X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/a24e5e842b10085c859c6988099e815964db2eda..82904dcc37f75453c571a1a74e67c1ba282ebe23:/src/msg/gos.c diff --git a/src/msg/gos.c b/src/msg/gos.c index 92dfac4a6a..cb6c929ae1 100644 --- a/src/msg/gos.c +++ b/src/msg/gos.c @@ -37,23 +37,24 @@ MSG_error_t MSG_get_errno(void) MSG_error_t MSG_task_execute(m_task_t task) { simdata_task_t simdata = NULL; - m_process_t self = MSG_process_self(); + simdata_process_t p_simdata; + e_smx_state_t comp_state; CHECK_HOST(); simdata = task->simdata; - xbt_assert0(simdata->host_nb == 0, + xbt_assert(simdata->host_nb == 0, "This is a parallel task. Go to hell."); #ifdef HAVE_TRACING TRACE_msg_task_execute_start(task); #endif - xbt_assert1((!simdata->compute) && (task->simdata->isused == 0), + xbt_assert((!simdata->compute) && (task->simdata->isused == 0), "This task is executed somewhere else. Go fix your code! %d", task->simdata->isused); - DEBUG1("Computing on %s", MSG_process_self()->simdata->m_host->name); + XBT_DEBUG("Computing on %s", MSG_process_get_name(MSG_process_self())); if (simdata->computation_amount == 0) { #ifdef HAVE_TRACING @@ -64,22 +65,22 @@ MSG_error_t MSG_task_execute(m_task_t task) simdata->isused=1; simdata->compute = SIMIX_req_host_execute(task->name, SIMIX_host_self(), - simdata->computation_amount); - SIMIX_req_host_execution_set_priority(simdata->compute, simdata->priority); + simdata->computation_amount, + simdata->priority); #ifdef HAVE_TRACING SIMIX_req_set_category(simdata->compute, task->category); #endif - self->simdata->waiting_action = simdata->compute; - SIMIX_req_host_execution_wait(simdata->compute); - self->simdata->waiting_action = NULL; + p_simdata = SIMIX_process_self_get_data(); + p_simdata->waiting_action = simdata->compute; + comp_state = SIMIX_req_host_execution_wait(simdata->compute); + p_simdata->waiting_action = NULL; simdata->isused=0; - DEBUG2("Execution task '%s' finished in state %d", task->name, SIMIX_req_host_execution_get_state(task->simdata->compute)); - if (SIMIX_req_host_execution_get_state(task->simdata->compute) == SIMIX_DONE) { + XBT_DEBUG("Execution task '%s' finished in state %d", task->name, comp_state); + if (comp_state == SIMIX_DONE) { /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */ - SIMIX_req_host_execution_destroy(task->simdata->compute); simdata->computation_amount = 0.0; simdata->comm = NULL; simdata->compute = NULL; @@ -89,7 +90,6 @@ MSG_error_t MSG_task_execute(m_task_t task) MSG_RETURN(MSG_OK); } else if (SIMIX_req_host_get_state(SIMIX_host_self()) == 0) { /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */ - SIMIX_req_host_execution_destroy(task->simdata->compute); simdata->comm = NULL; simdata->compute = NULL; #ifdef HAVE_TRACING @@ -98,7 +98,6 @@ MSG_error_t MSG_task_execute(m_task_t task) MSG_RETURN(MSG_HOST_FAILURE); } else { /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */ - SIMIX_req_host_execution_destroy(task->simdata->compute); simdata->comm = NULL; simdata->compute = NULL; #ifdef HAVE_TRACING @@ -167,19 +166,21 @@ MSG_parallel_task_create(const char *name, int host_nb, MSG_error_t MSG_parallel_task_execute(m_task_t task) { simdata_task_t simdata = NULL; - m_process_t self = MSG_process_self(); + e_smx_state_t comp_state; + simdata_process_t p_simdata; CHECK_HOST(); simdata = task->simdata; + p_simdata = SIMIX_process_self_get_data(); - xbt_assert0((!simdata->compute) + xbt_assert((!simdata->compute) && (task->simdata->isused == 0), "This task is executed somewhere else. Go fix your