X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/5d67901dba3dfd8e75f708c329b8144287490077..566c00275e9f0f06561e882c7c304f96e3f968a7:/src/simix/smx_network.cpp diff --git a/src/simix/smx_network.cpp b/src/simix/smx_network.cpp index 8d808d7136..722116e54c 100644 --- a/src/simix/smx_network.cpp +++ b/src/simix/smx_network.cpp @@ -3,6 +3,14 @@ /* 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 + +#include + +#include + #include "src/surf/surf_interface.hpp" #include "src/simix/smx_private.h" #include "xbt/log.h" @@ -11,7 +19,7 @@ #include "xbt/dict.h" #include "simgrid/s4u/mailbox.hpp" -#include "src/simix/SynchroComm.hpp" +#include "src/kernel/activity/SynchroComm.hpp" XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix, "SIMIX network-related synchronization"); @@ -20,7 +28,6 @@ static xbt_dict_t mailboxes = xbt_dict_new_homogeneous(SIMIX_mbox_free); static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall); static void SIMIX_comm_copy_data(smx_synchro_t comm); -static smx_synchro_t SIMIX_comm_new(e_smx_comm_type_t type); static inline void SIMIX_mbox_push(smx_mailbox_t mbox, smx_synchro_t comm); 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 *user_data, smx_synchro_t my_synchro, bool remove_matching); @@ -40,16 +47,10 @@ smx_mailbox_t SIMIX_mbox_create(const char *name) xbt_assert(name, "Mailboxes must have a name"); /* two processes may have pushed the same mbox_create simcall at the same time */ smx_mailbox_t mbox = (smx_mailbox_t) xbt_dict_get_or_null(mailboxes, name); - if (!mbox) { - mbox = xbt_new0(s_smx_mailbox_t, 1); - 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 = new simgrid::simix::Mailbox(name); 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; } @@ -58,11 +59,7 @@ void SIMIX_mbox_free(void *data) { XBT_DEBUG("mbox free %p", data); smx_mailbox_t mbox = (smx_mailbox_t) data; - xbt_free(mbox->name); - delete mbox->comm_queue; - delete mbox->done_comm_queue; - - xbt_free(mbox); + delete mbox; } smx_mailbox_t SIMIX_mbox_get_by_name(const char *name) @@ -70,21 +67,6 @@ smx_mailbox_t SIMIX_mbox_get_by_name(const char *name) return (smx_mailbox_t) xbt_dict_get_or_null(mailboxes, name); } -smx_synchro_t SIMIX_mbox_get_head(smx_mailbox_t mbox) -{ - return mbox->comm_queue->empty()? nullptr:mbox->comm_queue->front(); -} - -/** - * \brief get the receiver (process associated to the mailbox) - * \param mbox The rendez-vous point - * \return process The receiving process (NULL if not set) - */ -smx_process_t SIMIX_mbox_get_receiver(smx_mailbox_t mbox) -{ - return mbox->permanent_receiver; -} - /** * \brief set the receiver of the rendez vous point to allow eager sends * \param mbox The rendez-vous point @@ -92,55 +74,52 @@ smx_process_t SIMIX_mbox_get_receiver(smx_mailbox_t mbox) */ void SIMIX_mbox_set_receiver(smx_mailbox_t mbox, smx_process_t process) { - mbox->permanent_receiver=process; - if (mbox->done_comm_queue == nullptr) - mbox->done_comm_queue = new std::deque(); + mbox->permanent_receiver = 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) { - simgrid::simix::Comm *comm = static_cast(synchro); - - mbox->comm_queue->push_back(comm); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); + mbox->comm_queue.push_back(comm); comm->mbox = mbox; } /** * \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); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); - comm->mbox = NULL; - for (auto it = mbox->comm_queue->begin(); it != mbox->comm_queue->end(); it++) + comm->mbox = nullptr; + for (auto it = mbox->comm_queue.begin(); it != mbox->comm_queue.end(); it++) if (*it == comm) { - mbox->comm_queue->erase(it); + mbox->comm_queue. erase(it); return; } - xbt_die("Cannot remove this comm that is not part of the mailbox"); + xbt_die("Cannot remove the comm %p that is not part of the mailbox %s",comm, mbox->name); } /** * \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; - simgrid::simix::Comm *comm = static_cast(synchro); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); if (comm->type == SIMIX_COMM_SEND) { other_user_data = comm->src_data; @@ -153,11 +132,11 @@ static smx_synchro_t _find_matching_comm(std::deque *deque, e_smx XBT_DEBUG("Found a matching communication synchro %p", comm); if (remove_matching) deque->erase(it); - comm->refcount++; + comm->ref(); #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:" @@ -165,101 +144,25 @@ 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 */ /******************************************************************************/ - -/** - * \brief Creates a new communicate synchro - * \param type The direction of communication (comm_send, comm_recv) - * \return The new communicate synchro - */ -smx_synchro_t SIMIX_comm_new(e_smx_comm_type_t type) -{ - simgrid::simix::Comm *comm = new simgrid::simix::Comm(); - comm->state = SIMIX_WAITING; - comm->type = type; - comm->refcount = 1; - comm->src_data=NULL; - comm->dst_data=NULL; - - XBT_DEBUG("Create communicate synchro %p", comm); - - return comm; -} - -/** - * \brief Destroy a communicate synchro - * \param synchro The communicate synchro to be destroyed - */ -void SIMIX_comm_destroy(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - - XBT_DEBUG("Destroy synchro %p (refcount: %d), state: %d", comm, comm->refcount, (int)comm->state); - - if (comm->refcount <= 0) { - xbt_backtrace_display_current(); - xbt_die("The refcount of comm %p is already 0 before decreasing it. " - "That's a bug! If you didn't test and/or wait the same communication twice in your code, then the bug is SimGrid's...", synchro); - } - comm->refcount--; - if (comm->refcount > 0) - return; - XBT_DEBUG("Really free communication %p; refcount is now %d", comm, comm->refcount); - - SIMIX_comm_destroy_internal_actions(synchro); - - if (comm->detached && comm->state != SIMIX_DONE) { - /* the communication has failed and was detached: - * we have to free the buffer */ - if (comm->clean_fun) { - comm->clean_fun(comm->src_buff); - } - comm->src_buff = NULL; - } - - if(comm->mbox) - SIMIX_mbox_remove(comm->mbox, comm); - - delete comm; -} - -void SIMIX_comm_destroy_internal_actions(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - if (comm->surf_comm){ - comm->surf_comm->unref(); - comm->surf_comm = NULL; - } - - if (comm->src_timeout){ - comm->src_timeout->unref(); - comm->src_timeout = NULL; - } - - if (comm->dst_timeout){ - comm->dst_timeout->unref(); - comm->dst_timeout = NULL; - } -} - -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), @@ -270,28 +173,27 @@ smx_synchro_t simcall_HANDLER_comm_isend(smx_simcall_t simcall, smx_process_t sr XBT_DEBUG("send from %p", mbox); /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */ - smx_synchro_t this_synchro = SIMIX_comm_new(SIMIX_COMM_SEND); + simgrid::kernel::activity::Comm* this_synchro = new simgrid::kernel::activity::Comm(SIMIX_COMM_SEND); /* Look for communication synchro matching our needs. We also provide a description of * ourself so that the other side also gets a chance of choosing if it wants to match with us. * * If it is not found then push our communication into the rendez-vous point */ smx_synchro_t other_synchro = - _find_matching_comm(mbox->comm_queue, SIMIX_COMM_RECEIVE, match_fun, data, this_synchro, /*remove_matching*/true); - simgrid::simix::Comm *other_comm = static_cast(other_synchro); + _find_matching_comm(&mbox->comm_queue, SIMIX_COMM_RECEIVE, match_fun, data, this_synchro, /*remove_matching*/true); + simgrid::kernel::activity::Comm *other_comm = static_cast(other_synchro); if (!other_synchro) { other_synchro = this_synchro; - other_comm = static_cast(other_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; - other_comm->refcount++; - mbox->done_comm_queue->push_back(other_synchro); - other_comm->mbox=mbox; + other_comm->dst_proc=mbox->permanent_receiver.get(); + other_comm->ref(); + mbox->done_comm_queue.push_back(other_synchro); XBT_DEBUG("pushing a message into the permanent receive fifo %p, comm %p", mbox, &(other_comm)); }else{ @@ -299,8 +201,7 @@ smx_synchro_t simcall_HANDLER_comm_isend(smx_simcall_t simcall, smx_process_t sr } } else { XBT_DEBUG("Receive already pushed"); - - SIMIX_comm_destroy(this_synchro); + this_synchro->unref(); other_comm->state = SIMIX_READY; other_comm->type = SIMIX_COMM_READY; @@ -308,14 +209,12 @@ 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 = 1; - other_comm->refcount--; + other_comm->detached = true; other_comm->clean_fun = clean_fun; } else { - other_comm->clean_fun = NULL; + other_comm->clean_fun = nullptr; } /* Setup the communication synchro */ @@ -332,14 +231,14 @@ 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), @@ -350,7 +249,7 @@ 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), @@ -364,32 +263,32 @@ smx_synchro_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_mailbox_t mbox, void void (*copy_data_fun)(smx_synchro_t, void*, size_t), // used to copy data if not default one void *data, double rate) { - XBT_DEBUG("recv from %p %p", mbox, mbox->comm_queue); - smx_synchro_t this_synchro = SIMIX_comm_new(SIMIX_COMM_RECEIVE); + XBT_DEBUG("recv from %p %p", mbox, &mbox->comm_queue); + simgrid::kernel::activity::Comm* this_synchro = new simgrid::kernel::activity::Comm(SIMIX_COMM_RECEIVE); smx_synchro_t other_synchro; //communication already done, get it inside the fifo of completed comms - if (mbox->permanent_receiver && ! mbox->done_comm_queue->empty()) { + if (mbox->permanent_receiver != nullptr && ! mbox->done_comm_queue.empty()) { XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication"); //find a match in the already received fifo - other_synchro = _find_matching_comm(mbox->done_comm_queue, SIMIX_COMM_SEND, match_fun, data, this_synchro,/*remove_matching*/true); + other_synchro = _find_matching_comm(&mbox->done_comm_queue, SIMIX_COMM_SEND, match_fun, data, this_synchro,/*remove_matching*/true); //if not found, assume the receiver came first, register it to the mailbox in the classical way if (!other_synchro) { XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request into fifo"); other_synchro = this_synchro; SIMIX_mbox_push(mbox, this_synchro); } else { - simgrid::simix::Comm *other_comm = static_cast(other_synchro); + simgrid::kernel::activity::Comm *other_comm = static_cast(other_synchro); - if(other_comm->surf_comm && SIMIX_comm_get_remains(other_comm)==0.0) { + if(other_comm->surf_comm && other_comm->remains()==0.0) { 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->refcount--; - SIMIX_comm_destroy(this_synchro); + other_comm->unref(); + static_cast(this_synchro)->unref(); } } else { /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */ @@ -398,15 +297,15 @@ smx_synchro_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_mailbox_t mbox, void * ourself so that the other side also gets a chance of choosing if it wants to match with us. * * If it is not found then push our communication into the rendez-vous point */ - other_synchro = _find_matching_comm(mbox->comm_queue, SIMIX_COMM_SEND, match_fun, data, this_synchro,/*remove_matching*/true); + other_synchro = _find_matching_comm(&mbox->comm_queue, SIMIX_COMM_SEND, match_fun, data, this_synchro,/*remove_matching*/true); if (!other_synchro) { - XBT_DEBUG("Receive pushed first %zu", mbox->comm_queue->size()); + XBT_DEBUG("Receive pushed first %zu", mbox->comm_queue.size()); other_synchro = this_synchro; SIMIX_mbox_push(mbox, this_synchro); } else { - SIMIX_comm_destroy(this_synchro); - simgrid::simix::Comm *other_comm = static_cast(other_synchro); + this_synchro->unref(); + simgrid::kernel::activity::Comm *other_comm = static_cast(other_synchro); other_comm->state = SIMIX_READY; other_comm->type = SIMIX_COMM_READY; @@ -415,7 +314,7 @@ smx_synchro_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_mailbox_t mbox, void } /* Setup communication synchro */ - simgrid::simix::Comm *other_comm = static_cast(other_synchro); + simgrid::kernel::activity::Comm *other_comm = static_cast(other_synchro); other_comm->dst_proc = dst_proc; other_comm->dst_buff = dst_buff; other_comm->dst_buff_size = dst_buff_size; @@ -446,45 +345,41 @@ smx_synchro_t simcall_HANDLER_comm_iprobe(smx_simcall_t simcall, smx_mailbox_t m smx_synchro_t SIMIX_comm_iprobe(smx_process_t dst_proc, smx_mailbox_t mbox, int type, int src, int tag, int (*match_fun)(void *, void *, smx_synchro_t), void *data) { - XBT_DEBUG("iprobe from %p %p", mbox, mbox->comm_queue); - smx_synchro_t this_synchro; + XBT_DEBUG("iprobe from %p %p", mbox, &mbox->comm_queue); + simgrid::kernel::activity::Comm* this_comm; int smx_type; if(type == 1){ - this_synchro=SIMIX_comm_new(SIMIX_COMM_SEND); + this_comm = new simgrid::kernel::activity::Comm(SIMIX_COMM_SEND); smx_type = SIMIX_COMM_RECEIVE; } else{ - this_synchro=SIMIX_comm_new(SIMIX_COMM_RECEIVE); + this_comm = new simgrid::kernel::activity::Comm(SIMIX_COMM_RECEIVE); smx_type = SIMIX_COMM_SEND; } - smx_synchro_t other_synchro=NULL; - if(mbox->permanent_receiver && ! mbox->done_comm_queue->empty()){ + smx_synchro_t other_synchro=nullptr; + if (mbox->permanent_receiver != nullptr && !mbox->done_comm_queue.empty()) { XBT_DEBUG("first check in the permanent recv mailbox, to see if we already got something"); - other_synchro = - _find_matching_comm(mbox->done_comm_queue, (e_smx_comm_type_t) smx_type, match_fun, data, this_synchro,/*remove_matching*/false); + other_synchro = _find_matching_comm(&mbox->done_comm_queue, + (e_smx_comm_type_t) smx_type, match_fun, data, this_comm,/*remove_matching*/false); } if (!other_synchro){ XBT_DEBUG("check if we have more luck in the normal mailbox"); - other_synchro = _find_matching_comm(mbox->comm_queue, (e_smx_comm_type_t) smx_type, match_fun, data, this_synchro,/*remove_matching*/false); + other_synchro = _find_matching_comm(&mbox->comm_queue, + (e_smx_comm_type_t) smx_type, match_fun, data, this_comm,/*remove_matching*/false); } - if(other_synchro) { - simgrid::simix::Comm *other_comm = static_cast(other_synchro); - other_comm->refcount--; - } + if(other_synchro) + other_synchro->unref(); - SIMIX_comm_destroy(this_synchro); + this_comm->unref(); return other_synchro; } void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_synchro_t synchro, double timeout) { - /* the simcall may be a wait, a send or a recv */ - surf_action_t sleep; - /* 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()) { @@ -497,7 +392,7 @@ void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_synchro_t synchro, dou if (timeout == -1) THROW_IMPOSSIBLE; - simgrid::simix::Comm *comm = static_cast(synchro); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); if (comm->src_proc == simcall->issuer) comm->state = SIMIX_SRC_TIMEOUT; else @@ -513,10 +408,10 @@ void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_synchro_t synchro, dou if (synchro->state != SIMIX_WAITING && synchro->state != SIMIX_RUNNING) { SIMIX_comm_finish(synchro); } else { /* if (timeout >= 0) { we need a surf sleep action even when there is no timeout, otherwise surf won't tell us when the host fails */ - sleep = surf_host_sleep(simcall->issuer->host, timeout); + surf_action_t sleep = surf_host_sleep(simcall->issuer->host, timeout); sleep->setData(synchro); - simgrid::simix::Comm *comm = static_cast(synchro); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); if (simcall->issuer == comm->src_proc) comm->src_timeout = sleep; else @@ -526,15 +421,15 @@ void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_synchro_t synchro, dou void simcall_HANDLER_comm_test(smx_simcall_t simcall, smx_synchro_t synchro) { - simgrid::simix::Comm *comm = static_cast(synchro); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); - if(MC_is_active() || MC_record_replay_is_active()){ + if (MC_is_active() || MC_record_replay_is_active()){ simcall_comm_test__set__result(simcall, comm->src_proc && comm->dst_proc); - if(simcall_comm_test__get__result(simcall)){ + 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{ + } else { SIMIX_simcall_answer(simcall); } return; @@ -542,17 +437,18 @@ 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); } } -void simcall_HANDLER_comm_testany(smx_simcall_t simcall, xbt_dynar_t synchros) +void simcall_HANDLER_comm_testany( + smx_simcall_t simcall, simgrid::kernel::activity::Synchro* comms[], size_t count) { - 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()){ @@ -560,19 +456,20 @@ void simcall_HANDLER_comm_testany(smx_simcall_t simcall, xbt_dynar_t synchros) if(idx == -1){ SIMIX_simcall_answer(simcall); }else{ - synchro = xbt_dynar_get_as(synchros, idx, smx_synchro_t); + simgrid::kernel::activity::Synchro* synchro = comms[idx]; 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); } return; } - xbt_dynar_foreach(simcall_comm_testany__get__comms(simcall), cursor,synchro) { + for (std::size_t i = 0; i != count; ++i) { + simgrid::kernel::activity::Synchro* synchro = comms[i]; if (synchro->state != SIMIX_WAITING && synchro->state != SIMIX_RUNNING) { - simcall_comm_testany__set__result(simcall, cursor); - xbt_fifo_push(synchro->simcalls, simcall); + simcall_comm_testany__set__result(simcall, i); + synchro->simcalls.push_back(simcall); SIMIX_comm_finish(synchro); return; } @@ -580,24 +477,36 @@ void simcall_HANDLER_comm_testany(smx_simcall_t simcall, xbt_dynar_t synchros) SIMIX_simcall_answer(simcall); } -void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, xbt_dynar_t