X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/1af04af978a0d15f9e4e16db1270e42ff0fc6d11..28e032fe2974bf84e0414e4d3d4cd89a38752dae:/src/msg/msg_gos.c diff --git a/src/msg/msg_gos.c b/src/msg/msg_gos.c index 2b28c971cb..4aa46346da 100644 --- a/src/msg/msg_gos.c +++ b/src/msg/msg_gos.c @@ -1,4 +1,5 @@ -/* Copyright (c) 2004-2012. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2004-2015. The SimGrid Team. + * All rights reserved. */ /* 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. */ @@ -23,7 +24,15 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(msg_gos, msg, */ msg_error_t MSG_task_execute(msg_task_t task) { - return MSG_parallel_task_execute(task); + /* TODO: add this to other locations */ + msg_host_t host = MSG_process_get_host(MSG_process_self()); + MSG_host_add_task(host, task); + + msg_error_t ret = MSG_parallel_task_execute(task); + + MSG_host_del_task(host, task); + + return ret; } /** \ingroup msg_task_usage @@ -43,52 +52,56 @@ msg_error_t MSG_parallel_task_execute(msg_task_t task) e_smx_state_t comp_state; msg_error_t status = MSG_OK; -#ifdef HAVE_TRACING TRACE_msg_task_execute_start(task); -#endif xbt_assert((!simdata->compute) && (task->simdata->isused == 0), "This task is executed somewhere else. Go fix your code! %d", - task->simdata->isused); + task->simdata->isused!=NULL); XBT_DEBUG("Computing on %s", MSG_process_get_name(MSG_process_self())); - if (simdata->computation_amount == 0 && !simdata->host_nb) { -#ifdef HAVE_TRACING + if (simdata->flops_amount == 0 && !simdata->host_nb) { TRACE_msg_task_execute_end(task); -#endif return MSG_OK; } TRY { - - simdata->isused=1; + if (msg_global->debug_multiple_use) + MSG_BT(simdata->isused, "Using Backtrace"); + else + simdata->isused = (void*)1; if (simdata->host_nb > 0) { simdata->compute = simcall_host_parallel_execute(task->name, simdata->host_nb, simdata->host_list, - simdata->comp_amount, - simdata->comm_amount, + simdata->flops_parallel_amount, + simdata->bytes_parallel_amount, 1.0, -1.0); XBT_DEBUG("Parallel execution action created: %p", simdata->compute); } else { + unsigned long affinity_mask = (unsigned long) xbt_dict_get_or_null_ext(simdata->affinity_mask_db, (char *) p_simdata->m_host, sizeof(msg_host_t)); + XBT_DEBUG("execute %s@%s with affinity(0x%04lx)", MSG_task_get_name(task), MSG_host_get_name(p_simdata->m_host), affinity_mask); + simdata->compute = simcall_host_execute(task->name, p_simdata->m_host, - simdata->computation_amount, - simdata->priority); + simdata->flops_amount, + simdata->priority, + simdata->bound, + affinity_mask + ); } -#ifdef HAVE_TRACING simcall_set_category(simdata->compute, task->category); -#endif p_simdata->waiting_action = simdata->compute; comp_state = simcall_host_execution_wait(simdata->compute); p_simdata->waiting_action = NULL; - simdata->isused=0; + if (msg_global->debug_multiple_use && simdata->isused!