-/* Copyright (c) 2009, 2010. The SimGrid Team.
+/* Copyright (c) 2009-2014. 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 "smx_private.h"
#include "xbt/log.h"
#include "mc/mc.h"
+#include "mc/mc_replay.h"
#include "xbt/dict.h"
+#include "smpi/private.h"
XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
- "Logging specific to SIMIX (network)");
+ "SIMIX network-related synchronization");
static xbt_dict_t rdv_points = NULL;
-unsigned long int smx_total_comms = 0;
-
-static void SIMIX_waitany_req_remove_from_actions(smx_req_t req);
-static void SIMIX_comm_copy_data(smx_action_t comm);
-static smx_action_t SIMIX_comm_new(e_smx_comm_type_t type);
-static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm);
-static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm);
-static smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type,
- int (*match_fun)(void *, void *), void *);
+XBT_EXPORT_NO_IMPORT(unsigned long int) smx_total_comms = 0;
+
+static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall);
+static void SIMIX_comm_copy_data(smx_synchro_t comm);
+static smx_synchro_t SIMIX_comm_new(e_smx_comm_type_t type);
+static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_synchro_t comm);
+static smx_synchro_t SIMIX_fifo_probe_comm(xbt_fifo_t fifo, e_smx_comm_type_t type,
+ int (*match_fun)(void *, void *,smx_synchro_t),
+ void *user_data, smx_synchro_t my_synchro);
+static smx_synchro_t SIMIX_fifo_get_comm(xbt_fifo_t fifo, e_smx_comm_type_t type,
+ int (*match_fun)(void *, void *,smx_synchro_t),
+ void *user_data, smx_synchro_t my_synchro);
static void SIMIX_rdv_free(void *data);
+static void SIMIX_comm_start(smx_synchro_t synchro);
void SIMIX_network_init(void)
{
- rdv_points = xbt_dict_new();
+ rdv_points = xbt_dict_new_homogeneous(SIMIX_rdv_free);
}
void SIMIX_network_exit(void)
smx_rdv_t SIMIX_rdv_create(const char *name)
{
- /* two processes may have pushed the same rdv_create request at the same time */
+ /* two processes may have pushed the same rdv_create simcall at the same time */
smx_rdv_t rdv = name ? xbt_dict_get_or_null(rdv_points, name) : NULL;
if (!rdv) {
rdv = xbt_new0(s_smx_rvpoint_t, 1);
rdv->name = name ? xbt_strdup(name) : NULL;
rdv->comm_fifo = xbt_fifo_new();
+ rdv->done_comm_fifo = xbt_fifo_new();
+ rdv->permanent_receiver=NULL;
+
+ XBT_DEBUG("Creating a mailbox at %p with name %s", rdv, name);
if (rdv->name)
- xbt_dict_set(rdv_points, rdv->name, rdv, SIMIX_rdv_free);
+ xbt_dict_set(rdv_points, rdv->name, rdv, NULL);
}
return rdv;
}
void SIMIX_rdv_free(void *data)
{
+ XBT_DEBUG("rdv free %p", data);
smx_rdv_t rdv = (smx_rdv_t) data;
- if (rdv->name)
- xbt_free(rdv->name);
+ xbt_free(rdv->name);
xbt_fifo_free(rdv->comm_fifo);
- xbt_free(rdv);
+ xbt_fifo_free(rdv->done_comm_fifo);
+
+ xbt_free(rdv);
+}
+
+xbt_dict_t SIMIX_get_rdv_points()
+{
+ return rdv_points;
}
smx_rdv_t SIMIX_rdv_get_by_name(const char *name)
int SIMIX_rdv_comm_count_by_host(smx_rdv_t rdv, smx_host_t host)
{
- smx_action_t comm = NULL;
+ smx_synchro_t comm = NULL;
xbt_fifo_item_t item = NULL;
int count = 0;
- xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_action_t) {
+ xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_synchro_t) {
if (comm->comm.src_proc->smx_host == host)
count++;
}
return count;
}
-smx_action_t SIMIX_rdv_get_head(smx_rdv_t rdv)
+smx_synchro_t SIMIX_rdv_get_head(smx_rdv_t rdv)
{
return xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
}
/**
- * \brief Push a communication request into a rendez-vous point
+ * \brief get the receiver (process associated to the mailbox)
* \param rdv The rendez-vous point
- * \param comm The communication request
+ * \return process The receiving process (NULL if not set)
*/
-static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_action_t comm)
+smx_process_t SIMIX_rdv_get_receiver(smx_rdv_t rdv)
{
- xbt_fifo_push(rdv->comm_fifo, comm);
- comm->comm.rdv = rdv;
+ return rdv->permanent_receiver;
}
/**
- * \brief Remove a communication request from a rendez-vous point
+ * \brief set the receiver of the rendez vous point to allow eager sends
* \param rdv The rendez-vous point
- * \param comm The communication request
+ * \param process The receiving process
*/
-static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm)
+void SIMIX_rdv_set_receiver(smx_rdv_t rdv, smx_process_t process)
{
- xbt_fifo_remove(rdv->comm_fifo, comm);
- comm->comm.rdv = NULL;
+ rdv->permanent_receiver=process;
+}
+
+/**
+ * \brief Pushes a communication synchro into a rendez-vous point
+ * \param rdv The rendez-vous point
+ * \param comm The communication synchro
+ */
+static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_synchro_t comm)
+{
+ xbt_fifo_push(rdv->comm_fifo, comm);
+ comm->comm.rdv = rdv;
}
/**
- * \brief Wrapper to SIMIX_rdv_get_request
+ * \brief Removes a communication synchro from a rendez-vous point
+ * \param rdv The rendez-vous point
+ * \param comm The communication synchro
*/
-smx_action_t SIMIX_comm_get_send_match(smx_rdv_t rdv, int (*match_fun)(void*, void*), void* data) {
- return SIMIX_rdv_get_request(rdv, SIMIX_COMM_SEND, match_fun, data);
+XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_synchro_t comm)
+{
+ xbt_fifo_remove(rdv->comm_fifo, comm);
+ comm->comm.rdv = NULL;
}
/**
- * \brief Checks if there is a communication action queued in a rendez-vous matching our needs
+ * \brief Checks if there is a communication synchro queued in a fifo matching our needs
* \param type The type of communication we are looking for (comm_send, comm_recv)
- * \return The communication action if found, NULL otherwise
+ * \return The communication synchro if found, NULL otherwise
*/
-smx_action_t SIMIX_rdv_get_request(smx_rdv_t rdv, e_smx_comm_type_t type,
- int (*match_fun)(void *, void *), void *data)
+smx_synchro_t SIMIX_fifo_get_comm(xbt_fifo_t fifo, e_smx_comm_type_t type,
+ int (*match_fun)(void *, void *,smx_synchro_t),
+ void *this_user_data, smx_synchro_t my_synchro)
{
- // FIXME rewrite this function by using SIMIX_rdv_has_send/recv_match
- smx_action_t action;
+ smx_synchro_t synchro;
xbt_fifo_item_t item;
- void* req_data = NULL;
+ void* other_user_data = NULL;
- xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){
- if (action->comm.type == SIMIX_COMM_SEND) {
- req_data = action->comm.src_data;
- } else if (action->comm.type == SIMIX_COMM_RECEIVE) {
- req_data = action->comm.dst_data;
+ xbt_fifo_foreach(fifo, item, synchro, smx_synchro_t) {
+ if (synchro->comm.type == SIMIX_COMM_SEND) {
+ other_user_data = synchro->comm.src_data;
+ } else if (synchro->comm.type == SIMIX_COMM_RECEIVE) {
+ other_user_data = synchro->comm.dst_data;
}
- if (action->comm.type == type && (!match_fun || match_fun(data, req_data))) {
- XBT_DEBUG("Found a matching communication action %p", action);
- xbt_fifo_remove_item(rdv->comm_fifo, item);
+ if (synchro->comm.type == type &&
+ (!match_fun || match_fun(this_user_data, other_user_data, synchro)) &&
+ (!synchro->comm.match_fun || synchro->comm.match_fun(other_user_data, this_user_data, my_synchro))) {
+ XBT_DEBUG("Found a matching communication synchro %p", synchro);
+ xbt_fifo_remove_item(fifo, item);
xbt_fifo_free_item(item);
- action->comm.refcount++;
- action->comm.rdv = NULL;
- return action;
+ synchro->comm.refcount++;
+#ifdef HAVE_MC
+ synchro->comm.rdv_cpy = synchro->comm.rdv;
+#endif
+ synchro->comm.rdv = NULL;
+ return synchro;
}
- XBT_DEBUG("Sorry, communication action %p does not match our needs:"
- " its type is %d but we are looking for a comm of type %d",
- action, action->comm.type, type);
+ 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)",
+ synchro, (int)synchro->comm.type, (int)type);
}
- XBT_DEBUG("No matching communication action found");
+ XBT_DEBUG("No matching communication synchro found");
return NULL;
}
+
/**
- * \brief Checks if there is a send communication action
- * queued in a rendez-vous matching our needs.
