1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
4 * This source code is part of
8 * GROningen MAchine for Chemical Simulations
11 * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
12 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
13 * Copyright (c) 2001-2004, The GROMACS development team,
14 * check out http://www.gromacs.org for more information.
16 * This program is free software; you can redistribute it and/or
17 * modify it under the terms of the GNU General Public License
18 * as published by the Free Software Foundation; either version 2
19 * of the License, or (at your option) any later version.
21 * If you want to redistribute modifications, please consider that
22 * scientific software is very special. Version control is crucial -
23 * bugs must be traceable. We will be happy to consider code for
24 * inclusion in the official distribution, but derived work must not
25 * be called official GROMACS. Details are found in the README & COPYING
26 * files - if they are missing, get the official version at www.gromacs.org.
28 * To help us fund GROMACS development, we humbly ask that you cite
29 * the papers on the package - you can find them in the top README file.
31 * For more info, check our website at http://www.gromacs.org
34 * GROningen Mixture of Alchemy and Childrens' Stories
39 #include "gromacs/utility/gmx_header_config.h"
47 #ifdef HAVE_SYS_TIME_H
52 #include "gmx_fatal.h"
63 #include "thread_mpi.h"
66 /* The source code in this file should be thread-safe.
67 Please keep it that way. */
74 #ifdef GMX_NATIVE_WINDOWS
81 static void par_fn(char *base, int ftp, const t_commrec *cr,
82 gmx_bool bAppendSimId, gmx_bool bAppendNodeId,
83 char buf[], int bufsize)
85 if ((size_t)bufsize < (strlen(base)+10))
87 gmx_mem("Character buffer too small!");
90 /* Copy to buf, and strip extension */
92 buf[strlen(base) - strlen(ftp2ext(fn2ftp(base))) - 1] = '\0';
96 sprintf(buf+strlen(buf), "%d", cr->ms->sim);
100 strcat(buf, "_node");
101 sprintf(buf+strlen(buf), "%d", cr->nodeid);
105 /* Add extension again */
106 strcat(buf, (ftp == efTPX) ? "tpr" : (ftp == efEDR) ? "edr" : ftp2ext(ftp));
109 fprintf(debug, "node %d par_fn '%s'\n", cr->nodeid, buf);
110 if (fn2ftp(buf) == efLOG)
112 fprintf(debug, "log\n");
117 void check_multi_int(FILE *log, const gmx_multisim_t *ms, int val,
122 gmx_bool bCompatible;
124 if (NULL != log && !bQuiet)
126 fprintf(log, "Multi-checking %s ... ", name);
132 "check_multi_int called with a NULL communication pointer");
135 snew(ibuf, ms->nsim);
137 gmx_sumi_sim(ms->nsim, ibuf, ms);
140 for (p = 1; p < ms->nsim; p++)
142 bCompatible = bCompatible && (ibuf[p-1] == ibuf[p]);
147 if (NULL != log && !bQuiet)
149 fprintf(log, "OK\n");
156 fprintf(log, "\n%s is not equal for all subsystems\n", name);
157 for (p = 0; p < ms->nsim; p++)
159 fprintf(log, " subsystem %d: %d\n", p, ibuf[p]);
162 gmx_fatal(FARGS, "The %d subsystems are not compatible\n", ms->nsim);
168 void check_multi_large_int(FILE *log, const gmx_multisim_t *ms,
169 gmx_large_int_t val, const char *name,
172 gmx_large_int_t *ibuf;
174 gmx_bool bCompatible;
176 if (NULL != log && !bQuiet)
178 fprintf(log, "Multi-checking %s ... ", name);
184 "check_multi_int called with a NULL communication pointer");
187 snew(ibuf, ms->nsim);
189 gmx_sumli_sim(ms->nsim, ibuf, ms);
192 for (p = 1; p < ms->nsim; p++)
194 bCompatible = bCompatible && (ibuf[p-1] == ibuf[p]);
199 if (NULL != log && !bQuiet)
201 fprintf(log, "OK\n");
208 fprintf(log, "\n%s is not equal for all subsystems\n", name);
209 for (p = 0; p < ms->nsim; p++)
212 /* first make the format string */
213 snprintf(strbuf, 255, " subsystem %%d: %s\n",
215 fprintf(log, strbuf, p, ibuf[p]);
218 gmx_fatal(FARGS, "The %d subsystems are not compatible\n", ms->nsim);
225 char *gmx_gethostname(char *name, size_t len)
229 gmx_incons("gmx_gethostname called with len<8");
232 if (gethostname(name, len-1) != 0)
234 strncpy(name, "unknown", 8);
237 strncpy(name, "unknown", 8);
244 void gmx_log_open(const char *lognm, const t_commrec *cr, gmx_bool bMasterOnly,
245 gmx_bool bAppendFiles, FILE** fplog)
248 char buf[256], host[256];
250 char timebuf[STRLEN];
256 /* Communicate the filename for logfile */
257 if (cr->nnodes > 1 && !bMasterOnly
258 #ifdef GMX_THREAD_MPI
259 /* With thread MPI the non-master log files are opened later
260 * when the files names are already known on all nodes.
