1 /* Copyright (c) 2007-2014. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
8 #include "smpi_mpi_dt_private.h"
10 #include "xbt/replay.h"
11 #include "surf/surf.h"
12 #include "simix/smx_private.h"
13 #include "simgrid/sg_config.h"
15 #include <float.h> /* DBL_MAX */
20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_kernel, smpi,
21 "Logging specific to SMPI (kernel)");
23 typedef struct s_smpi_process_data {
28 smx_rdv_t mailbox_small;
31 void *data; /* user data */
34 int sampling; /* inside an SMPI_SAMPLE_ block? */
35 } s_smpi_process_data_t;
37 static smpi_process_data_t *process_data = NULL;
38 static int process_count = 0;
40 MPI_Comm MPI_COMM_WORLD = MPI_COMM_NULL;
41 int MPI_UNIVERSE_SIZE;
43 MPI_Errhandler *MPI_ERRORS_RETURN = NULL;
44 MPI_Errhandler *MPI_ERRORS_ARE_FATAL = NULL;
45 MPI_Errhandler *MPI_ERRHANDLER_NULL = NULL;
47 #define MAILBOX_NAME_MAXLEN (5 + sizeof(int) * 2 + 1)
49 static char *get_mailbox_name(char *str, int index)
51 snprintf(str, MAILBOX_NAME_MAXLEN, "SMPI-%0*x", (int) (sizeof(int) * 2),
56 static char *get_mailbox_name_small(char *str, int index)
58 snprintf(str, MAILBOX_NAME_MAXLEN, "small%0*x", (int) (sizeof(int) * 2),
63 void smpi_process_init(int *argc, char ***argv)
66 smpi_process_data_t data;
70 proc = SIMIX_process_self();
71 index = atoi((*argv)[1]);
73 smpi_current_rank = index;
76 data = smpi_process_remote_data(index);
77 simcall_process_set_data(proc, data);
80 memmove(&(*argv)[1], &(*argv)[2], sizeof(char *) * (*argc - 2));
81 (*argv)[(*argc) - 1] = NULL;
86 // set the process attached to the mailbox
87 simcall_rdv_set_receiver(data->mailbox_small, proc);
88 XBT_DEBUG("<%d> New process in the game: %p", index, proc);
90 if(smpi_privatize_global_variables){
91 switch_data_segment(index);
95 if (smpi_process_data() == NULL)
96 xbt_die("smpi_process_data() returned NULL. You probably gave a NULL parameter to MPI_Init. Although it's required by MPI-2, this is currently not supported by SMPI.");
99 void smpi_process_destroy(void)
101 int index = smpi_process_index();
102 if(smpi_privatize_global_variables){
103 switch_data_segment(index);
105 process_data[index]->state = SMPI_FINALIZED;
106 XBT_DEBUG("<%d> Process left the game", index);
110 * @brief Prepares the current process for termination.
112 void smpi_process_finalize(void)
115 // wait for all pending asynchronous comms to finish
116 while (SIMIX_process_has_pending_comms(SIMIX_process_self())) {
117 simcall_process_sleep(0.01);
121 int size = smpi_comm_size(MPI_COMM_WORLD);
122 int rank = smpi_comm_rank(MPI_COMM_WORLD);
123 /* All non-root send & receive zero-length message. */
125 smpi_mpi_ssend (NULL, 0, MPI_BYTE, 0,
128 smpi_mpi_recv (NULL, 0, MPI_BYTE, 0,
130 MPI_COMM_WORLD, MPI_STATUS_IGNORE);
132 /* The root collects and broadcasts the messages. */
134 MPI_Request* requests;
135 requests = (MPI_Request*)malloc( size * sizeof(MPI_Request) );
136 for (i = 1; i < size; ++i) {
137 requests[i] = smpi_mpi_irecv(NULL, 0, MPI_BYTE, MPI_ANY_SOURCE,
138 COLL_TAG_BARRIER, MPI_COMM_WORLD
141 smpi_mpi_waitall( size-1, requests+1, MPI_STATUSES_IGNORE );
142 for (i = 1; i < size; ++i) {
143 requests[i] = smpi_mpi_issend(NULL, 0, MPI_BYTE, i,
148 smpi_mpi_waitall( size-1, requests+1, MPI_STATUSES_IGNORE );
155 * @brief Check if a process is finalized
157 int smpi_process_finalized()
