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. */
11 /* Pimple to get an histogram of message sizes in the simulation */
12 xbt_dict_t msg_sizes = NULL;
14 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
15 "Logging specific to SIMIX (network)");
17 /******************************************************************************/
18 /* Rendez-Vous Points */
19 /******************************************************************************/
22 * \brief Creates a new rendez-vous point
23 * \param name The name of the rendez-vous point
24 * \return The created rendez-vous point
26 smx_rdv_t SIMIX_rdv_create(const char *name)
28 smx_rdv_t rdv = xbt_new0(s_smx_rvpoint_t, 1);
29 rdv->name = name ? xbt_strdup(name) : NULL;
30 rdv->read = SIMIX_mutex_init();
31 rdv->write = SIMIX_mutex_init();
32 rdv->comm_fifo = xbt_fifo_new();
38 * \brief Destroy a rendez-vous point
39 * \param name The rendez-vous point to destroy
41 void SIMIX_rdv_destroy(smx_rdv_t rdv)
45 SIMIX_mutex_destroy(rdv->read);
46 SIMIX_mutex_destroy(rdv->write);
47 xbt_fifo_free(rdv->comm_fifo);
52 * \brief Push a communication request into a rendez-vous point
53 * \param rdv The rendez-vous point
54 * \param comm The communication request
56 static inline void SIMIX_rdv_push(smx_rdv_t rdv, smx_comm_t comm)
58 xbt_fifo_push(rdv->comm_fifo, comm);
63 * \brief Remove a communication request from a rendez-vous point
64 * \param rdv The rendez-vous point
65 * \param comm The communication request
67 static inline void SIMIX_rdv_remove(smx_rdv_t rdv, smx_comm_t comm)
69 xbt_fifo_remove(rdv->comm_fifo, comm);
74 * \brief Checks if there is a communication request queued in a rendez-vous matching our needs
75 * \param type The type of communication we are looking for (comm_send, comm_recv)
76 * \return The communication request if found, NULL otherwise.
78 smx_comm_t SIMIX_rdv_get_request(smx_rdv_t rdv, smx_comm_type_t type) {
79 smx_comm_t comm = (smx_comm_t)xbt_fifo_get_item_content(
80 xbt_fifo_get_first_item(rdv->comm_fifo));
82 if(comm && comm->type == type){
83 DEBUG0("Communication request found!");
84 xbt_fifo_shift(rdv->comm_fifo);
85 SIMIX_communication_use(comm);
90 DEBUG0("Communication request not found");
95 * \brief counts the number of communication requests of a given host pending
96 * on a rendez-vous point
97 * \param rdv The rendez-vous point
98 * \param host The host to be counted
99 * \return The number of comm request pending in the rdv
102 SIMIX_rdv_get_count_waiting_comm(smx_rdv_t rdv, smx_host_t host)
104 smx_comm_t comm = NULL;
105 xbt_fifo_item_t item = NULL;
108 xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_comm_t) {
109 if (comm->src_proc->smx_host == host)
117 * \brief returns the communication at the head of the rendez-vous
118 * \param rdv The rendez-vous point
119 * \return The communication or NULL if empty
121 XBT_INLINE smx_comm_t SIMIX_rdv_get_head(smx_rdv_t rdv)
123 return (smx_comm_t)xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
126 /** @brief adds some API-related data to the rendez-vous point */
127 XBT_INLINE void SIMIX_rdv_set_data(smx_rdv_t rdv,void *data) {
130 /** @brief gets API-related data from the rendez-vous point */
131 XBT_INLINE void *SIMIX_rdv_get_data(smx_rdv_t rdv) {
135 /******************************************************************************/
136 /* Communication Requests */
137 /******************************************************************************/
140 * \brief Creates a new communication request
141 * \param type The type of communication (comm_send, comm_recv)
142 * \return The new communication request
144 smx_comm_t SIMIX_communication_new(smx_comm_type_t type)
146 /* alloc structures */
147 smx_comm_t comm = xbt_new0(s_smx_comm_t, 1);
149 comm->sem = SIMIX_sem_init(0);
156 * \brief Destroy a communication request
157 * \param comm The request to be destroyed
159 void SIMIX_communication_destroy(smx_comm_t comm)
162 if(comm->refcount > 0)
166 SIMIX_sem_destroy(comm->sem);
171 SIMIX_action_destroy(comm->act);
175 if(comm->src_timeout){
176 SIMIX_action_destroy(comm->src_timeout);
177 comm->src_timeout = NULL;
180 if(comm->dst_timeout){
181 SIMIX_action_destroy(comm->dst_timeout);
182 comm->dst_timeout = NULL;
189 * \brief Increase the number of users of the communication.
