X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/ebd53cac243186b83457179668123ae1ddbd47a2..d73cc63ed2740fe67bf9c0ada4e70ca826600963:/src/smpi/smpi_global.c diff --git a/src/smpi/smpi_global.c b/src/smpi/smpi_global.c index 4dac16053e..1b98966471 100644 --- a/src/smpi/smpi_global.c +++ b/src/smpi/smpi_global.c @@ -1,377 +1,297 @@ -#include +/* Copyright (c) 2007, 2008, 2009, 2010. 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. */ + +#include +#include #include "private.h" +#include "smpi_mpi_dt_private.h" +#include "mc/mc.h" +#include "surf/surf.h" XBT_LOG_NEW_CATEGORY(smpi, "All SMPI categories"); XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_kernel, smpi, "Logging specific to SMPI (kernel)"); -smpi_global_t smpi_global = NULL; +typedef struct s_smpi_process_data { + int index; + int* argc; + char*** argv; + smx_rdv_t mailbox; + xbt_os_timer_t timer; + double simulated; + MPI_Comm comm_self; +} s_smpi_process_data_t; -void *smpi_request_new(void); +static smpi_process_data_t *process_data = NULL; +static int process_count = 0; -void *smpi_request_new() -{ - smpi_mpi_request_t request = xbt_new(s_smpi_mpi_request_t, 1); +MPI_Comm MPI_COMM_WORLD = MPI_COMM_NULL; - request->buf = NULL; - request->completed = 0; - request->mutex = SIMIX_mutex_init(); - request->cond = SIMIX_cond_init(); - request->data = NULL; - request->forward = 0; +#define MAILBOX_NAME_MAXLEN (5 + sizeof(int) * 2 + 1) - return request; +static char* get_mailbox_name(char* str, int index) { + snprintf(str, MAILBOX_NAME_MAXLEN, "SMPI-%0*x", (int)(sizeof(int) * 2), index); + return str; } -void smpi_request_free(void *pointer); - -void smpi_request_free(void *pointer) +void smpi_process_init(int *argc, char ***argv) { + int index; + smpi_process_data_t data; + smx_process_t proc; + + if(argc && argv) { + proc = SIMIX_process_self(); + index = atoi((*argv)[1]); + data = smpi_process_remote_data(index); + SIMIX_req_process_set_data(proc, data); + if (*argc > 2) { + free((*argv)[1]); + memmove(&(*argv)[1], &(*argv)[2], sizeof(char *) * (*argc - 2)); + (*argv)[(*argc) - 1] = NULL; + } + (*argc)--; + data->argc = argc; + data->argv = argv; + XBT_DEBUG("<%d> New process in the game: %p", index, proc); + } +} - smpi_mpi_request_t request = pointer; - - SIMIX_cond_destroy(request->cond); - SIMIX_mutex_destroy(request->mutex); - xbt_free(request); +void smpi_process_destroy(void) +{ + int index = smpi_process_index(); - return; + XBT_DEBUG("<%d> Process left the game", index); } -void smpi_request_reset(void *pointer); +int smpi_process_argc(void) { + smpi_process_data_t data = smpi_process_data(); -void smpi_request_reset(void *pointer) -{ - smpi_mpi_request_t request = pointer; + return data->argc ? *(data->argc) - 1 : 0; +} - request->buf = NULL; - request->completed = 0; - request->data = NULL; - request->forward = 0; +int smpi_process_getarg(integer* index, char* dst, ftnlen len) { + smpi_process_data_t data = smpi_process_data(); + char* arg; + ftnlen i; - return; + if(!data->argc || !data->argv + || *index < 1 || *index >= *(data->argc)) { + return -1; + } + arg = (*data->argv)[*index]; + for(i = 0; i < len && arg[i] != '\0'; i++) { + dst[i] = arg[i]; + } + for(; i < len; i++) { + dst[i] = ' '; + } + return 0; } +int smpi_global_size(void) { + char* value = getenv("SMPI_GLOBAL_SIZE"); -void *smpi_message_new(void); - -void *smpi_message_new() -{ - smpi_received_message_t message = xbt_new(s_smpi_received_message_t, 1); - message->buf = NULL; - return