X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/bbfde1ef5427e4e7e84ca8d38aff13507947c194..2f391a5ab078df821362ea4d82338e4259028630:/src/simix/smx_network.c diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c index 0802fb7233..923173b5d7 100644 --- a/src/simix/smx_network.c +++ b/src/simix/smx_network.c @@ -54,7 +54,7 @@ void SIMIX_rdv_destroy(smx_rdv_t rdv) * \param rdv The rendez-vous point * \param comm The communication request */ -static inline void SIMIX_rdv_push(smx_rdv_t rdv, smx_comm_t comm) +static XBT_INLINE void SIMIX_rdv_push(smx_rdv_t rdv, smx_comm_t comm) { xbt_fifo_push(rdv->comm_fifo, comm); comm->rdv = rdv; @@ -65,7 +65,7 @@ static inline void SIMIX_rdv_push(smx_rdv_t rdv, smx_comm_t comm) * \param rdv The rendez-vous point * \param comm The communication request */ -static inline void SIMIX_rdv_remove(smx_rdv_t rdv, smx_comm_t comm) +static XBT_INLINE void SIMIX_rdv_remove(smx_rdv_t rdv, smx_comm_t comm) { xbt_fifo_remove(rdv->comm_fifo, comm); comm->rdv = NULL; @@ -159,6 +159,7 @@ smx_comm_t SIMIX_communication_new(smx_comm_type_t type) */ void SIMIX_communication_destroy(smx_comm_t comm) { + VERB2("Destroy communication %p; refcount initially %d",comm,comm->refcount); comm->refcount--; if(comm->refcount > 0) return; @@ -194,7 +195,7 @@ void SIMIX_communication_destroy(smx_comm_t comm) * maybe is in use by others. * \ */ -static inline void SIMIX_communication_use(smx_comm_t comm) +static XBT_INLINE void SIMIX_communication_use(smx_comm_t comm) { comm->refcount++; } @@ -203,7 +204,7 @@ static inline void SIMIX_communication_use(smx_comm_t comm) * \brief Start the simulation of a communication request * \param comm The communication request */ -static inline void SIMIX_communication_start(smx_comm_t comm) +static XBT_INLINE void SIMIX_communication_start(smx_comm_t comm) { /* If both the sender and the receiver are already there, start the communication */ if(comm->src_proc && comm->dst_proc){ @@ -233,7 +234,44 @@ static inline void SIMIX_communication_start(smx_comm_t comm) } /** - * \brief Waits for communication completion and performs error checking + * \brief Performs error checking and cleanup + * \param comm The communication + */ +static XBT_INLINE void SIMIX_communication_cleanup(smx_comm_t comm) +{ + DEBUG1("Checking errors and cleaning communication %p", comm); + + /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */ + SIMIX_sem_release_forever(comm->sem); + + /* Check for errors other than timeouts */ + if (!SIMIX_host_get_state(SIMIX_host_self())){ + if(comm->rdv) + SIMIX_rdv_remove(comm->rdv, comm); + SIMIX_communication_destroy(comm); + THROW0(host_error, 0, "Host failed"); + } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){ + SIMIX_communication_destroy(comm); + THROW0(network_error, 0, "Link failure"); + } else if (!SIMIX_host_get_state(SIMIX_process_get_host(comm->dst_proc)) || + !SIMIX_host_get_state(SIMIX_process_get_host(comm->src_proc))) { + /* We test both src&dst because we dunno who we are today, and we already tested myself above. + * So, at the end, we test the remote peer only + * Moreover, we have to test it because if the remote peer fails, the action comm->act is not done nor failed. + * In that case, we got awaken by the little endless actions created in the SIMIX_sem_acquire(comm->sem) + * at the beginning of this function. */ + SIMIX_communication_destroy(comm); + THROW0(network_error, 0, "Remote peer failed"); + + } + /* Copy network data */ + SIMIX_network_copy_data(comm); + + SIMIX_communication_destroy(comm); +} + +/** + * \brief Waits for communication completion * \param comm The communication * \param timeout The max amount of time to wait for the communication to finish * @@ -242,7 +280,7 @@ static inline void SIMIX_communication_start(smx_comm_t comm) * - timeout_error if communication reached the timeout specified (either because of local peer or remote peer) * - network_error if network failed