code!"); - xbt_assert0(simdata->host_nb, + xbt_assert(simdata->host_nb, "This is not a parallel task. Go to hell."); - DEBUG1("Parallel computing on %s", MSG_process_self()->simdata->m_host->name); + XBT_DEBUG("Parallel computing on %s", p_simdata->m_host->name); simdata->isused=1; @@ -188,32 +189,29 @@ MSG_error_t MSG_parallel_task_execute(m_task_t task) simdata->host_list, simdata->comp_amount, simdata->comm_amount, 1.0, -1.0); - DEBUG1("Parallel execution action created: %p", simdata->compute); + XBT_DEBUG("Parallel execution action created: %p", simdata->compute); - self->simdata->waiting_action = simdata->compute; - SIMIX_req_host_execution_wait(simdata->compute); - self->simdata->waiting_action = NULL; + p_simdata->waiting_action = simdata->compute; + comp_state = SIMIX_req_host_execution_wait(simdata->compute); + p_simdata->waiting_action = NULL; - DEBUG2("Finished waiting for execution of action %p, state = %d", simdata->compute, SIMIX_req_host_execution_get_state(task->simdata->compute)); + XBT_DEBUG("Finished waiting for execution of action %p, state = %d", simdata->compute, comp_state); simdata->isused=0; - if (SIMIX_req_host_execution_get_state(task->simdata->compute) == SIMIX_DONE) { + if (comp_state == SIMIX_DONE) { /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */ - SIMIX_req_host_execution_destroy(task->simdata->compute); simdata->computation_amount = 0.0; simdata->comm = NULL; simdata->compute = NULL; MSG_RETURN(MSG_OK); } else if (SIMIX_req_host_get_state(SIMIX_host_self()) == 0) { /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */ - SIMIX_req_host_execution_destroy(task->simdata->compute); simdata->comm = NULL; simdata->compute = NULL; MSG_RETURN(MSG_HOST_FAILURE); } else { /* action ended, set comm and compute = NULL, the actions is already destroyed in the main function */ - SIMIX_req_host_execution_destroy(task->simdata->compute); simdata->comm = NULL; simdata->compute = NULL; MSG_RETURN(MSG_TASK_CANCELLED); @@ -247,22 +245,16 @@ MSG_error_t MSG_process_sleep(double nb_sec) proc->simdata->waiting_action = NULL;*/ if (state == SIMIX_DONE) { - if (SIMIX_req_host_get_state(SIMIX_host_self()) == SURF_RESOURCE_OFF) { #ifdef HAVE_TRACING - TRACE_msg_process_sleep_out(MSG_process_self()); + TRACE_msg_process_sleep_out(MSG_process_self()); #endif - MSG_RETURN(MSG_HOST_FAILURE); - } + MSG_RETURN(MSG_OK); } else { #ifdef HAVE_TRACING TRACE_msg_process_sleep_out(MSG_process_self()); #endif MSG_RETURN(MSG_HOST_FAILURE); } -#ifdef HAVE_TRACING - TRACE_msg_process_sleep_out(MSG_process_self()); -#endif - MSG_RETURN(MSG_OK); } /** \ingroup msg_gos_functions @@ -339,7 +331,7 @@ MSG_error_t MSG_task_get_ext(m_task_t * task, m_channel_t channel, double timeout, m_host_t host) { - xbt_assert1((channel >= 0) + xbt_assert((channel >= 0) && (channel < msg_global->max_channel), "Invalid channel %d", channel); @@ -372,7 +364,7 @@ MSG_error_t MSG_task_receive_ext(m_task_t * task, const char *alias, double timeout, m_host_t host) { - DEBUG1 + XBT_DEBUG ("MSG_task_receive_ext: Trying to receive a message on mailbox '%s'", alias); return MSG_mailbox_get_task_ext(MSG_mailbox_get_by_alias(alias), task, @@ -380,16 +372,34 @@ MSG_task_receive_ext(m_task_t * task, const char *alias, double timeout, } /** \ingroup msg_gos_functions - * \brief Send a task on a channel. + * \brief Sends a task on a mailbox. + * + * This is a non blocking function: use MSG_comm_wait() or MSG_comm_test() + * to end the communication. * - * This function takes two parameter. * \param task a #m_task_t to send on another