3 /* Copyright (c) 2009 Cristian Rosa.
4 All rights reserved. */
6 /* This program is free software; you can redistribute it and/or modify it
7 * under the terms of the license (GNU LGPL) which comes with this package. */
13 /* Pimple to get an histogram of message sizes in the simulation */
14 xbt_dict_t msg_sizes = NULL;
16 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
17 "Logging specific to SIMIX (network)");
19 /******************************************************************************/
20 /* Rendez-Vous Points */
21 /******************************************************************************/
24 * \brief Creates a new rendez-vous point
25 * \param name The name of the rendez-vous point
26 * \return The created rendez-vous point
28 smx_rdv_t SIMIX_rdv_create(const char *name)
30 smx_rdv_t rdv = xbt_new0(s_smx_rvpoint_t, 1);
31 rdv->name = name ? xbt_strdup(name) : NULL;
32 rdv->read = SIMIX_mutex_init();
33 rdv->write = SIMIX_mutex_init();
34 rdv->comm_fifo = xbt_fifo_new();
40 * \brief Destroy a rendez-vous point
41 * \param name The rendez-vous point to destroy
43 void SIMIX_rdv_destroy(smx_rdv_t rdv)
47 SIMIX_mutex_destroy(rdv->read);
48 SIMIX_mutex_destroy(rdv->write);
49 xbt_fifo_free(rdv->comm_fifo);
54 * \brief Push a communication request into a rendez-vous point
55 * \param rdv The rendez-vous point
56 * \param comm The communication request
58 static inline void SIMIX_rdv_push(smx_rdv_t rdv, smx_comm_t comm)
60 xbt_fifo_push(rdv->comm_fifo, comm);
65 * \brief Remove a communication request from a rendez-vous point
66 * \param rdv The rendez-vous point
67 * \param comm The communication request
69 static inline void SIMIX_rdv_remove(smx_rdv_t rdv, smx_comm_t comm)
71 xbt_fifo_remove(rdv->comm_fifo, comm);
76 * \brief Checks if there is a communication request queued in a rendez-vous matching our needs
77 * \param type The type of communication we are looking for (comm_send, comm_recv)
78 * \return The communication request if found, NULL otherwise.
80 smx_comm_t SIMIX_rdv_get_request(smx_rdv_t rdv, smx_comm_type_t type) {
81 smx_comm_t comm = (smx_comm_t)xbt_fifo_get_item_content(
82 xbt_fifo_get_first_item(rdv->comm_fifo));
84 if(comm && comm->type == type){
85 DEBUG0("Communication request found!");
86 xbt_fifo_shift(rdv->comm_fifo);
87 SIMIX_communication_use(comm);
92 DEBUG0("Communication request not found");
97 * \brief counts the number of communication requests of a given host pending
98 * on a rendez-vous point
99 * \param rdv The rendez-vous point
100 * \param host The host to be counted
101 * \return The number of comm request pending in the rdv
104 SIMIX_rdv_get_count_waiting_comm(smx_rdv_t rdv, smx_host_t host)
106 smx_comm_t comm = NULL;
107 xbt_fifo_item_t item = NULL;
110 xbt_fifo_foreach(rdv->comm_fifo, item, comm, smx_comm_t) {
111 if (comm->src_proc->smx_host == host)
119 * \brief returns the communication at the head of the rendez-vous
120 * \param rdv The rendez-vous point
121 * \return The communication or NULL if empty
123 XBT_INLINE smx_comm_t SIMIX_rdv_get_head(smx_rdv_t rdv)
125 return (smx_comm_t)xbt_fifo_get_item_content(xbt_fifo_get_first_item(rdv->comm_fifo));
128 /** @brief adds some API-related data to the rendez-vous point */
129 XBT_INLINE void SIMIX_rdv_set_data(smx_rdv_t rdv,void *data) {
132 /** @brief gets API-related data from the rendez-vous point */
133 XBT_INLINE void *SIMIX_rdv_get_data(smx_rdv_t rdv) {
137 /******************************************************************************/
138 /* Communication Requests */
139 /******************************************************************************/
142 * \brief Creates a new communication request
143 * \param type The type of communication (comm_send, comm_recv)
144 * \return The new communication request
146 smx_comm_t SIMIX_communication_new(smx_comm_type_t type)
148 /* alloc structures */
149 smx_comm_t comm = xbt_new0(s_smx_comm_t, 1);
151 comm->sem = SIMIX_sem_init(0);
158 * \brief Destroy a communication request
159 * \param comm The request to be destroyed
161 void SIMIX_communication_destroy(smx_comm_t comm)
164 if(comm->refcount > 0)
168 SIMIX_sem_destroy(comm->sem);
173 SIMIX_action_destroy(comm->act);
177 if(comm->src_timeout){
178 SIMIX_action_destroy(comm->src_timeout);
179 comm->src_timeout = NULL;
182 if(comm->dst_timeout){
183 SIMIX_action_destroy(comm->dst_timeout);
184 comm->dst_timeout = NULL;
191 * \brief Increase the number of users of the communication.