268 len = strlen(lognm) + 1;
270 gmx_bcast(sizeof(len), &len, cr);
277 tmpnm = gmx_strdup(lognm);
279 gmx_bcast(len*sizeof(*tmpnm), tmpnm, cr);
283 tmpnm = gmx_strdup(lognm);
288 if (!bMasterOnly && !MASTER(cr))
290 /* Since log always ends with '.log' let's use this info */
291 par_fn(tmpnm, efLOG, cr, FALSE, !bMasterOnly, buf, 255);
292 fp = gmx_fio_fopen(buf, bAppendFiles ? "a+" : "w+" );
294 else if (!bAppendFiles)
296 fp = gmx_fio_fopen(tmpnm, bAppendFiles ? "a+" : "w+" );
301 gmx_fatal_set_log_file(fp);
303 /* Get some machine parameters */
304 gmx_gethostname(host, 256);
309 # ifdef GMX_NATIVE_WINDOWS
323 "-----------------------------------------------------------\n"
324 "Restarting from checkpoint, appending to previous log file.\n"
329 gmx_ctime_r(&t, timebuf, STRLEN);
332 "Log file opened on %s"
333 "Host: %s pid: %d nodeid: %d nnodes: %d\n",
334 timebuf, host, pid, cr->nodeid, cr->nnodes);
335 gmx_print_version_info(fp);
344 void gmx_log_close(FILE *fp)
348 gmx_fatal_set_log_file(NULL);
353 static void comm_args(const t_commrec *cr, int *argc, char ***argv)
359 gmx_bcast(sizeof(*argc), argc, cr);
364 snew(*argv, *argc+1);
368 fprintf(debug, "NODEID=%d argc=%d\n", cr->nodeid, *argc);
370 for (i = 0; (i < *argc); i++)
374 len = strlen((*argv)[i])+1;
376 gmx_bcast(sizeof(len), &len, cr);
379 snew((*argv)[i], len);
381 /*gmx_bcast(len*sizeof((*argv)[i][0]),(*argv)[i],cr);*/
382 gmx_bcast(len*sizeof(char), (*argv)[i], cr);
387 void init_multisystem(t_commrec *cr, int nsim, char **multidirs,
388 int nfile, const t_filenm fnm[], gmx_bool bParFn)
391 int nnodes, nnodpersim, sim, i, ftp;
394 MPI_Group mpi_group_world;
401 gmx_fatal(FARGS, "This binary is compiled without MPI support, can not do multiple simulations.");
406 if (nnodes % nsim != 0)
408 gmx_fatal(FARGS, "The number of nodes (%d) is not a multiple of the number of simulations (%d)", nnodes, nsim);
411 nnodpersim = nnodes/nsim;
412 sim = cr->nodeid/nnodpersim;
416 fprintf(debug, "We have %d simulations, %d nodes per simulation, local simulation is %d\n", nsim, nnodpersim, sim);
424 /* Create a communicator for the master nodes */
425 snew(rank, ms->nsim);
426 for (i = 0; i < ms->nsim; i++)
428 rank[i] = i*nnodpersim;
430 MPI_Comm_group(MPI_COMM_WORLD, &mpi_group_world);
431 MPI_Group_incl(mpi_group_world, nsim, rank, &ms->mpi_group_masters);
433 MPI_Comm_create(MPI_COMM_WORLD, ms->mpi_group_masters,
434 &ms->mpi_comm_masters);
436 #if !defined(GMX_THREAD_MPI) && !defined(MPI_IN_PLACE_EXISTS)
437 /* initialize the MPI_IN_PLACE replacement buffers */
439 ms->mpb->ibuf = NULL;
440 ms->mpb->libuf = NULL;
441 ms->mpb->fbuf = NULL;
442 ms->mpb->dbuf = NULL;
443 ms->mpb->ibuf_alloc = 0;
444 ms->mpb->libuf_alloc = 0;
445 ms->mpb->fbuf_alloc = 0;
446 ms->mpb->dbuf_alloc = 0;
451 /* Reduce the intra-simulation communication */