location. - * \param alias the channel on which the agent should put this - task. This value has to be >=0 and < than the maximal number of - channels fixed with MSG_set_channel_number(). - * \return the msg_comm_t communication. + * \param alias name of the mailbox to sent the task to + * \return the msg_comm_t communication created */ msg_comm_t MSG_task_isend(m_task_t task, const char *alias) +{ + return MSG_task_isend_with_matching(task,alias,NULL,NULL); +} +/** \ingroup msg_gos_functions + * \brief Sends a task on a mailbox, with support for matching requests + * + * This is a non blocking function: use MSG_comm_wait() or MSG_comm_test() + * to end the communication. + * + * \param task a #m_task_t to send on another location. + * \param alias name of the mailbox to sent the task to + * \param match_fun boolean function taking the #match_data provided by sender (here), and the one of the receiver (if any) and returning whether they match + * \param match_data user provided data passed to match_fun + * \return the msg_comm_t communication created + */ +XBT_INLINE msg_comm_t MSG_task_isend_with_matching(m_task_t task, const char *alias, + int (*match_fun)(void*,void*), + void *match_data) { simdata_task_t t_simdata = NULL; m_process_t process = MSG_process_self(); @@ -397,84 +407,223 @@ msg_comm_t MSG_task_isend(m_task_t task, const char *alias) CHECK_HOST(); - /* FIXME: these functions are not tracable */ + /* FIXME: these functions are not traceable */ /* Prepare the task to send */ t_simdata = task->simdata; t_simdata->sender = process; t_simdata->source = MSG_host_self(); - xbt_assert0(t_simdata->isused == 0, + xbt_assert(t_simdata->isused == 0, "This task is still being used somewhere else. You cannot send it now. Go fix your code!"); - t_simdata->isused=1; + t_simdata->isused = 1; msg_global->sent_msg++; /* Send it by calling SIMIX network layer */ + msg_comm_t comm = xbt_new0(s_msg_comm_t, 1); + comm->task_sent = task; + comm->task_received = NULL; + comm->status = MSG_OK; + comm->s_comm = + SIMIX_req_comm_isend(mailbox, t_simdata->message_size, + t_simdata->rate, task, sizeof(void *), match_fun, match_data, 0); + t_simdata->comm = comm->s_comm; /* FIXME: is the field t_simdata->comm still useful? */ - return SIMIX_req_comm_isend(mailbox, t_simdata->message_size, - t_simdata->rate, task, sizeof(void *), NULL, - &t_simdata->comm); + return comm; } /** \ingroup msg_gos_functions - * \brief Listen on a channel for receiving a task from an asynchronous communication. + * \brief Sends a task on a mailbox. + * + * This is a non blocking detached send function. + * Think of it as a best effort send. The task should + * be destroyed by the receiver. + * + * \param task a #m_task_t to send on another location. + * \param alias name of the mailbox to sent the task to + * \param cleanup a function to destroy the task if the + * communication fails (if NULL, MSG_task_destroy() will + * be used by default) + */ +void MSG_task_dsend(m_task_t task, const char *alias, void_f_pvoid_t cleanup) +{ + simdata_task_t t_simdata = NULL; + m_process_t process = MSG_process_self(); + msg_mailbox_t mailbox = MSG_mailbox_get_by_alias(alias); + + CHECK_HOST(); + + if (cleanup == NULL) { + cleanup = (void_f_pvoid_t) MSG_task_destroy; + } + + /* FIXME: these functions are not traceable */ + + /* Prepare the task to send */ + t_simdata = task->simdata; + t_simdata->sender = process; + t_simdata->source = MSG_host_self(); + + xbt_assert(t_simdata->isused == 0, + "This task is still being used somewhere else. You cannot send