src/gmxlib/main.c

   1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
   2  *
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  *          GROningen MAchine for Chemical Simulations
   9  *
  10  *                        VERSION 3.2.0
  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.
  15
  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.
  20  *
  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.
  27  *
  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.
  30  *
  31  * For more info, check our website at http://www.gromacs.org
  32  *
  33  * And Hey:
  34  * GROningen Mixture of Alchemy and Childrens' Stories
  35  */
  36 #ifdef HAVE_CONFIG_H
  37 #include <config.h>
  38 #endif
  39
  40 #include <stdio.h>
  41 #include <stdlib.h>
  42 #include <string.h>
  43 #include <limits.h>
  44 #include <time.h>
  45
  46 #ifdef HAVE_SYS_TIME_H
  47 #include <sys/time.h>
  48 #endif
  49
  50 #ifdef HAVE_DIRECT_H
  51 /* windows-specific include for _chdir() */
  52 #include <direct.h>
  53 #endif
  54
  55
  56 #include "smalloc.h"
  57 #include "gmx_fatal.h"
  58 #include "network.h"
  59 #include "main.h"
  60 #include "macros.h"
  61 #include "futil.h"
  62 #include "filenm.h"
  63 #include "mdrun.h"
  64 #include "gmxfio.h"
  65 #include "string2.h"
  66
  67 #ifdef GMX_THREADS
  68 #include "thread_mpi.h"
  69 #endif
  70
  71 /* The source code in this file should be thread-safe.
  72          Please keep it that way. */
  73
  74
  75 #ifdef HAVE_UNISTD_H
  76 #include <unistd.h>
  77 #endif
  78
  79 #if ((defined WIN32 || defined _WIN32 || defined WIN64 || defined _WIN64) && !defined __CYGWIN__ && !defined __CYGWIN32__)
  80 #include <process.h>
  81 #endif
  82
  83
  84 /* Portable version of ctime_r implemented in src/gmxlib/string2.c, but we do not want it declared in public installed headers */
  85 char *
  86 gmx_ctime_r(const time_t *clock,char *buf, int n);
  87
  88
  89 #define BUFSIZE 1024
  90
  91 /* this is not strictly thread-safe, but it's only written to at the beginning
  92    of the simulation, once by each thread with the same value. We assume
  93    that writing to an int is atomic.*/
  94 static gmx_bool parallel_env_val;
  95 #ifdef GMX_THREADS
  96 tMPI_Thread_mutex_t parallel_env_mutex=TMPI_THREAD_MUTEX_INITIALIZER;
  97 #endif
  98
  99
 100 /* returns 1 when running in a parallel environment, so could also be 1 if
 101    mdrun was started with: mpirun -np 1.
 102
 103    Use this function only to check whether a parallel environment has
 104    been initialized, for example when checking whether gmx_finalize()
 105    needs to be called. Use PAR(cr) to check whether the simulation actually
 106    has more than one node/thread.  */
 107 gmx_bool gmx_parallel_env_initialized(void)
 108 {
 109     gmx_bool ret;
 110 #ifdef GMX_THREADS
 111     tMPI_Thread_mutex_lock(&parallel_env_mutex);
 112 #endif
 113     ret=parallel_env_val;
 114 #ifdef GMX_THREADS
 115     tMPI_Thread_mutex_unlock(&parallel_env_mutex);
 116 #endif
 117     return ret;
 118 }
 119
 120 static void set_parallel_env(gmx_bool val)
 121 {
 122 #ifdef GMX_THREADS
 123     tMPI_Thread_mutex_lock(&parallel_env_mutex);
 124 #endif
 125     if (!parallel_env_val)
 126     {
 127         /* we only allow it to be set, not unset */
 128         parallel_env_val=val;
 129     }
 130 #ifdef GMX_THREADS
 131     tMPI_Thread_mutex_unlock(&parallel_env_mutex);
 132 #endif
 133 }
 134
 135
 136 static void par_fn(char *base,int ftp,const t_commrec *cr,
 137                    gmx_bool bAppendSimId,gmx_bool bAppendNodeId,
 138                    char buf[],int bufsize)
 139 {
 140   int n;
 141
 142   if((size_t)bufsize<(strlen(base)+10))
 143      gmx_mem("Character buffer too small!");
 144
 145   /* Copy to buf, and strip extension */