=0) + xbt_ex_free(*(xbt_ex_t*)simdata->isused); + simdata->isused = 0; XBT_DEBUG("Execution task '%s' finished in state %d", task->name, (int)comp_state); @@ -98,6 +111,9 @@ msg_error_t MSG_parallel_task_execute(msg_task_t task) case cancel_error: status = MSG_TASK_CANCELED; break; + case host_error: + status = MSG_HOST_FAILURE; + break; default: RETHROW; } @@ -105,12 +121,10 @@ msg_error_t MSG_parallel_task_execute(msg_task_t task) } /* action ended, set comm and compute = NULL, the actions is already destroyed * in the main function */ - simdata->computation_amount = 0.0; + simdata->flops_amount = 0.0; simdata->comm = NULL; simdata->compute = NULL; -#ifdef HAVE_TRACING TRACE_msg_task_execute_end(task); -#endif MSG_RETURN(status); } @@ -129,17 +143,7 @@ msg_error_t MSG_process_sleep(double nb_sec) msg_error_t status = MSG_OK; /*msg_process_t proc = MSG_process_self();*/ -#ifdef HAVE_TRACING TRACE_msg_process_sleep_in(MSG_process_self()); -#endif - - /* create action to sleep */ - - /*proc->simdata->waiting_action = act_sleep; - - FIXME: check if not setting the waiting_action breaks something on msg - - proc->simdata->waiting_action = NULL;*/ TRY { simcall_process_sleep(nb_sec); @@ -147,6 +151,13 @@ msg_error_t MSG_process_sleep(double nb_sec) CATCH(e) { switch (e.category) { case cancel_error: + XBT_DEBUG("According to the JAVA API, a sleep call should only deal with HostFailureException, WTF here ?"); + // adsein: MSG_TASK_CANCELED is assigned when someone kills the process that made the sleep, this is not + // correct. For instance, when the node is turned off, the error should be MSG_HOST_FAILURE, which is by the way + // and according to the JAVA document, the only exception that can be triggered by MSG_Process_sleep call. + // To avoid possible impacts in the code, I just raised a host_failure exception for the moment in the JAVA code + // and did not change anythings at the C level. + // See comment in the jmsg_process.c file, function JNIEXPORT void JNICALL Java_org_simgrid_msg_Process_sleep(JNIEnv *env, jclass cls, jlong jmillis, jint jnanos) status = MSG_TASK_CANCELED; break; default: @@ -155,9 +166,7 @@ msg_error_t MSG_process_sleep(double nb_sec) xbt_ex_free(e); } - #ifdef HAVE_TRACING - TRACE_msg_process_sleep_out(MSG_process_self()); - #endif + TRACE_msg_process_sleep_out(MSG_process_self()); MSG_RETURN(status); } @@ -214,7 +223,7 @@ MSG_task_receive_from_host(msg_task_t * task, const char *alias, */ msg_error_t MSG_task_receive_from_host_bounded(msg_task_t * task, const char *alias, - msg_host_t host, double rate) + msg_host_t host, double rate) { return MSG_task_receive_ext_bounded(task, alias, -1, host, rate); } @@ -249,7 +258,8 @@ msg_error_t MSG_task_receive(msg_task_t * task, const char *alias) * #MSG_OK if the task was successfully received, * #MSG_HOST_FAILURE, or #MSG_TRANSFER_FAILURE otherwise. */ -msg_error_t MSG_task_receive_bounded(msg_task_t * task, const char *alias, double rate) +msg_error_t MSG_task_receive_bounded(msg_task_t * task, const char *alias, + double rate) { return MSG_task_receive_with_timeout_bounded(task, alias, -1, rate); } @@ -291,9 +301,9 @@ MSG_task_receive_with_timeout(msg_task_t * task, const char *alias, */ msg_error_t MSG_task_receive_with_timeout_bounded(msg_task_t * task, const char *alias, - double timeout,double rate) + double timeout,double rate) { - return MSG_task_receive_ext_bounded(task, alias, timeout, NULL,rate); + return MSG_task_receive_ext_bounded(task, alias, timeout, NULL, rate); } /** \ingroup msg_task_usage @@ -328,6 +338,9 @@ MSG_task_receive_ext(msg_task_t * task, const char *alias, double timeout, } CATCH(e) { switch (e.category) { + case host_error: + ret = MSG_HOST_FAILURE; + break; case cancel_error: /* may be thrown by MSG_mailbox_get_by_alias */ ret = MSG_HOST_FAILURE; break; @@ -355,15 +368,79 @@ MSG_task_receive_ext(msg_task_t * task, const char *alias, double timeout, */ msg_error_t MSG_task_receive_ext_bounded(msg_task_t * task, const char *alias, double timeout, - msg_host_t host, double rate) + msg_host_t host, double rate) { XBT_DEBUG ("MSG_task_receive_ext: Trying to receive a message on mailbox '%s'", alias); return MSG_mailbox_get_task_ext_bounded(MSG_mailbox_get_by_alias(alias), task, - host, timeout, rate); + host, timeout, rate); } +/* Internal function used to factorize code between + * MSG_task_isend_with_matching() and MSG_task_dsend(). + */ +static XBT_INLINE +msg_comm_t MSG_task_isend_internal(msg_task_t task, const char *alias, + int (*match_fun)(void*,void*, smx_synchro_t), + void *match_data, void_f_pvoid_t cleanup, + int detached) +{ + simdata_task_t t_simdata = NULL; + msg_process_t process = MSG_process_self(); + msg_mailbox_t mailbox = MSG_mailbox_get_by_alias(alias); + int call_end = TRACE_msg_task_put_start(task); + + /* Prepare the task to send */ + t_simdata = task->simdata; + t_simdata->sender = process; + t_simdata->source = ((simdata_process_t) SIMIX_process_self_get_data(process))->m_host; + + if (t_simdata->isused != 0) { + if (msg_global->debug_multiple_use){ + XBT_ERROR("This task is already used in there:"); + xbt_backtrace_display(t_simdata->isused); + XBT_ERROR("And you try to reuse it from here:"); + xbt_backtrace_display_current(); + } else { + xbt_assert(t_simdata->isused == 0, + "This task is still being used somewhere else. You cannot send it now. Go fix your code! (use --cfg=msg/debug_multiple_use:on to get the backtrace of the other process)"); + } + } + + if (msg_global->debug_multiple_use) + MSG_BT(t_simdata->isused, "Using Backtrace"); + else + t_simdata->isused = (void*)1; + t_simdata->comm = NULL; + msg_global->sent_msg++; + + /* Send it by calling SIMIX network layer */ + smx_synchro_t act = simcall_comm_isend(SIMIX_process_self(), mailbox, t_simdata->bytes_amount, + t_simdata->rate, task, sizeof(void *), + match_fun, cleanup, NULL, match_data,detached); + t_simdata->comm = act; /* FIXME: is the field t_simdata->comm still useful? */ + + msg_comm_t comm; + if (detached) { + comm = NULL; + } else { + comm = xbt_new0(s_msg_comm_t, 1); + comm->task_sent = task; + comm->task_received = NULL; + comm->status = MSG_OK; + comm->s_comm = act; + } + + if (TRACE_is_enabled()) + simcall_set_category(act, task->category); + if (call_end) + TRACE_msg_task_put_end(); + + return comm; +} + + /** \ingroup msg_task_usage * \brief Sends a task on a mailbox. * @@ -376,7 +453,7 @@ MSG_task_receive_ext_bounded(msg_task_t * task, const char *alias, double timeou */ msg_comm_t MSG_task_isend(msg_task_t task, const char *alias) { - return MSG_task_isend_with_matching(task,alias,NULL,NULL); + return MSG_task_isend_internal(task, alias, NULL, NULL, NULL, 0); } /** \ingroup msg_task_usage @@ -391,10 +468,11 @@ msg_comm_t MSG_task_isend(msg_task_t task, const char *alias) * \param maxrate the maximum communication rate for sending this task . * \return the msg_comm_t communication created */ -msg_comm_t MSG_task_isend_bounded(msg_task_t task, const char *alias, double maxrate) +msg_comm_t MSG_task_isend_bounded(msg_task_t task, const char *alias, + double maxrate) { task->simdata->rate = maxrate; - return MSG_task_isend_with_matching(task,alias,NULL,NULL); + return