X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/62923a33166a3bca4b65351fbe26f170e42dd24f..deba0f7f0441a96f742af845cfceabf340d0f27f:/src/simix/smx_network.c

diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c
index 0f2bd0b4d5..4987f8f20e 100644
--- a/src/simix/smx_network.c
+++ b/src/simix/smx_network.c
@@ -1,13 +1,15 @@
-/* 	$Id$	 */
-
-/* Copyright (c) 2009 Cristian Rosa.
-   All rights reserved.                                          */
+/* Copyright (c) 2009, 2010. 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 "private.h"
 #include "xbt/log.h"
+#include "xbt/dict.h"
+
+/* Pimple to get an histogram of message sizes in the simulation */
+xbt_dict_t msg_sizes = NULL;
 
 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
                                 "Logging specific to SIMIX (network)");
@@ -73,8 +75,7 @@ static inline void SIMIX_rdv_remove(smx_rdv_t rdv, smx_comm_t comm)
  *  \param type The type of communication we are looking for (comm_send, comm_recv)
  *  \return The communication request if found, NULL otherwise.
  */
-smx_comm_t SIMIX_rdv_get_request(smx_rdv_t rdv, smx_comm_type_t type)
-{
+smx_comm_t SIMIX_rdv_get_request(smx_rdv_t rdv, smx_comm_type_t type) {
   smx_comm_t comm = (smx_comm_t)xbt_fifo_get_item_content(
                                   xbt_fifo_get_first_item(rdv->comm_fifo));
 
@@ -86,7 +87,6 @@ smx_comm_t SIMIX_rdv_get_request(smx_rdv_t rdv, smx_comm_type_t type)
     return comm;
   }
 
-  /* no relevant request found. Return NULL */
   DEBUG0("Communication request not found");
   return NULL;
 }
@@ -118,17 +118,17 @@ 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 */
-void SIMIX_rdv_set_data(smx_rdv_t rdv,void *data) {
+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 */
-void *SIMIX_rdv_get_data(smx_rdv_t rdv) {
+XBT_INLINE void *SIMIX_rdv_get_data(smx_rdv_t rdv) {
   return rdv->data;
 }
 
@@ -171,7 +171,17 @@ void SIMIX_communication_destroy(smx_comm_t comm)
     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);
 }
 
@@ -200,12 +210,15 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
     comm->act = SIMIX_action_communicate(comm->src_proc->smx_host, 
                                          comm->dst_proc->smx_host, NULL, 
                                          comm->task_size, comm->rate);
+#ifdef HAVE_TRACING
+    TRACE_smx_action_communicate (comm->act, comm->src_proc);
+#endif
 
     /* If any of the process is suspend, create the action but stop its execution,
        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);
+      SIMIX_action_suspend(comm->act);
     }
     
     /* Add the communication as user data of the action */
@@ -224,48 +237,60 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
  *  \param timeout The max amount of time to wait for the communication to finish
  *
  *  Throws:
- *   - host_error if peer failed
- *   - timeout_error if communication reached the timeout specified
- *   - network_error if network failed or peer issued a timeout
+ *   - host_error if local peer failed
+ *   - timeout_error if communication reached the timeout specified (either because of local peer or remote peer)
+ *   - network_error if network failed or remote peer failed
  */
 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_sem_acquire_timeout(comm->sem, 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);
-
-        /* 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);
-      }
-      RETHROW;
-    }
+  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())){
+  if (!SIMIX_host_get_state(SIMIX_host_self())){
     if(comm->rdv)
       SIMIX_rdv_remove(comm->rdv, comm);
     SIMIX_communication_destroy(comm);
@@ -273,55 +298,109 @@ static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, doub
   } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
     SIMIX_communication_destroy(comm);
     THROW0(network_error, 0, "Link failure");
+  } else if (!SIMIX_host_get_state(SIMIX_process_get_host(comm->dst_proc)) ||
+      !SIMIX_host_get_state(SIMIX_process_get_host(comm->src_proc))) {
+    /* We test both src&dst because we dunno who we are today, and we already tested myself above.
+     *    So, at the end, we test the remote peer only
+     * Moreover, we have to test it because if the remote peer fails, the action comm->act is not done nor failed.
+     *    In that case, we got awaken by the little endless actions created in the SIMIX_sem_acquire(comm->sem)
+     *    at the beginning of this function. */
+    SIMIX_communication_destroy(comm);
+    THROW0(network_error, 0, "Remote peer failed");
+
   }
+  SIMIX_communication_destroy(comm);
 }
 