 146   strcpy(buf,base);
 147   buf[strlen(base) - strlen(ftp2ext(fn2ftp(base))) - 1] = '\0';
 148
 149   if (bAppendSimId) {
 150     sprintf(buf+strlen(buf),"%d",cr->ms->sim);
 151   }
 152   if (bAppendNodeId) {
 153     strcat(buf,"_node");
 154     sprintf(buf+strlen(buf),"%d",cr->nodeid);
 155   }
 156   strcat(buf,".");
 157
 158   /* Add extension again */
 159   strcat(buf,(ftp == efTPX) ? "tpr" : (ftp == efEDR) ? "edr" : ftp2ext(ftp));
 160   if (cr->nodeid == 0) {
 161     printf("node %d par_fn '%s'\n",cr->nodeid,buf);
 162     if (fn2ftp(buf) == efLOG) {
 163       printf("log\n");
 164     }
 165   }
 166 }
 167
 168 void check_multi_int(FILE *log,const gmx_multisim_t *ms,int val,
 169                      const char *name)
 170 {
 171   int  *ibuf,p;
 172   gmx_bool bCompatible;
 173
 174   if (NULL != log)
 175       fprintf(log,"Multi-checking %s ... ",name);
 176
 177   if (ms == NULL)
 178     gmx_fatal(FARGS,
 179               "check_multi_int called with a NULL communication pointer");
 180
 181   snew(ibuf,ms->nsim);
 182   ibuf[ms->sim] = val;
 183   gmx_sumi_sim(ms->nsim,ibuf,ms);
 184
 185   bCompatible = TRUE;
 186   for(p=1; p<ms->nsim; p++)
 187     bCompatible = bCompatible && (ibuf[p-1] == ibuf[p]);
 188
 189   if (bCompatible)
 190   {
 191       if (NULL != log)
 192           fprintf(log,"OK\n");
 193   }
 194   else
 195   {
 196       if (NULL != log)
 197       {
 198           fprintf(log,"\n%s is not equal for all subsystems\n",name);
 199           for(p=0; p<ms->nsim; p++)
 200               fprintf(log,"  subsystem %d: %d\n",p,ibuf[p]);
 201       }
 202       gmx_fatal(FARGS,"The %d subsystems are not compatible\n",ms->nsim);
 203   }
 204
 205   sfree(ibuf);
 206 }
 207
 208 void check_multi_large_int(FILE *log,const gmx_multisim_t *ms,
 209                            gmx_large_int_t val, const char *name)
 210 {
 211   gmx_large_int_t  *ibuf;
 212   int p;
 213   gmx_bool bCompatible;
 214
 215   if (NULL != log)
 216       fprintf(log,"Multi-checking %s ... ",name);
 217
 218   if (ms == NULL)
 219     gmx_fatal(FARGS,
 220               "check_multi_int called with a NULL communication pointer");
 221
 222   snew(ibuf,ms->nsim);
 223   ibuf[ms->sim] = val;
 224   gmx_sumli_sim(ms->nsim,ibuf,ms);
 225
 226   bCompatible = TRUE;
 227   for(p=1; p<ms->nsim; p++)
 228     bCompatible = bCompatible && (ibuf[p-1] == ibuf[p]);
 229
 230   if (bCompatible)
 231   {
 232       if (NULL != log)
 233           fprintf(log,"OK\n");
 234   }
 235   else
 236   {
 237       if (NULL != log)
 238       {
 239           fprintf(log,"\n%s is not equal for all subsystems\n",name);
 240           for(p=0; p<ms->nsim; p++)
 241           {
 242               char strbuf[255];
 243               /* first make the format string */
 244               snprintf(strbuf, 255, "  subsystem %%d: %s\n",
 245                        gmx_large_int_pfmt);
 246               fprintf(log,strbuf,p,ibuf[p]);
 247           }
 248       }
 249       gmx_fatal(FARGS,"The %d subsystems are not compatible\n",ms->nsim);
 250   }
 251
 252   sfree(ibuf);
 253 }
 254
 255
 256 void gmx_log_open(const char *lognm,const t_commrec *cr,gmx_bool bMasterOnly,
 257                    unsigned long Flags, FILE** fplog)
 258 {
 259     int  len,testlen,pid;
 260     char buf[256],host[256];
 261     time_t t;
 262     char timebuf[STRLEN];
 263     FILE *fp=*fplog;
 264     char *tmpnm;
 265
 266     gmx_bool bAppend = Flags & MD_APPENDFILES;
 267
 268     debug_gmx();
 269
 270     /* Communicate the filename for logfile */
 271     if (cr->nnodes > 1 && !bMasterOnly
 272 #ifdef GMX_THREADS
 273         /* With thread MPI the non-master log files are opened later
 274          * when the files names are already known on all nodes.
