src/mdlib/sim_util.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  * GROwing Monsters And Cloning Shrimps
  35  */
  36 #ifdef HAVE_CONFIG_H
  37 #include <config.h>
  38 #endif
  39
  40 #ifdef GMX_CRAY_XT3
  41 #include<catamount/dclock.h>
  42 #endif
  43
  44
  45 #include <stdio.h>
  46 #include <time.h>
  47 #ifdef HAVE_SYS_TIME_H
  48 #include <sys/time.h>
  49 #endif
  50 #include <math.h>
  51 #include "typedefs.h"
  52 #include "string2.h"
  53 #include "gmxfio.h"
  54 #include "smalloc.h"
  55 #include "names.h"
  56 #include "confio.h"
  57 #include "mvdata.h"
  58 #include "txtdump.h"
  59 #include "pbc.h"
  60 #include "chargegroup.h"
  61 #include "vec.h"
  62 #include "time.h"
  63 #include "nrnb.h"
  64 #include "mshift.h"
  65 #include "mdrun.h"
  66 #include "update.h"
  67 #include "physics.h"
  68 #include "main.h"
  69 #include "mdatoms.h"
  70 #include "force.h"
  71 #include "bondf.h"
  72 #include "pme.h"
  73 #include "pppm.h"
  74 #include "disre.h"
  75 #include "orires.h"
  76 #include "network.h"
  77 #include "calcmu.h"
  78 #include "constr.h"
  79 #include "xvgr.h"
  80 #include "trnio.h"
  81 #include "xtcio.h"
  82 #include "copyrite.h"
  83
  84 #include "mpelogging.h"
  85 #include "domdec.h"
  86 #include "partdec.h"
  87 #include "gmx_wallcycle.h"
  88 #include "genborn.h"
  89
  90 #ifdef GMX_LIB_MPI
  91 #include <mpi.h>
  92 #endif
  93 #ifdef GMX_THREADS
  94 #include "tmpi.h"
  95 #endif
  96
  97 #include "qmmm.h"
  98
  99 #if 0
 100 typedef struct gmx_timeprint {
 101
 102 } t_gmx_timeprint;
 103 #endif
 104
 105 /* Portable version of ctime_r implemented in src/gmxlib/string2.c, but we do not want it declared in public installed headers */
 106 char *
 107 gmx_ctime_r(const time_t *clock,char *buf, int n);
 108
 109
 110 double
 111 gmx_gettime()
 112 {
 113 #ifdef HAVE_GETTIMEOFDAY
 114         struct timeval t;
 115         struct timezone tz = { 0,0 };
 116         double seconds;
 117
 118         gettimeofday(&t,&tz);
 119
 120         seconds = (double) t.tv_sec + 1e-6*(double)t.tv_usec;
 121
 122         return seconds;
 123 #else
 124         double  seconds;
 125
 126         seconds = time(NULL);
 127
 128         return seconds;
 129 #endif
 130 }
 131
 132
 133 #define difftime(end,start) ((double)(end)-(double)(start))
 134
 135 void print_time(FILE *out,gmx_runtime_t *runtime,gmx_large_int_t step,
 136                 t_inputrec *ir, t_commrec *cr)
 137 {
 138     time_t finish;
 139     char   timebuf[STRLEN];
 140     double dt;
 141     char buf[48];
 142
 143 #ifndef GMX_THREADS
 144     if (!PAR(cr))
 145 #endif
 146     {
 147         fprintf(out,"\r");
 148     }
 149     fprintf(out,"step %s",gmx_step_str(step,buf));
 150     if ((step >= ir->nstlist))
 151     {
 152         if ((ir->nstlist == 0) || ((step % ir->nstlist) == 0))
 153         {
 154             /* We have done a full cycle let's update time_per_step */
 155             runtime->last = gmx_gettime();
 156             dt = difftime(runtime->last,runtime->real);
 157             runtime->time_per_step = dt/(step - ir->init_step + 1);
 158         }
 159         dt = (ir->nsteps + ir->init_step - step)*runtime->time_per_step;
 160
 161         if (ir->nsteps >= 0)
 162         {
 163             if (dt >= 300)
 164             {
 165                 finish = (time_t) (runtime->last + dt);
 166                 gmx_ctime_r(&finish,timebuf,STRLEN);
 167                 sprintf(buf,"%s",timebuf);
 168                 buf[strlen(buf)-1]='\0';
 169                 fprintf(out,", will finish %s",buf);
 170             }
 171             else
 172                 fprintf(out,", remaining runtime: %5d s          ",(int)dt);
 173         }
 174         else
 175         {
 176             fprintf(out," performance: %.1f ns/day    ",
 177                     ir->delta_t/1000*24*60*60/runtime->time_per_step);
 178         }
 179     }
 180 #ifndef GMX_THREADS
 181     if (PAR(cr))
 182     {
 183         fprintf(out,"\n");
 184     }
 185 #endif
 186
 187     fflush(out);
 188 }
 189
 190 #ifdef NO_CLOCK
 191 #define clock() -1
 192 #endif
 193
 194 static double set_proctime(gmx_runtime_t *runtime)
 195 {
 196     double diff;
 197 #ifdef GMX_CRAY_XT3
 198     double prev;
 199
 200     prev = runtime->proc;
 201     runtime->proc = dclock();
 202
 203     diff = runtime->proc - prev;
 204 #else
 205     clock_t prev;
 206
 207     prev = runtime->proc;
 208     runtime->proc = clock();
 209
 210     diff = (double)(runtime->proc - prev)/(double)CLOCKS_PER_SEC;
 211 #endif
 212     if (diff < 0)
 213     {
 214         /* The counter has probably looped, ignore this data */
 215         diff = 0;
 216     }
 217
 218     return diff;
 219 }
 220
 221 void runtime_start(gmx_runtime_t *runtime)
 222 {
 223     runtime->real = gmx_gettime();
 224     runtime->proc          = 0;
 225     set_proctime(runtime);
 226     runtime->realtime      = 0;
 227     runtime->proctime      = 0;
 228     runtime->last          = 0;
 229     runtime->time_per_step = 0;
 230 }
 231
 232 void runtime_end(gmx_runtime_t *runtime)
 233 {
 234     double now;
 235
 236     now = gmx_gettime();
 237
 238     runtime->proctime += set_proctime(runtime);
 239     runtime->realtime  = now - runtime->real;
 240     runtime->real      = now;
 241 }
 242
 243 void runtime_upd_proc(gmx_runtime_t *runtime)
 244 {
 245     runtime->proctime += set_proctime(runtime);
 246 }
 247
 248 void print_date_and_time(FILE *fplog,int nodeid,const char *title,
 249                          const gmx_runtime_t *runtime)
 250 {
 251     int i;
 252     char timebuf[STRLEN];
 253     char time_string[STRLEN];
 254     time_t tmptime;
 255
 256     if (fplog)
 257     {
 258         if (runtime != NULL)
 259         {
 260             tmptime = (time_t) runtime->real;
 261             gmx_ctime_r(&tmptime,timebuf,STRLEN);
 262         }
 263         else
 264         {
 265             tmptime = (time_t) gmx_gettime();
 266             gmx_ctime_r(&tmptime,timebuf,STRLEN);
 267         }
 268         for(i=0; timebuf[i]>=' '; i++)
 269         {
 270             time_string[i]=timebuf[i];
 271         }
 272         time_string[i]='\0';
 273
 274         fprintf(fplog,"%s on node %d %s\n",title,nodeid,time_string);
 275     }
 276 }
 277
 278 static void sum_forces(int start,int end,rvec f[],rvec flr[])
 279 {
 280   int i;
 281
 282   if (gmx_debug_at) {
 283     pr_rvecs(debug,0,"fsr",f+start,end-start);
 284     pr_rvecs(debug,0,"flr",flr+start,end-start);
 285   }
 286   for(i=start; (i<end); i++)
 287     rvec_inc(f[i],flr[i]);
 288 }
 289
 290 /*
 291  * calc_f_el calculates forces due to an electric field.
