X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/d672a6447aca931be2b670c5c5e177013513c7ee..22356ab5dbecbe29e4f06dda4d3000f9cff68414:/src/simix/smx_network.c diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c index f874dcc1f5..f2b2258087 100644 --- a/src/simix/smx_network.c +++ b/src/simix/smx_network.c @@ -13,18 +13,15 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix, "Logging specific to SIMIX (network)"); static xbt_dict_t rdv_points = NULL; +unsigned long int smx_total_comms = 0; -static XBT_INLINE void SIMIX_comm_start(smx_action_t action); -static void SIMIX_comm_finish(smx_action_t action); 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_comm_wait_for_completion(smx_action_t comm, - double timeout); 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 *); + int (*match_fun)(void *, void *), void *); static void SIMIX_rdv_free(void *data); void SIMIX_network_init(void) @@ -51,8 +48,8 @@ smx_rdv_t SIMIX_rdv_create(const char *name) rdv->name = name ? xbt_strdup(name) : NULL; rdv->comm_fifo = xbt_fifo_new(); - if (name) - xbt_dict_set(rdv_points, name, rdv, SIMIX_rdv_free); + if (rdv->name) + xbt_dict_set(rdv_points, rdv->name, rdv, SIMIX_rdv_free); } return rdv; } @@ -60,7 +57,7 @@ smx_rdv_t SIMIX_rdv_create(const char *name) void SIMIX_rdv_destroy(smx_rdv_t rdv) { if (rdv->name) - xbt_dict_remove(rdv_points, rdv->name); + xbt_dict_remove(rdv_points, rdv->name); } void SIMIX_rdv_free(void *data) @@ -119,29 +116,89 @@ static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_action_t comm) } /** - * \brief Checks if there is a communication request queued in a rendez-vous matching our needs + * \brief Wrapper to SIMIX_rdv_get_request + */ +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); +} + +/** + * \brief Checks if there is a communication action queued in a rendez-vous matching our needs * \param type The type of communication we are looking for (comm_send, comm_recv) - * \return The communication request if found, NULL otherwise + * \return The communication action 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) + int (*match_fun)(void *, void *), void *data) { - smx_action_t req; + // FIXME rewrite this function by using SIMIX_rdv_has_send/recv_match + smx_action_t action; xbt_fifo_item_t item; + void* req_data = NULL; - xbt_fifo_foreach(rdv->comm_fifo, item, req, smx_action_t){ - if(req->comm.type == type && (!match_fun || match_fun(data, req->comm.data))){ - xbt_fifo_remove_item(rdv->comm_fifo, item); - req->comm.refcount++; - req->comm.rdv = NULL; - return req; - } + 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; + } + 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); + xbt_fifo_free_item(item); + action->comm.refcount++; + action->comm.rdv = NULL; + return action; + } + 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); } - - DEBUG0("Communication request not found"); + XBT_DEBUG("No matching communication action 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 + */ +int SIMIX_comm_has_send_match(smx_rdv_t rdv, int (*match_fun)(void*, void*), void* data) { + + 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_SEND + && (!match_fun || match_fun(data, action->comm.src_data))) { + XBT_DEBUG("Found a matching communication action %p", action); + return 1; + } + } + 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) { + + 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; + } + } + XBT_DEBUG("No matching communication action found"); + return 0; +} + /******************************************************************************/ /* Comunication Actions */ /******************************************************************************/ @@ -156,20 +213,25 @@ smx_action_t SIMIX_comm_new(e_smx_comm_type_t type) smx_action_t act; /* alloc structures */ - act = xbt_new0(s_smx_action_t, 1); + act = xbt_mallocator_get(simix_global->action_mallocator); act->type = SIMIX_ACTION_COMMUNICATE; act->state = SIMIX_WAITING; - act->request_list = xbt_fifo_new(); /* set communication */ act->comm.type = type; act->comm.refcount = 1; +#ifdef HAVE_LATENCY_BOUND_TRACKING + //initialize with unknown value + act->latency_limited = -1; +#endif + #ifdef HAVE_TRACING act->category = NULL; #endif - DEBUG1("Create communicate action %p", act); + XBT_DEBUG("Create communicate action %p", act); + ++smx_total_comms; return act; } @@ -180,44 +242,44 @@ smx_action_t SIMIX_comm_new(e_smx_comm_type_t type) */ void SIMIX_comm_destroy(smx_action_t action) { - DEBUG1("Destroy action %p", action); + XBT_DEBUG("Destroy action %p (refcount:%d)", action, action->comm.refcount); if (action->comm.refcount <= 0) - xbt_die(bprintf("the refcount of comm %p is already 0 before decreasing it. That's a bug!",action)); - -#ifdef HAVE_LATENCY_BOUND_TRACKING - //save is latency limited flag to use afterwards - if (action->comm.surf_comm) { - DEBUG2("adding key %p