 275          */
 276         && FALSE
 277 #endif
 278         )
 279     {
 280         if (MASTER(cr))
 281         {
 282             len = strlen(lognm) + 1;
 283         }
 284         gmx_bcast(sizeof(len),&len,cr);
 285         if (!MASTER(cr))
 286         {
 287             snew(tmpnm,len+8);
 288         }
 289         else
 290         {
 291             tmpnm=gmx_strdup(lognm);
 292         }
 293         gmx_bcast(len*sizeof(*tmpnm),tmpnm,cr);
 294     }
 295     else
 296     {
 297         tmpnm=gmx_strdup(lognm);
 298     }
 299
 300     debug_gmx();
 301
 302     if (!bMasterOnly && !MASTER(cr))
 303     {
 304         /* Since log always ends with '.log' let's use this info */
 305         par_fn(tmpnm,efLOG,cr,FALSE,!bMasterOnly,buf,255);
 306         fp = gmx_fio_fopen(buf, bAppend ? "a+" : "w+" );
 307     }
 308     else if (!bAppend)
 309     {
 310         fp = gmx_fio_fopen(tmpnm, bAppend ? "a+" : "w+" );
 311     }
 312
 313     sfree(tmpnm);
 314
 315     gmx_fatal_set_log_file(fp);
 316
 317     /* Get some machine parameters */
 318 #ifdef HAVE_UNISTD_H
 319     if (gethostname(host,255) != 0)
 320     {
 321         sprintf(host,"unknown");
 322     }
 323 #else
 324     sprintf(host,"unknown");
 325 #endif
 326
 327     time(&t);
 328
 329 #ifndef NO_GETPID
 330 #   if ((defined WIN32 || defined _WIN32 || defined WIN64 || defined _WIN64) && !defined __CYGWIN__ && !defined __CYGWIN32__)
 331     pid = _getpid();
 332 #   else
 333     pid = getpid();
 334 #   endif
 335 #else
 336         pid = 0;
 337 #endif
 338
 339     if (bAppend)
 340     {
 341         fprintf(fp,
 342                 "\n"
 343                 "\n"
 344                 "-----------------------------------------------------------\n"
 345                 "Restarting from checkpoint, appending to previous log file.\n"
 346                 "\n"
 347             );
 348     }
 349
 350     gmx_ctime_r(&t,timebuf,STRLEN);
 351
 352     fprintf(fp,
 353             "Log file opened on %s"
 354             "Host: %s  pid: %d  nodeid: %d  nnodes:  %d\n",
 355             timebuf,host,pid,cr->nodeid,cr->nnodes);
 356
 357 #if (defined BUILD_MACHINE && defined BUILD_TIME && defined BUILD_USER)
 358     fprintf(fp,
 359             "The Gromacs distribution was built %s by\n"
 360             "%s (%s)\n\n\n",BUILD_TIME,BUILD_USER,BUILD_MACHINE);
 361 #endif
 362
 363     fflush(fp);
 364     debug_gmx();
 365
 366     *fplog = fp;
 367 }
 368
 369 void gmx_log_close(FILE *fp)
 370 {
 371   if (fp) {
 372     gmx_fatal_set_log_file(NULL);
 373     gmx_fio_fclose(fp);
 374   }
 375 }
 376
 377 static void comm_args(const t_commrec *cr,int *argc,char ***argv)
 378 {
 379   int i,len;
 380
 381   if ((cr) && PAR(cr))
 382     gmx_bcast(sizeof(*argc),argc,cr);
 383
 384   if (!MASTER(cr))
 385     snew(*argv,*argc+1);
 386   fprintf(stderr,"NODEID=%d argc=%d\n",cr->nodeid,*argc);
 387   for(i=0; (i<*argc); i++) {
 388     if (MASTER(cr))
 389       len = strlen((*argv)[i])+1;
 390     gmx_bcast(sizeof(len),&len,cr);
 391     if (!MASTER(cr))
 392       snew((*argv)[i],len);
 393     /*gmx_bcast(len*sizeof((*argv)[i][0]),(*argv)[i],cr);*/
 394     gmx_bcast(len*sizeof(char),(*argv)[i],cr);
 395   }
 396   debug_gmx();
 397 }
 398
 399 void init_multisystem(t_commrec *cr,int nsim, char **multidirs,
 400                       int nfile, const t_filenm fnm[],gmx_bool bParFn)
 401 {
 402     gmx_multisim_t *ms;
 403     int  nnodes,nnodpersim,sim,i,ftp;