MSG_task_isend_internal(task, alias, NULL, NULL, NULL, 0); } @@ -409,55 +487,16 @@ msg_comm_t MSG_task_isend_bounded(msg_task_t task, const char *alias, double max * \param match_fun boolean function which parameters are: * - match_data_provided_here * - match_data_provided_by_other_side_if_any - * - the_smx_action_describing_the_other_side + * - the_smx_synchro_describing_the_other_side * \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(msg_task_t task, const char *alias, - int (*match_fun)(void*,void*, smx_action_t), - void *match_data) +msg_comm_t MSG_task_isend_with_matching(msg_task_t task, const char *alias, + int (*match_fun)(void*, void*, + smx_synchro_t), + void *match_data) { - simdata_task_t t_simdata = NULL; - msg_process_t process = MSG_process_self(); - msg_mailbox_t mailbox = MSG_mailbox_get_by_alias(alias); - -#ifdef HAVE_TRACING - int call_end = TRACE_msg_task_put_start(task); -#endif - - /* Prepare the task to send */ - t_simdata = task->simdata; - t_simdata->sender = process; - t_simdata->source = ((simdata_process_t) SIMIX_process_self_get_data(process))->m_host; - - 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->comm = NULL; - 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 = - simcall_comm_isend(mailbox, t_simdata->message_size, - t_simdata->rate, task, sizeof(void *), match_fun, NULL, match_data, 0); - t_simdata->comm = comm->s_comm; /* FIXME: is the field t_simdata->comm still useful? */ -#ifdef HAVE_TRACING - if (TRACE_is_enabled()) { - simcall_set_category(comm->s_comm, task->category); - } -#endif - -#ifdef HAVE_TRACING - if (call_end) - TRACE_msg_task_put_end(); -#endif - - return comm; + return MSG_task_isend_internal(task, alias, match_fun, match_data, NULL, 0); } /** \ingroup msg_task_usage @@ -479,43 +518,9 @@ XBT_INLINE msg_comm_t MSG_task_isend_with_matching(msg_task_t task, const char * */ void MSG_task_dsend(msg_task_t task, const char *alias, void_f_pvoid_t cleanup) { - simdata_task_t t_simdata = NULL; - msg_process_t process = MSG_process_self(); - msg_mailbox_t mailbox = MSG_mailbox_get_by_alias(alias); - - /* Prepare the task to send */ - t_simdata = task->simdata; - t_simdata->sender = process; - t_simdata->source = ((simdata_process_t) SIMIX_process_self_get_data(process))->m_host; - - 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->comm = NULL; - msg_global->sent_msg++; - -#ifdef HAVE_TRACING - int call_end = TRACE_msg_task_put_start(task); -#endif - - /* Send it by calling SIMIX network layer */ - smx_action_t comm = simcall_comm_isend(mailbox, t_simdata->message_size, - t_simdata->rate, task, sizeof(void *), NULL, cleanup, NULL, 1); - t_simdata->comm = comm; -#ifdef HAVE_TRACING - if (TRACE_is_enabled()) { - simcall_set_category(comm, task->category); - } -#endif - -#ifdef HAVE_TRACING - if (call_end) - TRACE_msg_task_put_end(); -#endif + MSG_task_isend_internal(task, alias, NULL, NULL, cleanup, 1); } - /** \ingroup msg_task_usage * \brief Sends a task on a mailbox with a maximal rate. * @@ -533,46 +538,13 @@ void MSG_task_dsend(msg_task_t task, const char *alias, void_f_pvoid_t cleanup) * communication fails, e.g. MSG_task_destroy * (if NULL, no function will be called) * \param maxrate the maximum communication rate for sending this task - * + * */ -void MSG_task_dsend_bounded(msg_task_t task, const char *alias, void_f_pvoid_t cleanup, double maxrate) +void MSG_task_dsend_bounded(msg_task_t task, const char *alias, + void_f_pvoid_t cleanup, double maxrate) { task->simdata->rate = maxrate; - - simdata_task_t t_simdata = NULL; - msg_process_t process = MSG_process_self(); - msg_mailbox_t mailbox = MSG_mailbox_get_by_alias(alias); - - /* Prepare the task to send */ - t_simdata = task->simdata; - t_simdata->sender = process; - t_simdata->source = ((simdata_process_t) SIMIX_process_self_get_data(process))->m_host; - - 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->comm = NULL; - msg_global->sent_msg++; - -#ifdef HAVE_TRACING - int call_end = TRACE_msg_task_put_start(task); -#endif - - /* Send it by calling SIMIX network layer */ - smx_action_t comm = simcall_comm_isend(mailbox, t_simdata->message_size, - t_simdata->rate, task, sizeof(void *), NULL, cleanup, NULL, 1); - t_simdata->comm = comm; -#ifdef HAVE_TRACING - if (TRACE_is_enabled()) { - simcall_set_category(comm, task->category); - } -#endif - -#ifdef HAVE_TRACING - if (call_end) - TRACE_msg_task_put_end(); -#endif + MSG_task_dsend(task, alias, cleanup); } /** \ingroup msg_task_usage @@ -580,32 +552,14 @@ void MSG_task_dsend_bounded(msg_task_t task, const char *alias, void_f_pvoid_t c * * This is a non blocking function: use MSG_comm_wait() or MSG_comm_test() * to end the communication. - * + * * \param task a memory location for storing a #msg_task_t. has to be valid until the end of the communication. * \param name of the mailbox to receive the task on * \return the msg_comm_t communication created */ msg_comm_t MSG_task_irecv(msg_task_t *task, const char *name) { - smx_rdv_t rdv = MSG_mailbox_get_by_alias(name); - - /* FIXME: these functions are not traceable */ - - /* Sanity check */ - xbt_assert(task, "Null pointer for the task storage"); - - if (*task) - XBT_CRITICAL - ("MSG_task_irecv() was asked to write in a non empty task struct."); - - /* Try to receive it by calling SIMIX network layer */ - 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 = simcall_comm_irecv(rdv, task, NULL, NULL, NULL); - - return comm; + return MSG_task_irecv_bounded(task, name, -1.0); } /** \ingroup msg_task_usage @@ -617,10 +571,9 @@ msg_comm_t MSG_task_irecv(msg_task_t *task, const char *name) * \param rate limit the bandwidth to the given rate * \return the msg_comm_t communication created */ -msg_comm_t MSG_task_irecv_bounded(msg_task_t *task, const char *name, double rate) +msg_comm_t MSG_task_irecv_bounded(msg_task_t *task, const char *name, + double rate) { - - smx_rdv_t rdv = MSG_mailbox_get_by_alias(name); /* FIXME: these functions are not traceable */ @@ -637,7 +590,7 @@ msg_comm_t MSG_task_irecv_bounded(msg_task_t *task, const char *name, double rat comm->task_sent = NULL; comm->task_received = task; comm->status = MSG_OK; - comm->s_comm = simcall_comm_irecv_bounded(rdv, task, NULL, NULL, NULL, rate); + comm->s_comm = simcall_comm_irecv(rdv, task, NULL, NULL, NULL, NULL, rate); return comm; } @@ -660,6 +613,8 @@ int MSG_comm_test(msg_comm_t comm) if (finished && comm->task_received != NULL) { /* I am the receiver */ + if (msg_global->debug_multiple_use && (*comm->task_received)->simdata->isused!