with latency limited value %d to the dict", action, - SIMIX_comm_is_latency_bounded(action)); - xbt_dicti_set(simix_global->latency_limited_dict, (uintptr_t) action, - SIMIX_comm_is_latency_bounded(action)); - } -#endif + 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; - VERB2("Really free communication %p; refcount is now %d", action, + XBT_DEBUG("Really free communication %p; refcount is now %d", action, action->comm.refcount); +#ifdef HAVE_LATENCY_BOUND_TRACKING + action->latency_limited = SIMIX_comm_is_latency_bounded( action ) ; +#endif + #ifdef HAVE_TRACING TRACE_smx_action_destroy(action); #endif - if (action->name) - xbt_free(action->name); - - xbt_fifo_free(action->request_list); - + xbt_free(action->name); SIMIX_comm_destroy_internal_actions(action); - xbt_free(action); + if (action->comm.detached && action->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); + } + + xbt_mallocator_release(simix_global->action_mallocator, action); } void SIMIX_comm_destroy_internal_actions(smx_action_t action) { if (action->comm.surf_comm){ +#ifdef HAVE_LATENCY_BOUND_TRACKING + action->latency_limited = SIMIX_comm_is_latency_bounded(action); +#endif action->comm.surf_comm->model_type->action_unref(action->comm.surf_comm); action->comm.surf_comm = NULL; } @@ -236,7 +298,8 @@ void SIMIX_comm_destroy_internal_actions(smx_action_t action) smx_action_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 *), void *data, + int detached) { smx_action_t action; @@ -252,13 +315,20 @@ smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv, action->comm.type = SIMIX_COMM_READY; } + /* If the communication action is detached then decrease the refcount + * by one, so it will be eliminated by the receivers destroy call */ + if (detached) { + action->comm.detached = 1; + action->comm.refcount--; + } + /* 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.data = data; + action->comm.src_data = data; if (MC_IS_ENABLED) { action->state = SIMIX_RUNNING; @@ -292,6 +362,7 @@ smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv, 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; if (MC_IS_ENABLED) { action->state = SIMIX_RUNNING; @@ -302,19 +373,32 @@ smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv, return action; } -void SIMIX_pre_comm_wait(smx_req_t req) +void SIMIX_pre_comm_wait(smx_req_t req, smx_action_t action, double timeout, int idx) { - smx_action_t action = req->comm_wait.comm; - double timeout = req->comm_wait.timeout; + /* the request 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; - if (MC_IS_ENABLED){ - action->state = SIMIX_DONE; + if (MC_IS_ENABLED) { + if (idx == 0) { + action->state = SIMIX_DONE; + } else { + /* If we reached this point, the wait request 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; + else + action->state = SIMIX_DST_TIMEOUT; + } + SIMIX_comm_finish(action); + return; } /* If the action has already finish perform the error handling, */ @@ -335,38 +419,74 @@ void SIMIX_pre_comm_wait(smx_req_t req) void SIMIX_pre_comm_test(smx_req_t req) { smx_action_t action = req->comm_test.comm; - req->comm_test.result = (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING); + 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); + }else{ + SIMIX_request_answer(req); + } + 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); - } - else { + } else { SIMIX_request_answer(req); } } -void SIMIX_pre_comm_testany(smx_req_t req) +void SIMIX_pre_comm_testany(smx_req_t req, int idx) { unsigned int cursor; smx_action_t action; + xbt_dynar_t actions = req->comm_testany.comms; req->comm_testany.result = -1; + + if (MC_IS_ENABLED){ + if(idx == -1){ + SIMIX_request_answer(req); + }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); + } + 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); - break; + return; } } SIMIX_request_answer(req); } -void SIMIX_pre_comm_waitany(smx_req_t req) +void SIMIX_pre_comm_waitany(smx_req_t req, int idx) { smx_action_t action; 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); + return; + } + xbt_dynar_foreach(actions, cursor, action){ /* Associate this request to the action */ xbt_fifo_push(action->request_list, req); @@ -392,14 +512,15 @@ void SIMIX_waitany_req_remove_from_actions(smx_req_t req) * \brief Start the simulation of a communication request * \param action The communication action */ -static XBT_INLINE void SIMIX_comm_start(smx_action_t action) + +XBT_INLINE void SIMIX_comm_start(smx_action_t action) { /* 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; - DEBUG3("Starting communication %p from '%s' to '%s'", action, + XBT_DEBUG("Starting communication %p from '%s' to '%s'", action, SIMIX_host_get_name(sender), SIMIX_host_get_name(receiver)); action->comm.surf_comm = surf_workstation_model->extension.workstation. @@ -415,7 +536,7 @@ static XBT_INLINE void SIMIX_comm_start(smx_action_t action) /* If a link