or remote peer failed */ -static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout) +static XBT_INLINE void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout) { smx_action_t act_sleep = NULL; int src_timeout = 0; @@ -285,35 +323,8 @@ static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, doub THROW1(timeout_error, 0, "Communication timeouted because of %s",src_timeout?"the source":"the destination"); } - DEBUG1("Communication %p complete! Let's check for errors", comm); - - /* Make sure that everyone sleeping on that semaphore is awake, and that nobody will ever block on it */ - SIMIX_sem_release_forever(comm->sem); - - /* Check for errors other than timeouts (they are catched above) */ - if (!SIMIX_host_get_state(SIMIX_host_self())){ - if(comm->rdv) - SIMIX_rdv_remove(comm->rdv, comm); - SIMIX_communication_destroy(comm); - THROW0(host_error, 0, "Host failed"); - } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){ - SIMIX_communication_destroy(comm); - THROW0(network_error, 0, "Link failure"); - } else if (!SIMIX_host_get_state(SIMIX_process_get_host(comm->dst_proc)) || - !SIMIX_host_get_state(SIMIX_process_get_host(comm->src_proc))) { - /* We test both src&dst because we dunno who we are today, and we already tested myself above. - * So, at the end, we test the remote peer only - * Moreover, we have to test it because if the remote peer fails, the action comm->act is not done nor failed. - * In that case, we got awaken by the little endless actions created in the SIMIX_sem_acquire(comm->sem) - * at the beginning of this function. */ - SIMIX_communication_destroy(comm); - THROW0(network_error, 0, "Remote peer failed"); - - } - /* Copy network data */ - SIMIX_network_copy_data(comm); - - SIMIX_communication_destroy(comm); + DEBUG1("Communication %p complete!", comm); + SIMIX_communication_cleanup(comm); } /** @@ -360,7 +371,8 @@ void SIMIX_network_copy_buffer_callback(smx_comm_t comm, size_t buff_size) { void SIMIX_network_copy_data(smx_comm_t comm) { size_t buff_size = comm->src_buff_size; - + uintptr_t casted_size = 0; + uintptr_t amount = 0; /* If there is no data to be copy then return */ if(!comm->src_buff || !comm->dst_buff || comm->copied == 1) return; @@ -391,8 +403,8 @@ void SIMIX_network_copy_data(smx_comm_t comm) { if (msg_sizes == NULL) msg_sizes = xbt_dict_new(); - uintptr_t casted_size = comm->task_size; - uintptr_t amount = xbt_dicti_get(msg_sizes, casted_size); + casted_size = comm->task_size; + amount = xbt_dicti_get(msg_sizes, casted_size); amount++; xbt_dicti_set(msg_sizes,casted_size, amount); @@ -401,10 +413,13 @@ void SIMIX_network_copy_data(smx_comm_t comm) #include "xbt.h" /* pimple to display the message sizes */ void SIMIX_message_sizes_output(const char *filename) { - FILE * out = fopen(filename,"w"); - xbt_assert1(out,"Cannot open file %s",filename); - uintptr_t key,data; + uintptr_t key = 0; + uintptr_t data = 0; xbt_dict_cursor_t cursor; + FILE * out = NULL; + out = fopen(filename,"w"); + xbt_assert1(out,"Cannot open file %s",filename); + xbt_dict_foreach(msg_sizes,cursor,key,data) { fprintf(out,"%zu %zu\n",key,data); } @@ -476,15 +491,11 @@ smx_comm_t SIMIX_network_isend(smx_rdv_t rdv, double task_size, double rate, { smx_comm_t comm; - /* If running in model-checking mode then intercept the communication action. - * There can be two situations, this is a replay (back-track process) so no - * transition needs to be created, or we are expanding a new state and in - * consecuence a transition needs to be created and added to the set of - * transitions associated to the current state. - */ - if (_surf_do_model_check) - MC_trans_intercept_isend(rdv); - + /*If running in model-checking mode then intercept the communication action*/ + #ifdef HAVE_MC + if (_surf_do_model_check) + MC_trans_intercept_isend(rdv); + #endif /* Look for communication request matching our needs. If it is not found then create it and push it into the rendez-vous