1 /* Copyright (c) 2009, 2010. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
12 /* Pimple to get an histogram of message sizes in the simulation */
13 xbt_dict_t msg_sizes = NULL;
14 #ifdef HAVE_LATENCY_BOUND_TRACKING
15 xbt_dict_t latency_limited_dict = NULL;
18 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
19 "Logging specific to SIMIX (network)");
21 /******************************************************************************/
22 /* Rendez-Vous Points */
23 /******************************************************************************/
26 * \brief Creates a new rendez-vous point
27 * \param name The name of the rendez-vous point
28 * \return The created rendez-vous point
30 smx_rdv_t SIMIX_rdv_create(const char *name)
32 smx_rdv_t rdv = xbt_new0(s_smx_rvpoint_t, 1);
33 rdv->name = name ? xbt_strdup(name) : NULL;
34 rdv->read = SIMIX_mutex_init();
35 rdv->write = SIMIX_mutex_init();
36 rdv->comm_fifo = xbt_fifo_new();
42 * \brief Destroy a rendez-vous point
43 * \param name The rendez-vous point to destroy
45 void SIMIX_rdv_destroy(smx_rdv_t rdv)
49 SIMIX_mutex_destroy(rdv->read);
50 SIMIX_mutex_destroy(rdv->write);
51 xbt_fifo_free(rdv->comm_fifo);
56 * \brief Push a communication request into a rendez-vous point
57 * \param rdv The rendez-vous point
58 * \param comm The communication request
60 static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_comm_t comm)
62 xbt_fifo_push(rdv->comm_fifo, comm);
67 * \brief Remove a communication request from a rendez-vous point
68 * \param rdv The rendez-vous point
69 * \param comm The communication request
71 static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_comm_t comm)
73 xbt_fifo_remove(rdv->comm_fifo, comm);
78 * \brief Checks if there is a communication request queued in a rendez-vous matching our needs
79 * \param type The type of communication we are looking for (comm_send, comm_recv)
80 * \return The communication request if found, NULL otherwise.
82 smx_comm_t SIMIX_rdv_get_request(smx_rdv_t rdv, smx_comm_type_t type)
84 smx_comm_t comm = (smx_comm_t)
85 xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
87 if (comm && comm->type == type) {
88 DEBUG0("Communication request found!");
89 xbt_fifo_shift(rdv->comm_fifo);
90 SIMIX_communication_use(comm);
95 DEBUG0("Communication request not found");
100 * \brief counts the number of communication requests of a given host pending
101 * on a rendez-vous point
102 * \param rdv The rendez-vous point
103 * \param host The host to be counted
104 * \return The number of comm request pending in the rdv
106 int SIMIX_rdv_get_count_waiting_comm(smx_rdv_t rdv, smx_host_t host)
108 smx_comm_t comm = NULL;
109 xbt_fifo_item_t item = NULL;
112 xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_comm_t) {
113 if (comm->src_proc->smx_host == host)
121 * \brief returns the communication at the head of the rendez-vous
122 * \param rdv The rendez-vous point
123 * \return The communication or NULL if empty
125 XBT_INLINE smx_comm_t SIMIX_rdv_get_head(smx_rdv_t rdv)
128 xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
131 /** @brief adds some API-related data to the rendez-vous point */
132 XBT_INLINE void SIMIX_rdv_set_data(smx_rdv_t rdv, void *data)
137 /** @brief gets API-related data from the rendez-vous point */
138 XBT_INLINE void *SIMIX_rdv_get_data(smx_rdv_t rdv)
143 /******************************************************************************/
144 /* Communication Requests */
145 /******************************************************************************/
148 * \brief Creates a new communication request
149 * \param type The type of communication (comm_send, comm_recv)
150 * \return The new communication request
152 smx_comm_t SIMIX_communication_new(smx_comm_type_t type)
154 /* alloc structures */
155 smx_comm_t comm = xbt_new0(s_smx_comm_t, 1);
157 comm->sem = SIMIX_sem_init(0);
164 * \brief Destroy a communication request
165 * \param comm The request to be destroyed
167 void SIMIX_communication_destroy(smx_comm_t comm)
169 VERB2("Destroy communication %p; refcount initially %d", comm,
172 #ifdef HAVE_LATENCY_BOUND_TRACKING
173 //save is latency limited flag to use afterwards
174 if (latency_limited_dict == NULL) {
175 latency_limited_dict = xbt_dict_new();
178 DEBUG2("adding key %p with latency limited value %d to the dict", comm,
179 SIMIX_action_is_latency_bounded(comm->act));
180 xbt_dicti_set(latency_limited_dict, (uintptr_t) comm,
181 SIMIX_action_is_latency_bounded(comm->act));
186 if (comm->refcount > 0)
190 SIMIX_sem_destroy(comm->sem);
195 SIMIX_action_destroy(comm->act);
199 if (comm->src_timeout) {
200 SIMIX_action_destroy(comm->src_timeout);
201 comm->src_timeout = NULL;
204 if (comm->dst_timeout) {
205 SIMIX_action_destroy(comm->dst_timeout);
206 comm->dst_timeout = NULL;
215 * \brief Increase the number of users of the communication.