message; + if(!value) { + abort(); + } + return atoi(value); } -void smpi_message_free(void *pointer); - -void smpi_message_free(void *pointer) +smpi_process_data_t smpi_process_data(void) { - xbt_free(pointer); - return; + return SIMIX_process_self_get_data(); } -void smpi_message_reset(void *pointer); - -void smpi_message_reset(void *pointer) +smpi_process_data_t smpi_process_remote_data(int index) { - smpi_received_message_t message = pointer; - message->buf = NULL; - return; + return process_data[index]; } -int smpi_create_request(void *buf, int count, smpi_mpi_datatype_t datatype, - int src, int dst, int tag, - smpi_mpi_communicator_t comm, - smpi_mpi_request_t * requestptr) +int smpi_process_count(void) { - int retval = MPI_SUCCESS; - - smpi_mpi_request_t request = NULL; - - // parameter checking prob belongs in smpi_mpi, but this is less repeat code - if (NULL == buf) { - retval = MPI_ERR_INTERN; - } else if (0 > count) { - retval = MPI_ERR_COUNT; - } else if (NULL == datatype) { - retval = MPI_ERR_TYPE; - } else if (MPI_ANY_SOURCE != src && (0 > src || comm->size <= src)) { - retval = MPI_ERR_RANK; - } else if (0 > dst || comm->size <= dst) { - retval = MPI_ERR_RANK; - } else if (MPI_ANY_TAG != tag && 0 > tag) { - retval = MPI_ERR_TAG; - } else if (NULL == comm) { - retval = MPI_ERR_COMM; - } else if (NULL == requestptr) { - retval = MPI_ERR_ARG; - } else { - request = xbt_mallocator_get(smpi_global->request_mallocator); - request->comm = comm; - request->src = src; - request->dst = dst; - request->tag = tag; - request->buf = buf; - request->datatype = datatype; - request->count = count; - - *requestptr = request; - } - return retval; + return process_count; } -/* FIXME: understand what they do and put the prototypes in a header file (live in smpi_base.c) */ -void smpi_mpi_land_func(void *a, void *b, int *length, - MPI_Datatype * datatype); -void smpi_mpi_sum_func(void *a, void *b, int *length, - MPI_Datatype * datatype); -void smpi_global_init() +int smpi_process_index(void) { - int i; - - int size = SIMIX_host_get_number(); - - /* Connect our log channels: that must be done manually under windows */ -#ifdef XBT_LOG_CONNECT - XBT_LOG_CONNECT(smpi_base, smpi); - XBT_LOG_CONNECT(smpi_bench, smpi); - XBT_LOG_CONNECT(smpi_kernel, smpi); - XBT_LOG_CONNECT(smpi_mpi, smpi); - XBT_LOG_CONNECT(smpi_receiver, smpi); - XBT_LOG_CONNECT(smpi_sender, smpi); - XBT_LOG_CONNECT(smpi_util, smpi); -#endif + smpi_process_data_t data = smpi_process_data(); - smpi_global = xbt_new(s_smpi_global_t, 1); - // config variable - smpi_global->reference_speed = SMPI_DEFAULT_SPEED; - - smpi_global->root_ready = 0; - smpi_global->ready_process_count = 0; - - // start/stop - smpi_global->start_stop_mutex = SIMIX_mutex_init(); - smpi_global->start_stop_cond = SIMIX_cond_init(); - - // host info blank until sim starts - // FIXME: is this okay? - smpi_global->hosts = NULL; - smpi_global->host_count = 0; - - // running hosts - smpi_global->running_hosts_count_mutex = SIMIX_mutex_init(); - smpi_global->running_hosts_count = 0; - - // mallocators - smpi_global->request_mallocator = - xbt_mallocator_new(SMPI_REQUEST_MALLOCATOR_SIZE, smpi_request_new, - smpi_request_free, smpi_request_reset); - smpi_global->message_mallocator = - xbt_mallocator_new(SMPI_MESSAGE_MALLOCATOR_SIZE, smpi_message_new, - smpi_message_free, smpi_message_reset); - - // queues - smpi_global->pending_send_request_queues = xbt_new(xbt_fifo_t, size); - smpi_global->pending_send_request_queues_mutexes = - xbt_new(smx_mutex_t, size); - smpi_global->pending_recv_request_queues = xbt_new(xbt_fifo_t, size); - smpi_global->pending_recv_request_queues_mutexes = - xbt_new(smx_mutex_t, size); - smpi_global->received_message_queues = xbt_new(xbt_fifo_t, size); - smpi_global->received_message_queues_mutexes = xbt_new(smx_mutex_t, size); - - // sender/receiver processes - smpi_global->sender_processes = xbt_new(smx_process_t, size); - smpi_global->receiver_processes = xbt_new(smx_process_t, size); - - // timers - smpi_global->timer = xbt_os_timer_new(); - smpi_global->timer_mutex = SIMIX_mutex_init(); - smpi_global->timer_cond = SIMIX_cond_init(); - - smpi_global->do_once_duration_nodes = NULL; - smpi_global->do_once_duration = NULL; - smpi_global->do_once_mutex = SIMIX_mutex_init(); - - for (i = 0; i < size; i++) { - smpi_global->pending_send_request_queues[i] = xbt_fifo_new(); - smpi_global->pending_send_request_queues_mutexes[i] = SIMIX_mutex_init(); - smpi_global->pending_recv_request_queues[i] = xbt_fifo_new(); - smpi_global->pending_recv_request_queues_mutexes[i] = SIMIX_mutex_init(); - smpi_global->received_message_queues[i] = xbt_fifo_new(); - smpi_global->received_message_queues_mutexes[i] = SIMIX_mutex_init(); - } + return data->index; +} - smpi_global->hosts = SIMIX_host_get_table(); - smpi_global->host_count = SIMIX_host_get_number(); - - smpi_mpi_global = xbt_new(s_smpi_mpi_global_t, 1); - - // global communicator - smpi_mpi_global->mpi_comm_world = xbt_new(s_smpi_mpi_communicator_t, 1); - smpi_mpi_global->mpi_comm_world->size = smpi_global->host_count; - smpi_mpi_global->mpi_comm_world->barrier_count = 0; - smpi_mpi_global->mpi_comm_world->barrier_mutex = SIMIX_mutex_init(); - smpi_mpi_global->mpi_comm_world->barrier_cond = SIMIX_cond_init(); - smpi_mpi_global->mpi_comm_world->rank_to_index_map = - xbt_new(int, smpi_global->host_count); - smpi_mpi_global->mpi_comm_world->index_to_rank_map = - xbt_new(int, smpi_global->host_count); - for (i = 0; i < smpi_global->host_count; i++) { - smpi_mpi_global->mpi_comm_world->rank_to_index_map[i] = i; - smpi_mpi_global->mpi_comm_world->index_to_rank_map[i] = i; - } +smx_rdv_t smpi_process_mailbox(void) { + smpi_process_data_t data = smpi_process_data(); - // mpi datatypes - smpi_mpi_global->mpi_byte = xbt_new(s_smpi_mpi_datatype_t, 1); - smpi_mpi_global->mpi_byte->size = (size_t) 1; - smpi_mpi_global->mpi_int = xbt_new(s_smpi_mpi_datatype_t, 1); - smpi_mpi_global->mpi_int->size = sizeof(int); - smpi_mpi_global->mpi_double = xbt_new(s_smpi_mpi_datatype_t, 1); - smpi_mpi_global->mpi_double->size = sizeof(double); + return data->mailbox; +} - // mpi operations - smpi_mpi_global->mpi_land = xbt_new(s_smpi_mpi_op_t, 1); - smpi_mpi_global->mpi_land->func = smpi_mpi_land_func; - smpi_mpi_global->mpi_sum = xbt_new(s_smpi_mpi_op_t, 1); - smpi_mpi_global->mpi_sum->func = smpi_mpi_sum_func; +smx_rdv_t smpi_process_remote_mailbox(int index) { + smpi_process_data_t data = smpi_process_remote_data(index); + return data->mailbox; } -void smpi_global_destroy() +xbt_os_timer_t smpi_process_timer(void) { - int i; - - int size = SIMIX_host_get_number(); - - smpi_do_once_duration_node_t curr, next; - - // start/stop - SIMIX_mutex_destroy(smpi_global->start_stop_mutex); - SIMIX_cond_destroy(smpi_global->start_stop_cond); - - // processes - xbt_free(smpi_global->sender_processes); - xbt_free(smpi_global->receiver_processes); + smpi_process_data_t data = smpi_process_data(); - // running