 292  *
 293  * force is kJ mol^-1 nm^-1 = e * kJ mol^-1 nm^-1 / e
 294  *
 295  * Et[] contains the parameters for the time dependent
 296  * part of the field (not yet used).
 297  * Ex[] contains the parameters for
 298  * the spatial dependent part of the field. You can have cool periodic
 299  * fields in principle, but only a constant field is supported
 300  * now.
 301  * The function should return the energy due to the electric field
 302  * (if any) but for now returns 0.
 303  *
 304  * WARNING:
 305  * There can be problems with the virial.
 306  * Since the field is not self-consistent this is unavoidable.
 307  * For neutral molecules the virial is correct within this approximation.
 308  * For neutral systems with many charged molecules the error is small.
 309  * But for systems with a net charge or a few charged molecules
 310  * the error can be significant when the field is high.
 311  * Solution: implement a self-consitent electric field into PME.
 312  */
 313 static void calc_f_el(FILE *fp,int  start,int homenr,
 314                       real charge[],rvec x[],rvec f[],
 315                       t_cosines Ex[],t_cosines Et[],double t)
 316 {
 317     rvec Ext;
 318     real t0;
 319     int  i,m;
 320
 321     for(m=0; (m<DIM); m++)
 322     {
 323         if (Et[m].n > 0)
 324         {
 325             if (Et[m].n == 3)
 326             {
 327                 t0 = Et[m].a[1];
 328                 Ext[m] = cos(Et[m].a[0]*(t-t0))*exp(-sqr(t-t0)/(2.0*sqr(Et[m].a[2])));
 329             }
 330             else
 331             {
 332                 Ext[m] = cos(Et[m].a[0]*t);
 333             }
 334         }
 335         else
 336         {
 337             Ext[m] = 1.0;
 338         }
 339         if (Ex[m].n > 0)
 340         {
 341             /* Convert the field strength from V/nm to MD-units */
 342             Ext[m] *= Ex[m].a[0]*FIELDFAC;
 343             for(i=start; (i<start+homenr); i++)
 344                 f[i][m] += charge[i]*Ext[m];
 345         }
 346         else
 347         {
 348             Ext[m] = 0;
 349         }
 350     }
 351     if (fp != NULL)
 352     {
 353         fprintf(fp,"%10g  %10g  %10g  %10g #FIELD\n",t,
 354                 Ext[XX]/FIELDFAC,Ext[YY]/FIELDFAC,Ext[ZZ]/FIELDFAC);
 355     }
 356 }
 357
 358 static void calc_virial(FILE *fplog,int start,int homenr,rvec x[],rvec f[],
 359                         tensor vir_part,t_graph *graph,matrix box,
 360                         t_nrnb *nrnb,const t_forcerec *fr,int ePBC)
 361 {
 362   int i,j;
 363   tensor virtest;
 364
 365   /* The short-range virial from surrounding boxes */
 366   clear_mat(vir_part);
 367   calc_vir(fplog,SHIFTS,fr->shift_vec,fr->fshift,vir_part,ePBC==epbcSCREW,box);
 368   inc_nrnb(nrnb,eNR_VIRIAL,SHIFTS);
 369
 370   /* Calculate partial virial, for local atoms only, based on short range.
 371    * Total virial is computed in global_stat, called from do_md
 372    */
 373   f_calc_vir(fplog,start,start+homenr,x,f,vir_part,graph,box);
 374   inc_nrnb(nrnb,eNR_VIRIAL,homenr);
 375
 376   /* Add position restraint contribution */
 377   for(i=0; i<DIM; i++) {
 378     vir_part[i][i] += fr->vir_diag_posres[i];
 379   }
 380
 381   /* Add wall contribution */
 382   for(i=0; i<DIM; i++) {
 383     vir_part[i][ZZ] += fr->vir_wall_z[i];
 384   }
 385
 386   if (debug)
 387     pr_rvecs(debug,0,"vir_part",vir_part,DIM);
 388 }
 389
 390 static void print_large_forces(FILE *fp,t_mdatoms *md,t_commrec *cr,
 391                                gmx_large_int_t step,real pforce,rvec *x,rvec *f)
 392 {
 393   int  i;
 394   real pf2,fn2;
 395   char buf[STEPSTRSIZE];
 396
 397   pf2 = sqr(pforce);
 398   for(i=md->start; i<md->start+md->homenr; i++) {
 399     fn2 = norm2(f[i]);
 400     /* We also catch NAN, if the compiler does not optimize this away. */
 401     if (fn2 >= pf2 || fn2 != fn2) {
 402       fprintf(fp,"step %s  atom %6d  x %8.3f %8.3f %8.3f  force %12.5e\n",
 403               gmx_step_str(step,buf),
 404               ddglatnr(cr->dd,i),x[i][XX],x[i][YY],x[i][ZZ],sqrt(fn2));
 405     }
 406   }
 407 }
 408
 409 void do_force(FILE *fplog,t_commrec *cr,
 410               t_inputrec *inputrec,
 411               gmx_large_int_t step,t_nrnb *nrnb,gmx_wallcycle_t wcycle,
 412               gmx_localtop_t *top,
 413               gmx_mtop_t *mtop,
 414               gmx_groups_t *groups,
 415               matrix box,rvec x[],history_t *hist,
 416               rvec f[],
 417               tensor vir_force,
 418               t_mdatoms *mdatoms,
 419               gmx_enerdata_t *enerd,t_fcdata *fcd,
 420               real lambda,t_graph *graph,
 421               t_forcerec *fr,gmx_vsite_t *vsite,rvec mu_tot,
 422               double t,FILE *field,gmx_edsam_t ed,
 423               gmx_bool bBornRadii,
 424               int flags)
 425 {
 426     int    cg0,cg1,i,j;
 427     int    start,homenr;
 428     double mu[2*DIM];
 429     gmx_bool   bSepDVDL,bStateChanged,bNS,bFillGrid,bCalcCGCM,bBS;
 430     gmx_bool   bDoLongRange,bDoForces,bSepLRF;
 431     matrix boxs;
 432     real   e,v,dvdl;
 433     t_pbc  pbc;
 434     float  cycles_ppdpme,cycles_pme,cycles_seppme,cycles_force;
 435
 436     start  = mdatoms->start;
 437     homenr = mdatoms->homenr;
 438
 439     bSepDVDL = (fr->bSepDVDL && do_per_step(step,inputrec->nstlog));
 440
 441     clear_mat(vir_force);
 442
 443     if (PARTDECOMP(cr))
 444     {
 445         pd_cg_range(cr,&cg0,&cg1);
 446     }
 447     else
 448     {
 449         cg0 = 0;
 450         if (DOMAINDECOMP(cr))
 451         {
 452             cg1 = cr->dd->ncg_tot;
 453         }
 454         else
 455         {
 456             cg1 = top->cgs.nr;
 457         }
 458         if (fr->n_tpi > 0)
 459         {
 460             cg1--;
 461         }
 462     }
 463
 464     bStateChanged = (flags & GMX_FORCE_STATECHANGED);
 465     bNS           = (flags & GMX_FORCE_NS) && (fr->bAllvsAll==FALSE);
 466     bFillGrid     = (bNS && bStateChanged);
 467     bCalcCGCM     = (bFillGrid && !DOMAINDECOMP(cr));
 468     bDoLongRange  = (fr->bTwinRange && bNS && (flags & GMX_FORCE_DOLR));
 469     bDoForces     = (flags & GMX_FORCE_FORCES);
 470     bSepLRF       = (bDoLongRange && bDoForces && (flags & GMX_FORCE_SEPLRF));
 471
 472     if (bStateChanged)
 473     {
 474         update_forcerec(fplog,fr,box);
 475
 476         /* Calculate total (local) dipole moment in a temporary common array.