- * \return 1 if found, 0 otherwise
+ * \brief Checks if there is a communication synchro queued in a fifo matching our needs, but leave it there
+ * \param type The type of communication we are looking for (comm_send, comm_recv)
+ * \return The communication synchro if found, NULL otherwise
*/
-int SIMIX_comm_has_send_match(smx_rdv_t rdv, int (*match_fun)(void*, void*), void* data) {
-
- smx_action_t action;
+smx_synchro_t SIMIX_fifo_probe_comm(xbt_fifo_t fifo, e_smx_comm_type_t type,
+ int (*match_fun)(void *, void *,smx_synchro_t),
+ void *this_user_data, smx_synchro_t my_synchro)
+{
+ smx_synchro_t synchro;
xbt_fifo_item_t item;
+ void* other_user_data = NULL;
- xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){
- if (action->comm.type == SIMIX_COMM_SEND
- && (!match_fun || match_fun(data, action->comm.src_data))) {
- XBT_DEBUG("Found a matching communication action %p", action);
- return 1;
+ xbt_fifo_foreach(fifo, item, synchro, smx_synchro_t) {
+ if (synchro->comm.type == SIMIX_COMM_SEND) {
+ other_user_data = synchro->comm.src_data;
+ } else if (synchro->comm.type == SIMIX_COMM_RECEIVE) {
+ other_user_data = synchro->comm.dst_data;
}
- }
- XBT_DEBUG("No matching communication action found");
- return 0;
-}
-
-/**
- * \brief Checks if there is a recv communication action
- * queued in a rendez-vous matching our needs.
- * \return 1 if found, 0 otherwise
- */
-int SIMIX_comm_has_recv_match(smx_rdv_t rdv, int (*match_fun)(void*, void*), void* data) {
+ if (synchro->comm.type == type &&
+ (!match_fun || match_fun(this_user_data, other_user_data, synchro)) &&
+ (!synchro->comm.match_fun || synchro->comm.match_fun(other_user_data, this_user_data, my_synchro))) {
+ XBT_DEBUG("Found a matching communication synchro %p", synchro);
+ synchro->comm.refcount++;
- smx_action_t action;
- xbt_fifo_item_t item;
-
- xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){
- if (action->comm.type == SIMIX_COMM_RECEIVE
- && (!match_fun || match_fun(data, action->comm.dst_data))) {
- XBT_DEBUG("Found a matching communication action %p", action);
- return 1;
+ return synchro;
}
+ 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)",
+ synchro, (int)synchro->comm.type, (int)type);
}
- XBT_DEBUG("No matching communication action found");
- return 0;
+ XBT_DEBUG("No matching communication synchro found");
+ return NULL;
}
-
/******************************************************************************/
-/* Comunication Actions */
+/* Communication synchros */
/******************************************************************************/
/**
- * \brief Creates a new comunicate action
- * \param type The type of request (comm_send, comm_recv)
- * \return The new comunicate action
+ * \brief Creates a new communicate synchro
+ * \param type The direction of communication (comm_send, comm_recv)
+ * \return The new communicate synchro
*/
-smx_action_t SIMIX_comm_new(e_smx_comm_type_t type)
+smx_synchro_t SIMIX_comm_new(e_smx_comm_type_t type)
{
- smx_action_t act;
+ smx_synchro_t synchro;
/* alloc structures */
- act = xbt_mallocator_get(simix_global->action_mallocator);
- act->type = SIMIX_ACTION_COMMUNICATE;
- act->state = SIMIX_WAITING;
+ synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
+
+ synchro->type = SIMIX_SYNC_COMMUNICATE;
+ synchro->state = SIMIX_WAITING;
/* set communication */
- act->comm.type = type;
- act->comm.refcount = 1;
+ synchro->comm.type = type;
+ synchro->comm.refcount = 1;
+ synchro->comm.src_data=NULL;
+ synchro->comm.dst_data=NULL;
+
#ifdef HAVE_LATENCY_BOUND_TRACKING
//initialize with unknown value
- act->latency_limited = -1;
+ synchro->latency_limited = -1;
#endif
-#ifdef HAVE_TRACING
- act->category = NULL;
-#endif
+ synchro->category = NULL;
- XBT_DEBUG("Create communicate action %p", act);
+ XBT_DEBUG("Create communicate synchro %p", synchro);
++smx_total_comms;
- return act;
+ return synchro;
}
/**
- * \brief Destroy a communicate action
- * \param action The communicate action to be destroyed
+ * \brief Destroy a communicate synchro
+ * \param synchro The communicate synchro to be destroyed
*/
-void SIMIX_comm_destroy(smx_action_t action)
+void SIMIX_comm_destroy(smx_synchro_t synchro)
{
- XBT_DEBUG("Destroy action %p (refcount:%d)", action, action->comm.refcount);
+ XBT_DEBUG("Destroy synchro %p (refcount: %d), state: %d",
+ synchro, synchro->comm.refcount, (int)synchro->state);
- if (action->comm.refcount <= 0)
- xbt_die("the refcount of comm %p is already 0 before decreasing it. "
- "That's a bug!", action);
-
- action->comm.refcount--;
- if (action->comm.refcount > 0)
- return;
- XBT_DEBUG("Really free communication %p; refcount is now %d", action,
- action->comm.refcount);
+ if (synchro->comm.refcount <= 0) {
+ xbt_backtrace_display_current();
+ xbt_die("The refcount of comm %p is already 0 before decreasing it. "
+ "That's a bug! If you didn't test and/or wait the same communication twice in your code, then the bug is SimGrid's...", synchro);
+ }
+ synchro->comm.refcount--;
+ if (synchro->comm.refcount > 0)
+ return;
+ XBT_DEBUG("Really free communication %p; refcount is now %d", synchro,
+ synchro->comm.refcount);
#ifdef HAVE_LATENCY_BOUND_TRACKING
- action->latency_limited = SIMIX_comm_is_latency_bounded( action ) ;
+ synchro->latency_limited = SIMIX_comm_is_latency_bounded( synchro ) ;
#endif
- xbt_free(action->name);
- SIMIX_comm_destroy_internal_actions(action);
+ xbt_free(synchro->name);
+ SIMIX_comm_destroy_internal_actions(synchro);
- if (action->comm.detached && action->state != SIMIX_DONE) {
+ if (synchro->comm.detached && synchro->state != SIMIX_DONE) {
/* the communication has failed and was detached:
* we have to free the buffer */
- ((void_f_pvoid_t) action->comm.src_data)(action->comm.src_buff);
+ if (synchro->comm.clean_fun) {
+ synchro->comm.clean_fun(synchro->comm.src_buff);
+ }
+ synchro->comm.src_buff = NULL;
}
- xbt_mallocator_release(simix_global->action_mallocator, action);
+ if(synchro->comm.rdv)
+ SIMIX_rdv_remove(synchro->comm.rdv, synchro);
+
+ xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
}
-void SIMIX_comm_destroy_internal_actions(smx_action_t action)
+void SIMIX_comm_destroy_internal_actions(smx_synchro_t synchro)
