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 "gmx_header_config.h"
47 #ifdef HAVE_SYS_TIME_H
52 /* windows-specific include for _chdir() */
58 #include "gmx_fatal.h"
69 #include "thread_mpi.h"
72 /* The source code in this file should be thread-safe.
73 Please keep it that way. */
80 #ifdef GMX_NATIVE_WINDOWS
85 /* Portable version of ctime_r implemented in src/gmxlib/string2.c, but we do not want it declared in public installed headers */
87 gmx_ctime_r(const time_t *clock,char *buf, int n);
93 static void par_fn(char *base,int ftp,const t_commrec *cr,
94 gmx_bool bAppendSimId,gmx_bool bAppendNodeId,
95 char buf[],int bufsize)
99 if((size_t)bufsize<(strlen(base)+10))
100 gmx_mem("Character buffer too small!");
102 /* Copy to buf, and strip extension */
104 buf[strlen(base) - strlen(ftp2ext(fn2ftp(base))) - 1] = '\0';
107 sprintf(buf+strlen(buf),"%d",cr->ms->sim);
111 sprintf(buf+strlen(buf),"%d",cr->nodeid);
115 /* Add extension again */
116 strcat(buf,(ftp == efTPX) ? "tpr" : (ftp == efEDR) ? "edr" : ftp2ext(ftp));
119 fprintf(debug, "node %d par_fn '%s'\n",cr->nodeid,buf);
120 if (fn2ftp(buf) == efLOG)
122 fprintf(debug,"log\n");
127 void check_multi_int(FILE *log,const gmx_multisim_t *ms,int val,
131 gmx_bool bCompatible;
134 fprintf(log,"Multi-checking %s ... ",name);
138 "check_multi_int called with a NULL communication pointer");
142 gmx_sumi_sim(ms->nsim,ibuf,ms);
145 for(p=1; p<ms->nsim; p++)
146 bCompatible = bCompatible && (ibuf[p-1] == ibuf[p]);
157 fprintf(log,"\n%s is not equal for all subsystems\n",name);
158 for(p=0; p<ms->nsim; p++)
159 fprintf(log," subsystem %d: %d\n",p,ibuf[p]);
161 gmx_fatal(FARGS,"The %d subsystems are not compatible\n",ms->nsim);
167 void check_multi_large_int(FILE *log,const gmx_multisim_t *ms,
168 gmx_large_int_t val, const char *name)
170 gmx_large_int_t *ibuf;
172 gmx_bool bCompatible;
175 fprintf(log,"Multi-checking %s ... ",name);
179 "check_multi_int called with a NULL communication pointer");
183 gmx_sumli_sim(ms->nsim,ibuf,ms);
186 for(p=1; p<ms->nsim; p++)
187 bCompatible = bCompatible && (ibuf[p-1] == ibuf[p]);
198 fprintf(log,"\n%s is not equal for all subsystems\n",name);
199 for(p=0; p<ms->nsim; p++)
202 /* first make the format string */
203 snprintf(strbuf, 255, " subsystem %%d: %s\n",
205 fprintf(log,strbuf,p,ibuf[p]);
208 gmx_fatal(FARGS,"The %d subsystems are not compatible\n",ms->nsim);
215 char *gmx_gethostname(char *name, size_t len)
219 gmx_incons("gmx_gethostname called with len<8");
222 if (gethostname(name, len-1) != 0)
224 strncpy(name, "unknown",8);
227 strncpy(name, "unknown",8);
234 void gmx_log_open(const char *lognm,const t_commrec *cr,gmx_bool bMasterOnly,
235 gmx_bool bAppendFiles, FILE** fplog)
238 char buf[256],host[256];
240 char timebuf[STRLEN];
246 /* Communicate the filename for logfile */
247 if (cr->nnodes > 1 && !bMasterOnly
248 #ifdef GMX_THREAD_MPI
249 /* With thread MPI the non-master log files are opened later
250 * when the files names are already known on all nodes.
258 len = strlen(lognm) + 1;
260 gmx_bcast(sizeof(len),&len,cr);
267 tmpnm=gmx_strdup(lognm);
269 gmx_bcast(len*sizeof(*tmpnm),tmpnm,cr);
273 tmpnm=gmx_strdup(lognm);
278 if (!bMasterOnly && !MASTER(cr))
280 /* Since log always ends with '.log' let's use this info */
281 par_fn(tmpnm,efLOG,cr,FALSE,!bMasterOnly,buf,255);
282 fp = gmx_fio_fopen(buf, bAppendFiles ? "a+" : "w+" );
284 else if (!bAppendFiles)
286 fp = gmx_fio_fopen(tmpnm, bAppendFiles ? "a+" : "w+" );
291 gmx_fatal_set_log_file(fp);
293 /* Get some machine parameters */
294 gmx_gethostname(host,256);
299 # ifdef GMX_NATIVE_WINDOWS
313 "-----------------------------------------------------------\n"
314 "Restarting from checkpoint, appending to previous log file.\n"
319 gmx_ctime_r(&t,timebuf,STRLEN);
322 "Log file opened on %s"
323 "Host: %s pid: %d nodeid: %d nnodes: %d\n",
324 timebuf,host,pid,cr->nodeid,cr->nnodes);
325 gmx_print_version_info(fp);
334 void gmx_log_close(FILE *fp)
337 gmx_fatal_set_log_file(NULL);
342 static void comm_args(const t_commrec *cr,int *argc,char ***argv)
347 gmx_bcast(sizeof(*argc),argc,cr);
351 fprintf(stderr,"NODEID=%d argc=%d\n",cr->nodeid,*argc);