 477          * This makes it possible to sum them over nodes faster.
 478          */
 479         calc_mu(start,homenr,
 480                 x,mdatoms->chargeA,mdatoms->chargeB,mdatoms->nChargePerturbed,
 481                 mu,mu+DIM);
 482     }
 483
 484   if (fr->ePBC != epbcNONE) {
 485     /* Compute shift vectors every step,
 486      * because of pressure coupling or box deformation!
 487      */
 488     if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
 489       calc_shifts(box,fr->shift_vec);
 490
 491     if (bCalcCGCM) {
 492       put_charge_groups_in_box(fplog,cg0,cg1,fr->ePBC,box,
 493                                &(top->cgs),x,fr->cg_cm);
 494       inc_nrnb(nrnb,eNR_CGCM,homenr);
 495       inc_nrnb(nrnb,eNR_RESETX,cg1-cg0);
 496     }
 497     else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph) {
 498       unshift_self(graph,box,x);
 499     }
 500   }
 501   else if (bCalcCGCM) {
 502     calc_cgcm(fplog,cg0,cg1,&(top->cgs),x,fr->cg_cm);
 503     inc_nrnb(nrnb,eNR_CGCM,homenr);
 504   }
 505
 506   if (bCalcCGCM) {
 507     if (PAR(cr)) {
 508       move_cgcm(fplog,cr,fr->cg_cm);
 509     }
 510     if (gmx_debug_at)
 511       pr_rvecs(debug,0,"cgcm",fr->cg_cm,top->cgs.nr);
 512   }
 513
 514 #ifdef GMX_MPI
 515   if (!(cr->duty & DUTY_PME)) {
 516     /* Send particle coordinates to the pme nodes.
 517      * Since this is only implemented for domain decomposition
 518      * and domain decomposition does not use the graph,
 519      * we do not need to worry about shifting.
 520      */
 521
 522     wallcycle_start(wcycle,ewcPP_PMESENDX);
 523     GMX_MPE_LOG(ev_send_coordinates_start);
 524
 525     bBS = (inputrec->nwall == 2);
 526     if (bBS) {
 527       copy_mat(box,boxs);
 528       svmul(inputrec->wall_ewald_zfac,boxs[ZZ],boxs[ZZ]);
 529     }
 530
 531     gmx_pme_send_x(cr,bBS ? boxs : box,x,
 532                    mdatoms->nChargePerturbed,lambda,
 533 /* FIX ME after 4.5 */
 534 /* we are using gmx_bool of type char */
 535                    ( flags & GMX_FORCE_VIRIAL) != 0,step);
 536
 537     GMX_MPE_LOG(ev_send_coordinates_finish);
 538     wallcycle_stop(wcycle,ewcPP_PMESENDX);
 539   }
 540 #endif /* GMX_MPI */
 541
 542     /* Communicate coordinates and sum dipole if necessary */
 543     if (PAR(cr))
 544     {
 545         wallcycle_start(wcycle,ewcMOVEX);
 546         if (DOMAINDECOMP(cr))
 547         {
 548             dd_move_x(cr->dd,box,x);
 549         }
 550         else
 551         {
 552             move_x(fplog,cr,GMX_LEFT,GMX_RIGHT,x,nrnb);
 553         }
 554         /* When we don't need the total dipole we sum it in global_stat */
 555         if (bStateChanged && NEED_MUTOT(*inputrec))
 556         {
 557             gmx_sumd(2*DIM,mu,cr);
 558         }
 559         wallcycle_stop(wcycle,ewcMOVEX);
 560     }
 561     if (bStateChanged)
 562     {
 563         for(i=0; i<2; i++)
 564         {
 565             for(j=0;j<DIM;j++)
 566             {
 567                 fr->mu_tot[i][j] = mu[i*DIM + j];
 568             }
 569         }
 570     }
 571     if (fr->efep == efepNO)
 572     {
 573         copy_rvec(fr->mu_tot[0],mu_tot);
 574     }
 575     else
 576     {
 577         for(j=0; j<DIM; j++)
 578         {
 579             mu_tot[j] =
 580                 (1.0 - lambda)*fr->mu_tot[0][j] + lambda*fr->mu_tot[1][j];
 581         }
 582     }
 583
 584     /* Reset energies */
 585     reset_enerdata(&(inputrec->opts),fr,bNS,enerd,MASTER(cr));
 586     clear_rvecs(SHIFTS,fr->fshift);
 587
 588     if (bNS)
 589     {
 590         wallcycle_start(wcycle,ewcNS);
 591
 592         if (graph && bStateChanged)
 593         {
 594             /* Calculate intramolecular shift vectors to make molecules whole */
 595             mk_mshift(fplog,graph,fr->ePBC,box,x);
 596         }
 597
 598         /* Reset long range forces if necessary */
 599         if (fr->bTwinRange)
 600         {
 601             /* Reset the (long-range) forces if necessary */
 602             clear_rvecs(fr->natoms_force_constr,bSepLRF ? fr->f_twin : f);
 603         }
 604
 605         /* Do the actual neighbour searching and if twin range electrostatics
 606          * also do the calculation of long range forces and energies.
 607          */
 608         dvdl = 0;
 609         ns(fplog,fr,x,box,
 610            groups,&(inputrec->opts),top,mdatoms,
 611            cr,nrnb,lambda,&dvdl,&enerd->grpp,bFillGrid,
 612            bDoLongRange,bDoForces,bSepLRF ? fr->f_twin : f);
 613         if (bSepDVDL)
 614         {
 615             fprintf(fplog,sepdvdlformat,"LR non-bonded",0.0,dvdl);
 616         }
 617         enerd->dvdl_lin += dvdl;
 618
 619         wallcycle_stop(wcycle,ewcNS);
 620     }
 621
 622     if (inputrec->implicit_solvent && bNS)
 623     {
 624         make_gb_nblist(cr,inputrec->gb_algorithm,inputrec->rlist,
 625                        x,box,fr,&top->idef,graph,fr->born);
 626     }
 627
 628     if (DOMAINDECOMP(cr))
 629     {
 630         if (!(cr->duty & DUTY_PME))
 631         {
 632             wallcycle_start(wcycle,ewcPPDURINGPME);
 633             dd_force_flop_start(cr->dd,nrnb);
 634         }
 635     }
 636
 637     /* Start the force cycle counter.
 638      * This counter is stopped in do_forcelow_level.