192 * \param comm The communication request
193 * Each communication request can be used by more than one process, so it is
194 * necessary to know number of them at destroy time, to avoid freeing stuff that
195 * maybe is in use by others.
198 static inline void SIMIX_communication_use(smx_comm_t comm)
204 * \brief Start the simulation of a communication request
205 * \param comm The comm->rdv = NULL;communication request
207 static inline void SIMIX_communication_start(smx_comm_t comm)
209 /* If both the sender and the receiver are already there, start the communication */
210 if(comm->src_proc && comm->dst_proc){
211 DEBUG1("Starting communication %p", comm);
212 comm->act = SIMIX_action_communicate(comm->src_proc->smx_host,
213 comm->dst_proc->smx_host, NULL,
214 comm->task_size, comm->rate);
216 TRACE_smx_action_communicate (comm->act, comm->src_proc);
219 /* If any of the process is suspend, create the action but stop its execution,
220 it will be restarted when the sender process resume */
221 if(SIMIX_process_is_suspended(comm->src_proc) ||
222 SIMIX_process_is_suspended(comm->dst_proc)) {
223 SIMIX_action_suspend(comm->act);
226 /* Add the communication as user data of the action */
227 comm->act->data = comm;
229 /* The semaphore will only get signaled once, but since the first unlocked guy will
230 * release_forever() the semaphore, that will unlock the second (and any other)
231 * communication partner */
232 SIMIX_register_action_to_semaphore(comm->act, comm->sem);
237 * \brief Waits for communication completion and performs error checking
238 * \param comm The communication
239 * \param timeout The max amount of time to wait for the communication to finish
242 * - host_error if peer failed
243 * - timeout_error if communication reached the timeout specified
244 * - network_error if network failed or peer issued a timeout
246 static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
248 smx_action_t act_sleep = NULL;
252 DEBUG1("Waiting for the completion of communication %p", comm);
255 act_sleep = SIMIX_action_sleep(SIMIX_host_self(), timeout);
256 if(SIMIX_process_self()==comm->src_proc)
257 comm->src_timeout = act_sleep;
259 comm->dst_timeout = act_sleep;
260 SIMIX_action_set_name(act_sleep,bprintf("Timeout for comm %p and wait on semaphore %p (max_duration:%f)", comm, comm->sem,timeout));
261 SIMIX_register_action_to_semaphore(act_sleep, comm->sem);
262 SIMIX_process_self()->waiting_action = act_sleep;
263 SIMIX_sem_block_onto(comm->sem);
264 SIMIX_process_self()->waiting_action = NULL;
265 SIMIX_unregister_action_to_semaphore(act_sleep, comm->sem);
267 SIMIX_sem_acquire(comm->sem);
270 /* Check for timeouts */
271 if ((src_timeout = ((comm->src_timeout) && (SIMIX_action_get_state(comm->src_timeout) == SURF_ACTION_DONE))) ||
272 (dst_timeout = ((comm->dst_timeout) && (SIMIX_action_get_state(comm->dst_timeout) == SURF_ACTION_DONE))) ) {
273 /* Somebody did a timeout! */
274 if (src_timeout) DEBUG1("Communication timeout from the src! %p", comm);
275 if (dst_timeout) DEBUG1("Communication timeout from the dst! %p", comm);
277 if(comm->act && SIMIX_action_get_state(comm->act) == SURF_ACTION_RUNNING)
278 SIMIX_communication_cancel(comm);
280 SIMIX_rdv_remove(comm->rdv, comm);
282 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
283 SIMIX_sem_release_forever(comm->sem);
284 SIMIX_communication_destroy(comm);
286 THROW1(timeout_error, 0, "Communication timeouted because of %s",src_timeout?"the source":"the destination");
289 DEBUG1("Communication %p complete! Let's check for errors", comm);
291 /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */
292 SIMIX_sem_release_forever(comm->sem);
294 /* Check for errors other than timeouts (they are catched above) */
295 if(!SIMIX_host_get_state(SIMIX_host_self())){
297 SIMIX_rdv_remove(comm->rdv, comm);
298 SIMIX_communication_destroy(comm);
299 THROW0(host_error, 0, "Host failed");
300 } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
301 SIMIX_communication_destroy(comm);
302 THROW0(network_error, 0, "Link failure");
304 SIMIX_communication_destroy(comm);
308 * \brief Cancels a communication
309 * \brief comm The communication to cancel
311 XBT_INLINE void SIMIX_communication_cancel(smx_comm_t comm)
314 SIMIX_action_cancel(comm->act);
318 * \brief get the amount remaining from the communication
319 * \param comm The communication
321 XBT_INLINE double SIMIX_communication_get_remains(smx_comm_t comm)
323 return SIMIX_action_get_remains(comm->act);
326 /******************************************************************************/
327 /* SIMIX_network_copy_data callbacks */
328 /******************************************************************************/