216 * \param comm The communication request
217 * Each communication request can be used by more than one process, so it is
218 * necessary to know number of them at destroy time, to avoid freeing stuff that
219 * maybe is in use by others.
222 static XBT_INLINE void SIMIX_communication_use(smx_comm_t comm)
228 * \brief Start the simulation of a communication request
229 * \param comm The communication request
231 static XBT_INLINE void SIMIX_communication_start(smx_comm_t comm)
233 /* If both the sender and the receiver are already there, start the communication */
234 if (comm->src_proc && comm->dst_proc) {
235 DEBUG1("Starting communication %p", comm);
236 comm->act = SIMIX_action_communicate(comm->src_proc->smx_host,
237 comm->dst_proc->smx_host, NULL,
238 comm->task_size, comm->rate);
240 TRACE_smx_action_communicate(comm->act, comm->src_proc);
241 TRACE_surf_action(comm->act->surf_action, comm->act->category);
244 /* If any of the process is suspend, create the action but stop its execution,
245 it will be restarted when the sender process resume */
246 if (SIMIX_process_is_suspended(comm->src_proc) ||
247 SIMIX_process_is_suspended(comm->dst_proc)) {
248 SIMIX_action_suspend(comm->act);
251 /* Add the communication as user data of the action */
252 comm->act->data = comm;
254 /* The semaphore will only get signaled once, but since the first unlocked guy will
255 * release_forever() the semaphore, that will unlock the second (and any other)
256 * communication partner */
257 SIMIX_register_action_to_semaphore(comm->act, comm->sem);
262 * \brief Performs error checking and cleanup
263 * \param comm The communication
265 static XBT_INLINE void SIMIX_communication_cleanup(smx_comm_t comm)
267 DEBUG1("Checking errors and cleaning communication %p", comm);
269 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
270 SIMIX_sem_release_forever(comm->sem);
272 /* Check for errors other than timeouts */
273 if (!SIMIX_host_get_state(SIMIX_host_self())) {
275 SIMIX_rdv_remove(comm->rdv, comm);
276 SIMIX_communication_destroy(comm);
277 THROW0(host_error, 0, "Host failed");
278 } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED) {
279 SIMIX_communication_destroy(comm);
280 THROW0(network_error, 0, "Link failure");
281 } else if (!SIMIX_host_get_state(SIMIX_process_get_host(comm->dst_proc))
283 !SIMIX_host_get_state(SIMIX_process_get_host(comm->src_proc)))
285 /* We test both src&dst because we dunno who we are today, and we already tested myself above.
286 * So, at the end, we test the remote peer only
287 * Moreover, we have to test it because if the remote peer fails, the action comm->act is not done nor failed.