hosts - SIMIX_mutex_destroy(smpi_global->running_hosts_count_mutex); - - // mallocators - xbt_mallocator_free(smpi_global->request_mallocator); - xbt_mallocator_free(smpi_global->message_mallocator); - - xbt_os_timer_free(smpi_global->timer); - SIMIX_mutex_destroy(smpi_global->timer_mutex); - SIMIX_cond_destroy(smpi_global->timer_cond); + return data->timer; +} - for (curr = smpi_global->do_once_duration_nodes; NULL != curr; curr = next) { - next = curr->next; - xbt_free(curr->file); - xbt_free(curr); - } +void smpi_process_simulated_start(void) +{ + smpi_process_data_t data = smpi_process_data(); - SIMIX_mutex_destroy(smpi_global->do_once_mutex); + data->simulated = SIMIX_get_clock(); +} - for (i = 0; i < size; i++) { - xbt_fifo_free(smpi_global->pending_send_request_queues[i]); - SIMIX_mutex_destroy(smpi_global->pending_send_request_queues_mutexes[i]); - xbt_fifo_free(smpi_global->pending_recv_request_queues[i]); - SIMIX_mutex_destroy(smpi_global->pending_recv_request_queues_mutexes[i]); - xbt_fifo_free(smpi_global->received_message_queues[i]); - SIMIX_mutex_destroy(smpi_global->received_message_queues_mutexes[i]); - } +double smpi_process_simulated_elapsed(void) +{ + smpi_process_data_t data = smpi_process_data(); - xbt_free(smpi_global->pending_send_request_queues); - xbt_free(smpi_global->pending_send_request_queues_mutexes); - xbt_free(smpi_global->pending_recv_request_queues); - xbt_free(smpi_global->pending_recv_request_queues_mutexes); - xbt_free(smpi_global->received_message_queues); - xbt_free(smpi_global->received_message_queues_mutexes); + return SIMIX_get_clock() - data->simulated; +} - xbt_free(smpi_global); +MPI_Comm smpi_process_comm_self(void) +{ + smpi_process_data_t data = smpi_process_data(); - smpi_global = NULL; + return data->comm_self; } -int smpi_host_index() +void print_request(const char *message, MPI_Request request) { - smx_host_t host = SIMIX_host_self(); - smpi_host_data_t hdata = (smpi_host_data_t) SIMIX_host_get_data(host); - return hdata->index; + XBT_DEBUG("%s request %p [buf = %p, size = %zu, src = %d, dst = %d, tag = %d, flags = %x]", + message, request, request->buf, request->size, + request->src, request->dst, request->tag, request->flags); } -smx_mutex_t smpi_host_mutex() +void smpi_global_init(void) { - smx_host_t host = SIMIX_host_self(); - smpi_host_data_t hdata = (smpi_host_data_t) SIMIX_host_get_data(host); - return hdata->mutex; + int i; + MPI_Group group; + char name[MAILBOX_NAME_MAXLEN]; + + SIMIX_comm_set_copy_data_callback + (&SIMIX_comm_copy_buffer_callback); + process_count = SIMIX_process_count(); + process_data = xbt_new(smpi_process_data_t, process_count); + for (i = 0; i < process_count; i++) { + process_data[i] = xbt_new(s_smpi_process_data_t, 1); + process_data[i]->index = i; + process_data[i]->argc = NULL; + process_data[i]->argv = NULL; + process_data[i]->mailbox = SIMIX_req_rdv_create(get_mailbox_name(name, i)); + process_data[i]->timer = xbt_os_timer_new(); + group = smpi_group_new(1); + process_data[i]->comm_self = smpi_comm_new(group); + smpi_group_set_mapping(group, i, 0); + } + group = smpi_group_new(process_count); + MPI_COMM_WORLD = smpi_comm_new(group); + for (i = 0; i < process_count; i++) { + smpi_group_set_mapping(group, i, i); + } } -smx_cond_t smpi_host_cond() +void smpi_global_destroy(void) { - smx_host_t host = SIMIX_host_self(); - smpi_host_data_t hdata = (smpi_host_data_t) SIMIX_host_get_data(host); - return hdata->cond; + int count = smpi_process_count(); + int i; + + smpi_bench_destroy(); + smpi_comm_destroy(MPI_COMM_WORLD); + MPI_COMM_WORLD = MPI_COMM_NULL; + for (i = 0; i < count; i++) { + smpi_comm_destroy(process_data[i]->comm_self); + xbt_os_timer_free(process_data[i]->timer); + SIMIX_req_rdv_destroy(process_data[i]->mailbox); + xbt_free(process_data[i]); + } + xbt_free(process_data); + process_data = NULL; } -int smpi_run_simulation(int *argc, char **argv) -{ - smx_cond_t cond = NULL; - smx_action_t action = NULL; +/* Fortran specific stuff */ +/* With smpicc, the following weak symbols are used */ +/* With smpiff, the following weak symbols are replaced by those in libf2c */ +int __attribute__((weak)) xargc; +char** __attribute__((weak)) xargv; - xbt_fifo_t actions_failed = xbt_fifo_new(); - xbt_fifo_t actions_done = xbt_fifo_new(); +int __attribute__((weak)) main(int argc, char** argv) { + xargc = argc; + xargv = argv; + return MAIN__(); +} +int MAIN__(void) +{ srand(SMPI_RAND_SEED); - SIMIX_global_init(argc, argv); + double default_reference_speed = 20000.0; + xbt_cfg_register(&_surf_cfg_set, "smpi/running_power", + "Power of the host running the simulation (in flop/s). Used to bench the operations.", + xbt_cfgelm_double, &default_reference_speed, 1, 1, NULL, + NULL); + + int default_display_timing = 0; + xbt_cfg_register(&_surf_cfg_set, "smpi/display_timing", + "Boolean indicating whether we should display the timing after simulation.", + xbt_cfgelm_int, &default_display_timing, 1, 1, NULL, + NULL); + + double default_threshold = 1e-6; + xbt_cfg_register(&_surf_cfg_set, "smpi/cpu_threshold", + "Minimal computation time (in seconds) not discarded.", + xbt_cfgelm_double, &default_threshold, 1, 1, NULL, + NULL); + + if(getenv("SMPI_PRETEND_CC") != NULL) { + return 0; + } + +#ifdef HAVE_TRACING + TRACE_global_init(&xargc, xargv); +#endif + + SIMIX_global_init(&xargc, xargv); + +#ifdef HAVE_TRACING + TRACE_smpi_start(); +#endif // parse the platform file: get the host list - SIMIX_create_environment(argv[1]); + SIMIX_create_environment(xargv[1]); SIMIX_function_register("smpi_simulated_main", smpi_simulated_main); - SIMIX_function_register("smpi_sender", smpi_sender); - SIMIX_function_register("smpi_receiver", smpi_receiver); - SIMIX_launch_application(argv[2]); + SIMIX_launch_application(xargv[2]); - // must initialize globals between creating environment and launching app.... smpi_global_init(); - /* Prepare to display some more info when dying on Ctrl-C pressing */ - // FIXME: doesn't work - //signal(SIGINT, inthandler); - /* Clean IO before the run */ fflush(stdout); fflush(stderr); - SIMIX_init(); - - while (SIMIX_solve(actions_done, actions_failed) != -1.0) { - while ((action = xbt_fifo_pop(actions_failed))) { - DEBUG1("** %s failed **", action->name); - while ((cond = xbt_fifo_pop(action->cond_list))) { - SIMIX_cond_broadcast(cond); - } - } - while ((action = xbt_fifo_pop(actions_done))) { - DEBUG1("** %s done **", action->name); - while ((cond = xbt_fifo_pop(action->cond_list))) { - SIMIX_cond_broadcast(cond); - } - } - } - // FIXME: cleanup incomplete - xbt_fifo_free(actions_failed); - xbt_fifo_free(actions_done); + if (MC_IS_ENABLED) + MC_modelcheck(); + else + SIMIX_run(); - INFO1("simulation time %g", SIMIX_get_clock()); + if (xbt_cfg_get_int(_surf_cfg_set, "smpi/display_timing")) + XBT_INFO("simulation time %g", SIMIX_get_clock()); smpi_global_destroy(); - SIMIX_clean(); +#ifdef HAVE_TRACING + TRACE_smpi_release(); +#endif + SIMIX_clean(); return 0; }