X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/78c279dafeaf98e75a7db7ccb30c5179dd446353..401ba15e5e4da730ac3239424534c28fa51eda02:/src/simix/smx_network.cpp diff --git a/src/simix/smx_network.cpp b/src/simix/smx_network.cpp index bec9527d1d..7da114eba2 100644 --- a/src/simix/smx_network.cpp +++ b/src/simix/smx_network.cpp @@ -3,6 +3,8 @@ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ +#include + #include "src/surf/surf_interface.hpp" #include "src/simix/smx_private.h" #include "xbt/log.h" @@ -45,10 +47,10 @@ smx_mailbox_t SIMIX_mbox_create(const char *name) mbox->name = xbt_strdup(name); mbox->comm_queue = new std::deque(); mbox->done_comm_queue = nullptr; // Allocated on need only - mbox->permanent_receiver=NULL; + mbox->permanent_receiver=nullptr; XBT_DEBUG("Creating a mailbox at %p with name %s", mbox, name); - xbt_dict_set(mailboxes, mbox->name, mbox, NULL); + xbt_dict_set(mailboxes, mbox->name, mbox, nullptr); } return mbox; } @@ -84,7 +86,7 @@ void SIMIX_mbox_set_receiver(smx_mailbox_t mbox, smx_process_t process) /** * \brief Pushes a communication synchro into a rendez-vous point * \param mbox The mailbox - * \param comm The communication synchro + * \param synchro The communication synchro */ static inline void SIMIX_mbox_push(smx_mailbox_t mbox, smx_synchro_t synchro) { @@ -97,13 +99,13 @@ static inline void SIMIX_mbox_push(smx_mailbox_t mbox, smx_synchro_t synchro) /** * \brief Removes a communication synchro from a rendez-vous point * \param mbox The rendez-vous point - * \param comm The communication synchro + * \param synchro The communication synchro */ void SIMIX_mbox_remove(smx_mailbox_t mbox, smx_synchro_t synchro) { simgrid::simix::Comm *comm = static_cast(synchro); - comm->mbox = NULL; + comm->mbox = nullptr; for (auto it = mbox->comm_queue->begin(); it != mbox->comm_queue->end(); it++) if (*it == comm) { mbox->comm_queue->erase(it); @@ -115,12 +117,12 @@ void SIMIX_mbox_remove(smx_mailbox_t mbox, smx_synchro_t synchro) /** * \brief Checks if there is a communication synchro queued in a deque matching our needs * \param type The type of communication we are looking for (comm_send, comm_recv) - * \return The communication synchro if found, NULL otherwise + * \return The communication synchro if found, nullptr otherwise */ static smx_synchro_t _find_matching_comm(std::deque *deque, e_smx_comm_type_t type, int (*match_fun)(void *, void *,smx_synchro_t), void *this_user_data, smx_synchro_t my_synchro, bool remove_matching) { - void* other_user_data = NULL; + void* other_user_data = nullptr; for(auto it = deque->begin(); it != deque->end(); it++){ smx_synchro_t synchro = *it; @@ -141,7 +143,7 @@ static smx_synchro_t _find_matching_comm(std::deque *deque, e_smx #if HAVE_MC comm->mbox_cpy = comm->mbox; #endif - comm->mbox = NULL; + comm->mbox = nullptr; return comm; } XBT_DEBUG("Sorry, communication synchro %p does not match our needs:" @@ -149,25 +151,25 @@ static smx_synchro_t _find_matching_comm(std::deque *deque, e_smx comm, (int)comm->type, (int)type); } XBT_DEBUG("No matching communication synchro found"); - return NULL; + return nullptr; } /******************************************************************************/ /* Communication synchros */ /******************************************************************************/ -void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_process_t src, smx_mailbox_t mbox, +XBT_PRIVATE void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_process_t src, smx_mailbox_t mbox, double task_size, double rate, void *src_buff, size_t src_buff_size, int (*match_fun)(void *, void *,smx_synchro_t), void (*copy_data_fun)(smx_synchro_t, void*, size_t), void *data, double timeout){ smx_synchro_t comm = simcall_HANDLER_comm_isend(simcall, src, mbox, task_size, rate, - src_buff, src_buff_size, match_fun, NULL, copy_data_fun, + src_buff, src_buff_size, match_fun, nullptr, copy_data_fun, data, 0); SIMCALL_SET_MC_VALUE(simcall, 