{
- if (action->comm.surf_comm){
+ if (synchro->comm.surf_comm){
#ifdef HAVE_LATENCY_BOUND_TRACKING
- action->latency_limited = SIMIX_comm_is_latency_bounded(action);
+ synchro->latency_limited = SIMIX_comm_is_latency_bounded(synchro);
#endif
- action->comm.surf_comm->model_type->action_unref(action->comm.surf_comm);
- action->comm.surf_comm = NULL;
+ surf_action_unref(synchro->comm.surf_comm);
+ synchro->comm.surf_comm = NULL;
}
- if (action->comm.src_timeout){
- action->comm.src_timeout->model_type->action_unref(action->comm.src_timeout);
- action->comm.src_timeout = NULL;
+ if (synchro->comm.src_timeout){
+ surf_action_unref(synchro->comm.src_timeout);
+ synchro->comm.src_timeout = NULL;
}
- if (action->comm.dst_timeout){
- action->comm.dst_timeout->model_type->action_unref(action->comm.dst_timeout);
- action->comm.dst_timeout = NULL;
+ if (synchro->comm.dst_timeout){
+ surf_action_unref(synchro->comm.dst_timeout);
+ synchro->comm.dst_timeout = NULL;
}
}
-smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv,
+void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_process_t src, smx_rdv_t rdv,
+ double task_size, double rate,
+ void *src_buff, size_t src_buff_size,
+ int (*match_fun)(void *, void *,smx_synchro_t),
+ void (*copy_data_fun)(smx_synchro_t, void*, size_t),
+ void *data, double timeout){
+ smx_synchro_t comm = SIMIX_comm_isend(src, rdv, task_size, rate,
+ src_buff, src_buff_size, match_fun, NULL, copy_data_fun,
+ data, 0);
+ SIMCALL_SET_MC_VALUE(simcall, 0);
+ simcall_HANDLER_comm_wait(simcall, comm, timeout);
+}
+smx_synchro_t simcall_HANDLER_comm_isend(smx_simcall_t simcall, smx_process_t src, smx_rdv_t rdv,
+ double task_size, double rate,
+ void *src_buff, size_t src_buff_size,
+ int (*match_fun)(void *, void *,smx_synchro_t),
+ void (*clean_fun)(void *),
+ void (*copy_data_fun)(smx_synchro_t, void*, size_t),
+ void *data, int detached){
+ return SIMIX_comm_isend(src, rdv, task_size, rate, src_buff,
+ src_buff_size, match_fun, clean_fun, copy_data_fun, data, detached);
+
+}
+smx_synchro_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv,
double task_size, double rate,
void *src_buff, size_t src_buff_size,
- int (*match_fun)(void *, void *), void *data,
+ int (*match_fun)(void *, void *,smx_synchro_t),
+ void (*clean_fun)(void *), // used to free the synchro in case of problem after a detached send
+ void (*copy_data_fun)(smx_synchro_t, void*, size_t), // used to copy data if not default one
+ void *data,
int detached)
{
- smx_action_t action;
+ XBT_DEBUG("send from %p", rdv);
+
+ /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
+ smx_synchro_t this_synchro = SIMIX_comm_new(SIMIX_COMM_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 */
+ smx_synchro_t other_synchro = SIMIX_fifo_get_comm(rdv->comm_fifo, SIMIX_COMM_RECEIVE, match_fun, data, this_synchro);
- /* Look for communication request matching our needs.
- If it is not found then create it and push it into the rendez-vous point */
- action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_RECEIVE, match_fun, data);
+ if (!other_synchro) {
+ other_synchro = this_synchro;
- if (!action) {
- action = SIMIX_comm_new(SIMIX_COMM_SEND);
- SIMIX_rdv_push(rdv, action);
+ if (rdv->permanent_receiver!=NULL){
+ //this mailbox is for small messages, which have to be sent right now
+ other_synchro->state = SIMIX_READY;
+ other_synchro->comm.dst_proc=rdv->permanent_receiver;
+ other_synchro->comm.refcount++;
+ xbt_fifo_push(rdv->done_comm_fifo,other_synchro);
+ other_synchro->comm.rdv=rdv;
+ XBT_DEBUG("pushing a message into the permanent receive fifo %p, comm %p", rdv, &(other_synchro->comm));
+
+ }else{
+ SIMIX_rdv_push(rdv, this_synchro);
+ }
} else {
- action->state = SIMIX_READY;
- action->comm.type = SIMIX_COMM_READY;
+ XBT_DEBUG("Receive already pushed");
+
+ SIMIX_comm_destroy(this_synchro);
+ --smx_total_comms; // this creation was a pure waste
+
+ other_synchro->state = SIMIX_READY;
+ other_synchro->comm.type = SIMIX_COMM_READY;
+
}
+ xbt_fifo_push(src_proc->comms, other_synchro);
- /* If the communication action is detached then decrease the refcount
- * by one, so it will be eliminated by the receivers destroy call */
+ /* if the communication synchro is detached then decrease the refcount
+ * by one, so it will be eliminated by the receiver's destroy call */
if (detached) {
- action->comm.detached = 1;
- action->comm.refcount--;
+ other_synchro->comm.detached = 1;
+ other_synchro->comm.refcount--;
+ other_synchro->comm.clean_fun = clean_fun;
+ } else {
+ other_synchro->comm.clean_fun = NULL;
}
- /* Setup the communication request */
- action->comm.src_proc = src_proc;
- action->comm.task_size = task_size;
- action->comm.rate = rate;
- action->comm.src_buff = src_buff;
- action->comm.src_buff_size = src_buff_size;
- action->comm.src_data = data;
+ /* Setup the communication synchro */
+ other_synchro->comm.src_proc = src_proc;
+ other_synchro->comm.task_size = task_size;
+ other_synchro->comm.rate = rate;
+ other_synchro->comm.src_buff = src_buff;
+ other_synchro->comm.src_buff_size = src_buff_size;
+ other_synchro->comm.src_data = data;
+
+ other_synchro->comm.match_fun = match_fun;
+ other_synchro->comm.copy_data_fun = copy_data_fun;
+
- if (MC_IS_ENABLED) {
- action->state = SIMIX_RUNNING;
- return action;
+ if (MC_is_active() || MC_record_replay_is_active()) {
+ other_synchro->state = SIMIX_RUNNING;
+ return (detached ? NULL : other_synchro);
}
- SIMIX_comm_start(action);
- return action;
+ SIMIX_comm_start(other_synchro);
+ return (detached ? NULL : other_synchro);
}
-smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv,
- void *dst_buff, size_t *dst_buff_size,
- int (*match_fun)(void *, void *), void *data)
+void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_rdv_t rdv,
+ void *dst_buff, size_t *dst_buff_size,
+ int (*match_fun)(void *, void *, smx_synchro_t),
+ void (*copy_data_fun)(smx_synchro_t, void*, size_t),
+ void *data, double timeout, double rate)
{
- smx_action_t action;
+ smx_synchro_t comm = SIMIX_comm_irecv(simcall->issuer, rdv, 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);
+}
- /* Look for communication request matching our needs.
