,50 @@ void *SIMIX_communication_get_data(smx_comm_t comm)
  *  \param timeout The timeout used for the waiting the completion 
  *  \param src_buff The source buffer containing the message to be sent
  *  \param src_buff_size The size of the source buffer
- *  \param comm_ref The communication object used for the send
+ *  \param comm_ref The communication object used for the send  (useful if someone else wants to cancel this communication afterward)
  *  \param data User data associated to the communication object
  *  Throws:
  *   - host_error if peer failed
  *   - timeout_error if communication reached the timeout specified
  *   - network_error if network failed or peer issued a timeout
  */
-void SIMIX_network_send(smx_rdv_t rdv, double task_size, double rate, 
+XBT_INLINE void SIMIX_network_send(smx_rdv_t rdv, double task_size, double rate,
                         double timeout, void *src_buff, size_t src_buff_size,
                         smx_comm_t *comm_ref, void *data)
+{
+  *comm_ref = SIMIX_network_isend(rdv,task_size,rate,src_buff,src_buff_size,data);
+  SIMIX_network_wait(*comm_ref,timeout);
+}
+
+/**
+ *  \brief Put a receive communication request in a rendez-vous point and waits
+ *         for its completion (blocking)
+ *  \param rdv The rendez-vous point
+ *  \param timeout The timeout used for the waiting the completion 
+ *  \param dst_buff The destination buffer to copy the received message
+ *  \param src_buff_size The size of the destination buffer
+ *  \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
+ *  Throws:
+ *   - host_error if peer failed
+ *   - timeout_error if communication reached the timeout specified
+ *   - network_error if network failed or peer issued a timeout
+ */
+XBT_INLINE void SIMIX_network_recv(smx_rdv_t rdv, double timeout, void *dst_buff,
+                        size_t *dst_buff_size, smx_comm_t *comm_ref)
+{
+  *comm_ref = SIMIX_network_irecv(rdv,dst_buff,dst_buff_size);
+  SIMIX_network_wait(*comm_ref,timeout);
+}
+
+/******************************************************************************/
+/*                        Asynchronous Communication                          */
+/******************************************************************************/
+smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size, double rate,
+    void *src_buff, size_t src_buff_size, void *data)
 {
   smx_comm_t comm;
-  
-  /* Look for communication request matching our needs. 
+
+  /* Look for communication request matching our needs.
      If it is not found then create it and push it into the rendez-vous point */
   comm = SIMIX_rdv_get_request(rdv, comm_recv);
 
@@ -351,9 +430,6 @@ void SIMIX_network_send(smx_rdv_t rdv, double task_size, double rate,
     SIMIX_rdv_push(rdv, comm);
   }
 
-  /* Update the communication reference with the comm to be used */
-  *comm_ref = comm;
-  
   /* Setup the communication request */
   comm->src_proc = SIMIX_process_self();
   comm->task_size = task_size;
@@ -363,32 +439,13 @@ void SIMIX_network_send(smx_rdv_t rdv, double task_size, double rate,
   comm->data = data;
 
   SIMIX_communication_start(comm);
-
-  /* Wait for communication completion */
-  SIMIX_communication_wait_for_completion(comm, timeout);
-
-  SIMIX_communication_destroy(comm);
+  return comm;
 }
 
-/**
- *  \brief Put a receive communication request in a rendez-vous point and waits
- *         for its completion (blocking)
- *  \param rdv The rendez-vous point
- *  \param timeout The timeout used for the waiting the completion 
- *  \param dst_buff The destination buffer to copy the received message
- *  \param src_buff_size The size of the destination buffer
- *  \param comm_ref The communication object used for the send
- *  Throws:
- *   - host_error if peer failed
- *   - timeout_error if communication reached the timeout specified
- *   - network_error if network failed or peer issued a timeout
- */
-void SIMIX_network_recv(smx_rdv_t rdv, double timeout, void *dst_buff, 
-                        size_t *dst_buff_size, smx_comm_t *comm_ref)
-{
+smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_size) {
   smx_comm_t comm;
 
-  /* Look for communication request matching our needs. 
+  /* Look for communication request matching our needs.
      If it is not found then create it and push it into the rendez-vous point */
   comm = SIMIX_rdv_get_request(rdv, comm_send);
 
@@ -397,40 +454,63 @@ void SIMIX_network_recv(smx_rdv_t rdv, double timeout, void *dst_buff,
     SIMIX_rdv_push(rdv, comm);
   }
 
-  /* Update the communication reference with the comm to be used */
-  *comm_ref = comm;
- 
   /* Setup communication request */
   comm->dst_proc = SIMIX_process_self();
   comm->dst_buff = dst_buff;
   comm->dst_buff_size = dst_buff_size;
 
   SIMIX_communication_start(comm);
+  return comm;
+}
 
+/** @brief blocks until the communication terminates or the timeout occurs */
+XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout) {
   /* Wait for communication completion */
   SIMIX_communication_wait_for_completion(comm, timeout);
 
   SIMIX_communication_destroy(comm);
 }
 
-/******************************************************************************/
-/*                        Asynchronous Communication                          */
-/******************************************************************************/
-
-/*
-void SIMIX_network_wait(smx_action_t comm, double timeout)
-{
-  TO BE IMPLEMENTED
+/** @Returns whether the (asynchronous) communication is done yet or not */
+XBT_INLINE int SIMIX_network_test(smx_comm_t comm) {
+  return comm->sem?SIMIX_sem_would_block(comm->sem):0;
 }
 
-XBT_PUBLIC(int) SIMIX_network_test(smx_action_t comm)
-{
-  TO BE IMPLEMENTED
-}*/
+/** @brief wait for the completion of any communication of a set
+ *
+ *  @Returns the rank in the dynar of communication which finished; destroy it after identifying which one it is
+ */
+unsigned int SIMIX_network_waitany(xbt_dynar_t comms) {
+  xbt_dynar_t sems = xbt_dynar_new(sizeof(smx_sem_t),NULL);
+  unsigned int cursor, found_comm=-1;
+  smx_comm_t comm,comm_finished=NULL;
 
+  xbt_dynar_foreach(comms,cursor,comm){
+    xbt_dynar_push(sems,&(comm->sem));
+  }
 
+  DEBUG1("Waiting for the completion of communication set %p", comms);
 
+  found_comm = SIMIX_sem_acquire_any(sems);
+  xbt_assert0(found_comm!=-1,"Cannot find which communication finished");
+  xbt_dynar_get_cpy(comms,found_comm,&comm_finished);
 
+  DEBUG1("Communication %p complete! Let's check for errors", comm_finished);
 
+  /* Make sure that everyone sleeping on that semaphore is awake,
+   * and that nobody will ever block on it */
+  SIMIX_sem_release_forever(comm_finished->sem);
 
+  /* Check for errors */
+  if(!SIMIX_host_get_state(SIMIX_host_self())){
+    if(comm_finished->rdv)
+      SIMIX_rdv_remove(comm_finished->rdv, comm_finished);
+    SIMIX_communication_destroy(comm_finished);
+    THROW0(host_error, 0, "Host failed");
+  } else if (SIMIX_action_get_state(comm_finished->act) == SURF_ACTION_FAILED){
+    SIMIX_communication_destroy(comm_finished);
+    THROW0(network_error, 0, "Link failure");
+  }
 
+  return found_comm;
+}