 639      * No parallel communication should occur while this counter is running,
 640      * since that will interfere with the dynamic load balancing.
 641      */
 642     wallcycle_start(wcycle,ewcFORCE);
 643
 644     if (bDoForces)
 645     {
 646         /* Reset forces for which the virial is calculated separately:
 647          * PME/Ewald forces if necessary */
 648         if (fr->bF_NoVirSum)
 649         {
 650             if (flags & GMX_FORCE_VIRIAL)
 651             {
 652                 fr->f_novirsum = fr->f_novirsum_alloc;
 653                 GMX_BARRIER(cr->mpi_comm_mygroup);
 654                 if (fr->bDomDec)
 655                 {
 656                     clear_rvecs(fr->f_novirsum_n,fr->f_novirsum);
 657                 }
 658                 else
 659                 {
 660                     clear_rvecs(homenr,fr->f_novirsum+start);
 661                 }
 662                 GMX_BARRIER(cr->mpi_comm_mygroup);
 663             }
 664             else
 665             {
 666                 /* We are not calculating the pressure so we do not need
 667                  * a separate array for forces that do not contribute
 668                  * to the pressure.
 669                  */
 670                 fr->f_novirsum = f;
 671             }
 672         }
 673
 674         if (bSepLRF)
 675         {
 676             /* Add the long range forces to the short range forces */
 677             for(i=0; i<fr->natoms_force_constr; i++)
 678             {
 679                 copy_rvec(fr->f_twin[i],f[i]);
 680             }
 681         }
 682         else if (!(fr->bTwinRange && bNS))
 683         {
 684             /* Clear the short-range forces */
 685             clear_rvecs(fr->natoms_force_constr,f);
 686         }
 687
 688         clear_rvec(fr->vir_diag_posres);
 689
 690         GMX_BARRIER(cr->mpi_comm_mygroup);
 691     }
 692     if (inputrec->ePull == epullCONSTRAINT)
 693     {
 694         clear_pull_forces(inputrec->pull);
 695     }
 696
 697     /* update QMMMrec, if necessary */
 698     if(fr->bQMMM)
 699     {
 700         update_QMMMrec(cr,fr,x,mdatoms,box,top);
 701     }
 702
 703     if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
 704     {
 705         /* Position restraints always require full pbc */
 706         set_pbc(&pbc,inputrec->ePBC,box);
 707         v = posres(top->idef.il[F_POSRES].nr,top->idef.il[F_POSRES].iatoms,
 708                    top->idef.iparams_posres,
 709                    (const rvec*)x,fr->f_novirsum,fr->vir_diag_posres,
 710                    inputrec->ePBC==epbcNONE ? NULL : &pbc,lambda,&dvdl,
 711                    fr->rc_scaling,fr->ePBC,fr->posres_com,fr->posres_comB);
 712         if (bSepDVDL)
 713         {
 714             fprintf(fplog,sepdvdlformat,
 715                     interaction_function[F_POSRES].longname,v,dvdl);
 716         }
 717         enerd->term[F_POSRES] += v;
 718         /* This linear lambda dependence assumption is only correct
 719          * when only k depends on lambda,
 720          * not when the reference position depends on lambda.
 721          * grompp checks for this.
 722          */
 723         enerd->dvdl_lin += dvdl;
 724         inc_nrnb(nrnb,eNR_POSRES,top->idef.il[F_POSRES].nr/2);
 725     }
 726
 727     /* Compute the bonded and non-bonded energies and optionally forces */
 728     do_force_lowlevel(fplog,step,fr,inputrec,&(top->idef),
 729                       cr,nrnb,wcycle,mdatoms,&(inputrec->opts),
 730                       x,hist,f,enerd,fcd,mtop,top,fr->born,
 731                       &(top->atomtypes),bBornRadii,box,
 732                       lambda,graph,&(top->excls),fr->mu_tot,
 733                       flags,&cycles_pme);
 734
 735     cycles_force = wallcycle_stop(wcycle,ewcFORCE);
 736     GMX_BARRIER(cr->mpi_comm_mygroup);
 737
 738     if (ed)
 739     {
 740         do_flood(fplog,cr,x,f,ed,box,step);
 741     }
 742
 743     if (DOMAINDECOMP(cr))
 744     {
 745         dd_force_flop_stop(cr->dd,nrnb);
 746         if (wcycle)
 747         {
 748             dd_cycles_add(cr->dd,cycles_force-cycles_pme,ddCyclF);
 749         }
 750     }
 751
 752     if (bDoForces)
 753     {
 754         if (IR_ELEC_FIELD(*inputrec))
 755         {
 756             /* Compute forces due to electric field */
 757             calc_f_el(MASTER(cr) ? field : NULL,
 758                       start,homenr,mdatoms->chargeA,x,fr->f_novirsum,
 759                       inputrec->ex,inputrec->et,t);
 760         }
 761
 762         /* Communicate the forces */
 763         if (PAR(cr))
 764         {
 765             wallcycle_start(wcycle,ewcMOVEF);
 766             if (DOMAINDECOMP(cr))
 767             {
 768                 dd_move_f(cr->dd,f,fr->fshift);
 769                 /* Do we need to communicate the separate force array
 770                  * for terms that do not contribute to the single sum virial?
 771                  * Position restraints and electric fields do not introduce
 772                  * inter-cg forces, only full electrostatics methods do.
 773                  * When we do not calculate the virial, fr->f_novirsum = f,
 774                  * so we have already communicated these forces.
 775                  */
 776                 if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
 777                     (flags & GMX_FORCE_VIRIAL))
 778                 {
 779                     dd_move_f(cr->dd,fr->f_novirsum,NULL);
 780                 }
 781                 if (bSepLRF)
 782                 {
 783                     /* We should not update the shift forces here,
 784                      * since f_twin is already included in f.
 785                      */
 786                     dd_move_f(cr->dd,fr->f_twin,NULL);
 787                 }
 788             }
 789             else
 790             {
 791                 pd_move_f(cr,f,nrnb);
 792                 if (bSepLRF)
 793                 {
 794                     pd_move_f(cr,fr->f_twin,nrnb);
 795                 }
 796             }
 797             wallcycle_stop(wcycle,ewcMOVEF);
 798         }
 799
 800         /* If we have NoVirSum forces, but we do not calculate the virial,
 801          * we sum fr->f_novirum=f later.
 802          */
 803         if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
 804         {
 805             wallcycle_start(wcycle,ewcVSITESPREAD);
 806             spread_vsite_f(fplog,vsite,x,f,fr->fshift,nrnb,
 807                            &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
 808             wallcycle_stop(wcycle,ewcVSITESPREAD);
 809
 810             if (bSepLRF)
 811             {
 812                 wallcycle_start(wcycle,ewcVSITESPREAD);
 813                 spread_vsite_f(fplog,vsite,x,fr->f_twin,NULL,
 814                                nrnb,
 815                                &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
 816                 wallcycle_stop(wcycle,ewcVSITESPREAD);
 817             }
 818         }
 819
 820         if (flags & GMX_FORCE_VIRIAL)
 821         {
 822             /* Calculation of the virial must be done after vsites! */
 823             calc_virial(fplog,mdatoms->start,mdatoms->homenr,x,f,
 824                         vir_force,graph,box,nrnb,fr,inputrec->ePBC);
 825         }
 826     }
 827
 828     if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
 829     {
 830         /* Calculate the center of mass forces, this requires communication,
 831          * which is why pull_potential is called close to other communication.