0); simcall_HANDLER_comm_wait(simcall, comm, timeout); } -smx_synchro_t simcall_HANDLER_comm_isend(smx_simcall_t simcall, smx_process_t src_proc, smx_mailbox_t mbox, +XBT_PRIVATE smx_synchro_t simcall_HANDLER_comm_isend(smx_simcall_t simcall, smx_process_t src_proc, smx_mailbox_t mbox, double task_size, double rate, void *src_buff, size_t src_buff_size, int (*match_fun)(void *, void *,smx_synchro_t), @@ -193,7 +195,7 @@ smx_synchro_t simcall_HANDLER_comm_isend(smx_simcall_t simcall, smx_process_t sr other_synchro = this_synchro; other_comm = static_cast(other_synchro); - if (mbox->permanent_receiver!=NULL){ + if (mbox->permanent_receiver!=nullptr){ //this mailbox is for small messages, which have to be sent right now other_synchro->state = SIMIX_READY; other_comm->dst_proc=mbox->permanent_receiver; @@ -215,14 +217,12 @@ smx_synchro_t simcall_HANDLER_comm_isend(smx_simcall_t simcall, smx_process_t sr } xbt_fifo_push(src_proc->comms, other_synchro); - /* if the communication synchro is detached then decrease the refcount - * by one, so it will be eliminated by the receiver's destroy call */ + if (detached) { other_comm->detached = true; - other_comm->ref(); other_comm->clean_fun = clean_fun; } else { - other_comm->clean_fun = NULL; + other_comm->clean_fun = nullptr; } /* Setup the communication synchro */ @@ -239,14 +239,14 @@ smx_synchro_t simcall_HANDLER_comm_isend(smx_simcall_t simcall, smx_process_t sr if (MC_is_active() || MC_record_replay_is_active()) { other_comm->state = SIMIX_RUNNING; - return (detached ? NULL : other_comm); + return (detached ? nullptr : other_comm); } SIMIX_comm_start(other_comm); - return (detached ? NULL : other_comm); + return (detached ? nullptr : other_comm); } -void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_process_t receiver, smx_mailbox_t mbox, +XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_process_t receiver, smx_mailbox_t mbox, void *dst_buff, size_t *dst_buff_size, int (*match_fun)(void *, void *, smx_synchro_t), void (*copy_data_fun)(smx_synchro_t, void*, size_t), @@ -257,7 +257,7 @@ void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_process_t receiver, sm simcall_HANDLER_comm_wait(simcall, comm, timeout); } -smx_synchro_t simcall_HANDLER_comm_irecv(smx_simcall_t simcall, smx_process_t receiver, smx_mailbox_t mbox, +XBT_PRIVATE smx_synchro_t simcall_HANDLER_comm_irecv(smx_simcall_t simcall, smx_process_t receiver, smx_mailbox_t mbox, void *dst_buff, size_t *dst_buff_size, int (*match_fun)(void *, void *, smx_synchro_t), void (*copy_data_fun)(smx_synchro_t, void*, size_t), @@ -293,7 +293,7 @@ smx_synchro_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_mailbox_t mbox, void XBT_DEBUG("comm %p has been already sent, and is finished, destroy it",other_comm); other_comm->state = SIMIX_DONE; other_comm->type = SIMIX_COMM_DONE; - other_comm->mbox = NULL; + other_comm->mbox = nullptr; } other_comm->unref(); static_cast(this_synchro)->unref(); @@ -363,7 +363,7 @@ smx_synchro_t SIMIX_comm_iprobe(smx_process_t dst_proc, smx_mailbox_t mbox, int this_comm = new simgrid::simix::Comm(SIMIX_COMM_RECEIVE); smx_type = SIMIX_COMM_SEND; } - smx_synchro_t other_synchro=NULL; + smx_synchro_t other_synchro=nullptr; if(mbox->permanent_receiver && ! mbox->done_comm_queue->empty()){ XBT_DEBUG("first check in the permanent recv mailbox, to see if we already got something"); other_synchro = @@ -386,7 +386,7 @@ void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_synchro_t synchro, dou /* Associate this simcall to the wait synchro */ XBT_DEBUG("simcall_HANDLER_comm_wait, %p", synchro); - xbt_fifo_push(synchro->simcalls, simcall); + synchro->simcalls.push_back(simcall); simcall->issuer->waiting_synchro = synchro; if (MC_is_active() || MC_record_replay_is_active()) { @@ -434,7 +434,7 @@ void simcall_HANDLER_comm_test(smx_simcall_t simcall, smx_synchro_t synchro) simcall_comm_test__set__result(simcall, comm->src_proc && comm->dst_proc); if (simcall_comm_test__get__result(simcall)){ synchro->state = SIMIX_DONE; - xbt_fifo_push(synchro->simcalls, simcall); + synchro->simcalls.push_back(simcall); SIMIX_comm_finish(synchro); } else { SIMIX_simcall_answer(simcall); @@ -444,7 +444,7 @@ void simcall_HANDLER_comm_test(smx_simcall_t simcall, smx_synchro_t synchro) simcall_comm_test__set__result(simcall, (synchro->state != SIMIX_WAITING && synchro->state != SIMIX_RUNNING)); if (simcall_comm_test__get__result(simcall)) { - xbt_fifo_push(synchro->simcalls, simcall); + synchro->simcalls.push_back(simcall); SIMIX_comm_finish(synchro); } else { SIMIX_simcall_answer(simcall); @@ -455,6 +455,8 @@ void simcall_HANDLER_comm_testany(smx_simcall_t simcall, xbt_dynar_t synchros) { unsigned int cursor; smx_synchro_t synchro; + // The default result is -1 -- this means, "nothing is ready". + // It can be changed below, but only if something matches. simcall_comm_testany__set__result(simcall, -1); if (MC_is_active() || MC_record_replay_is_active()){ @@ -464,7 +466,7 @@ void simcall_HANDLER_comm_testany(smx_simcall_t simcall, xbt_dynar_t synchros) }else{ synchro = xbt_dynar_get_as(synchros, idx, smx_synchro_t); simcall_comm_testany__set__result(simcall, idx); - xbt_fifo_push(synchro->simcalls, simcall); + synchro->simcalls.push_back(simcall); synchro->state = SIMIX_DONE; SIMIX_comm_finish(synchro); } @@ -474,7 +476,7 @@ void simcall_HANDLER_comm_testany(smx_simcall_t simcall, xbt_dynar_t synchros) xbt_dynar_foreach(simcall_comm_testany__get__comms(simcall), cursor,synchro) { if (synchro->state != SIMIX_WAITING && synchro->state != SIMIX_RUNNING) { simcall_comm_testany__set__result(simcall, cursor); - xbt_fifo_push(synchro->simcalls, simcall); + synchro->simcalls.push_back(simcall); SIMIX_comm_finish(synchro); return; } @@ -490,7 +492,7 @@ void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, xbt_dynar_t synchros) if (MC_is_active() || MC_record_replay_is_active()){ int idx = SIMCALL_GET_MC_VALUE(simcall); synchro = xbt_dynar_get_as(synchros, idx, smx_synchro_t); - xbt_fifo_push(synchro->simcalls, simcall); + synchro->simcalls.push_back(simcall); simcall_comm_waitany__set__result(simcall, idx); synchro->state = SIMIX_DONE; SIMIX_comm_finish(synchro); @@ -499,7 +501,7 @@ void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, xbt_dynar_t synchros) xbt_dynar_foreach(synchros, cursor, synchro){ /* associate this simcall to the the synchro */ - xbt_fifo_push(synchro->simcalls, simcall); + synchro->simcalls.push_back(simcall); /* see if the synchro is already finished */ if (synchro->state != SIMIX_WAITING && synchro->state != SIMIX_RUNNING){ @@ -515,8 +517,12 @@ void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall) unsigned int cursor = 0; xbt_dynar_t synchros = simcall_comm_waitany__get__comms(simcall); - xbt_dynar_foreach(synchros, cursor, synchro) - xbt_fifo_remove(synchro->simcalls, simcall); + xbt_dynar_foreach(synchros, cursor, synchro) { + // Remove the first occurence of simcall: + auto i = boost::range::find(synchro->simcalls, simcall); + if (i != synchro->simcalls.end()) + synchro->simcalls.erase(i); + } } /** @@ -552,10 +558,10 @@ static inline void SIMIX_comm_start(smx_synchro_t synchro) if (SIMIX_process_is_suspended(comm->src_proc) || SIMIX_process_is_suspended(comm->dst_proc)) { if (SIMIX_process_is_suspended(comm->src_proc)) XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the communication", - comm->src_proc->name, sg_host_get_name(comm->src_proc->host)); + comm->src_proc->name.c_str(), sg_host_get_name(comm->src_proc->host)); else XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the communication", - comm->dst_proc->name, sg_host_get_name(comm->dst_proc->host)); + comm->dst_proc->name.c_str(), sg_host_get_name(comm->dst_proc->host)); comm->surf_comm->suspend(); } @@ -570,9 +576,10 @@ void SIMIX_comm_finish(smx_synchro_t synchro) { simgrid::simix::Comm *comm = static_cast(synchro); unsigned int destroy_count = 0; - smx_simcall_t