 404     char buf[256];
 405 #ifdef GMX_MPI
 406     MPI_Group mpi_group_world;
 407 #endif
 408     int *rank;
 409
 410 #ifndef GMX_MPI
 411     if (nsim > 1)
 412     {
 413         gmx_fatal(FARGS,"This binary is compiled without MPI support, can not do multiple simulations.");
 414     }
 415 #endif
 416
 417     nnodes  = cr->nnodes;
 418     if (nnodes % nsim != 0)
 419     {
 420         gmx_fatal(FARGS,"The number of nodes (%d) is not a multiple of the number of simulations (%d)",nnodes,nsim);
 421     }
 422
 423     nnodpersim = nnodes/nsim;
 424     sim = cr->nodeid/nnodpersim;
 425
 426     if (debug)
 427     {
 428         fprintf(debug,"We have %d simulations, %d nodes per simulation, local simulation is %d\n",nsim,nnodpersim,sim);
 429     }
 430
 431     snew(ms,1);
 432     cr->ms = ms;
 433     ms->nsim = nsim;
 434     ms->sim  = sim;
 435 #ifdef GMX_MPI
 436     /* Create a communicator for the master nodes */
 437     snew(rank,ms->nsim);
 438     for(i=0; i<ms->nsim; i++)
 439     {
 440         rank[i] = i*nnodpersim;
 441     }
 442     MPI_Comm_group(MPI_COMM_WORLD,&mpi_group_world);
 443     MPI_Group_incl(mpi_group_world,nsim,rank,&ms->mpi_group_masters);
 444     sfree(rank);
 445     MPI_Comm_create(MPI_COMM_WORLD,ms->mpi_group_masters,
 446                     &ms->mpi_comm_masters);
 447
 448 #if !defined(GMX_THREADS) && !defined(MPI_IN_PLACE_EXISTS)
 449     /* initialize the MPI_IN_PLACE replacement buffers */
 450     snew(ms->mpb, 1);
 451     ms->mpb->ibuf=NULL;
 452     ms->mpb->fbuf=NULL;
 453     ms->mpb->dbuf=NULL;
 454     ms->mpb->ibuf_alloc=0;
 455     ms->mpb->fbuf_alloc=0;
 456     ms->mpb->dbuf_alloc=0;
 457 #endif
 458
 459 #endif
 460
 461     /* Reduce the intra-simulation communication */
 462     cr->sim_nodeid = cr->nodeid % nnodpersim;
 463     cr->nnodes = nnodpersim;
 464 #ifdef GMX_MPI
 465     MPI_Comm_split(MPI_COMM_WORLD,sim,cr->sim_nodeid,&cr->mpi_comm_mysim);
 466     cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
 467     cr->nodeid = cr->sim_nodeid;
 468 #endif
 469
 470     if (debug)
 471     {
 472         fprintf(debug,"This is simulation %d",cr->ms->sim);
 473         if (PAR(cr))
 474         {
 475             fprintf(debug,", local number of nodes %d, local nodeid %d",
 476                     cr->nnodes,cr->sim_nodeid);
 477         }
 478         fprintf(debug,"\n\n");
 479     }
 480
 481     if (multidirs)
 482     {
 483         int ret;
 484         if (debug)
 485         {
 486             fprintf(debug,"Changing to directory %s\n",multidirs[cr->ms->sim]);
 487         }
 488         if (chdir(multidirs[cr->ms->sim]) != 0)
 489         {
 490             gmx_fatal(FARGS, "Couldn't change directory to %s: %s",
 491                       multidirs[cr->ms->sim],
 492                       strerror(errno));
 493         }
 494     }
 495     else if (bParFn)
 496     {
 497         /* Patch output and tpx, cpt and rerun input file names */
 498         for(i=0; (i<nfile); i++)
 499         {
 500             /* Because of possible multiple extensions per type we must look
 501              * at the actual file name
 502              */
 503             if (is_output(&fnm[i]) ||
 504                 fnm[i].ftp == efTPX || fnm[i].ftp == efCPT ||
 505                 strcmp(fnm[i].opt,"-rerun") == 0)
 506             {
 507                 ftp = fn2ftp(fnm[i].fns[0]);
 508                 par_fn(fnm[i].fns[0],ftp,cr,TRUE,FALSE,buf,255);
