X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/8787f75c9d14a09fa99c40380a579b2b8372b40a..93a44592f2e075710723814e39fc3ec3fe93888b:/src/smpi/smpi_global.c diff --git a/src/smpi/smpi_global.c b/src/smpi/smpi_global.c index d9da58990c..36ecab65b9 100644 --- a/src/smpi/smpi_global.c +++ b/src/smpi/smpi_global.c @@ -1,321 +1,186 @@ -#include +#include #include "private.h" +#include "smpi_mpi_dt_private.h" XBT_LOG_NEW_CATEGORY(smpi, "All SMPI categories"); -XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_kernel, smpi, "Logging specific to SMPI (kernel)"); +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; + xbt_fifo_t pending_sent; + xbt_fifo_t pending_recv; + xbt_os_timer_t timer; +} 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); - - request->buf = NULL; - request->completed = 0; - request->mutex = SIMIX_mutex_init(); - request->cond = SIMIX_cond_init(); - request->data = NULL; - request->forward = 0; +MPI_Comm MPI_COMM_WORLD = MPI_COMM_NULL; - return request; +smpi_process_data_t smpi_process_data(void) { + return SIMIX_process_get_data(SIMIX_process_self()); } -void smpi_request_free(void *pointer); - -void smpi_request_free(void *pointer) -{ - - smpi_mpi_request_t request = pointer; - - SIMIX_cond_destroy(request->cond); - SIMIX_mutex_destroy(request->mutex); - xbt_free(request); - - return; +smpi_process_data_t smpi_process_remote_data(int index) { + return process_data[index]; } -void smpi_request_reset(void *pointer); - -void smpi_request_reset(void *pointer) -{ - smpi_mpi_request_t request = pointer; - - request->buf = NULL; - request->completed = 0; - request->data = NULL; - request->forward = 0; - - return; +int smpi_process_count(void) { + return process_count; } +int smpi_process_index(void) { + smpi_process_data_t data = smpi_process_data(); -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; -} - -void smpi_message_free(void *pointer); - -void smpi_message_free(void *pointer) -{ - xbt_free(pointer); - return; -} - -void smpi_message_reset(void *pointer); - -void smpi_message_reset(void *pointer) -{ - smpi_received_message_t message = pointer; - message->buf = NULL; - return; -} - -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 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 data->index; } -void smpi_global_init() -{ - int i; - - int size = SIMIX_host_get_number(); - - 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(); - } +xbt_os_timer_t smpi_process_timer(void) { + smpi_process_data_t data = smpi_process_data(); + return data->timer; } -void smpi_global_destroy() -{ - 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); - - // 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); - - for(curr = smpi_global->do_once_duration_nodes; NULL != curr; curr = next) { - next = curr->next; - xbt_free(curr->file); - xbt_free(curr); - } - - SIMIX_mutex_destroy(smpi_global->do_once_mutex); - - 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]); - } - - 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); - - xbt_free(smpi_global); - - smpi_global = NULL; +void smpi_process_post_send(MPI_Comm comm, MPI_Request request) { + int index = smpi_group_index(smpi_comm_group(comm), request->dst); + smpi_process_data_t data = smpi_process_remote_data(index); + xbt_fifo_item_t item; + MPI_Request req; + + DEBUG4("isend for request %p [src = %d, dst = %d, tag = %d]", + request, request->src, request->dst, request->tag); + xbt_fifo_foreach(data->pending_recv, item, req, MPI_Request) { + if(req->comm == request->comm + && (req->src == MPI_ANY_SOURCE || req->src == request->src) + && (req->tag == MPI_ANY_TAG || req->tag == request->tag)){ + DEBUG4("find matching request %p [src = %d, dst = %d, tag = %d]", + req, req->src, req->dst, req->tag); + xbt_fifo_remove_item(data->pending_recv, item); + /* Materialize the *_ANY_* fields from corresponding irecv request */ + req->src = request->src; + req->tag = request->tag; + req->data = request->data; + request->rdv = req->rdv; + return; + } else { + DEBUG4("not matching request %p [src = %d, dst = %d, tag = %d]", + req, req->src, req->dst, req->tag); + } + } + request->rdv = SIMIX_rdv_create(NULL); + xbt_fifo_push(data->pending_sent, request); } -int smpi_host_index() -{ - smx_host_t host = SIMIX_host_self(); - smpi_host_data_t hdata = (smpi_host_data_t)SIMIX_host_get_data(host); - return hdata->index; +void smpi_process_post_recv(MPI_Request