src/smpi/smpi_global.c

   1 #include <stdio.h>
   2
   3 #include "private.h"
   4
   5 XBT_LOG_NEW_CATEGORY(smpi, "All SMPI categories");
   6
   7 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_kernel, smpi,
   8                                 "Logging specific to SMPI (kernel)");
   9
  10 smpi_global_t smpi_global = NULL;
  11
  12 void *smpi_request_new(void);
  13
  14 void *smpi_request_new()
  15 {
  16   smpi_mpi_request_t request = xbt_new(s_smpi_mpi_request_t, 1);
  17
  18   request->buf = NULL;
  19   request->completed = 0;
  20   request->mutex = SIMIX_mutex_init();
  21   request->cond = SIMIX_cond_init();
  22   request->data = NULL;
  23   request->forward = 0;
  24
  25   return request;
  26 }
  27
  28 void smpi_request_free(void *pointer);
  29
  30 void smpi_request_free(void *pointer)
  31 {
  32
  33   smpi_mpi_request_t request = pointer;
  34
  35   SIMIX_cond_destroy(request->cond);
  36   SIMIX_mutex_destroy(request->mutex);
  37   xbt_free(request);
  38
  39   return;
  40 }
  41
  42 void smpi_request_reset(void *pointer);
  43
  44 void smpi_request_reset(void *pointer)
  45 {
  46   smpi_mpi_request_t request = pointer;
  47
  48   request->buf = NULL;
  49   request->completed = 0;
  50   request->consumed = 0;
  51   request->data = NULL;
  52   request->forward = 0;
  53
  54   return;
  55 }
  56
  57
  58 void *smpi_message_new(void);
  59
  60 void *smpi_message_new()
  61 {
  62   smpi_received_message_t message = xbt_new(s_smpi_received_message_t, 1);
  63   message->buf = NULL;
  64   return message;
  65 }
  66
  67 void smpi_message_free(void *pointer);
  68
  69 void smpi_message_free(void *pointer)
  70 {
  71   xbt_free(pointer);
  72   return;
  73 }
  74
  75 void smpi_message_reset(void *pointer);
  76
  77 void smpi_message_reset(void *pointer)
  78 {
  79   smpi_received_message_t message = pointer;
  80   message->buf = NULL;
  81   return;
  82 }
  83
  84 int smpi_create_request(void *buf, int count, smpi_mpi_datatype_t datatype,
  85                         int src, int dst, int tag,
  86                         smpi_mpi_communicator_t comm,
  87                         smpi_mpi_request_t * requestptr)
  88 {
  89   int retval = MPI_SUCCESS;
  90
  91   smpi_mpi_request_t request = NULL;
  92
  93   // parameter checking prob belongs in smpi_mpi, but this is less repeat code
  94   if (NULL == buf) {
  95     retval = MPI_ERR_INTERN;
  96   } else if (0 > count) {
  97     retval = MPI_ERR_COUNT;
  98   } else if (NULL == datatype) {
  99     retval = MPI_ERR_TYPE;
 100   } else if (MPI_ANY_SOURCE != src && (0 > src || comm->size <= src)) {
 101     retval = MPI_ERR_RANK;
 102   } else if (0 > dst || comm->size <= dst) {
 103     retval = MPI_ERR_RANK;
 104   } else if (MPI_ANY_TAG != tag && 0 > tag) {
 105     retval = MPI_ERR_TAG;
 106   } else if (NULL == comm) {
 107     retval = MPI_ERR_COMM;
 108   } else if (NULL == requestptr) {
 109     retval = MPI_ERR_ARG;
 110   } else {
 111     request = xbt_mallocator_get(smpi_global->request_mallocator);
 112     request->comm = comm;
 113     request->src = src;
 114     request->dst = dst;
 115     request->tag = tag;
 116     request->buf = buf;
 117     request->datatype = datatype;
 118     request->count = count;
 119
 120     *requestptr = request;
 121   }
 122   return retval;
 123 }
 124
 125 /* FIXME: understand what they do and put the prototypes in a header file (live in smpi_base.c) */
 126 void smpi_mpi_land_func(void *a, void *b, int *length,
 127                         MPI_Datatype * datatype);
 128 void smpi_mpi_sum_func(void *a, void *b, int *length,
 129                        MPI_Datatype * datatype);
 130 void smpi_mpi_min_func(void *a, void *b, int *length,
 131                        MPI_Datatype * datatype);
 132 void smpi_mpi_max_func(void *a, void *b, int *length,
 133                        MPI_Datatype * datatype);
 134
 135 void smpi_global_init()
 136 {
 137   int i;
 138
 139   /* Connect our log channels: that must be done manually under windows */