 /**
  *  \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);
 }  
 
+/******************************************************************************/
+/*                    SIMIX_network_copy_data callbacks                       */
+/******************************************************************************/
+static void (*SIMIX_network_copy_data_callback)(smx_comm_t, size_t) = &SIMIX_network_copy_pointer_callback;
+
+void SIMIX_network_set_copy_data_callback(void (*callback)(smx_comm_t, size_t)) {
+  SIMIX_network_copy_data_callback = callback;
+}
+
+void SIMIX_network_copy_pointer_callback(smx_comm_t comm, size_t buff_size) {
+  xbt_assert1((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)",buff_size);
+  *(void**)(comm->dst_buff) = comm->src_buff;
+}
+
+void SIMIX_network_copy_buffer_callback(smx_comm_t comm, size_t buff_size) {
+  memcpy(comm->dst_buff, comm->src_buff, buff_size);
+}
+
 /**
  *  \brief Copy the communication data from the sender's buffer to the receiver's one
  *  \param comm The communication
  */
 void SIMIX_network_copy_data(smx_comm_t comm)
 {
+  size_t buff_size = comm->src_buff_size;
+
+  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);
+
   /* If there is no data to be copy then return */
   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);
+  if (comm->dst_buff_size)
+    buff_size = 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 = dst_buff_size;
+    *comm->dst_buff_size = buff_size;
 
-  if(dst_buff_size == 0)
+  if(buff_size == 0)
     return;
+  (*SIMIX_network_copy_data_callback)(comm, buff_size);
 
-  memcpy(comm->dst_buff, comm->src_buff, dst_buff_size);
+  /* pimple to display the message sizes */
+  {
+    if (msg_sizes == NULL)
+      msg_sizes = xbt_dict_new();
+    uintptr_t casted_size = comm->task_size;
+    uintptr_t amount = xbt_dicti_get(msg_sizes, casted_size);
+    amount++;
 
-  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);
+    xbt_dicti_set(msg_sizes,casted_size, amount);
+  }
+}
+#include "xbt.h"
+/* pimple to display the message sizes */
+void SIMIX_message_sizes_output(const char *filename) {
+  FILE * out = fopen(filename,"w");
+  xbt_assert1(out,"Cannot open file %s",filename);
+  uintptr_t key,data;
+  xbt_dict_cursor_t cursor;
+  xbt_dict_foreach(msg_sizes,cursor,key,data) {
+    fprintf(out,"%zu %zu\n",key,data);
+  }
+  fclose(out);
 }
 
 /**
@@ -329,7 +408,7 @@ 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;
 }
@@ -353,7 +432,7 @@ void *SIMIX_communication_get_data(smx_comm_t comm)
  *   - 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)
 {
@@ -374,7 +453,7 @@ void SIMIX_network_send(smx_rdv_t rdv, double task_size, double rate,
  *   - 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, 
+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);
@@ -432,27 +511,24 @@ smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_s
 }
 
 /** @brief blocks until the communication terminates or the timeout occurs */
-void SIMIX_network_wait(smx_comm_t comm, double timeout) {
+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);
 }
 
 /** @Returns whether the (asynchronous) communication is done yet or not */
-int SIMIX_network_test(smx_comm_t comm) {
+XBT_INLINE int SIMIX_network_test(smx_comm_t comm) {
   return comm->sem?SIMIX_sem_would_block(comm->sem):0;
 }
 
 /** @brief wait for the completion of any communication of a set
  *
- *  @Returns the communication which finished, destroy it after identifying which one it is (not removed from the dynar)
+ *  @Returns the rank in the dynar of communication which finished; destroy it after identifying which one it is
  */
-smx_comm_t SIMIX_network_waitany(xbt_dynar_t comms) {
+unsigned int SIMIX_network_waitany(xbt_dynar_t comms) {
   xbt_dynar_t sems = xbt_dynar_new(sizeof(smx_sem_t),NULL);
-  unsigned int cursor;
+  unsigned int cursor, found_comm=-1;
   smx_comm_t comm,comm_finished=NULL;
-  smx_sem_t sem;
 
   xbt_dynar_foreach(comms,cursor,comm){
     xbt_dynar_push(sems,&(comm->sem));
@@ -460,13 +536,9 @@ smx_comm_t SIMIX_network_waitany(xbt_dynar_t comms) {
 
   DEBUG1("Waiting for the completion of communication set %p", comms);
 
-  sem = SIMIX_sem_acquire_any(sems);
-  xbt_dynar_foreach(comms,cursor,comm){
-    if (comm->sem == sem) {
-      comm_finished = comm;
-    }
-  }
-  xbt_assert0(comm_finished,"Cannot find which communication finished");
+  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);
 
@@ -474,7 +546,7 @@ smx_comm_t SIMIX_network_waitany(xbt_dynar_t comms) {
    * and that nobody will ever block on it */
   SIMIX_sem_release_forever(comm_finished->sem);
 
-  /* Check for errors other than timeouts (they are catched above) */
+  /* Check for errors */
   if(!SIMIX_host_get_state(SIMIX_host_self())){
     if(comm_finished->rdv)
       SIMIX_rdv_remove(comm_finished->rdv, comm_finished);
@@ -484,6 +556,7 @@ smx_comm_t SIMIX_network_waitany(xbt_dynar_t comms) {
     SIMIX_communication_destroy(comm_finished);
     THROW0(network_error, 0, "Link failure");
   }
+  SIMIX_communication_destroy(comm_finished);
 
-  return comm_finished;
+  return found_comm;
 }