288 * In that case, we got awaken by the little endless actions created in the SIMIX_sem_acquire(comm->sem)
289 * at the beginning of this function. */
290 SIMIX_communication_destroy(comm);
291 THROW0(network_error, 0, "Remote peer failed");
294 /* Copy network data */
295 SIMIX_network_copy_data(comm);
297 SIMIX_communication_destroy(comm);
301 * \brief Waits for communication completion
302 * \param comm The communication
303 * \param timeout The max amount of time to wait for the communication to finish
306 * - host_error if local peer failed
307 * - timeout_error if communication reached the timeout specified (either because of local peer or remote peer)
308 * - network_error if network failed or remote peer failed
310 static XBT_INLINE void SIMIX_communication_wait_for_completion(smx_comm_t
315 smx_action_t act_sleep = NULL;
319 DEBUG1("Waiting for the completion of communication %p", comm);
322 act_sleep = SIMIX_action_sleep(SIMIX_host_self(), timeout);
323 if (SIMIX_process_self() == comm->src_proc)
324 comm->src_timeout = act_sleep;
326 comm->dst_timeout = act_sleep;
327 SIMIX_action_set_name(act_sleep,
329 ("Timeout for comm %p and wait on semaphore %p (max_duration:%f)",
330 comm, comm->sem, timeout));
331 SIMIX_register_action_to_semaphore(act_sleep, comm->sem);
332 SIMIX_process_self()->waiting_action = act_sleep;
333 SIMIX_sem_block_onto(comm->sem);
334 SIMIX_process_self()->waiting_action = NULL;
335 SIMIX_unregister_action_to_semaphore(act_sleep, comm->sem);
337 SIMIX_sem_acquire(comm->sem);
340 /* Check for timeouts */
341 if ((src_timeout = ((comm->src_timeout)
342 && (SIMIX_action_get_state(comm->src_timeout) ==
344 || (dst_timeout = ((comm->dst_timeout)
345 && (SIMIX_action_get_state(comm->dst_timeout) ==
346 SURF_ACTION_DONE)))) {
347 /* Somebody did a timeout! */
349 DEBUG1("Communication timeout from the src! %p", comm);
351 DEBUG1("Communication timeout from the dst! %p", comm);
354 && SIMIX_action_get_state(comm->act) == SURF_ACTION_RUNNING)
355 SIMIX_communication_cancel(comm);
357 SIMIX_rdv_remove(comm->rdv, comm);
359 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
360 SIMIX_sem_release_forever(comm->sem);
361 SIMIX_communication_destroy(comm);
363 THROW1(timeout_error, 0, "Communication timeouted because of %s",
364 src_timeout ? "the source" : "the destination");
367 DEBUG1("Communication %p complete!", comm);
368 SIMIX_communication_cleanup(comm);
372 * \brief Cancels a communication
373 * \brief comm The communication to cancel
375 XBT_INLINE void SIMIX_communication_cancel(smx_comm_t comm)
378 SIMIX_action_cancel(comm->act);
382 * \brief get the amount remaining from the communication
383 * \param comm The communication
385 XBT_INLINE double SIMIX_communication_get_remains(smx_comm_t comm)
387 DEBUG1("calling SIMIX_action_get_remains(%p)", comm->act);
388 return SIMIX_action_get_remains(comm->act);
391 #ifdef HAVE_LATENCY_BOUND_TRACKING
393 * \brief verify if communication is latency bounded
394 * \param comm The communication
396 XBT_INLINE int SIMIX_communication_is_latency_bounded(smx_comm_t comm)
398 //try to find comm on the list of finished flows
401 xbt_dict_cursor_t cursor;
402 xbt_dict_foreach(latency_limited_dict, cursor, key, data) {
403 DEBUG2("comparing key=%p with comm=%p", (void *) key, (void *) comm);
404 if ((void *) comm == (void *) key) {
405 DEBUG2("key %p found, return value latency limited value %d",
406 (void *) key, (int) data);
411 DEBUG1("calling SIMIX_action_is_latency_bounded(%p)", comm->act);
412 return SIMIX_action_is_latency_bounded(comm->act);
416 /******************************************************************************/
417 /* SIMIX_network_copy_data callbacks */
418 /******************************************************************************/
419 static void (*SIMIX_network_copy_data_callback) (smx_comm_t, size_t) =
420 &SIMIX_network_copy_pointer_callback;
423 SIMIX_network_set_copy_data_callback(void (*callback) (smx_comm_t, size_t))
425 SIMIX_network_copy_data_callback = callback;
428 void SIMIX_network_copy_pointer_callback(smx_comm_t comm, size_t buff_size)
430 xbt_assert1((buff_size == sizeof(void *)),
431 "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
432 *(void **) (comm->dst_buff) = comm->src_buff;
435 void SIMIX_network_copy_buffer_callback(smx_comm_t comm, size_t buff_size)
437 memcpy(comm->dst_buff, comm->src_buff, buff_size);
441 * \brief Copy the communication data from the sender's buffer to the receiver's one
442 * \param comm The communication
444 void SIMIX_network_copy_data(smx_comm_t comm)
446 size_t buff_size = comm->src_buff_size;
447 uintptr_t casted_size = 0;