329 static void (*SIMIX_network_copy_data_callback)(smx_comm_t, size_t) = &SIMIX_network_copy_pointer_callback;
331 void SIMIX_network_set_copy_data_callback(void (*callback)(smx_comm_t, size_t)) {
332 SIMIX_network_copy_data_callback = callback;
335 void SIMIX_network_copy_pointer_callback(smx_comm_t comm, size_t buff_size) {
336 xbt_assert1((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)",buff_size);
337 *(void**)(comm->dst_buff) = comm->src_buff;
340 void SIMIX_network_copy_buffer_callback(smx_comm_t comm, size_t buff_size) {
341 memcpy(comm->dst_buff, comm->src_buff, buff_size);
345 * \brief Copy the communication data from the sender's buffer to the receiver's one
346 * \param comm The communication
348 void SIMIX_network_copy_data(smx_comm_t comm)
350 size_t buff_size = comm->src_buff_size;
352 DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
354 comm->src_proc->smx_host->name, comm->src_buff,
355 comm->dst_proc->smx_host->name, comm->dst_buff,
358 /* If there is no data to be copy then return */
359 if(!comm->src_buff || !comm->dst_buff)
362 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
363 if (comm->dst_buff_size)
364 buff_size = MIN(buff_size,*(comm->dst_buff_size));
366 /* Update the receiver's buffer size to the copied amount */
367 if (comm->dst_buff_size)
368 *comm->dst_buff_size = buff_size;
372 (*SIMIX_network_copy_data_callback)(comm, buff_size);
374 /* pimple to display the message sizes */
376 if (msg_sizes == NULL)
377 msg_sizes = xbt_dict_new();
378 uintptr_t casted_size = comm->task_size;
379 uintptr_t amount = xbt_dicti_get(msg_sizes, casted_size);
382 xbt_dicti_set(msg_sizes,casted_size, amount);
386 /* pimple to display the message sizes */
387 void SIMIX_message_sizes_output(const char *filename) {
388 FILE * out = fopen(filename,"w");
389 xbt_assert1(out,"Cannot open file %s",filename);
391 xbt_dict_cursor_t cursor;
392 xbt_dict_foreach(msg_sizes,cursor,key,data) {
393 fprintf(out,"%zu %zu\n",key,data);
399 * \brief Return the user data associated to the communication
400 * \param comm The communication
401 * \return the user data
403 XBT_INLINE void *SIMIX_communication_get_data(smx_comm_t comm)
408 /******************************************************************************/
409 /* Synchronous Communication */
410 /******************************************************************************/
412 * \brief Put a send communication request in a rendez-vous point and waits for
413 * its completion (blocking)
414 * \param rdv The rendez-vous point
415 * \param task_size The size of the communication action (for surf simulation)
416 * \param rate The rate of the communication action (for surf)
417 * \param timeout The timeout used for the waiting the completion
418 * \param src_buff The source buffer containing the message to be sent
419 * \param src_buff_size The size of the source buffer
420 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
421 * \param data User data associated to the communication object
423 * - host_error if peer failed
424 * - timeout_error if communication reached the timeout specified
425 * - network_error if network failed or peer issued a timeout
427 XBT_INLINE void SIMIX_network_send(smx_rdv_t rdv, double task_size, double rate,
428 double timeout, void *src_buff, size_t src_buff_size,
429 smx_comm_t *comm_ref, void *data)
431 *comm_ref = SIMIX_network_isend(rdv,task_size,rate,src_buff,src_buff_size,data);
432 SIMIX_network_wait(*comm_ref,timeout);
436 * \brief Put a receive communication request in a rendez-vous point and waits
437 * for its completion (blocking)
438 * \param rdv The rendez-vous point
439 * \param timeout The timeout used for the waiting the completion
440 * \param dst_buff The destination buffer to copy the received message
441 * \param src_buff_size The size of the destination buffer
442 * \param comm_ref The communication object used for the send (useful if someone else wants to cancel this communication afterward)
444 * - host_error if peer failed
445 * - timeout_error if communication reached the timeout specified
446 * - network_error if network failed or peer issued a timeout
448 XBT_INLINE void SIMIX_network_recv(smx_rdv_t rdv, double timeout, void *dst_buff,
449 size_t *dst_buff_size, smx_comm_t *comm_ref)
451 *comm_ref = SIMIX_network_irecv(rdv,dst_buff,dst_buff_size);
452 SIMIX_network_wait(*comm_ref,timeout);
455 /******************************************************************************/
456 /* Asynchronous Communication */
457 /******************************************************************************/
458 smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size, double rate,
459 void *src_buff, size_t src_buff_size, void *data)
463 /* Look for communication request matching our needs.