 140 #ifdef XBT_LOG_CONNECT
 141   XBT_LOG_CONNECT(smpi_base, smpi);
 142   XBT_LOG_CONNECT(smpi_bench, smpi);
 143   XBT_LOG_CONNECT(smpi_kernel, smpi);
 144   XBT_LOG_CONNECT(smpi_mpi, smpi);
 145   XBT_LOG_CONNECT(smpi_receiver, smpi);
 146   XBT_LOG_CONNECT(smpi_sender, smpi);
 147   XBT_LOG_CONNECT(smpi_util, smpi);
 148 #endif
 149
 150   smpi_global = xbt_new(s_smpi_global_t, 1);
 151
 152   // mallocators
 153   smpi_global->request_mallocator =
 154     xbt_mallocator_new(SMPI_REQUEST_MALLOCATOR_SIZE, smpi_request_new,
 155                        smpi_request_free, smpi_request_reset);
 156   smpi_global->message_mallocator =
 157     xbt_mallocator_new(SMPI_MESSAGE_MALLOCATOR_SIZE, smpi_message_new,
 158                        smpi_message_free, smpi_message_reset);
 159
 160   smpi_global->process_count = SIMIX_process_count();
 161   DEBUG1("There is %d processes", smpi_global->process_count);
 162
 163   // sender/receiver processes
 164   smpi_global->main_processes =
 165     xbt_new(smx_process_t, smpi_global->process_count);
 166
 167   // timers
 168   smpi_global->timer = xbt_os_timer_new();
 169   smpi_global->timer_mutex = SIMIX_mutex_init();
 170   smpi_global->timer_cond = SIMIX_cond_init();
 171
 172   smpi_global->do_once_duration_nodes = NULL;
 173   smpi_global->do_once_duration = NULL;
 174   smpi_global->do_once_mutex = SIMIX_mutex_init();
 175
 176
 177   smpi_mpi_global = xbt_new(s_smpi_mpi_global_t, 1);
 178
 179   // global communicator
 180   smpi_mpi_global->mpi_comm_world = xbt_new(s_smpi_mpi_communicator_t, 1);
 181   smpi_mpi_global->mpi_comm_world->size = smpi_global->process_count;
 182   smpi_mpi_global->mpi_comm_world->barrier_count = 0;
 183   smpi_mpi_global->mpi_comm_world->barrier_mutex = SIMIX_mutex_init();
 184   smpi_mpi_global->mpi_comm_world->barrier_cond = SIMIX_cond_init();
 185   smpi_mpi_global->mpi_comm_world->rank_to_index_map =
 186     xbt_new(int, smpi_global->process_count);
 187   smpi_mpi_global->mpi_comm_world->index_to_rank_map =
 188     xbt_new(int, smpi_global->process_count);
 189   for (i = 0; i < smpi_global->process_count; i++) {
 190     smpi_mpi_global->mpi_comm_world->rank_to_index_map[i] = i;
 191     smpi_mpi_global->mpi_comm_world->index_to_rank_map[i] = i;
 192   }
 193
 194   // mpi datatypes
 195   smpi_mpi_global->mpi_byte = xbt_new(s_smpi_mpi_datatype_t, 1);
 196   smpi_mpi_global->mpi_byte->size = (size_t) 1;
 197   smpi_mpi_global->mpi_char = xbt_new(s_smpi_mpi_datatype_t, 1);
 198   smpi_mpi_global->mpi_char->size = (size_t) 1;
 199   smpi_mpi_global->mpi_int = xbt_new(s_smpi_mpi_datatype_t, 1);
 200   smpi_mpi_global->mpi_int->size = sizeof(int);
 201   smpi_mpi_global->mpi_float = xbt_new(s_smpi_mpi_datatype_t, 1);
 202   smpi_mpi_global->mpi_float->size = sizeof(float);
 203   smpi_mpi_global->mpi_double = xbt_new(s_smpi_mpi_datatype_t, 1);
 204   smpi_mpi_global->mpi_double->size = sizeof(double);
 205
 206   // mpi operations
 207   smpi_mpi_global->mpi_land = xbt_new(s_smpi_mpi_op_t, 1);
 208   smpi_mpi_global->mpi_land->func = smpi_mpi_land_func;
 209   smpi_mpi_global->mpi_sum = xbt_new(s_smpi_mpi_op_t, 1);
 210   smpi_mpi_global->mpi_sum->func = smpi_mpi_sum_func;
 211   smpi_mpi_global->mpi_min = xbt_new(s_smpi_mpi_op_t, 1);
 212   smpi_mpi_global->mpi_min->func = smpi_mpi_min_func;
 213   smpi_mpi_global->mpi_max = xbt_new(s_smpi_mpi_op_t, 1);
 214   smpi_mpi_global->mpi_max->func = smpi_mpi_max_func;
 215
 216 }
 217
 218 void smpi_global_destroy()
 219 {
 220   smpi_do_once_duration_node_t curr, next;
 221
 222   // processes
 223   xbt_free(smpi_global->main_processes);
 224
 225   // mallocators
 226   xbt_mallocator_free(smpi_global->request_mallocator);
 227   xbt_mallocator_free(smpi_global->message_mallocator);
 228
 229   xbt_os_timer_free(smpi_global->timer);
 230   SIMIX_mutex_destroy(smpi_global->timer_mutex);
 231   SIMIX_cond_destroy(smpi_global->timer_cond);
 232
 233   for (curr = smpi_global->do_once_duration_nodes; NULL != curr; curr = next) {