- * If it is not found then create it and push it into the rendez-vous point
- */
- action = SIMIX_rdv_get_request(rdv, SIMIX_COMM_SEND, match_fun, data);
+smx_synchro_t simcall_HANDLER_comm_irecv(smx_simcall_t simcall, smx_rdv_t rdv,
+ void *dst_buff, size_t *dst_buff_size,
+ int (*match_fun)(void *, void *, smx_synchro_t),
+ void (*copy_data_fun)(smx_synchro_t, void*, size_t),
+ void *data, double rate)
+{
+ return SIMIX_comm_irecv(simcall->issuer, rdv, dst_buff, dst_buff_size,
+ match_fun, copy_data_fun, data, rate);
+}
- if (!action) {
- action = SIMIX_comm_new(SIMIX_COMM_RECEIVE);
- SIMIX_rdv_push(rdv, action);
- } else {
- action->state = SIMIX_READY;
- action->comm.type = SIMIX_COMM_READY;
+smx_synchro_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv,
+ void *dst_buff, size_t *dst_buff_size,
+ int (*match_fun)(void *, void *, smx_synchro_t),
+ void (*copy_data_fun)(smx_synchro_t, void*, size_t), // used to copy data if not default one
+ void *data, double rate)
+{
+ XBT_DEBUG("recv from %p %p", rdv, rdv->comm_fifo);
+ smx_synchro_t this_synchro = SIMIX_comm_new(SIMIX_COMM_RECEIVE);
+
+ smx_synchro_t other_synchro;
+ //communication already done, get it inside the fifo of completed comms
+ //permanent receive v1
+ //int already_received=0;
+ if(rdv->permanent_receiver && xbt_fifo_size(rdv->done_comm_fifo)!=0){
+
+ 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 already received fifo
+ other_synchro = SIMIX_fifo_get_comm(rdv->done_comm_fifo, SIMIX_COMM_SEND, match_fun, data, this_synchro);
+ //if not found, assume the receiver came first, register it to the mailbox in the classical way
+ if (!other_synchro) {
+ XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request into fifo");
+ other_synchro = this_synchro;
+ SIMIX_rdv_push(rdv, this_synchro);
+ }else{
+ if(other_synchro->comm.surf_comm && SIMIX_comm_get_remains(other_synchro)==0.0)
+ {
+ XBT_DEBUG("comm %p has been already sent, and is finished, destroy it",&(other_synchro->comm));
+ other_synchro->state = SIMIX_DONE;
+ other_synchro->comm.type = SIMIX_COMM_DONE;
+ other_synchro->comm.rdv = NULL;
+ }/*else{
+ XBT_DEBUG("Not yet finished, we have to wait %d", xbt_fifo_size(rdv->comm_fifo));
+ }*/
+ other_synchro->comm.refcount--;
+ SIMIX_comm_destroy(this_synchro);
+ --smx_total_comms; // this creation was a pure waste
+ }
+ }else{
+ /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
+
+ /* 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 */
+ other_synchro = SIMIX_fifo_get_comm(rdv->comm_fifo, SIMIX_COMM_SEND, match_fun, data, this_synchro);
+
+ if (!other_synchro) {
+ XBT_DEBUG("Receive pushed first %d", xbt_fifo_size(rdv->comm_fifo));
+ other_synchro = this_synchro;
+ SIMIX_rdv_push(rdv, this_synchro);
+ } else {
+ SIMIX_comm_destroy(this_synchro);
+ --smx_total_comms; // this creation was a pure waste
+ other_synchro->state = SIMIX_READY;
+ other_synchro->comm.type = SIMIX_COMM_READY;
+ //other_synchro->comm.refcount--;
+ }
+ xbt_fifo_push(dst_proc->comms, other_synchro);
}
- /* Setup communication request */
- action->comm.dst_proc = dst_proc;
- action->comm.dst_buff = dst_buff;
- action->comm.dst_buff_size = dst_buff_size;
- action->comm.dst_data = data;
+ /* Setup communication synchro */
+ other_synchro->comm.dst_proc = dst_proc;
+ other_synchro->comm.dst_buff = dst_buff;
+ other_synchro->comm.dst_buff_size = dst_buff_size;
+ other_synchro->comm.dst_data = data;
+
+ if (rate != -1.0 &&
+ (other_synchro->comm.rate == -1.0 || rate < other_synchro->comm.rate))
+ other_synchro->comm.rate = rate;
+
+ other_synchro->comm.match_fun = match_fun;
+ other_synchro->comm.copy_data_fun = copy_data_fun;
+
- if (MC_IS_ENABLED) {
- action->state = SIMIX_RUNNING;
- return action;
+ /*if(already_received)//do the actual copy, because the first one after the comm didn't have all the info
+ SIMIX_comm_copy_data(other_synchro);*/
+
+
+ if (MC_is_active() || MC_record_replay_is_active()) {
+ other_synchro->state = SIMIX_RUNNING;
+ return other_synchro;
+ }
+
+ SIMIX_comm_start(other_synchro);
+ // }
+ return other_synchro;
+}
+
+smx_synchro_t simcall_HANDLER_comm_iprobe(smx_simcall_t simcall, smx_rdv_t rdv,
+ int type, int src, int tag,
+ int (*match_fun)(void *, void *, smx_synchro_t),
+ void *data){
+ return SIMIX_comm_iprobe(simcall->issuer, rdv, type, src, tag, match_fun, data);
+}
+
+smx_synchro_t SIMIX_comm_iprobe(smx_process_t dst_proc, smx_rdv_t rdv, int type, int src,
+ int tag, int (*match_fun)(void *, void *, smx_synchro_t), void *data)
+{
+ XBT_DEBUG("iprobe from %p %p", rdv, rdv->comm_fifo);
+ smx_synchro_t this_synchro;
+ int smx_type;
+ if(type == 1){
+ this_synchro=SIMIX_comm_new(SIMIX_COMM_SEND);
+ smx_type = SIMIX_COMM_RECEIVE;
+ } else{
+ this_synchro=SIMIX_comm_new(SIMIX_COMM_RECEIVE);
+ smx_type = SIMIX_COMM_SEND;
+ }
+ smx_synchro_t other_synchro=NULL;
+ if(rdv->permanent_receiver && xbt_fifo_size(rdv->done_comm_fifo)!=0){
+ //find a match in the already received fifo
+ XBT_DEBUG("first try in the perm recv mailbox");
+
+ other_synchro = SIMIX_fifo_probe_comm(rdv->done_comm_fifo, smx_type, match_fun, data, this_synchro);
}
+ // }else{
+ if(!other_synchro){
+ XBT_DEBUG("try in the normal mailbox");
+ other_synchro = SIMIX_fifo_probe_comm(rdv->comm_fifo, smx_type, match_fun, data, this_synchro);
+ }
+// }
+ if(other_synchro)other_synchro->comm.refcount--;
- SIMIX_comm_start(action);
- return action;
+ SIMIX_comm_destroy(this_synchro);
+ --smx_total_comms;
+ return other_synchro;
}
-void SIMIX_pre_comm_wait(smx_req_t req, smx_action_t action, double timeout, int idx)
+void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_synchro_t synchro, double timeout)
{
- /* the request may be a wait, a send or a recv */
+ /* the simcall may be a wait, a send or a recv */
surf_action_t sleep;
- /* Associate this request to the action */
- xbt_fifo_push(action->request_list, req);
- req->issuer->waiting_action = action;
+ /* Associate this simcall to the wait synchro */
+ XBT_DEBUG("simcall_HANDLER_comm_wait, %p", synchro);
+
+ xbt_fifo_push(synchro->simcalls, simcall);
+ simcall->issuer->waiting_synchro = synchro;
- if (MC_IS_ENABLED) {
+ if (MC_is_active() || MC_record_replay_is_active()) {
+ int idx = SIMCALL_GET_MC_VALUE(simcall);
if (idx == 0) {
- action->state = SIMIX_DONE;
+ synchro->state = SIMIX_DONE;
} else {
- /* If we reached this point, the wait request must have a timeout */