it now. Go fix your code!"); + + t_simdata->isused = 1; + msg_global->sent_msg++; + + /* Send it by calling SIMIX network layer */ + SIMIX_req_comm_isend(mailbox, t_simdata->message_size, + t_simdata->rate, task, sizeof(void *), NULL, cleanup, 1); +} + +/** \ingroup msg_gos_functions + * \brief Starts listening for receiving a task from an asynchronous communication. + * + * This is a non blocking function: use MSG_comm_wait() or MSG_comm_test() + * to end the communication. * - * It takes two parameters. * \param task a memory location for storing a #m_task_t. - * \param alias the channel on which the agent should be - listening. This value has to be >=0 and < than the maximal - number of channels fixed with MSG_set_channel_number(). - * \return the msg_comm_t communication. + * \param name of the mailbox to receive the task on + * \return the msg_comm_t communication created */ -msg_comm_t MSG_task_irecv(m_task_t * task, const char *alias) +msg_comm_t MSG_task_irecv(m_task_t *task, const char *name) { - smx_rdv_t rdv = MSG_mailbox_get_by_alias(alias); + smx_rdv_t rdv = MSG_mailbox_get_by_alias(name); CHECK_HOST(); /* FIXME: these functions are not tracable */ /* Sanity check */ - xbt_assert0(task, "Null pointer for the task storage"); + xbt_assert(task, "Null pointer for the task storage"); if (*task) - CRITICAL0 + XBT_CRITICAL ("MSG_task_get() was asked to write in a non empty task struct."); /* Try to receive it by calling SIMIX network layer */ - return SIMIX_req_comm_irecv(rdv, task, NULL, NULL, NULL); + msg_comm_t comm = xbt_new0(s_msg_comm_t, 1); + comm->task_sent = NULL; + comm->task_received = task; + comm->status = MSG_OK; + comm->s_comm = SIMIX_req_comm_irecv(rdv, task, NULL, NULL, NULL); + + return comm; } /** \ingroup msg_gos_functions - * \brief Test the status of a communication. - * - * It takes one parameter. - * \param comm the communication to test. - * \return the status of the communication: - * TRUE : the communication is completed - * FALSE: the communication is incompleted + * \brief Checks whether a communication is done, and if yes, finalizes it. + * \param comm the communication to test + * \return TRUE if the communication is finished + * (but it may have failed, use MSG_comm_get_status() to know its status) + * or FALSE if the communication is not finished yet * If the status is FALSE, don't forget to use MSG_process_sleep() after the test. */ int MSG_comm_test(msg_comm_t comm) { - return SIMIX_req_comm_test(comm); + xbt_ex_t e; + int finished = 0; + TRY { + finished = SIMIX_req_comm_test(comm->s_comm); + } + CATCH(e) { + switch (e.category) { + + case host_error: + comm->status = MSG_HOST_FAILURE; + finished = 1; + break; + + case network_error: + comm->status = MSG_TRANSFER_FAILURE; + finished = 1; + break; + + case timeout_error: + comm->status = MSG_TIMEOUT; + finished = 1; + break; + + default: + RETHROW; + } + xbt_ex_free(e); + } + + return finished; } /** \ingroup msg_gos_functions - * \brief After received TRUE to MSG_comm_test(), the communication must be destroyed. - * - * It takes one parameter. + * \brief This function checks if a communication is finished. + * \param comms a vector of communications + * \return the position of the finished communication if any + * (but it may have failed, use MSG_comm_get_status() to know its status), + * or -1 if none is finished + */ +int MSG_comm_testany(xbt_dynar_t comms) +{ + xbt_ex_t e; + int finished_index = -1; + + /* create the equivalent dynar with SIMIX objects */ + xbt_dynar_t s_comms = xbt_dynar_new(sizeof(smx_action_t), NULL); + msg_comm_t comm; + unsigned int cursor; + xbt_dynar_foreach(comms, cursor, comm) { + xbt_dynar_push(s_comms, &comm->s_comm); + } + + MSG_error_t status = MSG_OK; + TRY { + finished_index = SIMIX_req_comm_testany(s_comms); + } + CATCH(e) { + switch (e.category) { + + case host_error: + finished_index = e.value; + status = MSG_HOST_FAILURE; + break; + + case network_error: + finished_index = e.value; + status = MSG_TRANSFER_FAILURE; + break; + + case timeout_error: + finished_index = e.value; + status = MSG_TIMEOUT; + break; + + default: + RETHROW; + } + xbt_ex_free(e); + } + xbt_dynar_free(&s_comms); + + if (finished_index != -1) { + comm = xbt_dynar_get_as(comms, finished_index, msg_comm_t); + /* the communication is finished */ + comm->status = status; + } + + return finished_index; +} + +/** \ingroup msg_gos_functions + * \brief Destroys a communication. * \param comm the communication to destroy. */ void MSG_comm_destroy(msg_comm_t comm) { - if (SIMIX_req_comm_get_src_proc(comm) != SIMIX_process_self()) { - m_task_t task; - task = (m_task_t) SIMIX_req_comm_get_src_buff(comm); - task->simdata->isused=0; + if (comm->task_received != NULL + && *comm->task_received != NULL + && MSG_comm_get_status(comm) == MSG_OK) { + (*comm->task_received)->simdata->isused = 0; } - SIMIX_req_comm_destroy(comm); + + xbt_free(comm); } /** \ingroup msg_gos_functions @@ -488,44 +637,42 @@ void MSG_comm_destroy(msg_comm_t comm) MSG_error_t MSG_comm_wait(msg_comm_t comm, double timeout) { xbt_ex_t e; - MSG_error_t res = MSG_OK; TRY { - SIMIX_req_comm_wait(comm, timeout); + SIMIX_req_comm_wait(comm->s_comm, timeout); - if (SIMIX_req_comm_get_src_proc(comm) != SIMIX_process_self()) { - m_task_t task; - task = (m_task_t) SIMIX_req_comm_get_src_buff(comm); - task->simdata->isused=0; + if (comm->task_received != NULL) { + /* I am the receiver */ + (*comm->task_received)->simdata->isused = 0; } - /* FIXME: these functions are not tracable */ + /* FIXME: these functions are not traceable */ } CATCH(e) { switch (e.category) { case host_error: - res = MSG_HOST_FAILURE; + comm->status = MSG_HOST_FAILURE; break; case network_error: - res = MSG_TRANSFER_FAILURE; + comm->status = MSG_TRANSFER_FAILURE; break; case timeout_error: - res = MSG_TIMEOUT; + comm->status = MSG_TIMEOUT; break; default: - xbt_die(bprintf("Unhandled SIMIX network exception: %s", e.msg)); + RETHROW; } xbt_ex_free(e); } - return res; + + return comm->status; } /** \ingroup msg_gos_functions * \brief This function is called by a sender and permit to wait for each communication * -* It takes three parameters. * \param comm a vector of communication * \param nb_elem is the size of the comm vector -* \param timeout for each call of MSG_comm_wait +* \param timeout for each call of MSG_comm_wait */ void MSG_comm_waitall(msg_comm_t * comm, int nb_elem, double timeout) { @@ -536,33 +683,79 @@ void MSG_comm_waitall(msg_comm_t * comm, int nb_elem, double timeout) } /** \ingroup msg_gos_functions -* \brief This function wait for the first completed communication -* -* It takes on parameter. -* \param comms a vector of communication -* \return the position of the completed communication from the xbt_dynar_t. -*/ + * \brief This function waits for the first communication finished in a list. + * \param comms a vector of communications + * \return the position of the first finished communication + * (but it may have failed, use MSG_comm_get_status() to know its status) + */ int MSG_comm_waitany(xbt_dynar_t comms) { - return