synchros) +void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, xbt_dynar_t synchros, double timeout) { smx_synchro_t synchro; unsigned int cursor = 0; if (MC_is_active() || MC_record_replay_is_active()){ + if (timeout != -1) + xbt_die("Timeout not implemented for waitany in the model-checker"); 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); return; } - + + if (timeout == -1 ){ + simcall->timer = NULL; + } else { + simcall->timer = SIMIX_timer_set(timeout, [simcall]() { + SIMIX_waitany_remove_simcall_from_actions(simcall); + simcall_comm_waitany__set__result(simcall, -1); + SIMIX_simcall_answer(simcall); + }); + } + 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){ @@ -614,7 +523,10 @@ void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall) xbt_dynar_t synchros = simcall_comm_waitany__get__comms(simcall); xbt_dynar_foreach(synchros, cursor, synchro) { - xbt_fifo_remove(synchro->simcalls, simcall); + // Remove the first occurence of simcall: + auto i = boost::range::find(synchro->simcalls, simcall); + if (i != synchro->simcalls.end()) + synchro->simcalls.erase(i); } } @@ -624,7 +536,7 @@ void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall) */ static inline void SIMIX_comm_start(smx_synchro_t synchro) { - simgrid::simix::Comm *comm = static_cast(synchro); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); /* If both the sender and the receiver are already there, start the communication */ if (synchro->state == SIMIX_READY) { @@ -643,21 +555,18 @@ static inline void SIMIX_comm_start(smx_synchro_t synchro) XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", sg_host_get_name(sender), sg_host_get_name(receiver)); comm->state = SIMIX_LINK_FAILURE; - SIMIX_comm_destroy_internal_actions(synchro); + comm->cleanupSurf(); } /* If any of the process is suspend, create the synchro but stop its execution, it will be restarted when the sender process resume */ - if (SIMIX_process_is_suspended(comm->src_proc) || - SIMIX_process_is_suspended(comm->dst_proc)) { - /* FIXME: check what should happen with the synchro state */ - + 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) were suspended since it initiated the communication", - sg_host_get_name(comm->src_proc->host), comm->src_proc->name); + XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the communication", + 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) were suspended since it initiated the communication", - sg_host_get_name(comm->dst_proc->host), comm->dst_proc->name); + XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the communication", + comm->dst_proc->name.c_str(), sg_host_get_name(comm->dst_proc->host)); comm->surf_comm->suspend(); } @@ -670,11 +579,12 @@ static inline void SIMIX_comm_start(smx_synchro_t synchro) */ void SIMIX_comm_finish(smx_synchro_t synchro) { - simgrid::simix::Comm *comm = static_cast(synchro); + simgrid::kernel::activity::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 @@ -685,6 +595,10 @@ void SIMIX_comm_finish(smx_synchro_t synchro) continue; // if process handling comm is killed if (simcall->call == SIMCALL_COMM_WAITANY) { SIMIX_waitany_remove_simcall_from_actions(simcall); + if (simcall->timer) { + SIMIX_timer_remove(simcall->timer); + simcall->timer = nullptr; + } if (!MC_is_active() && !MC_record_replay_is_active()) simcall_comm_waitany__set__result(simcall, xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro)); } @@ -700,74 +614,91 @@ 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); + } + 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 = -1; + auto comms = simcall_comm_testany__get__comms(simcall); + auto count = simcall_comm_testany__get__count(simcall); + auto element = std::find(comms, comms + count, synchro); + if (element == comms + count) + e.value = -1; + else + e.value = element - comms; + } + simcall->issuer->exception = std::make_exception_ptr(e); } - else if (simcall->call == SIMCALL_COMM_TESTANY) { - simcall->issuer->running_ctx->exception.value = xbt_dynar_search(simcall_comm_testany__get__comms(simcall), &synchro); + catch(...) { + // Nothing to do } } @@ -775,7 +706,7 @@ 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){ @@ -785,6 +716,8 @@ 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); @@ -792,159 +725,7 @@ void SIMIX_comm_finish(smx_synchro_t synchro) } while (destroy_count-- > 0) - SIMIX_comm_destroy(synchro); -} - -/** - * \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 */ - SIMIX_comm_destroy_internal_actions(comm); - - /* if there are simcalls associated with the synchro, then answer them */ - if (xbt_fifo_size(synchro->simcalls)) { - SIMIX_comm_finish(comm); - } -} - -void SIMIX_comm_cancel(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - - /* if the synchro is a waiting state means that it is still in a mbox */ - /* so remove from it and delete it */ - if (comm->state == SIMIX_WAITING) { - SIMIX_mbox_remove(comm->mbox, synchro); - comm->state = SIMIX_CANCELED; - } - else if (!MC_is_active() /* when running the MC there are no surf actions */ - && !MC_record_replay_is_active() - && (comm->state == SIMIX_READY || comm->state == SIMIX_RUNNING)) { - - comm->surf_comm->cancel(); - } -} - -void SIMIX_comm_suspend(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - - /*FIXME: shall we suspend also the timeout synchro? */ - if (comm->surf_comm) - comm->surf_comm->suspend(); - /* in the other case, the action will be suspended on creation, in SIMIX_comm_start() */ -} - -void SIMIX_comm_resume(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - - /*FIXME: check what happen with the timeouts */ - if (comm->surf_comm) - comm->surf_comm->resume(); - /* in the other case, the synchro were not really suspended yet, see SIMIX_comm_suspend() and SIMIX_comm_start() */ -} - - -/************* synchro Getters **************/ - -/** - * \brief get the amount remaining from the communication - * \param synchro The communication - */ -double SIMIX_comm_get_remains(smx_synchro_t synchro) -{ - if(!synchro) - return 0; - simgrid::simix::Comm *comm = static_cast(synchro); - - double remains; - switch (synchro->state) { - - case SIMIX_RUNNING: - remains = comm->surf_comm->getRemains(); - break; - - case SIMIX_WAITING: - case SIMIX_READY: - remains = 0; /*FIXME: check what should be returned */ - break; - - default: - remains = 0; /*FIXME: is this correct? */ - break; - } - return remains; -} - -e_smx_state_t SIMIX_comm_get_state(smx_synchro_t synchro) -{ - return synchro->state; -} - -/** - * \brief Return the user data associated to the sender of the communication - * \param synchro The communication - * \return the user data - */ -void* SIMIX_comm_get_src_data(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - - return comm->src_data; -} - -/** - * \brief Return the user data associated to the receiver of the communication - * \param synchro The communication - * \return the user data - */ -void* SIMIX_comm_get_dst_data(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - - return comm->dst_data; -} - -smx_process_t SIMIX_comm_get_src_proc(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - - return comm->src_proc; -} - -smx_process_t SIMIX_comm_get_dst_proc(smx_synchro_t synchro) -{ - simgrid::simix::Comm *comm = static_cast(synchro); - - return comm->dst_proc; + static_cast(synchro)->unref(); } /******************************************************************************/ @@ -959,7 +740,7 @@ void SIMIX_comm_set_copy_data_callback(void (*callback) (smx_synchro_t, void*, s void SIMIX_comm_copy_pointer_callback(smx_synchro_t synchro, void* buff, size_t buff_size) { - simgrid::simix::Comm *comm = static_cast(synchro); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); xbt_assert((buff_size == sizeof(void *)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size); *(void **) (comm->dst_buff) = buff; @@ -967,13 +748,13 @@ void SIMIX_comm_copy_pointer_callback(smx_synchro_t synchro, void* buff, size_t void SIMIX_comm_copy_buffer_callback(smx_synchro_t synchro, void* buff, size_t buff_size) { - simgrid::simix::Comm *comm = static_cast(synchro); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); XBT_DEBUG("Copy the data over"); 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; } } @@ -984,7 +765,7 @@ void SIMIX_comm_copy_buffer_callback(smx_synchro_t synchro, void* buff, size_t b */ void SIMIX_comm_copy_data(smx_synchro_t synchro) { - simgrid::simix::Comm *comm = static_cast(synchro); + simgrid::kernel::activity::Comm *comm = static_cast(synchro); size_t buff_size = comm->src_buff_size; /* If there is no data to copy then return */