+ /* If we reached this point, the wait simcall must have a timeout */
/* Otherwise it shouldn't be enabled and executed by the MC */
if (timeout == -1)
THROW_IMPOSSIBLE;
- if (action->comm.src_proc == req->issuer)
- action->state = SIMIX_SRC_TIMEOUT;
+ if (synchro->comm.src_proc == simcall->issuer)
+ synchro->state = SIMIX_SRC_TIMEOUT;
else
- action->state = SIMIX_DST_TIMEOUT;
+ synchro->state = SIMIX_DST_TIMEOUT;
}
- SIMIX_comm_finish(action);
+ SIMIX_comm_finish(synchro);
return;
}
- /* If the action has already finish perform the error handling, */
- /* otherwise set up a waiting timeout on the right side */
- if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
- SIMIX_comm_finish(action);
+ /* If the synchro has already finish perform the error handling, */
+ /* otherwise set up a waiting timeout on the right side */
+ if (synchro->state != SIMIX_WAITING && synchro->state != SIMIX_RUNNING) {
+ SIMIX_comm_finish(synchro);
} else { /* if (timeout >= 0) { we need a surf sleep action even when there is no timeout, otherwise surf won't tell us when the host fails */
- sleep = surf_workstation_model->extension.workstation.sleep(req->issuer->smx_host->host, timeout);
- surf_workstation_model->action_data_set(sleep, action);
+ sleep = surf_workstation_sleep(simcall->issuer->smx_host, timeout);
+ surf_action_set_data(sleep, synchro);
- if (req->issuer == action->comm.src_proc)
- action->comm.src_timeout = sleep;
+ if (simcall->issuer == synchro->comm.src_proc)
+ synchro->comm.src_timeout = sleep;
else
- action->comm.dst_timeout = sleep;
+ synchro->comm.dst_timeout = sleep;
}
}
-void SIMIX_pre_comm_test(smx_req_t req)
+void simcall_HANDLER_comm_test(smx_simcall_t simcall, smx_synchro_t synchro)
{
- smx_action_t action = req->comm_test.comm;
-
- if(MC_IS_ENABLED){
- req->comm_test.result = action->comm.src_proc && action->comm.dst_proc;
- if(req->comm_test.result){
- action->state = SIMIX_DONE;
- xbt_fifo_push(action->request_list, req);
- SIMIX_comm_finish(action);
+ if(MC_is_active() || MC_record_replay_is_active()){
+ simcall_comm_test__set__result(simcall, synchro->comm.src_proc && synchro->comm.dst_proc);
+ if(simcall_comm_test__get__result(simcall)){
+ synchro->state = SIMIX_DONE;
+ xbt_fifo_push(synchro->simcalls, simcall);
+ SIMIX_comm_finish(synchro);
}else{
- SIMIX_request_answer(req);
+ SIMIX_simcall_answer(simcall);
}
return;
}
- req->comm_test.result = (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING);
- if (req->comm_test.result) {
- xbt_fifo_push(action->request_list, req);
- SIMIX_comm_finish(action);
+ simcall_comm_test__set__result(simcall, (synchro->state != SIMIX_WAITING && synchro->state != SIMIX_RUNNING));
+ if (simcall_comm_test__get__result(simcall)) {
+ xbt_fifo_push(synchro->simcalls, simcall);
+ SIMIX_comm_finish(synchro);
} else {
- SIMIX_request_answer(req);
+ SIMIX_simcall_answer(simcall);
}
}
-void SIMIX_pre_comm_testany(smx_req_t req, int idx)
+void simcall_HANDLER_comm_testany(smx_simcall_t simcall, xbt_dynar_t synchros)
{
unsigned int cursor;
- smx_action_t action;
- xbt_dynar_t actions = req->comm_testany.comms;
- req->comm_testany.result = -1;
+ smx_synchro_t synchro;
+ simcall_comm_testany__set__result(simcall, -1);
- if (MC_IS_ENABLED){
+ if (MC_is_active() || MC_record_replay_is_active()){
+ int idx = SIMCALL_GET_MC_VALUE(simcall);
if(idx == -1){
- SIMIX_request_answer(req);
+ SIMIX_simcall_answer(simcall);
}else{
- action = xbt_dynar_get_as(actions, idx, smx_action_t);
- req->comm_testany.result = idx;
- xbt_fifo_push(action->request_list, req);
- action->state = SIMIX_DONE;
- SIMIX_comm_finish(action);
+ synchro = xbt_dynar_get_as(synchros, idx, smx_synchro_t);
+ simcall_comm_testany__set__result(simcall, idx);
+ xbt_fifo_push(synchro->simcalls, simcall);
+ synchro->state = SIMIX_DONE;
+ SIMIX_comm_finish(synchro);
}
return;
}
- xbt_dynar_foreach(req->comm_testany.comms,cursor,action) {
- if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
- req->comm_testany.result = cursor;
- xbt_fifo_push(action->request_list, req);
- SIMIX_comm_finish(action);
+ xbt_dynar_foreach(simcall_comm_testany__get__comms(simcall), cursor,synchro) {
+ if (synchro->state != SIMIX_WAITING && synchro->state != SIMIX_RUNNING) {
+ simcall_comm_testany__set__result(simcall, cursor);
+ xbt_fifo_push(synchro->simcalls, simcall);
+ SIMIX_comm_finish(synchro);
return;
}
}
- SIMIX_request_answer(req);
+ SIMIX_simcall_answer(simcall);
}
-void SIMIX_pre_comm_waitany(smx_req_t req, int idx)
+void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, xbt_dynar_t synchros)
{
- smx_action_t action;
+ smx_synchro_t synchro;
unsigned int cursor = 0;
- xbt_dynar_t actions = req->comm_waitany.comms;
-
- if (MC_IS_ENABLED){
- action = xbt_dynar_get_as(actions, idx, smx_action_t);
- xbt_fifo_push(action->request_list, req);
- req->comm_waitany.result = idx;
- action->state = SIMIX_DONE;
- SIMIX_comm_finish(action);
+
+ if (MC_is_active() || MC_record_replay_is_active()){
+ int idx = SIMCALL_GET_MC_VALUE(simcall);
+ synchro = xbt_dynar_get_as(synchros, idx, smx_synchro_t);
+ xbt_fifo_push(synchro->simcalls, simcall);
+ simcall_comm_waitany__set__result(simcall, idx);
+ synchro->state = SIMIX_DONE;
+ SIMIX_comm_finish(synchro);
return;
}
- xbt_dynar_foreach(actions, cursor, action){
- /* Associate this request to the action */
- xbt_fifo_push(action->request_list, req);
- if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING){
- SIMIX_comm_finish(action);
+ xbt_dynar_foreach(synchros, cursor, synchro){
+ /* associate this simcall to the the synchro */
+ xbt_fifo_push(synchro->simcalls, simcall);
+
+ /* see if the synchro is already finished */
+ if (synchro->state != SIMIX_WAITING && synchro->state != SIMIX_RUNNING){
+ SIMIX_comm_finish(synchro);
break;
}
}
}
-void SIMIX_waitany_req_remove_from_actions(smx_req_t req)
+void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall)
{
- smx_action_t action;
+ smx_synchro_t synchro;
unsigned int cursor = 0;
- xbt_dynar_t actions = req->comm_waitany.comms;
+ xbt_dynar_t synchros = simcall_comm_waitany__get__comms(simcall);
- xbt_dynar_foreach(actions, cursor, action){
- xbt_fifo_remove(action->request_list, req);
+ xbt_dynar_foreach(synchros, cursor, synchro) {
+ xbt_fifo_remove(synchro->simcalls, simcall);
}
}
/**
- * \brief Start the simulation of a communication request
- * \param action The communication action
+ * \brief Starts the simulation of a communication synchro.