159 int index = smpi_process_index();
160 if (index != MPI_UNDEFINED)
161 return (process_data[index]->state == SMPI_FINALIZED);
167 * @brief Check if a process is initialized
169 int smpi_process_initialized(void)
171 int index = smpi_process_index();
172 return ( (index != MPI_UNDEFINED)
173 && (process_data[index]->state == SMPI_INITIALIZED));
177 * @brief Mark a process as initialized (=MPI_Init called)
179 void smpi_process_mark_as_initialized(void)
181 int index = smpi_process_index();
182 if ((index != MPI_UNDEFINED) && (!process_data[index]->state != SMPI_FINALIZED))
183 process_data[index]->state = SMPI_INITIALIZED;
188 int smpi_process_argc(void)
190 smpi_process_data_t data = smpi_process_data();
191 return data->argc ? *(data->argc) - 1 : 0;
194 int smpi_process_getarg(integer * index, char *dst, ftnlen len)
196 smpi_process_data_t data = smpi_process_data();
200 if (!data->argc || !data->argv || *index < 1 || *index >= *(data->argc)) {
203 arg = (*data->argv)[*index];
204 for (i = 0; i < len && arg[i] != '\0'; i++) {
207 for (; i < len; i++) {
214 int smpi_global_size(void)
216 char *value = getenv("SMPI_GLOBAL_SIZE");
220 "Please set env var SMPI_GLOBAL_SIZE to expected number of processes.\n");
226 smpi_process_data_t smpi_process_data(void)
228 return SIMIX_process_self_get_data(SIMIX_process_self());
231 smpi_process_data_t smpi_process_remote_data(int index)
233 return process_data[index];
236 void smpi_process_set_user_data(void *data)
238 smpi_process_data_t process_data = smpi_process_data();
239 process_data->data = data;
242 void *smpi_process_get_user_data()
244 smpi_process_data_t process_data = smpi_process_data();
245 return process_data->data;
248 int smpi_process_count(void)
250 return process_count;
253 int smpi_process_index(void)
255 smpi_process_data_t data = smpi_process_data();
256 //return -1 if not initialized
257 return data ? data->index : MPI_UNDEFINED;
260 smx_rdv_t smpi_process_mailbox(void)
262 smpi_process_data_t data = smpi_process_data();
263 return data->mailbox;
266 smx_rdv_t smpi_process_mailbox_small(void)
268 smpi_process_data_t data = smpi_process_data();
269 return data->mailbox_small;
272 smx_rdv_t smpi_process_remote_mailbox(int index)
274 smpi_process_data_t data = smpi_process_remote_data(index);
275 return data->mailbox;
279 smx_rdv_t smpi_process_remote_mailbox_small(int index)
281 smpi_process_data_t data = smpi_process_remote_data(index);
282 return data->mailbox_small;
285 xbt_os_timer_t smpi_process_timer(void)
287 smpi_process_data_t data = smpi_process_data();
291 void smpi_process_simulated_start(void)
293 smpi_process_data_t data = smpi_process_data();
294 data->simulated = SIMIX_get_clock();
297 double smpi_process_simulated_elapsed(void)
299 smpi_process_data_t data = smpi_process_data();
300 return SIMIX_get_clock() - data->simulated;
303 MPI_Comm smpi_process_comm_self(void)
305 smpi_process_data_t data = smpi_process_data();
306 if(data->comm_self==MPI_COMM_NULL){
307 MPI_Group group = smpi_group_new(1);
308 data->comm_self = smpi_comm_new(group);
309 smpi_group_set_mapping(group, smpi_process_index(), 0);
312 return data->comm_self;
315 void smpi_process_set_sampling(int s)
317 smpi_process_data_t data = smpi_process_data();
321 int smpi_process_get_sampling(void)
323 smpi_process_data_t data = smpi_process_data();