is failed, detect it immediately */ if (surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) { - DEBUG2("Communication from '%s' to '%s' failed to start because of a link failure", + 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); @@ -433,6 +554,7 @@ static XBT_INLINE void SIMIX_comm_start(smx_action_t action) void SIMIX_comm_finish(smx_action_t action) { + unsigned int destroy_count = 0; smx_req_t req; while ((req = xbt_fifo_shift(action->request_list))) { @@ -443,20 +565,21 @@ void SIMIX_comm_finish(smx_action_t action) return it as the result of the call */ if (req->call == REQ_COMM_WAITANY) { SIMIX_waitany_req_remove_from_actions(req); - req->comm_waitany.result = xbt_dynar_search(req->comm_waitany.comms, &action); + if (!MC_IS_ENABLED) + req->comm_waitany.result = xbt_dynar_search(req->comm_waitany.comms, &action); } /* 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); - DEBUG1("SIMIX_comm_finish: action state = %d", action->state); + XBT_DEBUG("SIMIX_comm_finish: action state = %d", action->state); /* Check out for errors */ switch (action->state) { case SIMIX_DONE: - DEBUG1("Communication %p complete!", action); + XBT_DEBUG("Communication %p complete!", action); SIMIX_comm_copy_data(action); break; @@ -504,7 +627,7 @@ void SIMIX_comm_finish(smx_action_t action) case SIMIX_LINK_FAILURE: TRY { - DEBUG5("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)", + 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); THROW0(network_error, 0, "Link failure"); @@ -517,9 +640,24 @@ void SIMIX_comm_finish(smx_action_t action) default: THROW_IMPOSSIBLE; } + + /* 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); + } + else if (req->call == REQ_COMM_TESTANY) { + req->issuer->running_ctx->exception.value = xbt_dynar_search(req->comm_testany.comms, &action); + } + } + req->issuer->waiting_action = NULL; SIMIX_request_answer(req); + destroy_count++; } + + while (destroy_count-- > 0) + SIMIX_comm_destroy(action); } void SIMIX_post_comm(smx_action_t action) @@ -543,7 +681,7 @@ void SIMIX_post_comm(smx_action_t action) else action->state = SIMIX_DONE; - DEBUG1("SIMIX_post_comm: action state = %d", action->state); + 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. */ @@ -562,7 +700,9 @@ void SIMIX_comm_cancel(smx_action_t action) SIMIX_rdv_remove(action->comm.rdv, action); action->state = SIMIX_FAILED; } else { - surf_workstation_model->action_cancel(action->comm.surf_comm); + /* When running the MC there are no surf actions */ + if(!MC_IS_ENABLED) + surf_workstation_model->action_cancel(action->comm.surf_comm); } } @@ -613,40 +753,23 @@ e_smx_state_t SIMIX_comm_get_state(smx_action_t action) } /** - * \brief Return the user data associated to the communication + * \brief Return the user data associated to the sender of the communication * \param action The communication * \return the user data */ -void* SIMIX_comm_get_data(smx_action_t action) -{ - return action->comm.data; -} - -void* SIMIX_comm_get_src_buff(smx_action_t action) -{ - return action->comm.src_buff; -} - -void* SIMIX_comm_get_dst_buff(smx_action_t action) +void* SIMIX_comm_get_src_data(smx_action_t action) { - return action->comm.dst_buff; + return action->comm.src_data; } -size_t SIMIX_comm_get_src_buff_size(smx_action_t action) +/** + * \brief Return the user data associated to the receiver of the communication + * \param action The communication + * \return the user data + */ +void* SIMIX_comm_get_dst_data(smx_action_t action) { - return action->comm.src_buff_size; -} - -size_t SIMIX_comm_get_dst_buff_size(smx_action_t action) -{ - size_t buff_size; - - if (action->comm.dst_buff_size) - buff_size = *(action->comm.dst_buff_size); - else - buff_size = 0; - - return buff_size; + return action->comm.dst_data; } smx_process_t SIMIX_comm_get_src_proc(smx_action_t action) @@ -666,21 +789,12 @@ smx_process_t SIMIX_comm_get_dst_proc(smx_action_t action) */ XBT_INLINE int SIMIX_comm_is_latency_bounded(smx_action_t action) { - //try to find comm on the list of finished flows - uintptr_t key = 0; - uintptr_t data = 0; - xbt_dict_cursor_t cursor; - xbt_dict_foreach(simix_global->latency_limited_dict, cursor, key, data) { - DEBUG2("comparing key=%p with comm=%p", (void *) key, (void *) action); - if ((void *) action == (void *) key) { - DEBUG2("key %p found, return value latency limited value %d", - (void *) key, (int) data); - xbt_dict_cursor_free(&cursor); - return (int) data; - } + 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); } - - return surf_workstation_model->get_latency_limited(action->comm.surf_comm); + return action->latency_limited; } #endif @@ -719,7 +833,7 @@ void SIMIX_comm_copy_data(smx_action_t comm) if (!comm->comm.src_buff || !comm->comm.dst_buff || comm->comm.copied == 1) return; - DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", + 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);