+ * \param synchro the communication synchro
*/
-
-XBT_INLINE void SIMIX_comm_start(smx_action_t action)
+static XBT_INLINE void SIMIX_comm_start(smx_synchro_t synchro)
{
/* If both the sender and the receiver are already there, start the communication */
- if (action->state == SIMIX_READY) {
- smx_host_t sender = action->comm.src_proc->smx_host;
- smx_host_t receiver = action->comm.dst_proc->smx_host;
+ if (synchro->state == SIMIX_READY) {
+
+ smx_host_t sender = synchro->comm.src_proc->smx_host;
+ smx_host_t receiver = synchro->comm.dst_proc->smx_host;
- XBT_DEBUG("Starting communication %p from '%s' to '%s'", action,
- SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
+ XBT_DEBUG("Starting communication %p from '%s' to '%s'", synchro,
+ SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
- action->comm.surf_comm = surf_workstation_model->extension.workstation.
- communicate(sender->host, receiver->host, action->comm.task_size, action->comm.rate);
+ synchro->comm.surf_comm = surf_workstation_model_communicate(surf_workstation_model,
+ sender, receiver,
+ synchro->comm.task_size, synchro->comm.rate);
- surf_workstation_model->action_data_set(action->comm.surf_comm, action);
+ surf_action_set_data(synchro->comm.surf_comm, synchro);
- action->state = SIMIX_RUNNING;
+ synchro->state = SIMIX_RUNNING;
/* If a link is failed, detect it immediately */
- if (surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) {
+ if (surf_action_get_state(synchro->comm.surf_comm) == SURF_ACTION_FAILED) {
XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure",
- SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
- action->state = SIMIX_LINK_FAILURE;
- SIMIX_comm_destroy_internal_actions(action);
+ SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver));
+ synchro->state = SIMIX_LINK_FAILURE;
+ SIMIX_comm_destroy_internal_actions(synchro);
}
- /* If any of the process is suspend, create the action but stop its execution,
+ /* If any of the process is suspend, create the synchro but stop its execution,
it will be restarted when the sender process resume */
- if (SIMIX_process_is_suspended(action->comm.src_proc) ||
- SIMIX_process_is_suspended(action->comm.dst_proc)) {
- /* FIXME: check what should happen with the action state */
- surf_workstation_model->suspend(action->comm.surf_comm);
+ if (SIMIX_process_is_suspended(synchro->comm.src_proc) ||
+ SIMIX_process_is_suspended(synchro->comm.dst_proc)) {
+ /* FIXME: check what should happen with the synchro state */
+
+ if (SIMIX_process_is_suspended(synchro->comm.src_proc))
+ XBT_DEBUG("The communication is suspended on startup because src (%s:%s) were suspended since it initiated the communication",
+ SIMIX_host_get_name(synchro->comm.src_proc->smx_host), synchro->comm.src_proc->name);
+ else
+ XBT_DEBUG("The communication is suspended on startup because dst (%s:%s) were suspended since it initiated the communication",
+ SIMIX_host_get_name(synchro->comm.dst_proc->smx_host), synchro->comm.dst_proc->name);
+
+ surf_action_suspend(synchro->comm.surf_comm);
+
}
}
}
-void SIMIX_comm_finish(smx_action_t action)
+/**
+ * \brief Answers the SIMIX simcalls associated to a communication synchro.
+ * \param synchro a finished communication synchro
+ */
+void SIMIX_comm_finish(smx_synchro_t synchro)
{
unsigned int destroy_count = 0;
- smx_req_t req;
-
- while ((req = xbt_fifo_shift(action->request_list))) {
-
- /* If a waitany request is waiting for this action to finish, then remove
- it from the other actions in the waitany list. Afterwards, get the
- position of the actual action in the waitany request's actions dynar and
- return it as the result of the call */
- if (req->call == REQ_COMM_WAITANY) {
- SIMIX_waitany_req_remove_from_actions(req);
- if (!MC_IS_ENABLED)
- req->comm_waitany.result = xbt_dynar_search(req->comm_waitany.comms, &action);
+ smx_simcall_t simcall;
+
+
+ while ((simcall = xbt_fifo_shift(synchro->simcalls))) {
+
+ /* If a waitany simcall is waiting for this synchro to finish, then remove
+ it from the other synchros in the waitany list. Afterwards, get the
+ position of the actual synchro in the waitany dynar 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_COMM_WAITANY) {
+ SIMIX_waitany_remove_simcall_from_actions(simcall);
+ if (!MC_is_active() && !MC_record_replay_is_active())
+ simcall_comm_waitany__set__result(simcall, xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro));
}
- /* If the action is still in a rendez-vous point then remove from it */
- if (action->comm.rdv)
- SIMIX_rdv_remove(action->comm.rdv, action);
+ /* If the synchro is still in a rendez-vous point then remove from it */
+ if (synchro->comm.rdv)
+ SIMIX_rdv_remove(synchro->comm.rdv, synchro);
- XBT_DEBUG("SIMIX_comm_finish: action state = %d", action->state);
+ XBT_DEBUG("SIMIX_comm_finish: synchro state = %d", (int)synchro->state);
/* Check out for errors */
- switch (action->state) {
-
- case SIMIX_DONE:
- XBT_DEBUG("Communication %p complete!", action);
- SIMIX_comm_copy_data(action);
- break;
-
- case SIMIX_SRC_TIMEOUT:
- TRY {
- THROWF(timeout_error, 0, "Communication timeouted because of sender");
- }
- CATCH(req->issuer->running_ctx->exception) {
- req->issuer->doexception = 1;
- }
- break;
-
- case SIMIX_DST_TIMEOUT:
- TRY {
- THROWF(timeout_error, 0, "Communication timeouted because of receiver");
- }
- CATCH(req->issuer->running_ctx->exception) {
- req->issuer->doexception = 1;
- }
- break;
-
- case SIMIX_SRC_HOST_FAILURE:
- TRY {
- if (req->issuer == action->comm.src_proc)
- THROWF(host_error, 0, "Host failed");
- else
- THROWF(network_error, 0, "Remote peer failed");
- }
- CATCH(req->issuer->running_ctx->exception) {
- req->issuer->doexception = 1;
- }
- break;
-
- case SIMIX_DST_HOST_FAILURE:
- TRY {
- if (req->issuer == action->comm.dst_proc)
- THROWF(host_error, 0, "Host failed");
- else
- THROWF(network_error, 0, "Remote peer failed");
- }
- CATCH(req->issuer->running_ctx->exception) {
- req->issuer->doexception = 1;
- }
- break;
-
- case SIMIX_LINK_FAILURE:
- TRY {
- XBT_DEBUG("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)",
- action, action->comm.src_proc->smx_host->name, action->comm.dst_proc->smx_host->name,
- req->issuer->name, req->issuer);
- THROWF(network_error, 0, "Link failure");
- }
- CATCH(req->issuer->running_ctx->exception) {
- req->issuer->doexception = 1;
- }
- break;
-
- default:
- THROW_IMPOSSIBLE;
+ switch (synchro->state) {
+
+ case SIMIX_DONE:
+ XBT_DEBUG("Communication %p complete!", synchro);
+ SIMIX_comm_copy_data(synchro);
+ break;
+
+ case SIMIX_SRC_TIMEOUT:
+ SMX_EXCEPTION(simcall->issuer, timeout_error, 0,
+ "Communication timeouted because of sender");
+ break;
+
+ case SIMIX_DST_TIMEOUT:
+ SMX_EXCEPTION(simcall->issuer, timeout_error, 0,
+ "Communication timeouted because of receiver");
+ break;
+
+ case SIMIX_SRC_HOST_FAILURE:
