#include "private.h"
#include "xbt/log.h"
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
+ "Logging specific to SIMIX (network)");
+
/******************************************************************************/
/* Rendez-Vous Points */
/******************************************************************************/
*/
smx_rdv_t SIMIX_rdv_create(const char *name)
{
- smx_rdv_t rvp = xbt_new0(s_smx_rvpoint_t, 1);
- rvp->name = name ? xbt_strdup(name) : NULL;
- rvp->read = SIMIX_mutex_init();
- rvp->write = SIMIX_mutex_init();
- rvp->comm_fifo = xbt_fifo_new();
+ smx_rdv_t rdv = xbt_new0(s_smx_rvpoint_t, 1);
+ rdv->name = name ? xbt_strdup(name) : NULL;
+ rdv->read = SIMIX_mutex_init();
+ rdv->write = SIMIX_mutex_init();
+ rdv->comm_fifo = xbt_fifo_new();
- return rvp;
+ return rdv;
}
/**
* \brief Destroy a rendez-vous point
* \param name The rendez-vous point to destroy
*/
-void SIMIX_rdv_destroy(smx_rdv_t rvp)
+void SIMIX_rdv_destroy(smx_rdv_t rdv)
{
- if(rvp->name)
- xbt_free(rvp->name);
- SIMIX_mutex_destroy(rvp->read);
- SIMIX_mutex_destroy(rvp->write);
- xbt_fifo_free(rvp->comm_fifo);
- xbt_free(rvp);
+ if(rdv->name)
+ xbt_free(rdv->name);
+ SIMIX_mutex_destroy(rdv->read);
+ SIMIX_mutex_destroy(rdv->write);
+ xbt_fifo_free(rdv->comm_fifo);
+ xbt_free(rdv);
}
/**
* \brief Push a communication request into a rendez-vous point
- * The communications request are dequeued by the two functions below
- * \param rvp The rendez-vous point
+ * \param rdv The rendez-vous point
* \param comm The communication request
*/
-static inline void SIMIX_rdv_push(smx_rdv_t rvp, smx_comm_t comm)
+static inline void SIMIX_rdv_push(smx_rdv_t rdv, smx_comm_t comm)
{
- xbt_fifo_push(rvp->comm_fifo, comm);
+ xbt_fifo_push(rdv->comm_fifo, comm);
+ comm->rdv = rdv;
}
+/**
+ * \brief Remove a communication request from a rendez-vous point
+ * \param rdv The rendez-vous point
+ * \param comm The communication request
+ */
+static inline void SIMIX_rdv_remove(smx_rdv_t rdv, smx_comm_t comm)
+{
+ xbt_fifo_remove(rdv->comm_fifo, comm);
+ comm->rdv = NULL;
+}
+
/**
* \brief Checks if there is a communication request queued in a rendez-vous matching our needs
- * \param rvp The rendez-vous with the queue
- * \param look_for_src boolean. True: we are receiver looking for sender; False: other way round
- * \return The communication request if found, or a newly created one otherwise.
+ * \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 rvp, int (filter)(smx_comm_t, void*), void *arg) {
- smx_comm_t comm;
- xbt_fifo_item_t item;
-
- /* Traverse the rendez-vous queue looking for a comm request matching the
- filter conditions. If found return it and remove it from the list. */
- xbt_fifo_foreach(rvp->comm_fifo, item, comm, smx_comm_t) {
- if(filter(comm, arg)){
- SIMIX_communication_use(comm);
- xbt_fifo_remove_item(rvp->comm_fifo, item);
- return comm;
- }
+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));
+
+ if(comm && comm->type == type){
+ DEBUG0("Communication request found!");
+ xbt_fifo_shift(rdv->comm_fifo);
+ SIMIX_communication_use(comm);
+ comm->rdv = NULL;
+ return comm;
}
/* no relevant request found. Return NULL */
+ DEBUG0("Communication request not found");
return NULL;
}
+/**
+ * \brief counts the number of communication requests of a given host pending
+ * on a rendez-vous point
+ * \param rdv The rendez-vous point
+ * \param host The host to be counted
+ * \return The number of comm request pending in the rdv
+ */
+int
+SIMIX_rdv_get_count_waiting_comm(smx_rdv_t rdv, smx_host_t host)
+{
+ smx_comm_t comm = NULL;
+ xbt_fifo_item_t item = NULL;
+ int count = 0;
+
+ xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_comm_t) {
+ if (comm->src_proc->smx_host == host)
+ count++;
+ }
+
+ return count;
+}
+
+/**
+ * \brief returns the communication at the head of the rendez-vous
+ * \param rdv The rendez-vous point
+ * \return The communication or NULL if empty
+ */
+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) {
+ rdv->data=data;
+}