simcall; - while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) { + while (!synchro->simcalls.empty()) { + smx_simcall_t simcall = synchro->simcalls.front(); + synchro->simcalls.pop_front(); /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany list. Afterwards, get the @@ -598,74 +605,84 @@ void SIMIX_comm_finish(smx_synchro_t synchro) if (simcall->issuer->host->isOff()) { simcall->issuer->context->iwannadie = 1; SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed"); - } else - - switch (synchro->state) { - - case SIMIX_DONE: - XBT_DEBUG("Communication %p complete!", synchro); - SIMIX_comm_copy_data(synchro); - break; - - case SIMIX_SRC_TIMEOUT: - SMX_EXCEPTION(simcall->issuer, timeout_error, 0, "Communication timeouted because of sender"); - break; - - case SIMIX_DST_TIMEOUT: - SMX_EXCEPTION(simcall->issuer, timeout_error, 0, "Communication timeouted because of receiver"); - break; - - case SIMIX_SRC_HOST_FAILURE: - if (simcall->issuer == comm->src_proc) - simcall->issuer->context->iwannadie = 1; -// SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed"); - else - SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed"); - break; - - case SIMIX_DST_HOST_FAILURE: - if (simcall->issuer == comm->dst_proc) - simcall->issuer->context->iwannadie = 1; -// SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed"); - else - SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed"); - break; - - case SIMIX_LINK_FAILURE: - - XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) detached:%d", - synchro, - comm->src_proc ? sg_host_get_name(comm->src_proc->host) : NULL, - comm->dst_proc ? sg_host_get_name(comm->dst_proc->host) : NULL, - simcall->issuer->name, simcall->issuer, comm->detached); - if (comm->src_proc == simcall->issuer) { - XBT_DEBUG("I'm source"); - } else if (comm->dst_proc == simcall->issuer) { - XBT_DEBUG("I'm dest"); - } else { - XBT_DEBUG("I'm neither source nor dest"); + } else { + switch (synchro->state) { + + case SIMIX_DONE: + XBT_DEBUG("Communication %p complete!", synchro); + SIMIX_comm_copy_data(synchro); + break; + + case SIMIX_SRC_TIMEOUT: + SMX_EXCEPTION(simcall->issuer, timeout_error, 0, "Communication timeouted because of sender"); + break; + + case SIMIX_DST_TIMEOUT: + SMX_EXCEPTION(simcall->issuer, timeout_error, 0, "Communication timeouted because of receiver"); + break; + + case SIMIX_SRC_HOST_FAILURE: + if (simcall->issuer == comm->src_proc) + simcall->issuer->context->iwannadie = 1; + // SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed"); + else + SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed"); + break; + + case SIMIX_DST_HOST_FAILURE: + if (simcall->issuer == comm->dst_proc) + simcall->issuer->context->iwannadie = 1; + // SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed"); + else + SMX_EXCEPTION(simcall->issuer, network_error, 0, "Remote peer failed"); + break; + + case SIMIX_LINK_FAILURE: + + XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) detached:%d", + synchro, + comm->src_proc ? sg_host_get_name(comm->src_proc->host) : nullptr, + comm->dst_proc ? sg_host_get_name(comm->dst_proc->host) : nullptr, + simcall->issuer->name.c_str(), simcall->issuer, comm->detached); + if (comm->src_proc == simcall->issuer) { + XBT_DEBUG("I'm source"); + } else if (comm->dst_proc == simcall->issuer) { + XBT_DEBUG("I'm dest"); + } else { + XBT_DEBUG("I'm neither source nor dest"); + } + SMX_EXCEPTION(simcall->issuer, network_error, 0, "Link failure"); + break; + + case SIMIX_CANCELED: + if (simcall->issuer == comm->dst_proc) + SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Communication canceled by the sender"); + else + SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Communication canceled by the receiver"); + break; + + default: + xbt_die("Unexpected synchro state in SIMIX_comm_finish: %d", (int)synchro->state); } - SMX_EXCEPTION(simcall->issuer, network_error, 0, "Link failure"); - break; - - case SIMIX_CANCELED: - if (simcall->issuer == comm->dst_proc) - SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Communication canceled by the sender"); - else - SMX_EXCEPTION(simcall->issuer, cancel_error, 0, "Communication canceled by the receiver"); - break; - - default: - xbt_die("Unexpected synchro state in SIMIX_comm_finish: %d", (int)synchro->state); } /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */ - if (simcall->issuer->doexception) { - if (simcall->call == SIMCALL_COMM_WAITANY) { - simcall->issuer->running_ctx->exception.value = xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro); + if (simcall->issuer->exception) { + // In order to modify the exception we have to rethrow it: + try { + std::rethrow_exception(simcall->issuer->exception); } - else if (simcall->call == SIMCALL_COMM_TESTANY) { - simcall->issuer->running_ctx->exception.value = xbt_dynar_search(simcall_comm_testany__get__comms(simcall), &synchro); + catch(xbt_ex& e) { + if (simcall->call == SIMCALL_COMM_WAITANY) { + e.value = xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro); + } + else if (simcall->call == SIMCALL_COMM_TESTANY) { + e.value = xbt_dynar_search(simcall_comm_testany__get__comms(simcall), &synchro); + } + simcall->issuer->exception = std::make_exception_ptr(e); + } + catch(...) { + // Nothing to do } } @@ -673,7 +690,7 @@ void SIMIX_comm_finish(smx_synchro_t synchro) simcall->issuer->context->iwannadie = 1; } - simcall->issuer->waiting_synchro = NULL; + simcall->issuer->waiting_synchro = nullptr; xbt_fifo_remove(simcall->issuer->comms, synchro); if(comm->detached){ if(simcall->issuer == comm->src_proc){ @@ -683,6 +700,8 @@ void SIMIX_comm_finish(smx_synchro_t synchro) if(simcall->issuer == comm->dst_proc){ if(comm->src_proc) xbt_fifo_remove(comm->src_proc->comms, synchro); + //in case of a detached comm we have an extra ref to remove, as the sender won't do it + destroy_count++; } } SIMIX_simcall_answer(simcall); @@ -693,44 +712,6 @@ void SIMIX_comm_finish(smx_synchro_t synchro) static_cast(synchro)->unref(); } -/** - * \brief This function is called when a Surf communication synchro is finished. - * \param synchro the corresponding Simix communication - */ -void SIMIX_post_comm(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - - /* Update synchro state */ - if (comm->src_timeout && - comm->src_timeout->getState() == simgrid::surf::Action::State::done) - synchro->state = SIMIX_SRC_TIMEOUT; - else if (comm->dst_timeout && - comm->dst_timeout->getState() == simgrid::surf::Action::State::done) - synchro->state = SIMIX_DST_TIMEOUT; - else if (comm->src_timeout && - comm->src_timeout->getState() == simgrid::surf::Action::State::failed) - synchro->state = SIMIX_SRC_HOST_FAILURE; - else if (comm->dst_timeout && - comm->dst_timeout->getState() == simgrid::surf::Action::State::failed) - synchro->state = SIMIX_DST_HOST_FAILURE; - else if (comm->surf_comm && - comm->surf_comm->getState() == simgrid::surf::Action::State::failed) { - synchro->state = SIMIX_LINK_FAILURE; - } else - synchro->state = SIMIX_DONE; - - XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d", - comm, (int)comm->state, comm->src_proc, comm->dst_proc, comm->detached); - - /* destroy the surf actions associated with the Simix communication */ - comm->cleanupSurf(); - - /* if there are simcalls associated with the synchro, then answer them */ - if (xbt_fifo_size(synchro->simcalls)) - SIMIX_comm_finish(comm); -} - /******************************************************************************/ /* SIMIX_comm_copy_data callbacks */ /******************************************************************************/ @@ -757,7 +738,7 @@ void SIMIX_comm_copy_buffer_callback(smx_synchro_t synchro, void* buff, size_t b memcpy(comm->dst_buff, buff, buff_size); if (comm->detached) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the original buffer available to the application ASAP xbt_free(buff); - comm->src_buff = NULL; + comm->src_buff = nullptr; } }