+ if (simcall->issuer == synchro->comm.src_proc)
+ simcall->issuer->context->iwannadie = 1;
+// SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
+ else
+ SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed");
+ break;
+
+ case SIMIX_DST_HOST_FAILURE:
+ if (simcall->issuer == synchro->comm.dst_proc)
+ simcall->issuer->context->iwannadie = 1;
+// SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
+ else
+ SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed");
+ break;
+
+ case SIMIX_LINK_FAILURE:
+ XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) detached:%d",
+ synchro,
+ synchro->comm.src_proc ? sg_host_name(synchro->comm.src_proc->smx_host) : NULL,
+ synchro->comm.dst_proc ? sg_host_name(synchro->comm.dst_proc->smx_host) : NULL,
+ simcall->issuer->name, simcall->issuer, synchro->comm.detached);
+ if (synchro->comm.src_proc == simcall->issuer) {
+ XBT_DEBUG("I'm source");
+ } else if (synchro->comm.dst_proc == simcall->issuer) {
+ XBT_DEBUG("I'm dest");
+ } else {
+ XBT_DEBUG("I'm neither source nor dest");
+ }
+ SMX_EXCEPTION(simcall->issuer, network_error, 0, "Link failure");
+ break;
+
+ case SIMIX_CANCELED:
+ if (simcall->issuer == synchro->comm.dst_proc)
+ SMX_EXCEPTION(simcall->issuer, cancel_error, 0,
+ "Communication canceled by the sender");
+ else
+ SMX_EXCEPTION(simcall->issuer, cancel_error, 0,
+ "Communication canceled by the receiver");
+ break;
+
+ default:
+ xbt_die("Unexpected synchro state in SIMIX_comm_finish: %d", (int)synchro->state);
}
/* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
- if (req->issuer->doexception) {
- if (req->call == REQ_COMM_WAITANY) {
- req->issuer->running_ctx->exception.value = xbt_dynar_search(req->comm_waitany.comms, &action);
+ if (simcall->issuer->doexception) {
+ if (simcall->call == SIMCALL_COMM_WAITANY) {
+ simcall->issuer->running_ctx->exception.value = xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro);
}
- else if (req->call == REQ_COMM_TESTANY) {
- req->issuer->running_ctx->exception.value = xbt_dynar_search(req->comm_testany.comms, &action);
+ else if (simcall->call == SIMCALL_COMM_TESTANY) {
+ simcall->issuer->running_ctx->exception.value = xbt_dynar_search(simcall_comm_testany__get__comms(simcall), &synchro);
}
}
- req->issuer->waiting_action = NULL;
- SIMIX_request_answer(req);
+ if (surf_resource_get_state(surf_workstation_resource_priv(simcall->issuer->smx_host)) != SURF_RESOURCE_ON) {
+ simcall->issuer->context->iwannadie = 1;
+ }
+
+ simcall->issuer->waiting_synchro = NULL;
+ xbt_fifo_remove(simcall->issuer->comms, synchro);
+ if(synchro->comm.detached){
+ if(simcall->issuer == synchro->comm.src_proc){
+ if(synchro->comm.dst_proc)
+ xbt_fifo_remove(synchro->comm.dst_proc->comms, synchro);
+ }
+ if(simcall->issuer == synchro->comm.dst_proc){
+ if(synchro->comm.src_proc)
+ xbt_fifo_remove(synchro->comm.src_proc->comms, synchro);
+ }
+ }
+ SIMIX_simcall_answer(simcall);
destroy_count++;
}
while (destroy_count-- > 0)
- SIMIX_comm_destroy(action);
+ SIMIX_comm_destroy(synchro);
}
-void SIMIX_post_comm(smx_action_t action)
+/**
+ * \brief This function is called when a Surf communication synchro is finished.
+ * \param synchro the corresponding Simix communication
+ */
+void SIMIX_post_comm(smx_synchro_t synchro)
{
- /* Update action state */
- if (action->comm.src_timeout &&
- surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_DONE)
- action->state = SIMIX_SRC_TIMEOUT;
- else if (action->comm.dst_timeout &&
- surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_DONE)
- action->state = SIMIX_DST_TIMEOUT;
- else if (action->comm.src_timeout &&
- surf_workstation_model->action_state_get(action->comm.src_timeout) == SURF_ACTION_FAILED)
- action->state = SIMIX_SRC_HOST_FAILURE;
- else if (action->comm.dst_timeout &&
- surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_FAILED)
- action->state = SIMIX_DST_HOST_FAILURE;
- else if (action->comm.surf_comm &&
- surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED)
- action->state = SIMIX_LINK_FAILURE;
- else
- action->state = SIMIX_DONE;
-
- XBT_DEBUG("SIMIX_post_comm: action state = %d", action->state);
-
- /* After this point the surf actions associated with the simix communicate
- action are no longer needed, thus we delete them. */
- SIMIX_comm_destroy_internal_actions(action);
-
- /* If there are requests associated with the action, then answer them */
- if (xbt_fifo_size(action->request_list))
- SIMIX_comm_finish(action);
+ /* Update synchro state */
+ if (synchro->comm.src_timeout &&
+ surf_action_get_state(synchro->comm.src_timeout) == SURF_ACTION_DONE)
+ synchro->state = SIMIX_SRC_TIMEOUT;
+ else if (synchro->comm.dst_timeout &&
+ surf_action_get_state(synchro->comm.dst_timeout) == SURF_ACTION_DONE)
+ synchro->state = SIMIX_DST_TIMEOUT;
+ else if (synchro->comm.src_timeout &&
+ surf_action_get_state(synchro->comm.src_timeout) == SURF_ACTION_FAILED)
+ synchro->state = SIMIX_SRC_HOST_FAILURE;
+ else if (synchro->comm.dst_timeout &&
+ surf_action_get_state(synchro->comm.dst_timeout) == SURF_ACTION_FAILED)
+ synchro->state = SIMIX_DST_HOST_FAILURE;
+ else if (synchro->comm.surf_comm &&
+ surf_action_get_state(synchro->comm.surf_comm) == SURF_ACTION_FAILED) {
+ XBT_DEBUG("Puta madre. Surf says that the link broke");
+ synchro->state = SIMIX_LINK_FAILURE;
+ } else
+ synchro->state = SIMIX_DONE;
+
+ XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d",
+ synchro, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc, synchro->comm.detached);
+
+ /* destroy the surf actions associated with the Simix communication */
+ SIMIX_comm_destroy_internal_actions(synchro);
+
+ /* if there are simcalls associated with the synchro, then answer them */
+ if (xbt_fifo_size(synchro->simcalls)) {
+ SIMIX_comm_finish(synchro);
+ }
}
-void SIMIX_comm_cancel(smx_action_t action)
+void SIMIX_comm_cancel(smx_synchro_t synchro)
{
- /* If the action is a waiting state means that it is still in a rdv */
+ /* if the synchro is a waiting state means that it is still in a rdv */
/* so remove from it and delete it */
- if (action->state == SIMIX_WAITING) {
- SIMIX_rdv_remove(action->comm.rdv, action);
- action->state = SIMIX_FAILED;
- } else {
- /* When running the MC there are no surf actions */
- if(!MC_IS_ENABLED)
- surf_workstation_model->action_cancel(action->comm.surf_comm);
+ if (synchro->state == SIMIX_WAITING) {
+ SIMIX_rdv_remove(synchro->comm.rdv, synchro);
+ synchro->state = SIMIX_CANCELED;
+ }
+ else if (!MC_is_active() /* when running the MC there are no surf actions */
+ && !MC_record_replay_is_active()
+ && (synchro->state == SIMIX_READY || synchro->state == SIMIX_RUNNING)) {
+
+ surf_action_cancel(synchro->comm.surf_comm);
}
}
-void SIMIX_comm_suspend(smx_action_t action)
+void SIMIX_comm_suspend(smx_synchro_t synchro)
{
- /*FIXME: shall we suspend also the timeout actions? */