190 * \param comm The communication request
191 * Each communication request can be used by more than one process, so it is
192 * necessary to know number of them at destroy time, to avoid freeing stuff that
193 * maybe is in use by others.
196 static inline void SIMIX_communication_use(smx_comm_t comm)
202 * \brief Start the simulation of a communication request
203 * \param comm The comm->rdv = NULL;communication request
205 static inline void SIMIX_communication_start(smx_comm_t comm)
207 /* If both the sender and the receiver are already there, start the communication */
208 if(comm->src_proc && comm->dst_proc){
209 DEBUG1("Starting communication %p", comm);
210 comm->act = SIMIX_action_communicate(comm->src_proc->smx_host,
211 comm->dst_proc->smx_host, NULL,
212 comm->task_size, comm->rate);
214 TRACE_smx_action_communicate (comm->act, comm->src_proc);
217 /* If any of the process is suspend, create the action but stop its execution,
218 it will be restarted when the sender process resume */
219 if(SIMIX_process_is_suspended(comm->src_proc) ||
220 SIMIX_process_is_suspended(comm->dst_proc)) {
221 SIMIX_action_suspend(comm->act);
224 /* Add the communication as user data of the action */
225 comm->act->data = comm;
227 /* The semaphore will only get signaled once, but since the first unlocked guy will
228 * release_forever() the semaphore, that will unlock the second (and any other)
229 * communication partner */
230 SIMIX_register_action_to_semaphore(comm->act, comm->sem);
235 * \brief Waits for communication completion and performs error checking
236 * \param comm The communication
237 * \param timeout The max amount of time to wait for the communication to finish
240 * - host_error if peer failed
241 * - timeout_error if communication reached the timeout specified
242 * - network_error if network failed or peer issued a timeout
244 static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
246 smx_action_t act_sleep = NULL;
250 DEBUG1("Waiting for the completion of communication %p", comm);
253 act_sleep = SIMIX_action_sleep(SIMIX_host_self(), timeout);
254 if(SIMIX_process_self()==comm->src_proc)
255 comm->src_timeout = act_sleep;
257 comm->dst_timeout = act_sleep;
258 SIMIX_action_set_name(act_sleep,bprintf("Timeout for comm %p and wait on semaphore %p (max_duration:%f)", comm, comm->sem,timeout));
259 SIMIX_register_action_to_semaphore(act_sleep, comm->sem);
260 SIMIX_process_self()->waiting_action = act_sleep;
261 SIMIX_sem_block_onto(comm->sem);
262 SIMIX_process_self()->waiting_action = NULL;
263 SIMIX_unregister_action_to_semaphore(act_sleep, comm->sem);
265 SIMIX_sem_acquire(comm->sem);
268 /* Check for timeouts */
269 if ((src_timeout = ((comm->src_timeout) && (SIMIX_action_get_state(comm->src_timeout) == SURF_ACTION_DONE))) ||
270 (dst_timeout = ((comm->dst_timeout) && (SIMIX_action_get_state(comm->dst_timeout) == SURF_ACTION_DONE))) ) {
271 /* Somebody did a timeout! */
272 if (src_timeout) DEBUG1("Communication timeout from the src! %p", comm);
273 if (dst_timeout) DEBUG1("Communication timeout from the dst! %p", comm);
275 if(comm->act && SIMIX_action_get_state(comm->act) == SURF_ACTION_RUNNING)
276 SIMIX_communication_cancel(comm);
278 SIMIX_rdv_remove(comm->rdv, comm);
280 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
281 SIMIX_sem_release_forever(comm->sem);
282 SIMIX_communication_destroy(comm);
284 THROW1(timeout_error, 0, "Communication timeouted because of %s",src_timeout?"the source":"the destination");
287 DEBUG1("Communication %p complete! Let's check for errors", comm);
289 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
290 SIMIX_sem_release_forever(comm->sem);
292 /* Check for errors other than timeouts (they are catched above) */
293 if(!SIMIX_host_get_state(SIMIX_host_self())){
295 SIMIX_rdv_remove(comm->rdv, comm);
296 SIMIX_communication_destroy(comm);
297 THROW0(host_error, 0, "Host failed");
298 } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
299 SIMIX_communication_destroy(comm);
300 THROW0(network_error, 0, "Link failure");
302 SIMIX_communication_destroy(comm);
306 * \brief Cancels a communication
307 * \brief comm The communication to cancel
309 XBT_INLINE void SIMIX_communication_cancel(smx_comm_t comm)
312 SIMIX_action_cancel(comm->act);
316 * \brief get the amount remaining from the communication
317 * \param comm The communication
319 XBT_INLINE double SIMIX_communication_get_remains(smx_comm_t comm)
321 return SIMIX_action_get_remains(comm->act);
324 /******************************************************************************/
325 /* SIMIX_network_copy_data callbacks */
326 /******************************************************************************/