point */ comm = SIMIX_rdv_get_request(rdv, comm_recv); @@ -510,15 +521,11 @@ smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_s { smx_comm_t comm; - /* If running in model-checking mode then intercept the communication action. - * There can be two situations, this is a replay (back-track process) so no - * transition needs to be created, or we are expanding a new state and in - * consecuence a transition needs to be created and added to the set of - * transitions associated to the current state. - */ + /*If running in model-checking mode then intercept the communication action*/ +#ifdef HAVE_MC if (_surf_do_model_check) MC_trans_intercept_irecv(rdv); - +#endif /* Look for communication request matching our needs. * If it is not found then create it and push it into the rendez-vous point */ @@ -541,15 +548,11 @@ smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_s /** @brief blocks until the communication terminates or the timeout occurs */ XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout) { - /* If running in model-checking mode then intercept the communication action. - * There can be two situations, this is a replay (back-track process) so no - * transition needs to be created, or we are expanding a new state and in - * consecuence a transition needs to be created and added to the set of - * transitions associated to the current state. - */ + /*If running in model-checking mode then intercept the communication action*/ +#ifdef HAVE_MC if (_surf_do_model_check) MC_trans_intercept_wait(comm); - +#endif /* Wait for communication completion */ SIMIX_communication_wait_for_completion(comm, timeout); } @@ -557,14 +560,11 @@ XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout) /** @Returns whether the (asynchronous) communication is done yet or not */ XBT_INLINE int SIMIX_network_test(smx_comm_t comm) { - /* If running in model-checking mode then intercept the communication action. - * There can be two situations, this is a replay (back-track process) so no - * transition needs to be created, or we are expanding a new state and in - * consecuence a transition needs to be created and added to the set of - * transitions associated to the current state. - */ + /*If running in model-checking mode then intercept the communication action*/ +#ifdef HAVE_MC if (_surf_do_model_check) MC_trans_intercept_test(comm); +#endif /* Copy data if the communication is done */ if(comm->sem && !SIMIX_sem_would_block(comm->sem)){ @@ -585,15 +585,11 @@ unsigned int SIMIX_network_waitany(xbt_dynar_t comms) unsigned int cursor, found_comm=-1; smx_comm_t comm,comm_finished=NULL; - /* If running in model-checking mode then intercept the communication action. - * There can be two situations, this is a replay (back-track process) so no - * transition needs to be created, or we are expanding a new state and in - * consecuence a transition needs to be created and added to the set of - * transitions associated to the current state. - */ + /*If running in model-checking mode then intercept the communication action*/ +#ifdef HAVE_MC if (_surf_do_model_check) MC_trans_intercept_waitany(comms); - +#endif xbt_dynar_foreach(comms,cursor,comm) xbt_dynar_push(sems,&(comm->sem)); @@ -604,7 +600,11 @@ unsigned int SIMIX_network_waitany(xbt_dynar_t comms) xbt_assert0(found_comm!=-1,"Cannot find which communication finished"); xbt_dynar_get_cpy(comms,found_comm,&comm_finished); - /* Check for errors and cleanup the comm */ - SIMIX_communication_wait_for_completion(comm_finished,-1); + DEBUG2("Communication %p of communication set %p finished", comm_finished, comms); + + /* let the regular code deal with the communication end (errors checking and cleanup). + * A bit of useless work will be done, but that's good for source factorization */ + SIMIX_sem_release_forever(comm_finished->sem); + SIMIX_communication_wait_for_completion(comm_finished, -1); return found_comm; -} \ No newline at end of file +}