452 cr->sim_nodeid = cr->nodeid % nnodpersim;
453 cr->nnodes = nnodpersim;
455 MPI_Comm_split(MPI_COMM_WORLD, sim, cr->sim_nodeid, &cr->mpi_comm_mysim);
456 cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
457 cr->nodeid = cr->sim_nodeid;
462 fprintf(debug, "This is simulation %d", cr->ms->sim);
465 fprintf(debug, ", local number of nodes %d, local nodeid %d",
466 cr->nnodes, cr->sim_nodeid);
468 fprintf(debug, "\n\n");
475 fprintf(debug, "Changing to directory %s\n", multidirs[cr->ms->sim]);
477 gmx_chdir(multidirs[cr->ms->sim]);
481 /* Patch output and tpx, cpt and rerun input file names */
482 for (i = 0; (i < nfile); i++)
484 /* Because of possible multiple extensions per type we must look
485 * at the actual file name
487 if (is_output(&fnm[i]) ||
488 fnm[i].ftp == efTPX || fnm[i].ftp == efCPT ||
489 strcmp(fnm[i].opt, "-rerun") == 0)
491 ftp = fn2ftp(fnm[i].fns[0]);
492 par_fn(fnm[i].fns[0], ftp, cr, TRUE, FALSE, buf, 255);
493 sfree(fnm[i].fns[0]);
494 fnm[i].fns[0] = gmx_strdup(buf);
500 t_commrec *init_par(int gmx_unused *argc, char ***argv_ptr)
506 #if defined GMX_MPI && !defined GMX_THREAD_MPI
507 char **argv = argv_ptr ? *argv_ptr : NULL;
508 cr->sim_nodeid = gmx_setup(argc, argv, &cr->nnodes);
510 if (!PAR(cr) && (cr->sim_nodeid != 0))
512 gmx_comm("(!PAR(cr) && (cr->sim_nodeid != 0))");
515 cr->mpi_comm_mysim = MPI_COMM_WORLD;
516 cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
518 /* These should never be accessed */
519 cr->mpi_comm_mysim = NULL;
520 cr->mpi_comm_mygroup = NULL;
525 cr->nodeid = cr->sim_nodeid;
527 cr->duty = (DUTY_PP | DUTY_PME);
529 /* Communicate arguments if parallel */
530 #ifndef GMX_THREAD_MPI
533 comm_args(cr, argc, argv_ptr);
535 #endif /* GMX_THREAD_MPI */
538 #if !defined(GMX_THREAD_MPI) && !defined(MPI_IN_PLACE_EXISTS)
539 /* initialize the MPI_IN_PLACE replacement buffers */
541 cr->mpb->ibuf = NULL;
542 cr->mpb->libuf = NULL;
543 cr->mpb->fbuf = NULL;
544 cr->mpb->dbuf = NULL;
545 cr->mpb->ibuf_alloc = 0;
546 cr->mpb->libuf_alloc = 0;
547 cr->mpb->fbuf_alloc = 0;
548 cr->mpb->dbuf_alloc = 0;
555 t_commrec *init_par_threads(const t_commrec *cro)
557 #ifdef GMX_THREAD_MPI
561 /* make a thread-specific commrec */
563 /* now copy the whole thing, so settings like the number of PME nodes
567 /* and we start setting our own thread-specific values for things */
568 MPI_Initialized(&initialized);
571 gmx_comm("Initializing threads without comm");
573 /* once threads will be used together with MPI, we'll
574 fill the cr structure with distinct data here. This might even work: */
575 cr->sim_nodeid = gmx_setup(0, NULL, &cr->nnodes);
577 cr->mpi_comm_mysim = MPI_COMM_WORLD;
578 cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
579 cr->nodeid = cr->sim_nodeid;
580 cr->duty = (DUTY_PP | DUTY_PME);