SIMIX_req_comm_waitany(comms); + xbt_ex_t e; + int finished_index = -1; + + /* create the equivalent dynar with SIMIX objects */ + xbt_dynar_t s_comms = xbt_dynar_new(sizeof(smx_action_t), NULL); + msg_comm_t comm; + unsigned int cursor; + xbt_dynar_foreach(comms, cursor, comm) { + xbt_dynar_push(s_comms, &comm->s_comm); + } + + MSG_error_t status = MSG_OK; + TRY { + finished_index = SIMIX_req_comm_waitany(s_comms); + } + CATCH(e) { + switch (e.category) { + + case host_error: + finished_index = e.value; + status = MSG_HOST_FAILURE; + break; + + case network_error: + finished_index = e.value; + status = MSG_TRANSFER_FAILURE; + break; + + case timeout_error: + finished_index = e.value; + status = MSG_TIMEOUT; + break; + + default: + RETHROW; + } + xbt_ex_free(e); + } + + xbt_assert(finished_index != -1, "WaitAny returned -1"); + xbt_dynar_free(&s_comms); + + comm = xbt_dynar_get_as(comms, finished_index, msg_comm_t); + /* the communication is finished */ + comm->status = status; + + return finished_index; } -/** \ingroup msg_gos_functions -* \brief This function wait for the first completed communication -* -* It takes on parameter. -* \param comms a vector of communication -* \return the position of the completed communication, if any, or -1 if none was completed -*/ -int MSG_comm_testany(xbt_dynar_t comms) -{ - return SIMIX_req_comm_testany(comms); +/** + * \ingroup msg_gos_functions + * \brief Returns the error (if any) that occured during a finished communication. + * \param comm a finished communication + * \return the status of the communication, or MSG_OK if no error occured + * during the communication + */ +MSG_error_t MSG_comm_get_status(msg_comm_t comm) { + + return comm->status; } m_task_t MSG_comm_get_task(msg_comm_t comm) { - xbt_assert0(comm, "Invalid parameters"); - return (m_task_t) SIMIX_req_comm_get_src_buff(comm); + xbt_assert(comm, "Invalid parameter"); + + return comm->task_received ? *comm->task_received : comm->task_sent; } /** \ingroup msg_gos_functions @@ -643,11 +836,11 @@ MSG_error_t MSG_task_put_with_timeout(m_task_t task, m_host_t dest, m_channel_t channel, double timeout) { - xbt_assert1((channel >= 0) + xbt_assert((channel >= 0) && (channel < msg_global->max_channel), "Invalid channel %d", channel); - DEBUG1("MSG_task_put_with_timout: Trying to send a task to '%s'", dest->name); + XBT_DEBUG("MSG_task_put_with_timout: Trying to send a task to '%s'", dest->name); return MSG_mailbox_put_with_timeout(MSG_mailbox_get_by_channel (dest, channel), task, timeout); @@ -655,7 +848,7 @@ MSG_task_put_with_timeout(m_task_t task, m_host_t dest, MSG_error_t MSG_task_send(m_task_t task, const char *alias) { - DEBUG1("MSG_task_send: Trying to send a message on mailbox '%s'", alias); + XBT_DEBUG("MSG_task_send: Trying to send a message on mailbox '%s'", alias); return MSG_task_send_with_timeout(task, alias, -1); } @@ -694,7 +887,7 @@ int MSG_task_listen(const char *alias) */ int MSG_task_Iprobe(m_channel_t channel) { - xbt_assert1((channel >= 0) + xbt_assert((channel >= 0) && (channel < msg_global->max_channel), "Invalid channel %d", channel); @@ -720,7 +913,7 @@ int MSG_task_Iprobe(m_channel_t channel) */ int MSG_task_probe_from_host(int channel, m_host_t host) { - xbt_assert1((channel >= 0) + xbt_assert((channel >= 0) && (channel < msg_global->max_channel), "Invalid channel %d", channel); @@ -757,7 +950,7 @@ int MSG_task_probe_from(m_channel_t channel) CHECK_HOST(); - xbt_assert1((channel >= 0) + xbt_assert((channel >= 0) && (channel < msg_global->max_channel), "Invalid channel %d", channel);