352 for(i=0; (i<*argc); i++) {
354 len = strlen((*argv)[i])+1;
355 gmx_bcast(sizeof(len),&len,cr);
357 snew((*argv)[i],len);
358 /*gmx_bcast(len*sizeof((*argv)[i][0]),(*argv)[i],cr);*/
359 gmx_bcast(len*sizeof(char),(*argv)[i],cr);
364 void init_multisystem(t_commrec *cr,int nsim, char **multidirs,
365 int nfile, const t_filenm fnm[],gmx_bool bParFn)
368 int nnodes,nnodpersim,sim,i,ftp;
371 MPI_Group mpi_group_world;
378 gmx_fatal(FARGS,"This binary is compiled without MPI support, can not do multiple simulations.");
383 if (nnodes % nsim != 0)
385 gmx_fatal(FARGS,"The number of nodes (%d) is not a multiple of the number of simulations (%d)",nnodes,nsim);
388 nnodpersim = nnodes/nsim;
389 sim = cr->nodeid/nnodpersim;
393 fprintf(debug,"We have %d simulations, %d nodes per simulation, local simulation is %d\n",nsim,nnodpersim,sim);
401 /* Create a communicator for the master nodes */
403 for(i=0; i<ms->nsim; i++)
405 rank[i] = i*nnodpersim;
407 MPI_Comm_group(MPI_COMM_WORLD,&mpi_group_world);
408 MPI_Group_incl(mpi_group_world,nsim,rank,&ms->mpi_group_masters);
410 MPI_Comm_create(MPI_COMM_WORLD,ms->mpi_group_masters,
411 &ms->mpi_comm_masters);
413 #if !defined(GMX_THREAD_MPI) && !defined(MPI_IN_PLACE_EXISTS)
414 /* initialize the MPI_IN_PLACE replacement buffers */
420 ms->mpb->ibuf_alloc=0;
421 ms->mpb->libuf_alloc=0;
422 ms->mpb->fbuf_alloc=0;
423 ms->mpb->dbuf_alloc=0;
428 /* Reduce the intra-simulation communication */
429 cr->sim_nodeid = cr->nodeid % nnodpersim;
430 cr->nnodes = nnodpersim;
432 MPI_Comm_split(MPI_COMM_WORLD,sim,cr->sim_nodeid,&cr->mpi_comm_mysim);
433 cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
434 cr->nodeid = cr->sim_nodeid;
439 fprintf(debug,"This is simulation %d",cr->ms->sim);
442 fprintf(debug,", local number of nodes %d, local nodeid %d",
443 cr->nnodes,cr->sim_nodeid);
445 fprintf(debug,"\n\n");
453 fprintf(debug,"Changing to directory %s\n",multidirs[cr->ms->sim]);
455 if (chdir(multidirs[cr->ms->sim]) != 0)
457 gmx_fatal(FARGS, "Couldn't change directory to %s: %s",
458 multidirs[cr->ms->sim],
464 /* Patch output and tpx, cpt and rerun input file names */
465 for(i=0; (i<nfile); i++)
467 /* Because of possible multiple extensions per type we must look
468 * at the actual file name
470 if (is_output(&fnm[i]) ||
471 fnm[i].ftp == efTPX || fnm[i].ftp == efCPT ||
472 strcmp(fnm[i].opt,"-rerun") == 0)
474 ftp = fn2ftp(fnm[i].fns[0]);
475 par_fn(fnm[i].fns[0],ftp,cr,TRUE,FALSE,buf,255);
476 sfree(fnm[i].fns[0]);
477 fnm[i].fns[0] = gmx_strdup(buf);
483 t_commrec *init_par(int *argc,char ***argv_ptr)
492 argv = argv_ptr ? *argv_ptr : NULL;
494 #if defined GMX_MPI && !defined GMX_THREAD_MPI
495 cr->sim_nodeid = gmx_setup(argc,argv,&cr->nnodes);
497 if (!PAR(cr) && (cr->sim_nodeid != 0))
499 gmx_comm("(!PAR(cr) && (cr->sim_nodeid != 0))");
502 cr->mpi_comm_mysim = MPI_COMM_WORLD;
503 cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
505 /* These should never be accessed */
506 cr->mpi_comm_mysim = NULL;
507 cr->mpi_comm_mygroup = NULL;
512 cr->nodeid = cr->sim_nodeid;
514 cr->duty = (DUTY_PP | DUTY_PME);
516 /* Communicate arguments if parallel */
517 #ifndef GMX_THREAD_MPI
520 comm_args(cr,argc,argv_ptr);
522 #endif /* GMX_THREAD_MPI */
525 #if !defined(GMX_THREAD_MPI) && !defined(MPI_IN_PLACE_EXISTS)
526 /* initialize the MPI_IN_PLACE replacement buffers */
532 cr->mpb->ibuf_alloc=0;
533 cr->mpb->libuf_alloc=0;
534 cr->mpb->fbuf_alloc=0;
535 cr->mpb->dbuf_alloc=0;
542 t_commrec *init_par_threads(const t_commrec *cro)
544 #ifdef GMX_THREAD_MPI
548 /* make a thread-specific commrec */
550 /* now copy the whole thing, so settings like the number of PME nodes
554 /* and we start setting our own thread-specific values for things */
555 MPI_Initialized(&initialized);
558 gmx_comm("Initializing threads without comm");
560 /* once threads will be used together with MPI, we'll
561 fill the cr structure with distinct data here. This might even work: */
562 cr->sim_nodeid = gmx_setup(0,NULL, &cr->nnodes);
564 cr->mpi_comm_mysim = MPI_COMM_WORLD;
565 cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
566 cr->nodeid = cr->sim_nodeid;
567 cr->duty = (DUTY_PP | DUTY_PME);