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;
75 data = smpi_process_remote_data(index);
76 simcall_process_set_data(proc, data);
79 memmove(&(*argv)[1], &(*argv)[2], sizeof(char *) * (*argc - 2));
80 (*argv)[(*argc) - 1] = NULL;
85 // set the process attached to the mailbox
86 simcall_rdv_set_receiver(data->mailbox_small, proc);
87 XBT_DEBUG("<%d> New process in the game: %p", index, proc);
89 if (smpi_process_data() == NULL)
90 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.");
93 void smpi_process_destroy(void)
95 int index = smpi_process_index();
96 process_data[index]->state = SMPI_FINALIZED;
97 XBT_DEBUG("<%d> Process left the game", index);
101 * @brief Prepares the current process for termination.
103 void smpi_process_finalize(void)
106 // wait for all pending asynchronous comms to finish
107 while (SIMIX_process_has_pending_comms(SIMIX_process_self())) {
108 simcall_process_sleep(0.01);
112 int size = smpi_comm_size(MPI_COMM_WORLD);
113 int rank = smpi_comm_rank(MPI_COMM_WORLD);
114 /* All non-root send & receive zero-length message. */
116 smpi_mpi_ssend (NULL, 0, MPI_BYTE, 0,
119 smpi_mpi_recv (NULL, 0, MPI_BYTE, 0,
121 MPI_COMM_WORLD, MPI_STATUS_IGNORE);
123 /* The root collects and broadcasts the messages. */
125 MPI_Request* requests;
126 requests = (MPI_Request*)malloc( size * sizeof(MPI_Request) );
127 for (i = 1; i < size; ++i) {
128 requests[i] = smpi_mpi_irecv(NULL, 0, MPI_BYTE, MPI_ANY_SOURCE,
129 COLL_TAG_BARRIER, MPI_COMM_WORLD
132 smpi_mpi_waitall( size-1, requests+1, MPI_STATUSES_IGNORE );
133 for (i = 1; i < size; ++i) {
134 requests[i] = smpi_mpi_issend(NULL, 0, MPI_BYTE, i,
139 smpi_mpi_waitall( size-1, requests+1, MPI_STATUSES_IGNORE );
146 * @brief Check if a process is finalized
148 int smpi_process_finalized()
150 int index = smpi_process_index();
151 if (index != MPI_UNDEFINED)
152 return (process_data[index]->state == SMPI_FINALIZED);
158 * @brief Check if a process is initialized
160 int smpi_process_initialized(void)
162 int index = smpi_process_index();
163 return ( (index != MPI_UNDEFINED)
164 && (process_data[index]->state == SMPI_INITIALIZED));
168 * @brief Mark a process as initialized (=MPI_Init called)
170 void smpi_process_mark_as_initialized(void)
172 int index = smpi_process_index();
173 if ((index != MPI_UNDEFINED) && (!process_data[index]->state != SMPI_FINALIZED))
174 process_data[index]->state = SMPI_INITIALIZED;
179 int smpi_process_argc(void)
181 smpi_process_data_t data = smpi_process_data();
182 return data->argc ? *(data->argc) - 1 : 0;
185 int smpi_process_getarg(integer * index, char *dst, ftnlen len)
187 smpi_process_data_t data = smpi_process_data();
191 if (!data->argc || !data->argv || *index < 1 || *index >= *(data->argc)) {
194 arg = (*data->argv)[*index];
195 for (i = 0; i < len && arg[i] != '\0'; i++) {
198 for (; i < len; i++) {
205 int smpi_global_size(void)
207 char *value = getenv("SMPI_GLOBAL_SIZE");
211 "Please set env var SMPI_GLOBAL_SIZE to expected number of processes.\n");
217 smpi_process_data_t smpi_process_data(void)
219 return SIMIX_process_self_get_data(SIMIX_process_self());
222 smpi_process_data_t smpi_process_remote_data(int index)
224 return process_data[index];
227 void smpi_process_set_user_data(void *data)
229 smpi_process_data_t process_data = smpi_process_data();
230 process_data->data = data;
233 void *smpi_process_get_user_data()
235 smpi_process_data_t process_data = smpi_process_data();
236 return process_data->data;
239 int smpi_process_count(void)
241 return process_count;
244 int smpi_process_index(void)
246 smpi_process_data_t data = smpi_process_data();
247 //return -1 if not initialized
248 return data ? data->index : MPI_UNDEFINED;
251 smx_rdv_t smpi_process_mailbox(void)
253 smpi_process_data_t data = smpi_process_data();
254 return data->mailbox;
257 smx_rdv_t smpi_process_mailbox_small(void)
259 smpi_process_data_t data = smpi_process_data();
260 return data->mailbox_small;
263 smx_rdv_t smpi_process_remote_mailbox(int index)
265 smpi_process_data_t data = smpi_process_remote_data(index);
266 return data->mailbox;
270 smx_rdv_t smpi_process_remote_mailbox_small(int index)
272 smpi_process_data_t data = smpi_process_remote_data(index);