448 uintptr_t amount = 0;
449 /* If there is no data to be copy then return */
450 if (!comm->src_buff || !comm->dst_buff || comm->copied == 1)
453 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
455 comm->src_proc->smx_host->name, comm->src_buff,
456 comm->dst_proc->smx_host->name, comm->dst_buff, buff_size);
458 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
459 if (comm->dst_buff_size)
460 buff_size = MIN(buff_size, *(comm->dst_buff_size));
462 /* Update the receiver's buffer size to the copied amount */
463 if (comm->dst_buff_size)
464 *comm->dst_buff_size = buff_size;
468 (*SIMIX_network_copy_data_callback) (comm, buff_size);
470 /* Set the copied flag so we copy data only once */
471 /* (this function might be called from both communication ends) */
474 /* pimple to display the message sizes */
476 if (msg_sizes == NULL)
477 msg_sizes = xbt_dict_new();
478 casted_size = comm->task_size;
479 amount = xbt_dicti_get(msg_sizes, casted_size);
482 xbt_dicti_set(msg_sizes, casted_size, amount);
487 /* pimple to display the message sizes */
488 void SIMIX_message_sizes_output(const char *filename)
492 xbt_dict_cursor_t cursor;
494 out = fopen(filename, "w");
495 xbt_assert1(out, "Cannot open file %s", filename);
497 xbt_dict_foreach(msg_sizes, cursor, key, data) {
498 fprintf(out, "%zu %zu\n", key, data);
504 * \brief Return the user data associated to the communication
505 * \param comm The communication
506 * \return the user data
508 XBT_INLINE void *SIMIX_communication_get_data(smx_comm_t comm)
513 XBT_PUBLIC(void *) SIMIX_communication_get_src_buf(smx_comm_t comm)
515 return comm->src_buff;
518 XBT_PUBLIC(void *) SIMIX_communication_get_dst_buf(smx_comm_t comm)
520 return comm->dst_buff;
523 XBT_PUBLIC(size_t) SIMIX_communication_get_src_buf_size(smx_comm_t comm)
525 return comm->src_buff_size;
528 XBT_PUBLIC(size_t) SIMIX_communication_get_dst_buf_size(smx_comm_t comm)
530 return *(comm->dst_buff_size);
533 /******************************************************************************/
534 /* Synchronous Communication */
535 /******************************************************************************/
537 * \brief Put a send communication request in a rendez-vous point and waits for
538 * its completion (blocking)
539 * \param rdv The rendez-vous point
540 * \param task_size The size of the communication action (for surf simulation)
541 * \param rate The rate of the communication action (for surf)
542 * \param timeout The timeout used for the waiting the completion
543 * \param src_buff The source buffer containing the message to be sent
544 * \param src_buff_size The size of the source buffer
545 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
546 * \param data User data associated to the communication object
548 * - host_error if peer failed
549 * - timeout_error if communication reached the timeout specified
550 * - network_error if network failed or peer issued a timeout
552 XBT_INLINE void SIMIX_network_send(smx_rdv_t rdv, double task_size,
553 double rate, double timeout,
554 void *src_buff, size_t src_buff_size,
555 smx_comm_t * comm_ref, void *data)
558 SIMIX_network_isend(rdv, task_size, rate, src_buff, src_buff_size,
560 SIMIX_network_wait(*comm_ref, timeout);
564 * \brief Put a receive communication request in a rendez-vous point and waits
565 * for its completion (blocking)
566 * \param rdv The rendez-vous point
567 * \param timeout The timeout used for the waiting the completion
568 * \param dst_buff The destination buffer to copy the received message
569 * \param src_buff_size The size of the destination buffer
570 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
572 * - host_error if peer failed
573 * - timeout_error if communication reached the timeout specified
574 * - network_error if network failed or peer issued a timeout
576 XBT_INLINE void SIMIX_network_recv(smx_rdv_t rdv, double timeout,
577 void *dst_buff, size_t * dst_buff_size,
578 smx_comm_t * comm_ref)
581 (smx_comm_t) SIMIX_network_irecv(rdv, dst_buff, dst_buff_size);
582 SIMIX_network_wait(*comm_ref, timeout);
585 /******************************************************************************/
586 /* Asynchronous Communication */
587 /******************************************************************************/
588 smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size,
589 double rate, void *src_buff,
590 size_t src_buff_size, void *data)
594 /*If running in model-checking mode then intercept the communication action */
596 if (_surf_do_model_check)
597 MC_trans_intercept_isend(rdv);
599 /* Look for communication request matching our needs.