 832          * The virial contribution is calculated directly,
 833          * which is why we call pull_potential after calc_virial.
 834          */
 835         set_pbc(&pbc,inputrec->ePBC,box);
 836         dvdl = 0;
 837         enerd->term[F_COM_PULL] =
 838             pull_potential(inputrec->ePull,inputrec->pull,mdatoms,&pbc,
 839                            cr,t,lambda,x,f,vir_force,&dvdl);
 840         if (bSepDVDL)
 841         {
 842             fprintf(fplog,sepdvdlformat,"Com pull",enerd->term[F_COM_PULL],dvdl);
 843         }
 844         enerd->dvdl_lin += dvdl;
 845     }
 846
 847     if (PAR(cr) && !(cr->duty & DUTY_PME))
 848     {
 849         cycles_ppdpme = wallcycle_stop(wcycle,ewcPPDURINGPME);
 850         dd_cycles_add(cr->dd,cycles_ppdpme,ddCyclPPduringPME);
 851
 852         /* In case of node-splitting, the PP nodes receive the long-range
 853          * forces, virial and energy from the PME nodes here.
 854          */
 855         wallcycle_start(wcycle,ewcPP_PMEWAITRECVF);
 856         dvdl = 0;
 857         gmx_pme_receive_f(cr,fr->f_novirsum,fr->vir_el_recip,&e,&dvdl,
 858                           &cycles_seppme);
 859         if (bSepDVDL)
 860         {
 861             fprintf(fplog,sepdvdlformat,"PME mesh",e,dvdl);
 862         }
 863         enerd->term[F_COUL_RECIP] += e;
 864         enerd->dvdl_lin += dvdl;
 865         if (wcycle)
 866         {
 867             dd_cycles_add(cr->dd,cycles_seppme,ddCyclPME);
 868         }
 869         wallcycle_stop(wcycle,ewcPP_PMEWAITRECVF);
 870     }
 871
 872     if (bDoForces && fr->bF_NoVirSum)
 873     {
 874         if (vsite)
 875         {
 876             /* Spread the mesh force on virtual sites to the other particles...
 877              * This is parallellized. MPI communication is performed
 878              * if the constructing atoms aren't local.
 879              */
 880             wallcycle_start(wcycle,ewcVSITESPREAD);
 881             spread_vsite_f(fplog,vsite,x,fr->f_novirsum,NULL,nrnb,
 882                            &top->idef,fr->ePBC,fr->bMolPBC,graph,box,cr);
 883             wallcycle_stop(wcycle,ewcVSITESPREAD);
 884         }
 885         if (flags & GMX_FORCE_VIRIAL)
 886         {
 887             /* Now add the forces, this is local */
 888             if (fr->bDomDec)
 889             {
 890                 sum_forces(0,fr->f_novirsum_n,f,fr->f_novirsum);
 891             }
 892             else
 893             {
 894                 sum_forces(start,start+homenr,f,fr->f_novirsum);
 895             }
 896             if (EEL_FULL(fr->eeltype))
 897             {
 898                 /* Add the mesh contribution to the virial */
 899                 m_add(vir_force,fr->vir_el_recip,vir_force);
 900             }
 901             if (debug)
 902             {
 903                 pr_rvecs(debug,0,"vir_force",vir_force,DIM);
 904             }
 905         }
 906     }
 907
 908     /* Sum the potential energy terms from group contributions */
 909     sum_epot(&(inputrec->opts),enerd);
 910
 911     if (fr->print_force >= 0 && bDoForces)
 912     {
 913         print_large_forces(stderr,mdatoms,cr,step,fr->print_force,x,f);
 914     }
 915 }
 916
 917 void do_constrain_first(FILE *fplog,gmx_constr_t constr,
 918                         t_inputrec *ir,t_mdatoms *md,
 919                         t_state *state,rvec *f,
 920                         t_graph *graph,t_commrec *cr,t_nrnb *nrnb,
 921                         t_forcerec *fr, gmx_localtop_t *top, tensor shake_vir)
 922 {
 923     int    i,m,start,end;
 924     gmx_large_int_t step;
 925     double mass,tmass,vcm[4];
 926     real   dt=ir->delta_t;
 927     real   dvdlambda;
 928     rvec   *savex;
 929
 930     snew(savex,state->natoms);
 931
 932     start = md->start;
 933     end   = md->homenr + start;
 934
 935     if (debug)
 936         fprintf(debug,"vcm: start=%d, homenr=%d, end=%d\n",
 937                 start,md->homenr,end);
 938     /* Do a first constrain to reset particles... */
 939     step = ir->init_step;
 940     if (fplog)
 941     {
 942         char buf[STEPSTRSIZE];
 943         fprintf(fplog,"\nConstraining the starting coordinates (step %s)\n",
 944                 gmx_step_str(step,buf));
 945     }
 946     dvdlambda = 0;
 947
 948     /* constrain the current position */
 949     constrain(NULL,TRUE,FALSE,constr,&(top->idef),
 950               ir,NULL,cr,step,0,md,
 951               state->x,state->x,NULL,
 952               state->box,state->lambda,&dvdlambda,
 953               NULL,NULL,nrnb,econqCoord,ir->epc==epcMTTK,state->veta,state->veta);
 954     if (EI_VV(ir->eI))
 955     {
 956         /* constrain the inital velocity, and save it */
 957         /* also may be useful if we need the ekin from the halfstep for velocity verlet */
 958         /* might not yet treat veta correctly */
 959         constrain(NULL,TRUE,FALSE,constr,&(top->idef),
 960                   ir,NULL,cr,step,0,md,
 961                   state->x,state->v,state->v,
 962                   state->box,state->lambda,&dvdlambda,
 963                   NULL,NULL,nrnb,econqVeloc,ir->epc==epcMTTK,state->veta,state->veta);
 964     }
 965     /* constrain the inital velocities at t-dt/2 */
 966     if (EI_STATE_VELOCITY(ir->eI) && ir->eI!=eiVV)
 967     {
 968         for(i=start; (i<end); i++)
 969         {
 970             for(m=0; (m<DIM); m++)
 971             {
 972                 /* Reverse the velocity */
 973                 state->v[i][m] = -state->v[i][m];
 974                 /* Store the position at t-dt in buf */
 975                 savex[i][m] = state->x[i][m] + dt*state->v[i][m];
 976             }
 977         }
 978     /* Shake the positions at t=-dt with the positions at t=0
 979      * as reference coordinates.