=0) + xbt_ex_free(*(xbt_ex_t*)(*comm->task_received)->simdata->isused); (*comm->task_received)->simdata->isused = 0; } } @@ -697,7 +652,7 @@ int MSG_comm_testany(xbt_dynar_t comms) int finished_index = -1; /* create the equivalent dynar with SIMIX objects */ - xbt_dynar_t s_comms = xbt_dynar_new(sizeof(smx_action_t), NULL); + xbt_dynar_t s_comms = xbt_dynar_new(sizeof(smx_synchro_t), NULL); msg_comm_t comm; unsigned int cursor; xbt_dynar_foreach(comms, cursor, comm) { @@ -734,6 +689,8 @@ int MSG_comm_testany(xbt_dynar_t comms) if (status == MSG_OK && comm->task_received != NULL) { /* I am the receiver */ + if (msg_global->debug_multiple_use && (*comm->task_received)->simdata->isused!=0) + xbt_ex_free(*(xbt_ex_t*)(*comm->task_received)->simdata->isused); (*comm->task_received)->simdata->isused = 0; } } @@ -755,7 +712,7 @@ void MSG_comm_destroy(msg_comm_t comm) * * It takes two parameters. * \param comm the communication to wait. - * \param timeout Wait until the communication terminates or the timeout + * \param timeout Wait until the communication terminates or the timeout * occurs. You can provide a -1 timeout to obtain an infinite timeout. * \return msg_error_t */ @@ -767,6 +724,8 @@ msg_error_t MSG_comm_wait(msg_comm_t comm, double timeout) if (comm->task_received != NULL) { /* I am the receiver */ + if (msg_global->debug_multiple_use && (*comm->task_received)->simdata->isused!=0) + xbt_ex_free(*(xbt_ex_t*)(*comm->task_received)->simdata->isused); (*comm->task_received)->simdata->isused = 0; } @@ -816,7 +775,7 @@ int MSG_comm_waitany(xbt_dynar_t comms) int finished_index = -1; /* create the equivalent dynar with SIMIX objects */ - xbt_dynar_t s_comms = xbt_dynar_new(sizeof(smx_action_t), NULL); + xbt_dynar_t s_comms = xbt_dynar_new(sizeof(smx_synchro_t), NULL); msg_comm_t comm; unsigned int cursor; xbt_dynar_foreach(comms, cursor, comm) { @@ -854,6 +813,8 @@ int MSG_comm_waitany(xbt_dynar_t comms) if (comm->task_received != NULL) { /* I am the receiver */ + if (msg_global->debug_multiple_use && (*comm->task_received)->simdata->isused!=0) + xbt_ex_free(*(xbt_ex_t*)(*comm->task_received)->simdata->isused); (*comm->task_received)->simdata->isused = 0; } @@ -891,7 +852,7 @@ msg_task_t MSG_comm_get_task(msg_comm_t comm) * \param buff the data copied * \param buff_size size of the buffer */ -void MSG_comm_copy_data_from_SIMIX(smx_action_t comm, void* buff, size_t buff_size) { +void MSG_comm_copy_data_from_SIMIX(smx_synchro_t comm, void* buff, size_t buff_size) { // copy the task SIMIX_comm_copy_pointer_callback(comm, buff, buff_size); @@ -1058,9 +1019,7 @@ int MSG_task_listen_from(const char *alias) */ void MSG_task_set_category (msg_task_t task, const char *category) { -#ifdef HAVE_TRACING TRACE_msg_set_task_category (task, category); -#endif } /** \ingroup msg_task_usage @@ -1075,11 +1034,7 @@ void MSG_task_set_category (msg_task_t task, const char *category) */ const char *MSG_task_get_category (msg_task_t task) { -#ifdef HAVE_TRACING return task->category; -#else - return NULL; -#endif } /** @@ -1214,7 +1169,7 @@ MSG_task_put_with_timeout(msg_task_t task, msg_host_t dest, && (channel < msg_global->max_channel), "Invalid channel %d", channel); - XBT_DEBUG("MSG_task_put_with_timout: Trying to send a task to '%s'", SIMIX_host_get_name(dest->smx_host)); + XBT_DEBUG("MSG_task_put_with_timout: Trying to send a task to '%s'", MSG_host_get_name(dest)); return MSG_mailbox_put_with_timeout(MSG_mailbox_get_by_channel (dest, channel), task, timeout);