273 return data->mailbox_small;
276 xbt_os_timer_t smpi_process_timer(void)
278 smpi_process_data_t data = smpi_process_data();
282 void smpi_process_simulated_start(void)
284 smpi_process_data_t data = smpi_process_data();
285 data->simulated = SIMIX_get_clock();
288 double smpi_process_simulated_elapsed(void)
290 smpi_process_data_t data = smpi_process_data();
291 return SIMIX_get_clock() - data->simulated;
294 MPI_Comm smpi_process_comm_self(void)
296 smpi_process_data_t data = smpi_process_data();
297 if(data->comm_self==MPI_COMM_NULL){
298 MPI_Group group = smpi_group_new(1);
299 data->comm_self = smpi_comm_new(group);
300 smpi_group_set_mapping(group, smpi_process_index(), 0);
303 return data->comm_self;
306 void smpi_process_set_sampling(int s)
308 smpi_process_data_t data = smpi_process_data();
312 int smpi_process_get_sampling(void)
314 smpi_process_data_t data = smpi_process_data();
315 return data->sampling;
318 void print_request(const char *message, MPI_Request request)
321 ("%s request %p [buf = %p, size = %zu, src = %d, dst = %d, tag = %d, flags = %x]",
322 message, request, request->buf, request->size, request->src,
323 request->dst, request->tag, request->flags);
326 static void smpi_comm_copy_buffer_callback(smx_action_t comm,
327 void *buff, size_t buff_size)
329 XBT_DEBUG("Copy the data over");
330 if(_xbt_replay_is_active()) return;
333 if((smpi_privatize_global_variables)
334 && ((char*)buff >= start_data_exe)
335 && ((char*)buff < start_data_exe + size_data_exe )
337 XBT_WARN("Privatization : We are copying from a zone inside global memory... Saving data to temp buffer !");
338 switch_data_segment(((smpi_process_data_t)SIMIX_process_get_data(comm->comm.src_proc))->index);
339 tmpbuff = (void*)xbt_malloc(buff_size);
340 memcpy(tmpbuff, buff, buff_size);
344 if((smpi_privatize_global_variables)
345 && ((char*)comm->comm.dst_buff >= start_data_exe)
346 && ((char*)comm->comm.dst_buff < start_data_exe + size_data_exe )
348 XBT_WARN("Privatization : We are copying to a zone inside global memory - Switch data segment");
349 switch_data_segment(((smpi_process_data_t)SIMIX_process_get_data(comm->comm.dst_proc))->index);
353 memcpy(comm->comm.dst_buff, tmpbuff, buff_size);
354 if (comm->comm.detached) {
355 // if this is a detached send, the source buffer was duplicated by SMPI
356 // sender to make the original buffer available to the application ASAP
358 //It seems that the request is used after the call there this should
359 //be free somewhereelse but where???
360 //xbt_free(comm->comm.src_data);// inside SMPI the request is keep
361 //inside the user data and should be free
362 comm->comm.src_buff = NULL;
365 if(tmpbuff!=buff)xbt_free(tmpbuff);
369 void smpi_global_init(void)
373 char name[MAILBOX_NAME_MAXLEN];
375 SIMIX_comm_set_copy_data_callback(&smpi_comm_copy_buffer_callback);
376 process_count = SIMIX_process_count();
377 process_data = xbt_new(smpi_process_data_t, process_count);
378 for (i = 0; i < process_count; i++) {
379 process_data[i] = xbt_new(s_smpi_process_data_t, 1);
380 process_data[i]->index = i;
381 process_data[i]->argc = NULL;
382 process_data[i]->argv = NULL;
383 process_data[i]->mailbox = simcall_rdv_create(get_mailbox_name(name, i));
384 process_data[i]->mailbox_small =
385 simcall_rdv_create(get_mailbox_name_small(name, i));
386 process_data[i]->timer = xbt_os_timer_new();
388 MC_ignore_heap(process_data[i]->timer, xbt_os_timer_size());
389 process_data[i]->comm_self = MPI_COMM_NULL;
390 process_data[i]->state = SMPI_UNINITIALIZED;
391 process_data[i]->sampling = 0;
393 group = smpi_group_new(process_count);
394 MPI_COMM_WORLD = smpi_comm_new(group);
395 MPI_UNIVERSE_SIZE = smpi_comm_size(MPI_COMM_WORLD);
396 for (i = 0; i < process_count; i++) {
397 smpi_group_set_mapping(group, i, i);
400 //check correctness of MPI parameters
402 xbt_assert(sg_cfg_get_int("smpi/async_small_thres") <=
403 sg_cfg_get_int("smpi/send_is_detached_thres"));
405 if (sg_cfg_is_default_value("smpi/running_power")) {
406 XBT_INFO("You did not set the power of the host running the simulation. "
407 "The timings will certainly not be accurate. "
408 "Use the option \"--cfg=smpi/running_power:<flops>\" to set its value."