+/** @brief gets API-related data from the rendez-vous point */
+void *SIMIX_rdv_get_data(smx_rdv_t rdv) {
+ return rdv->data;
+}
+
/******************************************************************************/
/* Communication Requests */
/******************************************************************************/
/**
* \brief Creates a new communication request
- * \param sender The process starting the communication (by send)
- * \param receiver The process receiving the communication (by recv)
- * \return the communication request
+ * \param type The type of communication (comm_send, comm_recv)
+ * \return The new communication request
*/
-smx_comm_t SIMIX_communication_new(smx_comm_type_t type, smx_rdv_t rdv)
+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->rdv = rdv;
+ comm->sem = SIMIX_sem_init(0);
comm->refcount = 1;
return comm;
void SIMIX_communication_destroy(smx_comm_t comm)
{
comm->refcount--;
- if(comm->refcount == 0){
- if(comm->act != NULL)
- SIMIX_action_destroy(comm->act);
+ if(comm->refcount > 0)
+ return;
- xbt_free(comm->cond);
- xbt_free(comm);
+ if(comm->sem){
+ SIMIX_sem_destroy(comm->sem);
+ comm->sem = NULL;
+ }
+
+ if(comm->act){
+ SIMIX_action_destroy(comm->act);
+ comm->act = NULL;
}
+
+ xbt_free(comm);
}
/**
/**
* \brief Start the simulation of a communication request
- * \param comm The communication request
+ * \param comm The comm->rdv = NULL;communication request
*/
static inline void SIMIX_communication_start(smx_comm_t comm)
{
/* If both the sender and the receiver are already there, start the communication */
- if(comm->src_host != NULL && comm->dst_host != NULL){
- comm->act = SIMIX_action_communicate(comm->src_host, comm->dst_host, NULL,
+ if(comm->src_proc && comm->dst_proc){
+ DEBUG1("Starting communication %p", comm);
+ comm->act = SIMIX_action_communicate(comm->src_proc->smx_host,
+ comm->dst_proc->smx_host, NULL,
comm->task_size, comm->rate);
- /* Add the communication as user data into the action, so it can be reached from it later */
- comm->act->data = comm;
+
+ /* 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_register_action_to_condition(comm->act, comm->cond);
+ /* Add the communication as user data of the action */
+ comm->act->data = comm;
+
+ /* 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);
}
}
static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
{
xbt_ex_t e;
+
+ DEBUG1("Waiting for the completion of communication %p", comm);
if(timeout > 0){
TRY{
- SIMIX_cond_wait_timeout(comm->cond, NULL, timeout);
+ SIMIX_sem_acquire_timeout(comm->sem, timeout);
}
CATCH(e){
- /* If it's a timeout then cancel the communication and signal the other peer */
- if(e.category == timeout_error)
- SIMIX_action_cancel(comm->act);
- SIMIX_cond_signal(comm->cond);
+ /* 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;
}
- }else{
- SIMIX_cond_wait(comm->cond, NULL);
+ } else {
+ SIMIX_sem_acquire(comm->sem);
}
- SIMIX_unregister_action_to_condition(comm->act, comm->cond);
+ 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 */
- if (SIMIX_host_get_state(comm->dst_host) == 0){
- THROW1(host_error, 0, "Destination host %s failed", comm->dst_host->name);
- } else if (SIMIX_host_get_state(comm->src_host) == 0){
- THROW1(host_error, 0, "Source host %s failed", comm->src_host->name);
+ /* Check for errors other than timeouts (they are catched above) */
+ if(!SIMIX_host_get_state(SIMIX_host_self())){
+ 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){
+ SIMIX_communication_destroy(comm);
THROW0(network_error, 0, "Link failure");
}
}
/**
- * \brief Copy the communication data from the sender's buffer to the receiver's one
- * \param comm The communication
+ * \brief Cancels a communication
+ * \brief comm The communication to cancel
*/
-void SIMIX_network_copy_data(smx_comm_t comm)
+void SIMIX_communication_cancel(smx_comm_t comm)
{
- /* Copy the minimum between the size of the sender's message and the size of the
- receiver's buffer */
- *comm->dest_buff_size = *comm->dest_buff_size < comm->data_size ?