request) { + smpi_process_data_t data = smpi_process_data(); + xbt_fifo_item_t item; + MPI_Request req; + + DEBUG4("irecv for request %p [src = %d, dst = %d, tag = %d]", + request, request->src, request->dst, request->tag); + xbt_fifo_foreach(data->pending_sent, item, req, MPI_Request) { + if(req->comm == request->comm + && (request->src == MPI_ANY_SOURCE || req->src == request->src) + && (request->tag == MPI_ANY_TAG || req->tag == request->tag)){ + DEBUG4("find matching request %p [src = %d, dst = %d, tag = %d]", + req, req->src, req->dst, req->tag); + xbt_fifo_remove_item(data->pending_sent, item); + /* Materialize the *_ANY_* fields from the irecv request */ + request->src = req->src; + request->tag = req->tag; + request->data = req->data; + request->rdv = req->rdv; + return; + } else { + DEBUG4("not matching request %p [src = %d, dst = %d, tag = %d]", + req, req->src, req->dst, req->tag); + } + } + request->rdv = SIMIX_rdv_create(NULL); + xbt_fifo_push(data->pending_recv, request); } -smx_mutex_t smpi_host_mutex() -{ - smx_host_t host = SIMIX_host_self(); - smpi_host_data_t hdata = (smpi_host_data_t)SIMIX_host_get_data(host); - return hdata->mutex; +void smpi_global_init(void) { + int i; + MPI_Group group; + + SIMIX_network_set_copy_data_callback(&SIMIX_network_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]->pending_sent = xbt_fifo_new(); + process_data[i]->pending_recv = xbt_fifo_new(); + process_data[i]->timer = xbt_os_timer_new(); + } + 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() -{ - smx_host_t host = SIMIX_host_self(); - smpi_host_data_t hdata = (smpi_host_data_t)SIMIX_host_get_data(host); - return hdata->cond; +void smpi_global_destroy(void) { + int count = smpi_process_count(); + int i; + + smpi_comm_destroy(MPI_COMM_WORLD); + MPI_COMM_WORLD = MPI_COMM_NULL; + for(i = 0; i < count; i++) { + xbt_os_timer_free(process_data[i]->timer); + xbt_fifo_free(process_data[i]->pending_recv); + xbt_fifo_free(process_data[i]->pending_sent); + xbt_free(process_data[i]); + } + xbt_free(process_data); + process_data = NULL; } -int smpi_run_simulation(int *argc, char **argv) +int main(int argc, char **argv) { - smx_cond_t cond = NULL; - smx_action_t action = NULL; - - xbt_fifo_t actions_failed = xbt_fifo_new(); - xbt_fifo_t actions_done = xbt_fifo_new(); - - srand(SMPI_RAND_SEED); - - SIMIX_global_init(argc, argv); - - SIMIX_function_register("smpi_simulated_main", smpi_simulated_main); - SIMIX_function_register("smpi_sender", smpi_sender); - SIMIX_function_register("smpi_receiver", smpi_receiver); + srand(SMPI_RAND_SEED); - // FIXME: ought to verify these files... - SIMIX_create_environment(argv[1]); + double default_reference_speed = 20000.0; + xbt_cfg_register(&_surf_cfg_set, "reference_speed", + "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); - // must initialize globals between creating environment and launching app.... - smpi_global_init(); + int default_display_timing = 0; + xbt_cfg_register(&_surf_cfg_set, "display_timing", + "Boolean indicating whether we should display the timing after simulation.", + xbt_cfgelm_int, &default_display_timing, 1, 1, NULL, NULL); - SIMIX_launch_application(argv[2]); + int default_display_smpe = 0; + xbt_cfg_register(&_surf_cfg_set, "SMPE", + "Boolean indicating whether we should display simulated time spent in MPI calls.", + xbt_cfgelm_int, &default_display_smpe, 1, 1, NULL, NULL); - /* Prepare to display some more info when dying on Ctrl-C pressing */ - // FIXME: doesn't work - //signal(SIGINT, inthandler); + SIMIX_global_init(&argc, argv); - /* Clean IO before the run */ - fflush(stdout); - fflush(stderr); + // parse the platform file: get the host list + SIMIX_create_environment(argv[1]); - 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); - } - } - } + SIMIX_function_register("smpi_simulated_main", smpi_simulated_main); + SIMIX_launch_application(argv[2]); - // FIXME: cleanup incomplete - xbt_fifo_free(actions_failed); - xbt_fifo_free(actions_done); + smpi_global_init(); - INFO1("simulation time %g", SIMIX_get_clock()); + /* Clean IO before the run */ + fflush(stdout); + fflush(stderr); + SIMIX_init(); - smpi_global_destroy(); + while (SIMIX_solve(NULL, NULL) != -1.0); + + if (xbt_cfg_get_int(_surf_cfg_set, "display_timing")) + INFO1("simulation time %g", SIMIX_get_clock()); - SIMIX_clean(); + smpi_global_destroy(); - return 0; + SIMIX_clean(); + return 0; }