409 "Check http://simgrid.org/simgrid/latest/doc/options.html#options_smpi_bench for more information. ");
411 if(smpi_privatize_global_variables)
412 smpi_initialize_global_memory_segments();
415 void smpi_global_destroy(void)
417 int count = smpi_process_count();
420 smpi_bench_destroy();
421 while (smpi_group_unuse(smpi_comm_group(MPI_COMM_WORLD)) > 0);
422 xbt_free(MPI_COMM_WORLD);
423 MPI_COMM_WORLD = MPI_COMM_NULL;
424 for (i = 0; i < count; i++) {
425 if(process_data[i]->comm_self!=MPI_COMM_NULL){
426 smpi_group_unuse(smpi_comm_group(process_data[i]->comm_self));
427 smpi_comm_destroy(process_data[i]->comm_self);
429 xbt_os_timer_free(process_data[i]->timer);
430 simcall_rdv_destroy(process_data[i]->mailbox);
431 simcall_rdv_destroy(process_data[i]->mailbox_small);
432 xbt_free(process_data[i]);
434 xbt_free(process_data);
436 if(smpi_privatize_global_variables)
437 smpi_destroy_global_memory_segments();
441 /* Fortran specific stuff */
442 /* With smpicc, the following weak symbols are used */
443 /* With smpiff, the following weak symbols are replaced by those in libf2c */
444 int __attribute__ ((weak)) xargc;
445 char ** __attribute__ ((weak)) xargv;
448 void __attribute__ ((weak)) user_main_()
450 xbt_die("Should not be in this smpi_simulated_main");
454 int __attribute__ ((weak)) smpi_simulated_main_(int argc, char **argv)
456 smpi_process_init(&argc, &argv);
461 int __attribute__ ((weak)) main(int argc, char **argv)
463 return smpi_main(smpi_simulated_main_, argc, argv);
466 int __attribute__ ((weak)) MAIN__()
468 return smpi_main(smpi_simulated_main_, xargc, xargv);
472 int smpi_main(int (*realmain) (int argc, char *argv[]), int argc, char *argv[])
474 srand(SMPI_RAND_SEED);
476 if (getenv("SMPI_PRETEND_CC") != NULL) {
477 /* Hack to ensure that smpicc can pretend to be a simple
478 * compiler. Particularly handy to pass it to the configuration tools */
482 /* Connect log categories. See xbt/log.c */
483 XBT_LOG_CONNECT(smpi); /* Keep this line as soon as possible in this
484 function: xbt_log_appender_file.c depends on it
485 DO NOT connect this in XBT or so, or it will be
486 useless to xbt_log_appender_file.c */
488 XBT_LOG_CONNECT(instr_smpi);
490 XBT_LOG_CONNECT(smpi_base);
491 XBT_LOG_CONNECT(smpi_bench);
492 XBT_LOG_CONNECT(smpi_coll);
493 XBT_LOG_CONNECT(smpi_colls);
494 XBT_LOG_CONNECT(smpi_comm);
495 XBT_LOG_CONNECT(smpi_dvfs);
496 XBT_LOG_CONNECT(smpi_group);
497 XBT_LOG_CONNECT(smpi_kernel);
498 XBT_LOG_CONNECT(smpi_mpi);
499 XBT_LOG_CONNECT(smpi_mpi_dt);
500 XBT_LOG_CONNECT(smpi_pmpi);
501 XBT_LOG_CONNECT(smpi_replay);
504 TRACE_global_init(&argc, argv);
506 TRACE_add_start_function(TRACE_smpi_alloc);
507 TRACE_add_end_function(TRACE_smpi_release);
510 SIMIX_global_init(&argc, argv);
516 // parse the platform file: get the host list
517 SIMIX_create_environment(argv[1]);
519 SIMIX_function_register_default(realmain);
520 SIMIX_launch_application(argv[2]);
522 int gather_id = find_coll_description(mpi_coll_gather_description,
523 sg_cfg_get_string("smpi/gather"));
524 mpi_coll_gather_fun = (int (*)(void *, int, MPI_Datatype,
525 void *, int, MPI_Datatype, int, MPI_Comm))
526 mpi_coll_gather_description[gather_id].coll;
528 int allgather_id = find_coll_description(mpi_coll_allgather_description,