 980          */
 981         if (fplog)
 982         {
 983             char buf[STEPSTRSIZE];
 984             fprintf(fplog,"\nConstraining the coordinates at t0-dt (step %s)\n",
 985                     gmx_step_str(step,buf));
 986         }
 987         dvdlambda = 0;
 988         constrain(NULL,TRUE,FALSE,constr,&(top->idef),
 989                   ir,NULL,cr,step,-1,md,
 990                   state->x,savex,NULL,
 991                   state->box,state->lambda,&dvdlambda,
 992                   state->v,NULL,nrnb,econqCoord,ir->epc==epcMTTK,state->veta,state->veta);
 993
 994         for(i=start; i<end; i++) {
 995             for(m=0; m<DIM; m++) {
 996                 /* Re-reverse the velocities */
 997                 state->v[i][m] = -state->v[i][m];
 998             }
 999         }
1000     }
1001
1002     for(m=0; (m<4); m++)
1003         vcm[m] = 0;
1004     for(i=start; i<end; i++) {
1005         mass = md->massT[i];
1006         for(m=0; m<DIM; m++) {
1007             vcm[m] += state->v[i][m]*mass;
1008         }
1009         vcm[3] += mass;
1010     }
1011
1012     if (ir->nstcomm != 0 || debug) {
1013         /* Compute the global sum of vcm */
1014         if (debug)
1015             fprintf(debug,"vcm: %8.3f  %8.3f  %8.3f,"
1016                     " total mass = %12.5e\n",vcm[XX],vcm[YY],vcm[ZZ],vcm[3]);
1017         if (PAR(cr))
1018             gmx_sumd(4,vcm,cr);
1019         tmass = vcm[3];
1020         for(m=0; (m<DIM); m++)
1021             vcm[m] /= tmass;
1022         if (debug)
1023             fprintf(debug,"vcm: %8.3f  %8.3f  %8.3f,"
1024                     " total mass = %12.5e\n",vcm[XX],vcm[YY],vcm[ZZ],tmass);
1025         if (ir->nstcomm != 0) {
1026             /* Now we have the velocity of center of mass, let's remove it */
1027             for(i=start; (i<end); i++) {
1028                 for(m=0; (m<DIM); m++)
1029                     state->v[i][m] -= vcm[m];
1030             }
1031
1032         }
1033     }
1034     sfree(savex);
1035 }
1036
1037 void calc_enervirdiff(FILE *fplog,int eDispCorr,t_forcerec *fr)
1038 {
1039   double eners[2],virs[2],enersum,virsum,y0,f,g,h;
1040   double r0,r1,r,rc3,rc9,ea,eb,ec,pa,pb,pc,pd;
1041   double invscale,invscale2,invscale3;
1042   int    ri0,ri1,ri,i,offstart,offset;
1043   real   scale,*vdwtab;
1044
1045   fr->enershiftsix = 0;
1046   fr->enershifttwelve = 0;
1047   fr->enerdiffsix = 0;
1048   fr->enerdifftwelve = 0;
1049   fr->virdiffsix = 0;
1050   fr->virdifftwelve = 0;
1051
1052   if (eDispCorr != edispcNO) {
1053     for(i=0; i<2; i++) {
1054       eners[i] = 0;
1055       virs[i]  = 0;
1056     }
1057     if ((fr->vdwtype == evdwSWITCH) || (fr->vdwtype == evdwSHIFT)) {
1058       if (fr->rvdw_switch == 0)
1059         gmx_fatal(FARGS,
1060                   "With dispersion correction rvdw-switch can not be zero "
1061                   "for vdw-type = %s",evdw_names[fr->vdwtype]);
1062
1063       scale  = fr->nblists[0].tab.scale;
1064       vdwtab = fr->nblists[0].vdwtab;
1065
1066       /* Round the cut-offs to exact table values for precision */
1067       ri0 = floor(fr->rvdw_switch*scale);
1068       ri1 = ceil(fr->rvdw*scale);
1069       r0  = ri0/scale;
1070       r1  = ri1/scale;
1071       rc3 = r0*r0*r0;
1072       rc9  = rc3*rc3*rc3;
1073
1074       if (fr->vdwtype == evdwSHIFT) {
1075         /* Determine the constant energy shift below rvdw_switch */
1076         fr->enershiftsix    = (real)(-1.0/(rc3*rc3)) - vdwtab[8*ri0];
1077         fr->enershifttwelve = (real)( 1.0/(rc9*rc3)) - vdwtab[8*ri0 + 4];
1078       }
1079       /* Add the constant part from 0 to rvdw_switch.
1080        * This integration from 0 to rvdw_switch overcounts the number
1081        * of interactions by 1, as it also counts the self interaction.
1082        * We will correct for this later.
1083        */
1084       eners[0] += 4.0*M_PI*fr->enershiftsix*rc3/3.0;
1085       eners[1] += 4.0*M_PI*fr->enershifttwelve*rc3/3.0;
1086
1087       invscale = 1.0/(scale);
1088       invscale2 = invscale*invscale;
1089       invscale3 = invscale*invscale2;
1090
1091       /* following summation derived from cubic spline definition,
1092         Numerical Recipies in C, second edition, p. 113-116.  Exact
1093         for the cubic spline.  We first calculate the negative of
1094         the energy from rvdw to rvdw_switch, assuming that g(r)=1,
1095         and then add the more standard, abrupt cutoff correction to
1096         that result, yielding the long-range correction for a
1097         switched function.  We perform both the pressure and energy
1098         loops at the same time for simplicity, as the computational
1099         cost is low. */
1100
1101       for (i=0;i<2;i++) {
1102         enersum = 0.0; virsum = 0.0;
1103         if (i==0)
1104           offstart = 0;
1105         else
1106           offstart = 4;
1107         for (ri=ri0; ri<ri1; ri++) {
1108           r = ri*invscale;
1109           ea = invscale3;
1110           eb = 2.0*invscale2*r;
1111           ec = invscale*r*r;
1112
1113           pa = invscale3;
1114           pb = 3.0*invscale2*r;
1115           pc = 3.0*invscale*r*r;
1116           pd = r*r*r;
1117
1118           /* this "8" is from the packing in the vdwtab array - perhaps
1119             should be #define'ed? */
1120           offset = 8*ri + offstart;
1121           y0 = vdwtab[offset];
1122           f = vdwtab[offset+1];
1123           g = vdwtab[offset+2];
1124           h = vdwtab[offset+3];
1125
1126           enersum += y0*(ea/3 + eb/2 + ec) + f*(ea/4 + eb/3 + ec/2)+
1127             g*(ea/5 + eb/4 + ec/3) + h*(ea/6 + eb/5 + ec/4);
1128           virsum  +=  f*(pa/4 + pb/3 + pc/2 + pd) +
1129             2*g*(pa/5 + pb/4 + pc/3 + pd/2) + 3*h*(pa/6 + pb/5 + pc/4 + pd/3);
1130
1131         }
1132         enersum *= 4.0*M_PI;
1133         virsum  *= 4.0*M_PI;
1134         eners[i] -= enersum;
1135         virs[i]  -= virsum;
1136       }
1137
1138       /* now add the correction for rvdw_switch to infinity */
1139       eners[0] += -4.0*M_PI/(3.0*rc3);
1140       eners[1] +=  4.0*M_PI/(9.0*rc9);
1141       virs[0]  +=  8.0*M_PI/rc3;
1142       virs[1]  += -16.0*M_PI/(3.0*rc9);
1143     }
1144     else if ((fr->vdwtype == evdwCUT) || (fr->vdwtype == evdwUSER)) {
1145       if (fr->vdwtype == evdwUSER && fplog)
1146         fprintf(fplog,
1147                 "WARNING: using dispersion correction with user tables\n");
1148       rc3  = fr->rvdw*fr->rvdw*fr->rvdw;
1149       rc9  = rc3*rc3*rc3;
1150       eners[0] += -4.0*M_PI/(3.0*rc3);