327 static void (*SIMIX_network_copy_data_callback)(smx_comm_t, size_t) = &SIMIX_network_copy_pointer_callback;
329 void SIMIX_network_set_copy_data_callback(void (*callback)(smx_comm_t, size_t)) {
330 SIMIX_network_copy_data_callback = callback;
333 void SIMIX_network_copy_pointer_callback(smx_comm_t comm, size_t buff_size) {
334 xbt_assert1((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)",buff_size);
335 *(void**)(comm->dst_buff) = comm->src_buff;
338 void SIMIX_network_copy_buffer_callback(smx_comm_t comm, size_t buff_size) {
339 memcpy(comm->dst_buff, comm->src_buff, buff_size);
343 * \brief Copy the communication data from the sender's buffer to the receiver's one
344 * \param comm The communication
346 void SIMIX_network_copy_data(smx_comm_t comm)
348 size_t buff_size = comm->src_buff_size;
350 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
352 comm->src_proc->smx_host->name, comm->src_buff,
353 comm->dst_proc->smx_host->name, comm->dst_buff,
356 /* If there is no data to be copy then return */
357 if(!comm->src_buff || !comm->dst_buff)
360 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
361 if (comm->dst_buff_size)
362 buff_size = MIN(buff_size,*(comm->dst_buff_size));
364 /* Update the receiver's buffer size to the copied amount */
365 if (comm->dst_buff_size)
366 *comm->dst_buff_size = buff_size;
370 (*SIMIX_network_copy_data_callback)(comm, buff_size);
372 /* pimple to display the message sizes */
374 if (msg_sizes == NULL)
375 msg_sizes = xbt_dict_new();
376 uintptr_t casted_size = comm->task_size;
377 uintptr_t amount = xbt_dicti_get(msg_sizes, casted_size);
380 xbt_dicti_set(msg_sizes,casted_size, amount);
384 /* pimple to display the message sizes */
385 void SIMIX_message_sizes_output(const char *filename) {
386 FILE * out = fopen(filename,"w");
387 xbt_assert1(out,"Cannot open file %s",filename);
389 xbt_dict_cursor_t cursor;
390 xbt_dict_foreach(msg_sizes,cursor,key,data) {
391 fprintf(out,"%zu %zu\n",key,data);
397 * \brief Return the user data associated to the communication
398 * \param comm The communication
399 * \return the user data
401 XBT_INLINE void *SIMIX_communication_get_data(smx_comm_t comm)
406 /******************************************************************************/
407 /* Synchronous Communication */
408 /******************************************************************************/
410 * \brief Put a send communication request in a rendez-vous point and waits for
411 * its completion (blocking)
412 * \param rdv The rendez-vous point
413 * \param task_size The size of the communication action (for surf simulation)
414 * \param rate The rate of the communication action (for surf)
415 * \param timeout The timeout used for the waiting the completion
416 * \param src_buff The source buffer containing the message to be sent
417 * \param src_buff_size The size of the source buffer
418 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
419 * \param data User data associated to the communication object
421 * - host_error if peer failed
422 * - timeout_error if communication reached the timeout specified
423 * - network_error if network failed or peer issued a timeout
425 XBT_INLINE void SIMIX_network_send(smx_rdv_t rdv, double task_size, double rate,
426 double timeout, void *src_buff, size_t src_buff_size,
427 smx_comm_t *comm_ref, void *data)
429 *comm_ref = SIMIX_network_isend(rdv,task_size,rate,src_buff,src_buff_size,data);
430 SIMIX_network_wait(*comm_ref,timeout);
434 * \brief Put a receive communication request in a rendez-vous point and waits
435 * for its completion (blocking)
436 * \param rdv The rendez-vous point
437 * \param timeout The timeout used for the waiting the completion
438 * \param dst_buff The destination buffer to copy the received message
439 * \param src_buff_size The size of the destination buffer
440 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
442 * - host_error if peer failed
443 * - timeout_error if communication reached the timeout specified
444 * - network_error if network failed or peer issued a timeout
446 XBT_INLINE void SIMIX_network_recv(smx_rdv_t rdv, double timeout, void *dst_buff,
447 size_t *dst_buff_size, smx_comm_t *comm_ref)
449 *comm_ref = SIMIX_network_irecv(rdv,dst_buff,dst_buff_size);
450 SIMIX_network_wait(*comm_ref,timeout);
453 /******************************************************************************/
454 /* Asynchronous Communication */
455 /******************************************************************************/
456 smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size, double rate,
457 void *src_buff, size_t src_buff_size, void *data)
461 /* Look for communication request matching our needs.