324 return data->sampling;
327 void print_request(const char *message, MPI_Request request)
330 ("%s request %p [buf = %p, size = %zu, src = %d, dst = %d, tag = %d, flags = %x]",
331 message, request, request->buf, request->size, request->src,
332 request->dst, request->tag, request->flags);
335 static void smpi_comm_copy_buffer_callback(smx_action_t comm,
336 void *buff, size_t buff_size)
338 XBT_DEBUG("Copy the data over");
339 if(_xbt_replay_is_active()) return;
342 if((smpi_privatize_global_variables)
343 && ((char*)buff >= start_data_exe)
344 && ((char*)buff < start_data_exe + size_data_exe )
346 XBT_DEBUG("Privatization : We are copying from a zone inside global memory... Saving data to temp buffer !");
347 switch_data_segment(((smpi_process_data_t)SIMIX_process_get_data(comm->comm.src_proc))->index);
348 tmpbuff = (void*)xbt_malloc(buff_size);
349 memcpy(tmpbuff, buff, buff_size);
353 if((smpi_privatize_global_variables)
354 && ((char*)comm->comm.dst_buff >= start_data_exe)
355 && ((char*)comm->comm.dst_buff < start_data_exe + size_data_exe )
357 XBT_DEBUG("Privatization : We are copying to a zone inside global memory - Switch data segment");
358 switch_data_segment(((smpi_process_data_t)SIMIX_process_get_data(comm->comm.dst_proc))->index);
362 memcpy(comm->comm.dst_buff, tmpbuff, buff_size);
363 if (comm->comm.detached) {
364 // if this is a detached send, the source buffer was duplicated by SMPI
365 // sender to make the original buffer available to the application ASAP
367 //It seems that the request is used after the call there this should
368 //be free somewhereelse but where???
369 //xbt_free(comm->comm.src_data);// inside SMPI the request is keep
370 //inside the user data and should be free
371 comm->comm.src_buff = NULL;
374 if(tmpbuff!=buff)xbt_free(tmpbuff);
378 void smpi_global_init(void)
382 char name[MAILBOX_NAME_MAXLEN];
384 SIMIX_comm_set_copy_data_callback(&smpi_comm_copy_buffer_callback);
385 process_count = SIMIX_process_count();
386 process_data = xbt_new(smpi_process_data_t, process_count);
387 for (i = 0; i < process_count; i++) {
388 process_data[i] = xbt_new(s_smpi_process_data_t, 1);
389 process_data[i]->index = i;
390 process_data[i]->argc = NULL;
391 process_data[i]->argv = NULL;
392 process_data[i]->mailbox = simcall_rdv_create(get_mailbox_name(name, i));
393 process_data[i]->mailbox_small =
394 simcall_rdv_create(get_mailbox_name_small(name, i));
395 process_data[i]->timer = xbt_os_timer_new();
397 MC_ignore_heap(process_data[i]->timer, xbt_os_timer_size());
398 process_data[i]->comm_self = MPI_COMM_NULL;
399 process_data[i]->state = SMPI_UNINITIALIZED;
400 process_data[i]->sampling = 0;
402 group = smpi_group_new(process_count);
403 MPI_COMM_WORLD = smpi_comm_new(group);
404 MPI_UNIVERSE_SIZE = smpi_comm_size(MPI_COMM_WORLD);
405 for (i = 0; i < process_count; i++) {
406 smpi_group_set_mapping(group, i, i);
409 //check correctness of MPI parameters
411 xbt_assert(sg_cfg_get_int("smpi/async_small_thres") <=
412 sg_cfg_get_int("smpi/send_is_detached_thres"));
414 if (sg_cfg_is_default_value("smpi/running_power")) {
415 XBT_INFO("You did not set the power of the host running the simulation. "
416 "The timings will certainly not be accurate. "
417 "Use the option \"--cfg=smpi/running_power:<flops>\" to set its value."