464 If it is not found then create it and push it into the rendez-vous point */
465 comm = SIMIX_rdv_get_request(rdv, comm_recv);
468 comm = SIMIX_communication_new(comm_send);
469 SIMIX_rdv_push(rdv, comm);
472 /* Setup the communication request */
473 comm->src_proc = SIMIX_process_self();
474 comm->task_size = task_size;
476 comm->src_buff = src_buff;
477 comm->src_buff_size = src_buff_size;
480 SIMIX_communication_start(comm);
484 smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_size) {
487 /* Look for communication request matching our needs.
488 If it is not found then create it and push it into the rendez-vous point */
489 comm = SIMIX_rdv_get_request(rdv, comm_send);
492 comm = SIMIX_communication_new(comm_recv);
493 SIMIX_rdv_push(rdv, comm);
496 /* Setup communication request */
497 comm->dst_proc = SIMIX_process_self();
498 comm->dst_buff = dst_buff;
499 comm->dst_buff_size = dst_buff_size;
501 SIMIX_communication_start(comm);
505 /** @brief blocks until the communication terminates or the timeout occurs */
506 XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout) {
507 /* Wait for communication completion */
508 SIMIX_communication_wait_for_completion(comm, timeout);
511 /** @Returns whether the (asynchronous) communication is done yet or not */
512 XBT_INLINE int SIMIX_network_test(smx_comm_t comm) {
513 return comm->sem?SIMIX_sem_would_block(comm->sem):0;
516 /** @brief wait for the completion of any communication of a set
518 * @Returns the rank in the dynar of communication which finished; destroy it after identifying which one it is
520 unsigned int SIMIX_network_waitany(xbt_dynar_t comms) {
521 xbt_dynar_t sems = xbt_dynar_new(sizeof(smx_sem_t),NULL);
522 unsigned int cursor, found_comm=-1;
523 smx_comm_t comm,comm_finished=NULL;
525 xbt_dynar_foreach(comms,cursor,comm){
526 xbt_dynar_push(sems,&(comm->sem));
529 DEBUG1("Waiting for the completion of communication set %p", comms);
531 found_comm = SIMIX_sem_acquire_any(sems);
532 xbt_assert0(found_comm!=-1,"Cannot find which communication finished");
533 xbt_dynar_get_cpy(comms,found_comm,&comm_finished);
535 DEBUG1("Communication %p complete! Let's check for errors", comm_finished);
537 /* Make sure that everyone sleeping on that semaphore is awake,
538 * and that nobody will ever block on it */
539 SIMIX_sem_release_forever(comm_finished->sem);
541 /* Check for errors */
542 if(!SIMIX_host_get_state(SIMIX_host_self())){
543 if(comm_finished->rdv)
544 SIMIX_rdv_remove(comm_finished->rdv, comm_finished);
545 SIMIX_communication_destroy(comm_finished);
546 THROW0(host_error, 0, "Host failed");
547 } else if (SIMIX_action_get_state(comm_finished->act) == SURF_ACTION_FAILED){
548 SIMIX_communication_destroy(comm_finished);
549 THROW0(network_error, 0, "Link failure");
551 SIMIX_communication_destroy(comm_finished);