462 If it is not found then create it and push it into the rendez-vous point */
463 comm = SIMIX_rdv_get_request(rdv, comm_recv);
466 comm = SIMIX_communication_new(comm_send);
467 SIMIX_rdv_push(rdv, comm);
470 /* Setup the communication request */
471 comm->src_proc = SIMIX_process_self();
472 comm->task_size = task_size;
474 comm->src_buff = src_buff;
475 comm->src_buff_size = src_buff_size;
478 SIMIX_communication_start(comm);
482 smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_size) {
485 /* Look for communication request matching our needs.
486 If it is not found then create it and push it into the rendez-vous point */
487 comm = SIMIX_rdv_get_request(rdv, comm_send);
490 comm = SIMIX_communication_new(comm_recv);
491 SIMIX_rdv_push(rdv, comm);
494 /* Setup communication request */
495 comm->dst_proc = SIMIX_process_self();
496 comm->dst_buff = dst_buff;
497 comm->dst_buff_size = dst_buff_size;
499 SIMIX_communication_start(comm);
503 /** @brief blocks until the communication terminates or the timeout occurs */
504 XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout) {
505 /* Wait for communication completion */
506 SIMIX_communication_wait_for_completion(comm, timeout);
509 /** @Returns whether the (asynchronous) communication is done yet or not */
510 XBT_INLINE int SIMIX_network_test(smx_comm_t comm) {
511 return comm->sem?SIMIX_sem_would_block(comm->sem):0;
514 /** @brief wait for the completion of any communication of a set
516 * @Returns the rank in the dynar of communication which finished; destroy it after identifying which one it is
518 unsigned int SIMIX_network_waitany(xbt_dynar_t comms) {
519 xbt_dynar_t sems = xbt_dynar_new(sizeof(smx_sem_t),NULL);
520 unsigned int cursor, found_comm=-1;
521 smx_comm_t comm,comm_finished=NULL;
523 xbt_dynar_foreach(comms,cursor,comm){
524 xbt_dynar_push(sems,&(comm->sem));
527 DEBUG1("Waiting for the completion of communication set %p", comms);
529 found_comm = SIMIX_sem_acquire_any(sems);
530 xbt_assert0(found_comm!=-1,"Cannot find which communication finished");
531 xbt_dynar_get_cpy(comms,found_comm,&comm_finished);
533 DEBUG1("Communication %p complete! Let's check for errors", comm_finished);
535 /* Make sure that everyone sleeping on that semaphore is awake,
536 * and that nobody will ever block on it */
537 SIMIX_sem_release_forever(comm_finished->sem);
539 /* Check for errors */
540 if(!SIMIX_host_get_state(SIMIX_host_self())){
541 if(comm_finished->rdv)
542 SIMIX_rdv_remove(comm_finished->rdv, comm_finished);
543 SIMIX_communication_destroy(comm_finished);
544 THROW0(host_error, 0, "Host failed");
545 } else if (SIMIX_action_get_state(comm_finished->act) == SURF_ACTION_FAILED){
546 SIMIX_communication_destroy(comm_finished);
547 THROW0(network_error, 0, "Link failure");
549 SIMIX_communication_destroy(comm_finished);