 509                 sfree(fnm[i].fns[0]);
 510                 fnm[i].fns[0] = gmx_strdup(buf);
 511             }
 512         }
 513     }
 514 }
 515
 516 t_commrec *init_par(int *argc,char ***argv_ptr)
 517 {
 518     t_commrec *cr;
 519     char      **argv;
 520     int       i;
 521     gmx_bool      pe=FALSE;
 522
 523     snew(cr,1);
 524
 525     argv = *argv_ptr;
 526
 527 #ifdef GMX_MPI
 528 #ifdef GMX_LIB_MPI
 529     pe = TRUE;
 530 #ifdef GMX_CHECK_MPI_ENV
 531     /* Do not use MPI calls when env.var. GMX_CHECK_MPI_ENV is not set */
 532     if (getenv(GMX_CHECK_MPI_ENV) == NULL)
 533         pe = FALSE;
 534 #endif /* GMX_CHECK_MPI_ENV */
 535 #endif /* GMX_LIB_MPI  */
 536     set_parallel_env(pe);
 537     if (pe) {
 538         cr->sim_nodeid = gmx_setup(argc,argv,&cr->nnodes);
 539     } else {
 540         cr->nnodes     = 1;
 541         cr->sim_nodeid = 0;
 542     }
 543 #else /* GMX_MPI */
 544     pe=FALSE;
 545     set_parallel_env(pe);
 546     cr->sim_nodeid   = 0;
 547     cr->nnodes       = 1;
 548 #endif /* GMX_MPI */
 549
 550     if (!PAR(cr) && (cr->sim_nodeid != 0))
 551         gmx_comm("(!PAR(cr) && (cr->sim_nodeid != 0))");
 552
 553     if (PAR(cr))
 554     {
 555 #ifdef GMX_MPI
 556         cr->mpi_comm_mysim = MPI_COMM_WORLD;
 557         cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
 558 #endif /* GMX_MPI */
 559     }
 560     cr->nodeid = cr->sim_nodeid;
 561
 562     cr->duty = (DUTY_PP | DUTY_PME);
 563
 564     /* Communicate arguments if parallel */
 565 #ifndef GMX_THREADS
 566     if (PAR(cr))
 567         comm_args(cr,argc,argv_ptr);
 568 #endif /* GMX_THREADS */
 569
 570 #ifdef GMX_MPI
 571 #if !defined(GMX_THREADS) && !defined(MPI_IN_PLACE_EXISTS)
 572   /* initialize the MPI_IN_PLACE replacement buffers */
 573   snew(cr->mpb, 1);
 574   cr->mpb->ibuf=NULL;
 575   cr->mpb->fbuf=NULL;
 576   cr->mpb->dbuf=NULL;
 577   cr->mpb->ibuf_alloc=0;
 578   cr->mpb->fbuf_alloc=0;
 579   cr->mpb->dbuf_alloc=0;
 580 #endif
 581 #endif
 582
 583     return cr;
 584 }
 585
 586 t_commrec *init_par_threads(const t_commrec *cro)
 587 {
 588 #ifdef GMX_THREADS
 589     int initialized;
 590     t_commrec *cr;
 591
 592     /* make a thread-specific commrec */
 593     snew(cr,1);
 594     /* now copy the whole thing, so settings like the number of PME nodes
 595        get propagated. */
 596     *cr=*cro;
 597
 598     /* and we start setting our own thread-specific values for things */
 599     MPI_Initialized(&initialized);
 600     if (!initialized)
 601         gmx_comm("Initializing threads without comm");
 602     set_parallel_env(TRUE);
 603     /* once threads will be used together with MPI, we'll
 604        fill the cr structure with distinct data here. This might even work: */
 605     cr->sim_nodeid = gmx_setup(0,NULL, &cr->nnodes);
 606
 607     cr->mpi_comm_mysim = MPI_COMM_WORLD;
 608     cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
 609     cr->nodeid = cr->sim_nodeid;
 610     cr->duty = (DUTY_PP | DUTY_PME);
 611
 612     return cr;
 613 #else
 614     return NULL;
 615 #endif
 616 }
 617
 618
 619 t_commrec *init_cr_nopar(void)
 620 {
 621     t_commrec *cr;
 622
 623     snew(cr,1);
 624
 625     cr->nnodes     = 1;
 626     /* cr->nthreads   = 1; */
 627     cr->sim_nodeid = 0;
 628     cr->nodeid     = 0;
 629     /* cr->threadid   = 0; */
 630     cr->duty       = (DUTY_PP | DUTY_PME);
 631
 632     return cr;
 633 }