1151       eners[1] +=  4.0*M_PI/(9.0*rc9);
1152       virs[0]  +=  8.0*M_PI/rc3;
1153       virs[1]  += -16.0*M_PI/(3.0*rc9);
1154     } else {
1155       gmx_fatal(FARGS,
1156                 "Dispersion correction is not implemented for vdw-type = %s",
1157                 evdw_names[fr->vdwtype]);
1158     }
1159     fr->enerdiffsix    = eners[0];
1160     fr->enerdifftwelve = eners[1];
1161     /* The 0.5 is due to the Gromacs definition of the virial */
1162     fr->virdiffsix     = 0.5*virs[0];
1163     fr->virdifftwelve  = 0.5*virs[1];
1164   }
1165 }
1166
1167 void calc_dispcorr(FILE *fplog,t_inputrec *ir,t_forcerec *fr,
1168                    gmx_large_int_t step,int natoms,
1169                    matrix box,real lambda,tensor pres,tensor virial,
1170                    real *prescorr, real *enercorr, real *dvdlcorr)
1171 {
1172     gmx_bool bCorrAll,bCorrPres;
1173     real dvdlambda,invvol,dens,ninter,avcsix,avctwelve,enerdiff,svir=0,spres=0;
1174     int  m;
1175
1176     *prescorr = 0;
1177     *enercorr = 0;
1178     *dvdlcorr = 0;
1179
1180     clear_mat(virial);
1181     clear_mat(pres);
1182
1183     if (ir->eDispCorr != edispcNO) {
1184         bCorrAll  = (ir->eDispCorr == edispcAllEner ||
1185                      ir->eDispCorr == edispcAllEnerPres);
1186         bCorrPres = (ir->eDispCorr == edispcEnerPres ||
1187                      ir->eDispCorr == edispcAllEnerPres);
1188
1189         invvol = 1/det(box);
1190         if (fr->n_tpi)
1191         {
1192             /* Only correct for the interactions with the inserted molecule */
1193             dens = (natoms - fr->n_tpi)*invvol;
1194             ninter = fr->n_tpi;
1195         }
1196         else
1197         {
1198             dens = natoms*invvol;
1199             ninter = 0.5*natoms;
1200         }
1201
1202         if (ir->efep == efepNO)
1203         {
1204             avcsix    = fr->avcsix[0];
1205             avctwelve = fr->avctwelve[0];
1206         }
1207         else
1208         {
1209             avcsix    = (1 - lambda)*fr->avcsix[0]    + lambda*fr->avcsix[1];
1210             avctwelve = (1 - lambda)*fr->avctwelve[0] + lambda*fr->avctwelve[1];
1211         }
1212
1213         enerdiff = ninter*(dens*fr->enerdiffsix - fr->enershiftsix);
1214         *enercorr += avcsix*enerdiff;
1215         dvdlambda = 0.0;
1216         if (ir->efep != efepNO)
1217         {
1218             dvdlambda += (fr->avcsix[1] - fr->avcsix[0])*enerdiff;
1219         }
1220         if (bCorrAll)
1221         {
1222             enerdiff = ninter*(dens*fr->enerdifftwelve - fr->enershifttwelve);
1223             *enercorr += avctwelve*enerdiff;
1224             if (fr->efep != efepNO)
1225             {
1226                 dvdlambda += (fr->avctwelve[1] - fr->avctwelve[0])*enerdiff;
1227             }
1228         }
1229
1230         if (bCorrPres)
1231         {
1232             svir = ninter*dens*avcsix*fr->virdiffsix/3.0;
1233             if (ir->eDispCorr == edispcAllEnerPres)
1234             {
1235                 svir += ninter*dens*avctwelve*fr->virdifftwelve/3.0;
1236             }
1237             /* The factor 2 is because of the Gromacs virial definition */
1238             spres = -2.0*invvol*svir*PRESFAC;
1239
1240             for(m=0; m<DIM; m++) {
1241                 virial[m][m] += svir;
1242                 pres[m][m] += spres;
1243             }
1244             *prescorr += spres;
1245         }
1246
1247         /* Can't currently control when it prints, for now, just print when degugging */
1248         if (debug)
1249         {
1250             if (bCorrAll) {
1251                 fprintf(debug,"Long Range LJ corr.: <C6> %10.4e, <C12> %10.4e\n",
1252                         avcsix,avctwelve);
1253             }
1254             if (bCorrPres)
1255             {
1256                 fprintf(debug,
1257                         "Long Range LJ corr.: Epot %10g, Pres: %10g, Vir: %10g\n",
1258                         *enercorr,spres,svir);
1259             }
1260             else
1261             {
1262                 fprintf(debug,"Long Range LJ corr.: Epot %10g\n",*enercorr);
1263             }
1264         }
1265
1266         if (fr->bSepDVDL && do_per_step(step,ir->nstlog))
1267         {
1268             fprintf(fplog,sepdvdlformat,"Dispersion correction",
1269                     *enercorr,dvdlambda);
1270         }
1271         if (fr->efep != efepNO)
1272         {
1273             *dvdlcorr += dvdlambda;
1274         }
1275     }
1276 }
1277
1278 void do_pbc_first(FILE *fplog,matrix box,t_forcerec *fr,
1279                   t_graph *graph,rvec x[])
1280 {
1281   if (fplog)
1282     fprintf(fplog,"Removing pbc first time\n");
1283   calc_shifts(box,fr->shift_vec);
1284   if (graph) {
1285     mk_mshift(fplog,graph,fr->ePBC,box,x);
1286     if (gmx_debug_at)
1287       p_graph(debug,"do_pbc_first 1",graph);
1288     shift_self(graph,box,x);
1289     /* By doing an extra mk_mshift the molecules that are broken
1290      * because they were e.g. imported from another software
1291      * will be made whole again. Such are the healing powers
1292      * of GROMACS.
1293      */
1294     mk_mshift(fplog,graph,fr->ePBC,box,x);
1295     if (gmx_debug_at)
1296       p_graph(debug,"do_pbc_first 2",graph);
1297   }
1298   if (fplog)
1299     fprintf(fplog,"Done rmpbc\n");
1300 }
1301
1302 static void low_do_pbc_mtop(FILE *fplog,int ePBC,matrix box,
1303                             gmx_mtop_t *mtop,rvec x[],
1304                             gmx_bool bFirst)
1305 {
1306   t_graph *graph;
1307   int mb,as,mol;
1308   gmx_molblock_t *molb;
1309
1310   if (bFirst && fplog)
1311     fprintf(fplog,"Removing pbc first time\n");
1312
1313   snew(graph,1);
1314   as = 0;
1315   for(mb=0; mb<mtop->nmolblock; mb++) {
1316     molb = &mtop->molblock[mb];
1317     if (molb->natoms_mol == 1 ||
1318         (!bFirst && mtop->moltype[molb->type].cgs.nr == 1)) {
1319       /* Just one atom or charge group in the molecule, no PBC required */
1320       as += molb->nmol*molb->natoms_mol;
1321     } else {
1322       /* Pass NULL iso fplog to avoid graph prints for each molecule type */
1323       mk_graph_ilist(NULL,mtop->moltype[molb->type].ilist,
1324                      0,molb->natoms_mol,FALSE,FALSE,graph);
1325
1326       for(mol=0; mol<molb->nmol; mol++) {
1327         mk_mshift(fplog,graph,ePBC,box,x+as);
1328
1329         shift_self(graph,box,x+as);
1330         /* The molecule is whole now.
1331          * We don't need the second mk_mshift call as in do_pbc_first,
1332          * since we no longer need this graph.