 234     next = curr->next;
 235     xbt_free(curr->file);
 236     xbt_free(curr);
 237   }
 238
 239   SIMIX_mutex_destroy(smpi_global->do_once_mutex);
 240
 241   xbt_free(smpi_global);
 242   smpi_global = NULL;
 243
 244   /* free smpi_mpi_global */
 245   SIMIX_mutex_destroy(smpi_mpi_global->mpi_comm_world->barrier_mutex);
 246   SIMIX_cond_destroy(smpi_mpi_global->mpi_comm_world->barrier_cond);
 247   xbt_free(smpi_mpi_global->mpi_comm_world->rank_to_index_map);
 248   xbt_free(smpi_mpi_global->mpi_comm_world);
 249
 250   xbt_free(smpi_mpi_global->mpi_byte);
 251   xbt_free(smpi_mpi_global->mpi_char);
 252   xbt_free(smpi_mpi_global->mpi_int);
 253   xbt_free(smpi_mpi_global->mpi_double);
 254   xbt_free(smpi_mpi_global->mpi_float);
 255
 256   xbt_free(smpi_mpi_global->mpi_land);
 257   xbt_free(smpi_mpi_global->mpi_sum);
 258   xbt_free(smpi_mpi_global->mpi_max);
 259   xbt_free(smpi_mpi_global->mpi_min);
 260
 261   xbt_free(smpi_mpi_global);
 262
 263 }
 264
 265 int smpi_process_index()
 266 {
 267   smpi_process_data_t pdata =
 268     (smpi_process_data_t) SIMIX_process_get_data(SIMIX_process_self());
 269   return pdata->index;
 270 }
 271
 272 smx_mutex_t smpi_process_mutex()
 273 {
 274   smpi_process_data_t pdata =
 275     (smpi_process_data_t) SIMIX_process_get_data(SIMIX_process_self());
 276   return pdata->mutex;
 277 }
 278
 279 smx_cond_t smpi_process_cond()
 280 {
 281   smpi_process_data_t pdata =
 282     (smpi_process_data_t) SIMIX_process_get_data(SIMIX_process_self());
 283   return pdata->cond;
 284 }
 285
 286 static void smpi_cfg_cb_host_speed(const char *name, int pos)
 287 {
 288   smpi_global->reference_speed =
 289     xbt_cfg_get_double_at(_surf_cfg_set, name, pos);
 290 }
 291
 292 int smpi_run_simulation(int *argc, char **argv)
 293 {
 294   smx_cond_t cond = NULL;
 295   smx_action_t action = NULL;
 296
 297   xbt_fifo_t actions_failed = xbt_fifo_new();
 298   xbt_fifo_t actions_done = xbt_fifo_new();
 299
 300   srand(SMPI_RAND_SEED);
 301
 302   double default_reference_speed = 20000.0;
 303   xbt_cfg_register(&_surf_cfg_set, "reference_speed",
 304                    "Power of the host running the simulation (in flop/s). Used to bench the operations.",
 305                    xbt_cfgelm_double, &default_reference_speed, 1, 1,
 306                    smpi_cfg_cb_host_speed, NULL);
 307
 308   int default_display_timing = 0;
 309   xbt_cfg_register(&_surf_cfg_set, "display_timing",
 310                    "Boolean indicating whether we should display the timing after simulation.",
 311                    xbt_cfgelm_int, &default_display_timing, 1, 1, NULL, NULL);
 312
 313   SIMIX_global_init(argc, argv);
 314
 315
 316   // parse the platform file: get the host list
 317   SIMIX_create_environment(argv[1]);
 318
 319   SIMIX_function_register("smpi_simulated_main", smpi_simulated_main);
 320   SIMIX_launch_application(argv[2]);
 321
 322   // must initialize globals between creating environment and launching app....
 323   smpi_global_init();
 324
 325   /* Clean IO before the run */
 326   fflush(stdout);
 327   fflush(stderr);
 328   SIMIX_init();
 329
 330   while (SIMIX_solve(actions_done, actions_failed) != -1.0) {
 331     while ((action = xbt_fifo_pop(actions_failed))) {
 332       DEBUG1("** %s failed **", action->name);
 333       while ((cond = xbt_fifo_pop(action->cond_list))) {
 334         SIMIX_cond_broadcast(cond);
 335       }
 336     }
 337     while ((action = xbt_fifo_pop(actions_done))) {
 338       DEBUG1("** %s done **", action->name);
 339       while ((cond = xbt_fifo_pop(action->cond_list))) {
 340         SIMIX_cond_broadcast(cond);
 341       }
 342     }
 343   }
 344
 345   // FIXME: cleanup incomplete
 346   xbt_fifo_free(actions_failed);
 347   xbt_fifo_free(actions_done);
 348
 349
 350   if (xbt_cfg_get_int(_surf_cfg_set, "display_timing"))
 351     INFO1("simulation time %g", SIMIX_get_clock());
 352
 353   smpi_global_destroy();
 354   SIMIX_clean();
 355
 356   return 0;
 357 }