418 "Check http://simgrid.org/simgrid/latest/doc/options.html#options_smpi_bench for more information. ");
420 if(smpi_privatize_global_variables)
421 smpi_initialize_global_memory_segments();
424 void smpi_global_destroy(void)
426 int count = smpi_process_count();
429 smpi_bench_destroy();
430 while (smpi_group_unuse(smpi_comm_group(MPI_COMM_WORLD)) > 0);
431 xbt_free(MPI_COMM_WORLD);
432 MPI_COMM_WORLD = MPI_COMM_NULL;
433 for (i = 0; i < count; i++) {
434 if(process_data[i]->comm_self!=MPI_COMM_NULL){
435 smpi_group_unuse(smpi_comm_group(process_data[i]->comm_self));
436 smpi_comm_destroy(process_data[i]->comm_self);
438 xbt_os_timer_free(process_data[i]->timer);
439 simcall_rdv_destroy(process_data[i]->mailbox);
440 simcall_rdv_destroy(process_data[i]->mailbox_small);
441 xbt_free(process_data[i]);
443 xbt_free(process_data);
445 if(smpi_privatize_global_variables)
446 smpi_destroy_global_memory_segments();
450 /* Fortran specific stuff */
451 /* With smpicc, the following weak symbols are used */
452 /* With smpiff, the following weak symbols are replaced by those in libf2c */
453 int __attribute__ ((weak)) xargc;
454 char ** __attribute__ ((weak)) xargv;
457 void __attribute__ ((weak)) user_main_()
459 xbt_die("Should not be in this smpi_simulated_main");
463 int __attribute__ ((weak)) smpi_simulated_main_(int argc, char **argv)
465 smpi_process_init(&argc, &argv);
470 int __attribute__ ((weak)) main(int argc, char **argv)
472 return smpi_main(smpi_simulated_main_, argc, argv);
475 int __attribute__ ((weak)) MAIN__()
477 return smpi_main(smpi_simulated_main_, xargc, xargv);
481 int smpi_main(int (*realmain) (int argc, char *argv[]), int argc, char *argv[])
483 srand(SMPI_RAND_SEED);
485 if (getenv("SMPI_PRETEND_CC") != NULL) {
486 /* Hack to ensure that smpicc can pretend to be a simple
487 * compiler. Particularly handy to pass it to the configuration tools */
491 /* Connect log categories. See xbt/log.c */
492 XBT_LOG_CONNECT(smpi); /* Keep this line as soon as possible in this
493 function: xbt_log_appender_file.c depends on it
494 DO NOT connect this in XBT or so, or it will be
495 useless to xbt_log_appender_file.c */
497 XBT_LOG_CONNECT(instr_smpi);
499 XBT_LOG_CONNECT(smpi_base);
500 XBT_LOG_CONNECT(smpi_bench);
501 XBT_LOG_CONNECT(smpi_coll);
502 XBT_LOG_CONNECT(smpi_colls);
503 XBT_LOG_CONNECT(smpi_comm);
504 XBT_LOG_CONNECT(smpi_dvfs);
505 XBT_LOG_CONNECT(smpi_group);
506 XBT_LOG_CONNECT(smpi_kernel);
507 XBT_LOG_CONNECT(smpi_mpi);
508 XBT_LOG_CONNECT(smpi_mpi_dt);
509 XBT_LOG_CONNECT(smpi_pmpi);
510 XBT_LOG_CONNECT(smpi_replay);
513 TRACE_global_init(&argc, argv);
515 TRACE_add_start_function(TRACE_smpi_alloc);
516 TRACE_add_end_function(TRACE_smpi_release);
519 SIMIX_global_init(&argc, argv);
525 // parse the platform file: get the host list
526 SIMIX_create_environment(argv[1]);
528 SIMIX_function_register_default(realmain);
529 SIMIX_launch_application(argv[2]);
531 int gather_id = find_coll_description(mpi_coll_gather_description,
532 sg_cfg_get_string("smpi/gather"));
533 mpi_coll_gather_fun = (int (*)(void *, int, MPI_Datatype,
534 void *, int, MPI_Datatype, int, MPI_Comm))
535 mpi_coll_gather_description[gather_id].coll;
537 int allgather_id = find_coll_description(mpi_coll_allgather_description,