1333          */
1334
1335         as += molb->natoms_mol;
1336       }
1337       done_graph(graph);
1338     }
1339   }
1340   sfree(graph);
1341 }
1342
1343 void do_pbc_first_mtop(FILE *fplog,int ePBC,matrix box,
1344                        gmx_mtop_t *mtop,rvec x[])
1345 {
1346   low_do_pbc_mtop(fplog,ePBC,box,mtop,x,TRUE);
1347 }
1348
1349 void do_pbc_mtop(FILE *fplog,int ePBC,matrix box,
1350                  gmx_mtop_t *mtop,rvec x[])
1351 {
1352   low_do_pbc_mtop(fplog,ePBC,box,mtop,x,FALSE);
1353 }
1354
1355 void finish_run(FILE *fplog,t_commrec *cr,const char *confout,
1356                 t_inputrec *inputrec,
1357                 t_nrnb nrnb[],gmx_wallcycle_t wcycle,
1358                 gmx_runtime_t *runtime,
1359                 gmx_bool bWriteStat)
1360 {
1361   int    i,j;
1362   t_nrnb *nrnb_tot=NULL;
1363   real   delta_t;
1364   double nbfs,mflop;
1365   double cycles[ewcNR];
1366
1367   wallcycle_sum(cr,wcycle,cycles);
1368
1369   if (cr->nnodes > 1) {
1370     if (SIMMASTER(cr))
1371       snew(nrnb_tot,1);
1372 #ifdef GMX_MPI
1373     MPI_Reduce(nrnb->n,nrnb_tot->n,eNRNB,MPI_DOUBLE,MPI_SUM,
1374                MASTERRANK(cr),cr->mpi_comm_mysim);
1375 #endif
1376   } else {
1377     nrnb_tot = nrnb;
1378   }
1379
1380   if (SIMMASTER(cr)) {
1381     print_flop(fplog,nrnb_tot,&nbfs,&mflop);
1382     if (cr->nnodes > 1) {
1383       sfree(nrnb_tot);
1384     }
1385   }
1386
1387   if ((cr->duty & DUTY_PP) && DOMAINDECOMP(cr)) {
1388     print_dd_statistics(cr,inputrec,fplog);
1389   }
1390
1391 #ifdef GMX_MPI
1392     if (PARTDECOMP(cr))
1393     {
1394         if (MASTER(cr))
1395         {
1396             t_nrnb     *nrnb_all;
1397             int        s;
1398             MPI_Status stat;
1399
1400             snew(nrnb_all,cr->nnodes);
1401             nrnb_all[0] = *nrnb;
1402             for(s=1; s<cr->nnodes; s++)
1403             {
1404                 MPI_Recv(nrnb_all[s].n,eNRNB,MPI_DOUBLE,s,0,
1405                          cr->mpi_comm_mysim,&stat);
1406             }
1407             pr_load(fplog,cr,nrnb_all);
1408             sfree(nrnb_all);
1409         }
1410         else
1411         {
1412             MPI_Send(nrnb->n,eNRNB,MPI_DOUBLE,MASTERRANK(cr),0,
1413                      cr->mpi_comm_mysim);
1414         }
1415     }
1416 #endif
1417
1418   if (SIMMASTER(cr)) {
1419     wallcycle_print(fplog,cr->nnodes,cr->npmenodes,runtime->realtime,
1420                     wcycle,cycles);
1421
1422     if (EI_DYNAMICS(inputrec->eI)) {
1423       delta_t = inputrec->delta_t;
1424     } else {
1425       delta_t = 0;
1426     }
1427
1428     if (fplog) {
1429         print_perf(fplog,runtime->proctime,runtime->realtime,
1430                    cr->nnodes-cr->npmenodes,
1431                    runtime->nsteps_done,delta_t,nbfs,mflop);
1432     }
1433     if (bWriteStat) {
1434         print_perf(stderr,runtime->proctime,runtime->realtime,
1435                    cr->nnodes-cr->npmenodes,
1436                    runtime->nsteps_done,delta_t,nbfs,mflop);
1437     }
1438
1439     /*
1440     runtime=inputrec->nsteps*inputrec->delta_t;
1441     if (bWriteStat) {
1442       if (cr->nnodes == 1)
1443         fprintf(stderr,"\n\n");
1444       print_perf(stderr,nodetime,realtime,runtime,&ntot,
1445                  cr->nnodes-cr->npmenodes,FALSE);
1446     }
1447     wallcycle_print(fplog,cr->nnodes,cr->npmenodes,realtime,wcycle,cycles);
1448     print_perf(fplog,nodetime,realtime,runtime,&ntot,cr->nnodes-cr->npmenodes,
1449                TRUE);
1450     if (PARTDECOMP(cr))
1451       pr_load(fplog,cr,nrnb_all);
1452     if (cr->nnodes > 1)
1453       sfree(nrnb_all);
1454     */
1455   }
1456 }
1457
1458 void init_md(FILE *fplog,
1459              t_commrec *cr,t_inputrec *ir,const output_env_t oenv,
1460              double *t,double *t0,
1461              real *lambda,double *lam0,
1462              t_nrnb *nrnb,gmx_mtop_t *mtop,
1463              gmx_update_t *upd,
1464              int nfile,const t_filenm fnm[],
1465              gmx_mdoutf_t **outf,t_mdebin **mdebin,
1466              tensor force_vir,tensor shake_vir,rvec mu_tot,
1467              gmx_bool *bSimAnn,t_vcm **vcm, t_state *state, unsigned long Flags)
1468 {
1469     int  i,j,n;
1470     real tmpt,mod;
1471
1472     /* Initial values */
1473     *t = *t0       = ir->init_t;
1474     if (ir->efep != efepNO)
1475     {
1476         *lam0 = ir->init_lambda;
1477         *lambda = *lam0 + ir->init_step*ir->delta_lambda;
1478     }
1479     else
1480     {
1481         *lambda = *lam0   = 0.0;
1482     }
1483
1484     *bSimAnn=FALSE;
1485     for(i=0;i<ir->opts.ngtc;i++)
1486     {
1487         /* set bSimAnn if any group is being annealed */
1488         if(ir->opts.annealing[i]!=eannNO)
1489         {
1490             *bSimAnn = TRUE;
1491         }
1492     }
1493     if (*bSimAnn)
1494     {
1495         update_annealing_target_temp(&(ir->opts),ir->init_t);
1496     }
1497
1498     if (upd)
1499     {
1500         *upd = init_update(fplog,ir);
1501     }
1502
1503     if (vcm != NULL)
1504     {
1505         *vcm = init_vcm(fplog,&mtop->groups,ir);
1506     }
1507
1508     if (EI_DYNAMICS(ir->eI) && !(Flags & MD_APPENDFILES))
1509     {
1510         if (ir->etc == etcBERENDSEN)
1511         {
1512             please_cite(fplog,"Berendsen84a");
1513         }
1514         if (ir->etc == etcVRESCALE)
1515         {
1516             please_cite(fplog,"Bussi2007a");
1517         }
1518     }
1519
1520     init_nrnb(nrnb);
1521
1522     if (nfile != -1)
1523     {
1524         *outf = init_mdoutf(nfile,fnm,Flags,cr,ir,oenv);
1525
1526         *mdebin = init_mdebin((Flags & MD_APPENDFILES) ? NULL : (*outf)->fp_ene,
1527                               mtop,ir, (*outf)->fp_dhdl);
1528     }
1529
1530     /* Initiate variables */
1531     clear_mat(force_vir);
1532     clear_mat(shake_vir);
1533     clear_rvec(mu_tot);
1534
1535     debug_gmx();
1536 }
1537