- *comm->dest_buff_size : comm->data_size;
-
- memcpy(comm->dest_buff, comm->data, *comm->dest_buff_size);
+ SIMIX_action_cancel(comm->act);
}
/**
- * \brief Checks if a communication is a send request
+ * \brief get the amount remaining from the communication
* \param comm The communication
- * \return Boolean value
*/
-int SIMIX_communication_isSend(smx_comm_t comm)
+double SIMIX_communication_get_remains(smx_comm_t comm)
{
- return comm->type == comm_send ? TRUE : FALSE;
-}
+ return SIMIX_action_get_remains(comm->act);
+}
/**
- * \brief Checks if a communication is a recv request
+ * \brief Copy the communication data from the sender's buffer to the receiver's one
* \param comm The communication
- * \return Boolean value
*/
-int SIMIX_communication_isRecv(smx_comm_t comm)
+void SIMIX_network_copy_data(smx_comm_t comm)
{
- return comm->type == comm_recv ? TRUE : FALSE;
-}
+ /* 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);
+
+ /* Update the receiver's buffer size to the copied amount */
+ if (comm->dst_buff_size)
+ *comm->dst_buff_size = dst_buff_size;
-/* FIXME: move to some other place */
-int comm_filter_get(smx_comm_t comm, void *arg)
-{
- if(comm->type == comm_send){
- if(arg && comm->src_host != (smx_host_t)arg)
- return FALSE;
- else
- return TRUE;
- }else{
- return FALSE;
- }
+ if(dst_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);
}
-int comm_filter_put(smx_comm_t comm, void *arg)
+/**
+ * \brief Return the user data associated to the communication
+ * \param comm The communication
+ * \return the user data
+ */
+void *SIMIX_communication_get_data(smx_comm_t comm)
{
- return comm->type == comm_recv ? TRUE : FALSE;
+ return comm->data;
}
+
/******************************************************************************/
/* Synchronous Communication */
/******************************************************************************/
-/* Throws:
+/**
+ * \brief Put a send communication request in a rendez-vous point and waits for
+ * its completion (blocking)
+ * \param rdv The rendez-vous point
+ * \param task_size The size of the communication action (for surf simulation)
+ * \param rate The rate of the communication action (for surf)
+ * \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 (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,
- double timeout, void *data, size_t data_size,
- int (filter)(smx_comm_t, void *), void *arg)
+ 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
+ */
+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, filter, arg);
+ comm = SIMIX_rdv_get_request(rdv, comm_recv);
- if(comm == NULL){
- comm = SIMIX_communication_new(comm_send, rdv);
+ if(!comm){
+ comm = SIMIX_communication_new(comm_send);
SIMIX_rdv_push(rdv, comm);
}
/* Setup the communication request */
- comm->src_host = SIMIX_host_self();
+ comm->src_proc = SIMIX_process_self();
comm->task_size = task_size;
comm->rate = rate;
+ comm->src_buff = src_buff;
+ comm->src_buff_size = src_buff_size;
comm->data = data;
- comm->data_size = data_size;
SIMIX_communication_start(comm);
-
- /* Wait for communication completion */
- /* FIXME: if the semantic is non blocking, it shouldn't wait on the condition here */
- /* FIXME: add timeout checking stuff */
- SIMIX_communication_wait_for_completion(comm, timeout);
-
- SIMIX_communication_destroy(comm);
+ return comm;
}
-/* 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 *data,
- size_t *data_size, int (filter)(smx_comm_t, void *), void *arg)
-{
+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, filter, arg);
+ comm = SIMIX_rdv_get_request(rdv, comm_send);
- if(comm == NULL){
- comm = SIMIX_communication_new(comm_send, rdv);
+ if(!comm){
+ comm = SIMIX_communication_new(comm_recv);
SIMIX_rdv_push(rdv, comm);
}
/* Setup communication request */
- comm->dst_host = SIMIX_host_self();
- comm->dest_buff = data;
- comm->dest_buff_size = data_size;
+ 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 */
+void SIMIX_network_wait(smx_comm_t comm, double timeout) {
/* Wait for communication completion */
- /* FIXME: if the semantic is non blocking, it shouldn't wait on the condition here */
- /* FIXME: add timeout checking stuff */
SIMIX_communication_wait_for_completion(comm, timeout);
SIMIX_communication_destroy(comm);
}
-/******************************************************************************/
-/* Asynchronous Communication */
-/******************************************************************************/
-
-/*
-void SIMIX_network_wait(smx_action_t comm, double timeout)
-{
- if (timeout > 0)
- SIMIX_cond_wait_timeout(rvp_cond, rvp_comm_mutex, timeout - start_time);
- else
- SIMIX_cond_wait(rvp_cond, rvp_comm_mutex);
-
+/** @Returns whether the (asynchronous) communication is done yet or not */
+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 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_PUBLIC(int) SIMIX_network_test(smx_action_t comm)
-{
- if(SIMIX_action_get_state (comm) == SURF_ACTION_DONE){
- memcpy(comm->data
-
- return SIMIX_action_get_state (comm) == SURF_ACTION_DONE ? TRUE : FALSE;
-}*/
-
+ 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;
+}