600 If it is not found then create it and push it into the rendez-vous point */
601 comm = SIMIX_rdv_get_request(rdv, comm_recv);
604 comm = SIMIX_communication_new(comm_send);
605 SIMIX_rdv_push(rdv, comm);
608 /* Setup the communication request */
609 comm->src_proc = SIMIX_process_self();
610 comm->task_size = task_size;
612 comm->src_buff = src_buff;
613 comm->src_buff_size = src_buff_size;
616 SIMIX_communication_start(comm);
620 smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff,
621 size_t * dst_buff_size)
625 /*If running in model-checking mode then intercept the communication action */
627 if (_surf_do_model_check)
628 MC_trans_intercept_irecv(rdv);
630 /* Look for communication request matching our needs.
631 * If it is not found then create it and push it into the rendez-vous point
633 comm = SIMIX_rdv_get_request(rdv, comm_send);
636 comm = SIMIX_communication_new(comm_recv);
637 SIMIX_rdv_push(rdv, comm);
640 /* Setup communication request */
641 comm->dst_proc = SIMIX_process_self();
642 comm->dst_buff = dst_buff;
643 comm->dst_buff_size = dst_buff_size;
645 SIMIX_communication_start(comm);
649 /** @brief blocks until the communication terminates or the timeout occurs */
650 XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout)
652 /*If running in model-checking mode then intercept the communication action */
654 if (_surf_do_model_check)
655 MC_trans_intercept_wait(comm);
657 /* Wait for communication completion */
658 SIMIX_communication_wait_for_completion(comm, timeout);
661 /** @Returns whether the (asynchronous) communication is done yet or not */
662 XBT_INLINE int SIMIX_network_test(smx_comm_t comm)
664 /*If running in model-checking mode then intercept the communication action */
666 if (_surf_do_model_check)
667 MC_trans_intercept_test(comm);
670 /* Copy data if the communication is done */
671 if (comm->sem && !SIMIX_sem_would_block(comm->sem)) {
672 /* Copy network data */
673 SIMIX_network_copy_data(comm);
679 /** @brief wait for the completion of any communication of a set
681 * @Returns the rank in the dynar of communication which finished; destroy it after identifying which one it is
683 unsigned int SIMIX_network_waitany(xbt_dynar_t comms)
685 xbt_dynar_t sems = xbt_dynar_new(sizeof(smx_sem_t), NULL);
686 unsigned int cursor, found_comm = -1;
687 smx_comm_t comm, comm_finished = NULL;
689 /*If running in model-checking mode then intercept the communication action */
691 if (_surf_do_model_check)
692 MC_trans_intercept_waitany(comms);
694 xbt_dynar_foreach(comms, cursor, comm)
695 xbt_dynar_push(sems, &(comm->sem));
697 DEBUG1("Waiting for the completion of communication set %p", comms);
699 found_comm = SIMIX_sem_acquire_any(sems);
700 xbt_dynar_free_container(&sems);
701 xbt_assert0(found_comm != -1,
702 "Cannot find which communication finished");
703 xbt_dynar_get_cpy(comms, found_comm, &comm_finished);
705 DEBUG2("Communication %p of communication set %p finished",
706 comm_finished, comms);
708 /* let the regular code deal with the communication end (errors checking and cleanup).
709 * A bit of useless work will be done, but that's good for source factorization */
710 SIMIX_sem_release_forever(comm_finished->sem);
711 SIMIX_communication_wait_for_completion(comm_finished, -1);