- surf_workstation_model->suspend(action->comm.surf_comm);
+ /*FIXME: shall we suspend also the timeout synchro? */
+ if (synchro->comm.surf_comm)
+ surf_action_suspend(synchro->comm.surf_comm);
+ /* in the other case, the action will be suspended on creation, in SIMIX_comm_start() */
}
-void SIMIX_comm_resume(smx_action_t action)
+void SIMIX_comm_resume(smx_synchro_t synchro)
{
/*FIXME: check what happen with the timeouts */
- surf_workstation_model->resume(action->comm.surf_comm);
+ if (synchro->comm.surf_comm)
+ surf_action_resume(synchro->comm.surf_comm);
+ /* in the other case, the synchro were not really suspended yet, see SIMIX_comm_suspend() and SIMIX_comm_start() */
}
-/************* Action Getters **************/
+/************* synchro Getters **************/
/**
* \brief get the amount remaining from the communication
- * \param action The communication
+ * \param synchro The communication
*/
-double SIMIX_comm_get_remains(smx_action_t action)
+double SIMIX_comm_get_remains(smx_synchro_t synchro)
{
double remains;
- if(!action){
- return 0;
+ if(!synchro){
+ return 0;
}
- switch (action->state) {
+ switch (synchro->state) {
- case SIMIX_RUNNING:
- remains = surf_workstation_model->get_remains(action->comm.surf_comm);
- break;
+ case SIMIX_RUNNING:
+ remains = surf_action_get_remains(synchro->comm.surf_comm);
+ break;
- case SIMIX_WAITING:
- case SIMIX_READY:
- remains = 0; /*FIXME: check what should be returned */
- break;
+ case SIMIX_WAITING:
+ case SIMIX_READY:
+ remains = 0; /*FIXME: check what should be returned */
+ break;
- default:
- remains = 0; /*FIXME: is this correct? */
- break;
+ default:
+ remains = 0; /*FIXME: is this correct? */
+ break;
}
return remains;
}
-e_smx_state_t SIMIX_comm_get_state(smx_action_t action)
+e_smx_state_t SIMIX_comm_get_state(smx_synchro_t synchro)
{
- return action->state;
+ return synchro->state;
}
/**
* \brief Return the user data associated to the sender of the communication
- * \param action The communication
+ * \param synchro The communication
* \return the user data
*/
-void* SIMIX_comm_get_src_data(smx_action_t action)
+void* SIMIX_comm_get_src_data(smx_synchro_t synchro)
{
- return action->comm.src_data;
+ return synchro->comm.src_data;
}
/**
* \brief Return the user data associated to the receiver of the communication
- * \param action The communication
+ * \param synchro The communication
* \return the user data
*/
-void* SIMIX_comm_get_dst_data(smx_action_t action)
+void* SIMIX_comm_get_dst_data(smx_synchro_t synchro)
{
- return action->comm.dst_data;
+ return synchro->comm.dst_data;
}
-smx_process_t SIMIX_comm_get_src_proc(smx_action_t action)
+smx_process_t SIMIX_comm_get_src_proc(smx_synchro_t synchro)
{
- return action->comm.src_proc;
+ return synchro->comm.src_proc;
}
-smx_process_t SIMIX_comm_get_dst_proc(smx_action_t action)
+smx_process_t SIMIX_comm_get_dst_proc(smx_synchro_t synchro)
{
- return action->comm.dst_proc;
+ return synchro->comm.dst_proc;
}
#ifdef HAVE_LATENCY_BOUND_TRACKING
* \brief verify if communication is latency bounded
* \param comm The communication
*/
-XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action)
+int SIMIX_comm_is_latency_bounded(smx_synchro_t synchro)
{
- if(!action){
- return 0;
+ if(!synchro){
+ return 0;
}
- if (action->comm.surf_comm){
- XBT_DEBUG("Getting latency limited for surf_action (%p)", action->comm.surf_comm);
- action->latency_limited = surf_workstation_model->get_latency_limited(action->comm.surf_comm);
- XBT_DEBUG("Action limited is %d", action->latency_limited);
+ if (synchro->comm.surf_comm){
+ XBT_DEBUG("Getting latency limited for surf_action (%p)", synchro->comm.surf_comm);
+ synchro->latency_limited = surf_network_action_get_latency_limited(synchro->comm.surf_comm);
+ XBT_DEBUG("synchro limited is %d", synchro->latency_limited);
}
- return action->latency_limited;
+ return synchro->latency_limited;
}
#endif
/******************************************************************************/
/* SIMIX_comm_copy_data callbacks */
/******************************************************************************/
-static void (*SIMIX_comm_copy_data_callback) (smx_action_t, size_t) =
- &SIMIX_comm_copy_pointer_callback;
+static void (*SIMIX_comm_copy_data_callback) (smx_synchro_t, void*, size_t) =
+ &SIMIX_comm_copy_pointer_callback;
void
-SIMIX_comm_set_copy_data_callback(void (*callback) (smx_action_t, size_t))
+SIMIX_comm_set_copy_data_callback(void (*callback) (smx_synchro_t, void*, size_t))
{
SIMIX_comm_copy_data_callback = callback;
}
-void SIMIX_comm_copy_pointer_callback(smx_action_t comm, size_t buff_size)
+void SIMIX_comm_copy_pointer_callback(smx_synchro_t comm, void* buff, size_t buff_size)
{
xbt_assert((buff_size == sizeof(void *)),
- "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
- *(void **) (comm->comm.dst_buff) = comm->comm.src_buff;
+ "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
+ *(void **) (comm->comm.dst_buff) = buff;
}
-void SIMIX_comm_copy_buffer_callback(smx_action_t comm, size_t buff_size)
+void SIMIX_comm_copy_buffer_callback(smx_synchro_t comm, void* buff, size_t buff_size)
{
- memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size);
+ XBT_DEBUG("Copy the data over");
+ memcpy(comm->comm.dst_buff, buff, buff_size);
+ if (comm->comm.detached) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the original buffer available to the application ASAP
+ xbt_free(buff);
+ comm->comm.src_buff = NULL;
+ }
}
+
/**
* \brief Copy the communication data from the sender's buffer to the receiver's one
* \param comm The communication
*/
-void SIMIX_comm_copy_data(smx_action_t comm)
+void SIMIX_comm_copy_data(smx_synchro_t comm)
{
size_t buff_size = comm->comm.src_buff_size;
/* If there is no data to be copy then return */
- if (!comm->comm.src_buff || !comm->comm.dst_buff || comm->comm.copied == 1)
+ if (!comm->comm.src_buff || !comm->comm.dst_buff || comm->comm.copied)
return;
XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
- comm,
- comm->comm.src_proc->smx_host->name, comm->comm.src_buff,
- comm->comm.dst_proc->smx_host->name, comm->comm.dst_buff, buff_size);
+ comm,
+ comm->comm.src_proc ? sg_host_name(comm->comm.src_proc->smx_host) : "a finished process",
+ comm->comm.src_buff,
+ comm->comm.dst_proc ? sg_host_name(comm->comm.dst_proc->smx_host) : "a finished process",
+ comm->comm.dst_buff, buff_size);
/* Copy at most dst_buff_size bytes of the message to receiver's buffer */
if (comm->comm.dst_buff_size)
if (comm->comm.dst_buff_size)
*comm->comm.dst_buff_size = buff_size;
- if (buff_size > 0)
- (*SIMIX_comm_copy_data_callback) (comm, buff_size);
+ if (buff_size > 0){
+ if(comm->comm.copy_data_fun)
+ comm->comm.copy_data_fun (comm, comm->comm.src_buff, buff_size);
+ else
+ SIMIX_comm_copy_data_callback (comm, 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) */