538 sg_cfg_get_string("smpi/allgather"));
539 mpi_coll_allgather_fun = (int (*)(void *, int, MPI_Datatype,
540 void *, int, MPI_Datatype, MPI_Comm))
541 mpi_coll_allgather_description[allgather_id].coll;
543 int allgatherv_id = find_coll_description(mpi_coll_allgatherv_description,
544 sg_cfg_get_string("smpi/allgatherv"));
545 mpi_coll_allgatherv_fun = (int (*)(void *, int, MPI_Datatype, void *, int *,
546 int *, MPI_Datatype, MPI_Comm))
547 mpi_coll_allgatherv_description[allgatherv_id].coll;
549 int allreduce_id = find_coll_description(mpi_coll_allreduce_description,
550 sg_cfg_get_string("smpi/allreduce"));
551 mpi_coll_allreduce_fun = (int (*)(void *sbuf, void *rbuf, int rcount,
552 MPI_Datatype dtype, MPI_Op op,
554 mpi_coll_allreduce_description[allreduce_id].coll;
556 int alltoall_id = find_coll_description(mpi_coll_alltoall_description,
557 sg_cfg_get_string("smpi/alltoall"));
558 mpi_coll_alltoall_fun = (int (*)(void *, int, MPI_Datatype,
559 void *, int, MPI_Datatype, MPI_Comm))
560 mpi_coll_alltoall_description[alltoall_id].coll;
562 int alltoallv_id = find_coll_description(mpi_coll_alltoallv_description,
563 sg_cfg_get_string("smpi/alltoallv"));
564 mpi_coll_alltoallv_fun = (int (*)(void *, int *, int *, MPI_Datatype,
565 void *, int *, int *, MPI_Datatype,
567 mpi_coll_alltoallv_description[alltoallv_id].coll;
569 int bcast_id = find_coll_description(mpi_coll_bcast_description,
570 sg_cfg_get_string("smpi/bcast"));
571 mpi_coll_bcast_fun = (int (*)(void *buf, int count, MPI_Datatype datatype,
572 int root, MPI_Comm com))
573 mpi_coll_bcast_description[bcast_id].coll;
575 int reduce_id = find_coll_description(mpi_coll_reduce_description,
576 sg_cfg_get_string("smpi/reduce"));
577 mpi_coll_reduce_fun = (int (*)(void *buf, void *rbuf, int count,
578 MPI_Datatype datatype, MPI_Op op,
579 int root, MPI_Comm comm))
580 mpi_coll_reduce_description[reduce_id].coll;
582 int reduce_scatter_id =
583 find_coll_description(mpi_coll_reduce_scatter_description,
584 sg_cfg_get_string("smpi/reduce_scatter"));
585 mpi_coll_reduce_scatter_fun = (int (*)(void *sbuf, void *rbuf, int *rcounts,
586 MPI_Datatype dtype, MPI_Op op,
588 mpi_coll_reduce_scatter_description[reduce_scatter_id].coll;
590 int scatter_id = find_coll_description(mpi_coll_scatter_description,
591 sg_cfg_get_string("smpi/scatter"));
592 mpi_coll_scatter_fun = (int (*)(void *sendbuf, int sendcount,
593 MPI_Datatype sendtype, void *recvbuf,
594 int recvcount, MPI_Datatype recvtype,
595 int root, MPI_Comm comm))
596 mpi_coll_scatter_description[scatter_id].coll;
598 int barrier_id = find_coll_description(mpi_coll_barrier_description,
599 sg_cfg_get_string("smpi/barrier"));
600 mpi_coll_barrier_fun = (int (*)(MPI_Comm comm))
601 mpi_coll_barrier_description[barrier_id].coll;
603 smpi_cpu_threshold = sg_cfg_get_double("smpi/cpu_threshold");
604 smpi_running_power = sg_cfg_get_double("smpi/running_power");
605 smpi_privatize_global_variables = sg_cfg_get_boolean("smpi/privatize_global_variables");
606 if (smpi_cpu_threshold < 0)
607 smpi_cpu_threshold = DBL_MAX;
611 /* Clean IO before the run */
616 MC_do_the_modelcheck_for_real();
620 if (sg_cfg_get_boolean("smpi/display_timing"))
621 XBT_INFO("Simulation time: %g seconds.", SIMIX_get_clock());
623 smpi_global_destroy();