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