529 sg_cfg_get_string("smpi/allgather"));
530 mpi_coll_allgather_fun = (int (*)(void *, int, MPI_Datatype,
531 void *, int, MPI_Datatype, MPI_Comm))
532 mpi_coll_allgather_description[allgather_id].coll;
534 int allgatherv_id = find_coll_description(mpi_coll_allgatherv_description,
535 sg_cfg_get_string("smpi/allgatherv"));
536 mpi_coll_allgatherv_fun = (int (*)(void *, int, MPI_Datatype, void *, int *,
537 int *, MPI_Datatype, MPI_Comm))
538 mpi_coll_allgatherv_description[allgatherv_id].coll;
540 int allreduce_id = find_coll_description(mpi_coll_allreduce_description,
541 sg_cfg_get_string("smpi/allreduce"));
542 mpi_coll_allreduce_fun = (int (*)(void *sbuf, void *rbuf, int rcount,
543 MPI_Datatype dtype, MPI_Op op,
545 mpi_coll_allreduce_description[allreduce_id].coll;
547 int alltoall_id = find_coll_description(mpi_coll_alltoall_description,
548 sg_cfg_get_string("smpi/alltoall"));
549 mpi_coll_alltoall_fun = (int (*)(void *, int, MPI_Datatype,
550 void *, int, MPI_Datatype, MPI_Comm))
551 mpi_coll_alltoall_description[alltoall_id].coll;
553 int alltoallv_id = find_coll_description(mpi_coll_alltoallv_description,
554 sg_cfg_get_string("smpi/alltoallv"));
555 mpi_coll_alltoallv_fun = (int (*)(void *, int *, int *, MPI_Datatype,
556 void *, int *, int *, MPI_Datatype,
558 mpi_coll_alltoallv_description[alltoallv_id].coll;
560 int bcast_id = find_coll_description(mpi_coll_bcast_description,
561 sg_cfg_get_string("smpi/bcast"));
562 mpi_coll_bcast_fun = (int (*)(void *buf, int count, MPI_Datatype datatype,
563 int root, MPI_Comm com))
564 mpi_coll_bcast_description[bcast_id].coll;
566 int reduce_id = find_coll_description(mpi_coll_reduce_description,
567 sg_cfg_get_string("smpi/reduce"));
568 mpi_coll_reduce_fun = (int (*)(void *buf, void *rbuf, int count,
569 MPI_Datatype datatype, MPI_Op op,
570 int root, MPI_Comm comm))
571 mpi_coll_reduce_description[reduce_id].coll;
573 int reduce_scatter_id =
574 find_coll_description(mpi_coll_reduce_scatter_description,
575 sg_cfg_get_string("smpi/reduce_scatter"));
576 mpi_coll_reduce_scatter_fun = (int (*)(void *sbuf, void *rbuf, int *rcounts,
577 MPI_Datatype dtype, MPI_Op op,
579 mpi_coll_reduce_scatter_description[reduce_scatter_id].coll;
581 int scatter_id = find_coll_description(mpi_coll_scatter_description,
582 sg_cfg_get_string("smpi/scatter"));
583 mpi_coll_scatter_fun = (int (*)(void *sendbuf, int sendcount,
584 MPI_Datatype sendtype, void *recvbuf,
585 int recvcount, MPI_Datatype recvtype,
586 int root, MPI_Comm comm))
587 mpi_coll_scatter_description[scatter_id].coll;
589 int barrier_id = find_coll_description(mpi_coll_barrier_description,
590 sg_cfg_get_string("smpi/barrier"));
591 mpi_coll_barrier_fun = (int (*)(MPI_Comm comm))
592 mpi_coll_barrier_description[barrier_id].coll;
594 smpi_cpu_threshold = sg_cfg_get_double("smpi/cpu_threshold");
595 smpi_running_power = sg_cfg_get_double("smpi/running_power");
596 smpi_privatize_global_variables = sg_cfg_get_boolean("smpi/privatize_global_variables");
597 if (smpi_cpu_threshold < 0)
598 smpi_cpu_threshold = DBL_MAX;
602 /* Clean IO before the run */
607 MC_do_the_modelcheck_for_real();
611 if (sg_cfg_get_boolean("smpi/display_timing"))
612 XBT_INFO("Simulation time: %g seconds.", SIMIX_get_clock());
614 smpi_global_destroy();